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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94
39 * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40 * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.135 2008/11/11 00:55:49 pavalos Exp $
43 #include <sys/param.h>
44 #include <sys/systm.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
56 #include <sys/linker.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
63 #include <sys/namei.h>
64 #include <sys/nlookup.h>
65 #include <sys/dirent.h>
66 #include <sys/extattr.h>
67 #include <sys/spinlock.h>
68 #include <sys/kern_syscall.h>
69 #include <sys/objcache.h>
70 #include <sys/sysctl.h>
73 #include <sys/file2.h>
74 #include <sys/spinlock2.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
80 #include <machine/limits.h>
81 #include <machine/stdarg.h>
83 #include <vfs/union/union.h>
85 static void mount_warning(struct mount *mp, const char *ctl, ...);
86 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
87 static int checkvp_chdir (struct vnode *vn, struct thread *td);
88 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
89 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
90 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
91 static int getutimes (const struct timeval *, struct timespec *);
92 static int setfown (struct vnode *, uid_t, gid_t);
93 static int setfmode (struct vnode *, int);
94 static int setfflags (struct vnode *, int);
95 static int setutimes (struct vnode *, struct vattr *,
96 const struct timespec *, int);
97 static int usermount = 0; /* if 1, non-root can mount fs. */
99 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
101 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
104 * Virtual File System System Calls
108 * Mount a file system.
111 * mount_args(char *type, char *path, int flags, caddr_t data)
115 sys_mount(struct mount_args *uap)
117 struct thread *td = curthread;
118 struct proc *p = td->td_proc;
122 struct vfsconf *vfsp;
123 int error, flag = 0, flag2 = 0;
126 struct nlookupdata nd;
127 char fstypename[MFSNAMELEN];
128 struct ucred *cred = p->p_ucred;
133 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
136 * Do not allow NFS export by non-root users.
138 if (uap->flags & MNT_EXPORTED) {
139 error = priv_check(td, PRIV_ROOT);
144 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
146 if (priv_check(td, PRIV_ROOT))
147 uap->flags |= MNT_NOSUID | MNT_NODEV;
150 * Lookup the requested path and extract the nch and vnode.
152 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
154 if ((error = nlookup(&nd)) == 0) {
155 if (nd.nl_nch.ncp->nc_vp == NULL)
165 * Extract the locked+refd ncp and cleanup the nd structure
168 cache_zero(&nd.nl_nch);
171 if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
178 * now we have the locked ref'd nch and unreferenced vnode.
181 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
188 * Now we have an unlocked ref'd nch and a locked ref'd vp
190 if (uap->flags & MNT_UPDATE) {
191 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
198 flag2 = mp->mnt_kern_flag;
200 * We only allow the filesystem to be reloaded if it
201 * is currently mounted read-only.
203 if ((uap->flags & MNT_RELOAD) &&
204 ((mp->mnt_flag & MNT_RDONLY) == 0)) {
207 return (EOPNOTSUPP); /* Needs translation */
210 * Only root, or the user that did the original mount is
211 * permitted to update it.
213 if (mp->mnt_stat.f_owner != cred->cr_uid &&
214 (error = priv_check(td, PRIV_ROOT))) {
219 if (vfs_busy(mp, LK_NOWAIT)) {
224 if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
230 vp->v_flag |= VMOUNT;
232 uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
237 * If the user is not root, ensure that they own the directory
238 * onto which we are attempting to mount.
240 if ((error = VOP_GETATTR(vp, &va)) ||
241 (va.va_uid != cred->cr_uid && (error = priv_check(td, PRIV_ROOT)))) {
246 if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
251 if (vp->v_type != VDIR) {
256 if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
261 if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
266 vfsp = vfsconf_find_by_name(fstypename);
270 /* Only load modules for root (very important!) */
271 if ((error = priv_check(td, PRIV_ROOT)) != 0) {
276 error = linker_load_file(fstypename, &lf);
277 if (error || lf == NULL) {
285 /* lookup again, see if the VFS was loaded */
286 vfsp = vfsconf_find_by_name(fstypename);
289 linker_file_unload(lf);
295 if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
300 vp->v_flag |= VMOUNT;
303 * Allocate and initialize the filesystem.
305 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
306 TAILQ_INIT(&mp->mnt_nvnodelist);
307 TAILQ_INIT(&mp->mnt_reservedvnlist);
308 TAILQ_INIT(&mp->mnt_jlist);
309 mp->mnt_nvnodelistsize = 0;
310 lockinit(&mp->mnt_lock, "vfslock", 0, 0);
311 vfs_busy(mp, LK_NOWAIT);
312 mp->mnt_op = vfsp->vfc_vfsops;
314 vfsp->vfc_refcount++;
315 mp->mnt_stat.f_type = vfsp->vfc_typenum;
316 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
317 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
318 mp->mnt_stat.f_owner = cred->cr_uid;
319 mp->mnt_iosize_max = DFLTPHYS;
323 * Set the mount level flags.
325 if (uap->flags & MNT_RDONLY)
326 mp->mnt_flag |= MNT_RDONLY;
327 else if (mp->mnt_flag & MNT_RDONLY)
328 mp->mnt_kern_flag |= MNTK_WANTRDWR;
329 mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
330 MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
331 MNT_NOSYMFOLLOW | MNT_IGNORE |
332 MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
333 mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
334 MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
335 MNT_NOSYMFOLLOW | MNT_IGNORE |
336 MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
338 * Mount the filesystem.
339 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
342 error = VFS_MOUNT(mp, uap->path, uap->data, cred);
343 if (mp->mnt_flag & MNT_UPDATE) {
344 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
345 mp->mnt_flag &= ~MNT_RDONLY;
346 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
347 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
350 mp->mnt_kern_flag = flag2;
353 vp->v_flag &= ~VMOUNT;
358 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
360 * Put the new filesystem on the mount list after root. The mount
361 * point gets its own mnt_ncmountpt (unless the VFS already set one
362 * up) which represents the root of the mount. The lookup code
363 * detects the mount point going forward and checks the root of
364 * the mount going backwards.
366 * It is not necessary to invalidate or purge the vnode underneath
367 * because elements under the mount will be given their own glue
371 if (mp->mnt_ncmountpt.ncp == NULL) {
373 * allocate, then unlock, but leave the ref intact
375 cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
376 cache_unlock(&mp->mnt_ncmountpt);
378 mp->mnt_ncmounton = nch; /* inherits ref */
379 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
381 /* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
382 vp->v_flag &= ~VMOUNT;
383 mountlist_insert(mp, MNTINS_LAST);
385 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
386 error = vfs_allocate_syncvnode(mp);
388 error = VFS_START(mp, 0);
391 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
392 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
393 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
394 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
395 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
396 vp->v_flag &= ~VMOUNT;
397 mp->mnt_vfc->vfc_refcount--;
407 * Scan all active processes to see if any of them have a current
408 * or root directory onto which the new filesystem has just been
409 * mounted. If so, replace them with the new mount point.
411 * The passed ncp is ref'd and locked (from the mount code) and
412 * must be associated with the vnode representing the root of the
415 struct checkdirs_info {
416 struct nchandle old_nch;
417 struct nchandle new_nch;
418 struct vnode *old_vp;
419 struct vnode *new_vp;
422 static int checkdirs_callback(struct proc *p, void *data);
425 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
427 struct checkdirs_info info;
433 * If the old mount point's vnode has a usecount of 1, it is not
434 * being held as a descriptor anywhere.
436 olddp = old_nch->ncp->nc_vp;
437 if (olddp == NULL || olddp->v_sysref.refcnt == 1)
441 * Force the root vnode of the new mount point to be resolved
442 * so we can update any matching processes.
445 if (VFS_ROOT(mp, &newdp))
446 panic("mount: lost mount");
447 cache_setunresolved(new_nch);
448 cache_setvp(new_nch, newdp);
451 * Special handling of the root node
453 if (rootvnode == olddp) {
455 vfs_cache_setroot(newdp, cache_hold(new_nch));
459 * Pass newdp separately so the callback does not have to access
460 * it via new_nch->ncp->nc_vp.
462 info.old_nch = *old_nch;
463 info.new_nch = *new_nch;
465 allproc_scan(checkdirs_callback, &info);
470 * NOTE: callback is not MP safe because the scanned process's filedesc
471 * structure can be ripped out from under us, amoung other things.
