2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993, 1995
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_quota.h"
43 #include "opt_suiddir.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/malloc.h>
50 #include <sys/namei.h>
51 #include <sys/kernel.h>
52 #include <sys/fcntl.h>
53 #include <sys/filio.h>
57 #include <sys/mount.h>
59 #include <sys/refcount.h>
60 #include <sys/unistd.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/lockf.h>
68 #include <security/audit/audit.h>
69 #include <security/mac/mac_framework.h>
71 #include <sys/file.h> /* XXX */
74 #include <vm/vm_extern.h>
76 #include <ufs/ufs/acl.h>
77 #include <ufs/ufs/extattr.h>
78 #include <ufs/ufs/quota.h>
79 #include <ufs/ufs/inode.h>
80 #include <ufs/ufs/dir.h>
81 #include <ufs/ufs/ufsmount.h>
82 #include <ufs/ufs/ufs_extern.h>
84 #include <ufs/ufs/dirhash.h>
87 #include <ufs/ufs/gjournal.h>
88 FEATURE(ufs_gjournal, "Journaling support through GEOM for UFS");
92 FEATURE(ufs_quota, "UFS disk quotas support");
93 FEATURE(ufs_quota64, "64bit UFS disk quotas support");
98 "Give all new files in directory the same ownership as the directory");
103 #include <ufs/ffs/ffs_extern.h>
105 static vop_accessx_t ufs_accessx;
106 static vop_fplookup_vexec_t ufs_fplookup_vexec;
107 static int ufs_chmod(struct vnode *, int, struct ucred *, struct thread *);
108 static int ufs_chown(struct vnode *, uid_t, gid_t, struct ucred *,
110 static vop_close_t ufs_close;
111 static vop_create_t ufs_create;
112 static vop_stat_t ufs_stat;
113 static vop_getattr_t ufs_getattr;
114 static vop_ioctl_t ufs_ioctl;
115 static vop_link_t ufs_link;
116 static int ufs_makeinode(int mode, struct vnode *, struct vnode **,
117 struct componentname *, const char *);
118 static vop_mmapped_t ufs_mmapped;
119 static vop_mkdir_t ufs_mkdir;
120 static vop_mknod_t ufs_mknod;
121 static vop_open_t ufs_open;
122 static vop_pathconf_t ufs_pathconf;
123 static vop_print_t ufs_print;
124 static vop_readlink_t ufs_readlink;
125 static vop_remove_t ufs_remove;
126 static vop_rename_t ufs_rename;
127 static vop_rmdir_t ufs_rmdir;
128 static vop_setattr_t ufs_setattr;
129 static vop_strategy_t ufs_strategy;
130 static vop_symlink_t ufs_symlink;
131 static vop_whiteout_t ufs_whiteout;
132 static vop_close_t ufsfifo_close;
134 SYSCTL_NODE(_vfs, OID_AUTO, ufs, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
138 * A virgin directory (no blushing please).
140 static struct dirtemplate mastertemplate = {
141 0, 12, DT_DIR, 1, ".",
142 0, DIRBLKSIZ - 12, DT_DIR, 2, ".."
144 static struct odirtemplate omastertemplate = {
146 0, DIRBLKSIZ - 12, 2, ".."
150 ufs_itimes_locked(struct vnode *vp)
155 ASSERT_VI_LOCKED(vp, __func__);
160 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
163 if ((vp->v_type == VBLK || vp->v_type == VCHR) && !DOINGSOFTDEP(vp))
164 UFS_INODE_SET_FLAG(ip, IN_LAZYMOD);
165 else if (((vp->v_mount->mnt_kern_flag &
166 (MNTK_SUSPENDED | MNTK_SUSPEND)) == 0) ||
167 (ip->i_flag & (IN_CHANGE | IN_UPDATE)))
168 UFS_INODE_SET_FLAG(ip, IN_MODIFIED);
169 else if (ip->i_flag & IN_ACCESS)
170 UFS_INODE_SET_FLAG(ip, IN_LAZYACCESS);
172 if (ip->i_flag & IN_ACCESS) {
173 DIP_SET(ip, i_atime, ts.tv_sec);
174 DIP_SET(ip, i_atimensec, ts.tv_nsec);
176 if (ip->i_flag & IN_UPDATE) {
177 DIP_SET(ip, i_mtime, ts.tv_sec);
178 DIP_SET(ip, i_mtimensec, ts.tv_nsec);
180 if (ip->i_flag & IN_CHANGE) {
181 DIP_SET(ip, i_ctime, ts.tv_sec);
182 DIP_SET(ip, i_ctimensec, ts.tv_nsec);
183 DIP_SET(ip, i_modrev, DIP(ip, i_modrev) + 1);
187 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE);
191 ufs_itimes(struct vnode *vp)
196 if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0)
200 ufs_itimes_locked(vp);
205 ufs_sync_nlink1(struct mount *mp)
209 error = vfs_busy(mp, 0);
211 VFS_SYNC(mp, MNT_WAIT);
220 ufs_sync_nlink(struct vnode *vp, struct vnode *vp1)
227 if (ip->i_nlink < UFS_LINK_MAX)
229 if (!DOINGSOFTDEP(vp) || ip->i_effnlink >= UFS_LINK_MAX)
237 error = ufs_sync_nlink1(mp);
238 vn_lock_pair(vp, false, vp1, false);
243 * Create a regular file
247 struct vop_create_args /* {
249 struct vnode **a_vpp;
250 struct componentname *a_cnp;
257 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
258 ap->a_dvp, ap->a_vpp, ap->a_cnp, "ufs_create");
261 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
262 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
272 struct vop_mknod_args /* {
274 struct vnode **a_vpp;
275 struct componentname *a_cnp;
279 struct vattr *vap = ap->a_vap;
280 struct vnode **vpp = ap->a_vpp;
285 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
286 ap->a_dvp, vpp, ap->a_cnp, "ufs_mknod");
290 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
291 if (vap->va_rdev != VNOVAL) {
293 * Want to be able to use this to make badblock
294 * inodes, so don't truncate the dev number.
296 DIP_SET(ip, i_rdev, vap->va_rdev);
299 * Remove inode, then reload it through VFS_VGET(). This is
300 * needed to do further inode initialization, for instance
301 * fifo, which was too early for VFS_VGET() done as part of
304 (*vpp)->v_type = VNON;
305 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
308 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
321 ufs_open(struct vop_open_args *ap)
323 struct vnode *vp = ap->a_vp;
326 if (vp->v_type == VCHR || vp->v_type == VBLK)
330 vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td);
331 if (vp->v_type == VREG && (vn_irflag_read(vp) & VIRF_PGREAD) == 0 &&
332 ip->i_ump->um_bsize >= PAGE_SIZE) {
333 vn_irflag_set_cond(vp, VIRF_PGREAD);
337 * Files marked append-only must be opened for appending.
339 if ((ip->i_flags & APPEND) &&
340 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
349 * Update the times on the inode.
354 struct vop_close_args /* {
357 struct ucred *a_cred;
361 struct vnode *vp = ap->a_vp;
369 struct vop_accessx_args /* {
372 struct ucred *a_cred;
376 struct vnode *vp = ap->a_vp;
377 struct inode *ip = VTOI(vp);
378 accmode_t accmode = ap->a_accmode;
386 * Disallow write attempts on read-only filesystems;
387 * unless the file is a socket, fifo, or a block or
388 * character device resident on the filesystem.
390 if (accmode & VMODIFY_PERMS) {
391 switch (vp->v_type) {
395 if (vp->v_mount->mnt_flag & MNT_RDONLY)
399 * Inode is accounted in the quotas only if struct
400 * dquot is attached to it. VOP_ACCESS() is called
401 * from vn_open_cred() and provides a convenient
402 * point to call getinoquota(). The lock mode is
403 * exclusive when the file is opening for write.
405 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
406 error = getinoquota(ip);
418 * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS"
419 * permits the owner of the file to remove the IMMUTABLE flag.
421 if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) &&
422 (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
426 if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) {
427 if (vp->v_mount->mnt_flag & MNT_NFS4ACLS)
428 type = ACL_TYPE_NFS4;
430 type = ACL_TYPE_ACCESS;
432 acl = acl_alloc(M_WAITOK);
433 if (type == ACL_TYPE_NFS4)
434 error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td);
436 error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td);
439 if (type == ACL_TYPE_NFS4) {
440 error = vaccess_acl_nfs4(vp->v_type, ip->i_uid,
441 ip->i_gid, acl, accmode, ap->a_cred);
443 error = vfs_unixify_accmode(&accmode);
445 error = vaccess_acl_posix1e(vp->v_type, ip->i_uid,
446 ip->i_gid, acl, accmode, ap->a_cred);
450 if (error != EOPNOTSUPP)
452 "ufs_accessx(): Error retrieving ACL on object (%d).\n",
455 * XXX: Fall back until debugged. Should
456 * eventually possibly log an error, and return
459 error = vfs_unixify_accmode(&accmode);
461 error = vaccess(vp->v_type, ip->i_mode,
462 ip->i_uid, ip->i_gid, accmode, ap->a_cred);
468 #endif /* !UFS_ACL */
469 error = vfs_unixify_accmode(&accmode);
471 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
472 accmode, ap->a_cred);
477 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
478 * the comment above cache_fplookup for details.
