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)
195 ufs_itimes_locked(vp);
200 ufs_sync_nlink1(struct mount *mp)
204 error = vfs_busy(mp, 0);
206 VFS_SYNC(mp, MNT_WAIT);
215 ufs_sync_nlink(struct vnode *vp, struct vnode *vp1)
222 if (ip->i_nlink < UFS_LINK_MAX)
224 if (!DOINGSOFTDEP(vp) || ip->i_effnlink >= UFS_LINK_MAX)
232 error = ufs_sync_nlink1(mp);
233 vn_lock_pair(vp, false, vp1, false);
238 * Create a regular file
242 struct vop_create_args /* {
244 struct vnode **a_vpp;
245 struct componentname *a_cnp;
252 ufs_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode),
253 ap->a_dvp, ap->a_vpp, ap->a_cnp, "ufs_create");
256 if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0)
257 cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp);
267 struct vop_mknod_args /* {
269 struct vnode **a_vpp;
270 struct componentname *a_cnp;
274 struct vattr *vap = ap->a_vap;
275 struct vnode **vpp = ap->a_vpp;
280 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode),
281 ap->a_dvp, vpp, ap->a_cnp, "ufs_mknod");
285 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
286 if (vap->va_rdev != VNOVAL) {
288 * Want to be able to use this to make badblock
289 * inodes, so don't truncate the dev number.
291 DIP_SET(ip, i_rdev, vap->va_rdev);
294 * Remove inode, then reload it through VFS_VGET(). This is
295 * needed to do further inode initialization, for instance
296 * fifo, which was too early for VFS_VGET() done as part of
299 (*vpp)->v_type = VNON;
300 ino = ip->i_number; /* Save this before vgone() invalidates ip. */
303 error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp);
316 ufs_open(struct vop_open_args *ap)
318 struct vnode *vp = ap->a_vp;
321 if (vp->v_type == VCHR || vp->v_type == VBLK)
325 vnode_create_vobject(vp, DIP(ip, i_size), ap->a_td);
326 if (vp->v_type == VREG && (vn_irflag_read(vp) & VIRF_PGREAD) == 0 &&
327 ip->i_ump->um_bsize >= PAGE_SIZE) {
328 vn_irflag_set_cond(vp, VIRF_PGREAD);
332 * Files marked append-only must be opened for appending.
334 if ((ip->i_flags & APPEND) &&
335 (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE)
344 * Update the times on the inode.
349 struct vop_close_args /* {
352 struct ucred *a_cred;
356 struct vnode *vp = ap->a_vp;
364 struct vop_accessx_args /* {
367 struct ucred *a_cred;
371 struct vnode *vp = ap->a_vp;
372 struct inode *ip = VTOI(vp);
373 accmode_t accmode = ap->a_accmode;
381 * Disallow write attempts on read-only filesystems;
382 * unless the file is a socket, fifo, or a block or
383 * character device resident on the filesystem.
385 if (accmode & VMODIFY_PERMS) {
386 switch (vp->v_type) {
390 if (vp->v_mount->mnt_flag & MNT_RDONLY)
394 * Inode is accounted in the quotas only if struct
395 * dquot is attached to it. VOP_ACCESS() is called
396 * from vn_open_cred() and provides a convenient
397 * point to call getinoquota(). The lock mode is
398 * exclusive when the file is opening for write.
400 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
401 error = getinoquota(ip);
413 * If immutable bit set, nobody gets to write it. "& ~VADMIN_PERMS"
414 * permits the owner of the file to remove the IMMUTABLE flag.
416 if ((accmode & (VMODIFY_PERMS & ~VADMIN_PERMS)) &&
417 (ip->i_flags & (IMMUTABLE | SF_SNAPSHOT)))
421 if ((vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS)) != 0) {
422 if (vp->v_mount->mnt_flag & MNT_NFS4ACLS)
423 type = ACL_TYPE_NFS4;
425 type = ACL_TYPE_ACCESS;
427 acl = acl_alloc(M_WAITOK);
428 if (type == ACL_TYPE_NFS4)
429 error = ufs_getacl_nfs4_internal(vp, acl, ap->a_td);
431 error = VOP_GETACL(vp, type, acl, ap->a_cred, ap->a_td);
434 if (type == ACL_TYPE_NFS4) {
435 error = vaccess_acl_nfs4(vp->v_type, ip->i_uid,
436 ip->i_gid, acl, accmode, ap->a_cred);
438 error = vfs_unixify_accmode(&accmode);
440 error = vaccess_acl_posix1e(vp->v_type, ip->i_uid,
441 ip->i_gid, acl, accmode, ap->a_cred);
445 if (error != EOPNOTSUPP)
447 "ufs_accessx(): Error retrieving ACL on object (%d).\n",
450 * XXX: Fall back until debugged. Should
451 * eventually possibly log an error, and return
454 error = vfs_unixify_accmode(&accmode);
456 error = vaccess(vp->v_type, ip->i_mode,
457 ip->i_uid, ip->i_gid, accmode, ap->a_cred);
463 #endif /* !UFS_ACL */
464 error = vfs_unixify_accmode(&accmode);
466 error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid,
467 accmode, ap->a_cred);
472 * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see
473 * the comment above cache_fplookup for details.
477 struct vop_fplookup_vexec_args /* {
479 struct ucred *a_cred;
490 if (__predict_false(ip == NULL))
496 * ACLs are not supported and UFS clears/sets this flag on mount and
497 * remount. However, we may still be racing with seeing them and there
498 * is no provision to make sure they were accounted for. This matches
499 * the behavior of the locked case, since the lookup there is also
500 * racy: mount takes no measures to block anyone from progressing.
502 all_x = S_IXUSR | S_IXGRP | S_IXOTH;
503 mode = atomic_load_short(&ip->i_mode);
504 if (__predict_true((mode & all_x) == all_x))
508 return (vaccess_vexec_smr(mode, ip->i_uid, ip->i_gid, cred));
513 ufs_stat(struct vop_stat_args *ap)
515 struct vnode *vp = ap->a_vp;
516 struct inode *ip = VTOI(vp);
517 struct stat *sb = ap->a_sb;
520 error = vop_stat_helper_pre(ap);
521 if (__predict_false(error))
525 ufs_itimes_locked(vp);
527 sb->st_atim.tv_sec = ip->i_din1->di_atime;
528 sb->st_atim.tv_nsec = ip->i_din1->di_atimensec;
530 sb->st_atim.tv_sec = ip->i_din2->di_atime;
531 sb->st_atim.tv_nsec = ip->i_din2->di_atimensec;
535 sb->st_dev = dev2udev(ITOUMP(ip)->um_dev);
536 sb->st_ino = ip->i_number;
537 sb->st_mode = (ip->i_mode & ~IFMT) | VTTOIF(vp->v_type);
538 sb->st_nlink = ip->i_effnlink;
539 sb->st_uid = ip->i_uid;
540 sb->st_gid = ip->i_gid;
542 sb->st_rdev = ip->i_din1->di_rdev;
543 sb->st_size = ip->i_din1->di_size;
544 sb->st_mtim.tv_sec = ip->i_din1->di_mtime;
545 sb->st_mtim.tv_nsec = ip->i_din1->di_mtimensec;
546 sb->st_ctim.tv_sec = ip->i_din1->di_ctime;
547 sb->st_ctim.tv_nsec = ip->i_din1->di_ctimensec;
548 sb->st_birthtim.tv_sec = -1;
549 sb->st_birthtim.tv_nsec = 0;
550 sb->st_blocks = dbtob((u_quad_t)ip->i_din1->di_blocks) / S_BLKSIZE;
552 sb->st_rdev = ip->i_din2->di_rdev;
553 sb->st_size = ip->i_din2->di_size;
554 sb->st_mtim.tv_sec = ip->i_din2->di_mtime;
555 sb->st_mtim.tv_nsec = ip->i_din2->di_mtimensec;
556 sb->st_ctim.tv_sec = ip->i_din2->di_ctime;
557 sb->st_ctim.tv_nsec = ip->i_din2->di_ctimensec;
558 sb->st_birthtim.tv_sec = ip->i_din2->di_birthtime;
559 sb->st_birthtim.tv_nsec = ip->i_din2->di_birthnsec;
560 sb->st_blocks = dbtob((u_quad_t)ip->i_din2->di_blocks) / S_BLKSIZE;
563 sb->st_blksize = max(PAGE_SIZE, vp->v_mount->mnt_stat.f_iosize);
564 sb->st_flags = ip->i_flags;
565 sb->st_gen = ip->i_gen;
567 return (vop_stat_helper_post(ap, error));
573 struct vop_getattr_args /* {
576 struct ucred *a_cred;
579 struct vnode *vp = ap->a_vp;
580 struct inode *ip = VTOI(vp);
581 struct vattr *vap = ap->a_vap;
584 ufs_itimes_locked(vp);
586 vap->va_atime.tv_sec = ip->i_din1->di_atime;
587 vap->va_atime.tv_nsec = ip->i_din1->di_atimensec;
589 vap->va_atime.tv_sec = ip->i_din2->di_atime;
590 vap->va_atime.tv_nsec = ip->i_din2->di_atimensec;
594 * Copy from inode table
596 vap->va_fsid = dev2udev(ITOUMP(ip)->um_dev);
597 vap->va_fileid = ip->i_number;
598 vap->va_mode = ip->i_mode & ~IFMT;
599 vap->va_nlink = ip->i_effnlink;
600 vap->va_uid = ip->i_uid;
601 vap->va_gid = ip->i_gid;
603 vap->va_rdev = ip->i_din1->di_rdev;
604 vap->va_size = ip->i_din1->di_size;
605 vap->va_mtime.tv_sec = ip->i_din1->di_mtime;
606 vap->va_mtime.tv_nsec = ip->i_din1->di_mtimensec;
607 vap->va_ctime.tv_sec = ip->i_din1->di_ctime;
608 vap->va_ctime.tv_nsec = ip->i_din1->di_ctimensec;
609 vap->va_bytes = dbtob((u_quad_t)ip->i_din1->di_blocks);
610 vap->va_filerev = ip->i_din1->di_modrev;
612 vap->va_rdev = ip->i_din2->di_rdev;
613 vap->va_size = ip->i_din2->di_size;
614 vap->va_mtime.tv_sec = ip->i_din2->di_mtime;
615 vap->va_mtime.tv_nsec = ip->i_din2->di_mtimensec;
616 vap->va_ctime.tv_sec = ip->i_din2->di_ctime;
617 vap->va_ctime.tv_nsec = ip->i_din2->di_ctimensec;
618 vap->va_birthtime.tv_sec = ip->i_din2->di_birthtime;
619 vap->va_birthtime.tv_nsec = ip->i_din2->di_birthnsec;
620 vap->va_bytes = dbtob((u_quad_t)ip->i_din2->di_blocks);
621 vap->va_filerev = ip->i_din2->di_modrev;
623 vap->va_flags = ip->i_flags;
624 vap->va_gen = ip->i_gen;
625 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize;
626 vap->va_type = IFTOVT(ip->i_mode);
631 * Set attribute vnode op. called from several syscalls
635 struct vop_setattr_args /* {
638 struct ucred *a_cred;
641 struct vattr *vap = ap->a_vap;
642 struct vnode *vp = ap->a_vp;
643 struct inode *ip = VTOI(vp);
644 struct ucred *cred = ap->a_cred;
645 struct thread *td = curthread;
649 * Check for unsettable attributes.