474 checkdirs_callback(struct proc *p, void *data)
476 struct checkdirs_info *info = data;
477 struct filedesc *fdp;
478 struct nchandle ncdrop1;
479 struct nchandle ncdrop2;
480 struct vnode *vprele1;
481 struct vnode *vprele2;
483 if ((fdp = p->p_fd) != NULL) {
484 cache_zero(&ncdrop1);
485 cache_zero(&ncdrop2);
490 * MPUNSAFE - XXX fdp can be pulled out from under a
493 * A shared filedesc is ok, we don't have to copy it
494 * because we are making this change globally.
496 spin_lock_wr(&fdp->fd_spin);
497 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
498 fdp->fd_ncdir.ncp == info->old_nch.ncp) {
499 vprele1 = fdp->fd_cdir;
501 fdp->fd_cdir = info->new_vp;
502 ncdrop1 = fdp->fd_ncdir;
503 cache_copy(&info->new_nch, &fdp->fd_ncdir);
505 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
506 fdp->fd_nrdir.ncp == info->old_nch.ncp) {
507 vprele2 = fdp->fd_rdir;
509 fdp->fd_rdir = info->new_vp;
510 ncdrop2 = fdp->fd_nrdir;
511 cache_copy(&info->new_nch, &fdp->fd_nrdir);
513 spin_unlock_wr(&fdp->fd_spin);
515 cache_drop(&ncdrop1);
517 cache_drop(&ncdrop2);
527 * Unmount a file system.
529 * Note: unmount takes a path to the vnode mounted on as argument,
530 * not special file (as before).
533 * umount_args(char *path, int flags)
537 sys_unmount(struct unmount_args *uap)
539 struct thread *td = curthread;
540 struct proc *p = td->td_proc;
541 struct mount *mp = NULL;
543 struct nlookupdata nd;
546 if (p->p_ucred->cr_prison != NULL)
548 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
551 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
553 error = nlookup(&nd);
557 mp = nd.nl_nch.mount;
560 * Only root, or the user that did the original mount is
561 * permitted to unmount this filesystem.
563 if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
564 (error = priv_check(td, PRIV_ROOT)))
568 * Don't allow unmounting the root file system.
570 if (mp->mnt_flag & MNT_ROOTFS) {
576 * Must be the root of the filesystem
578 if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
587 return (dounmount(mp, uap->flags));
591 * Do the actual file system unmount.
594 dounmount_interlock(struct mount *mp)
596 if (mp->mnt_kern_flag & MNTK_UNMOUNT)
598 mp->mnt_kern_flag |= MNTK_UNMOUNT;
603 dounmount(struct mount *mp, int flags)
605 struct namecache *ncp;
614 * Exclusive access for unmounting purposes
616 if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
620 * Allow filesystems to detect that a forced unmount is in progress.
622 if (flags & MNT_FORCE)
623 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
624 lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
625 error = lockmgr(&mp->mnt_lock, lflags);
627 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
628 if (mp->mnt_kern_flag & MNTK_MWAIT)
633 if (mp->mnt_flag & MNT_EXPUBLIC)
634 vfs_setpublicfs(NULL, NULL, NULL);
636 vfs_msync(mp, MNT_WAIT);
637 async_flag = mp->mnt_flag & MNT_ASYNC;
638 mp->mnt_flag &=~ MNT_ASYNC;
641 * If this filesystem isn't aliasing other filesystems,
642 * try to invalidate any remaining namecache entries and
643 * check the count afterwords.
645 if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
646 cache_lock(&mp->mnt_ncmountpt);
647 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
648 cache_unlock(&mp->mnt_ncmountpt);
650 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
651 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
653 if ((flags & MNT_FORCE) == 0) {
655 mount_warning(mp, "Cannot unmount: "
661 mount_warning(mp, "Forced unmount: "
672 * nchandle records ref the mount structure. Expect a count of 1
673 * (our mount->mnt_ncmountpt).
675 if (mp->mnt_refs != 1) {
676 if ((flags & MNT_FORCE) == 0) {
677 mount_warning(mp, "Cannot unmount: "
678 "%d process references still "
679 "present", mp->mnt_refs);
682 mount_warning(mp, "Forced unmount: "
683 "%d process references still "
684 "present", mp->mnt_refs);
690 * Decomission our special mnt_syncer vnode. This also stops
691 * the vnlru code. If we are unable to unmount we recommission
695 if ((vp = mp->mnt_syncer) != NULL) {
696 mp->mnt_syncer = NULL;
699 if (((mp->mnt_flag & MNT_RDONLY) ||
700 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
701 (flags & MNT_FORCE)) {
702 error = VFS_UNMOUNT(mp, flags);
706 if (mp->mnt_syncer == NULL)
707 vfs_allocate_syncvnode(mp);
708 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
709 mp->mnt_flag |= async_flag;
710 lockmgr(&mp->mnt_lock, LK_RELEASE);
711 if (mp->mnt_kern_flag & MNTK_MWAIT)
716 * Clean up any journals still associated with the mount after
717 * filesystem activity has ceased.
719 journal_remove_all_journals(mp,
720 ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
722 mountlist_remove(mp);
725 * Remove any installed vnode ops here so the individual VFSs don't
728 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
729 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
730 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
731 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
732 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
734 if (mp->mnt_ncmountpt.ncp != NULL) {
735 nch = mp->mnt_ncmountpt;
736 cache_zero(&mp->mnt_ncmountpt);
737 cache_clrmountpt(&nch);
740 if (mp->mnt_ncmounton.ncp != NULL) {
741 nch = mp->mnt_ncmounton;
742 cache_zero(&mp->mnt_ncmounton);
743 cache_clrmountpt(&nch);
747 mp->mnt_vfc->vfc_refcount--;
748 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
749 panic("unmount: dangling vnode");
750 lockmgr(&mp->mnt_lock, LK_RELEASE);
751 if (mp->mnt_kern_flag & MNTK_MWAIT)
760 mount_warning(struct mount *mp, const char *ctl, ...)
767 if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
768 kprintf("unmount(%s): ", ptr);
773 kprintf("unmount(%p", mp);
774 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
775 kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
784 * Shim cache_fullpath() to handle the case where a process is chrooted into
785 * a subdirectory of a mount. In this case if the root mount matches the
786 * process root directory's mount we have to specify the process's root
787 * directory instead of the mount point, because the mount point might
788 * be above the root directory.
792 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
794 struct nchandle *nch;
796 if (p && p->p_fd->fd_nrdir.mount == mp)
797 nch = &p->p_fd->fd_nrdir;
799 nch = &mp->mnt_ncmountpt;
800 return(cache_fullpath(p, nch, rb, fb));
804 * Sync each mounted filesystem.
808 static int syncprt = 0;
809 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
812 static int sync_callback(struct mount *mp, void *data);
816 sys_sync(struct sync_args *uap)
818 mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
821 * print out buffer pool stat information on each sync() call.
831 sync_callback(struct mount *mp, void *data __unused)
835 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
836 asyncflag = mp->mnt_flag & MNT_ASYNC;
837 mp->mnt_flag &= ~MNT_ASYNC;
838 vfs_msync(mp, MNT_NOWAIT);
839 VFS_SYNC(mp, MNT_NOWAIT);
840 mp->mnt_flag |= asyncflag;
845 /* XXX PRISON: could be per prison flag */
846 static int prison_quotas;
848 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
852 * quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
854 * Change filesystem quotas.
858 sys_quotactl(struct quotactl_args *uap)
860 struct nlookupdata nd;
868 if (p->p_ucred->cr_prison && !prison_quotas)
871 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
873 error = nlookup(&nd);
875 mp = nd.nl_nch.mount;
876 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
877 uap->arg, nd.nl_cred);
884 * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
885 * void *buf, int buflen)
887 * This function operates on a mount point and executes the specified
888 * operation using the specified control data, and possibly returns data.
890 * The actual number of bytes stored in the result buffer is returned, 0
891 * if none, otherwise an error is returned.
895 sys_mountctl(struct mountctl_args *uap)
897 struct thread *td = curthread;
898 struct proc *p = td->td_proc;
906 * Sanity and permissions checks. We must be root.
909 if (p->p_ucred->cr_prison != NULL)
911 if ((error = priv_check(td, PRIV_ROOT)) != 0)
915 * Argument length checks
917 if (uap->ctllen < 0 || uap->ctllen > 1024)
919 if (uap->buflen < 0 || uap->buflen > 16 * 1024)
921 if (uap->path == NULL)
925 * Allocate the necessary buffers and copyin data
927 path = objcache_get(namei_oc, M_WAITOK);
928 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
933 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
934 error = copyin(uap->ctl, ctl, uap->ctllen);
939 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
942 * Validate the descriptor
945 fp = holdfp(p->p_fd, uap->fd, -1);
955 * Execute the internal kernel function and clean up.