482 struct vop_fplookup_vexec_args /* {
484 struct ucred *a_cred;
495 if (__predict_false(ip == NULL))
501 * ACLs are not supported and UFS clears/sets this flag on mount and
502 * remount. However, we may still be racing with seeing them and there
503 * is no provision to make sure they were accounted for. This matches
504 * the behavior of the locked case, since the lookup there is also
505 * racy: mount takes no measures to block anyone from progressing.
507 all_x = S_IXUSR | S_IXGRP | S_IXOTH;
508 mode = atomic_load_short(&ip->i_mode);
509 if (__predict_true((mode & all_x) == all_x))
513 return (vaccess_vexec_smr(mode, ip->i_uid, ip->i_gid, cred));
518 ufs_stat(struct vop_stat_args *ap)
520 struct vnode *vp = ap->a_vp;
521 struct inode *ip = VTOI(vp);
522 struct stat *sb = ap->a_sb;
525 error = vop_stat_helper_pre(ap);
526 if (__predict_false(error))
530 ufs_itimes_locked(vp);
532 sb->st_atim.tv_sec = ip->i_din1->di_atime;
533 sb->st_atim.tv_nsec = ip->i_din1->di_atimensec;
535 sb->st_atim.tv_sec = ip->i_din2->di_atime;
536 sb->st_atim.tv_nsec = ip->i_din2->di_atimensec;
540 sb->st_dev = dev2udev(ITOUMP(ip)->um_dev);
541 sb->st_ino = ip->i_number;
542 sb->st_mode = (ip->i_mode & ~IFMT) | VTTOIF(vp->v_type);
543 sb->st_nlink = ip->i_effnlink;
544 sb->st_uid = ip->i_uid;
545 sb->st_gid = ip->i_gid;
547 sb->st_rdev = ip->i_din1->di_rdev;
548 sb->st_size = ip->i_din1->di_size;
549 sb->st_mtim.tv_sec = ip->i_din1->di_mtime;
550 sb->st_mtim.tv_nsec = ip->i_din1->di_mtimensec;
551 sb->st_ctim.tv_sec = ip->i_din1->di_ctime;
552 sb->st_ctim.tv_nsec = ip->i_din1->di_ctimensec;
553 sb->st_birthtim.tv_sec = -1;
554 sb->st_birthtim.tv_nsec = 0;
555 sb->st_blocks = dbtob((u_quad_t)ip->i_din1->di_blocks) / S_BLKSIZE;
557 sb->st_rdev = ip->i_din2->di_rdev;
558 sb->st_size = ip->i_din2->di_size;
559 sb->st_mtim.tv_sec = ip->i_din2->di_mtime;
560 sb->st_mtim.tv_nsec = ip->i_din2->di_mtimensec;
561 sb->st_ctim.tv_sec = ip->i_din2->di_ctime;
562 sb->st_ctim.tv_nsec = ip->i_din2->di_ctimensec;
563 sb->st_birthtim.tv_sec = ip->i_din2->di_birthtime;
564 sb->st_birthtim.tv_nsec = ip->i_din2->di_birthnsec;
565 sb->st_blocks = dbtob((u_quad_t)ip->i_din2->di_blocks) / S_BLKSIZE;
568 sb->st_blksize = max(PAGE_SIZE, vp->v_mount->mnt_stat.f_iosize);
569 sb->st_flags = ip->i_flags;
570 sb->st_gen = ip->i_gen;
572 return (vop_stat_helper_post(ap, error));
578 struct vop_getattr_args /* {
581 struct ucred *a_cred;
584 struct vnode *vp = ap->a_vp;
585 struct inode *ip = VTOI(vp);
586 struct vattr *vap = ap->a_vap;
589 ufs_itimes_locked(vp);
591 vap->va_atime.tv_sec = ip->i_din1->di_atime;
592 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec;
594 vap->va_atime.tv_sec = ip->i_din2->di_atime;
595 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec;
599 * Copy from inode table
601 vap->va_fsid = dev2udev(ITOUMP(ip)->um_dev);
602 vap->va_fileid = ip->i_number;
603 vap->va_mode = ip->i_mode & ~IFMT;
604 vap->va_nlink = ip->i_effnlink;
605 vap->va_uid = ip->i_uid;
606 vap->va_gid = ip->i_gid;
608 vap->va_rdev = ip->i_din1->di_rdev;
609 vap->va_size = ip->i_din1->di_size;
610 vap->va_mtime.tv_sec = ip->i_din1->di_mtime;
611 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec;
612 vap->va_ctime.tv_sec = ip->i_din1->di_ctime;
613 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec;
614 vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks);
615 vap->va_filerev = ip->i_din1->di_modrev;
617 vap->va_rdev = ip->i_din2->di_rdev;
618 vap->va_size = ip->i_din2->di_size;
619 vap->va_mtime.tv_sec = ip->i_din2->di_mtime;
620 vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec;
621 vap->va_ctime.tv_sec = ip->i_din2->di_ctime;
622 vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec;
623 vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime;
624 vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec;
625 vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks);
626 vap->va_filerev = ip->i_din2->di_modrev;
628 vap->va_flags = ip->i_flags;
629 vap->va_gen = ip->i_gen;
630 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
631 vap->va_type = IFTOVT(ip->i_mode);
636 * Set attribute vnode op. called from several syscalls
640 struct vop_setattr_args /* {
643 struct ucred *a_cred;
646 struct vattr *vap = ap->a_vap;
647 struct vnode *vp = ap->a_vp;
648 struct inode *ip = VTOI(vp);
649 struct ucred *cred = ap->a_cred;
650 struct thread *td = curthread;
654 * Check for unsettable attributes.
656 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
657 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
658 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
659 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
662 if (vap->va_flags != VNOVAL) {
663 if ((vap->va_flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE |
664 SF_NOUNLINK | SF_SNAPSHOT | UF_APPEND | UF_ARCHIVE |
665 UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | UF_NOUNLINK |
666 UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
667 UF_SPARSE | UF_SYSTEM)) != 0)
669 if (vp->v_mount->mnt_flag & MNT_RDONLY)
672 * Callers may only modify the file flags on objects they
673 * have VADMIN rights for.
675 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
678 * Unprivileged processes are not permitted to unset system
679 * flags, or modify flags if any system flags are set.
680 * Privileged non-jail processes may not modify system flags
681 * if securelevel > 0 and any existing system flags are set.
682 * Privileged jail processes behave like privileged non-jail
683 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
684 * otherwise, they behave like unprivileged processes.
686 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
688 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
689 error = securelevel_gt(cred, 0);
693 /* The snapshot flag cannot be toggled. */
694 if ((vap->va_flags ^ ip->i_flags) & SF_SNAPSHOT)
698 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
699 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
702 ip->i_flags = vap->va_flags;
703 DIP_SET(ip, i_flags, vap->va_flags);
704 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
705 error = UFS_UPDATE(vp, 0);
706 if (ip->i_flags & (IMMUTABLE | APPEND))
710 * If immutable or append, no one can change any of its attributes
711 * except the ones already handled (in some cases, file flags
712 * including the immutability flags themselves for the superuser).
714 if (ip->i_flags & (IMMUTABLE | APPEND))
717 * Go through the fields and update iff not VNOVAL.
719 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
720 if (vp->v_mount->mnt_flag & MNT_RDONLY)
722 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred,
726 if (vap->va_size != VNOVAL) {
728 * XXX most of the following special cases should be in
729 * callers instead of in N filesystems. The VDIR check
732 switch (vp->v_type) {
738 * Truncation should have an effect in these cases.
739 * Disallow it if the filesystem is read-only or
740 * the file is being snapshotted.
742 if (vp->v_mount->mnt_flag & MNT_RDONLY)
749 * According to POSIX, the result is unspecified
750 * for file types other than regular files,
751 * directories and shared memory objects. We
752 * don't support shared memory objects in the file
753 * system, and have dubious support for truncating
754 * symlinks. Just ignore the request in other cases.