651 if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) ||
652 (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) ||
653 (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) ||
654 ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) {
657 if (vap->va_flags != VNOVAL) {
658 if ((vap->va_flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE |
659 SF_NOUNLINK | SF_SNAPSHOT | UF_APPEND | UF_ARCHIVE |
660 UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | UF_NOUNLINK |
661 UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
662 UF_SPARSE | UF_SYSTEM)) != 0)
664 if (vp->v_mount->mnt_flag & MNT_RDONLY)
667 * Callers may only modify the file flags on objects they
668 * have VADMIN rights for.
670 if ((error = VOP_ACCESS(vp, VADMIN, cred, td)))
673 * Unprivileged processes are not permitted to unset system
674 * flags, or modify flags if any system flags are set.
675 * Privileged non-jail processes may not modify system flags
676 * if securelevel > 0 and any existing system flags are set.
677 * Privileged jail processes behave like privileged non-jail
678 * processes if the PR_ALLOW_CHFLAGS permission bit is set;
679 * otherwise, they behave like unprivileged processes.
681 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
683 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
684 error = securelevel_gt(cred, 0);
688 /* The snapshot flag cannot be toggled. */
689 if ((vap->va_flags ^ ip->i_flags) & SF_SNAPSHOT)
693 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
694 ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE))
697 ip->i_flags = vap->va_flags;
698 DIP_SET(ip, i_flags, vap->va_flags);
699 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
700 error = UFS_UPDATE(vp, 0);
701 if (ip->i_flags & (IMMUTABLE | APPEND))
705 * If immutable or append, no one can change any of its attributes
706 * except the ones already handled (in some cases, file flags
707 * including the immutability flags themselves for the superuser).
709 if (ip->i_flags & (IMMUTABLE | APPEND))
712 * Go through the fields and update iff not VNOVAL.
714 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
715 if (vp->v_mount->mnt_flag & MNT_RDONLY)
717 if ((error = ufs_chown(vp, vap->va_uid, vap->va_gid, cred,
721 if (vap->va_size != VNOVAL) {
723 * XXX most of the following special cases should be in
724 * callers instead of in N filesystems. The VDIR check
727 switch (vp->v_type) {
733 * Truncation should have an effect in these cases.
734 * Disallow it if the filesystem is read-only or
735 * the file is being snapshotted.
737 if (vp->v_mount->mnt_flag & MNT_RDONLY)
744 * According to POSIX, the result is unspecified
745 * for file types other than regular files,
746 * directories and shared memory objects. We
747 * don't support shared memory objects in the file
748 * system, and have dubious support for truncating
749 * symlinks. Just ignore the request in other cases.
753 if ((error = UFS_TRUNCATE(vp, vap->va_size, IO_NORMAL |
754 ((vap->va_vaflags & VA_SYNC) != 0 ? IO_SYNC : 0),
758 if (vap->va_atime.tv_sec != VNOVAL ||
759 vap->va_mtime.tv_sec != VNOVAL ||
760 vap->va_birthtime.tv_sec != VNOVAL) {
761 if (vp->v_mount->mnt_flag & MNT_RDONLY)
765 error = vn_utimes_perm(vp, vap, cred, td);
768 UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED);
769 if (vap->va_atime.tv_sec != VNOVAL) {
770 ip->i_flag &= ~IN_ACCESS;
771 DIP_SET(ip, i_atime, vap->va_atime.tv_sec);
772 DIP_SET(ip, i_atimensec, vap->va_atime.tv_nsec);
774 if (vap->va_mtime.tv_sec != VNOVAL) {
775 ip->i_flag &= ~IN_UPDATE;
776 DIP_SET(ip, i_mtime, vap->va_mtime.tv_sec);
777 DIP_SET(ip, i_mtimensec, vap->va_mtime.tv_nsec);
779 if (vap->va_birthtime.tv_sec != VNOVAL && I_IS_UFS2(ip)) {
780 ip->i_din2->di_birthtime = vap->va_birthtime.tv_sec;
781 ip->i_din2->di_birthnsec = vap->va_birthtime.tv_nsec;
783 error = UFS_UPDATE(vp, 0);
788 if (vap->va_mode != (mode_t)VNOVAL) {
789 if (vp->v_mount->mnt_flag & MNT_RDONLY)
791 if (IS_SNAPSHOT(ip) && (vap->va_mode & (S_IXUSR | S_IWUSR |
792 S_IXGRP | S_IWGRP | S_IXOTH | S_IWOTH)) != 0)
794 error = ufs_chmod(vp, (int)vap->va_mode, cred, td);
801 ufs_update_nfs4_acl_after_mode_change(struct vnode *vp, int mode,
802 int file_owner_id, struct ucred *cred, struct thread *td)
807 aclp = acl_alloc(M_WAITOK);
808 error = ufs_getacl_nfs4_internal(vp, aclp, td);
810 * We don't have to handle EOPNOTSUPP here, as the filesystem claims
816 acl_nfs4_sync_acl_from_mode(aclp, mode, file_owner_id);
817 error = ufs_setacl_nfs4_internal(vp, aclp, td);
827 struct vop_mmapped_args /* {
839 if ((mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
840 UFS_INODE_SET_FLAG_SHARED(ip, IN_ACCESS);
842 * XXXKIB No UFS_UPDATE(ap->a_vp, 0) there.
848 * Change the mode on a file.
849 * Inode must be locked before calling.
852 ufs_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td)
854 struct inode *ip = VTOI(vp);
858 * To modify the permissions on a file, must possess VADMIN
861 if ((error = VOP_ACCESSX(vp, VWRITE_ACL, cred, td)))
864 * Privileged processes may set the sticky bit on non-directories,
865 * as well as set the setgid bit on a file with a group that the
866 * process is not a member of. Both of these are allowed in
869 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
870 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
873 if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) {
874 error = priv_check_cred(cred, PRIV_VFS_SETGID);
880 * Deny setting setuid if we are not the file owner.
882 if ((mode & ISUID) && ip->i_uid != cred->cr_uid) {
883 error = priv_check_cred(cred, PRIV_VFS_ADMIN);
888 newmode = ip->i_mode & ~ALLPERMS;
889 newmode |= (mode & ALLPERMS);
890 UFS_INODE_SET_MODE(ip, newmode);
891 DIP_SET(ip, i_mode, ip->i_mode);
892 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
894 if ((vp->v_mount->mnt_flag & MNT_NFS4ACLS) != 0)
895 error = ufs_update_nfs4_acl_after_mode_change(vp, mode, ip->i_uid, cred, td);
897 if (error == 0 && (ip->i_flag & IN_CHANGE) != 0)
898 error = UFS_UPDATE(vp, 0);
904 * Perform chown operation on inode ip;
905 * inode must be locked prior to call.
