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 1.127 2008/05/18 05:54:25 dillon 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         struct vnode *vp;
 611         int error;
 612         int async_flag;
 613         int lflags;
 614         int freeok = 1;
 615
 616         /*
 617          * Exclusive access for unmounting purposes
 618          */
 619         if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
 620                 return (error);
 621
 622         /*
 623          * Allow filesystems to detect that a forced unmount is in progress.
 624          */
 625         if (flags & MNT_FORCE)
 626                 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
 627         lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
 628         error = lockmgr(&mp->mnt_lock, lflags);
 629         if (error) {
 630                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 631                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 632                         wakeup(mp);
 633                 return (error);
 634         }
 635
 636         if (mp->mnt_flag & MNT_EXPUBLIC)
 637                 vfs_setpublicfs(NULL, NULL, NULL);
 638
 639         vfs_msync(mp, MNT_WAIT);
 640         async_flag = mp->mnt_flag & MNT_ASYNC;
 641         mp->mnt_flag &=~ MNT_ASYNC;
 642
 643         /*
 644          * If this filesystem isn't aliasing other filesystems,
 645          * try to invalidate any remaining namecache entries and
 646          * check the count afterwords.
 647          */
 648         if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
 649                 cache_lock(&mp->mnt_ncmountpt);
 650                 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
 651                 cache_unlock(&mp->mnt_ncmountpt);
 652
 653                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 654                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 655
 656                         if ((flags & MNT_FORCE) == 0) {
 657                                 error = EBUSY;
 658                                 mount_warning(mp, "Cannot unmount: "
 659                                                   "%d namecache "
 660                                                   "references still "
 661                                                   "present",
 662                                                   ncp->nc_refs - 1);
 663                         } else {
 664                                 mount_warning(mp, "Forced unmount: "
 665                                                   "%d namecache "
 666                                                   "references still "
 667                                                   "present",
 668                                                   ncp->nc_refs - 1);
 669                                 freeok = 0;
 670                         }
 671                 }
 672         }
 673
 674         /*
 675          * nchandle records ref the mount structure.  Expect a count of 1
 676          * (our mount->mnt_ncmountpt).
 677          */
 678         if (mp->mnt_refs != 1) {
 679                 if ((flags & MNT_FORCE) == 0) {
 680                         mount_warning(mp, "Cannot unmount: "
 681                                           "%d process references still "
 682                                           "present", mp->mnt_refs);
 683                         error = EBUSY;
 684                 } else {
 685                         mount_warning(mp, "Forced unmount: "
 686                                           "%d process references still "
 687                                           "present", mp->mnt_refs);
 688                         freeok = 0;
 689                 }
 690         }
 691
 692         /*
 693          * Decomission our special mnt_syncer vnode.  This also stops
 694          * the vnlru code.  If we are unable to unmount we recommission
 695          * the vnode.
 696          */
 697         if (error == 0) {
 698                 if ((vp = mp->mnt_syncer) != NULL) {
 699                         mp->mnt_syncer = NULL;
 700                         vrele(vp);
 701                 }
 702                 if (((mp->mnt_flag & MNT_RDONLY) ||
 703                      (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
 704                     (flags & MNT_FORCE)) {
 705                         error = VFS_UNMOUNT(mp, flags);
 706                 }
 707         }
 708         if (error) {
 709                 if (mp->mnt_syncer == NULL)
 710                         vfs_allocate_syncvnode(mp);
 711                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 712                 mp->mnt_flag |= async_flag;
 713                 lockmgr(&mp->mnt_lock, LK_RELEASE);
 714                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 715                         wakeup(mp);
 716                 return (error);
 717         }
 718         /*
 719          * Clean up any journals still associated with the mount after
 720          * filesystem activity has ceased.
 721          */
 722         journal_remove_all_journals(mp,
 723             ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
 724
 725         mountlist_remove(mp);
 726
 727         /*
 728          * Remove any installed vnode ops here so the individual VFSs don't
 729          * have to.
