2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94
35 * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
38 #include <sys/param.h>
39 #include <sys/systm.h>
42 #include <sys/sysent.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mountctl.h>
46 #include <sys/sysmsg.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/fcntl.h>
51 #include <sys/linker.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
58 #include <sys/namei.h>
59 #include <sys/nlookup.h>
60 #include <sys/dirent.h>
61 #include <sys/extattr.h>
62 #include <sys/spinlock.h>
63 #include <sys/kern_syscall.h>
64 #include <sys/objcache.h>
65 #include <sys/sysctl.h>
68 #include <sys/file2.h>
69 #include <sys/spinlock2.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
75 #include <machine/limits.h>
76 #include <machine/stdarg.h>
78 static void mount_warning(struct mount *mp, const char *ctl, ...)
80 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
81 static int checkvp_chdir (struct vnode *vn, struct thread *td);
82 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
83 static int get_fspriv(const char *);
84 static int chroot_refuse_vdir_fds (thread_t td, struct filedesc *fdp);
85 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
86 static int getutimes (struct timeval *, struct timespec *);
87 static int getutimens (const struct timespec *, struct timespec *, int *);
88 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
89 static int setfmode (struct vnode *, int);
90 static int setfflags (struct vnode *, u_long);
91 static int setutimes (struct vnode *, struct vattr *,
92 const struct timespec *, int);
94 static int usermount = 0; /* if 1, non-root can mount fs. */
95 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
96 "Allow non-root users to mount filesystems");
98 static int debug_unmount = 0; /* if 1 loop until unmount success */
99 SYSCTL_INT(_vfs, OID_AUTO, debug_unmount, CTLFLAG_RW, &debug_unmount, 0,
100 "Stall failed unmounts in loop");
102 static struct krate krate_rename = { 1 };
105 * Virtual File System System Calls
109 * Mount a file system.
111 * mount_args(char *type, char *path, int flags, caddr_t data)
116 sys_mount(struct sysmsg *sysmsg, const struct mount_args *uap)
118 struct thread *td = curthread;
121 struct mount *mp, *nullmp;
122 struct vfsconf *vfsp;
123 int error, flag = 0, flag2 = 0;
126 int flags = uap->flags;
128 struct nlookupdata nd;
129 char fstypename[MFSNAMELEN];
134 /* We do not allow user mounts inside a jail for now */
135 if (usermount && jailed(cred)) {
141 * Extract the file system type. We need to know this early, to take
142 * appropriate actions for jails and nullfs mounts.
144 if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0)
148 * Select the correct priv according to the file system type.
150 priv = get_fspriv(fstypename);
152 if (usermount == 0 && (error = priv_check(td, priv)))
156 * Do not allow NFS export by non-root users.
158 if (flags & MNT_EXPORTED) {
159 error = priv_check(td, priv);
164 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
166 if (priv_check(td, priv))
167 flags |= MNT_NOSUID | MNT_NODEV;
170 * Lookup the requested path and extract the nch and vnode.
172 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
174 if ((error = nlookup(&nd)) == 0) {
175 if (nd.nl_nch.ncp->nc_vp == NULL)
185 * If the target filesystem is resolved via a nullfs mount, then
186 * nd.nl_nch.mount will be pointing to the nullfs mount structure
187 * instead of the target file system. We need it in case we are
190 nullmp = nd.nl_nch.mount;
193 * Extract the locked+refd ncp and cleanup the nd structure
196 cache_zero(&nd.nl_nch);
199 if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
200 (mp = cache_findmount(&nch)) != NULL) {
209 * now we have the locked ref'd nch and unreferenced vnode.
212 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
219 * Now we have an unlocked ref'd nch and a locked ref'd vp
221 if (flags & MNT_UPDATE) {
222 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
229 if (strncmp(fstypename, "null", 5) == 0) {
237 flag2 = mp->mnt_kern_flag;
239 * We only allow the filesystem to be reloaded if it
240 * is currently mounted read-only.
242 if ((flags & MNT_RELOAD) &&
243 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
246 error = EOPNOTSUPP; /* Needs translation */
250 * Only root, or the user that did the original mount is
251 * permitted to update it.
253 if (mp->mnt_stat.f_owner != cred->cr_uid &&
254 (error = priv_check(td, priv))) {
259 if (vfs_busy(mp, LK_NOWAIT)) {
272 mp->mnt_flag |= flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
273 lwkt_gettoken(&mp->mnt_token);
280 * If the user is not root, ensure that they own the directory
281 * onto which we are attempting to mount.
283 if ((error = VOP_GETATTR(vp, &va)) ||
284 (va.va_uid != cred->cr_uid &&
285 (error = priv_check(td, priv)))) {
290 if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
295 if (vp->v_type != VDIR) {
301 if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
307 vfsp = vfsconf_find_by_name(fstypename);
311 /* Only load modules for root (very important!) */
312 if ((error = priv_check(td, PRIV_ROOT)) != 0) {
317 error = linker_load_file(fstypename, &lf);
318 if (error || lf == NULL) {
326 /* lookup again, see if the VFS was loaded */
327 vfsp = vfsconf_find_by_name(fstypename);
330 linker_file_unload(lf);
345 * Allocate and initialize the filesystem.
347 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
348 mount_init(mp, vfsp->vfc_vfsops);
349 vfs_busy(mp, LK_NOWAIT);
351 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT;
352 vfsp->vfc_refcount++;
353 mp->mnt_stat.f_type = vfsp->vfc_typenum;
354 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
355 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
356 mp->mnt_stat.f_owner = cred->cr_uid;
357 lwkt_gettoken(&mp->mnt_token);
361 * (per-mount token acquired at this point)
363 * Set the mount level flags.
365 if (flags & MNT_RDONLY)
366 mp->mnt_flag |= MNT_RDONLY;
367 else if (mp->mnt_flag & MNT_RDONLY)
368 mp->mnt_kern_flag |= MNTK_WANTRDWR;
369 mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
370 MNT_SYNCHRONOUS | MNT_ASYNC | MNT_NOATIME |
371 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
372 MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
374 mp->mnt_flag |= flags & (MNT_NOSUID | MNT_NOEXEC |
375 MNT_NODEV | MNT_SYNCHRONOUS | MNT_ASYNC | MNT_FORCE |
376 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
377 MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
381 * Pre-set the mount's ALL_MPSAFE flags if specified in the vfsconf.
382 * This way the initial VFS_MOUNT() call will also be MPSAFE.
384 if (vfsp->vfc_flags & VFCF_MPSAFE)
385 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
388 * Mount the filesystem.
389 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
392 if (mp->mnt_flag & MNT_UPDATE) {
393 error = VFS_MOUNT(mp, uap->path, uap->data, cred);
394 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
395 mp->mnt_flag &= ~MNT_RDONLY;
396 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
397 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
400 mp->mnt_kern_flag = flag2;
402 lwkt_reltoken(&mp->mnt_token);
408 mp->mnt_ncmounton = nch;
409 error = VFS_MOUNT(mp, uap->path, uap->data, cred);
410 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
413 * Put the new filesystem on the mount list after root. The mount
414 * point gets its own mnt_ncmountpt (unless the VFS already set one
415 * up) which represents the root of the mount. The lookup code
416 * detects the mount point going forward and checks the root of
417 * the mount going backwards.
419 * It is not necessary to invalidate or purge the vnode underneath
420 * because elements under the mount will be given their own glue
424 if (mp->mnt_ncmountpt.ncp == NULL) {
426 * Allocate, then unlock, but leave the ref intact.
427 * This is the mnt_refs (1) that we will retain
428 * through to the unmount.
430 cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
431 cache_unlock(&mp->mnt_ncmountpt);
435 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
437 cache_ismounting(mp);
438 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
440 mountlist_insert(mp, MNTINS_LAST);
442 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
443 error = vfs_allocate_syncvnode(mp);
444 lwkt_reltoken(&mp->mnt_token);
446 error = VFS_START(mp, 0);
448 KNOTE(&fs_klist, VQ_MOUNT);
450 bzero(&mp->mnt_ncmounton, sizeof(mp->mnt_ncmounton));
451 vn_syncer_thr_stop(mp);
452 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
453 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
454 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
455 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
456 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
458 crfree(mp->mnt_cred);
461 mp->mnt_vfc->vfc_refcount--;
462 lwkt_reltoken(&mp->mnt_token);
473 * Scan all active processes to see if any of them have a current
474 * or root directory onto which the new filesystem has just been
475 * mounted. If so, replace them with the new mount point.
477 * Both old_nch and new_nch are ref'd on call but not locked.
478 * new_nch must be temporarily locked so it can be associated with the
479 * vnode representing the root of the mount point.
481 struct checkdirs_info {
482 struct nchandle old_nch;
483 struct nchandle new_nch;
484 struct vnode *old_vp;
485 struct vnode *new_vp;
488 static int checkdirs_callback(struct proc *p, void *data);
491 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
493 struct checkdirs_info info;
499 * If the old mount point's vnode has a usecount of 1, it is not
500 * being held as a descriptor anywhere.
502 olddp = old_nch->ncp->nc_vp;
503 if (olddp == NULL || VREFCNT(olddp) == 1)
507 * Force the root vnode of the new mount point to be resolved
508 * so we can update any matching processes.
511 if (VFS_ROOT(mp, &newdp))
512 panic("mount: lost mount");
515 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
516 cache_setunresolved(new_nch);
517 cache_setvp(new_nch, newdp);
518 cache_unlock(new_nch);
521 * Special handling of the root node
523 if (rootvnode == olddp) {
525 vfs_cache_setroot(newdp, cache_hold(new_nch));
529 * Pass newdp separately so the callback does not have to access
530 * it via new_nch->ncp->nc_vp.
532 info.old_nch = *old_nch;
533 info.new_nch = *new_nch;
535 allproc_scan(checkdirs_callback, &info, 0);
540 * NOTE: callback is not MP safe because the scanned process's filedesc
541 * structure can be ripped out from under us, amoung other things.
544 checkdirs_callback(struct proc *p, void *data)
546 struct checkdirs_info *info = data;
547 struct filedesc *fdp;
548 struct nchandle ncdrop1;
549 struct nchandle ncdrop2;
550 struct vnode *vprele1;
551 struct vnode *vprele2;
553 if ((fdp = p->p_fd) != NULL) {
554 cache_zero(&ncdrop1);
555 cache_zero(&ncdrop2);
560 * MPUNSAFE - XXX fdp can be pulled out from under a
563 * A shared filedesc is ok, we don't have to copy it
564 * because we are making this change globally.
566 spin_lock(&fdp->fd_spin);
567 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
568 fdp->fd_ncdir.ncp == info->old_nch.ncp) {
569 vprele1 = fdp->fd_cdir;
571 fdp->fd_cdir = info->new_vp;
572 ncdrop1 = fdp->fd_ncdir;
573 cache_copy(&info->new_nch, &fdp->fd_ncdir);
575 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
576 fdp->fd_nrdir.ncp == info->old_nch.ncp) {
577 vprele2 = fdp->fd_rdir;
579 fdp->fd_rdir = info->new_vp;
580 ncdrop2 = fdp->fd_nrdir;
581 cache_copy(&info->new_nch, &fdp->fd_nrdir);
583 spin_unlock(&fdp->fd_spin);
585 cache_drop(&ncdrop1);
587 cache_drop(&ncdrop2);
597 * Unmount a file system.
599 * Note: unmount takes a path to the vnode mounted on as argument,
600 * not special file (as before).
602 * umount_args(char *path, int flags)
607 sys_unmount(struct sysmsg *sysmsg, const struct unmount_args *uap)
609 struct thread *td = curthread;
610 struct proc *p __debugvar = td->td_proc;
611 struct mount *mp = NULL;
612 struct nlookupdata nd;
613 char fstypename[MFSNAMELEN];
622 /* We do not allow user umounts inside a jail for now */
623 if (usermount && jailed(cred)) {
628 error = nlookup_init(&nd, uap->path, UIO_USERSPACE,
629 NLC_FOLLOW | NLC_IGNBADDIR);
631 error = nlookup(&nd);
635 mp = nd.nl_nch.mount;
637 /* Figure out the fsname in order to select proper privs */
638 ksnprintf(fstypename, MFSNAMELEN, "%s", mp->mnt_vfc->vfc_name);
639 priv = get_fspriv(fstypename);
641 if (usermount == 0 && (error = priv_check(td, priv))) {
647 * Only root, or the user that did the original mount is
648 * permitted to unmount this filesystem.