908 ufs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
911 struct inode *ip = VTOI(vp);
920 if (uid == (uid_t)VNOVAL)
922 if (gid == (gid_t)VNOVAL)
925 * To modify the ownership of a file, must possess VADMIN for that
928 if ((error = VOP_ACCESSX(vp, VWRITE_OWNER, cred, td)))
931 * To change the owner of a file, or change the group of a file to a
932 * group of which we are not a member, the caller must have
935 if (((uid != ip->i_uid && uid != cred->cr_uid) ||
936 (gid != ip->i_gid && !groupmember(gid, cred))) &&
937 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
942 if ((error = getinoquota(ip)) != 0)
945 dqrele(vp, ip->i_dquot[USRQUOTA]);
946 ip->i_dquot[USRQUOTA] = NODQUOT;
949 dqrele(vp, ip->i_dquot[GRPQUOTA]);
950 ip->i_dquot[GRPQUOTA] = NODQUOT;
952 change = DIP(ip, i_blocks);
953 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
954 (void) chkiq(ip, -1, cred, CHOWN|FORCE);
955 for (i = 0; i < MAXQUOTAS; i++) {
956 dqrele(vp, ip->i_dquot[i]);
957 ip->i_dquot[i] = NODQUOT;
961 DIP_SET(ip, i_gid, gid);
963 DIP_SET(ip, i_uid, uid);
965 if ((error = getinoquota(ip)) == 0) {
967 dqrele(vp, ip->i_dquot[USRQUOTA]);
968 ip->i_dquot[USRQUOTA] = NODQUOT;
971 dqrele(vp, ip->i_dquot[GRPQUOTA]);
972 ip->i_dquot[GRPQUOTA] = NODQUOT;
974 if ((error = chkdq(ip, change, cred, CHOWN)) == 0) {
975 if ((error = chkiq(ip, 1, cred, CHOWN)) == 0)
978 (void) chkdq(ip, -change, cred, CHOWN|FORCE);
980 for (i = 0; i < MAXQUOTAS; i++) {
981 dqrele(vp, ip->i_dquot[i]);
982 ip->i_dquot[i] = NODQUOT;
986 DIP_SET(ip, i_gid, ogid);
988 DIP_SET(ip, i_uid, ouid);
989 if (getinoquota(ip) == 0) {
991 dqrele(vp, ip->i_dquot[USRQUOTA]);
992 ip->i_dquot[USRQUOTA] = NODQUOT;
995 dqrele(vp, ip->i_dquot[GRPQUOTA]);
996 ip->i_dquot[GRPQUOTA] = NODQUOT;
998 (void) chkdq(ip, change, cred, FORCE|CHOWN);
999 (void) chkiq(ip, 1, cred, FORCE|CHOWN);
1000 (void) getinoquota(ip);
1004 if (getinoquota(ip))
1005 panic("ufs_chown: lost quota");
1007 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1008 if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) {
1009 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1010 UFS_INODE_SET_MODE(ip, ip->i_mode & ~(ISUID | ISGID));
1011 DIP_SET(ip, i_mode, ip->i_mode);
1014 error = UFS_UPDATE(vp, 0);
1020 struct vop_remove_args /* {
1021 struct vnode *a_dvp;
1023 struct componentname *a_cnp;
1027 struct vnode *vp = ap->a_vp;
1028 struct vnode *dvp = ap->a_dvp;
1034 if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1035 (VTOI(dvp)->i_flags & APPEND))
1037 if (DOINGSUJ(dvp)) {
1038 error = softdep_prelink(dvp, vp, ap->a_cnp);
1040 MPASS(error == ERELOOKUP);
1046 ufs_gjournal_orphan(vp);
1048 error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0);
1049 if (ip->i_nlink <= 0)
1050 vp->v_vflag |= VV_NOSYNC;
1051 if (IS_SNAPSHOT(ip)) {
1053 * Avoid deadlock where another thread is trying to
1054 * update the inodeblock for dvp and is waiting on
1055 * snaplk. Temporary unlock the vnode lock for the
1056 * unlinked file and sync the directory. This should
1057 * allow vput() of the directory to not block later on
1058 * while holding the snapshot vnode locked, assuming
1059 * that the directory hasn't been unlinked too.
1062 (void) VOP_FSYNC(dvp, MNT_WAIT, td);
1063 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1069 print_bad_link_count(const char *funcname, struct vnode *dvp)
1074 uprintf("%s: Bad link count %d on parent inode %jd in file system %s\n",
1075 funcname, dip->i_effnlink, (intmax_t)dip->i_number,
1076 dvp->v_mount->mnt_stat.f_mntonname);
1084 struct vop_link_args /* {
1085 struct vnode *a_tdvp;
1087 struct componentname *a_cnp;
1090 struct vnode *vp = ap->a_vp;
1091 struct vnode *tdvp = ap->a_tdvp;
1092 struct componentname *cnp = ap->a_cnp;
1094 struct direct newdir;
1098 if ((cnp->cn_flags & HASBUF) == 0)
1099 panic("ufs_link: no name");
1102 if (DOINGSUJ(tdvp)) {
1103 error = softdep_prelink(tdvp, vp, cnp);
1105 MPASS(error == ERELOOKUP);
1110 if (VTOI(tdvp)->i_effnlink < 2) {
1111 print_bad_link_count("ufs_link", tdvp);
1115 error = ufs_sync_nlink(vp, tdvp);
1121 * The file may have been removed after namei dropped the original
1124 if (ip->i_effnlink == 0) {
1128 if (ip->i_flags & (IMMUTABLE | APPEND)) {
1135 DIP_SET(ip, i_nlink, ip->i_nlink);
1136 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1137 if (DOINGSOFTDEP(vp))
1138 softdep_setup_link(VTOI(tdvp), ip);
1139 error = UFS_UPDATE(vp, !DOINGSOFTDEP(vp) && !DOINGASYNC(vp));
1141 ufs_makedirentry(ip, cnp, &newdir);
1142 error = ufs_direnter(tdvp, vp, &newdir, cnp, NULL);
1148 DIP_SET(ip, i_nlink, ip->i_nlink);
1149 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
1150 if (DOINGSOFTDEP(vp))
1151 softdep_revert_link(VTOI(tdvp), ip);
1158 * whiteout vnode call
1162 struct vop_whiteout_args /* {
1163 struct vnode *a_dvp;
1164 struct componentname *a_cnp;
1168 struct vnode *dvp = ap->a_dvp;
1169 struct componentname *cnp = ap->a_cnp;
1170 struct direct newdir;
1173 if (DOINGSUJ(dvp) && (ap->a_flags == CREATE ||
1174 ap->a_flags == DELETE)) {
1175 error = softdep_prelink(dvp, NULL, cnp);
1177 MPASS(error == ERELOOKUP);
1182 switch (ap->a_flags) {
1184 /* 4.4 format directories support whiteout operations */
1187 return (EOPNOTSUPP);
1190 /* create a new directory whiteout */
1192 if ((cnp->cn_flags & SAVENAME) == 0)
1193 panic("ufs_whiteout: missing name");
1195 panic("ufs_whiteout: old format filesystem");
1198 newdir.d_ino = UFS_WINO;
1199 newdir.d_namlen = cnp->cn_namelen;
1200 bcopy(cnp->cn_nameptr, newdir.d_name, (unsigned)cnp->cn_namelen + 1);
1201 newdir.d_type = DT_WHT;
1202 error = ufs_direnter(dvp, NULL, &newdir, cnp, NULL);
1206 /* remove an existing directory whiteout */
1209 panic("ufs_whiteout: old format filesystem");
1212 cnp->cn_flags &= ~DOWHITEOUT;
1213 error = ufs_dirremove(dvp, NULL, cnp->cn_flags, 0);
1216 panic("ufs_whiteout: unknown op");
1221 static volatile int rename_restarts;
1222 SYSCTL_INT(_vfs_ufs, OID_AUTO, rename_restarts, CTLFLAG_RD,
1223 __DEVOLATILE(int *, &rename_restarts), 0,
1224 "Times rename had to restart due to lock contention");
1227 * Rename system call.
1228 * rename("foo", "bar");
1231 * link("foo", "bar");
1233 * but ``atomically''. Can't do full commit without saving state in the
1234 * inode on disk which isn't feasible at this time. Best we can do is
1235 * always guarantee the target exists.
1237 * Basic algorithm is:
1239 * 1) Bump link count on source while we're linking it to the
1240 * target. This also ensure the inode won't be deleted out
1241 * from underneath us while we work (it may be truncated by
1242 * a concurrent `trunc' or `open' for creation).
1243 * 2) Link source to destination. If destination already exists,
1245 * 3) Unlink source reference to inode if still around. If a
1246 * directory was moved and the parent of the destination
1247 * is different from the source, patch the ".." entry in the
1252 struct vop_rename_args /* {
1253 struct vnode *a_fdvp;
1254 struct vnode *a_fvp;
1255 struct componentname *a_fcnp;
1256 struct vnode *a_tdvp;
1257 struct vnode *a_tvp;
1258 struct componentname *a_tcnp;
1261 struct vnode *tvp = ap->a_tvp;
1262 struct vnode *tdvp = ap->a_tdvp;
1263 struct vnode *fvp = ap->a_fvp;
1264 struct vnode *fdvp = ap->a_fdvp;
1266 struct componentname *tcnp = ap->a_tcnp;
1267 struct componentname *fcnp = ap->a_fcnp;
1268 struct thread *td = curthread;
1269 struct inode *fip, *tip, *tdp, *fdp;
1270 struct direct newdir;
1272 int doingdirectory, newparent;
1276 seqc_t fdvp_s, fvp_s, tdvp_s, tvp_s;
1277 bool checkpath_locked, want_seqc_end;
1279 checkpath_locked = want_seqc_end = false;
1282 if ((tcnp->cn_flags & HASBUF) == 0 ||
1283 (fcnp->cn_flags & HASBUF) == 0)
1284 panic("ufs_rename: no name");
1289 if (tvp && tvp != tdvp)
1292 * Check for cross-device rename.
