sys/kern/vfs_syscalls.c

   1 /*
   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.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  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.
  25  *
  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
  36  * SUCH DAMAGE.
  37  *
  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 1.124 2008/01/04 12:16:19 matthias Exp $
  41  */
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/buf.h>
  46 #include <sys/conf.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>
  55 #include <sys/file.h>
  56 #include <sys/linker.h>
  57 #include <sys/stat.h>
  58 #include <sys/unistd.h>
  59 #include <sys/vnode.h>
  60 #include <sys/proc.h>
  61 #include <sys/namei.h>
  62 #include <sys/nlookup.h>
  63 #include <sys/dirent.h>
  64 #include <sys/extattr.h>
  65 #include <sys/spinlock.h>
  66 #include <sys/kern_syscall.h>
  67 #include <sys/objcache.h>
  68 #include <sys/sysctl.h>
  69
  70 #include <sys/buf2.h>
  71 #include <sys/file2.h>
  72 #include <sys/spinlock2.h>
  73
  74 #include <vm/vm.h>
  75 #include <vm/vm_object.h>
  76 #include <vm/vm_page.h>
  77
  78 #include <machine/limits.h>
  79 #include <machine/stdarg.h>
  80
  81 #include <vfs/union/union.h>
  82
  83 static void mount_warning(struct mount *mp, const char *ctl, ...);
  84 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
  85 static int checkvp_chdir (struct vnode *vn, struct thread *td);
  86 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
  87 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
  88 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
  89 static int getutimes (const struct timeval *, struct timespec *);
  90 static int setfown (struct vnode *, uid_t, gid_t);
  91 static int setfmode (struct vnode *, int);
  92 static int setfflags (struct vnode *, int);
  93 static int setutimes (struct vnode *, const struct timespec *, int);
  94 static int      usermount = 0;  /* if 1, non-root can mount fs. */
  95
  96 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
  97
  98 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
  99
 100 /*
 101  * Virtual File System System Calls
 102  */
 103
 104 /*
 105  * Mount a file system.
 106  */
 107 /*
 108  * mount_args(char *type, char *path, int flags, caddr_t data)
 109  */
 110 /* ARGSUSED */
 111 int
 112 sys_mount(struct mount_args *uap)
 113 {
 114         struct thread *td = curthread;
 115         struct proc *p = td->td_proc;
 116         struct vnode *vp;
 117         struct nchandle nch;
 118         struct mount *mp;
 119         struct vfsconf *vfsp;
 120         int error, flag = 0, flag2 = 0;
 121         int hasmount;
 122         struct vattr va;
 123         struct nlookupdata nd;
 124         char fstypename[MFSNAMELEN];
 125         struct ucred *cred = p->p_ucred;
 126
 127         KKASSERT(p);
 128         if (cred->cr_prison != NULL)
 129                 return (EPERM);
 130         if (usermount == 0 && (error = suser(td)))
 131                 return (error);
 132         /*
 133          * Do not allow NFS export by non-root users.
 134          */
 135         if (uap->flags & MNT_EXPORTED) {
 136                 error = suser(td);
 137                 if (error)
 138                         return (error);
 139         }
 140         /*
 141          * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
 142          */
 143         if (suser(td))
 144                 uap->flags |= MNT_NOSUID | MNT_NODEV;
 145
 146         /*
 147          * Lookup the requested path and extract the nch and vnode.
 148          */
 149         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 150         if (error == 0) {
 151                 if ((error = nlookup(&nd)) == 0) {
 152                         if (nd.nl_nch.ncp->nc_vp == NULL)
 153                                 error = ENOENT;
 154                 }
 155         }
 156         if (error) {
 157                 nlookup_done(&nd);
 158                 return (error);
 159         }
 160
 161         /*
 162          * Extract the locked+refd ncp and cleanup the nd structure
 163          */
 164         nch = nd.nl_nch;
 165         cache_zero(&nd.nl_nch);
 166         nlookup_done(&nd);
 167
 168         if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
 169                 hasmount = 1;
 170         else
 171                 hasmount = 0;
 172
 173
 174         /*
 175          * now we have the locked ref'd nch and unreferenced vnode.
 176          */
 177         vp = nch.ncp->nc_vp;
 178         if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
 179                 cache_put(&nch);
 180                 return (error);
 181         }
 182         cache_unlock(&nch);
 183
 184         /*
 185          * Now we have an unlocked ref'd nch and a locked ref'd vp
 186          */
 187         if (uap->flags & MNT_UPDATE) {
 188                 if ((vp->v_flag & VROOT) == 0) {
 189                         cache_drop(&nch);
 190                         vput(vp);
 191                         return (EINVAL);
 192                 }
 193                 mp = vp->v_mount;
 194                 flag = mp->mnt_flag;
 195                 flag2 = mp->mnt_kern_flag;
 196                 /*
 197                  * We only allow the filesystem to be reloaded if it
 198                  * is currently mounted read-only.
 199                  */
 200                 if ((uap->flags & MNT_RELOAD) &&
 201                     ((mp->mnt_flag & MNT_RDONLY) == 0)) {
 202                         cache_drop(&nch);
 203                         vput(vp);
 204                         return (EOPNOTSUPP);    /* Needs translation */
 205                 }
 206                 /*
 207                  * Only root, or the user that did the original mount is
 208                  * permitted to update it.
 209                  */
 210                 if (mp->mnt_stat.f_owner != cred->cr_uid &&
 211                     (error = suser(td))) {
 212                         cache_drop(&nch);
 213                         vput(vp);
 214                         return (error);
 215                 }
 216                 if (vfs_busy(mp, LK_NOWAIT)) {
 217                         cache_drop(&nch);
 218                         vput(vp);
 219                         return (EBUSY);
 220                 }
 221                 if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 222                         cache_drop(&nch);
 223                         vfs_unbusy(mp);
 224                         vput(vp);
 225                         return (EBUSY);
 226                 }
 227                 vp->v_flag |= VMOUNT;
 228                 mp->mnt_flag |=
 229                     uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
 230                 vn_unlock(vp);
 231                 goto update;
 232         }
 233         /*
 234          * If the user is not root, ensure that they own the directory
 235          * onto which we are attempting to mount.
 236          */
 237         if ((error = VOP_GETATTR(vp, &va)) ||
 238             (va.va_uid != cred->cr_uid && (error = suser(td)))) {
 239                 cache_drop(&nch);
 240                 vput(vp);
 241                 return (error);
 242         }
 243         if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
 244                 cache_drop(&nch);
 245                 vput(vp);
 246                 return (error);
 247         }
 248         if (vp->v_type != VDIR) {
 249                 cache_drop(&nch);
 250                 vput(vp);
 251                 return (ENOTDIR);
 252         }
 253         if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
 254                 cache_drop(&nch);
 255                 vput(vp);
 256                 return (EPERM);
 257         }
 258         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
 259                 cache_drop(&nch);
 260                 vput(vp);
 261                 return (error);
 262         }
 263         for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 264                 if (!strcmp(vfsp->vfc_name, fstypename))
 265                         break;
 266         }
 267         if (vfsp == NULL) {
 268                 linker_file_t lf;
 269
 270                 /* Only load modules for root (very important!) */
 271                 if ((error = suser(td)) != 0) {
 272                         cache_drop(&nch);
 273                         vput(vp);
 274                         return error;
 275                 }
 276                 error = linker_load_file(fstypename, &lf);
 277                 if (error || lf == NULL) {
 278                         cache_drop(&nch);
 279                         vput(vp);
 280                         if (lf == NULL)
 281                                 error = ENODEV;
 282                         return error;
 283                 }
 284                 lf->userrefs++;
 285                 /* lookup again, see if the VFS was loaded */
 286                 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 287                         if (!strcmp(vfsp->vfc_name, fstypename))
 288                                 break;
 289                 }
 290                 if (vfsp == NULL) {
 291                         lf->userrefs--;
 292                         linker_file_unload(lf);
 293                         cache_drop(&nch);
 294                         vput(vp);
 295                         return (ENODEV);
 296                 }
 297         }
 298         if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 299                 cache_drop(&nch);
 300                 vput(vp);
 301                 return (EBUSY);
 302         }
 303         vp->v_flag |= VMOUNT;
 304
 305         /*
 306          * Allocate and initialize the filesystem.
 307          */
 308         mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
 309         TAILQ_INIT(&mp->mnt_nvnodelist);
 310         TAILQ_INIT(&mp->mnt_reservedvnlist);
 311         TAILQ_INIT(&mp->mnt_jlist);
 312         mp->mnt_nvnodelistsize = 0;
 313         lockinit(&mp->mnt_lock, "vfslock", 0, 0);
 314         vfs_busy(mp, LK_NOWAIT);
 315         mp->mnt_op = vfsp->vfc_vfsops;
 316         mp->mnt_vfc = vfsp;
 317         vfsp->vfc_refcount++;
 318         mp->mnt_stat.f_type = vfsp->vfc_typenum;
 319         mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
 320         strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
 321         mp->mnt_stat.f_owner = cred->cr_uid;
 322         mp->mnt_iosize_max = DFLTPHYS;
 323         vn_unlock(vp);
 324 update:
 325         /*
 326          * Set the mount level flags.
 327          */
 328         if (uap->flags & MNT_RDONLY)
 329                 mp->mnt_flag |= MNT_RDONLY;
 330         else if (mp->mnt_flag & MNT_RDONLY)
 331                 mp->mnt_kern_flag |= MNTK_WANTRDWR;
 332         mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
 333             MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
 334             MNT_NOSYMFOLLOW | MNT_IGNORE |
 335             MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 336         mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
 337             MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
 338             MNT_NOSYMFOLLOW | MNT_IGNORE |
 339             MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 340         /*
 341          * Mount the filesystem.
 342          * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
 343          * get.
 344          */
 345         error = VFS_MOUNT(mp, uap->path, uap->data, cred);
 346         if (mp->mnt_flag & MNT_UPDATE) {
 347                 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
 348                         mp->mnt_flag &= ~MNT_RDONLY;
 349                 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
 350                 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
 351                 if (error) {
 352                         mp->mnt_flag = flag;
 353                         mp->mnt_kern_flag = flag2;
 354                 }
 355                 vfs_unbusy(mp);
 356                 vp->v_flag &= ~VMOUNT;
 357                 vrele(vp);
 358                 cache_drop(&nch);
 359                 return (error);
 360         }
 361         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 362         /*
 363          * Put the new filesystem on the mount list after root.  The mount
 364          * point gets its own mnt_ncmountpt (unless the VFS already set one
 365          * up) which represents the root of the mount.  The lookup code
 366          * detects the mount point going forward and checks the root of
 367          * the mount going backwards.
 368          *
 369          * It is not necessary to invalidate or purge the vnode underneath
 370          * because elements under the mount will be given their own glue
 371          * namecache record.
 372          */
 373         if (!error) {
 374                 if (mp->mnt_ncmountpt.ncp == NULL) {
 375                         /*
 376                          * allocate, then unlock, but leave the ref intact
 377                          */
 378                         cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
 379                         cache_unlock(&mp->mnt_ncmountpt);
 380                 }
 381                 mp->mnt_ncmounton = nch;                /* inherits ref */
 382                 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
 383
 384                 /* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
 385                 vp->v_flag &= ~VMOUNT;
 386                 mountlist_insert(mp, MNTINS_LAST);
 387                 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
 388                 vn_unlock(vp);
 389                 error = vfs_allocate_syncvnode(mp);
 390                 vfs_unbusy(mp);
 391                 error = VFS_START(mp, 0);
 392                 vrele(vp);
 393         } else {
 394                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 395                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 396                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 397                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 398                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 399                 vp->v_flag &= ~VMOUNT;
 400                 mp->mnt_vfc->vfc_refcount--;
 401                 vfs_unbusy(mp);
 402                 kfree(mp, M_MOUNT);
 403                 cache_drop(&nch);
 404                 vput(vp);
 405         }
 406         return (error);
 407 }
 408
 409 /*
 410  * Scan all active processes to see if any of them have a current
 411  * or root directory onto which the new filesystem has just been
 412  * mounted. If so, replace them with the new mount point.
 413  *
 414  * The passed ncp is ref'd and locked (from the mount code) and
 415  * must be associated with the vnode representing the root of the
 416  * mount point.
 417  */
 418 struct checkdirs_info {
 419         struct nchandle old_nch;
 420         struct nchandle new_nch;
 421         struct vnode *old_vp;
 422         struct vnode *new_vp;
 423 };
 424
 425 static int checkdirs_callback(struct proc *p, void *data);
 426
 427 static void
 428 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
 429 {
 430         struct checkdirs_info info;
 431         struct vnode *olddp;
 432         struct vnode *newdp;
 433         struct mount *mp;
 434
 435         /*
 436          * If the old mount point's vnode has a usecount of 1, it is not
 437          * being held as a descriptor anywhere.
 438          */
 439         olddp = old_nch->ncp->nc_vp;
 440         if (olddp == NULL || olddp->v_sysref.refcnt == 1)
 441                 return;
 442
 443         /*
 444          * Force the root vnode of the new mount point to be resolved
 445          * so we can update any matching processes.
 446          */
 447         mp = new_nch->mount;
 448         if (VFS_ROOT(mp, &newdp))
 449                 panic("mount: lost mount");
 450         cache_setunresolved(new_nch);
 451         cache_setvp(new_nch, newdp);
 452
 453         /*
 454          * Special handling of the root node
 455          */
 456         if (rootvnode == olddp) {
 457                 vref(newdp);
 458                 vfs_cache_setroot(newdp, cache_hold(new_nch));
 459         }
 460
 461         /*
 462          * Pass newdp separately so the callback does not have to access
 463          * it via new_nch->ncp->nc_vp.