650 if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
651 (error = priv_check(td, priv)))
655 * Don't allow unmounting the root file system.
657 if (mp->mnt_flag & MNT_ROOTFS) {
663 * Must be the root of the filesystem
665 if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
670 /* Check if this mount belongs to this prison */
671 if (jailed(cred) && mp->mnt_cred && (!mp->mnt_cred->cr_prison ||
672 mp->mnt_cred->cr_prison != cred->cr_prison)) {
673 kprintf("mountpoint %s does not belong to this jail\n",
680 * If no error try to issue the unmount. We lose our cache
681 * ref when we call nlookup_done so we must hold the mount point
682 * to prevent use-after-free races.
688 error = dounmount(mp, uap->flags, 0);
698 * Do the actual file system unmount (interlocked against the mountlist
699 * token and mp->mnt_token).
702 dounmount_interlock(struct mount *mp)
704 if (mp->mnt_kern_flag & MNTK_UNMOUNT)
706 mp->mnt_kern_flag |= MNTK_UNMOUNT;
711 unmount_allproc_cb(struct proc *p, void *arg)
715 if (p->p_textnch.ncp == NULL)
718 mp = (struct mount *)arg;
719 if (p->p_textnch.mount == mp)
720 cache_drop(&p->p_textnch);
726 * The guts of the unmount code. The mount owns one ref and one hold
727 * count. If we successfully interlock the unmount, those refs are ours.
728 * (The ref is from mnt_ncmountpt).
730 * When halting we shortcut certain mount types such as devfs by not actually
731 * issuing the VFS_SYNC() or VFS_UNMOUNT(). They are still disconnected
732 * from the mountlist so higher-level filesytems can unmount cleanly.
734 * The mount types that allow QUICKHALT are: devfs, tmpfs, procfs.
737 dounmount(struct mount *mp, int flags, int halting)
739 struct namecache *ncp;
750 lwkt_gettoken(&mp->mnt_token);
753 * When halting, certain mount points can essentially just
754 * be unhooked and otherwise ignored.
756 if (halting && (mp->mnt_kern_flag & MNTK_QUICKHALT)) {
765 * Exclusive access for unmounting purposes.
767 if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
771 * We now 'own' the last mp->mnt_refs
773 * Allow filesystems to detect that a forced unmount is in progress.
775 if (flags & MNT_FORCE)
776 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
777 lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK);
778 error = lockmgr(&mp->mnt_lock, lflags);
780 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
781 if (mp->mnt_kern_flag & MNTK_MWAIT) {
782 mp->mnt_kern_flag &= ~MNTK_MWAIT;
788 if (mp->mnt_flag & MNT_EXPUBLIC)
789 vfs_setpublicfs(NULL, NULL, NULL);
791 vfs_msync(mp, MNT_WAIT);
792 async_flag = mp->mnt_flag & MNT_ASYNC;
793 mp->mnt_flag &=~ MNT_ASYNC;
796 * Decomission our special mnt_syncer vnode. This also stops
797 * the vnlru code. If we are unable to unmount we recommission
800 * Then sync the filesystem.
802 if ((vp = mp->mnt_syncer) != NULL) {
803 mp->mnt_syncer = NULL;
804 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
810 * Sync normally-mounted filesystem.
812 if (quickhalt == 0) {
813 if ((mp->mnt_flag & MNT_RDONLY) == 0)
814 VFS_SYNC(mp, MNT_WAIT);
818 * nchandle records ref the mount structure. Expect a count of 1
819 * (our mount->mnt_ncmountpt).
821 * Scans can get temporary refs on a mountpoint (thought really
822 * heavy duty stuff like cache_findmount() do not).
824 for (retry = 0; (retry < 10 || debug_unmount); ++retry) {
826 * Invalidate the namecache topology under the mount.
827 * nullfs mounts alias a real mount's namecache topology
828 * and it should not be invalidated in that case.
830 if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
831 cache_lock(&mp->mnt_ncmountpt);
832 cache_inval(&mp->mnt_ncmountpt,
833 CINV_DESTROY|CINV_CHILDREN);
834 cache_unlock(&mp->mnt_ncmountpt);
840 cache_unmounting(mp);
841 if (mp->mnt_refs != 1)
842 cache_clearmntcache(mp);
845 * Break out if we are good. Don't count ncp refs if the
848 ncp = (mp->mnt_kern_flag & MNTK_NCALIASED) ?
849 NULL : mp->mnt_ncmountpt.ncp;
850 if (mp->mnt_refs == 1 &&
851 (ncp == NULL || (ncp->nc_refs == 1 &&
852 TAILQ_FIRST(&ncp->nc_list) == NULL))) {
857 * If forcing the unmount, clean out any p->p_textnch
858 * nchandles that match this mount.
860 if (flags & MNT_FORCE)
861 allproc_scan(&unmount_allproc_cb, mp, 0);
866 tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1);
867 if ((retry & 15) == 15) {
869 "(%p) debug - retry %d, "
870 "%d namecache refs, %d mount refs",
872 (ncp ? ncp->nc_refs - 1 : 0),
878 ncp = (mp->mnt_kern_flag & MNTK_NCALIASED) ?
879 NULL : mp->mnt_ncmountpt.ncp;
880 if (mp->mnt_refs != 1 ||
881 (ncp != NULL && (ncp->nc_refs != 1 ||
882 TAILQ_FIRST(&ncp->nc_list)))) {
884 "(%p): %d namecache refs, %d mount refs "
887 (ncp ? ncp->nc_refs - 1 : 0),
889 if (flags & MNT_FORCE) {
891 mount_warning(mp, "forcing unmount\n");
898 * So far so good, sync the filesystem once more and
899 * call the VFS unmount code if the sync succeeds.
901 if (error == 0 && quickhalt == 0) {
902 if (mp->mnt_flag & MNT_RDONLY) {
903 error = VFS_UNMOUNT(mp, flags);
905 error = VFS_SYNC(mp, MNT_WAIT);
906 if (error == 0 || /* no error */
907 error == EOPNOTSUPP || /* no sync avail */
908 (flags & MNT_FORCE)) { /* force anyway */
909 error = VFS_UNMOUNT(mp, flags);
914 "(%p) unmount: vfs refused to unmount, "
921 * If an error occurred we can still recover, restoring the
922 * syncer vnode and misc flags.
925 if (mp->mnt_syncer == NULL && hadsyncer)
926 vfs_allocate_syncvnode(mp);
927 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
928 mp->mnt_flag |= async_flag;
929 lockmgr(&mp->mnt_lock, LK_RELEASE);
930 if (mp->mnt_kern_flag & MNTK_MWAIT) {
931 mp->mnt_kern_flag &= ~MNTK_MWAIT;
937 * Clean up any journals still associated with the mount after
938 * filesystem activity has ceased.
940 journal_remove_all_journals(mp,
941 ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
943 mountlist_remove(mp);
946 * Remove any installed vnode ops here so the individual VFSs don't
949 * mnt_refs should go to zero when we scrap mnt_ncmountpt.
951 * When quickhalting we have to keep these intact because the
952 * underlying vnodes have not been destroyed, and some might be
955 if (quickhalt == 0) {
956 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
957 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
958 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
959 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
960 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
963 if (mp->mnt_ncmountpt.ncp != NULL) {
964 nch = mp->mnt_ncmountpt;
965 cache_zero(&mp->mnt_ncmountpt);
966 cache_clrmountpt(&nch);
969 if (mp->mnt_ncmounton.ncp != NULL) {
970 cache_unmounting(mp);
971 nch = mp->mnt_ncmounton;
972 cache_zero(&mp->mnt_ncmounton);
973 cache_clrmountpt(&nch);
978 crfree(mp->mnt_cred);
982 mp->mnt_vfc->vfc_refcount--;
985 * If not quickhalting the mount, we expect there to be no
988 if (quickhalt == 0 && !TAILQ_EMPTY(&mp->mnt_nvnodelist))
989 panic("unmount: dangling vnode");
994 lockmgr(&mp->mnt_lock, LK_RELEASE);
995 if (mp->mnt_kern_flag & MNTK_MWAIT) {
996 mp->mnt_kern_flag &= ~MNTK_MWAIT;
1001 * If we reach here and freeok != 0 we must free the mount.
1002 * mnt_refs should already have dropped to 0, so if it is not
1003 * zero we must cycle the caches and wait.
1005 * When we are satisfied that the mount has disconnected we can
1006 * drop the hold on the mp that represented the mount (though the
1007 * caller might actually have another, so the caller's drop may
1008 * do the actual free).
1011 if (mp->mnt_refs > 0)
1012 cache_clearmntcache(mp);
1013 while (mp->mnt_refs > 0) {
1014 cache_unmounting(mp);
1016 tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1);
1017 cache_clearmntcache(mp);
1019 lwkt_reltoken(&mp->mnt_token);
1023 cache_clearmntcache(mp);
1026 KNOTE(&fs_klist, VQ_UNMOUNT);
1029 lwkt_reltoken(&mp->mnt_token);
1035 mount_warning(struct mount *mp, const char *ctl, ...)
1041 __va_start(va, ctl);
1042 if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL,
1043 &ptr, &buf, 0) == 0) {
1044 kprintf("unmount(%s): ", ptr);
1049 kprintf("unmount(%p", mp);
1050 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
1051 kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
1060 * Shim cache_fullpath() to handle the case where a process is chrooted into
1061 * a subdirectory of a mount. In this case if the root mount matches the
1062 * process root directory's mount we have to specify the process's root
1063 * directory instead of the mount point, because the mount point might
1064 * be above the root directory.
1068 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
1070 struct nchandle *nch;
1072 if (p && p->p_fd->fd_nrdir.mount == mp)
1073 nch = &p->p_fd->fd_nrdir;
1075 nch = &mp->mnt_ncmountpt;
1076 return(cache_fullpath(p, nch, NULL, rb, fb, 0));
1080 * Sync each mounted filesystem.
1084 static int syncprt = 0;
1085 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
1088 static int sync_callback(struct mount *mp, void *data);
1091 sys_sync(struct sysmsg *sysmsg, const struct sync_args *uap)
1093 mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
1099 sync_callback(struct mount *mp, void *data __unused)
1103 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1104 lwkt_gettoken(&mp->mnt_token);
1105 asyncflag = mp->mnt_flag & MNT_ASYNC;
1106 mp->mnt_flag &= ~MNT_ASYNC;
1107 lwkt_reltoken(&mp->mnt_token);
1108 vfs_msync(mp, MNT_NOWAIT);
1109 VFS_SYNC(mp, MNT_NOWAIT);
1110 lwkt_gettoken(&mp->mnt_token);
1111 mp->mnt_flag |= asyncflag;
1112 lwkt_reltoken(&mp->mnt_token);
1117 /* XXX PRISON: could be per prison flag */
1118 static int prison_quotas;
1120 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
1124 * quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
1126 * Change filesystem quotas.
1131 sys_quotactl(struct sysmsg *sysmsg, const struct quotactl_args *uap)
1133 struct nlookupdata nd;
1139 if (td->td_ucred->cr_prison && !prison_quotas) {
1144 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1146 error = nlookup(&nd);
1148 mp = nd.nl_nch.mount;
1149 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
1150 uap->arg, nd.nl_cred);
1158 * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
1159 * void *buf, int buflen)
1161 * This function operates on a mount point and executes the specified
1162 * operation using the specified control data, and possibly returns data.
1164 * The actual number of bytes stored in the result buffer is returned, 0
1165 * if none, otherwise an error is returned.
1170 sys_mountctl(struct sysmsg *sysmsg, const struct mountctl_args *uap)
1172 struct thread *td = curthread;
1180 * Sanity and permissions checks. We must be root.
