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/sysproto.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 chroot_refuse_vdir_fds (struct filedesc *fdp);
84 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
85 static int getutimes (struct timeval *, struct timespec *);
86 static int getutimens (const struct timespec *, struct timespec *, int *);
87 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
88 static int setfmode (struct vnode *, int);
89 static int setfflags (struct vnode *, int);
90 static int setutimes (struct vnode *, struct vattr *,
91 const struct timespec *, int);
92 static int usermount = 0; /* if 1, non-root can mount fs. */
94 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
95 "Allow non-root users to mount filesystems");
98 * Virtual File System System Calls
102 * Mount a file system.
104 * mount_args(char *type, char *path, int flags, caddr_t data)
109 sys_mount(struct mount_args *uap)
111 struct thread *td = curthread;
114 struct mount *mp, *nullmp;
115 struct vfsconf *vfsp;
116 int error, flag = 0, flag2 = 0;
119 struct nlookupdata nd;
120 char fstypename[MFSNAMELEN];
128 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
132 * Do not allow NFS export by non-root users.
134 if (uap->flags & MNT_EXPORTED) {
135 error = priv_check(td, PRIV_ROOT);
140 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
142 if (priv_check(td, PRIV_ROOT))
143 uap->flags |= MNT_NOSUID | MNT_NODEV;
146 * Lookup the requested path and extract the nch and vnode.
148 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
150 if ((error = nlookup(&nd)) == 0) {
151 if (nd.nl_nch.ncp->nc_vp == NULL)
161 * If the target filesystem is resolved via a nullfs mount, then
162 * nd.nl_nch.mount will be pointing to the nullfs mount structure
163 * instead of the target file system. We need it in case we are
166 nullmp = nd.nl_nch.mount;
169 * Extract the locked+refd ncp and cleanup the nd structure
172 cache_zero(&nd.nl_nch);
175 if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
176 (mp = cache_findmount(&nch)) != NULL) {
185 * now we have the locked ref'd nch and unreferenced vnode.
188 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
195 * Extract the file system type. We need to know this early, to take
196 * appropriate actions if we are dealing with a nullfs.
198 if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
205 * Now we have an unlocked ref'd nch and a locked ref'd vp
207 if (uap->flags & MNT_UPDATE) {
208 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
215 if (strncmp(fstypename, "null", 5) == 0) {
223 flag2 = mp->mnt_kern_flag;
225 * We only allow the filesystem to be reloaded if it
226 * is currently mounted read-only.
228 if ((uap->flags & MNT_RELOAD) &&
229 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
232 error = EOPNOTSUPP; /* Needs translation */
236 * Only root, or the user that did the original mount is
237 * permitted to update it.
239 if (mp->mnt_stat.f_owner != cred->cr_uid &&
240 (error = priv_check(td, PRIV_ROOT))) {
245 if (vfs_busy(mp, LK_NOWAIT)) {
259 uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
260 lwkt_gettoken(&mp->mnt_token);
267 * If the user is not root, ensure that they own the directory
268 * onto which we are attempting to mount.
270 if ((error = VOP_GETATTR(vp, &va)) ||
271 (va.va_uid != cred->cr_uid &&
272 (error = priv_check(td, PRIV_ROOT)))) {
277 if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
282 if (vp->v_type != VDIR) {
288 if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
294 vfsp = vfsconf_find_by_name(fstypename);
298 /* Only load modules for root (very important!) */
299 if ((error = priv_check(td, PRIV_ROOT)) != 0) {
304 error = linker_load_file(fstypename, &lf);
305 if (error || lf == NULL) {
313 /* lookup again, see if the VFS was loaded */
314 vfsp = vfsconf_find_by_name(fstypename);
317 linker_file_unload(lf);
332 * Allocate and initialize the filesystem.
334 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
336 vfs_busy(mp, LK_NOWAIT);
337 mp->mnt_op = vfsp->vfc_vfsops;
339 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT;
340 vfsp->vfc_refcount++;
341 mp->mnt_stat.f_type = vfsp->vfc_typenum;
342 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
343 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
344 mp->mnt_stat.f_owner = cred->cr_uid;
345 lwkt_gettoken(&mp->mnt_token);
349 * (per-mount token acquired at this point)
351 * Set the mount level flags.
353 if (uap->flags & MNT_RDONLY)
354 mp->mnt_flag |= MNT_RDONLY;
355 else if (mp->mnt_flag & MNT_RDONLY)
356 mp->mnt_kern_flag |= MNTK_WANTRDWR;
357 mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
358 MNT_SYNCHRONOUS | MNT_ASYNC | MNT_NOATIME |
359 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
360 MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
362 mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
363 MNT_NODEV | MNT_SYNCHRONOUS | MNT_ASYNC | MNT_FORCE |
364 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
365 MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
369 * Pre-set the mount's ALL_MPSAFE flags if specified in the vfsconf.
370 * This way the initial VFS_MOUNT() call will also be MPSAFE.
372 if (vfsp->vfc_flags & VFCF_MPSAFE)
373 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
376 * Mount the filesystem.
377 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
380 error = VFS_MOUNT(mp, uap->path, uap->data, cred);
381 if (mp->mnt_flag & MNT_UPDATE) {
382 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
383 mp->mnt_flag &= ~MNT_RDONLY;
384 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
385 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
388 mp->mnt_kern_flag = flag2;
390 lwkt_reltoken(&mp->mnt_token);
396 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
399 * Put the new filesystem on the mount list after root. The mount
400 * point gets its own mnt_ncmountpt (unless the VFS already set one
401 * up) which represents the root of the mount. The lookup code
402 * detects the mount point going forward and checks the root of
403 * the mount going backwards.
405 * It is not necessary to invalidate or purge the vnode underneath
406 * because elements under the mount will be given their own glue
410 if (mp->mnt_ncmountpt.ncp == NULL) {
412 * Allocate, then unlock, but leave the ref intact.
413 * This is the mnt_refs (1) that we will retain
414 * through to the unmount.
416 cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
417 cache_unlock(&mp->mnt_ncmountpt);
420 mp->mnt_ncmounton = nch; /* inherits ref */
422 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
424 cache_ismounting(mp);
425 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
427 mountlist_insert(mp, MNTINS_LAST);
429 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
430 error = vfs_allocate_syncvnode(mp);
431 lwkt_reltoken(&mp->mnt_token);
433 error = VFS_START(mp, 0);
435 KNOTE(&fs_klist, VQ_MOUNT);
437 vn_syncer_thr_stop(mp);
438 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
439 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
440 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
441 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
442 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
443 mp->mnt_vfc->vfc_refcount--;
444 lwkt_reltoken(&mp->mnt_token);
455 * Scan all active processes to see if any of them have a current
456 * or root directory onto which the new filesystem has just been
457 * mounted. If so, replace them with the new mount point.
459 * Both old_nch and new_nch are ref'd on call but not locked.
460 * new_nch must be temporarily locked so it can be associated with the
461 * vnode representing the root of the mount point.
463 struct checkdirs_info {
464 struct nchandle old_nch;
465 struct nchandle new_nch;
466 struct vnode *old_vp;
467 struct vnode *new_vp;
470 static int checkdirs_callback(struct proc *p, void *data);
473 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
475 struct checkdirs_info info;
481 * If the old mount point's vnode has a usecount of 1, it is not
482 * being held as a descriptor anywhere.
484 olddp = old_nch->ncp->nc_vp;
485 if (olddp == NULL || VREFCNT(olddp) == 1)
489 * Force the root vnode of the new mount point to be resolved
490 * so we can update any matching processes.
493 if (VFS_ROOT(mp, &newdp))
494 panic("mount: lost mount");
497 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
498 cache_setunresolved(new_nch);
499 cache_setvp(new_nch, newdp);
500 cache_unlock(new_nch);
503 * Special handling of the root node
505 if (rootvnode == olddp) {
507 vfs_cache_setroot(newdp, cache_hold(new_nch));
511 * Pass newdp separately so the callback does not have to access
512 * it via new_nch->ncp->nc_vp.
514 info.old_nch = *old_nch;
515 info.new_nch = *new_nch;
517 allproc_scan(checkdirs_callback, &info, 0);
522 * NOTE: callback is not MP safe because the scanned process's filedesc
523 * structure can be ripped out from under us, amoung other things.
526 checkdirs_callback(struct proc *p, void *data)
528 struct checkdirs_info *info = data;
529 struct filedesc *fdp;
530 struct nchandle ncdrop1;
531 struct nchandle ncdrop2;
532 struct vnode *vprele1;
533 struct vnode *vprele2;
535 if ((fdp = p->p_fd) != NULL) {
536 cache_zero(&ncdrop1);
537 cache_zero(&ncdrop2);
542 * MPUNSAFE - XXX fdp can be pulled out from under a
545 * A shared filedesc is ok, we don't have to copy it
546 * because we are making this change globally.
548 spin_lock(&fdp->fd_spin);
549 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
550 fdp->fd_ncdir.ncp == info->old_nch.ncp) {
551 vprele1 = fdp->fd_cdir;
553 fdp->fd_cdir = info->new_vp;
554 ncdrop1 = fdp->fd_ncdir;
555 cache_copy(&info->new_nch, &fdp->fd_ncdir);
557 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
558 fdp->fd_nrdir.ncp == info->old_nch.ncp) {
559 vprele2 = fdp->fd_rdir;
561 fdp->fd_rdir = info->new_vp;
562 ncdrop2 = fdp->fd_nrdir;
563 cache_copy(&info->new_nch, &fdp->fd_nrdir);
565 spin_unlock(&fdp->fd_spin);
567 cache_drop(&ncdrop1);
569 cache_drop(&ncdrop2);
579 * Unmount a file system.
581 * Note: unmount takes a path to the vnode mounted on as argument,
582 * not special file (as before).
584 * umount_args(char *path, int flags)
589 sys_unmount(struct unmount_args *uap)
591 struct thread *td = curthread;
592 struct proc *p __debugvar = td->td_proc;
593 struct mount *mp = NULL;
594 struct nlookupdata nd;
598 if (td->td_ucred->cr_prison != NULL) {
602 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
605 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
607 error = nlookup(&nd);
611 mp = nd.nl_nch.mount;
614 * Only root, or the user that did the original mount is
615 * permitted to unmount this filesystem.
617 if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
618 (error = priv_check(td, PRIV_ROOT)))
622 * Don't allow unmounting the root file system.