 464          */
 465         info.old_nch = *old_nch;
 466         info.new_nch = *new_nch;
 467         info.new_vp = newdp;
 468         allproc_scan(checkdirs_callback, &info);
 469         vput(newdp);
 470 }
 471
 472 /*
 473  * NOTE: callback is not MP safe because the scanned process's filedesc
 474  * structure can be ripped out from under us, amoung other things.
 475  */
 476 static int
 477 checkdirs_callback(struct proc *p, void *data)
 478 {
 479         struct checkdirs_info *info = data;
 480         struct filedesc *fdp;
 481         struct nchandle ncdrop1;
 482         struct nchandle ncdrop2;
 483         struct vnode *vprele1;
 484         struct vnode *vprele2;
 485
 486         if ((fdp = p->p_fd) != NULL) {
 487                 cache_zero(&ncdrop1);
 488                 cache_zero(&ncdrop2);
 489                 vprele1 = NULL;
 490                 vprele2 = NULL;
 491
 492                 /*
 493                  * MPUNSAFE - XXX fdp can be pulled out from under a
 494                  * foreign process.
 495                  *
 496                  * A shared filedesc is ok, we don't have to copy it
 497                  * because we are making this change globally.
 498                  */
 499                 spin_lock_wr(&fdp->fd_spin);
 500                 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
 501                     fdp->fd_ncdir.ncp == info->old_nch.ncp) {
 502                         vprele1 = fdp->fd_cdir;
 503                         vref(info->new_vp);
 504                         fdp->fd_cdir = info->new_vp;
 505                         ncdrop1 = fdp->fd_ncdir;
 506                         cache_copy(&info->new_nch, &fdp->fd_ncdir);
 507                 }
 508                 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
 509                     fdp->fd_nrdir.ncp == info->old_nch.ncp) {
 510                         vprele2 = fdp->fd_rdir;
 511                         vref(info->new_vp);
 512                         fdp->fd_rdir = info->new_vp;
 513                         ncdrop2 = fdp->fd_nrdir;
 514                         cache_copy(&info->new_nch, &fdp->fd_nrdir);
 515                 }
 516                 spin_unlock_wr(&fdp->fd_spin);
 517                 if (ncdrop1.ncp)
 518                         cache_drop(&ncdrop1);
 519                 if (ncdrop2.ncp)
 520                         cache_drop(&ncdrop2);
 521                 if (vprele1)
 522                         vrele(vprele1);
 523                 if (vprele2)
 524                         vrele(vprele2);
 525         }
 526         return(0);
 527 }
 528
 529 /*
 530  * Unmount a file system.
 531  *
 532  * Note: unmount takes a path to the vnode mounted on as argument,
 533  * not special file (as before).
 534  */
 535 /*
 536  * umount_args(char *path, int flags)
 537  */
 538 /* ARGSUSED */
 539 int
 540 sys_unmount(struct unmount_args *uap)
 541 {
 542         struct thread *td = curthread;
 543         struct proc *p = td->td_proc;
 544         struct mount *mp = NULL;
 545         int error;
 546         struct nlookupdata nd;
 547
 548         KKASSERT(p);
 549         if (p->p_ucred->cr_prison != NULL)
 550                 return (EPERM);
 551         if (usermount == 0 && (error = suser(td)))
 552                 return (error);
 553
 554         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 555         if (error == 0)
 556                 error = nlookup(&nd);
 557         if (error)
 558                 goto out;
 559
 560         mp = nd.nl_nch.mount;
 561
 562         /*
 563          * Only root, or the user that did the original mount is
 564          * permitted to unmount this filesystem.
 565          */
 566         if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
 567             (error = suser(td)))
 568                 goto out;
 569
 570         /*
 571          * Don't allow unmounting the root file system.
 572          */
 573         if (mp->mnt_flag & MNT_ROOTFS) {
 574                 error = EINVAL;
 575                 goto out;
 576         }
 577
 578         /*
 579          * Must be the root of the filesystem
 580          */
 581         if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
 582                 error = EINVAL;
 583                 goto out;
 584         }
 585
 586 out:
 587         nlookup_done(&nd);
 588         if (error)
 589                 return (error);
 590         return (dounmount(mp, uap->flags));
 591 }
 592
 593 /*
 594  * Do the actual file system unmount.
 595  */
 596 static int
 597 dounmount_interlock(struct mount *mp)
 598 {
 599         if (mp->mnt_kern_flag & MNTK_UNMOUNT)
 600                 return (EBUSY);
 601         mp->mnt_kern_flag |= MNTK_UNMOUNT;
 602         return(0);
 603 }
 604
 605 int
 606 dounmount(struct mount *mp, int flags)
 607 {
 608         struct namecache *ncp;
 609         struct nchandle nch;
 610         int error;
 611         int async_flag;
 612         int lflags;
 613         int freeok = 1;
 614
 615         /*
 616          * Exclusive access for unmounting purposes
 617          */
 618         if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
 619                 return (error);
 620
 621         /*
 622          * Allow filesystems to detect that a forced unmount is in progress.
 623          */
 624         if (flags & MNT_FORCE)
 625                 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
 626         lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
 627         error = lockmgr(&mp->mnt_lock, lflags);
 628         if (error) {
 629                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 630                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 631                         wakeup(mp);
 632                 return (error);
 633         }
 634
 635         if (mp->mnt_flag & MNT_EXPUBLIC)
 636                 vfs_setpublicfs(NULL, NULL, NULL);
 637
 638         vfs_msync(mp, MNT_WAIT);
 639         async_flag = mp->mnt_flag & MNT_ASYNC;
 640         mp->mnt_flag &=~ MNT_ASYNC;
 641
 642         /*
 643          * If this filesystem isn't aliasing other filesystems,
 644          * try to invalidate any remaining namecache entries and
 645          * check the count afterwords.
 646          */
 647         if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
 648                 cache_lock(&mp->mnt_ncmountpt);
 649                 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
 650                 cache_unlock(&mp->mnt_ncmountpt);
 651
 652                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 653                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 654
 655                         if ((flags & MNT_FORCE) == 0) {
 656                                 error = EBUSY;
 657                                 mount_warning(mp, "Cannot unmount: "
 658                                                   "%d namecache "
 659                                                   "references still "
 660                                                   "present",
 661                                                   ncp->nc_refs - 1);
 662                         } else {
 663                                 mount_warning(mp, "Forced unmount: "
 664                                                   "%d namecache "
 665                                                   "references still "
 666                                                   "present",
 667                                                   ncp->nc_refs - 1);
 668                                 freeok = 0;
 669                         }
 670                 }
 671         }
 672
 673         /*
 674          * nchandle records ref the mount structure.  Expect a count of 1
 675          * (our mount->mnt_ncmountpt).
 676          */
 677         if (mp->mnt_refs != 1) {
 678                 if ((flags & MNT_FORCE) == 0) {
 679                         mount_warning(mp, "Cannot unmount: "
 680                                           "%d process references still "
 681                                           "present", mp->mnt_refs);
 682                         error = EBUSY;
 683                 } else {
 684                         mount_warning(mp, "Forced unmount: "
 685                                           "%d process references still "
 686                                           "present", mp->mnt_refs);
 687                         freeok = 0;
 688                 }
 689         }
 690
 691         if (error == 0) {
 692                 if (mp->mnt_syncer != NULL)
 693                         vrele(mp->mnt_syncer);
 694                 if (((mp->mnt_flag & MNT_RDONLY) ||
 695                      (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
 696                     (flags & MNT_FORCE)) {
 697                         error = VFS_UNMOUNT(mp, flags);
 698                 }
 699         }
 700         if (error) {
 701                 if (mp->mnt_syncer == NULL)
 702                         vfs_allocate_syncvnode(mp);
 703                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 704                 mp->mnt_flag |= async_flag;
 705                 lockmgr(&mp->mnt_lock, LK_RELEASE);
 706                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 707                         wakeup(mp);
 708                 return (error);
 709         }
 710         /*
 711          * Clean up any journals still associated with the mount after
 712          * filesystem activity has ceased.
 713          */
 714         journal_remove_all_journals(mp,
 715             ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
 716
 717         mountlist_remove(mp);
 718
 719         /*
 720          * Remove any installed vnode ops here so the individual VFSs don't
 721          * have to.
 722          */
 723         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 724         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 725         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 726         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 727         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 728
 729         if (mp->mnt_ncmountpt.ncp != NULL) {
 730                 nch = mp->mnt_ncmountpt;
 731                 cache_zero(&mp->mnt_ncmountpt);
 732                 cache_clrmountpt(&nch);
 733                 cache_drop(&nch);
 734         }
 735         if (mp->mnt_ncmounton.ncp != NULL) {
 736                 nch = mp->mnt_ncmounton;
 737                 cache_zero(&mp->mnt_ncmounton);
 738                 cache_clrmountpt(&nch);
 739                 cache_drop(&nch);
 740         }
 741
 742         mp->mnt_vfc->vfc_refcount--;
 743         if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 744                 panic("unmount: dangling vnode");
 745         lockmgr(&mp->mnt_lock, LK_RELEASE);
 746         if (mp->mnt_kern_flag & MNTK_MWAIT)
 747                 wakeup(mp);
 748         if (freeok)
 749                 kfree(mp, M_MOUNT);
 750         return (0);
 751 }
 752
 753 static
 754 void
 755 mount_warning(struct mount *mp, const char *ctl, ...)
 756 {
 757         char *ptr;
 758         char *buf;
 759         __va_list va;
 760
 761         __va_start(va, ctl);
 762         if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
 763                 kprintf("unmount(%s): ", ptr);
 764                 kvprintf(ctl, va);
 765                 kprintf("\n");
 766                 kfree(buf, M_TEMP);
 767         } else {
 768                 kprintf("unmount(%p", mp);
 769                 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
 770                         kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
 771                 kprintf("): ");
 772                 kvprintf(ctl, va);
 773                 kprintf("\n");
 774         }
 775         __va_end(va);
 776 }
 777
 778 /*
 779  * Shim cache_fullpath() to handle the case where a process is chrooted into
 780  * a subdirectory of a mount.  In this case if the root mount matches the
 781  * process root directory's mount we have to specify the process's root
 782  * directory instead of the mount point, because the mount point might
 783  * be above the root directory.
 784  */
 785 static
 786 int
 787 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
 788 {
 789         struct nchandle *nch;
 790
 791         if (p && p->p_fd->fd_nrdir.mount == mp)
 792                 nch = &p->p_fd->fd_nrdir;
 793         else
 794                 nch = &mp->mnt_ncmountpt;
 795         return(cache_fullpath(p, nch, rb, fb));
 796 }
 797
 798 /*
 799  * Sync each mounted filesystem.
 800  */
 801
 802 #ifdef DEBUG
 803 static int syncprt = 0;
 804 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
 805 #endif /* DEBUG */
 806
 807 static int sync_callback(struct mount *mp, void *data);
 808
 809 /* ARGSUSED */
 810 int
 811 sys_sync(struct sync_args *uap)
 812 {
 813         mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
 814 #ifdef DEBUG
 815         /*
 816          * print out buffer pool stat information on each sync() call.
 817          */
 818         if (syncprt)
 819                 vfs_bufstats();
 820 #endif /* DEBUG */
 821         return (0);
 822 }
 823
 824 static
 825 int
 826 sync_callback(struct mount *mp, void *data __unused)
 827 {
 828         int asyncflag;
 829
 830         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 831                 asyncflag = mp->mnt_flag & MNT_ASYNC;
 832                 mp->mnt_flag &= ~MNT_ASYNC;
 833                 vfs_msync(mp, MNT_NOWAIT);
 834                 VFS_SYNC(mp, MNT_NOWAIT);
 835                 mp->mnt_flag |= asyncflag;
 836         }
 837         return(0);
 838 }
 839
 840 /* XXX PRISON: could be per prison flag */
 841 static int prison_quotas;
 842 #if 0
 843 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
 844 #endif
 845
 846 /*
 847  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
 848  *
 849  * Change filesystem quotas.
 850  */
 851 /* ARGSUSED */
 852 int
 853 sys_quotactl(struct quotactl_args *uap)
 854 {
 855         struct nlookupdata nd;
 856         struct thread *td;
 857         struct proc *p;
 858         struct mount *mp;
 859         int error;
 860
 861         td = curthread;
 862         p = td->td_proc;
 863         if (p->p_ucred->cr_prison && !prison_quotas)
 864                 return (EPERM);
 865
 866         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 867         if (error == 0)
 868                 error = nlookup(&nd);
 869         if (error == 0) {
 870                 mp = nd.nl_nch.mount;
 871                 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
 872                                     uap->arg, nd.nl_cred);
 873         }
 874         nlookup_done(&nd);
 875         return (error);
 876 }
 877
 878 /*
 879  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
 880  *              void *buf, int buflen)
 881  *
 882  * This function operates on a mount point and executes the specified
 883  * operation using the specified control data, and possibly returns data.
 884  *
 885  * The actual number of bytes stored in the result buffer is returned, 0
 886  * if none, otherwise an error is returned.
 887  */
 888 /* ARGSUSED */
 889 int
 890 sys_mountctl(struct mountctl_args *uap)
 891 {
 892         struct thread *td = curthread;
 893         struct proc *p = td->td_proc;
 894         struct file *fp;
 895         void *ctl = NULL;
 896         void *buf = NULL;
 897         char *path = NULL;
 898         int error;
 899
 900         /*
 901          * Sanity and permissions checks.  We must be root.