1294 if ((fvp->v_mount != tdvp->v_mount) ||
1295 (tvp && (fvp->v_mount != tvp->v_mount))) {
1301 fdvp_s = fvp_s = tdvp_s = tvp_s = SEQC_MOD;
1304 * We need to acquire 2 to 4 locks depending on whether tvp is NULL
1305 * and fdvp and tdvp are the same directory. Subsequently we need
1306 * to double-check all paths and in the directory rename case we
1307 * need to verify that we are not creating a directory loop. To
1308 * handle this we acquire all but fdvp using non-blocking
1309 * acquisitions. If we fail to acquire any lock in the path we will
1310 * drop all held locks, acquire the new lock in a blocking fashion,
1311 * and then release it and restart the rename. This acquire/release
1312 * step ensures that we do not spin on a lock waiting for release.
1314 error = vn_lock(fdvp, LK_EXCLUSIVE);
1317 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1319 error = vn_lock(tdvp, LK_EXCLUSIVE);
1323 atomic_add_int(&rename_restarts, 1);
1327 * Re-resolve fvp to be certain it still exists and fetch the
1330 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1336 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1342 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1348 atomic_add_int(&rename_restarts, 1);
1354 * Re-resolve tvp and acquire the vnode lock if present.
1356 error = ufs_lookup_ino(tdvp, NULL, tcnp, &ino);
1357 if (error != 0 && error != EJUSTRETURN) {
1364 * If tvp disappeared we just carry on.
1366 if (error == EJUSTRETURN && tvp != NULL) {
1371 * Get the tvp ino if the lookup succeeded. We may have to restart
1372 * if the non-blocking acquire fails.
1376 error = VFS_VGET(mp, ino, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
1386 error = VFS_VGET(mp, ino, LK_EXCLUSIVE, &nvp);
1390 atomic_add_int(&rename_restarts, 1);
1395 if (DOINGSUJ(fdvp) &&
1396 (seqc_in_modify(fdvp_s) || !vn_seqc_consistent(fdvp, fdvp_s) ||
1397 seqc_in_modify(fvp_s) || !vn_seqc_consistent(fvp, fvp_s) ||
1398 seqc_in_modify(tdvp_s) || !vn_seqc_consistent(tdvp, tdvp_s) ||
1399 (tvp != NULL && (seqc_in_modify(tvp_s) ||
1400 !vn_seqc_consistent(tvp, tvp_s))))) {
1401 error = softdep_prerename(fdvp, fvp, tdvp, tvp);
1412 if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
1413 (VTOI(tdvp)->i_flags & APPEND))) {
1418 * Renaming a file to itself has no effect. The upper layers should
1419 * not call us in that case. However, things could change after
1420 * we drop the locks above.
1428 ino = fip->i_number;
1429 if (fip->i_nlink >= UFS_LINK_MAX) {
1430 if (!DOINGSOFTDEP(fvp) || fip->i_effnlink >= UFS_LINK_MAX) {
1435 MPASS(!want_seqc_end);
1436 if (checkpath_locked) {
1437 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1438 checkpath_locked = false;
1445 VOP_VPUT_PAIR(tdvp, &tvp, true);
1446 error = ufs_sync_nlink1(mp);
1454 if ((fip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))
1455 || (fdp->i_flags & APPEND)) {
1459 if ((fip->i_mode & IFMT) == IFDIR) {
1461 * Avoid ".", "..", and aliases of "." for obvious reasons.
1463 if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') ||
1465 (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT) {
1469 if (fdp->i_number != tdp->i_number)
1470 newparent = tdp->i_number;
1473 if ((fvp->v_type == VDIR && fvp->v_mountedhere != NULL) ||
1474 (tvp != NULL && tvp->v_type == VDIR &&
1475 tvp->v_mountedhere != NULL)) {
1481 * If ".." must be changed (ie the directory gets a new
1482 * parent) then the source directory must not be in the
1483 * directory hierarchy above the target, as this would
1484 * orphan everything below the source directory. Also
1485 * the user must have write permission in the source so
1486 * as to be able to change "..".
1488 if (doingdirectory && newparent) {
1489 error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, curthread);
1493 sx_xlock(&VFSTOUFS(mp)->um_checkpath_lock);
1494 checkpath_locked = true;
1495 error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred,
1498 * We encountered a lock that we have to wait for. Unlock
1499 * everything else and VGET before restarting.
1502 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1503 checkpath_locked = false;
1509 error = VFS_VGET(mp, ino, LK_SHARED, &nvp);
1512 atomic_add_int(&rename_restarts, 1);
1517 if ((tcnp->cn_flags & SAVESTART) == 0)
1518 panic("ufs_rename: lost to startdir");
1520 if (fip->i_effnlink == 0 || fdp->i_effnlink == 0 ||
1521 tdp->i_effnlink == 0)
1522 panic("Bad effnlink fip %p, fdp %p, tdp %p", fip, fdp, tdp);
1525 vn_seqc_write_begin(tvp);
1526 vn_seqc_write_begin(tdvp);
1527 vn_seqc_write_begin(fvp);
1528 vn_seqc_write_begin(fdvp);
1529 want_seqc_end = true;
1532 * 1) Bump link count while we're moving stuff
1533 * around. If we crash somewhere before
1534 * completing our work, the link count
1535 * may be wrong, but correctable.
1539 DIP_SET(fip, i_nlink, fip->i_nlink);
1540 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1541 if (DOINGSOFTDEP(fvp))
1542 softdep_setup_link(tdp, fip);
1543 error = UFS_UPDATE(fvp, !DOINGSOFTDEP(fvp) && !DOINGASYNC(fvp));
1548 * 2) If target doesn't exist, link the target
1549 * to the source and unlink the source.
1550 * Otherwise, rewrite the target directory
1551 * entry to reference the source inode and
1552 * expunge the original entry's existence.
1555 if (ITODEV(tdp) != ITODEV(fip))
1556 panic("ufs_rename: EXDEV");
1557 if (doingdirectory && newparent) {
1559 * Account for ".." in new directory.
1560 * When source and destination have the same
1561 * parent we don't adjust the link count. The
1562 * actual link modification is completed when
1563 * .. is rewritten below.
1565 if (tdp->i_nlink >= UFS_LINK_MAX) {
1568 DIP_SET(fip, i_nlink, fip->i_nlink);
1569 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1570 if (DOINGSOFTDEP(fvp))
1571 softdep_revert_link(tdp, fip);
1572 if (!DOINGSOFTDEP(tdvp) ||
1573 tdp->i_effnlink >= UFS_LINK_MAX) {
1577 MPASS(want_seqc_end);
1579 vn_seqc_write_end(tvp);
1580 vn_seqc_write_end(tdvp);
1581 vn_seqc_write_end(fvp);
1582 vn_seqc_write_end(fdvp);
1583 want_seqc_end = false;
1585 MPASS(checkpath_locked);
1586 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1587 checkpath_locked = false;
1593 VOP_VPUT_PAIR(tdvp, &tvp, true);
1594 error = ufs_sync_nlink1(mp);
1603 ufs_makedirentry(fip, tcnp, &newdir);
1604 error = ufs_direnter(tdvp, NULL, &newdir, tcnp, NULL);
1607 /* Setup tdvp for directory compaction if needed. */
1608 if (I_COUNT(tdp) != 0 && I_ENDOFF(tdp) != 0 &&
1609 I_ENDOFF(tdp) < tdp->i_size)
1610 endoff = I_ENDOFF(tdp);
1612 if (ITODEV(tip) != ITODEV(tdp) || ITODEV(tip) != ITODEV(fip))
1613 panic("ufs_rename: EXDEV");
1615 * Short circuit rename(foo, foo).
1617 if (tip->i_number == fip->i_number)
1618 panic("ufs_rename: same file");
1620 * If the parent directory is "sticky", then the caller
1621 * must possess VADMIN for the parent directory, or the
1622 * destination of the rename. This implements append-only
1625 if ((tdp->i_mode & S_ISTXT) &&
1626 VOP_ACCESS(tdvp, VADMIN, tcnp->cn_cred, td) &&
1627 VOP_ACCESS(tvp, VADMIN, tcnp->cn_cred, td)) {
1632 * Target must be empty if a directory and have no links
1633 * to it. Also, ensure source and target are compatible
1634 * (both directories, or both not directories).
1636 if ((tip->i_mode & IFMT) == IFDIR) {
1637 if ((tip->i_effnlink > 2) ||
1638 !ufs_dirempty(tip, tdp->i_number, tcnp->cn_cred)) {
1642 if (!doingdirectory) {
1647 } else if (doingdirectory) {
1651 if (doingdirectory) {
1654 if (DOINGSOFTDEP(tdvp))
1655 softdep_change_linkcnt(tdp);
1658 if (DOINGSOFTDEP(tvp))
1659 softdep_change_linkcnt(tip);
1661 error = ufs_dirrewrite(tdp, tip, fip->i_number,
1662 IFTODT(fip->i_mode),
1663 (doingdirectory && newparent) ? newparent : doingdirectory);
1665 if (doingdirectory) {
1668 if (DOINGSOFTDEP(tdvp))
1669 softdep_change_linkcnt(tdp);
1672 if (DOINGSOFTDEP(tvp))
1673 softdep_change_linkcnt(tip);
1677 if (doingdirectory && !DOINGSOFTDEP(tvp)) {
1679 * The only stuff left in the directory is "."