758 error = vn_rlimit_trunc(vap->va_size, td);
761 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL |
762 ((vap->va_vaflags & VA_SYNC) != 0 ? IO_SYNC : 0),
766 if (vap->va_atime.tv_sec != VNOVAL ||
767 vap->va_mtime.tv_sec != VNOVAL ||
768 vap->va_birthtime.tv_sec != VNOVAL) {
769 if (vp->v_mount->mnt_flag & MNT_RDONLY)
773 error = vn_utimes_perm(vp, vap, cred, td);
776 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED);
777 if (vap->va_atime.tv_sec != VNOVAL) {
778 ip->i_flag &= ~IN_ACCESS;
779 DIP_SET(ip, i_atime, vap->va_atime.tv_sec);
780 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec);
782 if (vap->va_mtime.tv_sec != VNOVAL) {
783 ip->i_flag &= ~IN_UPDATE;
784 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec);
785 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec);
787 if (vap->va_birthtime.tv_sec != VNOVAL && I_IS_UFS2(ip)) {
788 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec;
789 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec;
791 error = UFS_UPDATE(vp, 0);
796 if (vap->va_mode != (mode_t)VNOVAL) {
797 if (vp->v_mount->mnt_flag & MNT_RDONLY)
799 if (IS_SNAPSHOT(ip) && (vap->va_mode & (S_IXUSR | S_IWUSR |
800 S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)) != 0)
802 error = ufs_chmod(vp, (int)vap->va_mode, cred, td);
809 ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode,
810 int file_owner_id, struct ucred *cred, struct thread *td)
815 aclp = acl_alloc(M_WAITOK);
816 error = ufs_getacl_nfs4_internal(vp, aclp, td);
818 * We don't have to handle EOPNOTSUPP here, as the filesystem claims
824 acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id);
825 error = ufs_setacl_nfs4_internal(vp, aclp, td);
835 struct vop_mmapped_args /* {
847 if ((mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
848 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
850 * XXXKIB No UFS_UPDATE(ap->a_vp, 0) there.
856 * Change the mode on a file.
857 * Inode must be locked before calling.
860 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
862 struct inode *ip = VTOI(vp);
866 * To modify the permissions on a file, must possess VADMIN
869 if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td)))
872 * Privileged processes may set the sticky bit on non-directories,
873 * as well as set the setgid bit on a file with a group that the
874 * process is not a member of. Both of these are allowed in
877 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
878 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
881 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
882 error = priv_check_cred(cred, PRIV_VFS_SETGID);
888 * Deny setting setuid if we are not the file owner.
890 if ((mode & ISUID) && ip->i_uid != cred->cr_uid) {
891 error = priv_check_cred(cred, PRIV_VFS_ADMIN);
896 newmode = ip->i_mode & ~ALLPERMS;
897 newmode |= (mode & ALLPERMS);
898 UFS_INODE_SET_MODE(ip, newmode);
899 DIP_SET(ip, i_mode, ip->i_mode);
900 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
902 if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0)
903 error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td);
905 if (error == 0 && (ip->i_flag & IN_CHANGE) != 0)
906 error = UFS_UPDATE(vp, 0);
912 * Perform chown operation on inode ip;
913 * inode must be locked prior to call.
916 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
919 struct inode *ip = VTOI(vp);
928 if (uid == (uid_t)VNOVAL)
930 if (gid == (gid_t)VNOVAL)
933 * To modify the ownership of a file, must possess VADMIN for that
936 if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td)))
939 * To change the owner of a file, or change the group of a file to a
940 * group of which we are not a member, the caller must have
943 if (((uid != ip->i_uid && uid != cred->cr_uid) ||
944 (gid != ip->i_gid && !groupmember(gid, cred))) &&
945 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
950 if ((error = getinoquota(ip)) != 0)
953 dqrele(vp, ip->i_dquot[USRQUOTA]);
954 ip->i_dquot[USRQUOTA] = NODQUOT;
957 dqrele(vp, ip->i_dquot[GRPQUOTA]);
958 ip->i_dquot[GRPQUOTA] = NODQUOT;
960 change = DIP(ip, i_blocks);
961 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
962 (void) chkiq(ip, -1, cred, CHOWN|FORCE);
963 for (i = 0; i < MAXQUOTAS; i++) {
964 dqrele(vp, ip->i_dquot[i]);
965 ip->i_dquot[i] = NODQUOT;
969 DIP_SET(ip, i_gid, gid);
971 DIP_SET(ip, i_uid, uid);
973 if ((error = getinoquota(ip)) == 0) {
975 dqrele(vp, ip->i_dquot[USRQUOTA]);
976 ip->i_dquot[USRQUOTA] = NODQUOT;
979 dqrele(vp, ip->i_dquot[GRPQUOTA]);
980 ip->i_dquot[GRPQUOTA] = NODQUOT;
982 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
983 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
986 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
988 for (i = 0; i < MAXQUOTAS; i++) {
989 dqrele(vp, ip->i_dquot[i]);
990 ip->i_dquot[i] = NODQUOT;
994 DIP_SET(ip, i_gid, ogid);
996 DIP_SET(ip, i_uid, ouid);
997 if (getinoquota(ip) == 0) {
999 dqrele(vp, ip->i_dquot[USRQUOTA]);
1000 ip->i_dquot[USRQUOTA] = NODQUOT;
1003 dqrele(vp, ip->i_dquot[GRPQUOTA]);
1004 ip->i_dquot[GRPQUOTA] = NODQUOT;
1006 (void) chkdq(ip, change, cred, FORCE|CHOWN);
1007 (void) chkiq(ip, 1, cred, FORCE|CHOWN);
1008 (void) getinoquota(ip);
1012 if (getinoquota(ip))
1013 panic("ufs_chown: lost quota");
1015 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1016 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
1017 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1018 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1019 DIP_SET(ip, i_mode, ip->i_mode);
1022 error = UFS_UPDATE(vp, 0);
1028 struct vop_remove_args /* {
1029 struct vnode *a_dvp;
1031 struct componentname *a_cnp;
1035 struct vnode *vp = ap->a_vp;
1036 struct vnode *dvp = ap->a_dvp;
1042 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1043 (VTOI(dvp)->i_flags & APPEND))
1045 if (DOINGSUJ(dvp)) {
1046 error = softdep_prelink(dvp, vp, ap->a_cnp);
1048 MPASS(error == ERELOOKUP);
1054 ufs_gjournal_orphan(vp);
1056 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
1057 if (ip->i_nlink <= 0)
1058 vp->v_vflag |= VV_NOSYNC;
1059 if (IS_SNAPSHOT(ip)) {
1061 * Avoid deadlock where another thread is trying to
1062 * update the inodeblock for dvp and is waiting on
1063 * snaplk. Temporary unlock the vnode lock for the
1064 * unlinked file and sync the directory. This should
1065 * allow vput() of the directory to not block later on
1066 * while holding the snapshot vnode locked, assuming
1067 * that the directory hasn't been unlinked too.
1070 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
1071 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1077 print_bad_link_count(const char *funcname, struct vnode *dvp)
1082 uprintf("%s: Bad link count %d on parent inode %jd in file system %s\n",
1083 funcname, dip->i_effnlink, (intmax_t)dip->i_number,
1084 dvp->v_mount->mnt_stat.f_mntonname);
1092 struct vop_link_args /* {
1093 struct vnode *a_tdvp;
1095 struct componentname *a_cnp;
1098 struct vnode *vp = ap->a_vp;
1099 struct vnode *tdvp = ap->a_tdvp;
1100 struct componentname *cnp = ap->a_cnp;
1102 struct direct newdir;
1105 if (DOINGSUJ(tdvp)) {
1106 error = softdep_prelink(tdvp, vp, cnp);
1108 MPASS(error == ERELOOKUP);
1113 if (VTOI(tdvp)->i_effnlink < 2) {
1114 print_bad_link_count("ufs_link", tdvp);
1118 error = ufs_sync_nlink(vp, tdvp);
1124 * The file may have been removed after namei dropped the original
1127 if (ip->i_effnlink == 0) {
1131 if (ip->i_flags & (IMMUTABLE | APPEND)) {
1138 DIP_SET(ip, i_nlink, ip->i_nlink);
1139 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1140 if (DOINGSOFTDEP(vp))
1141 softdep_setup_link(VTOI(tdvp), ip);
1142 error = UFS_UPDATE(vp, !DOINGSOFTDEP(vp) && !DOINGASYNC(vp));
1144 ufs_makedirentry(ip, cnp, &newdir);
1145 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
1151 DIP_SET(ip, i_nlink, ip->i_nlink);
1152 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1153 if (DOINGSOFTDEP(vp))
1154 softdep_revert_link(VTOI(tdvp), ip);
1161 * whiteout vnode call
1165 struct vop_whiteout_args /* {
1166 struct vnode *a_dvp;
1167 struct componentname *a_cnp;
1171 struct vnode *dvp = ap->a_dvp;
1172 struct componentname *cnp = ap->a_cnp;
1173 struct direct newdir;
1176 if (DOINGSUJ(dvp) && (ap->a_flags == CREATE ||
1177 ap->a_flags == DELETE)) {
1178 error = softdep_prelink(dvp, NULL, cnp);
1180 MPASS(error == ERELOOKUP);
1185 switch (ap->a_flags) {
1187 /* 4.4 format directories support whiteout operations */
1190 return (EOPNOTSUPP);
1193 /* create a new directory whiteout */
1196 panic("ufs_whiteout: old format filesystem");
1199 newdir.d_ino = UFS_WINO;
1200 newdir.d_namlen = cnp->cn_namelen;
1201 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
1202 newdir.d_type = DT_WHT;
1203 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
1207 /* remove an existing directory whiteout */
1210 panic("ufs_whiteout: old format filesystem");
1213 cnp->cn_flags &= ~DOWHITEOUT;
1214 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
1217 panic("ufs_whiteout: unknown op");
1222 static volatile int rename_restarts;
1223 SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD,
1224 __DEVOLATILE(int *, &rename_restarts), 0,
1225 "Times rename had to restart due to lock contention");
1228 * Rename system call.