957 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
960 if (error == 0 && uap->sysmsg_result > 0)
961 error = copyout(buf, uap->buf, uap->sysmsg_result);
964 objcache_put(namei_oc, path);
973 * Execute a mount control operation by resolving the path to a mount point
974 * and calling vop_mountctl().
976 * Use the mount point from the nch instead of the vnode so nullfs mounts
977 * can properly spike the VOP.
980 kern_mountctl(const char *path, int op, struct file *fp,
981 const void *ctl, int ctllen,
982 void *buf, int buflen, int *res)
986 struct nlookupdata nd;
991 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
993 error = nlookup(&nd);
995 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
996 mp = nd.nl_nch.mount;
1002 * Must be the root of the filesystem
1004 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1008 error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1015 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1017 struct thread *td = curthread;
1018 struct proc *p = td->td_proc;
1021 char *fullpath, *freepath;
1024 if ((error = nlookup(nd)) != 0)
1026 mp = nd->nl_nch.mount;
1028 if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1031 error = mount_path(p, mp, &fullpath, &freepath);
1034 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1035 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1036 kfree(freepath, M_TEMP);
1038 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1039 bcopy(sp, buf, sizeof(*buf));
1040 /* Only root should have access to the fsid's. */
1041 if (priv_check(td, PRIV_ROOT))
1042 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1047 * statfs_args(char *path, struct statfs *buf)
1049 * Get filesystem statistics.
1052 sys_statfs(struct statfs_args *uap)
1054 struct nlookupdata nd;
1058 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1060 error = kern_statfs(&nd, &buf);
1063 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1068 kern_fstatfs(int fd, struct statfs *buf)
1070 struct thread *td = curthread;
1071 struct proc *p = td->td_proc;
1075 char *fullpath, *freepath;
1079 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1081 mp = ((struct vnode *)fp->f_data)->v_mount;
1086 if (fp->f_cred == NULL) {
1091 if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1094 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1096 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1097 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1098 kfree(freepath, M_TEMP);
1100 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1101 bcopy(sp, buf, sizeof(*buf));
1103 /* Only root should have access to the fsid's. */
1104 if (priv_check(td, PRIV_ROOT))
1105 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1113 * fstatfs_args(int fd, struct statfs *buf)
1115 * Get filesystem statistics.
1118 sys_fstatfs(struct fstatfs_args *uap)
1123 error = kern_fstatfs(uap->fd, &buf);
1126 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1131 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1137 if ((error = nlookup(nd)) != 0)
1139 mp = nd->nl_nch.mount;
1140 sp = &mp->mnt_vstat;
1141 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1145 if (mp->mnt_flag & MNT_RDONLY)
1146 sp->f_flag |= ST_RDONLY;
1147 if (mp->mnt_flag & MNT_NOSUID)
1148 sp->f_flag |= ST_NOSUID;
1149 bcopy(sp, buf, sizeof(*buf));
1154 * statfs_args(char *path, struct statfs *buf)
1156 * Get filesystem statistics.
1159 sys_statvfs(struct statvfs_args *uap)
1161 struct nlookupdata nd;
1165 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1167 error = kern_statvfs(&nd, &buf);
1170 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1175 kern_fstatvfs(int fd, struct statvfs *buf)
1177 struct thread *td = curthread;
1178 struct proc *p = td->td_proc;
1185 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1187 mp = ((struct vnode *)fp->f_data)->v_mount;
1192 if (fp->f_cred == NULL) {
1196 sp = &mp->mnt_vstat;
1197 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1201 if (mp->mnt_flag & MNT_RDONLY)
1202 sp->f_flag |= ST_RDONLY;
1203 if (mp->mnt_flag & MNT_NOSUID)
1204 sp->f_flag |= ST_NOSUID;
1206 bcopy(sp, buf, sizeof(*buf));
1214 * fstatfs_args(int fd, struct statfs *buf)
1216 * Get filesystem statistics.
1219 sys_fstatvfs(struct fstatvfs_args *uap)
1224 error = kern_fstatvfs(uap->fd, &buf);
1227 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1232 * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1234 * Get statistics on all filesystems.
1237 struct getfsstat_info {
1238 struct statfs *sfsp;
1246 static int getfsstat_callback(struct mount *, void *);
1250 sys_getfsstat(struct getfsstat_args *uap)
1252 struct thread *td = curthread;
1253 struct proc *p = td->td_proc;
1254 struct getfsstat_info info;
1256 bzero(&info, sizeof(info));
1258 info.maxcount = uap->bufsize / sizeof(struct statfs);
1259 info.sfsp = uap->buf;
1261 info.flags = uap->flags;
1264 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1265 if (info.sfsp && info.count > info.maxcount)
1266 uap->sysmsg_result = info.maxcount;
1268 uap->sysmsg_result = info.count;
1269 return (info.error);
1273 getfsstat_callback(struct mount *mp, void *data)
1275 struct getfsstat_info *info = data;
1281 if (info->sfsp && info->count < info->maxcount) {
1282 if (info->p && !chroot_visible_mnt(mp, info->p))
1287 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1288 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1289 * overrides MNT_WAIT.
1291 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1292 (info->flags & MNT_WAIT)) &&
1293 (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1296 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1298 error = mount_path(info->p, mp, &fullpath, &freepath);
1300 info->error = error;
1303 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1304 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1305 kfree(freepath, M_TEMP);
1307 error = copyout(sp, info->sfsp, sizeof(*sp));
1309 info->error = error;
1319 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1320 long bufsize, int flags)
1322 * Get statistics on all filesystems.
1325 struct getvfsstat_info {
1326 struct statfs *sfsp;
1327 struct statvfs *vsfsp;
1335 static int getvfsstat_callback(struct mount *, void *);
1339 sys_getvfsstat(struct getvfsstat_args *uap)
1341 struct thread *td = curthread;
1342 struct proc *p = td->td_proc;
1343 struct getvfsstat_info info;
1345 bzero(&info, sizeof(info));
1347 info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1348 info.sfsp = uap->buf;
1349 info.vsfsp = uap->vbuf;
1351 info.flags = uap->flags;
1354 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1355 if (info.vsfsp && info.count > info.maxcount)
1356 uap->sysmsg_result = info.maxcount;
1358 uap->sysmsg_result = info.count;
1359 return (info.error);
1363 getvfsstat_callback(struct mount *mp, void *data)
1365 struct getvfsstat_info *info = data;
1367 struct statvfs *vsp;
1372 if (info->vsfsp && info->count < info->maxcount) {
1373 if (info->p && !chroot_visible_mnt(mp, info->p))
1376 vsp = &mp->mnt_vstat;
1379 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1380 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1381 * overrides MNT_WAIT.
1383 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1384 (info->flags & MNT_WAIT)) &&
1385 (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1388 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1390 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1391 (info->flags & MNT_WAIT)) &&
1392 (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1396 if (mp->mnt_flag & MNT_RDONLY)
1397 vsp->f_flag |= ST_RDONLY;
1398 if (mp->mnt_flag & MNT_NOSUID)
1399 vsp->f_flag |= ST_NOSUID;
1401 error = mount_path(info->p, mp, &fullpath, &freepath);
1403 info->error = error;
1406 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1407 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1408 kfree(freepath, M_TEMP);
1410 error = copyout(sp, info->sfsp, sizeof(*sp));
1412 error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1414 info->error = error;
1426 * fchdir_args(int fd)
1428 * Change current working directory to a given file descriptor.
1432 sys_fchdir(struct fchdir_args *uap)
1434 struct thread *td = curthread;
1435 struct proc *p = td->td_proc;
1436 struct filedesc *fdp = p->p_fd;
1437 struct vnode *vp, *ovp;
1440 struct nchandle nch, onch, tnch;
1443 if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1445 vp = (struct vnode *)fp->f_data;
1447 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1448 if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1451 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1457 cache_copy(&fp->f_nchandle, &nch);
1460 * If the ncp has become a mount point, traverse through
1464 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1465 (mp = cache_findmount(&nch)) != NULL
1467 error = nlookup_mp(mp, &tnch);
1469 cache_unlock(&tnch); /* leave ref intact */
1471 vp = tnch.ncp->nc_vp;
1472 error = vget(vp, LK_SHARED);
1473 KKASSERT(error == 0);
1480 onch = fdp->fd_ncdir;
1481 vn_unlock(vp); /* leave ref intact */
1483 fdp->fd_ncdir = nch;
1495 kern_chdir(struct nlookupdata *nd)
1497 struct thread *td = curthread;
1498 struct proc *p = td->td_proc;
1499 struct filedesc *fdp = p->p_fd;
1500 struct vnode *vp, *ovp;
1501 struct nchandle onch;
1504 if ((error = nlookup(nd)) != 0)
1506 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1508 if ((error = vget(vp, LK_SHARED)) != 0)
1511 error = checkvp_chdir(vp, td);
1515 onch = fdp->fd_ncdir;
1516 cache_unlock(&nd->nl_nch); /* leave reference intact */
1517 fdp->fd_ncdir = nd->nl_nch;
1521 cache_zero(&nd->nl_nch);
1529 * chdir_args(char *path)
1531 * Change current working directory (``.'').