1680 * and "..". The "." reference is inconsequential
1681 * since we are quashing it. We have removed the "."
1682 * reference and the reference in the parent directory,
1683 * but there may be other hard links. The soft
1684 * dependency code will arrange to do these operations
1685 * after the parent directory entry has been deleted on
1686 * disk, so when running with that code we avoid doing
1691 DIP_SET(tdp, i_nlink, tdp->i_nlink);
1692 UFS_INODE_SET_FLAG(tdp, IN_CHANGE);
1695 DIP_SET(tip, i_nlink, tip->i_nlink);
1696 UFS_INODE_SET_FLAG(tip, IN_CHANGE);
1701 * 3) Unlink the source. We have to resolve the path again to
1702 * fixup the directory offset and count for ufs_dirremove.
1705 error = ufs_lookup_ino(fdvp, NULL, fcnp, &ino);
1707 panic("ufs_rename: from entry went away!");
1708 if (ino != fip->i_number)
1709 panic("ufs_rename: ino mismatch %ju != %ju\n",
1710 (uintmax_t)ino, (uintmax_t)fip->i_number);
1713 * If the source is a directory with a
1714 * new parent, the link count of the old
1715 * parent directory must be decremented
1716 * and ".." set to point to the new parent.
1718 if (doingdirectory && newparent) {
1720 * If tip exists we simply use its link, otherwise we must
1726 DIP_SET(tdp, i_nlink, tdp->i_nlink);
1727 UFS_INODE_SET_FLAG(tdp, IN_CHANGE);
1728 if (DOINGSOFTDEP(tdvp))
1729 softdep_setup_dotdot_link(tdp, fip);
1730 error = UFS_UPDATE(tdvp, !DOINGSOFTDEP(tdvp) &&
1732 /* Don't go to bad here as the new link exists. */
1735 } else if (DOINGSUJ(tdvp))
1736 /* Journal must account for each new link. */
1737 softdep_setup_dotdot_link(tdp, fip);
1738 SET_I_OFFSET(fip, mastertemplate.dot_reclen);
1739 ufs_dirrewrite(fip, fdp, newparent, DT_DIR, 0);
1742 error = ufs_dirremove(fdvp, fip, fcnp->cn_flags, 0);
1744 * The kern_renameat() looks up the fvp using the DELETE flag, which
1745 * causes the removal of the name cache entry for fvp.
1746 * As the relookup of the fvp is done in two steps:
1747 * ufs_lookup_ino() and then VFS_VGET(), another thread might do a
1748 * normal lookup of the from name just before the VFS_VGET() call,
1749 * causing the cache entry to be re-instantiated.
1751 * The same issue also applies to tvp if it exists as
1752 * otherwise we may have a stale name cache entry for the new
1753 * name that references the old i-node if it has other links
1754 * or open file descriptors.
1756 cache_vop_rename(fdvp, fvp, tdvp, tvp, fcnp, tcnp);
1759 if (want_seqc_end) {
1761 vn_seqc_write_end(tvp);
1762 vn_seqc_write_end(tdvp);
1763 vn_seqc_write_end(fvp);
1764 vn_seqc_write_end(fdvp);
1767 if (checkpath_locked)
1768 sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
1774 * If compaction or fsync was requested do it in
1775 * ffs_vput_pair() now that other locks are no longer needed.
1777 if (error == 0 && endoff != 0) {
1778 UFS_INODE_SET_FLAG(tdp, IN_ENDOFF);
1779 SET_I_ENDOFF(tdp, endoff);
1781 VOP_VPUT_PAIR(tdvp, &tvp, true);
1787 DIP_SET(fip, i_nlink, fip->i_nlink);
1788 UFS_INODE_SET_FLAG(fip, IN_CHANGE);
1789 if (DOINGSOFTDEP(fvp))
1790 softdep_revert_link(tdp, fip);
1794 if (want_seqc_end) {
1796 vn_seqc_write_end(tvp);
1797 vn_seqc_write_end(tdvp);
1798 vn_seqc_write_end(fvp);
1799 vn_seqc_write_end(fdvp);
1813 ufs_do_posix1e_acl_inheritance_dir(struct vnode *dvp, struct vnode *tvp,
1814 mode_t dmode, struct ucred *cred, struct thread *td)
1817 struct inode *ip = VTOI(tvp);
1818 struct acl *dacl, *acl;
1820 acl = acl_alloc(M_WAITOK);
1821 dacl = acl_alloc(M_WAITOK);
1824 * Retrieve default ACL from parent, if any.
1826 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1830 * Retrieved a default ACL, so merge mode and ACL if
1831 * necessary. If the ACL is empty, fall through to
1832 * the "not defined or available" case.
1834 if (acl->acl_cnt != 0) {
1835 dmode = acl_posix1e_newfilemode(dmode, acl);
1836 UFS_INODE_SET_MODE(ip, dmode);
1837 DIP_SET(ip, i_mode, dmode);
1839 ufs_sync_acl_from_inode(ip, acl);
1846 * Just use the mode as-is.
1848 UFS_INODE_SET_MODE(ip, dmode);
1849 DIP_SET(ip, i_mode, dmode);
1858 * XXX: If we abort now, will Soft Updates notify the extattr
1859 * code that the EAs for the file need to be released?
1861 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1863 error = VOP_SETACL(tvp, ACL_TYPE_DEFAULT, dacl, cred, td);
1870 * XXX: This should not happen, as EOPNOTSUPP above
1871 * was supposed to free acl.
1873 printf("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()\n");
1875 panic("ufs_mkdir: VOP_GETACL() but no VOP_SETACL()");
1891 ufs_do_posix1e_acl_inheritance_file(struct vnode *dvp, struct vnode *tvp,
1892 mode_t mode, struct ucred *cred, struct thread *td)
1895 struct inode *ip = VTOI(tvp);
1898 acl = acl_alloc(M_WAITOK);
1901 * Retrieve default ACL for parent, if any.
1903 error = VOP_GETACL(dvp, ACL_TYPE_DEFAULT, acl, cred, td);
1907 * Retrieved a default ACL, so merge mode and ACL if
1910 if (acl->acl_cnt != 0) {
1912 * Two possible ways for default ACL to not
1913 * be present. First, the EA can be
1914 * undefined, or second, the default ACL can
1915 * be blank. If it's blank, fall through to
1916 * the it's not defined case.
1918 mode = acl_posix1e_newfilemode(mode, acl);
1919 UFS_INODE_SET_MODE(ip, mode);
1920 DIP_SET(ip, i_mode, mode);
1921 ufs_sync_acl_from_inode(ip, acl);
1928 * Just use the mode as-is.
1930 UFS_INODE_SET_MODE(ip, mode);
1931 DIP_SET(ip, i_mode, mode);
1940 * XXX: If we abort now, will Soft Updates notify the extattr
1941 * code that the EAs for the file need to be released?
1943 error = VOP_SETACL(tvp, ACL_TYPE_ACCESS, acl, cred, td);
1950 * XXX: This should not happen, as EOPNOTSUPP above was
1951 * supposed to free acl.
1953 printf("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1954 "but no VOP_SETACL()\n");
1955 /* panic("ufs_do_posix1e_acl_inheritance_file: VOP_GETACL() "
1956 "but no VOP_SETACL()"); */
1970 ufs_do_nfs4_acl_inheritance(struct vnode *dvp, struct vnode *tvp,
1971 mode_t child_mode, struct ucred *cred, struct thread *td)
1974 struct acl *parent_aclp, *child_aclp;
1976 parent_aclp = acl_alloc(M_WAITOK);
1977 child_aclp = acl_alloc(M_WAITOK | M_ZERO);
1979 error = ufs_getacl_nfs4_internal(dvp, parent_aclp, td);
1982 acl_nfs4_compute_inherited_acl(parent_aclp, child_aclp,
1983 child_mode, VTOI(tvp)->i_uid, tvp->v_type == VDIR);
1984 error = ufs_setacl_nfs4_internal(tvp, child_aclp, td);
1988 acl_free(parent_aclp);
1989 acl_free(child_aclp);
2000 struct vop_mkdir_args /* {
2001 struct vnode *a_dvp;
2002 struct vnode **a_vpp;
2003 struct componentname *a_cnp;
2004 struct vattr *a_vap;
2007 struct vnode *dvp = ap->a_dvp;
2008 struct vattr *vap = ap->a_vap;
2009 struct componentname *cnp = ap->a_cnp;
2010 struct inode *ip, *dp;
2013 struct dirtemplate dirtemplate, *dtp;
2014 struct direct newdir;
2019 if ((cnp->cn_flags & HASBUF) == 0)
2020 panic("ufs_mkdir: no name");
2023 error = ufs_sync_nlink(dvp, NULL);
2026 dmode = vap->va_mode & 0777;
2030 * Must simulate part of ufs_makeinode here to acquire the inode,
2031 * but not have it entered in the parent directory. The entry is
2032 * made later after writing "." and ".." entries.