1182 if (td->td_ucred->cr_prison != NULL)
1184 if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
1185 (error = priv_check(td, PRIV_ROOT)) != 0)
1189 * Argument length checks
1191 if (uap->ctllen < 0 || uap->ctllen > 1024)
1193 if (uap->buflen < 0 || uap->buflen > 16 * 1024)
1195 if (uap->path == NULL)
1199 * Allocate the necessary buffers and copyin data
1201 path = objcache_get(namei_oc, M_WAITOK);
1202 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1207 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1208 error = copyin(uap->ctl, ctl, uap->ctllen);
1213 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1216 * Validate the descriptor
1219 fp = holdfp(td, uap->fd, -1);
1229 * Execute the internal kernel function and clean up.
1231 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen,
1232 buf, uap->buflen, &sysmsg->sysmsg_result);
1234 dropfp(td, uap->fd, fp);
1235 if (error == 0 && sysmsg->sysmsg_result > 0)
1236 error = copyout(buf, uap->buf, sysmsg->sysmsg_result);
1239 objcache_put(namei_oc, path);
1248 * Execute a mount control operation by resolving the path to a mount point
1249 * and calling vop_mountctl().
1251 * Use the mount point from the nch instead of the vnode so nullfs mounts
1252 * can properly spike the VOP.
1255 kern_mountctl(const char *path, int op, struct file *fp,
1256 const void *ctl, int ctllen,
1257 void *buf, int buflen, int *res)
1260 struct nlookupdata nd;
1261 struct nchandle nch;
1267 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1270 error = nlookup(&nd);
1275 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1282 * Yes, all this is needed to use the nch.mount below, because
1283 * we must maintain a ref on the mount to avoid ripouts (e.g.
1284 * due to heavy mount/unmount use by synth or poudriere).
1287 cache_zero(&nd.nl_nch);
1295 * Must be the root of the filesystem
1297 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1302 if (mp == NULL || mp->mnt_kern_flag & MNTK_UNMOUNT) {
1303 kprintf("kern_mountctl: Warning, \"%s\" racing unmount\n",
1309 error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1318 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1320 struct thread *td = curthread;
1321 struct proc *p = td->td_proc;
1324 char *fullpath, *freepath;
1327 if ((error = nlookup(nd)) != 0)
1329 mp = nd->nl_nch.mount;
1333 * Ignore refresh error, user should have visibility.
1334 * This can happen if a NFS mount goes bad (e.g. server
1335 * revokes perms or goes down).
1337 error = VFS_STATFS(mp, sp, nd->nl_cred);
1340 error = mount_path(p, mp, &fullpath, &freepath);
1343 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1344 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1345 kfree(freepath, M_TEMP);
1347 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1348 bcopy(sp, buf, sizeof(*buf));
1349 /* Only root should have access to the fsid's. */
1350 if (priv_check(td, PRIV_ROOT))
1351 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1356 * statfs_args(char *path, struct statfs *buf)
1358 * Get filesystem statistics.
1361 sys_statfs(struct sysmsg *sysmsg, const struct statfs_args *uap)
1363 struct nlookupdata nd;
1367 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1369 error = kern_statfs(&nd, &buf);
1372 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1377 kern_fstatfs(int fd, struct statfs *buf)
1379 struct thread *td = curthread;
1380 struct proc *p = td->td_proc;
1384 char *fullpath, *freepath;
1388 if ((error = holdvnode(td, fd, &fp)) != 0)
1392 * Try to use mount info from any overlays rather than the
1393 * mount info for the underlying vnode, otherwise we will
1394 * fail when operating on null-mounted paths inside a chroot.
1396 if ((mp = fp->f_nchandle.mount) == NULL)
1397 mp = ((struct vnode *)fp->f_data)->v_mount;
1402 if (fp->f_cred == NULL) {
1408 * Ignore refresh error, user should have visibility.
1409 * This can happen if a NFS mount goes bad (e.g. server
1410 * revokes perms or goes down).
1413 error = VFS_STATFS(mp, sp, fp->f_cred);
1415 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1417 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1418 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1419 kfree(freepath, M_TEMP);
1421 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1422 bcopy(sp, buf, sizeof(*buf));
1424 /* Only root should have access to the fsid's. */
1425 if (priv_check(td, PRIV_ROOT))
1426 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1434 * fstatfs_args(int fd, struct statfs *buf)
1436 * Get filesystem statistics.
1439 sys_fstatfs(struct sysmsg *sysmsg, const struct fstatfs_args *uap)
1444 error = kern_fstatfs(uap->fd, &buf);
1447 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1452 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1458 if ((error = nlookup(nd)) != 0)
1460 mp = nd->nl_nch.mount;
1461 sp = &mp->mnt_vstat;
1462 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1466 if (mp->mnt_flag & MNT_RDONLY)
1467 sp->f_flag |= ST_RDONLY;
1468 if (mp->mnt_flag & MNT_NOSUID)
1469 sp->f_flag |= ST_NOSUID;
1470 bcopy(sp, buf, sizeof(*buf));
1475 * statfs_args(char *path, struct statfs *buf)
1477 * Get filesystem statistics.
1480 sys_statvfs(struct sysmsg *sysmsg, const struct statvfs_args *uap)
1482 struct nlookupdata nd;
1486 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1488 error = kern_statvfs(&nd, &buf);
1491 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1496 kern_fstatvfs(int fd, struct statvfs *buf)
1498 struct thread *td = curthread;
1504 if ((error = holdvnode(td, fd, &fp)) != 0)
1506 if ((mp = fp->f_nchandle.mount) == NULL)
1507 mp = ((struct vnode *)fp->f_data)->v_mount;
1512 if (fp->f_cred == NULL) {
1516 sp = &mp->mnt_vstat;
1517 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1521 if (mp->mnt_flag & MNT_RDONLY)
1522 sp->f_flag |= ST_RDONLY;
1523 if (mp->mnt_flag & MNT_NOSUID)
1524 sp->f_flag |= ST_NOSUID;
1526 bcopy(sp, buf, sizeof(*buf));
1534 * fstatfs_args(int fd, struct statfs *buf)
1536 * Get filesystem statistics.
1539 sys_fstatvfs(struct sysmsg *sysmsg, const struct fstatvfs_args *uap)
1544 error = kern_fstatvfs(uap->fd, &buf);
1547 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1552 * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1554 * Get statistics on all filesystems.
1557 struct getfsstat_info {
1558 struct statfs *sfsp;
1566 static int getfsstat_callback(struct mount *, void *);
1569 sys_getfsstat(struct sysmsg *sysmsg, const struct getfsstat_args *uap)
1571 struct thread *td = curthread;
1572 struct getfsstat_info info;
1574 bzero(&info, sizeof(info));
1576 info.maxcount = uap->bufsize / sizeof(struct statfs);
1577 info.sfsp = uap->buf;
1579 info.flags = uap->flags;
1582 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1583 if (info.sfsp && info.count > info.maxcount)
1584 sysmsg->sysmsg_result = info.maxcount;
1586 sysmsg->sysmsg_result = info.count;
1587 return (info.error);
1591 getfsstat_callback(struct mount *mp, void *data)
1593 struct getfsstat_info *info = data;
1599 if (info->td->td_proc && !chroot_visible_mnt(mp, info->td->td_proc))
1602 if (info->sfsp && info->count < info->maxcount) {
1606 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1607 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1608 * overrides MNT_WAIT.
1610 * Ignore refresh error, user should have visibility.
1611 * This can happen if a NFS mount goes bad (e.g. server
1612 * revokes perms or goes down).
1614 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1615 (info->flags & MNT_WAIT)) &&
1616 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1619 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1621 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1623 info->error = error;
1626 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1627 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1628 kfree(freepath, M_TEMP);
1630 error = copyout(sp, info->sfsp, sizeof(*sp));
1632 info->error = error;
1642 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1643 long bufsize, int flags)
1645 * Get statistics on all filesystems.
1648 struct getvfsstat_info {
1649 struct statfs *sfsp;
1650 struct statvfs *vsfsp;
1658 static int getvfsstat_callback(struct mount *, void *);
1661 sys_getvfsstat(struct sysmsg *sysmsg, const struct getvfsstat_args *uap)
1663 struct thread *td = curthread;
1664 struct getvfsstat_info info;
1666 bzero(&info, sizeof(info));
1668 info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1669 info.sfsp = uap->buf;
1670 info.vsfsp = uap->vbuf;
1672 info.flags = uap->flags;
1675 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1676 if (info.vsfsp && info.count > info.maxcount)
1677 sysmsg->sysmsg_result = info.maxcount;
1679 sysmsg->sysmsg_result = info.count;
1680 return (info.error);
1684 getvfsstat_callback(struct mount *mp, void *data)
1686 struct getvfsstat_info *info = data;
1688 struct statvfs *vsp;
1693 if (info->td->td_proc && !chroot_visible_mnt(mp, info->td->td_proc))
1696 if (info->vsfsp && info->count < info->maxcount) {
1698 vsp = &mp->mnt_vstat;
1701 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1702 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1703 * overrides MNT_WAIT.
1705 * Ignore refresh error, user should have visibility.
1706 * This can happen if a NFS mount goes bad (e.g. server
1707 * revokes perms or goes down).
1709 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1710 (info->flags & MNT_WAIT)) &&
1711 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1714 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1716 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1717 (info->flags & MNT_WAIT)) &&
1718 (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1722 if (mp->mnt_flag & MNT_RDONLY)
1723 vsp->f_flag |= ST_RDONLY;
1724 if (mp->mnt_flag & MNT_NOSUID)
1725 vsp->f_flag |= ST_NOSUID;
1727 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1729 info->error = error;
1732 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1733 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1734 kfree(freepath, M_TEMP);
1736 error = copyout(sp, info->sfsp, sizeof(*sp));
1738 error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1740 info->error = error;
1752 * fchdir_args(int fd)
1754 * Change current working directory to a given file descriptor.
1757 sys_fchdir(struct sysmsg *sysmsg, const struct fchdir_args *uap)
1759 struct thread *td = curthread;
1760 struct proc *p = td->td_proc;
1761 struct filedesc *fdp = p->p_fd;
1762 struct vnode *vp, *ovp;
1765 struct nchandle nch, onch, tnch;
1768 if ((error = holdvnode(td, uap->fd, &fp)) != 0)
1770 lwkt_gettoken(&p->p_token);
1771 vp = (struct vnode *)fp->f_data;
1773 vn_lock(vp, LK_SHARED | LK_RETRY);
1774 if (fp->f_nchandle.ncp == NULL)
1777 error = checkvp_chdir(vp, td);
1782 cache_copy(&fp->f_nchandle, &nch);
1785 * If the ncp has become a mount point, traverse through
1789 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1790 (mp = cache_findmount(&nch)) != NULL
1792 error = nlookup_mp(mp, &tnch);
1794 cache_unlock(&tnch); /* leave ref intact */
1796 vp = tnch.ncp->nc_vp;
1797 error = vget(vp, LK_SHARED);
1798 KKASSERT(error == 0);
1802 cache_dropmount(mp);
1805 spin_lock(&fdp->fd_spin);
1807 onch = fdp->fd_ncdir;
1809 fdp->fd_ncdir = nch;
1810 spin_unlock(&fdp->fd_spin);
1811 vn_unlock(vp); /* leave ref intact */
1820 lwkt_reltoken(&p->p_token);
1825 kern_chdir(struct nlookupdata *nd)
1827 struct thread *td = curthread;
1828 struct proc *p = td->td_proc;
1829 struct filedesc *fdp = p->p_fd;
1830 struct vnode *vp, *ovp;
1831 struct nchandle onch;
1834 nd->nl_flags |= NLC_SHAREDLOCK;
1835 if ((error = nlookup(nd)) != 0)
1837 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1839 if ((error = vget(vp, LK_SHARED)) != 0)
1842 lwkt_gettoken(&p->p_token);
1843 error = checkvp_chdir(vp, td);
1846 spin_lock(&fdp->fd_spin);
1848 onch = fdp->fd_ncdir;
1849 fdp->fd_ncdir = nd->nl_nch;
1851 spin_unlock(&fdp->fd_spin);
1852 cache_unlock(&nd->nl_nch); /* leave reference intact */
1855 cache_zero(&nd->nl_nch);
1859 lwkt_reltoken(&p->p_token);
1864 * chdir_args(char *path)
1866 * Change current working directory (``.'').