1534 sys_chdir(struct chdir_args *uap)
1536 struct nlookupdata nd;
1539 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1541 error = kern_chdir(&nd);
1547 * Helper function for raised chroot(2) security function: Refuse if
1548 * any filedescriptors are open directories.
1551 chroot_refuse_vdir_fds(struct filedesc *fdp)
1558 for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1559 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1561 vp = (struct vnode *)fp->f_data;
1562 if (vp->v_type != VDIR) {
1573 * This sysctl determines if we will allow a process to chroot(2) if it
1574 * has a directory open:
1575 * 0: disallowed for all processes.
1576 * 1: allowed for processes that were not already chroot(2)'ed.
1577 * 2: allowed for all processes.
1580 static int chroot_allow_open_directories = 1;
1582 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1583 &chroot_allow_open_directories, 0, "");
1586 * chroot to the specified namecache entry. We obtain the vp from the
1587 * namecache data. The passed ncp must be locked and referenced and will
1588 * remain locked and referenced on return.
1591 kern_chroot(struct nchandle *nch)
1593 struct thread *td = curthread;
1594 struct proc *p = td->td_proc;
1595 struct filedesc *fdp = p->p_fd;
1600 * Only root can chroot
1602 if ((error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT)) != 0)
1606 * Disallow open directory descriptors (fchdir() breakouts).
1608 if (chroot_allow_open_directories == 0 ||
1609 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1610 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1613 if ((vp = nch->ncp->nc_vp) == NULL)
1616 if ((error = vget(vp, LK_SHARED)) != 0)
1620 * Check the validity of vp as a directory to change to and
1621 * associate it with rdir/jdir.
1623 error = checkvp_chdir(vp, td);
1624 vn_unlock(vp); /* leave reference intact */
1626 vrele(fdp->fd_rdir);
1627 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */
1628 cache_drop(&fdp->fd_nrdir);
1629 cache_copy(nch, &fdp->fd_nrdir);
1630 if (fdp->fd_jdir == NULL) {
1633 cache_copy(nch, &fdp->fd_njdir);
1642 * chroot_args(char *path)
1644 * Change notion of root (``/'') directory.
1648 sys_chroot(struct chroot_args *uap)
1650 struct thread *td = curthread;
1651 struct nlookupdata nd;
1654 KKASSERT(td->td_proc);
1655 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1660 nd.nl_flags |= NLC_EXEC;
1661 error = nlookup(&nd);
1663 error = kern_chroot(&nd.nl_nch);
1669 * Common routine for chroot and chdir. Given a locked, referenced vnode,
1670 * determine whether it is legal to chdir to the vnode. The vnode's state
1671 * is not changed by this call.
1674 checkvp_chdir(struct vnode *vp, struct thread *td)
1678 if (vp->v_type != VDIR)
1681 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1686 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1688 struct thread *td = curthread;
1689 struct proc *p = td->td_proc;
1690 struct lwp *lp = td->td_lwp;
1691 struct filedesc *fdp = p->p_fd;
1696 int type, indx, error;
1699 if ((oflags & O_ACCMODE) == O_ACCMODE)
1701 flags = FFLAGS(oflags);
1702 error = falloc(p, &nfp, NULL);
1706 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1709 * XXX p_dupfd is a real mess. It allows a device to return a
1710 * file descriptor to be duplicated rather then doing the open
1716 * Call vn_open() to do the lookup and assign the vnode to the
1717 * file pointer. vn_open() does not change the ref count on fp
1718 * and the vnode, on success, will be inherited by the file pointer
1721 nd->nl_flags |= NLC_LOCKVP;
1722 error = vn_open(nd, fp, flags, cmode);
1726 * handle special fdopen() case. bleh. dupfdopen() is
1727 * responsible for dropping the old contents of ofiles[indx]
1730 * Note that fsetfd() will add a ref to fp which represents
1731 * the fd_files[] assignment. We must still drop our
1734 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1735 if (fdalloc(p, 0, &indx) == 0) {
1736 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1739 fdrop(fp); /* our ref */
1742 fsetfd(p, NULL, indx);
1745 fdrop(fp); /* our ref */
1746 if (error == ERESTART)
1752 * ref the vnode for ourselves so it can't be ripped out from under
1753 * is. XXX need an ND flag to request that the vnode be returned
1756 * Reserve a file descriptor but do not assign it until the open
1759 vp = (struct vnode *)fp->f_data;
1761 if ((error = fdalloc(p, 0, &indx)) != 0) {
1768 * If no error occurs the vp will have been assigned to the file
1773 if (flags & (O_EXLOCK | O_SHLOCK)) {
1774 lf.l_whence = SEEK_SET;
1777 if (flags & O_EXLOCK)
1778 lf.l_type = F_WRLCK;
1780 lf.l_type = F_RDLCK;
1781 if (flags & FNONBLOCK)
1786 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1788 * lock request failed. Clean up the reserved
1792 fsetfd(p, NULL, indx);
1796 fp->f_flag |= FHASLOCK;
1800 * Assert that all regular file vnodes were created with a object.
1802 KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1803 ("open: regular file has no backing object after vn_open"));
1809 * release our private reference, leaving the one associated with the
1810 * descriptor table intact.
1812 fsetfd(p, fp, indx);
1819 * open_args(char *path, int flags, int mode)
1821 * Check permissions, allocate an open file structure,
1822 * and call the device open routine if any.
1825 sys_open(struct open_args *uap)
1827 struct nlookupdata nd;
1830 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1832 error = kern_open(&nd, uap->flags,
1833 uap->mode, &uap->sysmsg_result);
1840 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1842 struct thread *td = curthread;
1843 struct proc *p = td->td_proc;
1851 switch (mode & S_IFMT) {
1854 error = priv_check(td, PRIV_ROOT);
1857 error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT);
1864 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1865 if ((error = nlookup(nd)) != 0)
1867 if (nd->nl_nch.ncp->nc_vp)
1869 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1873 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1874 vattr.va_rmajor = rmajor;
1875 vattr.va_rminor = rminor;
1878 switch (mode & S_IFMT) {
1879 case S_IFMT: /* used by badsect to flag bad sectors */
1880 vattr.va_type = VBAD;
1883 vattr.va_type = VCHR;
1886 vattr.va_type = VBLK;
1892 /* special directories support for HAMMER */
1893 vattr.va_type = VDIR;
1901 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1902 nd->nl_cred, NAMEI_CREATE);
1905 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1906 &vp, nd->nl_cred, &vattr);
1915 * mknod_args(char *path, int mode, int dev)
1917 * Create a special file.
1920 sys_mknod(struct mknod_args *uap)
1922 struct nlookupdata nd;
1925 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1927 error = kern_mknod(&nd, uap->mode,
1928 umajor(uap->dev), uminor(uap->dev));
1935 kern_mkfifo(struct nlookupdata *nd, int mode)
1937 struct thread *td = curthread;
1938 struct proc *p = td->td_proc;
1945 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1946 if ((error = nlookup(nd)) != 0)
1948 if (nd->nl_nch.ncp->nc_vp)
1950 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1954 vattr.va_type = VFIFO;
1955 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1957 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1964 * mkfifo_args(char *path, int mode)
1966 * Create a named pipe.
1969 sys_mkfifo(struct mkfifo_args *uap)
1971 struct nlookupdata nd;
1974 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1976 error = kern_mkfifo(&nd, uap->mode);
1981 static int hardlink_check_uid = 0;
1982 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1983 &hardlink_check_uid, 0,
1984 "Unprivileged processes cannot create hard links to files owned by other "
1986 static int hardlink_check_gid = 0;
1987 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1988 &hardlink_check_gid, 0,
1989 "Unprivileged processes cannot create hard links to files owned by other "
1993 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1999 * Shortcut if disabled
2001 if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2005 * root cred can always hardlink
2007 if (priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT) == 0)
2011 * Otherwise only if the originating file is owned by the
2012 * same user or group. Note that any group is allowed if
2013 * the file is owned by the caller.