 902          */
 903         KKASSERT(p);
 904         if (p->p_ucred->cr_prison != NULL)
 905                 return (EPERM);
 906         if ((error = suser(td)) != 0)
 907                 return (error);
 908
 909         /*
 910          * Argument length checks
 911          */
 912         if (uap->ctllen < 0 || uap->ctllen > 1024)
 913                 return (EINVAL);
 914         if (uap->buflen < 0 || uap->buflen > 16 * 1024)
 915                 return (EINVAL);
 916         if (uap->path == NULL)
 917                 return (EINVAL);
 918
 919         /*
 920          * Allocate the necessary buffers and copyin data
 921          */
 922         path = objcache_get(namei_oc, M_WAITOK);
 923         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 924         if (error)
 925                 goto done;
 926
 927         if (uap->ctllen) {
 928                 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
 929                 error = copyin(uap->ctl, ctl, uap->ctllen);
 930                 if (error)
 931                         goto done;
 932         }
 933         if (uap->buflen)
 934                 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
 935
 936         /*
 937          * Validate the descriptor
 938          */
 939         if (uap->fd >= 0) {
 940                 fp = holdfp(p->p_fd, uap->fd, -1);
 941                 if (fp == NULL) {
 942                         error = EBADF;
 943                         goto done;
 944                 }
 945         } else {
 946                 fp = NULL;
 947         }
 948
 949         /*
 950          * Execute the internal kernel function and clean up.
 951          */
 952         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
 953         if (fp)
 954                 fdrop(fp);
 955         if (error == 0 && uap->sysmsg_result > 0)
 956                 error = copyout(buf, uap->buf, uap->sysmsg_result);
 957 done:
 958         if (path)
 959                 objcache_put(namei_oc, path);
 960         if (ctl)
 961                 kfree(ctl, M_TEMP);
 962         if (buf)
 963                 kfree(buf, M_TEMP);
 964         return (error);
 965 }
 966
 967 /*
 968  * Execute a mount control operation by resolving the path to a mount point
 969  * and calling vop_mountctl().
 970  */
 971 int
 972 kern_mountctl(const char *path, int op, struct file *fp,
 973                 const void *ctl, int ctllen,
 974                 void *buf, int buflen, int *res)
 975 {
 976         struct vnode *vp;
 977         struct mount *mp;
 978         struct nlookupdata nd;
 979         int error;
 980
 981         *res = 0;
 982         vp = NULL;
 983         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
 984         if (error == 0)
 985                 error = nlookup(&nd);
 986         if (error == 0)
 987                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 988         nlookup_done(&nd);
 989         if (error)
 990                 return (error);
 991
 992         mp = vp->v_mount;
 993
 994         /*
 995          * Must be the root of the filesystem
 996          */
 997         if ((vp->v_flag & VROOT) == 0) {
 998                 vput(vp);
 999                 return (EINVAL);
1000         }
1001         error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1002                                 buf, buflen, res);
1003         vput(vp);
1004         return (error);
1005 }
1006
1007 int
1008 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1009 {
1010         struct thread *td = curthread;
1011         struct proc *p = td->td_proc;
1012         struct mount *mp;
1013         struct statfs *sp;
1014         char *fullpath, *freepath;
1015         int error;
1016
1017         if ((error = nlookup(nd)) != 0)
1018                 return (error);
1019         mp = nd->nl_nch.mount;
1020         sp = &mp->mnt_stat;
1021         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1022                 return (error);
1023
1024         error = mount_path(p, mp, &fullpath, &freepath);
1025         if (error)
1026                 return(error);
1027         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1028         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1029         kfree(freepath, M_TEMP);
1030
1031         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1032         bcopy(sp, buf, sizeof(*buf));
1033         /* Only root should have access to the fsid's. */
1034         if (suser(td))
1035                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1036         return (0);
1037 }
1038
1039 /*
1040  * statfs_args(char *path, struct statfs *buf)
1041  *
1042  * Get filesystem statistics.
1043  */
1044 int
1045 sys_statfs(struct statfs_args *uap)
1046 {
1047         struct nlookupdata nd;
1048         struct statfs buf;
1049         int error;
1050
1051         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1052         if (error == 0)
1053                 error = kern_statfs(&nd, &buf);
1054         nlookup_done(&nd);
1055         if (error == 0)
1056                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1057         return (error);
1058 }
1059
1060 int
1061 kern_fstatfs(int fd, struct statfs *buf)
1062 {
1063         struct thread *td = curthread;
1064         struct proc *p = td->td_proc;
1065         struct file *fp;
1066         struct mount *mp;
1067         struct statfs *sp;
1068         char *fullpath, *freepath;
1069         int error;
1070
1071         KKASSERT(p);
1072         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1073                 return (error);
1074         mp = ((struct vnode *)fp->f_data)->v_mount;
1075         if (mp == NULL) {
1076                 error = EBADF;
1077                 goto done;
1078         }
1079         if (fp->f_cred == NULL) {
1080                 error = EINVAL;
1081                 goto done;
1082         }
1083         sp = &mp->mnt_stat;
1084         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1085                 goto done;
1086
1087         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1088                 goto done;
1089         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1090         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1091         kfree(freepath, M_TEMP);
1092
1093         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1094         bcopy(sp, buf, sizeof(*buf));
1095
1096         /* Only root should have access to the fsid's. */
1097         if (suser(td))
1098                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1099         error = 0;
1100 done:
1101         fdrop(fp);
1102         return (error);
1103 }
1104
1105 /*
1106  * fstatfs_args(int fd, struct statfs *buf)
1107  *
1108  * Get filesystem statistics.
1109  */
1110 int
1111 sys_fstatfs(struct fstatfs_args *uap)
1112 {
1113         struct statfs buf;
1114         int error;
1115
1116         error = kern_fstatfs(uap->fd, &buf);
1117
1118         if (error == 0)
1119                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1120         return (error);
1121 }
1122
1123 /*
1124  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1125  *
1126  * Get statistics on all filesystems.
1127  */
1128
1129 struct getfsstat_info {
1130         struct statfs *sfsp;
1131         long count;
1132         long maxcount;
1133         int error;
1134         int flags;
1135         struct proc *p;
1136 };
1137
1138 static int getfsstat_callback(struct mount *, void *);
1139
1140 /* ARGSUSED */
1141 int
1142 sys_getfsstat(struct getfsstat_args *uap)
1143 {
1144         struct thread *td = curthread;
1145         struct proc *p = td->td_proc;
1146         struct getfsstat_info info;
1147
1148         bzero(&info, sizeof(info));
1149
1150         info.maxcount = uap->bufsize / sizeof(struct statfs);
1151         info.sfsp = uap->buf;
1152         info.count = 0;
1153         info.flags = uap->flags;
1154         info.p = p;
1155
1156         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1157         if (info.sfsp && info.count > info.maxcount)
1158                 uap->sysmsg_result = info.maxcount;
1159         else
1160                 uap->sysmsg_result = info.count;
1161         return (info.error);
1162 }
1163
1164 static int
1165 getfsstat_callback(struct mount *mp, void *data)
1166 {
1167         struct getfsstat_info *info = data;
1168         struct statfs *sp;
1169         char *freepath;
1170         char *fullpath;
1171         int error;
1172
1173         if (info->sfsp && info->count < info->maxcount) {
1174                 if (info->p && !chroot_visible_mnt(mp, info->p))
1175                         return(0);
1176                 sp = &mp->mnt_stat;
1177
1178                 /*
1179                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1180                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1181                  * overrides MNT_WAIT.
1182                  */
1183                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1184                     (info->flags & MNT_WAIT)) &&
1185                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1186                         return(0);
1187                 }
1188                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1189
1190                 error = mount_path(info->p, mp, &fullpath, &freepath);
1191                 if (error) {
1192                         info->error = error;
1193                         return(-1);
1194                 }
1195                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1196                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1197                 kfree(freepath, M_TEMP);
1198
1199                 error = copyout(sp, info->sfsp, sizeof(*sp));
1200                 if (error) {
1201                         info->error = error;
1202                         return (-1);
1203                 }
1204                 ++info->sfsp;
1205         }
1206         info->count++;
1207         return(0);
1208 }
1209
1210 /*
1211  * fchdir_args(int fd)
1212  *
1213  * Change current working directory to a given file descriptor.
1214  */
1215 /* ARGSUSED */
1216 int
1217 sys_fchdir(struct fchdir_args *uap)
1218 {
1219         struct thread *td = curthread;
1220         struct proc *p = td->td_proc;
1221         struct filedesc *fdp = p->p_fd;
1222         struct vnode *vp, *ovp;
1223         struct mount *mp;
1224         struct file *fp;
1225         struct nchandle nch, onch, tnch;
1226         int error;
1227
1228         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1229                 return (error);
1230         vp = (struct vnode *)fp->f_data;
1231         vref(vp);
1232         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1233         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1234                 error = ENOTDIR;
1235         else
1236                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1237         if (error) {
1238                 vput(vp);
1239                 fdrop(fp);
1240                 return (error);
1241         }
1242         cache_copy(&fp->f_nchandle, &nch);
1243
1244         /*
1245          * If the ncp has become a mount point, traverse through
1246          * the mount point.
1247          */
1248
1249         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1250                (mp = cache_findmount(&nch)) != NULL
1251         ) {
1252                 error = nlookup_mp(mp, &tnch);
1253                 if (error == 0) {
1254                         cache_unlock(&tnch);    /* leave ref intact */
1255                         vput(vp);
1256                         vp = tnch.ncp->nc_vp;
1257                         error = vget(vp, LK_SHARED);
1258                         KKASSERT(error == 0);
1259                         cache_drop(&nch);
1260                         nch = tnch;
1261                 }
1262         }
1263         if (error == 0) {
1264                 ovp = fdp->fd_cdir;
1265                 onch = fdp->fd_ncdir;
1266                 vn_unlock(vp);          /* leave ref intact */
1267                 fdp->fd_cdir = vp;
1268                 fdp->fd_ncdir = nch;
1269                 cache_drop(&onch);
1270                 vrele(ovp);
1271         } else {
1272                 cache_drop(&nch);
1273                 vput(vp);
1274         }
1275         fdrop(fp);
1276         return (error);
1277 }
1278
1279 int
1280 kern_chdir(struct nlookupdata *nd)
1281 {
1282         struct thread *td = curthread;
1283         struct proc *p = td->td_proc;
1284         struct filedesc *fdp = p->p_fd;
1285         struct vnode *vp, *ovp;
1286         struct nchandle onch;
1287         int error;
1288
1289         if ((error = nlookup(nd)) != 0)
1290                 return (error);
1291         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1292                 return (ENOENT);
1293         if ((error = vget(vp, LK_SHARED)) != 0)
1294                 return (error);
1295
1296         error = checkvp_chdir(vp, td);
1297         vn_unlock(vp);
1298         if (error == 0) {
1299                 ovp = fdp->fd_cdir;
1300                 onch = fdp->fd_ncdir;
1301                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1302                 fdp->fd_ncdir = nd->nl_nch;
1303                 fdp->fd_cdir = vp;
1304                 cache_drop(&onch);
1305                 vrele(ovp);
1306                 cache_zero(&nd->nl_nch);
1307         } else {
1308                 vrele(vp);
1309         }
1310         return (error);
1311 }
1312
1313 /*
1314  * chdir_args(char *path)
1315  *
1316  * Change current working directory (``.'').
1317  */
1318 int
1319 sys_chdir(struct chdir_args *uap)
1320 {
1321         struct nlookupdata nd;
1322         int error;
1323
1324         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1325         if (error == 0)
1326                 error = kern_chdir(&nd);
1327         nlookup_done(&nd);
1328         return (error);
1329 }
1330
1331 /*
1332  * Helper function for raised chroot(2) security function:  Refuse if
1333  * any filedescriptors are open directories.
1334  */
1335 static int
1336 chroot_refuse_vdir_fds(struct filedesc *fdp)
1337 {
1338         struct vnode *vp;
1339         struct file *fp;
1340         int error;
1341         int fd;
1342
1343         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1344                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1345                         continue;
1346                 vp = (struct vnode *)fp->f_data;
1347                 if (vp->v_type != VDIR) {
1348                         fdrop(fp);
1349                         continue;
1350                 }
1351                 fdrop(fp);
1352                 return(EPERM);
1353         }
1354         return (0);
1355 }
1356
1357 /*
1358  * This sysctl determines if we will allow a process to chroot(2) if it
1359  * has a directory open:
1360  *      0: disallowed for all processes.
1361  *      1: allowed for processes that were not already chroot(2)'ed.
1362  *      2: allowed for all processes.
1363  */
1364
1365 static int chroot_allow_open_directories = 1;
1366
1367 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1368      &chroot_allow_open_directories, 0, "");
1369
1370 /*
1371  * chroot to the specified namecache entry.  We obtain the vp from the
1372  * namecache data.  The passed ncp must be locked and referenced and will
1373  * remain locked and referenced on return.
1374  */
1375 int
1376 kern_chroot(struct nchandle *nch)
1377 {
1378         struct thread *td = curthread;
1379         struct proc *p = td->td_proc;
1380         struct filedesc *fdp = p->p_fd;
1381         struct vnode *vp;
1382         int error;
1383
1384         /*
1385          * Only root can chroot
1386          */
1387         if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1388                 return (error);
1389
1390         /*
1391          * Disallow open directory descriptors (fchdir() breakouts).
1392          */
1393         if (chroot_allow_open_directories == 0 ||
1394            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1395                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1396                         return (error);
1397         }
1398         if ((vp = nch->ncp->nc_vp) == NULL)
1399                 return (ENOENT);
1400
1401         if ((error = vget(vp, LK_SHARED)) != 0)
1402                 return (error);
1403
1404         /*
1405          * Check the validity of vp as a directory to change to and
1406          * associate it with rdir/jdir.
1407          */
1408         error = checkvp_chdir(vp, td);
1409         vn_unlock(vp);                  /* leave reference intact */
1410         if (error == 0) {
1411                 vrele(fdp->fd_rdir);
1412                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1413                 cache_drop(&fdp->fd_nrdir);
1414                 cache_copy(nch, &fdp->fd_nrdir);
1415                 if (fdp->fd_jdir == NULL) {
1416                         fdp->fd_jdir = vp;
1417                         vref(fdp->fd_jdir);
1418                         cache_copy(nch, &fdp->fd_njdir);
1419                 }
1420         } else {
1421                 vrele(vp);
1422         }
1423         return (error);
1424 }
1425
1426 /*
1427  * chroot_args(char *path)
1428  *
1429  * Change notion of root (``/'') directory.