 730          */
 731         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 732         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 733         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 734         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 735         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 736
 737         if (mp->mnt_ncmountpt.ncp != NULL) {
 738                 nch = mp->mnt_ncmountpt;
 739                 cache_zero(&mp->mnt_ncmountpt);
 740                 cache_clrmountpt(&nch);
 741                 cache_drop(&nch);
 742         }
 743         if (mp->mnt_ncmounton.ncp != NULL) {
 744                 nch = mp->mnt_ncmounton;
 745                 cache_zero(&mp->mnt_ncmounton);
 746                 cache_clrmountpt(&nch);
 747                 cache_drop(&nch);
 748         }
 749
 750         mp->mnt_vfc->vfc_refcount--;
 751         if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 752                 panic("unmount: dangling vnode");
 753         lockmgr(&mp->mnt_lock, LK_RELEASE);
 754         if (mp->mnt_kern_flag & MNTK_MWAIT)
 755                 wakeup(mp);
 756         if (freeok)
 757                 kfree(mp, M_MOUNT);
 758         return (0);
 759 }
 760
 761 static
 762 void
 763 mount_warning(struct mount *mp, const char *ctl, ...)
 764 {
 765         char *ptr;
 766         char *buf;
 767         __va_list va;
 768
 769         __va_start(va, ctl);
 770         if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
 771                 kprintf("unmount(%s): ", ptr);
 772                 kvprintf(ctl, va);
 773                 kprintf("\n");
 774                 kfree(buf, M_TEMP);
 775         } else {
 776                 kprintf("unmount(%p", mp);
 777                 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
 778                         kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
 779                 kprintf("): ");
 780                 kvprintf(ctl, va);
 781                 kprintf("\n");
 782         }
 783         __va_end(va);
 784 }
 785
 786 /*
 787  * Shim cache_fullpath() to handle the case where a process is chrooted into
 788  * a subdirectory of a mount.  In this case if the root mount matches the
 789  * process root directory's mount we have to specify the process's root
 790  * directory instead of the mount point, because the mount point might
 791  * be above the root directory.
 792  */
 793 static
 794 int
 795 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
 796 {
 797         struct nchandle *nch;
 798
 799         if (p && p->p_fd->fd_nrdir.mount == mp)
 800                 nch = &p->p_fd->fd_nrdir;
 801         else
 802                 nch = &mp->mnt_ncmountpt;
 803         return(cache_fullpath(p, nch, rb, fb));
 804 }
 805
 806 /*
 807  * Sync each mounted filesystem.
 808  */
 809
 810 #ifdef DEBUG
 811 static int syncprt = 0;
 812 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
 813 #endif /* DEBUG */
 814
 815 static int sync_callback(struct mount *mp, void *data);
 816
 817 /* ARGSUSED */
 818 int
 819 sys_sync(struct sync_args *uap)
 820 {
 821         mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
 822 #ifdef DEBUG
 823         /*
 824          * print out buffer pool stat information on each sync() call.
 825          */
 826         if (syncprt)
 827                 vfs_bufstats();
 828 #endif /* DEBUG */
 829         return (0);
 830 }
 831
 832 static
 833 int
 834 sync_callback(struct mount *mp, void *data __unused)
 835 {
 836         int asyncflag;
 837
 838         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 839                 asyncflag = mp->mnt_flag & MNT_ASYNC;
 840                 mp->mnt_flag &= ~MNT_ASYNC;
 841                 vfs_msync(mp, MNT_NOWAIT);
 842                 VFS_SYNC(mp, MNT_NOWAIT);
 843                 mp->mnt_flag |= asyncflag;
 844         }
 845         return(0);
 846 }
 847
 848 /* XXX PRISON: could be per prison flag */
 849 static int prison_quotas;
 850 #if 0
 851 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
 852 #endif
 853
 854 /*
 855  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
 856  *
 857  * Change filesystem quotas.
 858  */
 859 /* ARGSUSED */
 860 int
 861 sys_quotactl(struct quotactl_args *uap)
 862 {
 863         struct nlookupdata nd;
 864         struct thread *td;
 865         struct proc *p;
 866         struct mount *mp;
 867         int error;
 868
 869         td = curthread;
 870         p = td->td_proc;
 871         if (p->p_ucred->cr_prison && !prison_quotas)
 872                 return (EPERM);
 873
 874         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 875         if (error == 0)
 876                 error = nlookup(&nd);
 877         if (error == 0) {
 878                 mp = nd.nl_nch.mount;
 879                 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
 880                                     uap->arg, nd.nl_cred);
 881         }
 882         nlookup_done(&nd);
 883         return (error);
 884 }
 885
 886 /*
 887  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
 888  *              void *buf, int buflen)
 889  *
 890  * This function operates on a mount point and executes the specified
 891  * operation using the specified control data, and possibly returns data.
 892  *
 893  * The actual number of bytes stored in the result buffer is returned, 0
 894  * if none, otherwise an error is returned.