2034 if (dp->i_effnlink < 2) {
2035 print_bad_link_count("ufs_mkdir", dvp);
2040 if (DOINGSUJ(dvp)) {
2041 error = softdep_prelink(dvp, NULL, cnp);
2043 MPASS(error == ERELOOKUP);
2048 error = UFS_VALLOC(dvp, dmode, cnp->cn_cred, &tvp);
2051 vn_seqc_write_begin(tvp);
2053 ip->i_gid = dp->i_gid;
2054 DIP_SET(ip, i_gid, dp->i_gid);
2058 struct ucred ucred, *ucp;
2063 * If we are hacking owners here, (only do this where told to)
2064 * and we are not giving it TO root, (would subvert quotas)
2065 * then go ahead and give it to the other user.
2066 * The new directory also inherits the SUID bit.
2067 * If user's UID and dir UID are the same,
2068 * 'give it away' so that the SUID is still forced on.
2070 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2071 (dp->i_mode & ISUID) && dp->i_uid) {
2073 ip->i_uid = dp->i_uid;
2074 DIP_SET(ip, i_uid, dp->i_uid);
2076 if (dp->i_uid != cnp->cn_cred->cr_uid) {
2078 * Make sure the correct user gets charged
2080 * Make a dummy credential for the victim.
2081 * XXX This seems to never be accessed out of
2082 * our context so a stack variable is ok.
2084 refcount_init(&ucred.cr_ref, 1);
2085 ucred.cr_uid = ip->i_uid;
2086 ucred.cr_ngroups = 1;
2087 ucred.cr_groups = &ucred_group;
2088 ucred.cr_groups[0] = dp->i_gid;
2093 ip->i_uid = cnp->cn_cred->cr_uid;
2094 DIP_SET(ip, i_uid, ip->i_uid);
2097 if ((error = getinoquota(ip)) ||
2098 (error = chkiq(ip, 1, ucp, 0))) {
2099 if (DOINGSOFTDEP(tvp))
2100 softdep_revert_link(dp, ip);
2101 UFS_VFREE(tvp, ip->i_number, dmode);
2102 vn_seqc_write_end(tvp);
2109 #else /* !SUIDDIR */
2110 ip->i_uid = cnp->cn_cred->cr_uid;
2111 DIP_SET(ip, i_uid, ip->i_uid);
2113 if ((error = getinoquota(ip)) ||
2114 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2115 if (DOINGSOFTDEP(tvp))
2116 softdep_revert_link(dp, ip);
2117 UFS_VFREE(tvp, ip->i_number, dmode);
2118 vn_seqc_write_end(tvp);
2124 #endif /* !SUIDDIR */
2125 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
2126 UFS_INODE_SET_MODE(ip, dmode);
2127 DIP_SET(ip, i_mode, dmode);
2128 tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */
2131 DIP_SET(ip, i_nlink, 2);
2133 if (cnp->cn_flags & ISWHITEOUT) {
2134 ip->i_flags |= UF_OPAQUE;
2135 DIP_SET(ip, i_flags, ip->i_flags);
2139 * Bump link count in parent directory to reflect work done below.
2140 * Should be done before reference is created so cleanup is
2141 * possible if we crash.
2145 DIP_SET(dp, i_nlink, dp->i_nlink);
2146 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2147 if (DOINGSOFTDEP(dvp))
2148 softdep_setup_mkdir(dp, ip);
2149 error = UFS_UPDATE(dvp, !DOINGSOFTDEP(dvp) && !DOINGASYNC(dvp));
2153 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2154 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2161 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2162 error = ufs_do_posix1e_acl_inheritance_dir(dvp, tvp, dmode,
2163 cnp->cn_cred, curthread);
2166 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2167 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, dmode,
2168 cnp->cn_cred, curthread);
2172 #endif /* !UFS_ACL */
2175 * Initialize directory with "." and ".." from static template.
2178 dtp = &mastertemplate;
2180 dtp = (struct dirtemplate *)&omastertemplate;
2182 dirtemplate.dot_ino = ip->i_number;
2183 dirtemplate.dotdot_ino = dp->i_number;
2184 vnode_pager_setsize(tvp, DIRBLKSIZ);
2185 if ((error = UFS_BALLOC(tvp, (off_t)0, DIRBLKSIZ, cnp->cn_cred,
2186 BA_CLRBUF, &bp)) != 0)
2188 ip->i_size = DIRBLKSIZ;
2189 DIP_SET(ip, i_size, DIRBLKSIZ);
2190 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
2191 bcopy((caddr_t)&dirtemplate, (caddr_t)bp->b_data, sizeof dirtemplate);
2192 if (DOINGSOFTDEP(tvp)) {
2194 * Ensure that the entire newly allocated block is a
2195 * valid directory so that future growth within the
2196 * block does not have to ensure that the block is
2197 * written before the inode.
2200 while (blkoff < bp->b_bcount) {
2202 (bp->b_data + blkoff))->d_reclen = DIRBLKSIZ;
2203 blkoff += DIRBLKSIZ;
2206 if ((error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) &&
2207 !DOINGASYNC(tvp))) != 0) {
2212 * Directory set up, now install its entry in the parent directory.
2214 * If we are not doing soft dependencies, then we must write out the
2215 * buffer containing the new directory body before entering the new
2216 * name in the parent. If we are doing soft dependencies, then the
2217 * buffer containing the new directory body will be passed to and
2218 * released in the soft dependency code after the code has attached
2219 * an appropriate ordering dependency to the buffer which ensures that
2220 * the buffer is written before the new name is written in the parent.
2222 if (DOINGASYNC(dvp))
2224 else if (!DOINGSOFTDEP(dvp) && ((error = bwrite(bp))))
2226 ufs_makedirentry(ip, cnp, &newdir);
2227 error = ufs_direnter(dvp, tvp, &newdir, cnp, bp);
2232 vn_seqc_write_end(tvp);
2236 DIP_SET(dp, i_nlink, dp->i_nlink);
2237 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2239 * No need to do an explicit VOP_TRUNCATE here, vrele will
2240 * do this for us because we set the link count to 0.
2244 DIP_SET(ip, i_nlink, 0);
2245 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2246 if (DOINGSOFTDEP(tvp))
2247 softdep_revert_mkdir(dp, ip);
2248 vn_seqc_write_end(tvp);
2257 * Rmdir system call.
2261 struct vop_rmdir_args /* {
2262 struct vnode *a_dvp;
2264 struct componentname *a_cnp;
2267 struct vnode *vp = ap->a_vp;
2268 struct vnode *dvp = ap->a_dvp;
2269 struct componentname *cnp = ap->a_cnp;
2270 struct inode *ip, *dp;
2277 * Do not remove a directory that is in the process of being renamed.
2278 * Verify the directory is empty (and valid). Rmdir ".." will not be
2279 * valid since ".." will contain a reference to the current directory
2280 * and thus be non-empty. Do not allow the removal of mounted on
2281 * directories (this can happen when an NFS exported filesystem
2282 * tries to remove a locally mounted on directory).
2285 if (dp->i_effnlink <= 2) {
2286 if (dp->i_effnlink == 2)
2287 print_bad_link_count("ufs_rmdir", dvp);
2291 if (!ufs_dirempty(ip, dp->i_number, cnp->cn_cred)) {
2295 if ((dp->i_flags & APPEND)
2296 || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
2300 if (vp->v_mountedhere != 0) {
2304 if (DOINGSUJ(dvp)) {
2305 error = softdep_prelink(dvp, vp, cnp);
2307 MPASS(error == ERELOOKUP);
2313 ufs_gjournal_orphan(vp);
2316 * Delete reference to directory before purging
2317 * inode. If we crash in between, the directory
2318 * will be reattached to lost+found,
2322 if (DOINGSOFTDEP(vp))
2323 softdep_setup_rmdir(dp, ip);
2324 error = ufs_dirremove(dvp, ip, cnp->cn_flags, 1);
2328 if (DOINGSOFTDEP(vp))
2329 softdep_revert_rmdir(dp, ip);
2333 * The only stuff left in the directory is "." and "..". The "."
2334 * reference is inconsequential since we are quashing it. The soft
2335 * dependency code will arrange to do these operations after
2336 * the parent directory entry has been deleted on disk, so
2337 * when running with that code we avoid doing them now.
2339 if (!DOINGSOFTDEP(vp)) {
2341 DIP_SET(dp, i_nlink, dp->i_nlink);
2342 UFS_INODE_SET_FLAG(dp, IN_CHANGE);
2343 error = UFS_UPDATE(dvp, 0);
2345 DIP_SET(ip, i_nlink, ip->i_nlink);
2346 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2348 cache_vop_rmdir(dvp, vp);
2350 /* Kill any active hash; i_effnlink == 0, so it will not come back. */
2351 if (ip->i_dirhash != NULL)
2352 ufsdirhash_free(ip);
2359 * symlink -- make a symbolic link
2363 struct vop_symlink_args /* {
2364 struct vnode *a_dvp;
2365 struct vnode **a_vpp;
2366 struct componentname *a_cnp;
2367 struct vattr *a_vap;
2368 const char *a_target;
2371 struct vnode *vp, **vpp = ap->a_vpp;
2375 error = ufs_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp,
2376 vpp, ap->a_cnp, "ufs_symlink");
2380 len = strlen(ap->a_target);
2381 if (len < VFSTOUFS(vp->v_mount)->um_maxsymlinklen) {
2383 bcopy(ap->a_target, DIP(ip, i_shortlink), len);
2385 DIP_SET(ip, i_size, len);
2386 UFS_INODE_SET_FLAG(ip, IN_SIZEMOD | IN_CHANGE | IN_UPDATE);
2387 error = UFS_UPDATE(vp, 0);
2389 error = vn_rdwr(UIO_WRITE, vp, __DECONST(void *, ap->a_target),
2390 len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK,
2391 ap->a_cnp->cn_cred, NOCRED, NULL, NULL);
2398 * Vnode op for reading directories.