1869 sys_chdir(struct sysmsg *sysmsg, const struct chdir_args *uap)
1871 struct nlookupdata nd;
1874 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1876 error = kern_chdir(&nd);
1882 * Helper function for raised chroot(2) security function: Refuse if
1883 * any filedescriptors are open directories.
1886 chroot_refuse_vdir_fds(thread_t td, struct filedesc *fdp)
1893 for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1894 if ((error = holdvnode(td, fd, &fp)) != 0)
1896 vp = (struct vnode *)fp->f_data;
1897 if (vp->v_type != VDIR) {
1908 * This sysctl determines if we will allow a process to chroot(2) if it
1909 * has a directory open:
1910 * 0: disallowed for all processes.
1911 * 1: allowed for processes that were not already chroot(2)'ed.
1912 * 2: allowed for all processes.
1915 static int chroot_allow_open_directories = 1;
1917 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1918 &chroot_allow_open_directories, 0, "");
1921 * chroot to the specified namecache entry. We obtain the vp from the
1922 * namecache data. The passed ncp must be locked and referenced and will
1923 * remain locked and referenced on return.
1926 kern_chroot(struct nchandle *nch)
1928 struct thread *td = curthread;
1929 struct proc *p = td->td_proc;
1930 struct filedesc *fdp = p->p_fd;
1935 * Only privileged user can chroot
1937 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1942 * Disallow open directory descriptors (fchdir() breakouts).
1944 if (chroot_allow_open_directories == 0 ||
1945 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1946 if ((error = chroot_refuse_vdir_fds(td, fdp)) != 0)
1949 if ((vp = nch->ncp->nc_vp) == NULL)
1952 if ((error = vget(vp, LK_SHARED)) != 0)
1956 * Check the validity of vp as a directory to change to and
1957 * associate it with rdir/jdir.
1959 error = checkvp_chdir(vp, td);
1960 vn_unlock(vp); /* leave reference intact */
1962 lwkt_gettoken(&p->p_token);
1963 vrele(fdp->fd_rdir);
1964 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */
1965 cache_drop(&fdp->fd_nrdir);
1966 cache_copy(nch, &fdp->fd_nrdir);
1967 if (fdp->fd_jdir == NULL) {
1970 cache_copy(nch, &fdp->fd_njdir);
1972 if ((p->p_flags & P_DIDCHROOT) == 0) {
1973 p->p_flags |= P_DIDCHROOT;
1974 if (p->p_depth <= 65535 - 32)
1977 lwkt_reltoken(&p->p_token);
1985 * chroot_args(char *path)
1987 * Change notion of root (``/'') directory.
1990 sys_chroot(struct sysmsg *sysmsg, const struct chroot_args *uap)
1992 struct thread *td __debugvar = curthread;
1993 struct nlookupdata nd;
1996 KKASSERT(td->td_proc);
1997 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1999 nd.nl_flags |= NLC_EXEC;
2000 error = nlookup(&nd);
2002 error = kern_chroot(&nd.nl_nch);
2009 sys_chroot_kernel(struct sysmsg *sysmsg, const struct chroot_kernel_args *uap)
2011 struct thread *td = curthread;
2012 struct nlookupdata nd;
2013 struct nchandle *nch;
2017 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2021 error = nlookup(&nd);
2027 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
2031 if ((vp = nch->ncp->nc_vp) == NULL) {
2036 if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
2039 vfs_cache_setroot(vp, cache_hold(nch));
2048 * Common routine for chroot and chdir. Given a locked, referenced vnode,
2049 * determine whether it is legal to chdir to the vnode. The vnode's state
2050 * is not changed by this call.
2053 checkvp_chdir(struct vnode *vp, struct thread *td)
2057 if (vp->v_type != VDIR)
2060 error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
2065 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
2067 struct thread *td = curthread;
2068 struct proc *p = td->td_proc;
2069 struct lwp *lp = td->td_lwp;
2070 struct filedesc *fdp = p->p_fd;
2075 int type, indx, error = 0;
2078 if ((oflags & O_ACCMODE) == O_ACCMODE)
2080 flags = FFLAGS(oflags);
2081 error = falloc(lp, &nfp, NULL);
2085 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
2088 * XXX p_dupfd is a real mess. It allows a device to return a
2089 * file descriptor to be duplicated rather then doing the open
2095 * Call vn_open() to do the lookup and assign the vnode to the
2096 * file pointer. vn_open() does not change the ref count on fp
2097 * and the vnode, on success, will be inherited by the file pointer
2100 * Request a shared lock on the vnode if possible.
2102 * When NLC_SHAREDLOCK is set we may still need an exclusive vnode
2103 * lock for O_RDWR opens on executables in order to avoid a VTEXT
2104 * detection race. The NLC_EXCLLOCK_IFEXEC handles this case.
2106 * NOTE: We need a flag to separate terminal vnode locking from
2107 * parent locking. O_CREAT needs parent locking, but O_TRUNC
2108 * and O_RDWR only need to lock the terminal vnode exclusively.
2110 nd->nl_flags |= NLC_LOCKVP;
2111 if ((flags & (O_CREAT|O_TRUNC)) == 0) {
2112 nd->nl_flags |= NLC_SHAREDLOCK;
2114 nd->nl_flags |= NLC_EXCLLOCK_IFEXEC;
2117 error = vn_open(nd, fp, flags, cmode);
2122 * handle special fdopen() case. bleh. dupfdopen() is
2123 * responsible for dropping the old contents of ofiles[indx]
2126 * Note that fsetfd() will add a ref to fp which represents
2127 * the fd_files[] assignment. We must still drop our
2130 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
2131 if (fdalloc(p, 0, &indx) == 0) {
2132 error = dupfdopen(td, indx, lp->lwp_dupfd, flags, error);
2135 fdrop(fp); /* our ref */
2138 fsetfd(fdp, NULL, indx);
2141 fdrop(fp); /* our ref */
2142 if (error == ERESTART)
2148 * ref the vnode for ourselves so it can't be ripped out from under
2149 * is. XXX need an ND flag to request that the vnode be returned
2152 * Reserve a file descriptor but do not assign it until the open
2155 vp = (struct vnode *)fp->f_data;
2157 if ((error = fdalloc(p, 0, &indx)) != 0) {
2164 * If no error occurs the vp will have been assigned to the file
2169 if (flags & (O_EXLOCK | O_SHLOCK)) {
2170 lf.l_whence = SEEK_SET;
2173 if (flags & O_EXLOCK)
2174 lf.l_type = F_WRLCK;
2176 lf.l_type = F_RDLCK;
2177 if (flags & FNONBLOCK)
2182 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
2184 * lock request failed. Clean up the reserved
2188 fsetfd(fdp, NULL, indx);
2192 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
2196 * Assert that all regular file vnodes were created with a object.
2198 KASSERT(vp->v_type != VREG || vp->v_object != NULL,
2199 ("open: regular file has no backing object after vn_open"));
2205 * release our private reference, leaving the one associated with the
2206 * descriptor table intact.
2208 if (oflags & O_CLOEXEC)
2209 fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
2210 fsetfd(fdp, fp, indx);
2218 * open_args(char *path, int flags, int mode)
2220 * Check permissions, allocate an open file structure,
2221 * and call the device open routine if any.
2224 sys_open(struct sysmsg *sysmsg, const struct open_args *uap)
2226 struct nlookupdata nd;
2229 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2231 error = kern_open(&nd, uap->flags,
2232 uap->mode, &sysmsg->sysmsg_result);
2239 * openat_args(int fd, char *path, int flags, int mode)
2242 sys_openat(struct sysmsg *sysmsg, const struct openat_args *uap)
2244 struct nlookupdata nd;
2248 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2250 error = kern_open(&nd, uap->flags, uap->mode,
2251 &sysmsg->sysmsg_result);
2253 nlookup_done_at(&nd, fp);
2258 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2260 struct thread *td = curthread;
2261 struct proc *p = td->td_proc;
2270 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2271 vattr.va_rmajor = rmajor;
2272 vattr.va_rminor = rminor;
2274 switch (mode & S_IFMT) {
2275 case S_IFMT: /* used by badsect to flag bad sectors */
2276 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2277 vattr.va_type = VBAD;
2280 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2281 vattr.va_type = VCHR;
2284 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2285 vattr.va_type = VBLK;
2288 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2291 case S_IFDIR: /* special directories support for HAMMER */
2292 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2293 vattr.va_type = VDIR;
2304 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2305 if ((error = nlookup(nd)) != 0)
2307 if (nd->nl_nch.ncp->nc_vp)
2309 if (nd->nl_dvp == NULL)
2311 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2315 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2316 nd->nl_cred, NAMEI_CREATE);
2319 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2320 &vp, nd->nl_cred, &vattr);
2328 * mknod_args(char *path, int mode, int dev)
2330 * Create a special file.
2333 sys_mknod(struct sysmsg *sysmsg, const struct mknod_args *uap)
2335 struct nlookupdata nd;
2338 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2340 error = kern_mknod(&nd, uap->mode,
2341 umajor(uap->dev), uminor(uap->dev));
2348 * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2350 * Create a special file. The path is relative to the directory associated
2354 sys_mknodat(struct sysmsg *sysmsg, const struct mknodat_args *uap)
2356 struct nlookupdata nd;
2360 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2362 error = kern_mknod(&nd, uap->mode,
2363 umajor(uap->dev), uminor(uap->dev));
2365 nlookup_done_at(&nd, fp);
2370 kern_mkfifo(struct nlookupdata *nd, int mode)
2372 struct thread *td = curthread;
2373 struct proc *p = td->td_proc;
2380 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2381 if ((error = nlookup(nd)) != 0)
2383 if (nd->nl_nch.ncp->nc_vp)
2385 if (nd->nl_dvp == NULL)
2387 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2391 vattr.va_type = VFIFO;
2392 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2394 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2401 * mkfifo_args(char *path, int mode)
2403 * Create a named pipe.
2406 sys_mkfifo(struct sysmsg *sysmsg, const struct mkfifo_args *uap)
2408 struct nlookupdata nd;
2411 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2413 error = kern_mkfifo(&nd, uap->mode);
2419 * mkfifoat_args(int fd, char *path, mode_t mode)
2421 * Create a named pipe. The path is relative to the directory associated
2425 sys_mkfifoat(struct sysmsg *sysmsg, const struct mkfifoat_args *uap)
2427 struct nlookupdata nd;
2431 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2433 error = kern_mkfifo(&nd, uap->mode);
2434 nlookup_done_at(&nd, fp);
2438 static int hardlink_check_uid = 0;
2439 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2440 &hardlink_check_uid, 0,
2441 "Unprivileged processes cannot create hard links to files owned by other "
2443 static int hardlink_check_gid = 0;
2444 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2445 &hardlink_check_gid, 0,
2446 "Unprivileged processes cannot create hard links to files owned by other "
2450 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2456 * Shortcut if disabled
2458 if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2462 * Privileged user can always hardlink
2464 if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2468 * Otherwise only if the originating file is owned by the
2469 * same user or group. Note that any group is allowed if
2470 * the file is owned by the caller.
2472 error = VOP_GETATTR(vp, &va);
2476 if (hardlink_check_uid) {
2477 if (cred->cr_uid != va.va_uid)
2481 if (hardlink_check_gid) {
2482 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2490 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2492 struct thread *td = curthread;
2497 * Lookup the source and obtained a locked vnode.
2499 * You may only hardlink a file which you have write permission
2500 * on or which you own.
2502 * XXX relookup on vget failure / race ?