 895  */
 896 /* ARGSUSED */
 897 int
 898 sys_mountctl(struct mountctl_args *uap)
 899 {
 900         struct thread *td = curthread;
 901         struct proc *p = td->td_proc;
 902         struct file *fp;
 903         void *ctl = NULL;
 904         void *buf = NULL;
 905         char *path = NULL;
 906         int error;
 907
 908         /*
 909          * Sanity and permissions checks.  We must be root.
 910          */
 911         KKASSERT(p);
 912         if (p->p_ucred->cr_prison != NULL)
 913                 return (EPERM);
 914         if ((error = suser(td)) != 0)
 915                 return (error);
 916
 917         /*
 918          * Argument length checks
 919          */
 920         if (uap->ctllen < 0 || uap->ctllen > 1024)
 921                 return (EINVAL);
 922         if (uap->buflen < 0 || uap->buflen > 16 * 1024)
 923                 return (EINVAL);
 924         if (uap->path == NULL)
 925                 return (EINVAL);
 926
 927         /*
 928          * Allocate the necessary buffers and copyin data
 929          */
 930         path = objcache_get(namei_oc, M_WAITOK);
 931         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 932         if (error)
 933                 goto done;
 934
 935         if (uap->ctllen) {
 936                 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
 937                 error = copyin(uap->ctl, ctl, uap->ctllen);
 938                 if (error)
 939                         goto done;
 940         }
 941         if (uap->buflen)
 942                 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
 943
 944         /*
 945          * Validate the descriptor
 946          */
 947         if (uap->fd >= 0) {
 948                 fp = holdfp(p->p_fd, uap->fd, -1);
 949                 if (fp == NULL) {
 950                         error = EBADF;
 951                         goto done;
 952                 }
 953         } else {
 954                 fp = NULL;
 955         }
 956
 957         /*
 958          * Execute the internal kernel function and clean up.
 959          */
 960         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
 961         if (fp)
 962                 fdrop(fp);
 963         if (error == 0 && uap->sysmsg_result > 0)
 964                 error = copyout(buf, uap->buf, uap->sysmsg_result);
 965 done:
 966         if (path)
 967                 objcache_put(namei_oc, path);
 968         if (ctl)
 969                 kfree(ctl, M_TEMP);
 970         if (buf)
 971                 kfree(buf, M_TEMP);
 972         return (error);
 973 }
 974
 975 /*
 976  * Execute a mount control operation by resolving the path to a mount point
 977  * and calling vop_mountctl().
 978  */
 979 int
 980 kern_mountctl(const char *path, int op, struct file *fp,
 981                 const void *ctl, int ctllen,
 982                 void *buf, int buflen, int *res)
 983 {
 984         struct vnode *vp;
 985         struct mount *mp;
 986         struct nlookupdata nd;
 987         int error;
 988
 989         *res = 0;
 990         vp = NULL;
 991         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
 992         if (error == 0)
 993                 error = nlookup(&nd);
 994         if (error == 0)
 995                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 996         nlookup_done(&nd);
 997         if (error)
 998                 return (error);
 999
1000         mp = vp->v_mount;
1001
1002         /*
1003          * Must be the root of the filesystem
1004          */
1005         if ((vp->v_flag & VROOT) == 0) {
1006                 vput(vp);
1007                 return (EINVAL);
1008         }
1009         error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1010                                 buf, buflen, res);
1011         vput(vp);
1012         return (error);
1013 }
1014
1015 int
1016 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1017 {
1018         struct thread *td = curthread;
1019         struct proc *p = td->td_proc;
1020         struct mount *mp;
1021         struct statfs *sp;
1022         char *fullpath, *freepath;
1023         int error;
1024
1025         if ((error = nlookup(nd)) != 0)
1026                 return (error);
1027         mp = nd->nl_nch.mount;
1028         sp = &mp->mnt_stat;
1029         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1030                 return (error);
1031
1032         error = mount_path(p, mp, &fullpath, &freepath);
1033         if (error)
1034                 return(error);
1035         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1036         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1037         kfree(freepath, M_TEMP);
1038
1039         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1040         bcopy(sp, buf, sizeof(*buf));
1041         /* Only root should have access to the fsid's. */
1042         if (suser(td))
1043                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1044         return (0);
1045 }
1046
1047 /*
1048  * statfs_args(char *path, struct statfs *buf)
1049  *
1050  * Get filesystem statistics.