624 if (mp->mnt_flag & MNT_ROOTFS) {
630 * Must be the root of the filesystem
632 if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
638 * If no error try to issue the unmount. We lose our cache
639 * ref when we call nlookup_done so we must hold the mount point
640 * to prevent use-after-free races.
646 error = dounmount(mp, uap->flags);
656 * Do the actual file system unmount (interlocked against the mountlist
657 * token and mp->mnt_token).
660 dounmount_interlock(struct mount *mp)
662 if (mp->mnt_kern_flag & MNTK_UNMOUNT)
664 mp->mnt_kern_flag |= MNTK_UNMOUNT;
669 unmount_allproc_cb(struct proc *p, void *arg)
673 if (p->p_textnch.ncp == NULL)
676 mp = (struct mount *)arg;
677 if (p->p_textnch.mount == mp)
678 cache_drop(&p->p_textnch);
684 * The guts of the unmount code. The mount owns one ref and one hold
685 * count. If we successfully interlock the unmount, those refs are ours.
686 * (The ref is from mnt_ncmountpt).
689 dounmount(struct mount *mp, int flags)
691 struct namecache *ncp;
700 lwkt_gettoken(&mp->mnt_token);
703 * Exclusive access for unmounting purposes.
705 if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
709 * We now 'own' the last mp->mnt_refs
711 * Allow filesystems to detect that a forced unmount is in progress.
713 if (flags & MNT_FORCE)
714 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
715 lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK);
716 error = lockmgr(&mp->mnt_lock, lflags);
718 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
719 if (mp->mnt_kern_flag & MNTK_MWAIT) {
720 mp->mnt_kern_flag &= ~MNTK_MWAIT;
726 if (mp->mnt_flag & MNT_EXPUBLIC)
727 vfs_setpublicfs(NULL, NULL, NULL);
729 vfs_msync(mp, MNT_WAIT);
730 async_flag = mp->mnt_flag & MNT_ASYNC;
731 mp->mnt_flag &=~ MNT_ASYNC;
734 * If this filesystem isn't aliasing other filesystems,
735 * try to invalidate any remaining namecache entries and
736 * check the count afterwords.
738 * We own the last mnt_refs by owning mnt_ncmountpt.
740 if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
741 cache_lock(&mp->mnt_ncmountpt);
742 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
743 cache_unlock(&mp->mnt_ncmountpt);
745 cache_clearmntcache();
746 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
747 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
748 allproc_scan(&unmount_allproc_cb, mp, 0);
751 cache_clearmntcache();
752 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
753 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
755 if ((flags & MNT_FORCE) == 0) {
757 mount_warning(mp, "Cannot unmount: "
763 mount_warning(mp, "Forced unmount: "
774 * Decomission our special mnt_syncer vnode. This also stops
775 * the vnlru code. If we are unable to unmount we recommission
778 * Then sync the filesystem.
780 if ((vp = mp->mnt_syncer) != NULL) {
781 mp->mnt_syncer = NULL;
782 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
785 if ((mp->mnt_flag & MNT_RDONLY) == 0)
786 VFS_SYNC(mp, MNT_WAIT);
789 * nchandle records ref the mount structure. Expect a count of 1
790 * (our mount->mnt_ncmountpt).
792 * Scans can get temporary refs on a mountpoint (thought really
793 * heavy duty stuff like cache_findmount() do not).
795 if (mp->mnt_refs != 1)
796 cache_clearmntcache();
797 for (retry = 0; retry < 10 && mp->mnt_refs != 1; ++retry) {
798 cache_unmounting(mp);
799 tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1);
800 cache_clearmntcache();
802 if (mp->mnt_refs != 1) {
803 if ((flags & MNT_FORCE) == 0) {
804 mount_warning(mp, "Cannot unmount: "
805 "%d mount refs still present",
809 mount_warning(mp, "Forced unmount: "
810 "%d mount refs still present",
817 * So far so good, sync the filesystem once more and
818 * call the VFS unmount code if the sync succeeds.
821 if (mp->mnt_flag & MNT_RDONLY) {
822 error = VFS_UNMOUNT(mp, flags);
824 error = VFS_SYNC(mp, MNT_WAIT);
826 (error == EOPNOTSUPP) || /* No sync */
827 (flags & MNT_FORCE)) {
828 error = VFS_UNMOUNT(mp, flags);
834 * If an error occurred we can still recover, restoring the
835 * syncer vnode and misc flags.
838 if (mp->mnt_syncer == NULL)
839 vfs_allocate_syncvnode(mp);
840 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
841 mp->mnt_flag |= async_flag;
842 lockmgr(&mp->mnt_lock, LK_RELEASE);
843 if (mp->mnt_kern_flag & MNTK_MWAIT) {
844 mp->mnt_kern_flag &= ~MNTK_MWAIT;
850 * Clean up any journals still associated with the mount after
851 * filesystem activity has ceased.
853 journal_remove_all_journals(mp,
854 ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
856 mountlist_remove(mp);
859 * Remove any installed vnode ops here so the individual VFSs don't
862 * mnt_refs should go to zero when we scrap mnt_ncmountpt.
864 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
865 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
866 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
867 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
868 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
870 if (mp->mnt_ncmountpt.ncp != NULL) {
871 nch = mp->mnt_ncmountpt;
872 cache_zero(&mp->mnt_ncmountpt);
873 cache_clrmountpt(&nch);
876 if (mp->mnt_ncmounton.ncp != NULL) {
877 cache_unmounting(mp);
878 nch = mp->mnt_ncmounton;
879 cache_zero(&mp->mnt_ncmounton);
880 cache_clrmountpt(&nch);
884 mp->mnt_vfc->vfc_refcount--;
885 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
886 panic("unmount: dangling vnode");
891 lockmgr(&mp->mnt_lock, LK_RELEASE);
892 if (mp->mnt_kern_flag & MNTK_MWAIT) {
893 mp->mnt_kern_flag &= ~MNTK_MWAIT;
898 * If we reach here and freeok != 0 we must free the mount.
899 * mnt_refs should already have dropped to 0, so if it is not
900 * zero we must cycle the caches and wait.
902 * When we are satisfied that the mount has disconnected we can
903 * drop the hold on the mp that represented the mount (though the
904 * caller might actually have another, so the caller's drop may
905 * do the actual free).
908 if (mp->mnt_refs > 0)
909 cache_clearmntcache();
910 while (mp->mnt_refs > 0) {
911 cache_unmounting(mp);
913 tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1);
914 cache_clearmntcache();
916 lwkt_reltoken(&mp->mnt_token);
921 KNOTE(&fs_klist, VQ_UNMOUNT);
924 lwkt_reltoken(&mp->mnt_token);
930 mount_warning(struct mount *mp, const char *ctl, ...)
937 if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL,
938 &ptr, &buf, 0) == 0) {
939 kprintf("unmount(%s): ", ptr);
944 kprintf("unmount(%p", mp);
945 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
946 kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
955 * Shim cache_fullpath() to handle the case where a process is chrooted into
956 * a subdirectory of a mount. In this case if the root mount matches the
957 * process root directory's mount we have to specify the process's root
958 * directory instead of the mount point, because the mount point might
959 * be above the root directory.
963 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
965 struct nchandle *nch;
967 if (p && p->p_fd->fd_nrdir.mount == mp)
968 nch = &p->p_fd->fd_nrdir;
970 nch = &mp->mnt_ncmountpt;
971 return(cache_fullpath(p, nch, NULL, rb, fb, 0));
975 * Sync each mounted filesystem.
979 static int syncprt = 0;
980 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
983 static int sync_callback(struct mount *mp, void *data);
986 sys_sync(struct sync_args *uap)
988 mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
994 sync_callback(struct mount *mp, void *data __unused)
998 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
999 asyncflag = mp->mnt_flag & MNT_ASYNC;
1000 mp->mnt_flag &= ~MNT_ASYNC;
1001 vfs_msync(mp, MNT_NOWAIT);
1002 VFS_SYNC(mp, MNT_NOWAIT);
1003 mp->mnt_flag |= asyncflag;
1008 /* XXX PRISON: could be per prison flag */
1009 static int prison_quotas;
1011 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
1015 * quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
1017 * Change filesystem quotas.
1022 sys_quotactl(struct quotactl_args *uap)
1024 struct nlookupdata nd;
1030 if (td->td_ucred->cr_prison && !prison_quotas) {
1035 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1037 error = nlookup(&nd);
1039 mp = nd.nl_nch.mount;
1040 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
1041 uap->arg, nd.nl_cred);
1049 * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
1050 * void *buf, int buflen)
1052 * This function operates on a mount point and executes the specified
1053 * operation using the specified control data, and possibly returns data.
1055 * The actual number of bytes stored in the result buffer is returned, 0
1056 * if none, otherwise an error is returned.
1061 sys_mountctl(struct mountctl_args *uap)
1063 struct thread *td = curthread;
1064 struct proc *p = td->td_proc;
1072 * Sanity and permissions checks. We must be root.
1075 if (td->td_ucred->cr_prison != NULL)
1077 if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
1078 (error = priv_check(td, PRIV_ROOT)) != 0)
1082 * Argument length checks
1084 if (uap->ctllen < 0 || uap->ctllen > 1024)
1086 if (uap->buflen < 0 || uap->buflen > 16 * 1024)
1088 if (uap->path == NULL)
1092 * Allocate the necessary buffers and copyin data
1094 path = objcache_get(namei_oc, M_WAITOK);
1095 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1100 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1101 error = copyin(uap->ctl, ctl, uap->ctllen);
1106 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1109 * Validate the descriptor
1112 fp = holdfp(p->p_fd, uap->fd, -1);
1122 * Execute the internal kernel function and clean up.
1124 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1127 if (error == 0 && uap->sysmsg_result > 0)
1128 error = copyout(buf, uap->buf, uap->sysmsg_result);
1131 objcache_put(namei_oc, path);
1140 * Execute a mount control operation by resolving the path to a mount point
1141 * and calling vop_mountctl().
1143 * Use the mount point from the nch instead of the vnode so nullfs mounts
1144 * can properly spike the VOP.
1147 kern_mountctl(const char *path, int op, struct file *fp,
1148 const void *ctl, int ctllen,
1149 void *buf, int buflen, int *res)
1152 struct nlookupdata nd;
1153 struct nchandle nch;
1159 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1162 error = nlookup(&nd);
1167 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1174 * Yes, all this is needed to use the nch.mount below, because
1175 * we must maintain a ref on the mount to avoid ripouts (e.g.