2015 error = VOP_GETATTR(vp, &va);
2019 if (hardlink_check_uid) {
2020 if (cred->cr_uid != va.va_uid)
2024 if (hardlink_check_gid) {
2025 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2033 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2035 struct thread *td = curthread;
2040 * Lookup the source and obtained a locked vnode.
2042 * You may only hardlink a file which you have write permission
2043 * on or which you own.
2045 * XXX relookup on vget failure / race ?
2048 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2049 if ((error = nlookup(nd)) != 0)
2051 vp = nd->nl_nch.ncp->nc_vp;
2052 KKASSERT(vp != NULL);
2053 if (vp->v_type == VDIR)
2054 return (EPERM); /* POSIX */
2055 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2057 if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2061 * Unlock the source so we can lookup the target without deadlocking
2062 * (XXX vp is locked already, possible other deadlock?). The target
2065 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2066 nd->nl_flags &= ~NLC_NCPISLOCKED;
2067 cache_unlock(&nd->nl_nch);
2069 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2070 if ((error = nlookup(linknd)) != 0) {
2074 if (linknd->nl_nch.ncp->nc_vp) {
2080 * Finally run the new API VOP.
2082 error = can_hardlink(vp, td, td->td_proc->p_ucred);
2084 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2085 vp, linknd->nl_cred);
2092 * link_args(char *path, char *link)
2094 * Make a hard file link.
2097 sys_link(struct link_args *uap)
2099 struct nlookupdata nd, linknd;
2102 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2104 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2106 error = kern_link(&nd, &linknd);
2107 nlookup_done(&linknd);
2114 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2122 nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2123 if ((error = nlookup(nd)) != 0)
2125 if (nd->nl_nch.ncp->nc_vp)
2127 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2131 vattr.va_mode = mode;
2132 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2139 * symlink(char *path, char *link)
2141 * Make a symbolic link.
2144 sys_symlink(struct symlink_args *uap)
2146 struct thread *td = curthread;
2147 struct nlookupdata nd;
2152 path = objcache_get(namei_oc, M_WAITOK);
2153 error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2155 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2157 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2158 error = kern_symlink(&nd, path, mode);
2162 objcache_put(namei_oc, path);
2167 * undelete_args(char *path)
2169 * Delete a whiteout from the filesystem.
2173 sys_undelete(struct undelete_args *uap)
2175 struct nlookupdata nd;
2178 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2180 nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2182 error = nlookup(&nd);
2184 error = ncp_writechk(&nd.nl_nch);
2186 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2194 kern_unlink(struct nlookupdata *nd)
2199 nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2200 if ((error = nlookup(nd)) != 0)
2202 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2204 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2209 * unlink_args(char *path)
2211 * Delete a name from the filesystem.
2214 sys_unlink(struct unlink_args *uap)
2216 struct nlookupdata nd;
2219 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2221 error = kern_unlink(&nd);
2227 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2229 struct thread *td = curthread;
2230 struct proc *p = td->td_proc;
2237 fp = holdfp(p->p_fd, fd, -1);
2240 if (fp->f_type != DTYPE_VNODE) {
2244 vp = (struct vnode *)fp->f_data;
2248 new_offset = fp->f_offset + offset;
2252 error = VOP_GETATTR(vp, &vattr);
2253 new_offset = offset + vattr.va_size;
2256 new_offset = offset;
2266 * Validate the seek position. Negative offsets are not allowed
2267 * for regular files or directories.
2269 * Normally we would also not want to allow negative offsets for
2270 * character and block-special devices. However kvm addresses
2271 * on 64 bit architectures might appear to be negative and must
2275 if (new_offset < 0 &&
2276 (vp->v_type == VREG || vp->v_type == VDIR)) {
2279 fp->f_offset = new_offset;
2282 *res = fp->f_offset;
2289 * lseek_args(int fd, int pad, off_t offset, int whence)
2291 * Reposition read/write file offset.
2294 sys_lseek(struct lseek_args *uap)
2298 error = kern_lseek(uap->fd, uap->offset, uap->whence,
2299 &uap->sysmsg_offset);
2305 kern_access(struct nlookupdata *nd, int aflags)
2310 if ((error = nlookup(nd)) != 0)
2313 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2317 /* Flags == 0 means only check for existence. */
2326 if ((flags & VWRITE) == 0 ||
2327 (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2328 error = VOP_ACCESS(vp, flags, nd->nl_cred);
2331 * If the file handle is stale we have to re-resolve the
2332 * entry. This is a hack at the moment.
2334 if (error == ESTALE) {
2336 cache_setunresolved(&nd->nl_nch);
2337 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2350 * access_args(char *path, int flags)
2352 * Check access permissions.
2355 sys_access(struct access_args *uap)
2357 struct nlookupdata nd;
2360 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2362 error = kern_access(&nd, uap->flags);
2368 kern_stat(struct nlookupdata *nd, struct stat *st)
2374 if ((error = nlookup(nd)) != 0)
2377 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2381 if ((error = vget(vp, LK_SHARED)) != 0)
2383 error = vn_stat(vp, st, nd->nl_cred);
2386 * If the file handle is stale we have to re-resolve the entry. This
2387 * is a hack at the moment.
2389 if (error == ESTALE) {
2391 cache_setunresolved(&nd->nl_nch);
2392 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2402 * stat_args(char *path, struct stat *ub)
2404 * Get file status; this version follows links.
2407 sys_stat(struct stat_args *uap)
2409 struct nlookupdata nd;
2413 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2415 error = kern_stat(&nd, &st);
2417 error = copyout(&st, uap->ub, sizeof(*uap->ub));
2424 * lstat_args(char *path, struct stat *ub)
2426 * Get file status; this version does not follow links.
2429 sys_lstat(struct lstat_args *uap)
2431 struct nlookupdata nd;
2435 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2437 error = kern_stat(&nd, &st);
2439 error = copyout(&st, uap->ub, sizeof(*uap->ub));
2446 * pathconf_Args(char *path, int name)
2448 * Get configurable pathname variables.
2452 sys_pathconf(struct pathconf_args *uap)
2454 struct nlookupdata nd;
2459 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2461 error = nlookup(&nd);
2463 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2466 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2474 * kern_readlink isn't properly split yet. There is a copyin burried
2475 * in VOP_READLINK().
2478 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2480 struct thread *td = curthread;
2481 struct proc *p = td->td_proc;
2487 if ((error = nlookup(nd)) != 0)
2489 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2492 if (vp->v_type != VLNK) {
2495 aiov.iov_base = buf;
2496 aiov.iov_len = count;
2497 auio.uio_iov = &aiov;
2498 auio.uio_iovcnt = 1;
2499 auio.uio_offset = 0;
2500 auio.uio_rw = UIO_READ;
2501 auio.uio_segflg = UIO_USERSPACE;
2503 auio.uio_resid = count;
2504 error = VOP_READLINK(vp, &auio, p->p_ucred);
2507 *res = count - auio.uio_resid;
2512 * readlink_args(char *path, char *buf, int count)
2514 * Return target name of a symbolic link.
2517 sys_readlink(struct readlink_args *uap)
2519 struct nlookupdata nd;
2522 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2524 error = kern_readlink(&nd, uap->buf, uap->count,
2525 &uap->sysmsg_result);
2532 setfflags(struct vnode *vp, int flags)
2534 struct thread *td = curthread;
2535 struct proc *p = td->td_proc;
2540 * Prevent non-root users from setting flags on devices. When
2541 * a device is reused, users can retain ownership of the device
2542 * if they are allowed to set flags and programs assume that
2543 * chown can't fail when done as root.
2545 if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2546 ((error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT)) != 0))
2550 * note: vget is required for any operation that might mod the vnode
2551 * so VINACTIVE is properly cleared.
2553 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2555 vattr.va_flags = flags;
2556 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2563 * chflags(char *path, int flags)
2565 * Change flags of a file given a path name.
2569 sys_chflags(struct chflags_args *uap)
2571 struct nlookupdata nd;
2576 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2578 error = nlookup(&nd);
2580 error = ncp_writechk(&nd.nl_nch);
2582 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2585 error = setfflags(vp, uap->flags);
2592 * lchflags(char *path, int flags)
2594 * Change flags of a file given a path name, but don't follow symlinks.