2505 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2506 if ((error = nlookup(nd)) != 0)
2508 vp = nd->nl_nch.ncp->nc_vp;
2509 KKASSERT(vp != NULL);
2510 if (vp->v_type == VDIR)
2511 return (EPERM); /* POSIX */
2512 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2514 if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2518 * Unlock the source so we can lookup the target without deadlocking
2519 * (XXX vp is locked already, possible other deadlock?). The target
2522 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2523 nd->nl_flags &= ~NLC_NCPISLOCKED;
2524 cache_unlock(&nd->nl_nch);
2527 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2528 if ((error = nlookup(linknd)) != 0) {
2532 if (linknd->nl_nch.ncp->nc_vp) {
2536 if (linknd->nl_dvp == NULL) {
2540 VFS_MODIFYING(vp->v_mount);
2541 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2548 * Finally run the new API VOP.
2550 error = can_hardlink(vp, td, td->td_ucred);
2552 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2553 vp, linknd->nl_cred);
2560 * link_args(char *path, char *link)
2562 * Make a hard file link.
2565 sys_link(struct sysmsg *sysmsg, const struct link_args *uap)
2567 struct nlookupdata nd, linknd;
2570 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2572 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2574 error = kern_link(&nd, &linknd);
2575 nlookup_done(&linknd);
2582 * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2584 * Make a hard file link. The path1 argument is relative to the directory
2585 * associated with fd1, and similarly the path2 argument is relative to
2586 * the directory associated with fd2.
2589 sys_linkat(struct sysmsg *sysmsg, const struct linkat_args *uap)
2591 struct nlookupdata nd, linknd;
2592 struct file *fp1, *fp2;
2595 error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2596 (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2598 error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2599 uap->path2, UIO_USERSPACE, 0);
2601 error = kern_link(&nd, &linknd);
2602 nlookup_done_at(&linknd, fp2);
2604 nlookup_done_at(&nd, fp1);
2609 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2617 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2618 if ((error = nlookup(nd)) != 0)
2620 if (nd->nl_nch.ncp->nc_vp)
2622 if (nd->nl_dvp == NULL)
2624 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2628 vattr.va_mode = mode;
2629 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2636 * symlink(char *path, char *link)
2638 * Make a symbolic link.
2641 sys_symlink(struct sysmsg *sysmsg, const struct symlink_args *uap)
2643 struct thread *td = curthread;
2644 struct nlookupdata nd;
2649 path = objcache_get(namei_oc, M_WAITOK);
2650 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2652 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2654 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2655 error = kern_symlink(&nd, path, mode);
2659 objcache_put(namei_oc, path);
2664 * symlinkat_args(char *path1, int fd, char *path2)
2666 * Make a symbolic link. The path2 argument is relative to the directory
2667 * associated with fd.
2670 sys_symlinkat(struct sysmsg *sysmsg, const struct symlinkat_args *uap)
2672 struct thread *td = curthread;
2673 struct nlookupdata nd;
2679 path1 = objcache_get(namei_oc, M_WAITOK);
2680 error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2682 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2685 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2686 error = kern_symlink(&nd, path1, mode);
2688 nlookup_done_at(&nd, fp);
2690 objcache_put(namei_oc, path1);
2695 * undelete_args(char *path)
2697 * Delete a whiteout from the filesystem.
2700 sys_undelete(struct sysmsg *sysmsg, const struct undelete_args *uap)
2702 struct nlookupdata nd;
2705 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2707 nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2709 error = nlookup(&nd);
2710 if (error == 0 && nd.nl_dvp == NULL)
2713 error = ncp_writechk(&nd.nl_nch);
2715 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2723 kern_unlink(struct nlookupdata *nd)
2728 nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2729 if ((error = nlookup(nd)) != 0)
2731 if (nd->nl_dvp == NULL)
2733 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2735 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2740 * unlink_args(char *path)
2742 * Delete a name from the filesystem.
2745 sys_unlink(struct sysmsg *sysmsg, const struct unlink_args *uap)
2747 struct nlookupdata nd;
2750 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2752 error = kern_unlink(&nd);
2759 * unlinkat_args(int fd, char *path, int flags)
2761 * Delete the file or directory entry pointed to by fd/path.
2764 sys_unlinkat(struct sysmsg *sysmsg, const struct unlinkat_args *uap)
2766 struct nlookupdata nd;
2770 if (uap->flags & ~AT_REMOVEDIR)
2773 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2775 if (uap->flags & AT_REMOVEDIR)
2776 error = kern_rmdir(&nd);
2778 error = kern_unlink(&nd);
2780 nlookup_done_at(&nd, fp);
2785 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2787 struct thread *td = curthread;
2790 struct vattr_lite lva;
2794 fp = holdfp(td, fd, -1);
2797 if (fp->f_type != DTYPE_VNODE) {
2801 vp = (struct vnode *)fp->f_data;
2805 spin_lock(&fp->f_spin);
2806 new_offset = fp->f_offset + offset;
2810 error = VOP_GETATTR_LITE(vp, &lva);
2811 spin_lock(&fp->f_spin);
2812 new_offset = offset + lva.va_size;
2815 new_offset = offset;
2817 spin_lock(&fp->f_spin);
2822 spin_lock(&fp->f_spin);
2827 * Validate the seek position. Negative offsets are not allowed
2828 * for regular files or directories.
2830 * Normally we would also not want to allow negative offsets for
2831 * character and block-special devices. However kvm addresses
2832 * on 64 bit architectures might appear to be negative and must
2836 if (new_offset < 0 &&
2837 (vp->v_type == VREG || vp->v_type == VDIR)) {
2840 fp->f_offset = new_offset;
2843 *res = fp->f_offset;
2844 spin_unlock(&fp->f_spin);
2852 * lseek_args(int fd, int pad, off_t offset, int whence)
2854 * Reposition read/write file offset.
2857 sys_lseek(struct sysmsg *sysmsg, const struct lseek_args *uap)
2861 error = kern_lseek(uap->fd, uap->offset, uap->whence,
2862 &sysmsg->sysmsg_offset);
2868 * Check if current process can access given file. amode is a bitmask of *_OK
2869 * access bits. flags is a bitmask of AT_* flags.
2872 kern_access(struct nlookupdata *nd, int amode, int flags)
2877 if (flags & ~AT_EACCESS)
2879 nd->nl_flags |= NLC_SHAREDLOCK;
2880 if ((error = nlookup(nd)) != 0)
2882 if ((amode & W_OK) && (error = ncp_writechk(&nd->nl_nch)) != 0)
2885 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2889 /* Flags == 0 means only check for existence. */
2898 if ((mode & VWRITE) == 0 ||
2899 (error = vn_writechk(vp)) == 0) {
2900 error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2904 * If the file handle is stale we have to re-resolve the
2905 * entry with the ncp held exclusively. This is a hack
2908 if (error == ESTALE) {
2910 cache_unlock(&nd->nl_nch);
2911 cache_lock(&nd->nl_nch);
2912 cache_setunresolved(&nd->nl_nch);
2913 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2926 * access_args(char *path, int flags)
2928 * Check access permissions.
2931 sys_access(struct sysmsg *sysmsg, const struct access_args *uap)
2933 struct nlookupdata nd;
2936 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2938 error = kern_access(&nd, uap->flags, 0);
2945 * eaccess_args(char *path, int flags)
2947 * Check access permissions.
2950 sys_eaccess(struct sysmsg *sysmsg, const struct eaccess_args *uap)
2952 struct nlookupdata nd;
2955 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2957 error = kern_access(&nd, uap->flags, AT_EACCESS);
2964 * faccessat_args(int fd, char *path, int amode, int flags)
2966 * Check access permissions.
2969 sys_faccessat(struct sysmsg *sysmsg, const struct faccessat_args *uap)
2971 struct nlookupdata nd;
2975 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2978 error = kern_access(&nd, uap->amode, uap->flags);
2979 nlookup_done_at(&nd, fp);
2984 kern_stat(struct nlookupdata *nd, struct stat *st)
2989 nd->nl_flags |= NLC_SHAREDLOCK;
2990 if ((error = nlookup(nd)) != 0)
2993 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2997 error = cache_vref(&nd->nl_nch, NULL, &vp);
2999 error = vget(vp, LK_SHARED);
3003 error = vn_stat(vp, st, nd->nl_cred);
3006 * If the file handle is stale we have to re-resolve the
3007 * entry with the ncp held exclusively. This is a hack
3010 if (error == ESTALE) {
3016 cache_unlock(&nd->nl_nch);
3017 cache_lock(&nd->nl_nch);
3018 cache_setunresolved(&nd->nl_nch);
3019 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
3033 * stat_args(char *path, struct stat *ub)
3035 * Get file status; this version follows links.
3038 sys_stat(struct sysmsg *sysmsg, const struct stat_args *uap)
3040 struct nlookupdata nd;
3044 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3046 error = kern_stat(&nd, &st);
3048 error = copyout(&st, uap->ub, sizeof(*uap->ub));
3055 * lstat_args(char *path, struct stat *ub)
3057 * Get file status; this version does not follow links.
3060 sys_lstat(struct sysmsg *sysmsg, const struct lstat_args *uap)
3062 struct nlookupdata nd;
3066 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3068 error = kern_stat(&nd, &st);
3070 error = copyout(&st, uap->ub, sizeof(*uap->ub));
3077 * fstatat_args(int fd, char *path, struct stat *sb, int flags)
3079 * Get status of file pointed to by fd/path.
3082 sys_fstatat(struct sysmsg *sysmsg, const struct fstatat_args *uap)
3084 struct nlookupdata nd;
3090 if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3093 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3095 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3096 UIO_USERSPACE, flags);
3098 error = kern_stat(&nd, &st);
3100 error = copyout(&st, uap->sb, sizeof(*uap->sb));
3102 nlookup_done_at(&nd, fp);
3107 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp)
3109 struct nlookupdata nd;
3114 error = nlookup_init(&nd, path, UIO_USERSPACE, flags);
3116 error = nlookup(&nd);
3118 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3121 error = VOP_PATHCONF(vp, name, sysmsg_regp);
3128 * pathconf_Args(char *path, int name)
3130 * Get configurable pathname variables.
3133 sys_pathconf(struct sysmsg *sysmsg, const struct pathconf_args *uap)
3135 return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW,
3136 &sysmsg->sysmsg_reg));
3140 * lpathconf_Args(char *path, int name)
3142 * Get configurable pathname variables, but don't follow symlinks.
3145 sys_lpathconf(struct sysmsg *sysmsg, const struct lpathconf_args *uap)
3147 return (kern_pathconf(uap->path, uap->name, 0, &sysmsg->sysmsg_reg));
3152 * kern_readlink isn't properly split yet. There is a copyin burried
3153 * in VOP_READLINK().
3156 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
3158 struct thread *td = curthread;
3164 nd->nl_flags |= NLC_SHAREDLOCK;
3165 if ((error = nlookup(nd)) != 0)
3167 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
3170 if (vp->v_type != VLNK) {
3173 aiov.iov_base = buf;
3174 aiov.iov_len = count;
3175 auio.uio_iov = &aiov;
3176 auio.uio_iovcnt = 1;
3177 auio.uio_offset = 0;
3178 auio.uio_rw = UIO_READ;
3179 auio.uio_segflg = UIO_USERSPACE;
3181 auio.uio_resid = count;
3182 error = VOP_READLINK(vp, &auio, td->td_ucred);
3185 *res = count - auio.uio_resid;
3190 * readlink_args(char *path, char *buf, int count)
3192 * Return target name of a symbolic link.
3195 sys_readlink(struct sysmsg *sysmsg, const struct readlink_args *uap)
3197 struct nlookupdata nd;
3200 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3202 error = kern_readlink(&nd, uap->buf, uap->count,
3203 &sysmsg->sysmsg_result);
3210 * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
3212 * Return target name of a symbolic link. The path is relative to the
3213 * directory associated with fd.
3216 sys_readlinkat(struct sysmsg *sysmsg, const struct readlinkat_args *uap)
3218 struct nlookupdata nd;
3222 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
3224 error = kern_readlink(&nd, uap->buf, uap->bufsize,
3225 &sysmsg->sysmsg_result);
3227 nlookup_done_at(&nd, fp);
3232 setfflags(struct vnode *vp, u_long flags)
3234 struct thread *td = curthread;
3239 * Prevent non-root users from setting flags on devices. When
3240 * a device is reused, users can retain ownership of the device
3241 * if they are allowed to set flags and programs assume that
3242 * chown can't fail when done as root.