1176 * due to heavy mount/unmount use by synth or poudriere).
1179 cache_zero(&nd.nl_nch);
1187 * Must be the root of the filesystem
1189 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1194 if (mp == NULL || mp->mnt_kern_flag & MNTK_UNMOUNT) {
1195 kprintf("kern_mountctl: Warning, \"%s\" racing unmount\n",
1201 error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1210 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1212 struct thread *td = curthread;
1213 struct proc *p = td->td_proc;
1216 char *fullpath, *freepath;
1219 if ((error = nlookup(nd)) != 0)
1221 mp = nd->nl_nch.mount;
1223 if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1226 error = mount_path(p, mp, &fullpath, &freepath);
1229 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1230 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1231 kfree(freepath, M_TEMP);
1233 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1234 bcopy(sp, buf, sizeof(*buf));
1235 /* Only root should have access to the fsid's. */
1236 if (priv_check(td, PRIV_ROOT))
1237 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1242 * statfs_args(char *path, struct statfs *buf)
1244 * Get filesystem statistics.
1247 sys_statfs(struct statfs_args *uap)
1249 struct nlookupdata nd;
1253 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1255 error = kern_statfs(&nd, &buf);
1258 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1263 kern_fstatfs(int fd, struct statfs *buf)
1265 struct thread *td = curthread;
1266 struct proc *p = td->td_proc;
1270 char *fullpath, *freepath;
1274 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1278 * Try to use mount info from any overlays rather than the
1279 * mount info for the underlying vnode, otherwise we will
1280 * fail when operating on null-mounted paths inside a chroot.
1282 if ((mp = fp->f_nchandle.mount) == NULL)
1283 mp = ((struct vnode *)fp->f_data)->v_mount;
1288 if (fp->f_cred == NULL) {
1293 if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1296 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1298 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1299 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1300 kfree(freepath, M_TEMP);
1302 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1303 bcopy(sp, buf, sizeof(*buf));
1305 /* Only root should have access to the fsid's. */
1306 if (priv_check(td, PRIV_ROOT))
1307 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1315 * fstatfs_args(int fd, struct statfs *buf)
1317 * Get filesystem statistics.
1320 sys_fstatfs(struct fstatfs_args *uap)
1325 error = kern_fstatfs(uap->fd, &buf);
1328 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1333 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1339 if ((error = nlookup(nd)) != 0)
1341 mp = nd->nl_nch.mount;
1342 sp = &mp->mnt_vstat;
1343 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1347 if (mp->mnt_flag & MNT_RDONLY)
1348 sp->f_flag |= ST_RDONLY;
1349 if (mp->mnt_flag & MNT_NOSUID)
1350 sp->f_flag |= ST_NOSUID;
1351 bcopy(sp, buf, sizeof(*buf));
1356 * statfs_args(char *path, struct statfs *buf)
1358 * Get filesystem statistics.
1361 sys_statvfs(struct statvfs_args *uap)
1363 struct nlookupdata nd;
1367 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1369 error = kern_statvfs(&nd, &buf);
1372 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1377 kern_fstatvfs(int fd, struct statvfs *buf)
1379 struct thread *td = curthread;
1380 struct proc *p = td->td_proc;
1387 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1389 if ((mp = fp->f_nchandle.mount) == NULL)
1390 mp = ((struct vnode *)fp->f_data)->v_mount;
1395 if (fp->f_cred == NULL) {
1399 sp = &mp->mnt_vstat;
1400 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1404 if (mp->mnt_flag & MNT_RDONLY)
1405 sp->f_flag |= ST_RDONLY;
1406 if (mp->mnt_flag & MNT_NOSUID)
1407 sp->f_flag |= ST_NOSUID;
1409 bcopy(sp, buf, sizeof(*buf));
1417 * fstatfs_args(int fd, struct statfs *buf)
1419 * Get filesystem statistics.
1422 sys_fstatvfs(struct fstatvfs_args *uap)
1427 error = kern_fstatvfs(uap->fd, &buf);
1430 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1435 * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1437 * Get statistics on all filesystems.
1440 struct getfsstat_info {
1441 struct statfs *sfsp;
1449 static int getfsstat_callback(struct mount *, void *);
1452 sys_getfsstat(struct getfsstat_args *uap)
1454 struct thread *td = curthread;
1455 struct getfsstat_info info;
1457 bzero(&info, sizeof(info));
1459 info.maxcount = uap->bufsize / sizeof(struct statfs);
1460 info.sfsp = uap->buf;
1462 info.flags = uap->flags;
1465 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1466 if (info.sfsp && info.count > info.maxcount)
1467 uap->sysmsg_result = info.maxcount;
1469 uap->sysmsg_result = info.count;
1470 return (info.error);
1474 getfsstat_callback(struct mount *mp, void *data)
1476 struct getfsstat_info *info = data;
1482 if (info->sfsp && info->count < info->maxcount) {
1483 if (info->td->td_proc &&
1484 !chroot_visible_mnt(mp, info->td->td_proc)) {
1490 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1491 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1492 * overrides MNT_WAIT.
1494 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1495 (info->flags & MNT_WAIT)) &&
1496 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1499 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1501 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1503 info->error = error;
1506 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1507 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1508 kfree(freepath, M_TEMP);
1510 error = copyout(sp, info->sfsp, sizeof(*sp));
1512 info->error = error;
1522 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1523 long bufsize, int flags)
1525 * Get statistics on all filesystems.
1528 struct getvfsstat_info {
1529 struct statfs *sfsp;
1530 struct statvfs *vsfsp;
1538 static int getvfsstat_callback(struct mount *, void *);
1541 sys_getvfsstat(struct getvfsstat_args *uap)
1543 struct thread *td = curthread;
1544 struct getvfsstat_info info;
1546 bzero(&info, sizeof(info));
1548 info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1549 info.sfsp = uap->buf;
1550 info.vsfsp = uap->vbuf;
1552 info.flags = uap->flags;
1555 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1556 if (info.vsfsp && info.count > info.maxcount)
1557 uap->sysmsg_result = info.maxcount;
1559 uap->sysmsg_result = info.count;
1560 return (info.error);
1564 getvfsstat_callback(struct mount *mp, void *data)
1566 struct getvfsstat_info *info = data;
1568 struct statvfs *vsp;
1573 if (info->vsfsp && info->count < info->maxcount) {
1574 if (info->td->td_proc &&
1575 !chroot_visible_mnt(mp, info->td->td_proc)) {
1579 vsp = &mp->mnt_vstat;
1582 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1583 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1584 * overrides MNT_WAIT.
1586 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1587 (info->flags & MNT_WAIT)) &&
1588 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1591 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1593 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1594 (info->flags & MNT_WAIT)) &&
1595 (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1599 if (mp->mnt_flag & MNT_RDONLY)
1600 vsp->f_flag |= ST_RDONLY;
1601 if (mp->mnt_flag & MNT_NOSUID)
1602 vsp->f_flag |= ST_NOSUID;
1604 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1606 info->error = error;
1609 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1610 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1611 kfree(freepath, M_TEMP);
1613 error = copyout(sp, info->sfsp, sizeof(*sp));
1615 error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1617 info->error = error;
1629 * fchdir_args(int fd)
1631 * Change current working directory to a given file descriptor.
1634 sys_fchdir(struct fchdir_args *uap)
1636 struct thread *td = curthread;
1637 struct proc *p = td->td_proc;
1638 struct filedesc *fdp = p->p_fd;
1639 struct vnode *vp, *ovp;
1642 struct nchandle nch, onch, tnch;
1645 if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1647 lwkt_gettoken(&p->p_token);
1648 vp = (struct vnode *)fp->f_data;
1650 vn_lock(vp, LK_SHARED | LK_RETRY);
1651 if (fp->f_nchandle.ncp == NULL)
1654 error = checkvp_chdir(vp, td);
1659 cache_copy(&fp->f_nchandle, &nch);
1662 * If the ncp has become a mount point, traverse through
1666 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1667 (mp = cache_findmount(&nch)) != NULL
1669 error = nlookup_mp(mp, &tnch);
1671 cache_unlock(&tnch); /* leave ref intact */
1673 vp = tnch.ncp->nc_vp;
1674 error = vget(vp, LK_SHARED);
1675 KKASSERT(error == 0);
1679 cache_dropmount(mp);
1682 spin_lock(&fdp->fd_spin);
1684 onch = fdp->fd_ncdir;
1686 fdp->fd_ncdir = nch;
1687 spin_unlock(&fdp->fd_spin);
1688 vn_unlock(vp); /* leave ref intact */
1697 lwkt_reltoken(&p->p_token);
1702 kern_chdir(struct nlookupdata *nd)
1704 struct thread *td = curthread;
1705 struct proc *p = td->td_proc;
1706 struct filedesc *fdp = p->p_fd;
1707 struct vnode *vp, *ovp;
1708 struct nchandle onch;
1711 nd->nl_flags |= NLC_SHAREDLOCK;
1712 if ((error = nlookup(nd)) != 0)
1714 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1716 if ((error = vget(vp, LK_SHARED)) != 0)
1719 lwkt_gettoken(&p->p_token);
1720 error = checkvp_chdir(vp, td);
1723 spin_lock(&fdp->fd_spin);
1725 onch = fdp->fd_ncdir;
1726 fdp->fd_ncdir = nd->nl_nch;
1728 spin_unlock(&fdp->fd_spin);
1729 cache_unlock(&nd->nl_nch); /* leave reference intact */
1732 cache_zero(&nd->nl_nch);
1736 lwkt_reltoken(&p->p_token);
1741 * chdir_args(char *path)
1743 * Change current working directory (``.'').
1746 sys_chdir(struct chdir_args *uap)
1748 struct nlookupdata nd;
1751 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1753 error = kern_chdir(&nd);
1759 * Helper function for raised chroot(2) security function: Refuse if
1760 * any filedescriptors are open directories.
1763 chroot_refuse_vdir_fds(struct filedesc *fdp)
1770 for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1771 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1773 vp = (struct vnode *)fp->f_data;
1774 if (vp->v_type != VDIR) {
1785 * This sysctl determines if we will allow a process to chroot(2) if it
1786 * has a directory open:
1787 * 0: disallowed for all processes.
1788 * 1: allowed for processes that were not already chroot(2)'ed.
1789 * 2: allowed for all processes.