1430  */
1431 /* ARGSUSED */
1432 int
1433 sys_chroot(struct chroot_args *uap)
1434 {
1435         struct thread *td = curthread;
1436         struct nlookupdata nd;
1437         int error;
1438
1439         KKASSERT(td->td_proc);
1440         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1441         if (error) {
1442                 nlookup_done(&nd);
1443                 return(error);
1444         }
1445         error = nlookup(&nd);
1446         if (error == 0)
1447                 error = kern_chroot(&nd.nl_nch);
1448         nlookup_done(&nd);
1449         return(error);
1450 }
1451
1452 /*
1453  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1454  * determine whether it is legal to chdir to the vnode.  The vnode's state
1455  * is not changed by this call.
1456  */
1457 int
1458 checkvp_chdir(struct vnode *vp, struct thread *td)
1459 {
1460         int error;
1461
1462         if (vp->v_type != VDIR)
1463                 error = ENOTDIR;
1464         else
1465                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1466         return (error);
1467 }
1468
1469 int
1470 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1471 {
1472         struct thread *td = curthread;
1473         struct proc *p = td->td_proc;
1474         struct lwp *lp = td->td_lwp;
1475         struct filedesc *fdp = p->p_fd;
1476         int cmode, flags;
1477         struct file *nfp;
1478         struct file *fp;
1479         struct vnode *vp;
1480         int type, indx, error;
1481         struct flock lf;
1482
1483         if ((oflags & O_ACCMODE) == O_ACCMODE)
1484                 return (EINVAL);
1485         flags = FFLAGS(oflags);
1486         error = falloc(p, &nfp, NULL);
1487         if (error)
1488                 return (error);
1489         fp = nfp;
1490         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1491
1492         /*
1493          * XXX p_dupfd is a real mess.  It allows a device to return a
1494          * file descriptor to be duplicated rather then doing the open
1495          * itself.
1496          */
1497         lp->lwp_dupfd = -1;
1498
1499         /*
1500          * Call vn_open() to do the lookup and assign the vnode to the
1501          * file pointer.  vn_open() does not change the ref count on fp
1502          * and the vnode, on success, will be inherited by the file pointer
1503          * and unlocked.
1504          */
1505         nd->nl_flags |= NLC_LOCKVP;
1506         error = vn_open(nd, fp, flags, cmode);
1507         nlookup_done(nd);
1508         if (error) {
1509                 /*
1510                  * handle special fdopen() case.  bleh.  dupfdopen() is
1511                  * responsible for dropping the old contents of ofiles[indx]
1512                  * if it succeeds.
1513                  *
1514                  * Note that fsetfd() will add a ref to fp which represents
1515                  * the fd_files[] assignment.  We must still drop our
1516                  * reference.
1517                  */
1518                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1519                         if (fdalloc(p, 0, &indx) == 0) {
1520                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1521                                 if (error == 0) {
1522                                         *res = indx;
1523                                         fdrop(fp);      /* our ref */
1524                                         return (0);
1525                                 }
1526                                 fsetfd(p, NULL, indx);
1527                         }
1528                 }
1529                 fdrop(fp);      /* our ref */
1530                 if (error == ERESTART)
1531                         error = EINTR;
1532                 return (error);
1533         }
1534
1535         /*
1536          * ref the vnode for ourselves so it can't be ripped out from under
1537          * is.  XXX need an ND flag to request that the vnode be returned
1538          * anyway.
1539          *
1540          * Reserve a file descriptor but do not assign it until the open
1541          * succeeds.
1542          */
1543         vp = (struct vnode *)fp->f_data;
1544         vref(vp);
1545         if ((error = fdalloc(p, 0, &indx)) != 0) {
1546                 fdrop(fp);
1547                 vrele(vp);
1548                 return (error);
1549         }
1550
1551         /*
1552          * If no error occurs the vp will have been assigned to the file
1553          * pointer.
1554          */
1555         lp->lwp_dupfd = 0;
1556
1557         if (flags & (O_EXLOCK | O_SHLOCK)) {
1558                 lf.l_whence = SEEK_SET;
1559                 lf.l_start = 0;
1560                 lf.l_len = 0;
1561                 if (flags & O_EXLOCK)
1562                         lf.l_type = F_WRLCK;
1563                 else
1564                         lf.l_type = F_RDLCK;
1565                 if (flags & FNONBLOCK)
1566                         type = 0;
1567                 else
1568                         type = F_WAIT;
1569
1570                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1571                         /*
1572                          * lock request failed.  Clean up the reserved
1573                          * descriptor.
1574                          */
1575                         vrele(vp);
1576                         fsetfd(p, NULL, indx);
1577                         fdrop(fp);
1578                         return (error);
1579                 }
1580                 fp->f_flag |= FHASLOCK;
1581         }
1582 #if 0
1583         /*
1584          * Assert that all regular file vnodes were created with a object.
1585          */
1586         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1587                 ("open: regular file has no backing object after vn_open"));
1588 #endif
1589
1590         vrele(vp);
1591
1592         /*
1593          * release our private reference, leaving the one associated with the
1594          * descriptor table intact.
1595          */
1596         fsetfd(p, fp, indx);
1597         fdrop(fp);
1598         *res = indx;
1599         return (0);
1600 }
1601
1602 /*
1603  * open_args(char *path, int flags, int mode)
1604  *
1605  * Check permissions, allocate an open file structure,
1606  * and call the device open routine if any.
1607  */
1608 int
1609 sys_open(struct open_args *uap)
1610 {
1611         struct nlookupdata nd;
1612         int error;
1613
1614         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1615         if (error == 0) {
1616                 error = kern_open(&nd, uap->flags,
1617                                     uap->mode, &uap->sysmsg_result);
1618         }
1619         nlookup_done(&nd);
1620         return (error);
1621 }
1622
1623 int
1624 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1625 {
1626         struct thread *td = curthread;
1627         struct proc *p = td->td_proc;
1628         struct vnode *vp;
1629         struct vnode *dvp;
1630         struct vattr vattr;
1631         int error;
1632         int whiteout = 0;
1633
1634         KKASSERT(p);
1635
1636         switch (mode & S_IFMT) {
1637         case S_IFCHR:
1638         case S_IFBLK:
1639                 error = suser(td);
1640                 break;
1641         default:
1642                 error = suser_cred(p->p_ucred, PRISON_ROOT);
1643                 break;
1644         }
1645         if (error)
1646                 return (error);
1647
1648         bwillwrite();
1649         nd->nl_flags |= NLC_CREATE;
1650         if ((error = nlookup(nd)) != 0)
1651                 return (error);
1652         if (nd->nl_nch.ncp->nc_vp)
1653                 return (EEXIST);
1654         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1655                 return (error);
1656         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
1657                 return (EPERM);
1658         /* vhold(dvp); - DVP can't go away */
1659
1660         VATTR_NULL(&vattr);
1661         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1662         vattr.va_rmajor = rmajor;
1663         vattr.va_rminor = rminor;
1664         whiteout = 0;
1665
1666         switch (mode & S_IFMT) {
1667         case S_IFMT:    /* used by badsect to flag bad sectors */
1668                 vattr.va_type = VBAD;
1669                 break;
1670         case S_IFCHR:
1671                 vattr.va_type = VCHR;
1672                 break;
1673         case S_IFBLK:
1674                 vattr.va_type = VBLK;
1675                 break;
1676         case S_IFWHT:
1677                 whiteout = 1;
1678                 break;
1679         default:
1680                 error = EINVAL;
1681                 break;
1682         }
1683         if (error == 0) {
1684                 if (whiteout) {
1685                         error = VOP_NWHITEOUT(&nd->nl_nch, dvp, nd->nl_cred, NAMEI_CREATE);
1686                 } else {
1687                         vp = NULL;
1688                         error = VOP_NMKNOD(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr);
1689                         if (error == 0)
1690                                 vput(vp);
1691                 }
1692         }
1693         /* vdrop(dvp); */
1694         return (error);
1695 }
1696
1697 /*
1698  * mknod_args(char *path, int mode, int dev)
1699  *
1700  * Create a special file.
1701  */
1702 int
1703 sys_mknod(struct mknod_args *uap)
1704 {
1705         struct nlookupdata nd;
1706         int error;
1707
1708         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1709         if (error == 0) {
1710                 error = kern_mknod(&nd, uap->mode,
1711                                    umajor(uap->dev), uminor(uap->dev));
1712         }
1713         nlookup_done(&nd);
1714         return (error);
1715 }
1716
1717 int
1718 kern_mkfifo(struct nlookupdata *nd, int mode)
1719 {
1720         struct thread *td = curthread;
1721         struct proc *p = td->td_proc;
1722         struct vattr vattr;
1723         struct vnode *vp;
1724         struct vnode *dvp;
1725         int error;
1726
1727         bwillwrite();
1728
1729         nd->nl_flags |= NLC_CREATE;
1730         if ((error = nlookup(nd)) != 0)
1731                 return (error);
1732         if (nd->nl_nch.ncp->nc_vp)
1733                 return (EEXIST);
1734         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1735                 return (error);
1736         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
1737                 return (EPERM);
1738         /* vhold(dvp); - DVP can't go away */
1739
1740         VATTR_NULL(&vattr);
1741         vattr.va_type = VFIFO;
1742         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1743         vp = NULL;
1744         error = VOP_NMKNOD(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr);
1745         /* vdrop(dvp); */
1746         if (error == 0)
1747                 vput(vp);
1748         return (error);
1749 }
1750
1751 /*
1752  * mkfifo_args(char *path, int mode)
1753  *
1754  * Create a named pipe.
1755  */
1756 int
1757 sys_mkfifo(struct mkfifo_args *uap)
1758 {
1759         struct nlookupdata nd;
1760         int error;
1761
1762         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1763         if (error == 0)
1764                 error = kern_mkfifo(&nd, uap->mode);
1765         nlookup_done(&nd);
1766         return (error);
1767 }
1768
1769 static int hardlink_check_uid = 0;
1770 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1771     &hardlink_check_uid, 0,
1772     "Unprivileged processes cannot create hard links to files owned by other "
1773     "users");
1774 static int hardlink_check_gid = 0;
1775 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1776     &hardlink_check_gid, 0,
1777     "Unprivileged processes cannot create hard links to files owned by other "
1778     "groups");
1779
1780 static int
1781 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1782 {
1783         struct vattr va;
1784         int error;
1785
1786         /*
1787          * Shortcut if disabled
1788          */
1789         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1790                 return (0);
1791
1792         /*
1793          * root cred can always hardlink
1794          */
1795         if (suser_cred(cred, PRISON_ROOT) == 0)
1796                 return (0);
1797
1798         /*
1799          * Otherwise only if the originating file is owned by the
1800          * same user or group.  Note that any group is allowed if
1801          * the file is owned by the caller.
1802          */
1803         error = VOP_GETATTR(vp, &va);
1804         if (error != 0)
1805                 return (error);
1806
1807         if (hardlink_check_uid) {
1808                 if (cred->cr_uid != va.va_uid)
1809                         return (EPERM);
1810         }
1811
1812         if (hardlink_check_gid) {
1813                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1814                         return (EPERM);
1815         }
1816
1817         return (0);
1818 }
1819
1820 int
1821 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1822 {
1823         struct thread *td = curthread;
1824         struct vnode *vp;
1825         struct vnode *dvp;
1826         int error;
1827
1828         /*
1829          * Lookup the source and obtained a locked vnode.
1830          *
1831          * XXX relookup on vget failure / race ?
1832          */
1833         bwillwrite();
1834         if ((error = nlookup(nd)) != 0)
1835                 return (error);
1836         vp = nd->nl_nch.ncp->nc_vp;
1837         KKASSERT(vp != NULL);
1838         if (vp->v_type == VDIR)
1839                 return (EPERM);         /* POSIX */
1840         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1841                 return (error);
1842         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1843                 return (error);
1844
1845         /*
1846          * Unlock the source so we can lookup the target without deadlocking
1847          * (XXX vp is locked already, possible other deadlock?).  The target
1848          * must not exist.
1849          */
1850         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1851         nd->nl_flags &= ~NLC_NCPISLOCKED;
1852         cache_unlock(&nd->nl_nch);
1853
1854         linknd->nl_flags |= NLC_CREATE;
1855         if ((error = nlookup(linknd)) != 0) {
1856                 vput(vp);
1857                 return (error);
1858         }
1859         if (linknd->nl_nch.ncp->nc_vp) {
1860                 vput(vp);
1861                 return (EEXIST);
1862         }
1863         if ((dvp = linknd->nl_nch.ncp->nc_parent->nc_vp) == NULL) {
1864                 vput(vp);
1865                 return (EPERM);
1866         }
1867         /* vhold(dvp); - dvp can't go away */
1868
1869         /*
1870          * Finally run the new API VOP.
1871          */
1872         error = can_hardlink(vp, td, td->td_proc->p_ucred);
1873         if (error == 0)
1874                 error = VOP_NLINK(&linknd->nl_nch, dvp, vp, linknd->nl_cred);
1875         /* vdrop(dvp); */
1876         vput(vp);
1877         return (error);
1878 }
1879
1880 /*
1881  * link_args(char *path, char *link)
1882  *
1883  * Make a hard file link.