3244 if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
3245 ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
3249 * note: vget is required for any operation that might mod the vnode
3250 * so VINACTIVE is properly cleared.
3252 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3254 vattr.va_flags = flags;
3255 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3262 * chflags(const char *path, u_long flags)
3264 * Change flags of a file given a path name.
3267 sys_chflags(struct sysmsg *sysmsg, const struct chflags_args *uap)
3269 struct nlookupdata nd;
3274 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3276 error = nlookup(&nd);
3278 error = ncp_writechk(&nd.nl_nch);
3280 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3283 error = setfflags(vp, uap->flags);
3290 * lchflags(const char *path, u_long flags)
3292 * Change flags of a file given a path name, but don't follow symlinks.
3295 sys_lchflags(struct sysmsg *sysmsg, const struct lchflags_args *uap)
3297 struct nlookupdata nd;
3302 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3304 error = nlookup(&nd);
3306 error = ncp_writechk(&nd.nl_nch);
3308 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3311 error = setfflags(vp, uap->flags);
3318 * fchflags_args(int fd, u_flags flags)
3320 * Change flags of a file given a file descriptor.
3323 sys_fchflags(struct sysmsg *sysmsg, const struct fchflags_args *uap)
3325 struct thread *td = curthread;
3329 if ((error = holdvnode(td, uap->fd, &fp)) != 0)
3331 if (fp->f_nchandle.ncp)
3332 error = ncp_writechk(&fp->f_nchandle);
3334 error = setfflags((struct vnode *) fp->f_data, uap->flags);
3340 * chflagsat_args(int fd, const char *path, u_long flags, int atflags)
3341 * change flags given a pathname relative to a filedescriptor
3344 sys_chflagsat(struct sysmsg *sysmsg, const struct chflagsat_args *uap)
3346 struct nlookupdata nd;
3352 if (uap->atflags & ~AT_SYMLINK_NOFOLLOW)
3355 lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3358 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, lookupflags);
3360 error = nlookup(&nd);
3362 error = ncp_writechk(&nd.nl_nch);
3364 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3365 nlookup_done_at(&nd, fp);
3367 error = setfflags(vp, uap->flags);
3375 setfmode(struct vnode *vp, int mode)
3377 struct thread *td = curthread;
3382 * note: vget is required for any operation that might mod the vnode
3383 * so VINACTIVE is properly cleared.
3385 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3387 vattr.va_mode = mode & ALLPERMS;
3388 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3389 cache_inval_wxok(vp);
3396 kern_chmod(struct nlookupdata *nd, int mode)
3401 if ((error = nlookup(nd)) != 0)
3403 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3405 if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3406 error = setfmode(vp, mode);
3412 * chmod_args(char *path, int mode)
3414 * Change mode of a file given path name.
3417 sys_chmod(struct sysmsg *sysmsg, const struct chmod_args *uap)
3419 struct nlookupdata nd;
3422 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3424 error = kern_chmod(&nd, uap->mode);
3430 * lchmod_args(char *path, int mode)
3432 * Change mode of a file given path name (don't follow links.)
3435 sys_lchmod(struct sysmsg *sysmsg, const struct lchmod_args *uap)
3437 struct nlookupdata nd;
3440 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3442 error = kern_chmod(&nd, uap->mode);
3448 * fchmod_args(int fd, int mode)
3450 * Change mode of a file given a file descriptor.
3453 sys_fchmod(struct sysmsg *sysmsg, const struct fchmod_args *uap)
3455 struct thread *td = curthread;
3459 if ((error = holdvnode(td, uap->fd, &fp)) != 0)
3461 if (fp->f_nchandle.ncp)
3462 error = ncp_writechk(&fp->f_nchandle);
3464 error = setfmode((struct vnode *)fp->f_data, uap->mode);
3470 * fchmodat_args(char *path, int mode)
3472 * Change mode of a file pointed to by fd/path.
3475 sys_fchmodat(struct sysmsg *sysmsg, const struct fchmodat_args *uap)
3477 struct nlookupdata nd;
3482 if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3484 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3486 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3487 UIO_USERSPACE, flags);
3489 error = kern_chmod(&nd, uap->mode);
3490 nlookup_done_at(&nd, fp);
3495 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3497 struct thread *td = curthread;
3505 * note: vget is required for any operation that might mod the vnode
3506 * so VINACTIVE is properly cleared.
3508 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3509 if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3511 o_uid = vattr.va_uid;
3512 o_gid = vattr.va_gid;
3513 size = vattr.va_size;
3518 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3527 VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3528 VFS_ACCOUNT(mp, uid, gid, size);
3535 kern_chown(struct nlookupdata *nd, int uid, int gid)
3540 if ((error = nlookup(nd)) != 0)
3542 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3544 if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3545 error = setfown(nd->nl_nch.mount, vp, uid, gid);
3551 * chown(char *path, int uid, int gid)
3553 * Set ownership given a path name.
3556 sys_chown(struct sysmsg *sysmsg, const struct chown_args *uap)
3558 struct nlookupdata nd;
3561 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3563 error = kern_chown(&nd, uap->uid, uap->gid);
3569 * lchown_args(char *path, int uid, int gid)
3571 * Set ownership given a path name, do not cross symlinks.
3574 sys_lchown(struct sysmsg *sysmsg, const struct lchown_args *uap)
3576 struct nlookupdata nd;
3579 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3581 error = kern_chown(&nd, uap->uid, uap->gid);
3587 * fchown_args(int fd, int uid, int gid)
3589 * Set ownership given a file descriptor.
3592 sys_fchown(struct sysmsg *sysmsg, const struct fchown_args *uap)
3594 struct thread *td = curthread;
3595 struct proc *p = td->td_proc;
3599 if ((error = holdvnode(td, uap->fd, &fp)) != 0)
3601 if (fp->f_nchandle.ncp)
3602 error = ncp_writechk(&fp->f_nchandle);
3604 error = setfown(p->p_fd->fd_ncdir.mount,
3605 (struct vnode *)fp->f_data, uap->uid, uap->gid);
3611 * fchownat(int fd, char *path, int uid, int gid, int flags)
3613 * Set ownership of file pointed to by fd/path.
3616 sys_fchownat(struct sysmsg *sysmsg, const struct fchownat_args *uap)
3618 struct nlookupdata nd;
3623 if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3625 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3627 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3628 UIO_USERSPACE, flags);
3630 error = kern_chown(&nd, uap->uid, uap->gid);
3631 nlookup_done_at(&nd, fp);
3637 getutimes(struct timeval *tvp, struct timespec *tsp)
3639 struct timeval tv[2];
3644 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3647 if ((error = itimerfix(tvp)) != 0)
3649 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3650 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3656 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag)
3658 struct timespec tsnow;
3672 if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT)
3674 if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW)
3677 if (newts[0].tv_nsec == UTIME_OMIT)
3678 newts[0].tv_sec = VNOVAL;
3679 else if (newts[0].tv_nsec == UTIME_NOW)
3681 else if ((error = itimespecfix(&newts[0])) != 0)
3684 if (newts[1].tv_nsec == UTIME_OMIT)
3685 newts[1].tv_sec = VNOVAL;
3686 else if (newts[1].tv_nsec == UTIME_NOW)
3688 else if ((error = itimespecfix(&newts[1])) != 0)
3695 setutimes(struct vnode *vp, struct vattr *vattr,
3696 const struct timespec *ts, int nullflag)
3698 struct thread *td = curthread;
3702 vattr->va_atime = ts[0];
3703 vattr->va_mtime = ts[1];
3705 vattr->va_vaflags |= VA_UTIMES_NULL;
3706 error = VOP_SETATTR(vp, vattr, td->td_ucred);
3712 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3714 struct timespec ts[2];
3718 if ((error = getutimes(tptr, ts)) != 0)
3721 error = kern_utimensat(nd, tptr ? ts : NULL, 0);
3726 * utimes_args(char *path, struct timeval *tptr)
3728 * Set the access and modification times of a file.
3731 sys_utimes(struct sysmsg *sysmsg, const struct utimes_args *uap)
3733 struct timeval tv[2];
3734 struct nlookupdata nd;
3738 error = copyin(uap->tptr, tv, sizeof(tv));
3742 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3744 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3750 * lutimes_args(char *path, struct timeval *tptr)
3752 * Set the access and modification times of a file.
3755 sys_lutimes(struct sysmsg *sysmsg, const struct lutimes_args *uap)
3757 struct timeval tv[2];
3758 struct nlookupdata nd;
3762 error = copyin(uap->tptr, tv, sizeof(tv));
3766 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3768 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3774 * Set utimes on a file descriptor. The creds used to open the
3775 * file are used to determine whether the operation is allowed
3779 kern_futimens(int fd, struct timespec *ts)
3781 struct thread *td = curthread;
3782 struct timespec newts[2];
3786 struct vattr_lite lva;
3790 error = getutimens(ts, newts, &nullflag);
3793 if ((error = holdvnode(td, fd, &fp)) != 0)
3795 if (fp->f_nchandle.ncp)
3796 error = ncp_writechk(&fp->f_nchandle);
3799 error = vget(vp, LK_EXCLUSIVE);
3801 error = VOP_GETATTR_FP(vp, &vattr, fp);
3803 lva.va_type = vattr.va_type;
3804 lva.va_nlink = vattr.va_nlink;
3805 lva.va_mode = vattr.va_mode;
3806 lva.va_uid = vattr.va_uid;
3807 lva.va_gid = vattr.va_gid;
3808 lva.va_size = vattr.va_size;
3809 lva.va_flags = vattr.va_flags;
3811 error = naccess_lva(&lva, NLC_OWN | NLC_WRITE,
3815 error = setutimes(vp, &vattr, newts, nullflag);
3825 * futimens_args(int fd, struct timespec *ts)
3827 * Set the access and modification times of a file.
3830 sys_futimens(struct sysmsg *sysmsg, const struct futimens_args *uap)
3832 struct timespec ts[2];
3836 error = copyin(uap->ts, ts, sizeof(ts));
3840 error = kern_futimens(uap->fd, uap->ts ? ts : NULL);
3845 kern_futimes(int fd, struct timeval *tptr)
3847 struct timespec ts[2];
3851 if ((error = getutimes(tptr, ts)) != 0)
3854 error = kern_futimens(fd, tptr ? ts : NULL);
3859 * futimes_args(int fd, struct timeval *tptr)
3861 * Set the access and modification times of a file.
3864 sys_futimes(struct sysmsg *sysmsg, const struct futimes_args *uap)
3866 struct timeval tv[2];
3870 error = copyin(uap->tptr, tv, sizeof(tv));
3874 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3879 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags)
3881 struct timespec newts[2];
3887 if (flags & ~AT_SYMLINK_NOFOLLOW)
3890 error = getutimens(ts, newts, &nullflag);
3894 nd->nl_flags |= NLC_OWN | NLC_WRITE;
3895 if ((error = nlookup(nd)) != 0)
3897 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3899 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3901 if ((error = vn_writechk(vp)) == 0) {
3902 error = vget(vp, LK_EXCLUSIVE);
3904 error = setutimes(vp, &vattr, newts, nullflag);
3913 * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags);
3915 * Set file access and modification times of a file.
3918 sys_utimensat(struct sysmsg *sysmsg, const struct utimensat_args *uap)
3920 struct timespec ts[2];
3921 struct nlookupdata nd;
3927 error = copyin(uap->ts, ts, sizeof(ts));
3932 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3933 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3934 UIO_USERSPACE, flags);
3936 error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags);
3937 nlookup_done_at(&nd, fp);
3942 kern_truncate(struct nlookupdata *nd, off_t length)
3949 uint64_t old_size = 0;
3953 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3954 if ((error = nlookup(nd)) != 0)
3956 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3958 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3960 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
3965 if (vp->v_type == VDIR) {
3969 if (vfs_quota_enabled) {
3970 error = VOP_GETATTR(vp, &vattr);
3971 KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3974 old_size = vattr.va_size;
3977 if ((error = vn_writechk(vp)) == 0) {
3979 vattr.va_size = length;
3980 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3981 VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3989 * truncate(char *path, int pad, off_t length)
3991 * Truncate a file given its path name.