1792 static int chroot_allow_open_directories = 1;
1794 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1795 &chroot_allow_open_directories, 0, "");
1798 * chroot to the specified namecache entry. We obtain the vp from the
1799 * namecache data. The passed ncp must be locked and referenced and will
1800 * remain locked and referenced on return.
1803 kern_chroot(struct nchandle *nch)
1805 struct thread *td = curthread;
1806 struct proc *p = td->td_proc;
1807 struct filedesc *fdp = p->p_fd;
1812 * Only privileged user can chroot
1814 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1819 * Disallow open directory descriptors (fchdir() breakouts).
1821 if (chroot_allow_open_directories == 0 ||
1822 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1823 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1826 if ((vp = nch->ncp->nc_vp) == NULL)
1829 if ((error = vget(vp, LK_SHARED)) != 0)
1833 * Check the validity of vp as a directory to change to and
1834 * associate it with rdir/jdir.
1836 error = checkvp_chdir(vp, td);
1837 vn_unlock(vp); /* leave reference intact */
1839 lwkt_gettoken(&p->p_token);
1840 vrele(fdp->fd_rdir);
1841 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */
1842 cache_drop(&fdp->fd_nrdir);
1843 cache_copy(nch, &fdp->fd_nrdir);
1844 if (fdp->fd_jdir == NULL) {
1847 cache_copy(nch, &fdp->fd_njdir);
1849 if ((p->p_flags & P_DIDCHROOT) == 0) {
1850 p->p_flags |= P_DIDCHROOT;
1851 if (p->p_depth <= 65535 - 32)
1854 lwkt_reltoken(&p->p_token);
1862 * chroot_args(char *path)
1864 * Change notion of root (``/'') directory.
1867 sys_chroot(struct chroot_args *uap)
1869 struct thread *td __debugvar = curthread;
1870 struct nlookupdata nd;
1873 KKASSERT(td->td_proc);
1874 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1876 nd.nl_flags |= NLC_EXEC;
1877 error = nlookup(&nd);
1879 error = kern_chroot(&nd.nl_nch);
1886 sys_chroot_kernel(struct chroot_kernel_args *uap)
1888 struct thread *td = curthread;
1889 struct nlookupdata nd;
1890 struct nchandle *nch;
1894 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1898 error = nlookup(&nd);
1904 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1908 if ((vp = nch->ncp->nc_vp) == NULL) {
1913 if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1916 kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1917 vfs_cache_setroot(vp, cache_hold(nch));
1926 * Common routine for chroot and chdir. Given a locked, referenced vnode,
1927 * determine whether it is legal to chdir to the vnode. The vnode's state
1928 * is not changed by this call.
1931 checkvp_chdir(struct vnode *vp, struct thread *td)
1935 if (vp->v_type != VDIR)
1938 error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1943 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1945 struct thread *td = curthread;
1946 struct proc *p = td->td_proc;
1947 struct lwp *lp = td->td_lwp;
1948 struct filedesc *fdp = p->p_fd;
1953 int type, indx, error = 0;
1956 if ((oflags & O_ACCMODE) == O_ACCMODE)
1958 flags = FFLAGS(oflags);
1959 error = falloc(lp, &nfp, NULL);
1963 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1966 * XXX p_dupfd is a real mess. It allows a device to return a
1967 * file descriptor to be duplicated rather then doing the open
1973 * Call vn_open() to do the lookup and assign the vnode to the
1974 * file pointer. vn_open() does not change the ref count on fp
1975 * and the vnode, on success, will be inherited by the file pointer
1978 * Request a shared lock on the vnode if possible.
1980 * Executable binaries can race VTEXT against O_RDWR opens, so
1981 * use an exclusive lock for O_RDWR opens as well.
1983 * NOTE: We need a flag to separate terminal vnode locking from
1984 * parent locking. O_CREAT needs parent locking, but O_TRUNC
1985 * and O_RDWR only need to lock the terminal vnode exclusively.
1987 nd->nl_flags |= NLC_LOCKVP;
1988 if ((flags & (O_CREAT|O_TRUNC|O_RDWR)) == 0)
1989 nd->nl_flags |= NLC_SHAREDLOCK;
1991 error = vn_open(nd, fp, flags, cmode);
1996 * handle special fdopen() case. bleh. dupfdopen() is
1997 * responsible for dropping the old contents of ofiles[indx]
2000 * Note that fsetfd() will add a ref to fp which represents
2001 * the fd_files[] assignment. We must still drop our
2004 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
2005 if (fdalloc(p, 0, &indx) == 0) {
2006 error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
2009 fdrop(fp); /* our ref */
2012 fsetfd(fdp, NULL, indx);
2015 fdrop(fp); /* our ref */
2016 if (error == ERESTART)
2022 * ref the vnode for ourselves so it can't be ripped out from under
2023 * is. XXX need an ND flag to request that the vnode be returned
2026 * Reserve a file descriptor but do not assign it until the open
2029 vp = (struct vnode *)fp->f_data;
2031 if ((error = fdalloc(p, 0, &indx)) != 0) {
2038 * If no error occurs the vp will have been assigned to the file
2043 if (flags & (O_EXLOCK | O_SHLOCK)) {
2044 lf.l_whence = SEEK_SET;
2047 if (flags & O_EXLOCK)
2048 lf.l_type = F_WRLCK;
2050 lf.l_type = F_RDLCK;
2051 if (flags & FNONBLOCK)
2056 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
2058 * lock request failed. Clean up the reserved
2062 fsetfd(fdp, NULL, indx);
2066 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
2070 * Assert that all regular file vnodes were created with a object.
2072 KASSERT(vp->v_type != VREG || vp->v_object != NULL,
2073 ("open: regular file has no backing object after vn_open"));
2079 * release our private reference, leaving the one associated with the
2080 * descriptor table intact.
2082 if (oflags & O_CLOEXEC)
2083 fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
2084 fsetfd(fdp, fp, indx);
2091 * open_args(char *path, int flags, int mode)
2093 * Check permissions, allocate an open file structure,
2094 * and call the device open routine if any.
2097 sys_open(struct open_args *uap)
2099 struct nlookupdata nd;
2102 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2104 error = kern_open(&nd, uap->flags,
2105 uap->mode, &uap->sysmsg_result);
2112 * openat_args(int fd, char *path, int flags, int mode)
2115 sys_openat(struct openat_args *uap)
2117 struct nlookupdata nd;
2121 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2123 error = kern_open(&nd, uap->flags, uap->mode,
2124 &uap->sysmsg_result);
2126 nlookup_done_at(&nd, fp);
2131 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2133 struct thread *td = curthread;
2134 struct proc *p = td->td_proc;
2143 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2144 vattr.va_rmajor = rmajor;
2145 vattr.va_rminor = rminor;
2147 switch (mode & S_IFMT) {
2148 case S_IFMT: /* used by badsect to flag bad sectors */
2149 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2150 vattr.va_type = VBAD;
2153 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2154 vattr.va_type = VCHR;
2157 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2158 vattr.va_type = VBLK;
2161 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2164 case S_IFDIR: /* special directories support for HAMMER */
2165 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2166 vattr.va_type = VDIR;
2177 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2178 if ((error = nlookup(nd)) != 0)
2180 if (nd->nl_nch.ncp->nc_vp)
2182 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2186 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2187 nd->nl_cred, NAMEI_CREATE);
2190 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2191 &vp, nd->nl_cred, &vattr);
2199 * mknod_args(char *path, int mode, int dev)
2201 * Create a special file.
2204 sys_mknod(struct mknod_args *uap)
2206 struct nlookupdata nd;
2209 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2211 error = kern_mknod(&nd, uap->mode,
2212 umajor(uap->dev), uminor(uap->dev));
2219 * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2221 * Create a special file. The path is relative to the directory associated
2225 sys_mknodat(struct mknodat_args *uap)
2227 struct nlookupdata nd;
2231 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2233 error = kern_mknod(&nd, uap->mode,
2234 umajor(uap->dev), uminor(uap->dev));
2236 nlookup_done_at(&nd, fp);
2241 kern_mkfifo(struct nlookupdata *nd, int mode)
2243 struct thread *td = curthread;
2244 struct proc *p = td->td_proc;
2251 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2252 if ((error = nlookup(nd)) != 0)
2254 if (nd->nl_nch.ncp->nc_vp)
2256 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2260 vattr.va_type = VFIFO;
2261 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2263 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2270 * mkfifo_args(char *path, int mode)
2272 * Create a named pipe.
2275 sys_mkfifo(struct mkfifo_args *uap)
2277 struct nlookupdata nd;
2280 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2282 error = kern_mkfifo(&nd, uap->mode);
2288 * mkfifoat_args(int fd, char *path, mode_t mode)
2290 * Create a named pipe. The path is relative to the directory associated
2294 sys_mkfifoat(struct mkfifoat_args *uap)
2296 struct nlookupdata nd;
2300 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2302 error = kern_mkfifo(&nd, uap->mode);
2303 nlookup_done_at(&nd, fp);
2307 static int hardlink_check_uid = 0;
2308 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2309 &hardlink_check_uid, 0,
2310 "Unprivileged processes cannot create hard links to files owned by other "
2312 static int hardlink_check_gid = 0;
2313 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2314 &hardlink_check_gid, 0,
2315 "Unprivileged processes cannot create hard links to files owned by other "
2319 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2325 * Shortcut if disabled
2327 if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2331 * Privileged user can always hardlink
2333 if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2337 * Otherwise only if the originating file is owned by the
2338 * same user or group. Note that any group is allowed if
2339 * the file is owned by the caller.
2341 error = VOP_GETATTR(vp, &va);
2345 if (hardlink_check_uid) {
2346 if (cred->cr_uid != va.va_uid)
2350 if (hardlink_check_gid) {
2351 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2359 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2361 struct thread *td = curthread;
2366 * Lookup the source and obtained a locked vnode.
2368 * You may only hardlink a file which you have write permission
2369 * on or which you own.
2371 * XXX relookup on vget failure / race ?
2374 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2375 if ((error = nlookup(nd)) != 0)
2377 vp = nd->nl_nch.ncp->nc_vp;
2378 KKASSERT(vp != NULL);
2379 if (vp->v_type == VDIR)
2380 return (EPERM); /* POSIX */
2381 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2383 if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2387 * Unlock the source so we can lookup the target without deadlocking
2388 * (XXX vp is locked already, possible other deadlock?). The target
2391 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2392 nd->nl_flags &= ~NLC_NCPISLOCKED;
2393 cache_unlock(&nd->nl_nch);
2396 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2397 if ((error = nlookup(linknd)) != 0) {
2401 if (linknd->nl_nch.ncp->nc_vp) {
2405 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2412 * Finally run the new API VOP.