1229 * rename("foo", "bar");
1232 * link("foo", "bar");
1234 * but ``atomically''. Can't do full commit without saving state in the
1235 * inode on disk which isn't feasible at this time. Best we can do is
1236 * always guarantee the target exists.
1238 * Basic algorithm is:
1240 * 1) Bump link count on source while we're linking it to the
1241 * target. This also ensure the inode won't be deleted out
1242 * from underneath us while we work (it may be truncated by
1243 * a concurrent `trunc' or `open' for creation).
1244 * 2) Link source to destination. If destination already exists,
1246 * 3) Unlink source reference to inode if still around. If a
1247 * directory was moved and the parent of the destination
1248 * is different from the source, patch the ".." entry in the
1253 struct vop_rename_args /* {
1254 struct vnode *a_fdvp;
1255 struct vnode *a_fvp;
1256 struct componentname *a_fcnp;
1257 struct vnode *a_tdvp;
1258 struct vnode *a_tvp;
1259 struct componentname *a_tcnp;
1262 struct vnode *tvp = ap->a_tvp;
1263 struct vnode *tdvp = ap->a_tdvp;
1264 struct vnode *fvp = ap->a_fvp;
1265 struct vnode *fdvp = ap->a_fdvp;
1267 struct componentname *tcnp = ap->a_tcnp;
1268 struct componentname *fcnp = ap->a_fcnp;
1269 struct thread *td = curthread;
1270 struct inode *fip, *tip, *tdp, *fdp;
1271 struct direct newdir;
1273 int doingdirectory, newparent;
1277 seqc_t fdvp_s, fvp_s, tdvp_s, tvp_s;
1278 bool checkpath_locked, want_seqc_end;
1280 checkpath_locked = want_seqc_end = false;
1285 if (tvp && tvp != tdvp)
1288 * Check for cross-device rename.
1290 if ((fvp->v_mount != tdvp->v_mount) ||
1291 (tvp && (fvp->v_mount != tvp->v_mount))) {
1297 fdvp_s = fvp_s = tdvp_s = tvp_s = SEQC_MOD;
1300 * We need to acquire 2 to 4 locks depending on whether tvp is NULL
1301 * and fdvp and tdvp are the same directory. Subsequently we need
1302 * to double-check all paths and in the directory rename case we
1303 * need to verify that we are not creating a directory loop. To
1304 * handle this we acquire all but fdvp using non-blocking
1305 * acquisitions. If we fail to acquire any lock in the path we will
1306 * drop all held locks, acquire the new lock in a blocking fashion,
1307 * and then release it and restart the rename. This acquire/release
1308 * step ensures that we do not spin on a lock waiting for release.
1310 error = vn_lock(fdvp, LK_EXCLUSIVE);
1313 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1315 error = vn_lock(tdvp, LK_EXCLUSIVE);
1319 atomic_add_int(&rename_restarts, 1);
1323 * Re-resolve fvp to be certain it still exists and fetch the
1326 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1332 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1338 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1344 atomic_add_int(&rename_restarts, 1);
1350 * Re-resolve tvp and acquire the vnode lock if present.
1352 error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino);
1353 if (error != 0 && error != EJUSTRETURN) {
1360 * If tvp disappeared we just carry on.
1362 if (error == EJUSTRETURN && tvp != NULL) {
1367 * Get the tvp ino if the lookup succeeded. We may have to restart
1368 * if the non-blocking acquire fails.
1372 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1382 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1386 atomic_add_int(&rename_restarts, 1);
1391 if (DOINGSUJ(fdvp) &&
1392 (seqc_in_modify(fdvp_s) || !vn_seqc_consistent(fdvp, fdvp_s) ||
1393 seqc_in_modify(fvp_s) || !vn_seqc_consistent(fvp, fvp_s) ||
1394 seqc_in_modify(tdvp_s) || !vn_seqc_consistent(tdvp, tdvp_s) ||
1395 (tvp != NULL && (seqc_in_modify(tvp_s) ||
1396 !vn_seqc_consistent(tvp, tvp_s))))) {
1397 error = softdep_prerename(fdvp, fvp, tdvp, tvp);
1408 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1409 (VTOI(tdvp)->i_flags & APPEND))) {
1414 * Renaming a file to itself has no effect. The upper layers should
1415 * not call us in that case. However, things could change after
1416 * we drop the locks above.
1424 ino = fip->i_number;
1425 if (fip->i_nlink >= UFS_LINK_MAX) {
1426 if (!DOINGSOFTDEP(fvp) || fip->i_effnlink >= UFS_LINK_MAX) {
1431 MPASS(!want_seqc_end);
1432 if (checkpath_locked) {
1433 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1434 checkpath_locked = false;
1441 VOP_VPUT_PAIR(tdvp, &tvp, true);
1442 error = ufs_sync_nlink1(mp);
1450 if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
1451 || (fdp->i_flags & APPEND)) {
1455 if ((fip->i_mode & IFMT) == IFDIR) {
1457 * Avoid ".", "..", and aliases of "." for obvious reasons.
1459 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1461 (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
1465 if (fdp->i_number != tdp->i_number)
1466 newparent = tdp->i_number;
1469 if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) ||
1470 (tvp != NULL && tvp->v_type == VDIR &&
1471 tvp->v_mountedhere != NULL)) {
1477 * If ".." must be changed (ie the directory gets a new
1478 * parent) then the source directory must not be in the
1479 * directory hierarchy above the target, as this would
1480 * orphan everything below the source directory. Also
1481 * the user must have write permission in the source so
1482 * as to be able to change "..".
1484 if (doingdirectory && newparent) {
1485 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
1489 sx_xlock(&VFSTOUFS(mp)->um_checkpath_lock);
1490 checkpath_locked = true;
1491 error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred,
1494 * We encountered a lock that we have to wait for. Unlock
1495 * everything else and VGET before restarting.
1498 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1499 checkpath_locked = false;
1505 error = VFS_VGET(mp, ino, LK_SHARED, &nvp);
1508 atomic_add_int(&rename_restarts, 1);
1514 if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 ||
1515 tdp->i_effnlink == 0)
1516 panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp);
1519 vn_seqc_write_begin(tvp);
1520 vn_seqc_write_begin(tdvp);
1521 vn_seqc_write_begin(fvp);
1522 vn_seqc_write_begin(fdvp);
1523 want_seqc_end = true;
1526 * 1) Bump link count while we're moving stuff
1527 * around. If we crash somewhere before
1528 * completing our work, the link count
1529 * may be wrong, but correctable.
1533 DIP_SET(fip, i_nlink, fip->i_nlink);
1534 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1535 if (DOINGSOFTDEP(fvp))
1536 softdep_setup_link(tdp, fip);
1537 error = UFS_UPDATE(fvp, !DOINGSOFTDEP(fvp) && !DOINGASYNC(fvp));
1542 * 2) If target doesn't exist, link the target
1543 * to the source and unlink the source.
1544 * Otherwise, rewrite the target directory
1545 * entry to reference the source inode and
1546 * expunge the original entry's existence.
1549 if (ITODEV(tdp) != ITODEV(fip))
1550 panic("ufs_rename: EXDEV");
1551 if (doingdirectory && newparent) {
1553 * Account for ".." in new directory.
1554 * When source and destination have the same
1555 * parent we don't adjust the link count. The
1556 * actual link modification is completed when
1557 * .. is rewritten below.