3994 sys_truncate(struct sysmsg *sysmsg, const struct truncate_args *uap)
3996 struct nlookupdata nd;
3999 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4001 error = kern_truncate(&nd, uap->length);
4007 kern_ftruncate(int fd, off_t length)
4009 struct thread *td = curthread;
4016 uint64_t old_size = 0;
4021 if ((error = holdvnode(td, fd, &fp)) != 0)
4023 if (fp->f_nchandle.ncp) {
4024 error = ncp_writechk(&fp->f_nchandle);
4028 if ((fp->f_flag & FWRITE) == 0) {
4032 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */
4036 vp = (struct vnode *)fp->f_data;
4037 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4038 if (vp->v_type == VDIR) {
4044 if (vfs_quota_enabled) {
4045 error = VOP_GETATTR_FP(vp, &vattr, fp);
4046 KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
4049 old_size = vattr.va_size;
4052 if ((error = vn_writechk(vp)) == 0) {
4054 vattr.va_size = length;
4055 error = VOP_SETATTR_FP(vp, &vattr, fp->f_cred, fp);
4057 VFS_ACCOUNT(mp, uid, gid, length - old_size);
4066 * ftruncate_args(int fd, int pad, off_t length)
4068 * Truncate a file given a file descriptor.
4071 sys_ftruncate(struct sysmsg *sysmsg, const struct ftruncate_args *uap)
4075 error = kern_ftruncate(uap->fd, uap->length);
4083 * Sync an open file.
4086 sys_fsync(struct sysmsg *sysmsg, const struct fsync_args *uap)
4088 struct thread *td = curthread;
4094 if ((error = holdvnode(td, uap->fd, &fp)) != 0)
4096 vp = (struct vnode *)fp->f_data;
4097 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4098 if ((obj = vp->v_object) != NULL) {
4099 if (vp->v_mount == NULL ||
4100 (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) {
4101 vm_object_page_clean(obj, 0, 0, 0);
4104 error = VOP_FSYNC_FP(vp, MNT_WAIT, VOP_FSYNC_SYSCALL, fp);
4105 if (error == 0 && vp->v_mount)
4106 error = buf_fsync(vp);
4114 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
4116 struct nchandle fnchd;
4117 struct nchandle tnchd;
4118 struct namecache *ncp;
4122 struct mount *userenlk;
4128 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
4129 if ((error = nlookup(fromnd)) != 0)
4131 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
4133 fnchd.mount = fromnd->nl_nch.mount;
4137 * unlock the source nch so we can lookup the target nch without
4138 * deadlocking. The target may or may not exist so we do not check
4139 * for a target vp like kern_mkdir() and other creation functions do.
4141 * The source and target directories are ref'd and rechecked after
4142 * everything is relocked to determine if the source or target file
4145 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
4146 fromnd->nl_flags &= ~NLC_NCPISLOCKED;
4147 fncp_gen = fromnd->nl_nch.ncp->nc_generation;
4149 if (fromnd->nl_nch.ncp->nc_vp &&
4150 fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
4151 userenlk = fnchd.mount;
4152 cache_unlock(&fromnd->nl_nch);
4153 lockmgr(&userenlk->mnt_renlock, LK_EXCLUSIVE);
4156 cache_unlock(&fromnd->nl_nch);
4162 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
4163 if ((error = nlookup(tond)) != 0) {
4167 tncp_gen = tond->nl_nch.ncp->nc_generation;
4169 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
4174 tnchd.mount = tond->nl_nch.mount;
4178 * If the source and target are the same there is nothing to do
4180 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
4188 * Mount points cannot be renamed or overwritten
4190 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
4200 * Lock all four namecache entries. tond is already locked.
4202 cache_lock4_tondlocked(&fnchd, &fromnd->nl_nch,
4203 &tnchd, &tond->nl_nch,
4204 fromnd->nl_cred, tond->nl_cred);
4205 fromnd->nl_flags |= NLC_NCPISLOCKED;
4208 * If the namecache generation changed for either fromnd or tond,
4211 if (fromnd->nl_nch.ncp->nc_generation != fncp_gen ||
4212 tond->nl_nch.ncp->nc_generation != tncp_gen) {
4213 krateprintf(&krate_rename,
4214 "kern_rename: retry due to race on: "
4215 "\"%s\" -> \"%s\"\n",
4216 fromnd->nl_nch.ncp->nc_name,
4217 tond->nl_nch.ncp->nc_name);
4223 * If either fromnd or tond are marked destroyed a ripout occured
4224 * out from under us and we must retry.
4226 if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) ||
4227 fromnd->nl_nch.ncp->nc_vp == NULL ||
4228 (tond->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED))) {
4229 krateprintf(&krate_rename,
4230 "kern_rename: retry due to ripout on: "
4231 "\"%s\" -> \"%s\"\n",
4232 fromnd->nl_nch.ncp->nc_name,
4233 tond->nl_nch.ncp->nc_name);
4239 * Make sure the parent directories linkages are the same. We have
4240 * already checked that fromnd and tond are not mount points so this
4241 * should not loop forever on a cross-mount.
4243 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
4244 tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
4250 * Both the source and target must be within the same filesystem and
4251 * in the same filesystem as their parent directories within the
4252 * namecache topology.
4254 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
4257 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
4258 mp != tond->nl_nch.mount) {
4264 * Make sure the mount point is writable
4266 if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
4271 * If the target exists and either the source or target is a directory,
4272 * then both must be directories.
4274 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
4277 if (tond->nl_nch.ncp->nc_vp) {
4278 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
4280 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
4281 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
4283 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
4289 * You cannot rename a source into itself or a subdirectory of itself.
4290 * We check this by travsersing the target directory upwards looking
4291 * for a match against the source.
4293 * Only required when renaming a directory, in which case userenlk is
4296 if (__predict_false(userenlk && error == 0)) {
4297 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
4298 if (fromnd->nl_nch.ncp == ncp) {
4306 * Even though the namespaces are different, they may still represent
4307 * hardlinks to the same file. The filesystem might have a hard time
4308 * with this so we issue a NREMOVE of the source instead of a NRENAME
4309 * when we detect the situation.
4312 fdvp = fromnd->nl_dvp;
4313 tdvp = tond->nl_dvp;
4314 if (fdvp == NULL || tdvp == NULL) {
4316 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
4317 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
4320 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
4321 fdvp, tdvp, tond->nl_cred);
4329 lockmgr(&userenlk->mnt_renlock, LK_RELEASE);
4334 * rename_args(char *from, char *to)
4336 * Rename files. Source and destination must either both be directories,
4337 * or both not be directories. If target is a directory, it must be empty.
4340 sys_rename(struct sysmsg *sysmsg, const struct rename_args *uap)
4342 struct nlookupdata fromnd, tond;
4346 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
4348 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
4350 error = kern_rename(&fromnd, &tond);
4351 nlookup_done(&tond);
4353 nlookup_done(&fromnd);
4354 } while (error == EAGAIN);
4359 * renameat_args(int oldfd, char *old, int newfd, char *new)
4361 * Rename files using paths relative to the directories associated with
4362 * oldfd and newfd. Source and destination must either both be directories,
4363 * or both not be directories. If target is a directory, it must be empty.
4366 sys_renameat(struct sysmsg *sysmsg, const struct renameat_args *uap)
4368 struct nlookupdata oldnd, newnd;
4369 struct file *oldfp, *newfp;
4373 error = nlookup_init_at(&oldnd, &oldfp,
4374 uap->oldfd, uap->old,
4377 error = nlookup_init_at(&newnd, &newfp,
4378 uap->newfd, uap->new,
4381 error = kern_rename(&oldnd, &newnd);
4382 nlookup_done_at(&newnd, newfp);
4384 nlookup_done_at(&oldnd, oldfp);
4385 } while (error == EAGAIN);
4390 kern_mkdir(struct nlookupdata *nd, int mode)
4392 struct thread *td = curthread;
4393 struct proc *p = td->td_proc;
4399 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
4400 if ((error = nlookup(nd)) != 0)
4403 if (nd->nl_nch.ncp->nc_vp)
4405 if (nd->nl_dvp == NULL)
4407 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4410 vattr.va_type = VDIR;
4411 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
4414 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
4421 * mkdir_args(char *path, int mode)
4423 * Make a directory file.
4426 sys_mkdir(struct sysmsg *sysmsg, const struct mkdir_args *uap)
4428 struct nlookupdata nd;
4431 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4433 error = kern_mkdir(&nd, uap->mode);
4439 * mkdirat_args(int fd, char *path, mode_t mode)
4441 * Make a directory file. The path is relative to the directory associated
4445 sys_mkdirat(struct sysmsg *sysmsg, const struct mkdirat_args *uap)
4447 struct nlookupdata nd;
4451 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
4453 error = kern_mkdir(&nd, uap->mode);
4454 nlookup_done_at(&nd, fp);
4459 kern_rmdir(struct nlookupdata *nd)
4464 nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4465 if ((error = nlookup(nd)) != 0)
4469 * Do not allow directories representing mount points to be
4470 * deleted, even if empty. Check write perms on mount point
4471 * in case the vnode is aliased (aka nullfs).
4473 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4475 if (nd->nl_dvp == NULL)
4477 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4479 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4484 * rmdir_args(char *path)
4486 * Remove a directory file.
4489 sys_rmdir(struct sysmsg *sysmsg, const struct rmdir_args *uap)
4491 struct nlookupdata nd;
4494 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4496 error = kern_rmdir(&nd);
4502 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4503 enum uio_seg direction)
4505 struct thread *td = curthread;
4513 if ((error = holdvnode(td, fd, &fp)) != 0)
4515 if ((fp->f_flag & FREAD) == 0) {
4519 vp = (struct vnode *)fp->f_data;
4520 if (vp->v_type != VDIR) {
4524 aiov.iov_base = buf;
4525 aiov.iov_len = count;
4526 auio.uio_iov = &aiov;
4527 auio.uio_iovcnt = 1;
4528 auio.uio_rw = UIO_READ;
4529 auio.uio_segflg = direction;
4531 auio.uio_resid = count;
4532 loff = auio.uio_offset = fp->f_offset;
4533 error = VOP_READDIR_FP(vp, &auio, fp->f_cred, &eofflag, NULL, NULL, fp);
4534 fp->f_offset = auio.uio_offset;
4539 * WARNING! *basep may not be wide enough to accomodate the
4540 * seek offset. XXX should we hack this to return the upper 32 bits
4541 * for offsets greater then 4G?
4544 *basep = (long)loff;
4546 *res = count - auio.uio_resid;
4553 * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4555 * Read a block of directory entries in a file system independent format.
4558 sys_getdirentries(struct sysmsg *sysmsg, const struct getdirentries_args *uap)
4563 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4564 &sysmsg->sysmsg_result, UIO_USERSPACE);
4566 if (error == 0 && uap->basep)
4567 error = copyout(&base, uap->basep, sizeof(*uap->basep));
4572 * getdents_args(int fd, char *buf, size_t count)
4575 sys_getdents(struct sysmsg *sysmsg, const struct getdents_args *uap)
4579 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4580 &sysmsg->sysmsg_result, UIO_USERSPACE);
4586 * Set the mode mask for creation of filesystem nodes.
4588 * umask(int newmask)
4591 sys_umask(struct sysmsg *sysmsg, const struct umask_args *uap)
4593 struct thread *td = curthread;
4594 struct proc *p = td->td_proc;
4595 struct filedesc *fdp;
4598 sysmsg->sysmsg_result = fdp->fd_cmask;
4599 fdp->fd_cmask = uap->newmask & ALLPERMS;
4604 * revoke(char *path)
4606 * Void all references to file by ripping underlying filesystem
4610 sys_revoke(struct sysmsg *sysmsg, const struct revoke_args *uap)
4612 struct nlookupdata nd;
4619 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4621 error = nlookup(&nd);
4623 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4624 cred = crhold(nd.nl_cred);
4628 error = VOP_GETATTR(vp, &vattr);
4629 if (error == 0 && cred->cr_uid != vattr.va_uid)
4630 error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4631 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4633 error = vrevoke(vp, cred);
4634 } else if (error == 0) {
4635 error = vrevoke(vp, cred);
4645 * getfh_args(char *fname, fhandle_t *fhp)
4647 * Get (NFS) file handle
4649 * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4650 * mount. This allows nullfs mounts to be explicitly exported.