2598 sys_lchflags(struct lchflags_args *uap)
2600 struct nlookupdata nd;
2605 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2607 error = nlookup(&nd);
2609 error = ncp_writechk(&nd.nl_nch);
2611 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2614 error = setfflags(vp, uap->flags);
2621 * fchflags_args(int fd, int flags)
2623 * Change flags of a file given a file descriptor.
2627 sys_fchflags(struct fchflags_args *uap)
2629 struct thread *td = curthread;
2630 struct proc *p = td->td_proc;
2634 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2636 if (fp->f_nchandle.ncp)
2637 error = ncp_writechk(&fp->f_nchandle);
2639 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2645 setfmode(struct vnode *vp, int mode)
2647 struct thread *td = curthread;
2648 struct proc *p = td->td_proc;
2653 * note: vget is required for any operation that might mod the vnode
2654 * so VINACTIVE is properly cleared.
2656 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2658 vattr.va_mode = mode & ALLPERMS;
2659 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2666 kern_chmod(struct nlookupdata *nd, int mode)
2671 if ((error = nlookup(nd)) != 0)
2673 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2675 if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2676 error = setfmode(vp, mode);
2682 * chmod_args(char *path, int mode)
2684 * Change mode of a file given path name.
2688 sys_chmod(struct chmod_args *uap)
2690 struct nlookupdata nd;
2693 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2695 error = kern_chmod(&nd, uap->mode);
2701 * lchmod_args(char *path, int mode)
2703 * Change mode of a file given path name (don't follow links.)
2707 sys_lchmod(struct lchmod_args *uap)
2709 struct nlookupdata nd;
2712 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2714 error = kern_chmod(&nd, uap->mode);
2720 * fchmod_args(int fd, int mode)
2722 * Change mode of a file given a file descriptor.
2726 sys_fchmod(struct fchmod_args *uap)
2728 struct thread *td = curthread;
2729 struct proc *p = td->td_proc;
2733 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2735 if (fp->f_nchandle.ncp)
2736 error = ncp_writechk(&fp->f_nchandle);
2738 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2744 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2746 struct thread *td = curthread;
2747 struct proc *p = td->td_proc;
2752 * note: vget is required for any operation that might mod the vnode
2753 * so VINACTIVE is properly cleared.
2755 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2759 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2766 kern_chown(struct nlookupdata *nd, int uid, int gid)
2771 if ((error = nlookup(nd)) != 0)
2773 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2775 if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2776 error = setfown(vp, uid, gid);
2782 * chown(char *path, int uid, int gid)
2784 * Set ownership given a path name.
2787 sys_chown(struct chown_args *uap)
2789 struct nlookupdata nd;
2792 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2794 error = kern_chown(&nd, uap->uid, uap->gid);
2800 * lchown_args(char *path, int uid, int gid)
2802 * Set ownership given a path name, do not cross symlinks.
2805 sys_lchown(struct lchown_args *uap)
2807 struct nlookupdata nd;
2810 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2812 error = kern_chown(&nd, uap->uid, uap->gid);
2818 * fchown_args(int fd, int uid, int gid)
2820 * Set ownership given a file descriptor.
2824 sys_fchown(struct fchown_args *uap)
2826 struct thread *td = curthread;
2827 struct proc *p = td->td_proc;
2831 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2833 if (fp->f_nchandle.ncp)
2834 error = ncp_writechk(&fp->f_nchandle);
2836 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2842 getutimes(const struct timeval *tvp, struct timespec *tsp)
2844 struct timeval tv[2];
2848 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2851 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2852 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2858 setutimes(struct vnode *vp, struct vattr *vattr,
2859 const struct timespec *ts, int nullflag)
2861 struct thread *td = curthread;
2862 struct proc *p = td->td_proc;
2866 vattr->va_atime = ts[0];
2867 vattr->va_mtime = ts[1];
2869 vattr->va_vaflags |= VA_UTIMES_NULL;
2870 error = VOP_SETATTR(vp, vattr, p->p_ucred);
2876 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2878 struct timespec ts[2];
2883 if ((error = getutimes(tptr, ts)) != 0)
2887 * NOTE: utimes() succeeds for the owner even if the file
2888 * is not user-writable.
2890 nd->nl_flags |= NLC_OWN | NLC_WRITE;
2892 if ((error = nlookup(nd)) != 0)
2894 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2896 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2900 * note: vget is required for any operation that might mod the vnode
2901 * so VINACTIVE is properly cleared.
2903 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
2904 error = vget(vp, LK_EXCLUSIVE);
2906 error = setutimes(vp, &vattr, ts, (tptr == NULL));
2915 * utimes_args(char *path, struct timeval *tptr)
2917 * Set the access and modification times of a file.
2920 sys_utimes(struct utimes_args *uap)
2922 struct timeval tv[2];
2923 struct nlookupdata nd;
2927 error = copyin(uap->tptr, tv, sizeof(tv));
2931 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2933 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2939 * lutimes_args(char *path, struct timeval *tptr)
2941 * Set the access and modification times of a file.
2944 sys_lutimes(struct lutimes_args *uap)
2946 struct timeval tv[2];
2947 struct nlookupdata nd;
2951 error = copyin(uap->tptr, tv, sizeof(tv));
2955 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2957 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2963 * Set utimes on a file descriptor. The creds used to open the
2964 * file are used to determine whether the operation is allowed
2968 kern_futimes(int fd, struct timeval *tptr)
2970 struct thread *td = curthread;
2971 struct proc *p = td->td_proc;
2972 struct timespec ts[2];
2978 error = getutimes(tptr, ts);
2981 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2983 if (fp->f_nchandle.ncp)
2984 error = ncp_writechk(&fp->f_nchandle);
2987 error = vget(vp, LK_EXCLUSIVE);
2989 error = VOP_GETATTR(vp, &vattr);
2991 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
2995 error = setutimes(vp, &vattr, ts,
3006 * futimes_args(int fd, struct timeval *tptr)
3008 * Set the access and modification times of a file.
3011 sys_futimes(struct futimes_args *uap)
3013 struct timeval tv[2];
3017 error = copyin(uap->tptr, tv, sizeof(tv));
3022 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3028 kern_truncate(struct nlookupdata *nd, off_t length)
3036 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3037 if ((error = nlookup(nd)) != 0)
3039 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3041 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3043 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3047 if (vp->v_type == VDIR) {
3049 } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3051 vattr.va_size = length;
3052 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3059 * truncate(char *path, int pad, off_t length)
3061 * Truncate a file given its path name.
3064 sys_truncate(struct truncate_args *uap)
3066 struct nlookupdata nd;
3069 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3071 error = kern_truncate(&nd, uap->length);
3077 kern_ftruncate(int fd, off_t length)
3079 struct thread *td = curthread;
3080 struct proc *p = td->td_proc;
3088 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3090 if (fp->f_nchandle.ncp) {
3091 error = ncp_writechk(&fp->f_nchandle);
3095 if ((fp->f_flag & FWRITE) == 0) {
3099 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */
3103 vp = (struct vnode *)fp->f_data;
3104 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3105 if (vp->v_type == VDIR) {
3107 } else if ((error = vn_writechk(vp, NULL)) == 0) {
3109 vattr.va_size = length;
3110 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3119 * ftruncate_args(int fd, int pad, off_t length)
3121 * Truncate a file given a file descriptor.
3124 sys_ftruncate(struct ftruncate_args *uap)
3128 error = kern_ftruncate(uap->fd, uap->length);
3136 * Sync an open file.
3140 sys_fsync(struct fsync_args *uap)
3142 struct thread *td = curthread;
3143 struct proc *p = td->td_proc;
3149 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3151 vp = (struct vnode *)fp->f_data;
3152 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3153 if ((obj = vp->v_object) != NULL)
3154 vm_object_page_clean(obj, 0, 0, 0);
3155 if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
3156 error = buf_fsync(vp);
3163 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3165 struct nchandle fnchd;
3166 struct nchandle tnchd;
3167 struct namecache *ncp;
3174 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3175 if ((error = nlookup(fromnd)) != 0)
3177 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3179 fnchd.mount = fromnd->nl_nch.mount;
3183 * unlock the source nch so we can lookup the target nch without
3184 * deadlocking. The target may or may not exist so we do not check
3185 * for a target vp like kern_mkdir() and other creation functions do.