2414 error = can_hardlink(vp, td, td->td_ucred);
2416 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2417 vp, linknd->nl_cred);
2424 * link_args(char *path, char *link)
2426 * Make a hard file link.
2429 sys_link(struct link_args *uap)
2431 struct nlookupdata nd, linknd;
2434 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2436 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2438 error = kern_link(&nd, &linknd);
2439 nlookup_done(&linknd);
2446 * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2448 * Make a hard file link. The path1 argument is relative to the directory
2449 * associated with fd1, and similarly the path2 argument is relative to
2450 * the directory associated with fd2.
2453 sys_linkat(struct linkat_args *uap)
2455 struct nlookupdata nd, linknd;
2456 struct file *fp1, *fp2;
2459 error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2460 (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2462 error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2463 uap->path2, UIO_USERSPACE, 0);
2465 error = kern_link(&nd, &linknd);
2466 nlookup_done_at(&linknd, fp2);
2468 nlookup_done_at(&nd, fp1);
2473 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2481 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2482 if ((error = nlookup(nd)) != 0)
2484 if (nd->nl_nch.ncp->nc_vp)
2486 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2490 vattr.va_mode = mode;
2491 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2498 * symlink(char *path, char *link)
2500 * Make a symbolic link.
2503 sys_symlink(struct symlink_args *uap)
2505 struct thread *td = curthread;
2506 struct nlookupdata nd;
2511 path = objcache_get(namei_oc, M_WAITOK);
2512 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2514 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2516 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2517 error = kern_symlink(&nd, path, mode);
2521 objcache_put(namei_oc, path);
2526 * symlinkat_args(char *path1, int fd, char *path2)
2528 * Make a symbolic link. The path2 argument is relative to the directory
2529 * associated with fd.
2532 sys_symlinkat(struct symlinkat_args *uap)
2534 struct thread *td = curthread;
2535 struct nlookupdata nd;
2541 path1 = objcache_get(namei_oc, M_WAITOK);
2542 error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2544 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2547 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2548 error = kern_symlink(&nd, path1, mode);
2550 nlookup_done_at(&nd, fp);
2552 objcache_put(namei_oc, path1);
2557 * undelete_args(char *path)
2559 * Delete a whiteout from the filesystem.
2562 sys_undelete(struct undelete_args *uap)
2564 struct nlookupdata nd;
2567 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2569 nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2571 error = nlookup(&nd);
2573 error = ncp_writechk(&nd.nl_nch);
2575 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2583 kern_unlink(struct nlookupdata *nd)
2588 nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2589 if ((error = nlookup(nd)) != 0)
2591 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2593 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2598 * unlink_args(char *path)
2600 * Delete a name from the filesystem.
2603 sys_unlink(struct unlink_args *uap)
2605 struct nlookupdata nd;
2608 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2610 error = kern_unlink(&nd);
2617 * unlinkat_args(int fd, char *path, int flags)
2619 * Delete the file or directory entry pointed to by fd/path.
2622 sys_unlinkat(struct unlinkat_args *uap)
2624 struct nlookupdata nd;
2628 if (uap->flags & ~AT_REMOVEDIR)
2631 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2633 if (uap->flags & AT_REMOVEDIR)
2634 error = kern_rmdir(&nd);
2636 error = kern_unlink(&nd);
2638 nlookup_done_at(&nd, fp);
2643 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2645 struct thread *td = curthread;
2646 struct proc *p = td->td_proc;
2653 fp = holdfp(p->p_fd, fd, -1);
2656 if (fp->f_type != DTYPE_VNODE) {
2660 vp = (struct vnode *)fp->f_data;
2664 spin_lock(&fp->f_spin);
2665 new_offset = fp->f_offset + offset;
2669 error = VOP_GETATTR(vp, &vattr);
2670 spin_lock(&fp->f_spin);
2671 new_offset = offset + vattr.va_size;
2674 new_offset = offset;
2676 spin_lock(&fp->f_spin);
2681 spin_lock(&fp->f_spin);
2686 * Validate the seek position. Negative offsets are not allowed
2687 * for regular files or directories.
2689 * Normally we would also not want to allow negative offsets for
2690 * character and block-special devices. However kvm addresses
2691 * on 64 bit architectures might appear to be negative and must
2695 if (new_offset < 0 &&
2696 (vp->v_type == VREG || vp->v_type == VDIR)) {
2699 fp->f_offset = new_offset;
2702 *res = fp->f_offset;
2703 spin_unlock(&fp->f_spin);
2710 * lseek_args(int fd, int pad, off_t offset, int whence)
2712 * Reposition read/write file offset.
2715 sys_lseek(struct lseek_args *uap)
2719 error = kern_lseek(uap->fd, uap->offset, uap->whence,
2720 &uap->sysmsg_offset);
2726 * Check if current process can access given file. amode is a bitmask of *_OK
2727 * access bits. flags is a bitmask of AT_* flags.
2730 kern_access(struct nlookupdata *nd, int amode, int flags)
2735 if (flags & ~AT_EACCESS)
2737 nd->nl_flags |= NLC_SHAREDLOCK;
2738 if ((error = nlookup(nd)) != 0)
2741 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2745 /* Flags == 0 means only check for existence. */
2754 if ((mode & VWRITE) == 0 ||
2755 (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2756 error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2759 * If the file handle is stale we have to re-resolve the
2760 * entry with the ncp held exclusively. This is a hack
2763 if (error == ESTALE) {
2765 cache_unlock(&nd->nl_nch);
2766 cache_lock(&nd->nl_nch);
2767 cache_setunresolved(&nd->nl_nch);
2768 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2781 * access_args(char *path, int flags)
2783 * Check access permissions.
2786 sys_access(struct access_args *uap)
2788 struct nlookupdata nd;
2791 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2793 error = kern_access(&nd, uap->flags, 0);
2800 * eaccess_args(char *path, int flags)
2802 * Check access permissions.
2805 sys_eaccess(struct eaccess_args *uap)
2807 struct nlookupdata nd;
2810 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2812 error = kern_access(&nd, uap->flags, AT_EACCESS);
2819 * faccessat_args(int fd, char *path, int amode, int flags)
2821 * Check access permissions.
2824 sys_faccessat(struct faccessat_args *uap)
2826 struct nlookupdata nd;
2830 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2833 error = kern_access(&nd, uap->amode, uap->flags);
2834 nlookup_done_at(&nd, fp);
2839 kern_stat(struct nlookupdata *nd, struct stat *st)
2844 nd->nl_flags |= NLC_SHAREDLOCK;
2845 if ((error = nlookup(nd)) != 0)
2848 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2851 if ((error = vget(vp, LK_SHARED)) != 0)
2853 error = vn_stat(vp, st, nd->nl_cred);
2856 * If the file handle is stale we have to re-resolve the
2857 * entry with the ncp held exclusively. This is a hack
2860 if (error == ESTALE) {
2862 cache_unlock(&nd->nl_nch);
2863 cache_lock(&nd->nl_nch);
2864 cache_setunresolved(&nd->nl_nch);
2865 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2875 * stat_args(char *path, struct stat *ub)
2877 * Get file status; this version follows links.
2880 sys_stat(struct stat_args *uap)
2882 struct nlookupdata nd;
2886 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2888 error = kern_stat(&nd, &st);
2890 error = copyout(&st, uap->ub, sizeof(*uap->ub));
2897 * lstat_args(char *path, struct stat *ub)
2899 * Get file status; this version does not follow links.
2902 sys_lstat(struct lstat_args *uap)
2904 struct nlookupdata nd;
2908 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2910 error = kern_stat(&nd, &st);
2912 error = copyout(&st, uap->ub, sizeof(*uap->ub));
2919 * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2921 * Get status of file pointed to by fd/path.
2924 sys_fstatat(struct fstatat_args *uap)
2926 struct nlookupdata nd;
2932 if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2935 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2937 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2938 UIO_USERSPACE, flags);
2940 error = kern_stat(&nd, &st);
2942 error = copyout(&st, uap->sb, sizeof(*uap->sb));
2944 nlookup_done_at(&nd, fp);
2949 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp)
2951 struct nlookupdata nd;
2956 error = nlookup_init(&nd, path, UIO_USERSPACE, flags);
2958 error = nlookup(&nd);
2960 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2963 error = VOP_PATHCONF(vp, name, sysmsg_regp);
2970 * pathconf_Args(char *path, int name)
2972 * Get configurable pathname variables.
2975 sys_pathconf(struct pathconf_args *uap)
2977 return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW,
2982 * lpathconf_Args(char *path, int name)
2984 * Get configurable pathname variables, but don't follow symlinks.
2987 sys_lpathconf(struct lpathconf_args *uap)
2989 return (kern_pathconf(uap->path, uap->name, 0, &uap->sysmsg_reg));
2994 * kern_readlink isn't properly split yet. There is a copyin burried
2995 * in VOP_READLINK().
2998 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
3000 struct thread *td = curthread;
3006 nd->nl_flags |= NLC_SHAREDLOCK;
3007 if ((error = nlookup(nd)) != 0)
3009 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
3012 if (vp->v_type != VLNK) {
3015 aiov.iov_base = buf;
3016 aiov.iov_len = count;
3017 auio.uio_iov = &aiov;
3018 auio.uio_iovcnt = 1;
3019 auio.uio_offset = 0;
3020 auio.uio_rw = UIO_READ;
3021 auio.uio_segflg = UIO_USERSPACE;
3023 auio.uio_resid = count;
3024 error = VOP_READLINK(vp, &auio, td->td_ucred);
3027 *res = count - auio.uio_resid;
3032 * readlink_args(char *path, char *buf, int count)
3034 * Return target name of a symbolic link.
3037 sys_readlink(struct readlink_args *uap)
3039 struct nlookupdata nd;
3042 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3044 error = kern_readlink(&nd, uap->buf, uap->count,
3045 &uap->sysmsg_result);
3052 * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
3054 * Return target name of a symbolic link. The path is relative to the
3055 * directory associated with fd.
3058 sys_readlinkat(struct readlinkat_args *uap)
3060 struct nlookupdata nd;
3064 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
3066 error = kern_readlink(&nd, uap->buf, uap->bufsize,
3067 &uap->sysmsg_result);
3069 nlookup_done_at(&nd, fp);
3074 setfflags(struct vnode *vp, int flags)
3076 struct thread *td = curthread;
3081 * Prevent non-root users from setting flags on devices. When
3082 * a device is reused, users can retain ownership of the device
3083 * if they are allowed to set flags and programs assume that
3084 * chown can't fail when done as root.