1051  */
1052 int
1053 sys_statfs(struct statfs_args *uap)
1054 {
1055         struct nlookupdata nd;
1056         struct statfs buf;
1057         int error;
1058
1059         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1060         if (error == 0)
1061                 error = kern_statfs(&nd, &buf);
1062         nlookup_done(&nd);
1063         if (error == 0)
1064                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1065         return (error);
1066 }
1067
1068 int
1069 kern_fstatfs(int fd, struct statfs *buf)
1070 {
1071         struct thread *td = curthread;
1072         struct proc *p = td->td_proc;
1073         struct file *fp;
1074         struct mount *mp;
1075         struct statfs *sp;
1076         char *fullpath, *freepath;
1077         int error;
1078
1079         KKASSERT(p);
1080         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1081                 return (error);
1082         mp = ((struct vnode *)fp->f_data)->v_mount;
1083         if (mp == NULL) {
1084                 error = EBADF;
1085                 goto done;
1086         }
1087         if (fp->f_cred == NULL) {
1088                 error = EINVAL;
1089                 goto done;
1090         }
1091         sp = &mp->mnt_stat;
1092         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1093                 goto done;
1094
1095         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1096                 goto done;
1097         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1098         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1099         kfree(freepath, M_TEMP);
1100
1101         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1102         bcopy(sp, buf, sizeof(*buf));
1103
1104         /* Only root should have access to the fsid's. */
1105         if (suser(td))
1106                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1107         error = 0;
1108 done:
1109         fdrop(fp);
1110         return (error);
1111 }
1112
1113 /*
1114  * fstatfs_args(int fd, struct statfs *buf)
1115  *
1116  * Get filesystem statistics.
1117  */
1118 int
1119 sys_fstatfs(struct fstatfs_args *uap)
1120 {
1121         struct statfs buf;
1122         int error;
1123
1124         error = kern_fstatfs(uap->fd, &buf);
1125
1126         if (error == 0)
1127                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1128         return (error);
1129 }
1130
1131 /*
1132  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1133  *
1134  * Get statistics on all filesystems.
1135  */
1136
1137 struct getfsstat_info {
1138         struct statfs *sfsp;
1139         long count;
1140         long maxcount;
1141         int error;
1142         int flags;
1143         struct proc *p;
1144 };
1145
1146 static int getfsstat_callback(struct mount *, void *);
1147
1148 /* ARGSUSED */
1149 int
1150 sys_getfsstat(struct getfsstat_args *uap)
1151 {
1152         struct thread *td = curthread;
1153         struct proc *p = td->td_proc;
1154         struct getfsstat_info info;
1155
1156         bzero(&info, sizeof(info));
1157
1158         info.maxcount = uap->bufsize / sizeof(struct statfs);
1159         info.sfsp = uap->buf;
1160         info.count = 0;
1161         info.flags = uap->flags;
1162         info.p = p;
1163
1164         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1165         if (info.sfsp && info.count > info.maxcount)
1166                 uap->sysmsg_result = info.maxcount;
1167         else
1168                 uap->sysmsg_result = info.count;
1169         return (info.error);
1170 }
1171
1172 static int
1173 getfsstat_callback(struct mount *mp, void *data)
1174 {
1175         struct getfsstat_info *info = data;
1176         struct statfs *sp;
1177         char *freepath;
1178         char *fullpath;
1179         int error;
1180
1181         if (info->sfsp && info->count < info->maxcount) {
1182                 if (info->p && !chroot_visible_mnt(mp, info->p))
1183                         return(0);
1184                 sp = &mp->mnt_stat;
1185
1186                 /*
1187                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1188                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1189                  * overrides MNT_WAIT.
1190                  */
1191                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1192                     (info->flags & MNT_WAIT)) &&
1193                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1194                         return(0);
1195                 }
1196                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1197
1198                 error = mount_path(info->p, mp, &fullpath, &freepath);
1199                 if (error) {
1200                         info->error = error;
1201                         return(-1);
1202                 }
1203                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1204                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1205                 kfree(freepath, M_TEMP);
1206
1207                 error = copyout(sp, info->sfsp, sizeof(*sp));
1208                 if (error) {
1209                         info->error = error;
1210                         return (-1);
1211                 }
1212                 ++info->sfsp;
1213         }
1214         info->count++;
1215         return(0);
1216 }
1217
1218 /*
1219  * fchdir_args(int fd)
1220  *
1221  * Change current working directory to a given file descriptor.