3187 * The source and target directories are ref'd and rechecked after
3188 * everything is relocked to determine if the source or target file
3191 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3192 fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3193 cache_unlock(&fromnd->nl_nch);
3195 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3196 if ((error = nlookup(tond)) != 0) {
3200 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3204 tnchd.mount = tond->nl_nch.mount;
3208 * If the source and target are the same there is nothing to do
3210 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3217 * Mount points cannot be renamed or overwritten
3219 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3228 * relock the source ncp. NOTE AFTER RELOCKING: the source ncp
3229 * may have become invalid while it was unlocked, nc_vp and nc_mount
3232 if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3233 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3234 } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3235 cache_lock(&fromnd->nl_nch);
3236 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3238 cache_unlock(&tond->nl_nch);
3239 cache_lock(&fromnd->nl_nch);
3240 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3241 cache_lock(&tond->nl_nch);
3242 cache_resolve(&tond->nl_nch, tond->nl_cred);
3244 fromnd->nl_flags |= NLC_NCPISLOCKED;
3247 * make sure the parent directories linkages are the same
3249 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3250 tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3257 * Both the source and target must be within the same filesystem and
3258 * in the same filesystem as their parent directories within the
3259 * namecache topology.
3261 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3264 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3265 mp != tond->nl_nch.mount) {
3272 * Make sure the mount point is writable
3274 if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3281 * If the target exists and either the source or target is a directory,
3282 * then both must be directories.
3284 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3287 if (tond->nl_nch.ncp->nc_vp) {
3288 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3290 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3291 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3293 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3299 * You cannot rename a source into itself or a subdirectory of itself.
3300 * We check this by travsersing the target directory upwards looking
3301 * for a match against the source.
3304 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3305 if (fromnd->nl_nch.ncp == ncp) {
3316 * Even though the namespaces are different, they may still represent
3317 * hardlinks to the same file. The filesystem might have a hard time
3318 * with this so we issue a NREMOVE of the source instead of a NRENAME
3319 * when we detect the situation.
3322 fdvp = fromnd->nl_dvp;
3323 tdvp = tond->nl_dvp;
3324 if (fdvp == NULL || tdvp == NULL) {
3326 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3327 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3330 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3331 fdvp, tdvp, tond->nl_cred);
3338 * rename_args(char *from, char *to)
3340 * Rename files. Source and destination must either both be directories,
3341 * or both not be directories. If target is a directory, it must be empty.
3344 sys_rename(struct rename_args *uap)
3346 struct nlookupdata fromnd, tond;
3349 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3351 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3353 error = kern_rename(&fromnd, &tond);
3354 nlookup_done(&tond);
3356 nlookup_done(&fromnd);
3361 kern_mkdir(struct nlookupdata *nd, int mode)
3363 struct thread *td = curthread;
3364 struct proc *p = td->td_proc;
3370 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3371 if ((error = nlookup(nd)) != 0)
3374 if (nd->nl_nch.ncp->nc_vp)
3376 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3379 vattr.va_type = VDIR;
3380 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3383 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3390 * mkdir_args(char *path, int mode)
3392 * Make a directory file.
3396 sys_mkdir(struct mkdir_args *uap)
3398 struct nlookupdata nd;
3401 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3403 error = kern_mkdir(&nd, uap->mode);
3409 kern_rmdir(struct nlookupdata *nd)
3414 nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3415 if ((error = nlookup(nd)) != 0)
3419 * Do not allow directories representing mount points to be
3420 * deleted, even if empty. Check write perms on mount point
3421 * in case the vnode is aliased (aka nullfs).
3423 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3425 if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3427 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3432 * rmdir_args(char *path)
3434 * Remove a directory file.
3438 sys_rmdir(struct rmdir_args *uap)
3440 struct nlookupdata nd;
3443 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3445 error = kern_rmdir(&nd);
3451 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3452 enum uio_seg direction)
3454 struct thread *td = curthread;
3455 struct proc *p = td->td_proc;
3463 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3465 if ((fp->f_flag & FREAD) == 0) {
3469 vp = (struct vnode *)fp->f_data;
3471 if (vp->v_type != VDIR) {
3475 aiov.iov_base = buf;
3476 aiov.iov_len = count;
3477 auio.uio_iov = &aiov;
3478 auio.uio_iovcnt = 1;
3479 auio.uio_rw = UIO_READ;
3480 auio.uio_segflg = direction;
3482 auio.uio_resid = count;
3483 loff = auio.uio_offset = fp->f_offset;
3484 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3485 fp->f_offset = auio.uio_offset;
3488 if (count == auio.uio_resid) {
3489 if (union_dircheckp) {
3490 error = union_dircheckp(td, &vp, fp);
3497 if ((vp->v_flag & VROOT) &&
3498 (vp->v_mount->mnt_flag & MNT_UNION)) {
3499 struct vnode *tvp = vp;
3500 vp = vp->v_mount->mnt_vnodecovered;
3511 * WARNING! *basep may not be wide enough to accomodate the
3512 * seek offset. XXX should we hack this to return the upper 32 bits
3513 * for offsets greater then 4G?
3516 *basep = (long)loff;
3518 *res = count - auio.uio_resid;
3525 * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3527 * Read a block of directory entries in a file system independent format.
3530 sys_getdirentries(struct getdirentries_args *uap)
3535 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3536 &uap->sysmsg_result, UIO_USERSPACE);
3538 if (error == 0 && uap->basep)
3539 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3544 * getdents_args(int fd, char *buf, size_t count)
3547 sys_getdents(struct getdents_args *uap)
3551 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3552 &uap->sysmsg_result, UIO_USERSPACE);
3558 * umask(int newmask)
3560 * Set the mode mask for creation of filesystem nodes.
3565 sys_umask(struct umask_args *uap)
3567 struct thread *td = curthread;
3568 struct proc *p = td->td_proc;
3569 struct filedesc *fdp;
3572 uap->sysmsg_result = fdp->fd_cmask;
3573 fdp->fd_cmask = uap->newmask & ALLPERMS;
3578 * revoke(char *path)
3580 * Void all references to file by ripping underlying filesystem
3585 sys_revoke(struct revoke_args *uap)
3587 struct nlookupdata nd;
3594 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3596 error = nlookup(&nd);
3598 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3599 cred = crhold(nd.nl_cred);
3603 error = VOP_GETATTR(vp, &vattr);
3604 if (error == 0 && cred->cr_uid != vattr.va_uid)
3605 error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT);
3606 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
3607 if (count_udev(vp->v_umajor, vp->v_uminor) > 0)
3608 error = vrevoke(vp, cred);
3609 } else if (error == 0) {
3610 error = vrevoke(vp, cred);
3620 * getfh_args(char *fname, fhandle_t *fhp)
3622 * Get (NFS) file handle
3624 * NOTE: We use the fsid of the covering mount, even if it is a nullfs
3625 * mount. This allows nullfs mounts to be explicitly exported.
3627 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
3629 * nullfs mounts of subdirectories are not safe. That is, it will
3630 * work, but you do not really have protection against access to
3631 * the related parent directories.
3634 sys_getfh(struct getfh_args *uap)
3636 struct thread *td = curthread;
3637 struct nlookupdata nd;
3644 * Must be super user
3646 if ((error = priv_check(td, PRIV_ROOT)) != 0)
3650 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3652 error = nlookup(&nd);
3654 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3655 mp = nd.nl_nch.mount;
3658 bzero(&fh, sizeof(fh));
3659 fh.fh_fsid = mp->mnt_stat.f_fsid;
3660 error = VFS_VPTOFH(vp, &fh.fh_fid);
3663 error = copyout(&fh, uap->fhp, sizeof(fh));
3669 * fhopen_args(const struct fhandle *u_fhp, int flags)
3671 * syscall for the rpc.lockd to use to translate a NFS file handle into
3672 * an open descriptor.
3674 * warning: do not remove the priv_check() call or this becomes one giant
3678 sys_fhopen(struct fhopen_args *uap)
3680 struct thread *td = curthread;
3681 struct proc *p = td->td_proc;
3686 struct vattr *vap = &vat;
3688 int fmode, mode, error, type;
3694 * Must be super user
3696 error = priv_check(td, PRIV_ROOT);
3700 fmode = FFLAGS(uap->flags);
3701 /* why not allow a non-read/write open for our lockd? */
3702 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3704 error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3707 /* find the mount point */
3708 mp = vfs_getvfs(&fhp.fh_fsid);
3711 /* now give me my vnode, it gets returned to me locked */
3712 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
3716 * from now on we have to make sure not
3717 * to forget about the vnode
3718 * any error that causes an abort must vput(vp)
3719 * just set error = err and 'goto bad;'.