3086 if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
3087 ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
3091 * note: vget is required for any operation that might mod the vnode
3092 * so VINACTIVE is properly cleared.
3094 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3096 vattr.va_flags = flags;
3097 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3104 * chflags(char *path, int flags)
3106 * Change flags of a file given a path name.
3109 sys_chflags(struct chflags_args *uap)
3111 struct nlookupdata nd;
3116 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3118 error = nlookup(&nd);
3120 error = ncp_writechk(&nd.nl_nch);
3122 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3125 error = setfflags(vp, uap->flags);
3132 * lchflags(char *path, int flags)
3134 * Change flags of a file given a path name, but don't follow symlinks.
3137 sys_lchflags(struct lchflags_args *uap)
3139 struct nlookupdata nd;
3144 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3146 error = nlookup(&nd);
3148 error = ncp_writechk(&nd.nl_nch);
3150 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3153 error = setfflags(vp, uap->flags);
3160 * fchflags_args(int fd, int flags)
3162 * Change flags of a file given a file descriptor.
3165 sys_fchflags(struct fchflags_args *uap)
3167 struct thread *td = curthread;
3168 struct proc *p = td->td_proc;
3172 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3174 if (fp->f_nchandle.ncp)
3175 error = ncp_writechk(&fp->f_nchandle);
3177 error = setfflags((struct vnode *) fp->f_data, uap->flags);
3183 * chflagsat_args(int fd, const char *path, int flags, int atflags)
3184 * change flags given a pathname relative to a filedescriptor
3186 int sys_chflagsat(struct chflagsat_args *uap)
3188 struct nlookupdata nd;
3194 if (uap->atflags & ~AT_SYMLINK_NOFOLLOW)
3197 lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3200 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, lookupflags);
3202 error = nlookup(&nd);
3204 error = ncp_writechk(&nd.nl_nch);
3206 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3207 nlookup_done_at(&nd, fp);
3209 error = setfflags(vp, uap->flags);
3217 setfmode(struct vnode *vp, int mode)
3219 struct thread *td = curthread;
3224 * note: vget is required for any operation that might mod the vnode
3225 * so VINACTIVE is properly cleared.
3227 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3229 vattr.va_mode = mode & ALLPERMS;
3230 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3231 cache_inval_wxok(vp);
3238 kern_chmod(struct nlookupdata *nd, int mode)
3243 if ((error = nlookup(nd)) != 0)
3245 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3247 if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3248 error = setfmode(vp, mode);
3254 * chmod_args(char *path, int mode)
3256 * Change mode of a file given path name.
3259 sys_chmod(struct chmod_args *uap)
3261 struct nlookupdata nd;
3264 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3266 error = kern_chmod(&nd, uap->mode);
3272 * lchmod_args(char *path, int mode)
3274 * Change mode of a file given path name (don't follow links.)
3277 sys_lchmod(struct lchmod_args *uap)
3279 struct nlookupdata nd;
3282 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3284 error = kern_chmod(&nd, uap->mode);
3290 * fchmod_args(int fd, int mode)
3292 * Change mode of a file given a file descriptor.
3295 sys_fchmod(struct fchmod_args *uap)
3297 struct thread *td = curthread;
3298 struct proc *p = td->td_proc;
3302 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3304 if (fp->f_nchandle.ncp)
3305 error = ncp_writechk(&fp->f_nchandle);
3307 error = setfmode((struct vnode *)fp->f_data, uap->mode);
3313 * fchmodat_args(char *path, int mode)
3315 * Change mode of a file pointed to by fd/path.
3318 sys_fchmodat(struct fchmodat_args *uap)
3320 struct nlookupdata nd;
3325 if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3327 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3329 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3330 UIO_USERSPACE, flags);
3332 error = kern_chmod(&nd, uap->mode);
3333 nlookup_done_at(&nd, fp);
3338 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3340 struct thread *td = curthread;
3348 * note: vget is required for any operation that might mod the vnode
3349 * so VINACTIVE is properly cleared.
3351 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3352 if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3354 o_uid = vattr.va_uid;
3355 o_gid = vattr.va_gid;
3356 size = vattr.va_size;
3361 error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3370 VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3371 VFS_ACCOUNT(mp, uid, gid, size);
3378 kern_chown(struct nlookupdata *nd, int uid, int gid)
3383 if ((error = nlookup(nd)) != 0)
3385 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3387 if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3388 error = setfown(nd->nl_nch.mount, vp, uid, gid);
3394 * chown(char *path, int uid, int gid)
3396 * Set ownership given a path name.
3399 sys_chown(struct chown_args *uap)
3401 struct nlookupdata nd;
3404 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3406 error = kern_chown(&nd, uap->uid, uap->gid);
3412 * lchown_args(char *path, int uid, int gid)
3414 * Set ownership given a path name, do not cross symlinks.
3417 sys_lchown(struct lchown_args *uap)
3419 struct nlookupdata nd;
3422 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3424 error = kern_chown(&nd, uap->uid, uap->gid);
3430 * fchown_args(int fd, int uid, int gid)
3432 * Set ownership given a file descriptor.
3435 sys_fchown(struct fchown_args *uap)
3437 struct thread *td = curthread;
3438 struct proc *p = td->td_proc;
3442 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3444 if (fp->f_nchandle.ncp)
3445 error = ncp_writechk(&fp->f_nchandle);
3447 error = setfown(p->p_fd->fd_ncdir.mount,
3448 (struct vnode *)fp->f_data, uap->uid, uap->gid);
3454 * fchownat(int fd, char *path, int uid, int gid, int flags)
3456 * Set ownership of file pointed to by fd/path.
3459 sys_fchownat(struct fchownat_args *uap)
3461 struct nlookupdata nd;
3466 if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3468 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3470 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3471 UIO_USERSPACE, flags);
3473 error = kern_chown(&nd, uap->uid, uap->gid);
3474 nlookup_done_at(&nd, fp);
3480 getutimes(struct timeval *tvp, struct timespec *tsp)
3482 struct timeval tv[2];
3487 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3490 if ((error = itimerfix(tvp)) != 0)
3492 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3493 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3499 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag)
3501 struct timespec tsnow;
3515 if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT)
3517 if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW)
3520 if (newts[0].tv_nsec == UTIME_OMIT)
3521 newts[0].tv_sec = VNOVAL;
3522 else if (newts[0].tv_nsec == UTIME_NOW)
3524 else if ((error = itimespecfix(&newts[0])) != 0)
3527 if (newts[1].tv_nsec == UTIME_OMIT)
3528 newts[1].tv_sec = VNOVAL;
3529 else if (newts[1].tv_nsec == UTIME_NOW)
3531 else if ((error = itimespecfix(&newts[1])) != 0)
3538 setutimes(struct vnode *vp, struct vattr *vattr,
3539 const struct timespec *ts, int nullflag)
3541 struct thread *td = curthread;
3545 vattr->va_atime = ts[0];
3546 vattr->va_mtime = ts[1];
3548 vattr->va_vaflags |= VA_UTIMES_NULL;
3549 error = VOP_SETATTR(vp, vattr, td->td_ucred);
3555 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3557 struct timespec ts[2];
3561 if ((error = getutimes(tptr, ts)) != 0)
3564 error = kern_utimensat(nd, tptr ? ts : NULL, 0);
3569 * utimes_args(char *path, struct timeval *tptr)
3571 * Set the access and modification times of a file.
3574 sys_utimes(struct utimes_args *uap)
3576 struct timeval tv[2];
3577 struct nlookupdata nd;
3581 error = copyin(uap->tptr, tv, sizeof(tv));
3585 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3587 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3593 * lutimes_args(char *path, struct timeval *tptr)
3595 * Set the access and modification times of a file.
3598 sys_lutimes(struct lutimes_args *uap)
3600 struct timeval tv[2];
3601 struct nlookupdata nd;
3605 error = copyin(uap->tptr, tv, sizeof(tv));
3609 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3611 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3617 * Set utimes on a file descriptor. The creds used to open the
3618 * file are used to determine whether the operation is allowed
3622 kern_futimens(int fd, struct timespec *ts)
3624 struct thread *td = curthread;
3625 struct proc *p = td->td_proc;
3626 struct timespec newts[2];
3633 error = getutimens(ts, newts, &nullflag);
3636 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3638 if (fp->f_nchandle.ncp)
3639 error = ncp_writechk(&fp->f_nchandle);
3642 error = vget(vp, LK_EXCLUSIVE);
3644 error = VOP_GETATTR(vp, &vattr);
3646 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3650 error = setutimes(vp, &vattr, newts, nullflag);
3660 * futimens_args(int fd, struct timespec *ts)
3662 * Set the access and modification times of a file.
3665 sys_futimens(struct futimens_args *uap)
3667 struct timespec ts[2];
3671 error = copyin(uap->ts, ts, sizeof(ts));
3675 error = kern_futimens(uap->fd, uap->ts ? ts : NULL);
3680 kern_futimes(int fd, struct timeval *tptr)
3682 struct timespec ts[2];
3686 if ((error = getutimes(tptr, ts)) != 0)
3689 error = kern_futimens(fd, tptr ? ts : NULL);
3694 * futimes_args(int fd, struct timeval *tptr)
3696 * Set the access and modification times of a file.
3699 sys_futimes(struct futimes_args *uap)
3701 struct timeval tv[2];
3705 error = copyin(uap->tptr, tv, sizeof(tv));
3709 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3714 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags)
3716 struct timespec newts[2];
3722 if (flags & ~AT_SYMLINK_NOFOLLOW)
3725 error = getutimens(ts, newts, &nullflag);
3729 nd->nl_flags |= NLC_OWN | NLC_WRITE;
3730 if ((error = nlookup(nd)) != 0)
3732 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3734 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3736 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3737 error = vget(vp, LK_EXCLUSIVE);
3739 error = setutimes(vp, &vattr, newts, nullflag);
3748 * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags);
3750 * Set file access and modification times of a file.