1559 if (tdp->i_nlink >= UFS_LINK_MAX) {
1562 DIP_SET(fip, i_nlink, fip->i_nlink);
1563 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1564 if (DOINGSOFTDEP(fvp))
1565 softdep_revert_link(tdp, fip);
1566 if (!DOINGSOFTDEP(tdvp) ||
1567 tdp->i_effnlink >= UFS_LINK_MAX) {
1571 MPASS(want_seqc_end);
1573 vn_seqc_write_end(tvp);
1574 vn_seqc_write_end(tdvp);
1575 vn_seqc_write_end(fvp);
1576 vn_seqc_write_end(fdvp);
1577 want_seqc_end = false;
1579 MPASS(checkpath_locked);
1580 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1581 checkpath_locked = false;
1587 VOP_VPUT_PAIR(tdvp, &tvp, true);
1588 error = ufs_sync_nlink1(mp);
1597 ufs_makedirentry(fip, tcnp, &newdir);
1598 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1601 /* Setup tdvp for directory compaction if needed. */
1602 if (I_COUNT(tdp) != 0 && I_ENDOFF(tdp) != 0 &&
1603 I_ENDOFF(tdp) < tdp->i_size)
1604 endoff = I_ENDOFF(tdp);
1606 if (ITODEV(tip) != ITODEV(tdp) || ITODEV(tip) != ITODEV(fip))
1607 panic("ufs_rename: EXDEV");
1609 * Short circuit rename(foo, foo).
1611 if (tip->i_number == fip->i_number)
1612 panic("ufs_rename: same file");
1614 * If the parent directory is "sticky", then the caller
1615 * must possess VADMIN for the parent directory, or the
1616 * destination of the rename. This implements append-only
1619 if ((tdp->i_mode & S_ISTXT) &&
1620 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) &&
1621 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) {
1626 * Target must be empty if a directory and have no links
1627 * to it. Also, ensure source and target are compatible
1628 * (both directories, or both not directories).
1630 if ((tip->i_mode & IFMT) == IFDIR) {
1631 if ((tip->i_effnlink > 2) ||
1632 !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) {
1636 if (!doingdirectory) {
1641 } else if (doingdirectory) {
1645 if (doingdirectory) {
1648 if (DOINGSOFTDEP(tdvp))
1649 softdep_change_linkcnt(tdp);
1652 if (DOINGSOFTDEP(tvp))
1653 softdep_change_linkcnt(tip);
1655 error = ufs_dirrewrite(tdp, tip, fip->i_number,
1656 IFTODT(fip->i_mode),
1657 (doingdirectory && newparent) ? newparent : doingdirectory);
1659 if (doingdirectory) {
1662 if (DOINGSOFTDEP(tdvp))
1663 softdep_change_linkcnt(tdp);
1666 if (DOINGSOFTDEP(tvp))
1667 softdep_change_linkcnt(tip);
1671 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1673 * The only stuff left in the directory is "."
1674 * and "..". The "." reference is inconsequential
1675 * since we are quashing it. We have removed the "."
1676 * reference and the reference in the parent directory,
1677 * but there may be other hard links. The soft
1678 * dependency code will arrange to do these operations
1679 * after the parent directory entry has been deleted on
1680 * disk, so when running with that code we avoid doing
1685 DIP_SET(tdp, i_nlink, tdp->i_nlink);
1686 UFS_INODE_SET_FLAG(tdp, IN_CHANGE);
1689 DIP_SET(tip, i_nlink, tip->i_nlink);
1690 UFS_INODE_SET_FLAG(tip, IN_CHANGE);
1695 * 3) Unlink the source. We have to resolve the path again to
1696 * fixup the directory offset and count for ufs_dirremove.
1699 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1701 panic("ufs_rename: from entry went away!");
1702 if (ino != fip->i_number)
1703 panic("ufs_rename: ino mismatch %ju != %ju\n",
1704 (uintmax_t)ino, (uintmax_t)fip->i_number);
1707 * If the source is a directory with a
1708 * new parent, the link count of the old
1709 * parent directory must be decremented
1710 * and ".." set to point to the new parent.
1712 if (doingdirectory && newparent) {
1714 * If tip exists we simply use its link, otherwise we must
1720 DIP_SET(tdp, i_nlink, tdp->i_nlink);
1721 UFS_INODE_SET_FLAG(tdp, IN_CHANGE);
1722 if (DOINGSOFTDEP(tdvp))
1723 softdep_setup_dotdot_link(tdp, fip);
1724 error = UFS_UPDATE(tdvp, !DOINGSOFTDEP(tdvp) &&
1726 /* Don't go to bad here as the new link exists. */
1729 } else if (DOINGSUJ(tdvp))
1730 /* Journal must account for each new link. */
1731 softdep_setup_dotdot_link(tdp, fip);
1732 SET_I_OFFSET(fip, mastertemplate.dot_reclen);
1733 ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0);
1736 error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0);
1738 * The kern_renameat() looks up the fvp using the DELETE flag, which
1739 * causes the removal of the name cache entry for fvp.
1740 * As the relookup of the fvp is done in two steps:
1741 * ufs_lookup_ino() and then VFS_VGET(), another thread might do a
1742 * normal lookup of the from name just before the VFS_VGET() call,
1743 * causing the cache entry to be re-instantiated.
1745 * The same issue also applies to tvp if it exists as
1746 * otherwise we may have a stale name cache entry for the new
1747 * name that references the old i-node if it has other links
1748 * or open file descriptors.
1750 cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp);
1753 if (want_seqc_end) {
1755 vn_seqc_write_end(tvp);
1756 vn_seqc_write_end(tdvp);
1757 vn_seqc_write_end(fvp);
1758 vn_seqc_write_end(fdvp);
1761 if (checkpath_locked)
1762 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1768 * If compaction or fsync was requested do it in
1769 * ffs_vput_pair() now that other locks are no longer needed.
1771 if (error == 0 && endoff != 0) {
1772 UFS_INODE_SET_FLAG(tdp, IN_ENDOFF);
1773 SET_I_ENDOFF(tdp, endoff);
1775 VOP_VPUT_PAIR(tdvp, &tvp, true);
1781 DIP_SET(fip, i_nlink, fip->i_nlink);
1782 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1783 if (DOINGSOFTDEP(fvp))
1784 softdep_revert_link(tdp, fip);
1788 if (want_seqc_end) {
1790 vn_seqc_write_end(tvp);
1791 vn_seqc_write_end(tdvp);
1792 vn_seqc_write_end(fvp);
1793 vn_seqc_write_end(fdvp);
1807 ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1808 mode_t dmode, struct ucred *cred, struct thread *td)
1811 struct inode *ip = VTOI(tvp);
1812 struct acl *dacl, *acl;
1814 acl = acl_alloc(M_WAITOK);
1815 dacl = acl_alloc(M_WAITOK);
1818 * Retrieve default ACL from parent, if any.
1820 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1824 * Retrieved a default ACL, so merge mode and ACL if
1825 * necessary. If the ACL is empty, fall through to
1826 * the "not defined or available" case.
1828 if (acl->acl_cnt != 0) {
1829 dmode = acl_posix1e_newfilemode(dmode, acl);
1830 UFS_INODE_SET_MODE(ip, dmode);
1831 DIP_SET(ip, i_mode, dmode);
1833 ufs_sync_acl_from_inode(ip, acl);
1840 * Just use the mode as-is.
1842 UFS_INODE_SET_MODE(ip, dmode);
1843 DIP_SET(ip, i_mode, dmode);
1852 * XXX: If we abort now, will Soft Updates notify the extattr
1853 * code that the EAs for the file need to be released?
1855 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1857 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1864 * XXX: This should not happen, as EOPNOTSUPP above
1865 * was supposed to free acl.
1867 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1869 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()");
1885 ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1886 mode_t mode, struct ucred *cred, struct thread *td)
1889 struct inode *ip = VTOI(tvp);
1892 acl = acl_alloc(M_WAITOK);
1895 * Retrieve default ACL for parent, if any.
1897 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1901 * Retrieved a default ACL, so merge mode and ACL if
1904 if (acl->acl_cnt != 0) {
1906 * Two possible ways for default ACL to not
1907 * be present. First, the EA can be
1908 * undefined, or second, the default ACL can
1909 * be blank. If it's blank, fall through to
1910 * the it's not defined case.
1912 mode = acl_posix1e_newfilemode(mode, acl);
1913 UFS_INODE_SET_MODE(ip, mode);
1914 DIP_SET(ip, i_mode, mode);
1915 ufs_sync_acl_from_inode(ip, acl);
1922 * Just use the mode as-is.
1924 UFS_INODE_SET_MODE(ip, mode);
1925 DIP_SET(ip, i_mode, mode);
1934 * XXX: If we abort now, will Soft Updates notify the extattr
1935 * code that the EAs for the file need to be released?
1937 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1944 * XXX: This should not happen, as EOPNOTSUPP above was
1945 * supposed to free acl.