2402 struct vop_readdir_args /* {
2405 struct ucred *a_cred;
2408 uint64_t **a_cookies;
2411 struct vnode *vp = ap->a_vp;
2412 struct uio *uio = ap->a_uio;
2415 struct direct *dp, *edp;
2417 struct dirent dstdp;
2418 off_t offset, startoffset;
2419 size_t readcnt, skipcnt;
2424 if (uio->uio_offset < 0)
2427 if (ip->i_effnlink == 0)
2429 if (ap->a_ncookies != NULL) {
2430 if (uio->uio_resid < 0)
2433 ncookies = uio->uio_resid;
2434 if (uio->uio_offset >= ip->i_size)
2436 else if (ip->i_size - uio->uio_offset < ncookies)
2437 ncookies = ip->i_size - uio->uio_offset;
2438 ncookies = ncookies / (offsetof(struct direct, d_name) + 4) + 1;
2439 cookies = malloc(ncookies * sizeof(*cookies), M_TEMP, M_WAITOK);
2440 *ap->a_ncookies = ncookies;
2441 *ap->a_cookies = cookies;
2446 offset = startoffset = uio->uio_offset;
2447 startresid = uio->uio_resid;
2449 while (error == 0 && uio->uio_resid > 0 &&
2450 uio->uio_offset < ip->i_size) {
2451 error = UFS_BLKATOFF(vp, uio->uio_offset, NULL, &bp);
2454 if (bp->b_offset + bp->b_bcount > ip->i_size)
2455 readcnt = ip->i_size - bp->b_offset;
2457 readcnt = bp->b_bcount;
2458 skipcnt = (size_t)(uio->uio_offset - bp->b_offset) &
2459 ~(size_t)(DIRBLKSIZ - 1);
2460 offset = bp->b_offset + skipcnt;
2461 dp = (struct direct *)&bp->b_data[skipcnt];
2462 edp = (struct direct *)&bp->b_data[readcnt];
2463 while (error == 0 && uio->uio_resid > 0 && dp < edp) {
2464 if (dp->d_reclen <= offsetof(struct direct, d_name) ||
2465 (caddr_t)dp + dp->d_reclen > (caddr_t)edp) {
2469 #if BYTE_ORDER == LITTLE_ENDIAN
2470 /* Old filesystem format. */
2472 dstdp.d_namlen = dp->d_type;
2473 dstdp.d_type = dp->d_namlen;
2477 dstdp.d_namlen = dp->d_namlen;
2478 dstdp.d_type = dp->d_type;
2480 if (offsetof(struct direct, d_name) + dstdp.d_namlen >
2485 if (offset < startoffset || dp->d_ino == 0)
2487 dstdp.d_fileno = dp->d_ino;
2488 dstdp.d_reclen = GENERIC_DIRSIZ(&dstdp);
2489 bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen);
2490 /* NOTE: d_off is the offset of the *next* entry. */
2491 dstdp.d_off = offset + dp->d_reclen;
2492 dirent_terminate(&dstdp);
2493 if (dstdp.d_reclen > uio->uio_resid) {
2494 if (uio->uio_resid == startresid)
2497 error = EJUSTRETURN;
2501 error = uiomove((caddr_t)&dstdp, dstdp.d_reclen, uio);
2504 if (cookies != NULL) {
2505 KASSERT(ncookies > 0,
2506 ("ufs_readdir: cookies buffer too small"));
2507 *cookies = offset + dp->d_reclen;
2512 offset += dp->d_reclen;
2513 dp = (struct direct *)((caddr_t)dp + dp->d_reclen);
2516 uio->uio_offset = offset;
2518 /* We need to correct uio_offset. */
2519 uio->uio_offset = offset;
2520 if (error == EJUSTRETURN)
2522 if (ap->a_ncookies != NULL) {
2524 ap->a_ncookies -= ncookies;
2526 free(*ap->a_cookies, M_TEMP);
2527 *ap->a_ncookies = 0;
2528 *ap->a_cookies = NULL;
2531 if (error == 0 && ap->a_eofflag)
2532 *ap->a_eofflag = ip->i_size <= uio->uio_offset;
2537 * Return target name of a symbolic link
2541 struct vop_readlink_args /* {
2544 struct ucred *a_cred;
2547 struct vnode *vp = ap->a_vp;
2548 struct inode *ip = VTOI(vp);
2552 if (isize < VFSTOUFS(vp->v_mount)->um_maxsymlinklen)
2553 return (uiomove(DIP(ip, i_shortlink), isize, ap->a_uio));
2554 return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred));
2558 * Calculate the logical to physical mapping if not done already,
2559 * then call the device strategy routine.
2561 * In order to be able to swap to a file, the ufs_bmaparray() operation may not
2562 * deadlock on memory. See ufs_bmap() for details.
2566 struct vop_strategy_args /* {
2571 struct buf *bp = ap->a_bp;
2572 struct vnode *vp = ap->a_vp;
2576 if (bp->b_blkno == bp->b_lblkno) {
2577 error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, bp, NULL, NULL);
2578 bp->b_blkno = blkno;
2580 bp->b_error = error;
2581 bp->b_ioflags |= BIO_ERROR;
2585 if ((long)bp->b_blkno == -1)
2588 if ((long)bp->b_blkno == -1) {
2592 bp->b_iooffset = dbtob(bp->b_blkno);
2593 BO_STRATEGY(VFSTOUFS(vp->v_mount)->um_bo, bp);
2598 * Print out the contents of an inode.
2602 struct vop_print_args /* {
2606 struct vnode *vp = ap->a_vp;
2607 struct inode *ip = VTOI(vp);
2609 printf("\tnlink=%d, effnlink=%d, size=%jd", ip->i_nlink,
2610 ip->i_effnlink, (intmax_t)ip->i_size);
2612 printf(", extsize %d", ip->i_din2->di_extsize);
2613 printf("\n\tgeneration=%jx, uid=%d, gid=%d, flags=0x%b\n",
2614 (uintmax_t)ip->i_gen, ip->i_uid, ip->i_gid,
2615 (u_int)ip->i_flags, PRINT_INODE_FLAGS);
2616 printf("\tino %lu, on dev %s", (u_long)ip->i_number,
2617 devtoname(ITODEV(ip)));
2618 if (vp->v_type == VFIFO)
2625 * Close wrapper for fifos.
2627 * Update the times on the inode then do device close.
2631 struct vop_close_args /* {
2634 struct ucred *a_cred;
2635 struct thread *a_td;
2640 return (fifo_specops.vop_close(ap));
2644 * Return POSIX pathconf information applicable to ufs filesystems.
2648 struct vop_pathconf_args /* {
2657 switch (ap->a_name) {
2659 *ap->a_retval = UFS_LINK_MAX;
2662 *ap->a_retval = UFS_MAXNAMLEN;
2665 if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
2666 *ap->a_retval = PIPE_BUF;
2670 case _PC_CHOWN_RESTRICTED:
2677 case _PC_ACL_EXTENDED:
2678 if (ap->a_vp->v_mount->mnt_flag & MNT_ACLS)
2684 if (ap->a_vp->v_mount->mnt_flag & MNT_NFS4ACLS)
2690 case _PC_ACL_PATH_MAX:
2692 if (ap->a_vp->v_mount->mnt_flag & (MNT_ACLS | MNT_NFS4ACLS))
2693 *ap->a_retval = ACL_MAX_ENTRIES;
2701 case _PC_MAC_PRESENT:
2702 if (ap->a_vp->v_mount->mnt_flag & MNT_MULTILABEL)
2708 case _PC_MIN_HOLE_SIZE:
2709 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2717 case _PC_ALLOC_SIZE_MIN:
2718 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize;
2720 case _PC_FILESIZEBITS:
2723 case _PC_REC_INCR_XFER_SIZE:
2724 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2726 case _PC_REC_MAX_XFER_SIZE:
2727 *ap->a_retval = -1; /* means ``unlimited'' */
2729 case _PC_REC_MIN_XFER_SIZE:
2730 *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize;
2732 case _PC_REC_XFER_ALIGN:
2733 *ap->a_retval = PAGE_SIZE;
2735 case _PC_SYMLINK_MAX:
2736 *ap->a_retval = MAXPATHLEN;
2740 error = vop_stdpathconf(ap);
2747 * Initialize the vnode associated with a new inode, handle aliased
2751 ufs_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp)
2757 ASSERT_VOP_LOCKED(vp, "ufs_vinit");
2759 vp->v_type = IFTOVT(ip->i_mode);
2761 * Only unallocated inodes should be of type VNON.
2763 if (ip->i_mode != 0 && vp->v_type == VNON)
2765 if (vp->v_type == VFIFO)
2767 if (ip->i_number == UFS_ROOTINO)
2768 vp->v_vflag |= VV_ROOT;
2774 * Allocate a new inode.
2775 * Vnode dvp must be locked.