3753 sys_utimensat(struct utimensat_args *uap)
3755 struct timespec ts[2];
3756 struct nlookupdata nd;
3762 error = copyin(uap->ts, ts, sizeof(ts));
3767 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3768 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3769 UIO_USERSPACE, flags);
3771 error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags);
3772 nlookup_done_at(&nd, fp);
3777 kern_truncate(struct nlookupdata *nd, off_t length)
3784 uint64_t old_size = 0;
3788 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3789 if ((error = nlookup(nd)) != 0)
3791 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3793 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3795 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
3800 if (vp->v_type == VDIR) {
3804 if (vfs_quota_enabled) {
3805 error = VOP_GETATTR(vp, &vattr);
3806 KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3809 old_size = vattr.va_size;
3812 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3814 vattr.va_size = length;
3815 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3816 VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3824 * truncate(char *path, int pad, off_t length)
3826 * Truncate a file given its path name.
3829 sys_truncate(struct truncate_args *uap)
3831 struct nlookupdata nd;
3834 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3836 error = kern_truncate(&nd, uap->length);
3842 kern_ftruncate(int fd, off_t length)
3844 struct thread *td = curthread;
3845 struct proc *p = td->td_proc;
3852 uint64_t old_size = 0;
3857 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3859 if (fp->f_nchandle.ncp) {
3860 error = ncp_writechk(&fp->f_nchandle);
3864 if ((fp->f_flag & FWRITE) == 0) {
3868 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */
3872 vp = (struct vnode *)fp->f_data;
3873 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3874 if (vp->v_type == VDIR) {
3880 if (vfs_quota_enabled) {
3881 error = VOP_GETATTR(vp, &vattr);
3882 KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
3885 old_size = vattr.va_size;
3888 if ((error = vn_writechk(vp, NULL)) == 0) {
3890 vattr.va_size = length;
3891 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3893 VFS_ACCOUNT(mp, uid, gid, length - old_size);
3902 * ftruncate_args(int fd, int pad, off_t length)
3904 * Truncate a file given a file descriptor.
3907 sys_ftruncate(struct ftruncate_args *uap)
3911 error = kern_ftruncate(uap->fd, uap->length);
3919 * Sync an open file.
3922 sys_fsync(struct fsync_args *uap)
3924 struct thread *td = curthread;
3925 struct proc *p = td->td_proc;
3931 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3933 vp = (struct vnode *)fp->f_data;
3934 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3935 if ((obj = vp->v_object) != NULL) {
3936 if (vp->v_mount == NULL ||
3937 (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) {
3938 vm_object_page_clean(obj, 0, 0, 0);
3941 error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3942 if (error == 0 && vp->v_mount)
3943 error = buf_fsync(vp);
3951 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3953 struct nchandle fnchd;
3954 struct nchandle tnchd;
3955 struct namecache *ncp;
3964 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3965 if ((error = nlookup(fromnd)) != 0)
3967 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3969 fnchd.mount = fromnd->nl_nch.mount;
3973 * unlock the source nch so we can lookup the target nch without
3974 * deadlocking. The target may or may not exist so we do not check
3975 * for a target vp like kern_mkdir() and other creation functions do.
3977 * The source and target directories are ref'd and rechecked after
3978 * everything is relocked to determine if the source or target file
3981 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3982 fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3984 fncp_gen = fromnd->nl_nch.ncp->nc_generation;
3986 cache_unlock(&fromnd->nl_nch);
3988 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3989 if ((error = nlookup(tond)) != 0) {
3993 tncp_gen = tond->nl_nch.ncp->nc_generation;
3995 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3999 tnchd.mount = tond->nl_nch.mount;
4003 * If the source and target are the same there is nothing to do
4005 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
4012 * Mount points cannot be renamed or overwritten
4014 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
4023 * Relock the source ncp. cache_relock() will deal with any
4024 * deadlocks against the already-locked tond and will also
4025 * make sure both are resolved.
4027 * NOTE AFTER RELOCKING: The source or target ncp may have become
4028 * invalid while they were unlocked, nc_vp and nc_mount could
4031 cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
4032 &tond->nl_nch, tond->nl_cred);
4033 fromnd->nl_flags |= NLC_NCPISLOCKED;
4036 * If the namecache generation changed for either fromnd or tond,
4039 if (fromnd->nl_nch.ncp->nc_generation != fncp_gen ||
4040 tond->nl_nch.ncp->nc_generation != tncp_gen) {
4041 kprintf("kern_rename: retry due to gen on: "
4042 "\"%s\" -> \"%s\"\n",
4043 fromnd->nl_nch.ncp->nc_name,
4044 tond->nl_nch.ncp->nc_name);
4051 * If either fromnd or tond are marked destroyed a ripout occured
4052 * out from under us and we must retry.
4054 if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) ||
4055 fromnd->nl_nch.ncp->nc_vp == NULL ||
4056 (tond->nl_nch.ncp->nc_flag & NCF_DESTROYED)) {
4057 kprintf("kern_rename: retry due to ripout on: "
4058 "\"%s\" -> \"%s\"\n",
4059 fromnd->nl_nch.ncp->nc_name,
4060 tond->nl_nch.ncp->nc_name);
4067 * Make sure the parent directories linkages are the same.
4068 * XXX shouldn't be needed any more w/ generation check above.
4070 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
4071 tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
4078 * Both the source and target must be within the same filesystem and
4079 * in the same filesystem as their parent directories within the
4080 * namecache topology.
4082 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
4085 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
4086 mp != tond->nl_nch.mount) {
4093 * Make sure the mount point is writable
4095 if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
4102 * If the target exists and either the source or target is a directory,
4103 * then both must be directories.
4105 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
4108 if (tond->nl_nch.ncp->nc_vp) {
4109 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
4111 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
4112 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
4114 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
4120 * You cannot rename a source into itself or a subdirectory of itself.
4121 * We check this by travsersing the target directory upwards looking
4122 * for a match against the source.
4127 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
4128 if (fromnd->nl_nch.ncp == ncp) {
4139 * Even though the namespaces are different, they may still represent
4140 * hardlinks to the same file. The filesystem might have a hard time
4141 * with this so we issue a NREMOVE of the source instead of a NRENAME
4142 * when we detect the situation.
4145 fdvp = fromnd->nl_dvp;
4146 tdvp = tond->nl_dvp;
4147 if (fdvp == NULL || tdvp == NULL) {
4149 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
4150 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
4153 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
4154 fdvp, tdvp, tond->nl_cred);
4161 * rename_args(char *from, char *to)
4163 * Rename files. Source and destination must either both be directories,
4164 * or both not be directories. If target is a directory, it must be empty.
4167 sys_rename(struct rename_args *uap)
4169 struct nlookupdata fromnd, tond;
4173 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
4175 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
4177 error = kern_rename(&fromnd, &tond);
4178 nlookup_done(&tond);
4180 nlookup_done(&fromnd);
4181 } while (error == EAGAIN);
4186 * renameat_args(int oldfd, char *old, int newfd, char *new)
4188 * Rename files using paths relative to the directories associated with
4189 * oldfd and newfd. Source and destination must either both be directories,
4190 * or both not be directories. If target is a directory, it must be empty.
4193 sys_renameat(struct renameat_args *uap)
4195 struct nlookupdata oldnd, newnd;
4196 struct file *oldfp, *newfp;
4200 error = nlookup_init_at(&oldnd, &oldfp,
4201 uap->oldfd, uap->old,
4204 error = nlookup_init_at(&newnd, &newfp,
4205 uap->newfd, uap->new,
4208 error = kern_rename(&oldnd, &newnd);
4209 nlookup_done_at(&newnd, newfp);
4211 nlookup_done_at(&oldnd, oldfp);
4212 } while (error == EAGAIN);
4217 kern_mkdir(struct nlookupdata *nd, int mode)
4219 struct thread *td = curthread;
4220 struct proc *p = td->td_proc;
4226 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
4227 if ((error = nlookup(nd)) != 0)
4230 if (nd->nl_nch.ncp->nc_vp)
4232 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4235 vattr.va_type = VDIR;
4236 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
4239 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
4246 * mkdir_args(char *path, int mode)
4248 * Make a directory file.
4251 sys_mkdir(struct mkdir_args *uap)
4253 struct nlookupdata nd;
4256 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4258 error = kern_mkdir(&nd, uap->mode);
4264 * mkdirat_args(int fd, char *path, mode_t mode)
4266 * Make a directory file. The path is relative to the directory associated
4270 sys_mkdirat(struct mkdirat_args *uap)
4272 struct nlookupdata nd;
4276 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
4278 error = kern_mkdir(&nd, uap->mode);
4279 nlookup_done_at(&nd, fp);
4284 kern_rmdir(struct nlookupdata *nd)
4289 nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4290 if ((error = nlookup(nd)) != 0)
4294 * Do not allow directories representing mount points to be
4295 * deleted, even if empty. Check write perms on mount point
4296 * in case the vnode is aliased (aka nullfs).
4298 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4300 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4302 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4307 * rmdir_args(char *path)
4309 * Remove a directory file.
4312 sys_rmdir(struct rmdir_args *uap)
4314 struct nlookupdata nd;
4317 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4319 error = kern_rmdir(&nd);
4325 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4326 enum uio_seg direction)
4328 struct thread *td = curthread;
4329 struct proc *p = td->td_proc;
4337 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
4339 if ((fp->f_flag & FREAD) == 0) {
4343 vp = (struct vnode *)fp->f_data;
4344 if (vp->v_type != VDIR) {
4348 aiov.iov_base = buf;
4349 aiov.iov_len = count;
4350 auio.uio_iov = &aiov;
4351 auio.uio_iovcnt = 1;
4352 auio.uio_rw = UIO_READ;
4353 auio.uio_segflg = direction;
4355 auio.uio_resid = count;
4356 loff = auio.uio_offset = fp->f_offset;
4357 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
4358 fp->f_offset = auio.uio_offset;
4363 * WARNING! *basep may not be wide enough to accomodate the
4364 * seek offset. XXX should we hack this to return the upper 32 bits
4365 * for offsets greater then 4G?
4368 *basep = (long)loff;
4370 *res = count - auio.uio_resid;
4377 * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4379 * Read a block of directory entries in a file system independent format.
4382 sys_getdirentries(struct getdirentries_args *uap)
4387 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4388 &uap->sysmsg_result, UIO_USERSPACE);
4390 if (error == 0 && uap->basep)
4391 error = copyout(&base, uap->basep, sizeof(*uap->basep));
4396 * getdents_args(int fd, char *buf, size_t count)
4399 sys_getdents(struct getdents_args *uap)
4403 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4404 &uap->sysmsg_result, UIO_USERSPACE);
4410 * Set the mode mask for creation of filesystem nodes.