4652 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4654 * nullfs mounts of subdirectories are not safe. That is, it will
4655 * work, but you do not really have protection against access to
4656 * the related parent directories.
4659 sys_getfh(struct sysmsg *sysmsg, const struct getfh_args *uap)
4661 struct thread *td = curthread;
4662 struct nlookupdata nd;
4669 * Must be super user
4671 if ((error = priv_check(td, PRIV_ROOT)) != 0)
4675 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4677 error = nlookup(&nd);
4679 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4680 mp = nd.nl_nch.mount;
4683 bzero(&fh, sizeof(fh));
4684 fh.fh_fsid = mp->mnt_stat.f_fsid;
4685 error = VFS_VPTOFH(vp, &fh.fh_fid);
4688 error = copyout(&fh, uap->fhp, sizeof(fh));
4694 * fhopen_args(const struct fhandle *u_fhp, int flags)
4696 * syscall for the rpc.lockd to use to translate a NFS file handle into
4697 * an open descriptor.
4699 * warning: do not remove the priv_check() call or this becomes one giant
4703 sys_fhopen(struct sysmsg *sysmsg, const struct fhopen_args *uap)
4705 struct thread *td = curthread;
4706 struct filedesc *fdp = td->td_proc->p_fd;
4711 struct vattr *vap = &vat;
4713 int fmode, mode, error = 0, type;
4719 * Must be super user
4721 error = priv_check(td, PRIV_ROOT);
4725 fmode = FFLAGS(uap->flags);
4728 * Why not allow a non-read/write open for our lockd?
4730 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4732 error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4737 * Find the mount point
4739 mp = vfs_getvfs(&fhp.fh_fsid);
4744 /* now give me my vnode, it gets returned to me locked */
4745 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4749 * from now on we have to make sure not
4750 * to forget about the vnode
4751 * any error that causes an abort must vput(vp)
4752 * just set error = err and 'goto bad;'.
4758 if (vp->v_type == VLNK) {
4762 if (vp->v_type == VSOCK) {
4767 if (fmode & (FWRITE | O_TRUNC)) {
4768 if (vp->v_type == VDIR) {
4772 error = vn_writechk(vp);
4780 error = VOP_ACCESS(vp, mode, td->td_ucred);
4784 if (fmode & O_TRUNC) {
4785 vn_unlock(vp); /* XXX */
4786 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */
4789 error = VOP_SETATTR(vp, vap, td->td_ucred);
4795 * VOP_OPEN needs the file pointer so it can potentially override
4798 * WARNING! no f_nchandle will be associated when fhopen()ing a
4801 if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4805 error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4808 * setting f_ops this way prevents VOP_CLOSE from being
4809 * called or fdrop() releasing the vp from v_data. Since
4810 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4812 fp->f_ops = &badfileops;
4818 * The fp is given its own reference, we still have our ref and lock.
4820 * Assert that all regular files must be created with a VM object.
4822 if (vp->v_type == VREG && vp->v_object == NULL) {
4823 kprintf("fhopen: regular file did not "
4824 "have VM object: %p\n",
4830 * The open was successful. Handle any locking requirements.
4832 if (fmode & (O_EXLOCK | O_SHLOCK)) {
4833 lf.l_whence = SEEK_SET;
4836 if (fmode & O_EXLOCK)
4837 lf.l_type = F_WRLCK;
4839 lf.l_type = F_RDLCK;
4840 if (fmode & FNONBLOCK)
4845 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK,
4848 * release our private reference.
4850 fsetfd(fdp, NULL, indx);
4855 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4856 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
4860 * Clean up. Associate the file pointer with the previously
4861 * reserved descriptor and return it.
4864 if (uap->flags & O_CLOEXEC)
4865 fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
4866 fsetfd(fdp, fp, indx);
4868 sysmsg->sysmsg_result = indx;
4874 fsetfd(fdp, NULL, indx);
4885 * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4888 sys_fhstat(struct sysmsg *sysmsg, const struct fhstat_args *uap)
4890 struct thread *td = curthread;
4898 * Must be super user
4900 error = priv_check(td, PRIV_ROOT);
4904 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4908 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4911 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4912 error = vn_stat(vp, &sb, td->td_ucred);
4917 error = copyout(&sb, uap->sb, sizeof(sb));
4925 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4928 sys_fhstatfs(struct sysmsg *sysmsg, const struct fhstatfs_args *uap)
4930 struct thread *td = curthread;
4931 struct proc *p = td->td_proc;
4936 char *fullpath, *freepath;
4941 * Must be super user
4943 if ((error = priv_check(td, PRIV_ROOT)))
4946 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4949 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4953 if (p != NULL && !chroot_visible_mnt(mp, p)) {
4958 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4963 if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4966 error = mount_path(p, mp, &fullpath, &freepath);
4969 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4970 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4971 kfree(freepath, M_TEMP);
4973 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4974 if (priv_check(td, PRIV_ROOT)) {
4975 bcopy(sp, &sb, sizeof(sb));
4976 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4979 error = copyout(sp, uap->buf, sizeof(*sp));
4988 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4991 sys_fhstatvfs(struct sysmsg *sysmsg, const struct fhstatvfs_args *uap)
4993 struct thread *td = curthread;
4994 struct proc *p = td->td_proc;
5002 * Must be super user
5004 if ((error = priv_check(td, PRIV_ROOT)))
5007 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
5010 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
5014 if (p != NULL && !chroot_visible_mnt(mp, p)) {
5019 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
5022 sp = &mp->mnt_vstat;
5024 if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
5028 if (mp->mnt_flag & MNT_RDONLY)
5029 sp->f_flag |= ST_RDONLY;
5030 if (mp->mnt_flag & MNT_NOSUID)
5031 sp->f_flag |= ST_NOSUID;
5032 error = copyout(sp, uap->buf, sizeof(*sp));
5041 * Syscall to push extended attribute configuration information into the
5042 * VFS. Accepts a path, which it converts to a mountpoint, as well as
5043 * a command (int cmd), and attribute name and misc data. For now, the
5044 * attribute name is left in userspace for consumption by the VFS_op.
5045 * It will probably be changed to be copied into sysspace by the
5046 * syscall in the future, once issues with various consumers of the
5047 * attribute code have raised their hands.
5049 * Currently this is used only by UFS Extended Attributes.
5052 sys_extattrctl(struct sysmsg *sysmsg, const struct extattrctl_args *uap)
5054 struct nlookupdata nd;
5056 char attrname[EXTATTR_MAXNAMELEN];
5064 if (error == 0 && uap->filename) {
5065 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
5068 error = nlookup(&nd);
5070 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
5074 if (error == 0 && uap->attrname) {
5075 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
5080 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5082 error = nlookup(&nd);
5084 error = ncp_writechk(&nd.nl_nch);
5086 error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
5088 uap->attrname, nd.nl_cred);
5097 * Syscall to get a named extended attribute on a file or directory.
5100 sys_extattr_set_file(struct sysmsg *sysmsg,
5101 const struct extattr_set_file_args *uap)
5103 char attrname[EXTATTR_MAXNAMELEN];
5104 struct nlookupdata nd;
5110 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5116 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5118 error = nlookup(&nd);
5120 error = ncp_writechk(&nd.nl_nch);
5122 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
5128 bzero(&auio, sizeof(auio));
5129 aiov.iov_base = uap->data;
5130 aiov.iov_len = uap->nbytes;
5131 auio.uio_iov = &aiov;
5132 auio.uio_iovcnt = 1;
5133 auio.uio_offset = 0;
5134 auio.uio_resid = uap->nbytes;
5135 auio.uio_rw = UIO_WRITE;
5136 auio.uio_td = curthread;
5138 error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
5147 * Syscall to get a named extended attribute on a file or directory.
5150 sys_extattr_get_file(struct sysmsg *sysmsg,
5151 const struct extattr_get_file_args *uap)
5153 char attrname[EXTATTR_MAXNAMELEN];
5154 struct nlookupdata nd;
5160 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5166 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5168 error = nlookup(&nd);
5170 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp);
5176 bzero(&auio, sizeof(auio));
5177 aiov.iov_base = uap->data;
5178 aiov.iov_len = uap->nbytes;
5179 auio.uio_iov = &aiov;
5180 auio.uio_iovcnt = 1;
5181 auio.uio_offset = 0;
5182 auio.uio_resid = uap->nbytes;
5183 auio.uio_rw = UIO_READ;
5184 auio.uio_td = curthread;
5186 error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
5188 sysmsg->sysmsg_result = uap->nbytes - auio.uio_resid;
5196 * Syscall to delete a named extended attribute from a file or directory.
5197 * Accepts attribute name. The real work happens in VOP_SETEXTATTR().
5200 sys_extattr_delete_file(struct sysmsg *sysmsg,
5201 const struct extattr_delete_file_args *uap)
5203 char attrname[EXTATTR_MAXNAMELEN];
5204 struct nlookupdata nd;
5208 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5212 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5214 error = nlookup(&nd);
5216 error = ncp_writechk(&nd.nl_nch);
5218 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
5220 error = VOP_SETEXTATTR(vp, uap->attrnamespace,
5221 attrname, NULL, nd.nl_cred);
5230 * Determine if the mount is visible to the process.
5233 chroot_visible_mnt(struct mount *mp, struct proc *p)
5235 struct nchandle nch;
5238 * Traverse from the mount point upwards. If we hit the process
5239 * root then the mount point is visible to the process.
5241 nch = mp->mnt_ncmountpt;
5243 if (nch.mount == p->p_fd->fd_nrdir.mount &&
5244 nch.ncp == p->p_fd->fd_nrdir.ncp) {
5247 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
5248 nch = nch.mount->mnt_ncmounton;
5250 nch.ncp = nch.ncp->nc_parent;
5255 * If the mount point is not visible to the process, but the
5256 * process root is in a subdirectory of the mount, return
5259 if (p->p_fd->fd_nrdir.mount == mp)
5265 /* Sets priv to PRIV_ROOT in case no matching fs */
5267 get_fspriv(const char *fsname)
5270 if (strncmp("null", fsname, 5) == 0) {
5271 return PRIV_VFS_MOUNT_NULLFS;
5272 } else if (strncmp(fsname, "tmpfs", 6) == 0) {
5273 return PRIV_VFS_MOUNT_TMPFS;
5280 sys___realpath(struct sysmsg *sysmsg, const struct __realpath_args *uap)
5282 struct nlookupdata nd;
5289 * Invalid length if less than 0. 0 is allowed
5291 if ((ssize_t)uap->len < 0)
5296 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5300 nd.nl_flags |= NLC_SHAREDLOCK;
5301 error = nlookup(&nd);
5305 if (nd.nl_nch.ncp->nc_vp == NULL) {
5311 * Shortcut test for existence.
5313 if (uap->len == 0) {
5314 error = ENAMETOOLONG;
5319 * Obtain the path relative to the process root. The nch must not
5320 * be locked for the cache_fullpath() call.
5322 if (nd.nl_flags & NLC_NCPISLOCKED) {
5323 nd.nl_flags &= ~NLC_NCPISLOCKED;
5324 cache_unlock(&nd.nl_nch);
5326 error = cache_fullpath(curproc, &nd.nl_nch, NULL, &rbuf, &fbuf, 0);
5330 rlen = (ssize_t)strlen(rbuf);
5331 if (rlen >= uap->len) {
5332 error = ENAMETOOLONG;
5335 error = copyout(rbuf, uap->buf, rlen + 1);
5337 sysmsg->sysmsg_szresult = rlen;
5341 kfree(fbuf, M_TEMP);