1884  */
1885 int
1886 sys_link(struct link_args *uap)
1887 {
1888         struct nlookupdata nd, linknd;
1889         int error;
1890
1891         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1892         if (error == 0) {
1893                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1894                 if (error == 0)
1895                         error = kern_link(&nd, &linknd);
1896                 nlookup_done(&linknd);
1897         }
1898         nlookup_done(&nd);
1899         return (error);
1900 }
1901
1902 int
1903 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1904 {
1905         struct vattr vattr;
1906         struct vnode *vp;
1907         struct vnode *dvp;
1908         int error;
1909
1910         bwillwrite();
1911         nd->nl_flags |= NLC_CREATE;
1912         if ((error = nlookup(nd)) != 0)
1913                 return (error);
1914         if (nd->nl_nch.ncp->nc_vp)
1915                 return (EEXIST);
1916         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1917                 return (error);
1918         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
1919                 return (EPERM);
1920         /* vhold(dvp); - dvp can't go away */
1921         VATTR_NULL(&vattr);
1922         vattr.va_mode = mode;
1923         error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
1924         /* vdrop(dvp); */
1925         if (error == 0)
1926                 vput(vp);
1927         return (error);
1928 }
1929
1930 /*
1931  * symlink(char *path, char *link)
1932  *
1933  * Make a symbolic link.
1934  */
1935 int
1936 sys_symlink(struct symlink_args *uap)
1937 {
1938         struct thread *td = curthread;
1939         struct nlookupdata nd;
1940         char *path;
1941         int error;
1942         int mode;
1943
1944         path = objcache_get(namei_oc, M_WAITOK);
1945         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1946         if (error == 0) {
1947                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1948                 if (error == 0) {
1949                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1950                         error = kern_symlink(&nd, path, mode);
1951                 }
1952                 nlookup_done(&nd);
1953         }
1954         objcache_put(namei_oc, path);
1955         return (error);
1956 }
1957
1958 /*
1959  * undelete_args(char *path)
1960  *
1961  * Delete a whiteout from the filesystem.
1962  */
1963 /* ARGSUSED */
1964 int
1965 sys_undelete(struct undelete_args *uap)
1966 {
1967         struct nlookupdata nd;
1968         struct vnode *dvp;
1969         int error;
1970
1971         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1972         bwillwrite();
1973         nd.nl_flags |= NLC_DELETE;
1974         if (error == 0)
1975                 error = nlookup(&nd);
1976         if (error == 0)
1977                 error = ncp_writechk(&nd.nl_nch);
1978         dvp = NULL;
1979         if (error == 0) {
1980                 if ((dvp = nd.nl_nch.ncp->nc_parent->nc_vp) == NULL)
1981                         error = EPERM;
1982         }
1983         if (error == 0) {
1984                 /* vhold(dvp); - dvp can't go away */
1985                 error = VOP_NWHITEOUT(&nd.nl_nch, dvp, nd.nl_cred, NAMEI_DELETE);
1986                 /* vdrop(dvp); */
1987         }
1988         nlookup_done(&nd);
1989         return (error);
1990 }
1991
1992 int
1993 kern_unlink(struct nlookupdata *nd)
1994 {
1995         struct vnode *dvp;
1996         int error;
1997
1998         bwillwrite();
1999         nd->nl_flags |= NLC_DELETE;
2000         if ((error = nlookup(nd)) != 0)
2001                 return (error);
2002         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2003                 return (error);
2004         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
2005                 return (EPERM);
2006         /* vhold(dvp); - dvp can't go away */
2007         error = VOP_NREMOVE(&nd->nl_nch, dvp, nd->nl_cred);
2008         /* vdrop(dvp); */
2009         return (error);
2010 }
2011
2012 /*
2013  * unlink_args(char *path)
2014  *
2015  * Delete a name from the filesystem.
2016  */
2017 int
2018 sys_unlink(struct unlink_args *uap)
2019 {
2020         struct nlookupdata nd;
2021         int error;
2022
2023         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2024         if (error == 0)
2025                 error = kern_unlink(&nd);
2026         nlookup_done(&nd);
2027         return (error);
2028 }
2029
2030 int
2031 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2032 {
2033         struct thread *td = curthread;
2034         struct proc *p = td->td_proc;
2035         struct file *fp;
2036         struct vattr vattr;
2037         int error;
2038
2039         fp = holdfp(p->p_fd, fd, -1);
2040         if (fp == NULL)
2041                 return (EBADF);
2042         if (fp->f_type != DTYPE_VNODE) {
2043                 error = ESPIPE;
2044                 goto done;
2045         }
2046
2047         switch (whence) {
2048         case L_INCR:
2049                 fp->f_offset += offset;
2050                 error = 0;
2051                 break;
2052         case L_XTND:
2053                 error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
2054                 if (error == 0)
2055                         fp->f_offset = offset + vattr.va_size;
2056                 break;
2057         case L_SET:
2058                 fp->f_offset = offset;
2059                 error = 0;
2060                 break;
2061         default:
2062                 error = EINVAL;
2063                 break;
2064         }
2065         *res = fp->f_offset;
2066 done:
2067         fdrop(fp);
2068         return (error);
2069 }
2070
2071 /*
2072  * lseek_args(int fd, int pad, off_t offset, int whence)
2073  *
2074  * Reposition read/write file offset.
2075  */
2076 int
2077 sys_lseek(struct lseek_args *uap)
2078 {
2079         int error;
2080
2081         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2082             &uap->sysmsg_offset);
2083
2084         return (error);
2085 }
2086
2087 int
2088 kern_access(struct nlookupdata *nd, int aflags)
2089 {
2090         struct vnode *vp;
2091         int error, flags;
2092
2093         if ((error = nlookup(nd)) != 0)
2094                 return (error);
2095 retry:
2096         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2097         if (error)
2098                 return (error);
2099
2100         /* Flags == 0 means only check for existence. */
2101         if (aflags) {
2102                 flags = 0;
2103                 if (aflags & R_OK)
2104                         flags |= VREAD;
2105                 if (aflags & W_OK)
2106                         flags |= VWRITE;
2107                 if (aflags & X_OK)
2108                         flags |= VEXEC;
2109                 if ((flags & VWRITE) == 0 ||
2110                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2111                         error = VOP_ACCESS(vp, flags, nd->nl_cred);
2112
2113                 /*
2114                  * If the file handle is stale we have to re-resolve the
2115                  * entry.  This is a hack at the moment.
2116                  */
2117                 if (error == ESTALE) {
2118                         vput(vp);
2119                         cache_setunresolved(&nd->nl_nch);
2120                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2121                         if (error == 0) {
2122                                 vp = NULL;
2123                                 goto retry;
2124                         }
2125                         return(error);
2126                 }
2127         }
2128         vput(vp);
2129         return (error);
2130 }
2131
2132 /*
2133  * access_args(char *path, int flags)
2134  *
2135  * Check access permissions.
2136  */
2137 int
2138 sys_access(struct access_args *uap)
2139 {
2140         struct nlookupdata nd;
2141         int error;
2142
2143         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2144         if (error == 0)
2145                 error = kern_access(&nd, uap->flags);
2146         nlookup_done(&nd);
2147         return (error);
2148 }
2149
2150 int
2151 kern_stat(struct nlookupdata *nd, struct stat *st)
2152 {
2153         int error;
2154         struct vnode *vp;
2155         thread_t td;
2156
2157         if ((error = nlookup(nd)) != 0)
2158                 return (error);
2159 again:
2160         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2161                 return (ENOENT);
2162
2163         td = curthread;
2164         if ((error = vget(vp, LK_SHARED)) != 0)
2165                 return (error);
2166         error = vn_stat(vp, st, nd->nl_cred);
2167
2168         /*
2169          * If the file handle is stale we have to re-resolve the entry.  This
2170          * is a hack at the moment.
2171          */
2172         if (error == ESTALE) {
2173                 vput(vp);
2174                 cache_setunresolved(&nd->nl_nch);
2175                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2176                 if (error == 0)
2177                         goto again;
2178         } else {
2179                 vput(vp);
2180         }
2181         return (error);
2182 }
2183
2184 /*
2185  * stat_args(char *path, struct stat *ub)
2186  *
2187  * Get file status; this version follows links.
2188  */
2189 int
2190 sys_stat(struct stat_args *uap)
2191 {
2192         struct nlookupdata nd;
2193         struct stat st;
2194         int error;
2195
2196         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2197         if (error == 0) {
2198                 error = kern_stat(&nd, &st);
2199                 if (error == 0)
2200                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2201         }
2202         nlookup_done(&nd);
2203         return (error);
2204 }
2205
2206 /*
2207  * lstat_args(char *path, struct stat *ub)
2208  *
2209  * Get file status; this version does not follow links.
2210  */
2211 int
2212 sys_lstat(struct lstat_args *uap)
2213 {
2214         struct nlookupdata nd;
2215         struct stat st;
2216         int error;
2217
2218         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2219         if (error == 0) {
2220                 error = kern_stat(&nd, &st);
2221                 if (error == 0)
2222                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2223         }
2224         nlookup_done(&nd);
2225         return (error);
2226 }
2227
2228 /*
2229  * pathconf_Args(char *path, int name)
2230  *
2231  * Get configurable pathname variables.
2232  */
2233 /* ARGSUSED */
2234 int
2235 sys_pathconf(struct pathconf_args *uap)
2236 {
2237         struct nlookupdata nd;
2238         struct vnode *vp;
2239         int error;
2240
2241         vp = NULL;
2242         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2243         if (error == 0)
2244                 error = nlookup(&nd);
2245         if (error == 0)
2246                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2247         nlookup_done(&nd);
2248         if (error == 0) {
2249                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2250                 vput(vp);
2251         }
2252         return (error);
2253 }
2254
2255 /*
2256  * XXX: daver
2257  * kern_readlink isn't properly split yet.  There is a copyin burried
2258  * in VOP_READLINK().
2259  */
2260 int
2261 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2262 {
2263         struct thread *td = curthread;
2264         struct proc *p = td->td_proc;
2265         struct vnode *vp;
2266         struct iovec aiov;
2267         struct uio auio;
2268         int error;
2269
2270         if ((error = nlookup(nd)) != 0)
2271                 return (error);
2272         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2273         if (error)
2274                 return (error);
2275         if (vp->v_type != VLNK) {
2276                 error = EINVAL;
2277         } else {
2278                 aiov.iov_base = buf;
2279                 aiov.iov_len = count;
2280                 auio.uio_iov = &aiov;
2281                 auio.uio_iovcnt = 1;
2282                 auio.uio_offset = 0;
2283                 auio.uio_rw = UIO_READ;
2284                 auio.uio_segflg = UIO_USERSPACE;
2285                 auio.uio_td = td;
2286                 auio.uio_resid = count;
2287                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2288         }
2289         vput(vp);
2290         *res = count - auio.uio_resid;
2291         return (error);
2292 }
2293
2294 /*
2295  * readlink_args(char *path, char *buf, int count)
2296  *
2297  * Return target name of a symbolic link.
2298  */
2299 int
2300 sys_readlink(struct readlink_args *uap)
2301 {
2302         struct nlookupdata nd;
2303         int error;
2304
2305         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2306         if (error == 0) {
2307                 error = kern_readlink(&nd, uap->buf, uap->count,
2308                                         &uap->sysmsg_result);
2309         }
2310         nlookup_done(&nd);
2311         return (error);
2312 }
2313
2314 static int
2315 setfflags(struct vnode *vp, int flags)
2316 {
2317         struct thread *td = curthread;
2318         struct proc *p = td->td_proc;
2319         int error;
2320         struct vattr vattr;
2321
2322         /*
2323          * Prevent non-root users from setting flags on devices.  When
2324          * a device is reused, users can retain ownership of the device
2325          * if they are allowed to set flags and programs assume that
2326          * chown can't fail when done as root.
2327          */
2328         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2329             ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2330                 return (error);
2331
2332         /*
2333          * note: vget is required for any operation that might mod the vnode
2334          * so VINACTIVE is properly cleared.
2335          */
2336         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2337                 VATTR_NULL(&vattr);
2338                 vattr.va_flags = flags;
2339                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2340                 vput(vp);
2341         }
2342         return (error);
2343 }
2344
2345 /*
2346  * chflags(char *path, int flags)
2347  *
2348  * Change flags of a file given a path name.
2349  */
2350 /* ARGSUSED */
2351 int
2352 sys_chflags(struct chflags_args *uap)
2353 {
2354         struct nlookupdata nd;
2355         struct vnode *vp;
2356         int error;
2357
2358         vp = NULL;
2359         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2360         /* XXX Add NLC flag indicating modifying operation? */
2361         if (error == 0)
2362                 error = nlookup(&nd);
2363         if (error == 0)
2364                 error = ncp_writechk(&nd.nl_nch);
2365         if (error == 0)
2366                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2367         nlookup_done(&nd);
2368         if (error == 0) {
2369                 error = setfflags(vp, uap->flags);
2370                 vrele(vp);
2371         }
2372         return (error);
2373 }
2374
2375 /*
2376  * fchflags_args(int fd, int flags)
2377  *
2378  * Change flags of a file given a file descriptor.
2379  */
2380 /* ARGSUSED */
2381 int
2382 sys_fchflags(struct fchflags_args *uap)
2383 {
2384         struct thread *td = curthread;
2385         struct proc *p = td->td_proc;
2386         struct file *fp;
2387         int error;
2388
2389         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2390                 return (error);
2391         if (fp->f_nchandle.ncp)
2392                 error = ncp_writechk(&fp->f_nchandle);
2393         if (error == 0)
2394                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2395         fdrop(fp);
2396         return (error);
2397 }
2398
2399 static int
2400 setfmode(struct vnode *vp, int mode)
2401 {
2402         struct thread *td = curthread;
2403         struct proc *p = td->td_proc;
2404         int error;
2405         struct vattr vattr;
2406
2407         /*
2408          * note: vget is required for any operation that might mod the vnode
2409          * so VINACTIVE is properly cleared.