4412 * umask(int newmask)
4415 sys_umask(struct umask_args *uap)
4417 struct thread *td = curthread;
4418 struct proc *p = td->td_proc;
4419 struct filedesc *fdp;
4422 uap->sysmsg_result = fdp->fd_cmask;
4423 fdp->fd_cmask = uap->newmask & ALLPERMS;
4428 * revoke(char *path)
4430 * Void all references to file by ripping underlying filesystem
4434 sys_revoke(struct revoke_args *uap)
4436 struct nlookupdata nd;
4443 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4445 error = nlookup(&nd);
4447 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4448 cred = crhold(nd.nl_cred);
4452 error = VOP_GETATTR(vp, &vattr);
4453 if (error == 0 && cred->cr_uid != vattr.va_uid)
4454 error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4455 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4457 error = vrevoke(vp, cred);
4458 } else if (error == 0) {
4459 error = vrevoke(vp, cred);
4469 * getfh_args(char *fname, fhandle_t *fhp)
4471 * Get (NFS) file handle
4473 * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4474 * mount. This allows nullfs mounts to be explicitly exported.
4476 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4478 * nullfs mounts of subdirectories are not safe. That is, it will
4479 * work, but you do not really have protection against access to
4480 * the related parent directories.
4483 sys_getfh(struct getfh_args *uap)
4485 struct thread *td = curthread;
4486 struct nlookupdata nd;
4493 * Must be super user
4495 if ((error = priv_check(td, PRIV_ROOT)) != 0)
4499 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4501 error = nlookup(&nd);
4503 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4504 mp = nd.nl_nch.mount;
4507 bzero(&fh, sizeof(fh));
4508 fh.fh_fsid = mp->mnt_stat.f_fsid;
4509 error = VFS_VPTOFH(vp, &fh.fh_fid);
4512 error = copyout(&fh, uap->fhp, sizeof(fh));
4518 * fhopen_args(const struct fhandle *u_fhp, int flags)
4520 * syscall for the rpc.lockd to use to translate a NFS file handle into
4521 * an open descriptor.
4523 * warning: do not remove the priv_check() call or this becomes one giant
4527 sys_fhopen(struct fhopen_args *uap)
4529 struct thread *td = curthread;
4530 struct filedesc *fdp = td->td_proc->p_fd;
4535 struct vattr *vap = &vat;
4537 int fmode, mode, error = 0, type;
4543 * Must be super user
4545 error = priv_check(td, PRIV_ROOT);
4549 fmode = FFLAGS(uap->flags);
4552 * Why not allow a non-read/write open for our lockd?
4554 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4556 error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4561 * Find the mount point
4563 mp = vfs_getvfs(&fhp.fh_fsid);
4568 /* now give me my vnode, it gets returned to me locked */
4569 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4573 * from now on we have to make sure not
4574 * to forget about the vnode
4575 * any error that causes an abort must vput(vp)
4576 * just set error = err and 'goto bad;'.
4582 if (vp->v_type == VLNK) {
4586 if (vp->v_type == VSOCK) {
4591 if (fmode & (FWRITE | O_TRUNC)) {
4592 if (vp->v_type == VDIR) {
4596 error = vn_writechk(vp, NULL);
4604 error = VOP_ACCESS(vp, mode, td->td_ucred);
4608 if (fmode & O_TRUNC) {
4609 vn_unlock(vp); /* XXX */
4610 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */
4613 error = VOP_SETATTR(vp, vap, td->td_ucred);
4619 * VOP_OPEN needs the file pointer so it can potentially override
4622 * WARNING! no f_nchandle will be associated when fhopen()ing a
4625 if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4629 error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4632 * setting f_ops this way prevents VOP_CLOSE from being
4633 * called or fdrop() releasing the vp from v_data. Since
4634 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4636 fp->f_ops = &badfileops;
4642 * The fp is given its own reference, we still have our ref and lock.
4644 * Assert that all regular files must be created with a VM object.
4646 if (vp->v_type == VREG && vp->v_object == NULL) {
4647 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4652 * The open was successful. Handle any locking requirements.
4654 if (fmode & (O_EXLOCK | O_SHLOCK)) {
4655 lf.l_whence = SEEK_SET;
4658 if (fmode & O_EXLOCK)
4659 lf.l_type = F_WRLCK;
4661 lf.l_type = F_RDLCK;
4662 if (fmode & FNONBLOCK)
4667 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4669 * release our private reference.
4671 fsetfd(fdp, NULL, indx);
4676 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4677 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
4681 * Clean up. Associate the file pointer with the previously
4682 * reserved descriptor and return it.
4685 if (uap->flags & O_CLOEXEC)
4686 fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
4687 fsetfd(fdp, fp, indx);
4689 uap->sysmsg_result = indx;
4693 fsetfd(fdp, NULL, indx);
4702 * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4705 sys_fhstat(struct fhstat_args *uap)
4707 struct thread *td = curthread;
4715 * Must be super user
4717 error = priv_check(td, PRIV_ROOT);
4721 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4725 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4728 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4729 error = vn_stat(vp, &sb, td->td_ucred);
4734 error = copyout(&sb, uap->sb, sizeof(sb));
4739 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4742 sys_fhstatfs(struct fhstatfs_args *uap)
4744 struct thread *td = curthread;
4745 struct proc *p = td->td_proc;
4750 char *fullpath, *freepath;
4755 * Must be super user
4757 if ((error = priv_check(td, PRIV_ROOT)))
4760 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4763 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4767 if (p != NULL && !chroot_visible_mnt(mp, p)) {
4772 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4777 if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4780 error = mount_path(p, mp, &fullpath, &freepath);
4783 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4784 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4785 kfree(freepath, M_TEMP);
4787 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4788 if (priv_check(td, PRIV_ROOT)) {
4789 bcopy(sp, &sb, sizeof(sb));
4790 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4793 error = copyout(sp, uap->buf, sizeof(*sp));
4799 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4802 sys_fhstatvfs(struct fhstatvfs_args *uap)
4804 struct thread *td = curthread;
4805 struct proc *p = td->td_proc;
4813 * Must be super user
4815 if ((error = priv_check(td, PRIV_ROOT)))
4818 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4821 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4825 if (p != NULL && !chroot_visible_mnt(mp, p)) {
4830 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4833 sp = &mp->mnt_vstat;
4835 if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4839 if (mp->mnt_flag & MNT_RDONLY)
4840 sp->f_flag |= ST_RDONLY;
4841 if (mp->mnt_flag & MNT_NOSUID)
4842 sp->f_flag |= ST_NOSUID;
4843 error = copyout(sp, uap->buf, sizeof(*sp));
4850 * Syscall to push extended attribute configuration information into the
4851 * VFS. Accepts a path, which it converts to a mountpoint, as well as
4852 * a command (int cmd), and attribute name and misc data. For now, the
4853 * attribute name is left in userspace for consumption by the VFS_op.
4854 * It will probably be changed to be copied into sysspace by the
4855 * syscall in the future, once issues with various consumers of the
4856 * attribute code have raised their hands.
4858 * Currently this is used only by UFS Extended Attributes.
4861 sys_extattrctl(struct extattrctl_args *uap)
4863 struct nlookupdata nd;
4865 char attrname[EXTATTR_MAXNAMELEN];
4873 if (error == 0 && uap->filename) {
4874 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4877 error = nlookup(&nd);
4879 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4883 if (error == 0 && uap->attrname) {
4884 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4889 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4891 error = nlookup(&nd);
4893 error = ncp_writechk(&nd.nl_nch);
4895 error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4897 uap->attrname, nd.nl_cred);
4906 * Syscall to get a named extended attribute on a file or directory.
4909 sys_extattr_set_file(struct extattr_set_file_args *uap)
4911 char attrname[EXTATTR_MAXNAMELEN];
4912 struct nlookupdata nd;
4918 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4924 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4926 error = nlookup(&nd);
4928 error = ncp_writechk(&nd.nl_nch);
4930 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4936 bzero(&auio, sizeof(auio));
4937 aiov.iov_base = uap->data;
4938 aiov.iov_len = uap->nbytes;
4939 auio.uio_iov = &aiov;
4940 auio.uio_iovcnt = 1;
4941 auio.uio_offset = 0;
4942 auio.uio_resid = uap->nbytes;
4943 auio.uio_rw = UIO_WRITE;
4944 auio.uio_td = curthread;
4946 error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4955 * Syscall to get a named extended attribute on a file or directory.
4958 sys_extattr_get_file(struct extattr_get_file_args *uap)
4960 char attrname[EXTATTR_MAXNAMELEN];
4961 struct nlookupdata nd;
4967 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4973 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4975 error = nlookup(&nd);
4977 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp);
4983 bzero(&auio, sizeof(auio));
4984 aiov.iov_base = uap->data;
4985 aiov.iov_len = uap->nbytes;
4986 auio.uio_iov = &aiov;
4987 auio.uio_iovcnt = 1;
4988 auio.uio_offset = 0;
4989 auio.uio_resid = uap->nbytes;
4990 auio.uio_rw = UIO_READ;
4991 auio.uio_td = curthread;
4993 error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4995 uap->sysmsg_result = uap->nbytes - auio.uio_resid;
5003 * Syscall to delete a named extended attribute from a file or directory.
5004 * Accepts attribute name. The real work happens in VOP_SETEXTATTR().
5007 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
5009 char attrname[EXTATTR_MAXNAMELEN];
5010 struct nlookupdata nd;
5014 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
5018 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
5020 error = nlookup(&nd);
5022 error = ncp_writechk(&nd.nl_nch);
5024 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
5026 error = VOP_SETEXTATTR(vp, uap->attrnamespace,
5027 attrname, NULL, nd.nl_cred);
5036 * Determine if the mount is visible to the process.
5039 chroot_visible_mnt(struct mount *mp, struct proc *p)
5041 struct nchandle nch;
5044 * Traverse from the mount point upwards. If we hit the process
5045 * root then the mount point is visible to the process.
5047 nch = mp->mnt_ncmountpt;
5049 if (nch.mount == p->p_fd->fd_nrdir.mount &&
5050 nch.ncp == p->p_fd->fd_nrdir.ncp) {
5053 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
5054 nch = nch.mount->mnt_ncmounton;
5056 nch.ncp = nch.ncp->nc_parent;
5061 * If the mount point is not visible to the process, but the
5062 * process root is in a subdirectory of the mount, return
5065 if (p->p_fd->fd_nrdir.mount == mp)