2778 ufs_makeinode(int mode, struct vnode *dvp, struct vnode **vpp,
2779 struct componentname *cnp, const char *callfunc)
2781 struct inode *ip, *pdir;
2782 struct direct newdir;
2788 if ((cnp->cn_flags & HASBUF) == 0)
2789 panic("%s: no name", callfunc);
2792 if ((mode & IFMT) == 0)
2795 if (pdir->i_effnlink < 2) {
2796 print_bad_link_count(callfunc, dvp);
2799 if (DOINGSUJ(dvp)) {
2800 error = softdep_prelink(dvp, NULL, cnp);
2802 MPASS(error == ERELOOKUP);
2806 error = UFS_VALLOC(dvp, mode, cnp->cn_cred, &tvp);
2810 ip->i_gid = pdir->i_gid;
2811 DIP_SET(ip, i_gid, pdir->i_gid);
2815 struct ucred ucred, *ucp;
2820 * If we are not the owner of the directory,
2821 * and we are hacking owners here, (only do this where told to)
2822 * and we are not giving it TO root, (would subvert quotas)
2823 * then go ahead and give it to the other user.
2824 * Note that this drops off the execute bits for security.
2826 if ((dvp->v_mount->mnt_flag & MNT_SUIDDIR) &&
2827 (pdir->i_mode & ISUID) &&
2828 (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) {
2829 ip->i_uid = pdir->i_uid;
2830 DIP_SET(ip, i_uid, ip->i_uid);
2834 * Make sure the correct user gets charged
2836 * Quickly knock up a dummy credential for the victim.
2837 * XXX This seems to never be accessed out of our
2838 * context so a stack variable is ok.
2840 refcount_init(&ucred.cr_ref, 1);
2841 ucred.cr_uid = ip->i_uid;
2842 ucred.cr_ngroups = 1;
2843 ucred.cr_groups = &ucred_group;
2844 ucred.cr_groups[0] = pdir->i_gid;
2848 ip->i_uid = cnp->cn_cred->cr_uid;
2849 DIP_SET(ip, i_uid, ip->i_uid);
2853 if ((error = getinoquota(ip)) ||
2854 (error = chkiq(ip, 1, ucp, 0))) {
2855 if (DOINGSOFTDEP(tvp))
2856 softdep_revert_link(pdir, ip);
2857 UFS_VFREE(tvp, ip->i_number, mode);
2864 #else /* !SUIDDIR */
2865 ip->i_uid = cnp->cn_cred->cr_uid;
2866 DIP_SET(ip, i_uid, ip->i_uid);
2868 if ((error = getinoquota(ip)) ||
2869 (error = chkiq(ip, 1, cnp->cn_cred, 0))) {
2870 if (DOINGSOFTDEP(tvp))
2871 softdep_revert_link(pdir, ip);
2872 UFS_VFREE(tvp, ip->i_number, mode);
2878 #endif /* !SUIDDIR */
2879 vn_seqc_write_begin(tvp); /* Mostly to cover asserts */
2880 UFS_INODE_SET_FLAG(ip, IN_ACCESS | IN_CHANGE | IN_UPDATE);
2881 UFS_INODE_SET_MODE(ip, mode);
2882 DIP_SET(ip, i_mode, mode);
2883 tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */
2886 DIP_SET(ip, i_nlink, 1);
2887 if (DOINGSOFTDEP(tvp))
2888 softdep_setup_create(VTOI(dvp), ip);
2889 if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred) &&
2890 priv_check_cred(cnp->cn_cred, PRIV_VFS_SETGID)) {
2891 UFS_INODE_SET_MODE(ip, ip->i_mode & ~ISGID);
2892 DIP_SET(ip, i_mode, ip->i_mode);
2895 if (cnp->cn_flags & ISWHITEOUT) {
2896 ip->i_flags |= UF_OPAQUE;
2897 DIP_SET(ip, i_flags, ip->i_flags);
2901 * Make sure inode goes to disk before directory entry.
2903 error = UFS_UPDATE(tvp, !DOINGSOFTDEP(tvp) && !DOINGASYNC(tvp));
2907 if (dvp->v_mount->mnt_flag & MNT_MULTILABEL) {
2908 error = mac_vnode_create_extattr(cnp->cn_cred, dvp->v_mount,
2915 if (dvp->v_mount->mnt_flag & MNT_ACLS) {
2916 error = ufs_do_posix1e_acl_inheritance_file(dvp, tvp, mode,
2917 cnp->cn_cred, curthread);
2920 } else if (dvp->v_mount->mnt_flag & MNT_NFS4ACLS) {
2921 error = ufs_do_nfs4_acl_inheritance(dvp, tvp, mode,
2922 cnp->cn_cred, curthread);
2926 #endif /* !UFS_ACL */
2927 ufs_makedirentry(ip, cnp, &newdir);
2928 error = ufs_direnter(dvp, tvp, &newdir, cnp, NULL);
2931 vn_seqc_write_end(tvp);
2937 * Write error occurred trying to update the inode
2938 * or the directory so must deallocate the inode.
2942 DIP_SET(ip, i_nlink, 0);
2943 UFS_INODE_SET_FLAG(ip, IN_CHANGE);
2944 if (DOINGSOFTDEP(tvp))
2945 softdep_revert_create(VTOI(dvp), ip);
2946 vn_seqc_write_end(tvp);
2953 ufs_ioctl(struct vop_ioctl_args *ap)
2959 switch (ap->a_command) {
2961 error = vn_lock(vp, LK_SHARED);
2963 error = ufs_bmap_seekdata(vp, (off_t *)ap->a_data);
2969 return (vn_bmap_seekhole(vp, ap->a_command, (off_t *)ap->a_data,
2977 ufs_read_pgcache(struct vop_read_pgcache_args *ap)
2984 VNPASS((vn_irflag_read(vp) & VIRF_PGREAD) != 0, vp);
2986 if (uio->uio_resid > ptoa(io_hold_cnt) || uio->uio_offset < 0 ||
2987 (ap->a_ioflag & IO_DIRECT) != 0)
2988 return (EJUSTRETURN);
2989 return (vn_read_from_obj(vp, uio));
2992 /* Global vfs data structures for ufs. */
2993 struct vop_vector ufs_vnodeops = {
2994 .vop_default = &default_vnodeops,
2995 .vop_fsync = VOP_PANIC,
2996 .vop_read = VOP_PANIC,
2997 .vop_reallocblks = VOP_PANIC,
2998 .vop_write = VOP_PANIC,
2999 .vop_accessx = ufs_accessx,
3000 .vop_bmap = ufs_bmap,
3001 .vop_fplookup_vexec = ufs_fplookup_vexec,
3002 .vop_fplookup_symlink = VOP_EAGAIN,
3003 .vop_cachedlookup = ufs_lookup,
3004 .vop_close = ufs_close,
3005 .vop_create = ufs_create,
3006 .vop_stat = ufs_stat,
3007 .vop_getattr = ufs_getattr,
3008 .vop_inactive = ufs_inactive,
3009 .vop_ioctl = ufs_ioctl,
3010 .vop_link = ufs_link,
3011 .vop_lookup = vfs_cache_lookup,
3012 .vop_mmapped = ufs_mmapped,
3013 .vop_mkdir = ufs_mkdir,
3014 .vop_mknod = ufs_mknod,
3015 .vop_need_inactive = ufs_need_inactive,
3016 .vop_open = ufs_open,
3017 .vop_pathconf = ufs_pathconf,
3018 .vop_poll = vop_stdpoll,
3019 .vop_print = ufs_print,
3020 .vop_read_pgcache = ufs_read_pgcache,
3021 .vop_readdir = ufs_readdir,
3022 .vop_readlink = ufs_readlink,
3023 .vop_reclaim = ufs_reclaim,
3024 .vop_remove = ufs_remove,
3025 .vop_rename = ufs_rename,
3026 .vop_rmdir = ufs_rmdir,
3027 .vop_setattr = ufs_setattr,
3029 .vop_setlabel = vop_stdsetlabel_ea,
3031 .vop_strategy = ufs_strategy,
3032 .vop_symlink = ufs_symlink,
3033 .vop_whiteout = ufs_whiteout,
3035 .vop_getextattr = ufs_getextattr,
3036 .vop_deleteextattr = ufs_deleteextattr,
3037 .vop_setextattr = ufs_setextattr,
3040 .vop_getacl = ufs_getacl,
3041 .vop_setacl = ufs_setacl,
3042 .vop_aclcheck = ufs_aclcheck,
3045 VFS_VOP_VECTOR_REGISTER(ufs_vnodeops);
3047 struct vop_vector ufs_fifoops = {
3048 .vop_default = &fifo_specops,
3049 .vop_fsync = VOP_PANIC,
3050 .vop_accessx = ufs_accessx,
3051 .vop_close = ufsfifo_close,
3052 .vop_getattr = ufs_getattr,
3053 .vop_inactive = ufs_inactive,
3054 .vop_pathconf = ufs_pathconf,
3055 .vop_print = ufs_print,
3056 .vop_read = VOP_PANIC,
3057 .vop_reclaim = ufs_reclaim,
3058 .vop_setattr = ufs_setattr,
3060 .vop_setlabel = vop_stdsetlabel_ea,
3062 .vop_write = VOP_PANIC,
3064 .vop_getextattr = ufs_getextattr,
3065 .vop_deleteextattr = ufs_deleteextattr,
3066 .vop_setextattr = ufs_setextattr,
3069 .vop_getacl = ufs_getacl,
3070 .vop_setacl = ufs_setacl,
3071 .vop_aclcheck = ufs_aclcheck,
3074 VFS_VOP_VECTOR_REGISTER(ufs_fifoops);