2410          */
2411         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2412                 VATTR_NULL(&vattr);
2413                 vattr.va_mode = mode & ALLPERMS;
2414                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2415                 vput(vp);
2416         }
2417         return error;
2418 }
2419
2420 int
2421 kern_chmod(struct nlookupdata *nd, int mode)
2422 {
2423         struct vnode *vp;
2424         int error;
2425
2426         /* XXX Add NLC flag indicating modifying operation? */
2427         if ((error = nlookup(nd)) != 0)
2428                 return (error);
2429         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2430                 return (error);
2431         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2432                 error = setfmode(vp, mode);
2433         vrele(vp);
2434         return (error);
2435 }
2436
2437 /*
2438  * chmod_args(char *path, int mode)
2439  *
2440  * Change mode of a file given path name.
2441  */
2442 /* ARGSUSED */
2443 int
2444 sys_chmod(struct chmod_args *uap)
2445 {
2446         struct nlookupdata nd;
2447         int error;
2448
2449         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2450         if (error == 0)
2451                 error = kern_chmod(&nd, uap->mode);
2452         nlookup_done(&nd);
2453         return (error);
2454 }
2455
2456 /*
2457  * lchmod_args(char *path, int mode)
2458  *
2459  * Change mode of a file given path name (don't follow links.)
2460  */
2461 /* ARGSUSED */
2462 int
2463 sys_lchmod(struct lchmod_args *uap)
2464 {
2465         struct nlookupdata nd;
2466         int error;
2467
2468         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2469         if (error == 0)
2470                 error = kern_chmod(&nd, uap->mode);
2471         nlookup_done(&nd);
2472         return (error);
2473 }
2474
2475 /*
2476  * fchmod_args(int fd, int mode)
2477  *
2478  * Change mode of a file given a file descriptor.
2479  */
2480 /* ARGSUSED */
2481 int
2482 sys_fchmod(struct fchmod_args *uap)
2483 {
2484         struct thread *td = curthread;
2485         struct proc *p = td->td_proc;
2486         struct file *fp;
2487         int error;
2488
2489         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2490                 return (error);
2491         if (fp->f_nchandle.ncp)
2492                 error = ncp_writechk(&fp->f_nchandle);
2493         if (error == 0)
2494                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2495         fdrop(fp);
2496         return (error);
2497 }
2498
2499 static int
2500 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2501 {
2502         struct thread *td = curthread;
2503         struct proc *p = td->td_proc;
2504         int error;
2505         struct vattr vattr;
2506
2507         /*
2508          * note: vget is required for any operation that might mod the vnode
2509          * so VINACTIVE is properly cleared.
2510          */
2511         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2512                 VATTR_NULL(&vattr);
2513                 vattr.va_uid = uid;
2514                 vattr.va_gid = gid;
2515                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2516                 vput(vp);
2517         }
2518         return error;
2519 }
2520
2521 int
2522 kern_chown(struct nlookupdata *nd, int uid, int gid)
2523 {
2524         struct vnode *vp;
2525         int error;
2526
2527         /* XXX Add NLC flag indicating modifying operation? */
2528         if ((error = nlookup(nd)) != 0)
2529                 return (error);
2530         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2531                 return (error);
2532         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2533                 error = setfown(vp, uid, gid);
2534         vrele(vp);
2535         return (error);
2536 }
2537
2538 /*
2539  * chown(char *path, int uid, int gid)
2540  *
2541  * Set ownership given a path name.
2542  */
2543 int
2544 sys_chown(struct chown_args *uap)
2545 {
2546         struct nlookupdata nd;
2547         int error;
2548
2549         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2550         if (error == 0)
2551                 error = kern_chown(&nd, uap->uid, uap->gid);
2552         nlookup_done(&nd);
2553         return (error);
2554 }
2555
2556 /*
2557  * lchown_args(char *path, int uid, int gid)
2558  *
2559  * Set ownership given a path name, do not cross symlinks.
2560  */
2561 int
2562 sys_lchown(struct lchown_args *uap)
2563 {
2564         struct nlookupdata nd;
2565         int error;
2566
2567         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2568         if (error == 0)
2569                 error = kern_chown(&nd, uap->uid, uap->gid);
2570         nlookup_done(&nd);
2571         return (error);
2572 }
2573
2574 /*
2575  * fchown_args(int fd, int uid, int gid)
2576  *
2577  * Set ownership given a file descriptor.
2578  */
2579 /* ARGSUSED */
2580 int
2581 sys_fchown(struct fchown_args *uap)
2582 {
2583         struct thread *td = curthread;
2584         struct proc *p = td->td_proc;
2585         struct file *fp;
2586         int error;
2587
2588         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2589                 return (error);
2590         if (fp->f_nchandle.ncp)
2591                 error = ncp_writechk(&fp->f_nchandle);
2592         if (error == 0)
2593                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2594         fdrop(fp);
2595         return (error);
2596 }
2597
2598 static int
2599 getutimes(const struct timeval *tvp, struct timespec *tsp)
2600 {
2601         struct timeval tv[2];
2602
2603         if (tvp == NULL) {
2604                 microtime(&tv[0]);
2605                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2606                 tsp[1] = tsp[0];
2607         } else {
2608                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2609                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2610         }
2611         return 0;
2612 }
2613
2614 static int
2615 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2616 {
2617         struct thread *td = curthread;
2618         struct proc *p = td->td_proc;
2619         int error;
2620         struct vattr vattr;
2621
2622         /*
2623          * note: vget is required for any operation that might mod the vnode
2624          * so VINACTIVE is properly cleared.
2625          */
2626         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2627                 VATTR_NULL(&vattr);
2628                 vattr.va_atime = ts[0];
2629                 vattr.va_mtime = ts[1];
2630                 if (nullflag)
2631                         vattr.va_vaflags |= VA_UTIMES_NULL;
2632                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2633                 vput(vp);
2634         }
2635         return error;
2636 }
2637
2638 int
2639 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2640 {
2641         struct timespec ts[2];
2642         struct vnode *vp;
2643         int error;
2644
2645         if ((error = getutimes(tptr, ts)) != 0)
2646                 return (error);
2647         /* XXX Add NLC flag indicating modifying operation? */
2648         if ((error = nlookup(nd)) != 0)
2649                 return (error);
2650         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2651                 return (error);
2652         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2653                 return (error);
2654         error = setutimes(vp, ts, tptr == NULL);
2655         vrele(vp);
2656         return (error);
2657 }
2658
2659 /*
2660  * utimes_args(char *path, struct timeval *tptr)
2661  *
2662  * Set the access and modification times of a file.
2663  */
2664 int
2665 sys_utimes(struct utimes_args *uap)
2666 {
2667         struct timeval tv[2];
2668         struct nlookupdata nd;
2669         int error;
2670
2671         if (uap->tptr) {
2672                 error = copyin(uap->tptr, tv, sizeof(tv));
2673                 if (error)
2674                         return (error);
2675         }
2676         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2677         if (error == 0)
2678                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2679         nlookup_done(&nd);
2680         return (error);
2681 }
2682
2683 /*
2684  * lutimes_args(char *path, struct timeval *tptr)
2685  *
2686  * Set the access and modification times of a file.
2687  */
2688 int
2689 sys_lutimes(struct lutimes_args *uap)
2690 {
2691         struct timeval tv[2];
2692         struct nlookupdata nd;
2693         int error;
2694
2695         if (uap->tptr) {
2696                 error = copyin(uap->tptr, tv, sizeof(tv));
2697                 if (error)
2698                         return (error);
2699         }
2700         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2701         if (error == 0)
2702                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2703         nlookup_done(&nd);
2704         return (error);
2705 }
2706
2707 int
2708 kern_futimes(int fd, struct timeval *tptr)
2709 {
2710         struct thread *td = curthread;
2711         struct proc *p = td->td_proc;
2712         struct timespec ts[2];
2713         struct file *fp;
2714         int error;
2715
2716         error = getutimes(tptr, ts);
2717         if (error)
2718                 return (error);
2719         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2720                 return (error);
2721         if (fp->f_nchandle.ncp)
2722                 error = ncp_writechk(&fp->f_nchandle);
2723         if (error == 0)
2724                 error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2725         fdrop(fp);
2726         return (error);
2727 }
2728
2729 /*
2730  * futimes_args(int fd, struct timeval *tptr)
2731  *
2732  * Set the access and modification times of a file.
2733  */
2734 int
2735 sys_futimes(struct futimes_args *uap)
2736 {
2737         struct timeval tv[2];
2738         int error;
2739
2740         if (uap->tptr) {
2741                 error = copyin(uap->tptr, tv, sizeof(tv));
2742                 if (error)
2743                         return (error);
2744         }
2745
2746         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2747
2748         return (error);
2749 }
2750
2751 int
2752 kern_truncate(struct nlookupdata *nd, off_t length)
2753 {
2754         struct vnode *vp;
2755         struct vattr vattr;
2756         int error;
2757
2758         if (length < 0)
2759                 return(EINVAL);
2760         /* XXX Add NLC flag indicating modifying operation? */
2761         if ((error = nlookup(nd)) != 0)
2762                 return (error);
2763         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2764                 return (error);
2765         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2766                 return (error);
2767         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2768                 vrele(vp);
2769                 return (error);
2770         }
2771         if (vp->v_type == VDIR) {
2772                 error = EISDIR;
2773         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2774             (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2775                 VATTR_NULL(&vattr);
2776                 vattr.va_size = length;
2777                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2778         }
2779         vput(vp);
2780         return (error);
2781 }
2782
2783 /*
2784  * truncate(char *path, int pad, off_t length)
2785  *
2786  * Truncate a file given its path name.
2787  */
2788 int
2789 sys_truncate(struct truncate_args *uap)
2790 {
2791         struct nlookupdata nd;
2792         int error;
2793
2794         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2795         if (error == 0)
2796                 error = kern_truncate(&nd, uap->length);
2797         nlookup_done(&nd);
2798         return error;
2799 }
2800
2801 int
2802 kern_ftruncate(int fd, off_t length)
2803 {
2804         struct thread *td = curthread;
2805         struct proc *p = td->td_proc;
2806         struct vattr vattr;
2807         struct vnode *vp;
2808         struct file *fp;
2809         int error;
2810
2811         if (length < 0)
2812                 return(EINVAL);
2813         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2814                 return (error);
2815         if (fp->f_nchandle.ncp) {
2816                 error = ncp_writechk(&fp->f_nchandle);
2817                 if (error)
2818                         goto done;
2819         }
2820         if ((fp->f_flag & FWRITE) == 0) {
2821                 error = EINVAL;
2822                 goto done;
2823         }
2824         vp = (struct vnode *)fp->f_data;
2825         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2826         if (vp->v_type == VDIR) {
2827                 error = EISDIR;
2828         } else if ((error = vn_writechk(vp, NULL)) == 0) {
2829                 VATTR_NULL(&vattr);
2830                 vattr.va_size = length;
2831                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2832         }
2833         vn_unlock(vp);
2834 done:
2835         fdrop(fp);
2836         return (error);
2837 }
2838
2839 /*
2840  * ftruncate_args(int fd, int pad, off_t length)
2841  *
2842  * Truncate a file given a file descriptor.
2843  */
2844 int
2845 sys_ftruncate(struct ftruncate_args *uap)
2846 {
2847         int error;
2848
2849         error = kern_ftruncate(uap->fd, uap->length);
2850
2851         return (error);
2852 }
2853
2854 /*
2855  * fsync(int fd)
2856  *
2857  * Sync an open file.
2858  */
2859 /* ARGSUSED */
2860 int
2861 sys_fsync(struct fsync_args *uap)
2862 {
2863         struct thread *td = curthread;
2864         struct proc *p = td->td_proc;
2865         struct vnode *vp;
2866         struct file *fp;
2867         vm_object_t obj;
2868         int error;
2869
2870         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2871                 return (error);
2872         vp = (struct vnode *)fp->f_data;
2873         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2874         if ((obj = vp->v_object) != NULL)
2875                 vm_object_page_clean(obj, 0, 0, 0);
2876         if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
2877                 error = buf_fsync(vp);
2878         vn_unlock(vp);
2879         fdrop(fp);
2880         return (error);
2881 }
2882
2883 int
2884 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2885 {
2886         struct nchandle fnchd;
2887         struct nchandle tnchd;
2888         struct namecache *ncp;
2889         struct vnode *fdvp;
2890         struct vnode *tdvp;
2891         struct mount *mp;
2892         int error;
2893
2894         bwillwrite();
2895         if ((error = nlookup(fromnd)) != 0)
2896                 return (error);
2897         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2898                 return (ENOENT);
2899         fnchd.mount = fromnd->nl_nch.mount;
2900         cache_hold(&fnchd);
2901
2902         /*
2903          * unlock the source nch so we can lookup the target nch without
2904          * deadlocking.  The target may or may not exist so we do not check
2905          * for a target vp like kern_mkdir() and other creation functions do.
2906          *
2907          * The source and target directories are ref'd and rechecked after
2908          * everything is relocked to determine if the source or target file
2909          * has been renamed.
2910          */
2911         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2912         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2913         cache_unlock(&fromnd->nl_nch);
2914
2915         tond->nl_flags |= NLC_CREATE;
2916         if ((error = nlookup(tond)) != 0) {
2917                 cache_drop(&fnchd);
2918                 return (error);
2919         }
2920         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2921                 cache_drop(&fnchd);
2922                 return (ENOENT);
2923         }
2924         tnchd.mount = tond->nl_nch.mount;
2925         cache_hold(&tnchd);
2926
2927         /*
2928          * If the source and target are the same there is nothing to do
2929          */
2930         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2931                 cache_drop(&fnchd);
2932                 cache_drop(&tnchd);
2933                 return (0);
2934         }
2935
2936         /*
2937          * Mount points cannot be renamed or overwritten
2938          */
2939         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2940             NCF_ISMOUNTPT
2941         ) {
2942                 cache_drop(&fnchd);
2943                 cache_drop(&tnchd);
2944                 return (EINVAL);
2945         }
2946
2947         /*
2948          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2949          * may have become invalid while it was unlocked, nc_vp and nc_mount
2950          * could be NULL.