1947 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1948 "but no VOP_SETACL()\n");
1949 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1950 "but no VOP_SETACL()"); */
1964 ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp,
1965 mode_t child_mode, struct ucred *cred, struct thread *td)
1968 struct acl *parent_aclp, *child_aclp;
1970 parent_aclp = acl_alloc(M_WAITOK);
1971 child_aclp = acl_alloc(M_WAITOK | M_ZERO);
1973 error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td);
1976 acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp,
1977 child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR);
1978 error = ufs_setacl_nfs4_internal(tvp, child_aclp, td);
1982 acl_free(parent_aclp);
1983 acl_free(child_aclp);
1994 struct vop_mkdir_args /* {
1995 struct vnode *a_dvp;
1996 struct vnode **a_vpp;
1997 struct componentname *a_cnp;
1998 struct vattr *a_vap;
2001 struct vnode *dvp = ap->a_dvp;
2002 struct vattr *vap = ap->a_vap;
2003 struct componentname *cnp = ap->a_cnp;
2004 struct inode *ip, *dp;
2007 struct dirtemplate dirtemplate, *dtp;
2008 struct direct newdir;
2013 error = ufs_sync_nlink(dvp, NULL);
2016 dmode = vap->va_mode & 0777;
2020 * Must simulate part of ufs_makeinode here to acquire the inode,
2021 * but not have it entered in the parent directory. The entry is
2022 * made later after writing "." and ".." entries.
2024 if (dp->i_effnlink < 2) {
2025 print_bad_link_count("ufs_mkdir", dvp);
2030 if (DOINGSUJ(dvp)) {
2031 error = softdep_prelink(dvp, NULL, cnp);
2033 MPASS(error == ERELOOKUP);
2038 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp);
2041 vn_seqc_write_begin(tvp);
2043 ip->i_gid = dp->i_gid;
2044 DIP_SET(ip, i_gid, dp->i_gid);
2048 struct ucred ucred, *ucp;
2053 * If we are hacking owners here, (only do this where told to)
2054 * and we are not giving it TO root, (would subvert quotas)
2055 * then go ahead and give it to the other user.
2056 * The new directory also inherits the SUID bit.
2057 * If user's UID and dir UID are the same,
2058 * 'give it away' so that the SUID is still forced on.
2060 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2061 (dp->i_mode & ISUID) && dp->i_uid) {
2063 ip->i_uid = dp->i_uid;
2064 DIP_SET(ip, i_uid, dp->i_uid);
2066 if (dp->i_uid != cnp->cn_cred->cr_uid) {
2068 * Make sure the correct user gets charged
2070 * Make a dummy credential for the victim.
2071 * XXX This seems to never be accessed out of
2072 * our context so a stack variable is ok.
2074 refcount_init(&ucred.cr_ref, 1);
2075 ucred.cr_uid = ip->i_uid;
2076 ucred.cr_ngroups = 1;
2077 ucred.cr_groups = &ucred_group;
2078 ucred.cr_groups[0] = dp->i_gid;
2083 ip->i_uid = cnp->cn_cred->cr_uid;
2084 DIP_SET(ip, i_uid, ip->i_uid);
2087 if ((error = getinoquota(ip)) ||
2088 (error = chkiq(ip, 1, ucp, 0))) {
2089 if (DOINGSOFTDEP(tvp))
2090 softdep_revert_link(dp, ip);
2091 UFS_VFREE(tvp, ip->i_number, dmode);
2092 vn_seqc_write_end(tvp);
2099 #else /* !SUIDDIR */
2100 ip->i_uid = cnp->cn_cred->cr_uid;
2101 DIP_SET(ip, i_uid, ip->i_uid);
2103 if ((error = getinoquota(ip)) ||
2104 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2105 if (DOINGSOFTDEP(tvp))
2106 softdep_revert_link(dp, ip);
2107 UFS_VFREE(tvp, ip->i_number, dmode);
2108 vn_seqc_write_end(tvp);
2114 #endif /* !SUIDDIR */
2115 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
2116 UFS_INODE_SET_MODE(ip, dmode);
2117 DIP_SET(ip, i_mode, dmode);
2118 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
2121 DIP_SET(ip, i_nlink, 2);
2123 if (cnp->cn_flags & ISWHITEOUT) {
2124 ip->i_flags |= UF_OPAQUE;
2125 DIP_SET(ip, i_flags, ip->i_flags);
2129 * Bump link count in parent directory to reflect work done below.
2130 * Should be done before reference is created so cleanup is
2131 * possible if we crash.
2135 DIP_SET(dp, i_nlink, dp->i_nlink);
2136 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2137 if (DOINGSOFTDEP(dvp))
2138 softdep_setup_mkdir(dp, ip);
2139 error = UFS_UPDATE(dvp, !DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp));
2143 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2144 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2151 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2152 error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
2153 cnp->cn_cred, curthread);
2156 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2157 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode,
2158 cnp->cn_cred, curthread);
2162 #endif /* !UFS_ACL */
2165 * Initialize directory with "." and ".." from static template.
2168 dtp = &mastertemplate;
2170 dtp = (struct dirtemplate *)&omastertemplate;
2172 dirtemplate.dot_ino = ip->i_number;
2173 dirtemplate.dotdot_ino = dp->i_number;
2174 vnode_pager_setsize(tvp, DIRBLKSIZ);
2175 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
2176 BA_CLRBUF, &bp)) != 0)
2178 ip->i_size = DIRBLKSIZ;
2179 DIP_SET(ip, i_size, DIRBLKSIZ);
2180 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
2181 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
2182 if (DOINGSOFTDEP(tvp)) {
2184 * Ensure that the entire newly allocated block is a
2185 * valid directory so that future growth within the
2186 * block does not have to ensure that the block is
2187 * written before the inode.
2190 while (blkoff < bp->b_bcount) {
2192 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
2193 blkoff += DIRBLKSIZ;
2196 if ((error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) &&
2197 !DOINGASYNC(tvp))) != 0) {
2202 * Directory set up, now install its entry in the parent directory.
2204 * If we are not doing soft dependencies, then we must write out the
2205 * buffer containing the new directory body before entering the new
2206 * name in the parent. If we are doing soft dependencies, then the
2207 * buffer containing the new directory body will be passed to and
2208 * released in the soft dependency code after the code has attached
2209 * an appropriate ordering dependency to the buffer which ensures that
2210 * the buffer is written before the new name is written in the parent.
2212 if (DOINGASYNC(dvp))
2214 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp))))
2216 ufs_makedirentry(ip, cnp, &newdir);
2217 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
2222 vn_seqc_write_end(tvp);
2226 DIP_SET(dp, i_nlink, dp->i_nlink);
2227 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2229 * No need to do an explicit VOP_TRUNCATE here, vrele will
2230 * do this for us because we set the link count to 0.
2234 DIP_SET(ip, i_nlink, 0);
2235 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2236 if (DOINGSOFTDEP(tvp))
2237 softdep_revert_mkdir(dp, ip);
2238 vn_seqc_write_end(tvp);
2247 * Rmdir system call.
2251 struct vop_rmdir_args /* {
2252 struct vnode *a_dvp;
2254 struct componentname *a_cnp;
2257 struct vnode *vp = ap->a_vp;
2258 struct vnode *dvp = ap->a_dvp;
2259 struct componentname *cnp = ap->a_cnp;
2260 struct inode *ip, *dp;
2267 * Do not remove a directory that is in the process of being renamed.
2268 * Verify the directory is empty (and valid). Rmdir ".." will not be
2269 * valid since ".." will contain a reference to the current directory
2270 * and thus be non-empty. Do not allow the removal of mounted on
2271 * directories (this can happen when an NFS exported filesystem
2272 * tries to remove a locally mounted on directory).
2275 if (dp->i_effnlink <= 2) {
2276 if (dp->i_effnlink == 2)
2277 print_bad_link_count("ufs_rmdir", dvp);
2281 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
2285 if ((dp->i_flags & APPEND)
2286 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
2290 if (vp->v_mountedhere != 0) {
2294 if (DOINGSUJ(dvp)) {
2295 error = softdep_prelink(dvp, vp, cnp);
2297 MPASS(error == ERELOOKUP);
2303 ufs_gjournal_orphan(vp);
2306 * Delete reference to directory before purging
2307 * inode. If we crash in between, the directory
2308 * will be reattached to lost+found,
2312 if (DOINGSOFTDEP(vp))
2313 softdep_setup_rmdir(dp, ip);
2314 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
2318 if (DOINGSOFTDEP(vp))
2319 softdep_revert_rmdir(dp, ip);
2323 * The only stuff left in the directory is "." and "..". The "."
2324 * reference is inconsequential since we are quashing it. The soft
2325 * dependency code will arrange to do these operations after
2326 * the parent directory entry has been deleted on disk, so
2327 * when running with that code we avoid doing them now.