3725 if (vp->v_type == VLNK) {
3729 if (vp->v_type == VSOCK) {
3734 if (fmode & (FWRITE | O_TRUNC)) {
3735 if (vp->v_type == VDIR) {
3739 error = vn_writechk(vp, NULL);
3747 error = VOP_ACCESS(vp, mode, p->p_ucred);
3751 if (fmode & O_TRUNC) {
3752 vn_unlock(vp); /* XXX */
3753 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */
3756 error = VOP_SETATTR(vp, vap, p->p_ucred);
3762 * VOP_OPEN needs the file pointer so it can potentially override
3765 * WARNING! no f_nchandle will be associated when fhopen()ing a
3768 if ((error = falloc(p, &nfp, &indx)) != 0)
3772 error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3775 * setting f_ops this way prevents VOP_CLOSE from being
3776 * called or fdrop() releasing the vp from v_data. Since
3777 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3779 fp->f_ops = &badfileops;
3785 * The fp is given its own reference, we still have our ref and lock.
3787 * Assert that all regular files must be created with a VM object.
3789 if (vp->v_type == VREG && vp->v_object == NULL) {
3790 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3795 * The open was successful. Handle any locking requirements.
3797 if (fmode & (O_EXLOCK | O_SHLOCK)) {
3798 lf.l_whence = SEEK_SET;
3801 if (fmode & O_EXLOCK)
3802 lf.l_type = F_WRLCK;
3804 lf.l_type = F_RDLCK;
3805 if (fmode & FNONBLOCK)
3810 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3812 * release our private reference.
3814 fsetfd(p, NULL, indx);
3819 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3820 fp->f_flag |= FHASLOCK;
3824 * Clean up. Associate the file pointer with the previously
3825 * reserved descriptor and return it.
3828 fsetfd(p, fp, indx);
3830 uap->sysmsg_result = indx;
3834 fsetfd(p, NULL, indx);
3842 * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3845 sys_fhstat(struct fhstat_args *uap)
3847 struct thread *td = curthread;
3855 * Must be super user
3857 error = priv_check(td, PRIV_ROOT);
3861 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3865 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3867 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3869 error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3873 error = copyout(&sb, uap->sb, sizeof(sb));
3878 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3881 sys_fhstatfs(struct fhstatfs_args *uap)
3883 struct thread *td = curthread;
3884 struct proc *p = td->td_proc;
3889 char *fullpath, *freepath;
3894 * Must be super user
3896 if ((error = priv_check(td, PRIV_ROOT)))
3899 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3902 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3905 if (p != NULL && !chroot_visible_mnt(mp, p))
3908 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3913 if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3916 error = mount_path(p, mp, &fullpath, &freepath);
3919 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3920 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3921 kfree(freepath, M_TEMP);
3923 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3924 if (priv_check(td, PRIV_ROOT)) {
3925 bcopy(sp, &sb, sizeof(sb));
3926 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3929 return (copyout(sp, uap->buf, sizeof(*sp)));
3933 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
3936 sys_fhstatvfs(struct fhstatvfs_args *uap)
3938 struct thread *td = curthread;
3939 struct proc *p = td->td_proc;
3947 * Must be super user
3949 if ((error = priv_check(td, PRIV_ROOT)))
3952 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3955 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3958 if (p != NULL && !chroot_visible_mnt(mp, p))
3961 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3964 sp = &mp->mnt_vstat;
3966 if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
3970 if (mp->mnt_flag & MNT_RDONLY)
3971 sp->f_flag |= ST_RDONLY;
3972 if (mp->mnt_flag & MNT_NOSUID)
3973 sp->f_flag |= ST_NOSUID;
3975 return (copyout(sp, uap->buf, sizeof(*sp)));
3980 * Syscall to push extended attribute configuration information into the
3981 * VFS. Accepts a path, which it converts to a mountpoint, as well as
3982 * a command (int cmd), and attribute name and misc data. For now, the
3983 * attribute name is left in userspace for consumption by the VFS_op.
3984 * It will probably be changed to be copied into sysspace by the
3985 * syscall in the future, once issues with various consumers of the
3986 * attribute code have raised their hands.
3988 * Currently this is used only by UFS Extended Attributes.
3991 sys_extattrctl(struct extattrctl_args *uap)
3993 struct nlookupdata nd;
3999 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4001 error = nlookup(&nd);
4003 mp = nd.nl_nch.mount;
4004 error = VFS_EXTATTRCTL(mp, uap->cmd,
4005 uap->attrname, uap->arg,
4013 * Syscall to set a named extended attribute on a file or directory.
4014 * Accepts attribute name, and a uio structure pointing to the data to set.
4015 * The uio is consumed in the style of writev(). The real work happens
4016 * in VOP_SETEXTATTR().
4019 sys_extattr_set_file(struct extattr_set_file_args *uap)
4021 char attrname[EXTATTR_MAXNAMELEN];
4022 struct iovec aiov[UIO_SMALLIOV];
4023 struct iovec *needfree;
4024 struct nlookupdata nd;
4033 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4038 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4040 error = nlookup(&nd);
4042 error = ncp_writechk(&nd.nl_nch);
4044 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4051 iovlen = uap->iovcnt * sizeof(struct iovec);
4052 if (uap->iovcnt > UIO_SMALLIOV) {
4053 if (uap->iovcnt > UIO_MAXIOV) {
4057 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4063 auio.uio_iovcnt = uap->iovcnt;
4064 auio.uio_rw = UIO_WRITE;
4065 auio.uio_segflg = UIO_USERSPACE;
4066 auio.uio_td = nd.nl_td;
4067 auio.uio_offset = 0;
4068 if ((error = copyin(uap->iovp, iov, iovlen)))
4071 for (i = 0; i < uap->iovcnt; i++) {
4072 if (iov->iov_len > INT_MAX - auio.uio_resid) {
4076 auio.uio_resid += iov->iov_len;
4079 cnt = auio.uio_resid;
4080 error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4081 cnt -= auio.uio_resid;
4082 uap->sysmsg_result = cnt;
4087 FREE(needfree, M_IOV);
4092 * Syscall to get a named extended attribute on a file or directory.
4093 * Accepts attribute name, and a uio structure pointing to a buffer for the
4094 * data. The uio is consumed in the style of readv(). The real work
4095 * happens in VOP_GETEXTATTR();
4098 sys_extattr_get_file(struct extattr_get_file_args *uap)
4100 char attrname[EXTATTR_MAXNAMELEN];
4101 struct iovec aiov[UIO_SMALLIOV];
4102 struct iovec *needfree;
4103 struct nlookupdata nd;
4112 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4117 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4119 error = nlookup(&nd);
4121 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4127 iovlen = uap->iovcnt * sizeof (struct iovec);
4129 if (uap->iovcnt > UIO_SMALLIOV) {
4130 if (uap->iovcnt > UIO_MAXIOV) {
4134 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4140 auio.uio_iovcnt = uap->iovcnt;
4141 auio.uio_rw = UIO_READ;
4142 auio.uio_segflg = UIO_USERSPACE;
4143 auio.uio_td = nd.nl_td;
4144 auio.uio_offset = 0;
4145 if ((error = copyin(uap->iovp, iov, iovlen)))
4148 for (i = 0; i < uap->iovcnt; i++) {
4149 if (iov->iov_len > INT_MAX - auio.uio_resid) {
4153 auio.uio_resid += iov->iov_len;
4156 cnt = auio.uio_resid;
4157 error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4158 cnt -= auio.uio_resid;
4159 uap->sysmsg_result = cnt;
4164 FREE(needfree, M_IOV);
4169 * Syscall to delete a named extended attribute from a file or directory.
4170 * Accepts attribute name. The real work happens in VOP_SETEXTATTR().
4173 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4175 char attrname[EXTATTR_MAXNAMELEN];
4176 struct nlookupdata nd;
4180 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4185 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4187 error = nlookup(&nd);
4189 error = ncp_writechk(&nd.nl_nch);
4191 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4197 error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4204 * Determine if the mount is visible to the process.
4207 chroot_visible_mnt(struct mount *mp, struct proc *p)
4209 struct nchandle nch;
4212 * Traverse from the mount point upwards. If we hit the process
4213 * root then the mount point is visible to the process.
4215 nch = mp->mnt_ncmountpt;
4217 if (nch.mount == p->p_fd->fd_nrdir.mount &&
4218 nch.ncp == p->p_fd->fd_nrdir.ncp) {
4221 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4222 nch = nch.mount->mnt_ncmounton;
4224 nch.ncp = nch.ncp->nc_parent;
4229 * If the mount point is not visible to the process, but the
4230 * process root is in a subdirectory of the mount, return
4233 if (p->p_fd->fd_nrdir.mount == mp)