2951          */
2952         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2953                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2954         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2955                 cache_lock(&fromnd->nl_nch);
2956                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2957         } else {
2958                 cache_unlock(&tond->nl_nch);
2959                 cache_lock(&fromnd->nl_nch);
2960                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2961                 cache_lock(&tond->nl_nch);
2962                 cache_resolve(&tond->nl_nch, tond->nl_cred);
2963         }
2964         fromnd->nl_flags |= NLC_NCPISLOCKED;
2965
2966         /*
2967          * make sure the parent directories linkages are the same
2968          */
2969         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2970             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2971                 cache_drop(&fnchd);
2972                 cache_drop(&tnchd);
2973                 return (ENOENT);
2974         }
2975
2976         /*
2977          * Both the source and target must be within the same filesystem and
2978          * in the same filesystem as their parent directories within the
2979          * namecache topology.
2980          *
2981          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2982          */
2983         mp = fnchd.mount;
2984         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2985             mp != tond->nl_nch.mount) {
2986                 cache_drop(&fnchd);
2987                 cache_drop(&tnchd);
2988                 return (EXDEV);
2989         }
2990
2991         /*
2992          * Make sure the mount point is writable
2993          */
2994         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2995                 cache_drop(&fnchd);
2996                 cache_drop(&tnchd);
2997                 return (error);
2998         }
2999
3000         /*
3001          * If the target exists and either the source or target is a directory,
3002          * then both must be directories.
3003          *
3004          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3005          * have become NULL.
3006          */
3007         if (tond->nl_nch.ncp->nc_vp) {
3008                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3009                         error = ENOENT;
3010                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3011                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3012                                 error = ENOTDIR;
3013                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3014                         error = EISDIR;
3015                 }
3016         }
3017
3018         /*
3019          * You cannot rename a source into itself or a subdirectory of itself.
3020          * We check this by travsersing the target directory upwards looking
3021          * for a match against the source.
3022          */
3023         if (error == 0) {
3024                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3025                         if (fromnd->nl_nch.ncp == ncp) {
3026                                 error = EINVAL;
3027                                 break;
3028                         }
3029                 }
3030         }
3031
3032         cache_drop(&fnchd);
3033         cache_drop(&tnchd);
3034
3035         /*
3036          * Even though the namespaces are different, they may still represent
3037          * hardlinks to the same file.  The filesystem might have a hard time
3038          * with this so we issue a NREMOVE of the source instead of a NRENAME
3039          * when we detect the situation.
3040          */
3041         if (error == 0) {
3042                 fdvp = fromnd->nl_nch.ncp->nc_parent->nc_vp;
3043                 tdvp = tond->nl_nch.ncp->nc_parent->nc_vp;
3044                 if (fdvp == NULL || tdvp == NULL) {
3045                         error = EPERM;
3046                 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3047                         /* vhold(fdvp); - dvp can't go away */
3048                         error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3049                                             fromnd->nl_cred);
3050                         /* vdrop(fdvp); */
3051                 } else {
3052                         /* vhold(fdvp); - dvp can't go away */
3053                         /* vhold(tdvp); - dvp can't go away */
3054                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3055                                             fdvp, tdvp, tond->nl_cred);
3056                         /* vdrop(fdvp); */
3057                         /* vdrop(tdvp); */
3058                 }
3059         }
3060         return (error);
3061 }
3062
3063 /*
3064  * rename_args(char *from, char *to)
3065  *
3066  * Rename files.  Source and destination must either both be directories,
3067  * or both not be directories.  If target is a directory, it must be empty.
3068  */
3069 int
3070 sys_rename(struct rename_args *uap)
3071 {
3072         struct nlookupdata fromnd, tond;
3073         int error;
3074
3075         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3076         if (error == 0) {
3077                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3078                 if (error == 0)
3079                         error = kern_rename(&fromnd, &tond);
3080                 nlookup_done(&tond);
3081         }
3082         nlookup_done(&fromnd);
3083         return (error);
3084 }
3085
3086 int
3087 kern_mkdir(struct nlookupdata *nd, int mode)
3088 {
3089         struct thread *td = curthread;
3090         struct proc *p = td->td_proc;
3091         struct vnode *vp;
3092         struct vnode *dvp;
3093         struct vattr vattr;
3094         int error;
3095
3096         bwillwrite();
3097         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
3098         if ((error = nlookup(nd)) != 0)
3099                 return (error);
3100
3101         if (nd->nl_nch.ncp->nc_vp)
3102                 return (EEXIST);
3103         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3104                 return (error);
3105         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
3106                 return (EPERM);
3107         /* vhold(dvp); - dvp can't go away */
3108         VATTR_NULL(&vattr);
3109         vattr.va_type = VDIR;
3110         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3111
3112         vp = NULL;
3113         error = VOP_NMKDIR(&nd->nl_nch, dvp, &vp, p->p_ucred, &vattr);
3114         /* vdrop(dvp); */
3115         if (error == 0)
3116                 vput(vp);
3117         return (error);
3118 }
3119
3120 /*
3121  * mkdir_args(char *path, int mode)
3122  *
3123  * Make a directory file.
3124  */
3125 /* ARGSUSED */
3126 int
3127 sys_mkdir(struct mkdir_args *uap)
3128 {
3129         struct nlookupdata nd;
3130         int error;
3131
3132         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3133         if (error == 0)
3134                 error = kern_mkdir(&nd, uap->mode);
3135         nlookup_done(&nd);
3136         return (error);
3137 }
3138
3139 int
3140 kern_rmdir(struct nlookupdata *nd)
3141 {
3142         struct vnode *dvp;
3143         int error;
3144
3145         bwillwrite();
3146         nd->nl_flags |= NLC_DELETE;
3147         if ((error = nlookup(nd)) != 0)
3148                 return (error);
3149
3150         /*
3151          * Do not allow directories representing mount points to be
3152          * deleted, even if empty.  Check write perms on mount point
3153          * in case the vnode is aliased (aka nullfs).
3154          */
3155         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3156                 return (EINVAL);
3157         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3158                 return (error);
3159         if ((dvp = nd->nl_nch.ncp->nc_parent->nc_vp) == NULL)
3160                 return (EPERM);
3161         /* vhold(dvp); - dvp can't go away */
3162         error = VOP_NRMDIR(&nd->nl_nch, dvp, nd->nl_cred);
3163         /* vdrop(dvp); */
3164         return (error);
3165 }
3166
3167 /*
3168  * rmdir_args(char *path)
3169  *
3170  * Remove a directory file.
3171  */
3172 /* ARGSUSED */
3173 int
3174 sys_rmdir(struct rmdir_args *uap)
3175 {
3176         struct nlookupdata nd;
3177         int error;
3178
3179         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3180         if (error == 0)
3181                 error = kern_rmdir(&nd);
3182         nlookup_done(&nd);
3183         return (error);
3184 }
3185
3186 int
3187 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3188     enum uio_seg direction)
3189 {
3190         struct thread *td = curthread;
3191         struct proc *p = td->td_proc;
3192         struct vnode *vp;
3193         struct file *fp;
3194         struct uio auio;
3195         struct iovec aiov;
3196         off_t loff;
3197         int error, eofflag;
3198
3199         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3200                 return (error);
3201         if ((fp->f_flag & FREAD) == 0) {
3202                 error = EBADF;
3203                 goto done;
3204         }
3205         vp = (struct vnode *)fp->f_data;
3206 unionread:
3207         if (vp->v_type != VDIR) {
3208                 error = EINVAL;
3209                 goto done;
3210         }
3211         aiov.iov_base = buf;
3212         aiov.iov_len = count;
3213         auio.uio_iov = &aiov;
3214         auio.uio_iovcnt = 1;
3215         auio.uio_rw = UIO_READ;
3216         auio.uio_segflg = direction;
3217         auio.uio_td = td;
3218         auio.uio_resid = count;
3219         loff = auio.uio_offset = fp->f_offset;
3220         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3221         fp->f_offset = auio.uio_offset;
3222         if (error)
3223                 goto done;
3224         if (count == auio.uio_resid) {
3225                 if (union_dircheckp) {
3226                         error = union_dircheckp(td, &vp, fp);
3227                         if (error == -1)
3228                                 goto unionread;
3229                         if (error)
3230                                 goto done;
3231                 }
3232 #if 0
3233                 if ((vp->v_flag & VROOT) &&
3234                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3235                         struct vnode *tvp = vp;
3236                         vp = vp->v_mount->mnt_vnodecovered;
3237                         vref(vp);
3238                         fp->f_data = vp;
3239                         fp->f_offset = 0;
3240                         vrele(tvp);
3241                         goto unionread;
3242                 }
3243 #endif
3244         }
3245
3246         /*
3247          * WARNING!  *basep may not be wide enough to accomodate the
3248          * seek offset.   XXX should we hack this to return the upper 32 bits
3249          * for offsets greater then 4G?
3250          */
3251         if (basep) {
3252                 *basep = (long)loff;
3253         }
3254         *res = count - auio.uio_resid;
3255 done:
3256         fdrop(fp);
3257         return (error);
3258 }
3259
3260 /*
3261  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3262  *
3263  * Read a block of directory entries in a file system independent format.
3264  */
3265 int
3266 sys_getdirentries(struct getdirentries_args *uap)
3267 {
3268         long base;
3269         int error;
3270
3271         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3272                                    &uap->sysmsg_result, UIO_USERSPACE);
3273
3274         if (error == 0 && uap->basep)
3275                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3276         return (error);
3277 }
3278
3279 /*
3280  * getdents_args(int fd, char *buf, size_t count)
3281  */
3282 int
3283 sys_getdents(struct getdents_args *uap)
3284 {
3285         int error;
3286
3287         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3288             &uap->sysmsg_result, UIO_USERSPACE);
3289
3290         return (error);
3291 }
3292
3293 /*
3294  * umask(int newmask)
3295  *
3296  * Set the mode mask for creation of filesystem nodes.
3297  *
3298  * MP SAFE
3299  */
3300 int
3301 sys_umask(struct umask_args *uap)
3302 {
3303         struct thread *td = curthread;
3304         struct proc *p = td->td_proc;
3305         struct filedesc *fdp;
3306
3307         fdp = p->p_fd;
3308         uap->sysmsg_result = fdp->fd_cmask;
3309         fdp->fd_cmask = uap->newmask & ALLPERMS;
3310         return (0);
3311 }
3312
3313 /*
3314  * revoke(char *path)
3315  *
3316  * Void all references to file by ripping underlying filesystem
3317  * away from vnode.
3318  */
3319 /* ARGSUSED */
3320 int
3321 sys_revoke(struct revoke_args *uap)
3322 {
3323         struct nlookupdata nd;
3324         struct vattr vattr;
3325         struct vnode *vp;
3326         struct ucred *cred;
3327         int error;
3328
3329         vp = NULL;
3330         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3331         if (error == 0)
3332                 error = nlookup(&nd);
3333         if (error == 0)
3334                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3335         cred = crhold(nd.nl_cred);
3336         nlookup_done(&nd);
3337         if (error == 0) {
3338                 if (vp->v_type != VCHR && vp->v_type != VBLK)
3339                         error = EINVAL;
3340                 if (error == 0)
3341                         error = VOP_GETATTR(vp, &vattr);
3342                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3343                         error = suser_cred(cred, PRISON_ROOT);
3344                 if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3345                         error = 0;
3346                         vx_lock(vp);
3347                         VOP_REVOKE(vp, REVOKEALL);
3348                         vx_unlock(vp);
3349                 }
3350                 vrele(vp);
3351         }
3352         if (cred)
3353                 crfree(cred);
3354         return (error);
3355 }
3356
3357 /*
3358  * getfh_args(char *fname, fhandle_t *fhp)
3359  *
3360  * Get (NFS) file handle
3361  */
3362 int
3363 sys_getfh(struct getfh_args *uap)
3364 {
3365         struct thread *td = curthread;
3366         struct nlookupdata nd;
3367         fhandle_t fh;
3368         struct vnode *vp;
3369         int error;
3370
3371         /*
3372          * Must be super user
3373          */
3374         if ((error = suser(td)) != 0)
3375                 return (error);
3376
3377         vp = NULL;
3378         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3379         if (error == 0)
3380                 error = nlookup(&nd);
3381         if (error == 0)
3382                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3383         nlookup_done(&nd);
3384         if (error == 0) {
3385                 bzero(&fh, sizeof(fh));
3386                 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3387                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3388                 vput(vp);
3389                 if (error == 0)
3390                         error = copyout(&fh, uap->fhp, sizeof(fh));
3391         }
3392         return (error);
3393 }
3394
3395 /*
3396  * fhopen_args(const struct fhandle *u_fhp, int flags)
3397  *
3398  * syscall for the rpc.lockd to use to translate a NFS file handle into
3399  * an open descriptor.
3400  *
3401  * warning: do not remove the suser() call or this becomes one giant
3402  * security hole.
3403  */
3404 int
3405 sys_fhopen(struct fhopen_args *uap)
3406 {
3407         struct thread *td = curthread;
3408         struct proc *p = td->td_proc;
3409         struct mount *mp;
3410         struct vnode *vp;
3411         struct fhandle fhp;
3412         struct vattr vat;
3413         struct vattr *vap = &vat;
3414         struct flock lf;
3415         int fmode, mode, error, type;
3416         struct file *nfp;
3417         struct file *fp;
3418         int indx;
3419
3420         /*
3421          * Must be super user
3422          */
3423         error = suser(td);
3424         if (error)
3425                 return (error);
3426
3427         fmode = FFLAGS(uap->flags);
3428         /* why not allow a non-read/write open for our lockd? */
3429         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3430                 return (EINVAL);
3431         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3432         if (error)
3433                 return(error);
3434         /* find the mount point */
3435         mp = vfs_getvfs(&fhp.fh_fsid);
3436         if (mp == NULL)
3437                 return (ESTALE);
3438         /* now give me my vnode, it gets returned to me locked */
3439         error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3440         if (error)
3441                 return (error);
3442         /*
3443          * from now on we have to make sure not
3444          * to forget about the vnode
3445          * any error that causes an abort must vput(vp)
3446          * just set error = err and 'goto bad;'.