2329 if (!DOINGSOFTDEP(vp)) {
2331 DIP_SET(dp, i_nlink, dp->i_nlink);
2332 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2333 error = UFS_UPDATE(dvp, 0);
2335 DIP_SET(ip, i_nlink, ip->i_nlink);
2336 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2338 cache_vop_rmdir(dvp, vp);
2340 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
2341 if (ip->i_dirhash != NULL)
2342 ufsdirhash_free(ip);
2349 * symlink -- make a symbolic link
2353 struct vop_symlink_args /* {
2354 struct vnode *a_dvp;
2355 struct vnode **a_vpp;
2356 struct componentname *a_cnp;
2357 struct vattr *a_vap;
2358 const char *a_target;
2361 struct vnode *vp, **vpp = ap->a_vpp;
2365 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
2366 vpp, ap->a_cnp, "ufs_symlink");
2370 len = strlen(ap->a_target);
2371 if (len < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) {
2373 bcopy(ap->a_target, DIP(ip, i_shortlink), len);
2375 DIP_SET(ip, i_size, len);
2376 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
2377 error = UFS_UPDATE(vp, 0);
2379 error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target),
2380 len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
2381 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
2388 * Vnode op for reading directories.
2392 struct vop_readdir_args /* {
2395 struct ucred *a_cred;
2398 uint64_t **a_cookies;
2401 struct vnode *vp = ap->a_vp;
2402 struct uio *uio = ap->a_uio;
2405 struct direct *dp, *edp;
2407 struct dirent dstdp;
2408 off_t offset, startoffset;
2409 size_t readcnt, skipcnt;
2414 if (uio->uio_offset < 0)
2417 if (ip->i_effnlink == 0)
2419 if (ap->a_ncookies != NULL) {
2420 if (uio->uio_resid < 0)
2423 ncookies = uio->uio_resid;
2424 if (uio->uio_offset >= ip->i_size)
2426 else if (ip->i_size - uio->uio_offset < ncookies)
2427 ncookies = ip->i_size - uio->uio_offset;
2428 ncookies = ncookies / (offsetof(struct direct, d_name) + 4) + 1;
2429 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
2430 *ap->a_ncookies = ncookies;
2431 *ap->a_cookies = cookies;
2436 offset = startoffset = uio->uio_offset;
2437 startresid = uio->uio_resid;
2439 while (error == 0 && uio->uio_resid > 0 &&
2440 uio->uio_offset < ip->i_size) {
2441 error = UFS_BLKATOFF(vp, uio->uio_offset, NULL, &bp);
2444 if (bp->b_offset + bp->b_bcount > ip->i_size)
2445 readcnt = ip->i_size - bp->b_offset;
2447 readcnt = bp->b_bcount;
2448 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
2449 ~(size_t)(DIRBLKSIZ - 1);
2450 offset = bp->b_offset + skipcnt;
2451 dp = (struct direct *)&bp->b_data[skipcnt];
2452 edp = (struct direct *)&bp->b_data[readcnt];
2453 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
2454 if (dp->d_reclen <= offsetof(struct direct, d_name) ||
2455 (caddr_t)dp + dp->d_reclen > (caddr_t)edp) {
2459 #if BYTE_ORDER == LITTLE_ENDIAN
2460 /* Old filesystem format. */
2462 dstdp.d_namlen = dp->d_type;
2463 dstdp.d_type = dp->d_namlen;
2467 dstdp.d_namlen = dp->d_namlen;
2468 dstdp.d_type = dp->d_type;
2470 if (offsetof(struct direct, d_name) + dstdp.d_namlen >
2475 if (offset < startoffset || dp->d_ino == 0)
2477 dstdp.d_fileno = dp->d_ino;
2478 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
2479 bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen);
2480 /* NOTE: d_off is the offset of the *next* entry. */
2481 dstdp.d_off = offset + dp->d_reclen;
2482 dirent_terminate(&dstdp);
2483 if (dstdp.d_reclen > uio->uio_resid) {
2484 if (uio->uio_resid == startresid)
2487 error = EJUSTRETURN;
2491 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
2494 if (cookies != NULL) {
2495 KASSERT(ncookies > 0,
2496 ("ufs_readdir: cookies buffer too small"));
2497 *cookies = offset + dp->d_reclen;
2502 offset += dp->d_reclen;
2503 dp = (struct direct *)((caddr_t)dp + dp->d_reclen);
2506 uio->uio_offset = offset;
2508 /* We need to correct uio_offset. */
2509 uio->uio_offset = offset;
2510 if (error == EJUSTRETURN)
2512 if (ap->a_ncookies != NULL) {
2514 *ap->a_ncookies -= ncookies;
2516 free(*ap->a_cookies, M_TEMP);
2517 *ap->a_ncookies = 0;
2518 *ap->a_cookies = NULL;
2521 if (error == 0 && ap->a_eofflag)
2522 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
2527 * Return target name of a symbolic link
2531 struct vop_readlink_args /* {
2534 struct ucred *a_cred;
2537 struct vnode *vp = ap->a_vp;
2538 struct inode *ip = VTOI(vp);
2542 if (isize < VFSTOUFS(vp->v_mount)->um_maxsymlinklen)
2543 return (uiomove(DIP(ip, i_shortlink), isize, ap->a_uio));
2544 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
2548 * Calculate the logical to physical mapping if not done already,
2549 * then call the device strategy routine.
2551 * In order to be able to swap to a file, the ufs_bmaparray() operation may not
2552 * deadlock on memory. See ufs_bmap() for details.
2556 struct vop_strategy_args /* {
2561 struct buf *bp = ap->a_bp;
2562 struct vnode *vp = ap->a_vp;
2566 if (bp->b_blkno == bp->b_lblkno) {
2567 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL);
2568 bp->b_blkno = blkno;
2570 bp->b_error = error;
2571 bp->b_ioflags |= BIO_ERROR;
2575 if ((long)bp->b_blkno == -1)
2578 if ((long)bp->b_blkno == -1) {
2582 bp->b_iooffset = dbtob(bp->b_blkno);
2583 BO_STRATEGY(VFSTOUFS(vp->v_mount)->um_bo, bp);
2588 * Print out the contents of an inode.
2592 struct vop_print_args /* {
2596 struct vnode *vp = ap->a_vp;
2597 struct inode *ip = VTOI(vp);
2599 printf("\tnlink=%d, effnlink=%d, size=%jd", ip->i_nlink,
2600 ip->i_effnlink, (intmax_t)ip->i_size);
2602 printf(", extsize %d", ip->i_din2->di_extsize);
2603 printf("\n\tgeneration=%jx, uid=%d, gid=%d, flags=0x%b\n",
2604 (uintmax_t)ip->i_gen, ip->i_uid, ip->i_gid,
2605 (u_int)ip->i_flags, PRINT_INODE_FLAGS);
2606 printf("\tino %lu, on dev %s", (u_long)ip->i_number,
2607 devtoname(ITODEV(ip)));
2608 if (vp->v_type == VFIFO)
2615 * Close wrapper for fifos.
2617 * Update the times on the inode then do device close.
2621 struct vop_close_args /* {
2624 struct ucred *a_cred;
2625 struct thread *a_td;
2630 return (fifo_specops.vop_close(ap));
2634 * Return POSIX pathconf information applicable to ufs filesystems.
2638 struct vop_pathconf_args /* {
2647 switch (ap->a_name) {
2649 *ap->a_retval = UFS_LINK_MAX;
2652 *ap->a_retval = UFS_MAXNAMLEN;
2655 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
2656 *ap->a_retval = PIPE_BUF;
2660 case _PC_CHOWN_RESTRICTED:
2667 case _PC_ACL_EXTENDED:
2668 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2674 if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS)
2680 case _PC_ACL_PATH_MAX:
2682 if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS))
2683 *ap->a_retval = ACL_MAX_ENTRIES;
2691 case _PC_MAC_PRESENT:
2692 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL)
2698 case _PC_MIN_HOLE_SIZE:
2699 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2707 case _PC_ALLOC_SIZE_MIN:
2708 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
2710 case _PC_FILESIZEBITS:
2713 case _PC_REC_INCR_XFER_SIZE:
2714 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2716 case _PC_REC_MAX_XFER_SIZE:
2717 *ap->a_retval = -1; /* means ``unlimited'' */
2719 case _PC_REC_MIN_XFER_SIZE:
2720 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2722 case _PC_REC_XFER_ALIGN:
2723 *ap->a_retval = PAGE_SIZE;
2725 case _PC_SYMLINK_MAX:
2726 *ap->a_retval = MAXPATHLEN;
2730 error = vop_stdpathconf(ap);
2737 * Initialize the vnode associated with a new inode, handle aliased
2741 ufs_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
2747 ASSERT_VOP_LOCKED(vp, "ufs_vinit");
2749 vp->v_type = IFTOVT(ip->i_mode);
2751 * Only unallocated inodes should be of type VNON.
2753 if (ip->i_mode != 0 && vp->v_type == VNON)
2755 if (vp->v_type == VFIFO)
2757 if (ip->i_number == UFS_ROOTINO)
2758 vp->v_vflag |= VV_ROOT;
2764 * Allocate a new inode.
2765 * Vnode dvp must be locked.
2768 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2769 struct componentname *cnp, const char *callfunc)
2771 struct inode *ip, *pdir;
2772 struct direct newdir;
2778 if ((mode & IFMT) == 0)
2781 if (pdir->i_effnlink < 2) {
2782 print_bad_link_count(callfunc, dvp);
2785 if (DOINGSUJ(dvp)) {
2786 error = softdep_prelink(dvp, NULL, cnp);
2788 MPASS(error == ERELOOKUP);
2792 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp);
2796 ip->i_gid = pdir->i_gid;
2797 DIP_SET(ip, i_gid, pdir->i_gid);
2801 struct ucred ucred, *ucp;
2806 * If we are not the owner of the directory,
2807 * and we are hacking owners here, (only do this where told to)
2808 * and we are not giving it TO root, (would subvert quotas)
2809 * then go ahead and give it to the other user.
2810 * Note that this drops off the execute bits for security.
2812 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2813 (pdir->i_mode & ISUID) &&
2814 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2815 ip->i_uid = pdir->i_uid;
2816 DIP_SET(ip, i_uid, ip->i_uid);
2820 * Make sure the correct user gets charged
2822 * Quickly knock up a dummy credential for the victim.
2823 * XXX This seems to never be accessed out of our
2824 * context so a stack variable is ok.
2826 refcount_init(&ucred.cr_ref, 1);
2827 ucred.cr_uid = ip->i_uid;
2828 ucred.cr_ngroups = 1;
2829 ucred.cr_groups = &ucred_group;
2830 ucred.cr_groups[0] = pdir->i_gid;
2834 ip->i_uid = cnp->cn_cred->cr_uid;
2835 DIP_SET(ip, i_uid, ip->i_uid);
2839 if ((error = getinoquota(ip)) ||
2840 (error = chkiq(ip, 1, ucp, 0))) {
2841 if (DOINGSOFTDEP(tvp))
2842 softdep_revert_link(pdir, ip);
2843 UFS_VFREE(tvp, ip->i_number, mode);
2850 #else /* !SUIDDIR */
2851 ip->i_uid = cnp->cn_cred->cr_uid;
2852 DIP_SET(ip, i_uid, ip->i_uid);
2854 if ((error = getinoquota(ip)) ||
2855 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2856 if (DOINGSOFTDEP(tvp))
2857 softdep_revert_link(pdir, ip);
2858 UFS_VFREE(tvp, ip->i_number, mode);
2864 #endif /* !SUIDDIR */
2865 vn_seqc_write_begin(tvp); /* Mostly to cover asserts */
2866 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
2867 UFS_INODE_SET_MODE(ip, mode);
2868 DIP_SET(ip, i_mode, mode);
2869 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2872 DIP_SET(ip, i_nlink, 1);
2873 if (DOINGSOFTDEP(tvp))
2874 softdep_setup_create(VTOI(dvp), ip);
2875 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2876 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID)) {
2877 UFS_INODE_SET_MODE(ip, ip->i_mode & ~ISGID);
2878 DIP_SET(ip, i_mode, ip->i_mode);
2881 if (cnp->cn_flags & ISWHITEOUT) {
2882 ip->i_flags |= UF_OPAQUE;
2883 DIP_SET(ip, i_flags, ip->i_flags);
2887 * Make sure inode goes to disk before directory entry.
2889 error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp));
2893 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2894 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2901 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2902 error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2903 cnp->cn_cred, curthread);
2906 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2907 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode,
2908 cnp->cn_cred, curthread);
2912 #endif /* !UFS_ACL */
2913 ufs_makedirentry(ip, cnp, &newdir);
2914 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2917 vn_seqc_write_end(tvp);
2923 * Write error occurred trying to update the inode
2924 * or the directory so must deallocate the inode.
2928 DIP_SET(ip, i_nlink, 0);
2929 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2930 if (DOINGSOFTDEP(tvp))
2931 softdep_revert_create(VTOI(dvp), ip);
2932 vn_seqc_write_end(tvp);
2939 ufs_ioctl(struct vop_ioctl_args *ap)
2945 switch (ap->a_command) {
2947 error = vn_lock(vp, LK_EXCLUSIVE);
2949 error = ufs_bmap_seekdata(vp, (off_t *)ap->a_data);
2955 return (vn_bmap_seekhole(vp, ap->a_command, (off_t *)ap->a_data,
2963 ufs_read_pgcache(struct vop_read_pgcache_args *ap)
2970 VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp);
2972 if (uio->uio_resid > ptoa(io_hold_cnt) || uio->uio_offset < 0 ||
2973 (ap->a_ioflag & IO_DIRECT) != 0)
2974 return (EJUSTRETURN);
2975 return (vn_read_from_obj(vp, uio));
2978 /* Global vfs data structures for ufs. */
2979 struct vop_vector ufs_vnodeops = {
2980 .vop_default = &default_vnodeops,
2981 .vop_fsync = VOP_PANIC,
2982 .vop_read = VOP_PANIC,
2983 .vop_reallocblks = VOP_PANIC,
2984 .vop_write = VOP_PANIC,
2985 .vop_accessx = ufs_accessx,
2986 .vop_bmap = ufs_bmap,
2987 .vop_fplookup_vexec = ufs_fplookup_vexec,
2988 .vop_fplookup_symlink = VOP_EAGAIN,
2989 .vop_cachedlookup = ufs_lookup,
2990 .vop_close = ufs_close,
2991 .vop_create = ufs_create,
2992 .vop_stat = ufs_stat,
2993 .vop_getattr = ufs_getattr,
2994 .vop_inactive = ufs_inactive,
2995 .vop_ioctl = ufs_ioctl,
2996 .vop_link = ufs_link,
2997 .vop_lookup = vfs_cache_lookup,
2998 .vop_mmapped = ufs_mmapped,
2999 .vop_mkdir = ufs_mkdir,
3000 .vop_mknod = ufs_mknod,
3001 .vop_need_inactive = ufs_need_inactive,
3002 .vop_open = ufs_open,
3003 .vop_pathconf = ufs_pathconf,
3004 .vop_poll = vop_stdpoll,
3005 .vop_print = ufs_print,
3006 .vop_read_pgcache = ufs_read_pgcache,
3007 .vop_readdir = ufs_readdir,
3008 .vop_readlink = ufs_readlink,
3009 .vop_reclaim = ufs_reclaim,
3010 .vop_remove = ufs_remove,
3011 .vop_rename = ufs_rename,
3012 .vop_rmdir = ufs_rmdir,
3013 .vop_setattr = ufs_setattr,
3015 .vop_setlabel = vop_stdsetlabel_ea,
3017 .vop_strategy = ufs_strategy,
3018 .vop_symlink = ufs_symlink,
3019 .vop_whiteout = ufs_whiteout,
3021 .vop_getextattr = ufs_getextattr,
3022 .vop_deleteextattr = ufs_deleteextattr,
3023 .vop_setextattr = ufs_setextattr,
3026 .vop_getacl = ufs_getacl,
3027 .vop_setacl = ufs_setacl,
3028 .vop_aclcheck = ufs_aclcheck,
3031 VFS_VOP_VECTOR_REGISTER(ufs_vnodeops);
3033 struct vop_vector ufs_fifoops = {
3034 .vop_default = &fifo_specops,
3035 .vop_fsync = VOP_PANIC,
3036 .vop_accessx = ufs_accessx,
3037 .vop_close = ufsfifo_close,
3038 .vop_getattr = ufs_getattr,
3039 .vop_inactive = ufs_inactive,
3040 .vop_pathconf = ufs_pathconf,
3041 .vop_print = ufs_print,
3042 .vop_read = VOP_PANIC,
3043 .vop_reclaim = ufs_reclaim,
3044 .vop_setattr = ufs_setattr,
3046 .vop_setlabel = vop_stdsetlabel_ea,
3048 .vop_write = VOP_PANIC,
3050 .vop_getextattr = ufs_getextattr,
3051 .vop_deleteextattr = ufs_deleteextattr,
3052 .vop_setextattr = ufs_setextattr,
3055 .vop_getacl = ufs_getacl,
3056 .vop_setacl = ufs_setacl,
3057 .vop_aclcheck = ufs_aclcheck,
3060 VFS_VOP_VECTOR_REGISTER(ufs_fifoops);