3447          */
3448
3449         /*
3450          * from vn_open
3451          */
3452         if (vp->v_type == VLNK) {
3453                 error = EMLINK;
3454                 goto bad;
3455         }
3456         if (vp->v_type == VSOCK) {
3457                 error = EOPNOTSUPP;
3458                 goto bad;
3459         }
3460         mode = 0;
3461         if (fmode & (FWRITE | O_TRUNC)) {
3462                 if (vp->v_type == VDIR) {
3463                         error = EISDIR;
3464                         goto bad;
3465                 }
3466                 error = vn_writechk(vp, NULL);
3467                 if (error)
3468                         goto bad;
3469                 mode |= VWRITE;
3470         }
3471         if (fmode & FREAD)
3472                 mode |= VREAD;
3473         if (mode) {
3474                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3475                 if (error)
3476                         goto bad;
3477         }
3478         if (fmode & O_TRUNC) {
3479                 vn_unlock(vp);                          /* XXX */
3480                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3481                 VATTR_NULL(vap);
3482                 vap->va_size = 0;
3483                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3484                 if (error)
3485                         goto bad;
3486         }
3487
3488         /*
3489          * VOP_OPEN needs the file pointer so it can potentially override
3490          * it.
3491          *
3492          * WARNING! no f_nchandle will be associated when fhopen()ing a
3493          * directory.  XXX
3494          */
3495         if ((error = falloc(p, &nfp, &indx)) != 0)
3496                 goto bad;
3497         fp = nfp;
3498
3499         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3500         if (error) {
3501                 /*
3502                  * setting f_ops this way prevents VOP_CLOSE from being
3503                  * called or fdrop() releasing the vp from v_data.   Since
3504                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3505                  */
3506                 fp->f_ops = &badfileops;
3507                 fp->f_data = NULL;
3508                 goto bad_drop;
3509         }
3510
3511         /*
3512          * The fp is given its own reference, we still have our ref and lock.
3513          *
3514          * Assert that all regular files must be created with a VM object.
3515          */
3516         if (vp->v_type == VREG && vp->v_object == NULL) {
3517                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3518                 goto bad_drop;
3519         }
3520
3521         /*
3522          * The open was successful.  Handle any locking requirements.
3523          */
3524         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3525                 lf.l_whence = SEEK_SET;
3526                 lf.l_start = 0;
3527                 lf.l_len = 0;
3528                 if (fmode & O_EXLOCK)
3529                         lf.l_type = F_WRLCK;
3530                 else
3531                         lf.l_type = F_RDLCK;
3532                 if (fmode & FNONBLOCK)
3533                         type = 0;
3534                 else
3535                         type = F_WAIT;
3536                 vn_unlock(vp);
3537                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3538                         /*
3539                          * release our private reference.
3540                          */
3541                         fsetfd(p, NULL, indx);
3542                         fdrop(fp);
3543                         vrele(vp);
3544                         return (error);
3545                 }
3546                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3547                 fp->f_flag |= FHASLOCK;
3548         }
3549
3550         /*
3551          * Clean up.  Associate the file pointer with the previously
3552          * reserved descriptor and return it.
3553          */
3554         vput(vp);
3555         fsetfd(p, fp, indx);
3556         fdrop(fp);
3557         uap->sysmsg_result = indx;
3558         return (0);
3559
3560 bad_drop:
3561         fsetfd(p, NULL, indx);
3562         fdrop(fp);
3563 bad:
3564         vput(vp);
3565         return (error);
3566 }
3567
3568 /*
3569  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3570  */
3571 int
3572 sys_fhstat(struct fhstat_args *uap)
3573 {
3574         struct thread *td = curthread;
3575         struct stat sb;
3576         fhandle_t fh;
3577         struct mount *mp;
3578         struct vnode *vp;
3579         int error;
3580
3581         /*
3582          * Must be super user
3583          */
3584         error = suser(td);
3585         if (error)
3586                 return (error);
3587
3588         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3589         if (error)
3590                 return (error);
3591
3592         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3593                 return (ESTALE);
3594         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3595                 return (error);
3596         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3597         vput(vp);
3598         if (error)
3599                 return (error);
3600         error = copyout(&sb, uap->sb, sizeof(sb));
3601         return (error);
3602 }
3603
3604 /*
3605  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3606  */
3607 int
3608 sys_fhstatfs(struct fhstatfs_args *uap)
3609 {
3610         struct thread *td = curthread;
3611         struct proc *p = td->td_proc;
3612         struct statfs *sp;
3613         struct mount *mp;
3614         struct vnode *vp;
3615         struct statfs sb;
3616         char *fullpath, *freepath;
3617         fhandle_t fh;
3618         int error;
3619
3620         /*
3621          * Must be super user
3622          */
3623         if ((error = suser(td)))
3624                 return (error);
3625
3626         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3627                 return (error);
3628
3629         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3630                 return (ESTALE);
3631
3632         if (p != NULL && !chroot_visible_mnt(mp, p))
3633                 return (ESTALE);
3634
3635         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3636                 return (error);
3637         mp = vp->v_mount;
3638         sp = &mp->mnt_stat;
3639         vput(vp);
3640         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3641                 return (error);
3642
3643         error = mount_path(p, mp, &fullpath, &freepath);
3644         if (error)
3645                 return(error);
3646         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3647         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3648         kfree(freepath, M_TEMP);
3649
3650         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3651         if (suser(td)) {
3652                 bcopy(sp, &sb, sizeof(sb));
3653                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3654                 sp = &sb;
3655         }
3656         return (copyout(sp, uap->buf, sizeof(*sp)));
3657 }
3658
3659 /*
3660  * Syscall to push extended attribute configuration information into the
3661  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3662  * a command (int cmd), and attribute name and misc data.  For now, the
3663  * attribute name is left in userspace for consumption by the VFS_op.
3664  * It will probably be changed to be copied into sysspace by the
3665  * syscall in the future, once issues with various consumers of the
3666  * attribute code have raised their hands.
3667  *
3668  * Currently this is used only by UFS Extended Attributes.
3669  */
3670 int
3671 sys_extattrctl(struct extattrctl_args *uap)
3672 {
3673         struct nlookupdata nd;
3674         struct mount *mp;
3675         struct vnode *vp;
3676         int error;
3677
3678         vp = NULL;
3679         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3680         if (error == 0)
3681                 error = nlookup(&nd);
3682         if (error == 0) {
3683                 mp = nd.nl_nch.mount;
3684                 error = VFS_EXTATTRCTL(mp, uap->cmd,
3685                                 uap->attrname, uap->arg,
3686                                 nd.nl_cred);
3687         }
3688         nlookup_done(&nd);
3689         return (error);
3690 }
3691
3692 /*
3693  * Syscall to set a named extended attribute on a file or directory.
3694  * Accepts attribute name, and a uio structure pointing to the data to set.
3695  * The uio is consumed in the style of writev().  The real work happens
3696  * in VOP_SETEXTATTR().
3697  */
3698 int
3699 sys_extattr_set_file(struct extattr_set_file_args *uap)
3700 {
3701         char attrname[EXTATTR_MAXNAMELEN];
3702         struct iovec aiov[UIO_SMALLIOV];
3703         struct iovec *needfree;
3704         struct nlookupdata nd;
3705         struct iovec *iov;
3706         struct vnode *vp;
3707         struct uio auio;
3708         u_int iovlen;
3709         u_int cnt;
3710         int error;
3711         int i;
3712
3713         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3714         if (error)
3715                 return (error);
3716
3717         vp = NULL;
3718         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3719         if (error == 0)
3720                 error = nlookup(&nd);
3721         if (error == 0)
3722                 error = ncp_writechk(&nd.nl_nch);
3723         if (error == 0)
3724                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3725         if (error) {
3726                 nlookup_done(&nd);
3727                 return (error);
3728         }
3729
3730         needfree = NULL;
3731         iovlen = uap->iovcnt * sizeof(struct iovec);
3732         if (uap->iovcnt > UIO_SMALLIOV) {
3733                 if (uap->iovcnt > UIO_MAXIOV) {
3734                         error = EINVAL;
3735                         goto done;
3736                 }
3737                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3738                 needfree = iov;
3739         } else {
3740                 iov = aiov;
3741         }
3742         auio.uio_iov = iov;
3743         auio.uio_iovcnt = uap->iovcnt;
3744         auio.uio_rw = UIO_WRITE;
3745         auio.uio_segflg = UIO_USERSPACE;
3746         auio.uio_td = nd.nl_td;
3747         auio.uio_offset = 0;
3748         if ((error = copyin(uap->iovp, iov, iovlen)))
3749                 goto done;
3750         auio.uio_resid = 0;
3751         for (i = 0; i < uap->iovcnt; i++) {
3752                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3753                         error = EINVAL;
3754                         goto done;
3755                 }
3756                 auio.uio_resid += iov->iov_len;
3757                 iov++;
3758         }
3759         cnt = auio.uio_resid;
3760         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3761         cnt -= auio.uio_resid;
3762         uap->sysmsg_result = cnt;
3763 done:
3764         vput(vp);
3765         nlookup_done(&nd);
3766         if (needfree)
3767                 FREE(needfree, M_IOV);
3768         return (error);
3769 }
3770
3771 /*
3772  * Syscall to get a named extended attribute on a file or directory.
3773  * Accepts attribute name, and a uio structure pointing to a buffer for the
3774  * data.  The uio is consumed in the style of readv().  The real work
3775  * happens in VOP_GETEXTATTR();
3776  */
3777 int
3778 sys_extattr_get_file(struct extattr_get_file_args *uap)
3779 {
3780         char attrname[EXTATTR_MAXNAMELEN];
3781         struct iovec aiov[UIO_SMALLIOV];
3782         struct iovec *needfree;
3783         struct nlookupdata nd;
3784         struct iovec *iov;
3785         struct vnode *vp;
3786         struct uio auio;
3787         u_int iovlen;
3788         u_int cnt;
3789         int error;
3790         int i;
3791
3792         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3793         if (error)
3794                 return (error);
3795
3796         vp = NULL;
3797         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3798         if (error == 0)
3799                 error = nlookup(&nd);
3800         if (error == 0)
3801                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3802         if (error) {
3803                 nlookup_done(&nd);
3804                 return (error);
3805         }
3806
3807         iovlen = uap->iovcnt * sizeof (struct iovec);
3808         needfree = NULL;
3809         if (uap->iovcnt > UIO_SMALLIOV) {
3810                 if (uap->iovcnt > UIO_MAXIOV) {
3811                         error = EINVAL;
3812                         goto done;
3813                 }
3814                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3815                 needfree = iov;
3816         } else {
3817                 iov = aiov;
3818         }
3819         auio.uio_iov = iov;
3820         auio.uio_iovcnt = uap->iovcnt;
3821         auio.uio_rw = UIO_READ;
3822         auio.uio_segflg = UIO_USERSPACE;
3823         auio.uio_td = nd.nl_td;
3824         auio.uio_offset = 0;
3825         if ((error = copyin(uap->iovp, iov, iovlen)))
3826                 goto done;
3827         auio.uio_resid = 0;
3828         for (i = 0; i < uap->iovcnt; i++) {
3829                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3830                         error = EINVAL;
3831                         goto done;
3832                 }
3833                 auio.uio_resid += iov->iov_len;
3834                 iov++;
3835         }
3836         cnt = auio.uio_resid;
3837         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3838         cnt -= auio.uio_resid;
3839         uap->sysmsg_result = cnt;
3840 done:
3841         vput(vp);
3842         nlookup_done(&nd);
3843         if (needfree)
3844                 FREE(needfree, M_IOV);
3845         return(error);
3846 }
3847
3848 /*
3849  * Syscall to delete a named extended attribute from a file or directory.
3850  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3851  */
3852 int
3853 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3854 {
3855         char attrname[EXTATTR_MAXNAMELEN];
3856         struct nlookupdata nd;
3857         struct vnode *vp;
3858         int error;
3859
3860         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3861         if (error)
3862                 return(error);
3863
3864         vp = NULL;
3865         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3866         if (error == 0)
3867                 error = nlookup(&nd);
3868         if (error == 0)
3869                 error = ncp_writechk(&nd.nl_nch);
3870         if (error == 0)
3871                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3872         if (error) {
3873                 nlookup_done(&nd);
3874                 return (error);
3875         }
3876
3877         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3878         vput(vp);
3879         nlookup_done(&nd);
3880         return(error);
3881 }
3882
3883 /*
3884  * Determine if the mount is visible to the process.
3885  */
3886 static int
3887 chroot_visible_mnt(struct mount *mp, struct proc *p)
3888 {
3889         struct nchandle nch;
3890
3891         /*
3892          * Traverse from the mount point upwards.  If we hit the process
3893          * root then the mount point is visible to the process.
3894          */
3895         nch = mp->mnt_ncmountpt;
3896         while (nch.ncp) {
3897                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
3898                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
3899                         return(1);
3900                 }
3901                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3902                         nch = nch.mount->mnt_ncmounton;
3903                 } else {
3904                         nch.ncp = nch.ncp->nc_parent;
3905                 }
3906         }
3907
3908         /*
3909          * If the mount point is not visible to the process, but the
3910          * process root is in a subdirectory of the mount, return
3911          * TRUE anyway.
3912          */
3913         if (p->p_fd->fd_nrdir.mount == mp)
3914                 return(1);
3915
3916         return(0);
3917 }
3918