1222  */
1223 /* ARGSUSED */
1224 int
1225 sys_fchdir(struct fchdir_args *uap)
1226 {
1227         struct thread *td = curthread;
1228         struct proc *p = td->td_proc;
1229         struct filedesc *fdp = p->p_fd;
1230         struct vnode *vp, *ovp;
1231         struct mount *mp;
1232         struct file *fp;
1233         struct nchandle nch, onch, tnch;
1234         int error;
1235
1236         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1237                 return (error);
1238         vp = (struct vnode *)fp->f_data;
1239         vref(vp);
1240         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1241         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1242                 error = ENOTDIR;
1243         else
1244                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1245         if (error) {
1246                 vput(vp);
1247                 fdrop(fp);
1248                 return (error);
1249         }
1250         cache_copy(&fp->f_nchandle, &nch);
1251
1252         /*
1253          * If the ncp has become a mount point, traverse through
1254          * the mount point.
1255          */
1256
1257         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1258                (mp = cache_findmount(&nch)) != NULL
1259         ) {
1260                 error = nlookup_mp(mp, &tnch);
1261                 if (error == 0) {
1262                         cache_unlock(&tnch);    /* leave ref intact */
1263                         vput(vp);
1264                         vp = tnch.ncp->nc_vp;
1265                         error = vget(vp, LK_SHARED);
1266                         KKASSERT(error == 0);
1267                         cache_drop(&nch);
1268                         nch = tnch;
1269                 }
1270         }
1271         if (error == 0) {
1272                 ovp = fdp->fd_cdir;
1273                 onch = fdp->fd_ncdir;
1274                 vn_unlock(vp);          /* leave ref intact */
1275                 fdp->fd_cdir = vp;
1276                 fdp->fd_ncdir = nch;
1277                 cache_drop(&onch);
1278                 vrele(ovp);
1279         } else {
1280                 cache_drop(&nch);
1281                 vput(vp);
1282         }
1283         fdrop(fp);
1284         return (error);
1285 }
1286
1287 int
1288 kern_chdir(struct nlookupdata *nd)
1289 {
1290         struct thread *td = curthread;
1291         struct proc *p = td->td_proc;
1292         struct filedesc *fdp = p->p_fd;
1293         struct vnode *vp, *ovp;
1294         struct nchandle onch;
1295         int error;
1296
1297         if ((error = nlookup(nd)) != 0)
1298                 return (error);
1299         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1300                 return (ENOENT);
1301         if ((error = vget(vp, LK_SHARED)) != 0)
1302                 return (error);
1303
1304         error = checkvp_chdir(vp, td);
1305         vn_unlock(vp);
1306         if (error == 0) {
1307                 ovp = fdp->fd_cdir;
1308                 onch = fdp->fd_ncdir;
1309                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1310                 fdp->fd_ncdir = nd->nl_nch;
1311                 fdp->fd_cdir = vp;
1312                 cache_drop(&onch);
1313                 vrele(ovp);
1314                 cache_zero(&nd->nl_nch);
1315         } else {
1316                 vrele(vp);
1317         }
1318         return (error);
1319 }
1320
1321 /*
1322  * chdir_args(char *path)
1323  *
1324  * Change current working directory (``.'').
1325  */
1326 int
1327 sys_chdir(struct chdir_args *uap)
1328 {
1329         struct nlookupdata nd;
1330         int error;
1331
1332         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1333         if (error == 0)
1334                 error = kern_chdir(&nd);
1335         nlookup_done(&nd);
1336         return (error);
1337 }
1338
1339 /*
1340  * Helper function for raised chroot(2) security function:  Refuse if
1341  * any filedescriptors are open directories.
1342  */
1343 static int
1344 chroot_refuse_vdir_fds(struct filedesc *fdp)
1345 {
1346         struct vnode *vp;
1347         struct file *fp;
1348         int error;
1349         int fd;
1350
1351         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1352                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1353                         continue;
1354                 vp = (struct vnode *)fp->f_data;
1355                 if (vp->v_type != VDIR) {
1356                         fdrop(fp);
1357                         continue;
1358                 }
1359                 fdrop(fp);
1360                 return(EPERM);
1361         }
1362         return (0);
1363 }
1364
1365 /*
1366  * This sysctl determines if we will allow a process to chroot(2) if it
1367  * has a directory open:
1368  *      0: disallowed for all processes.
1369  *      1: allowed for processes that were not already chroot(2)'ed.
1370  *      2: allowed for all processes.
1371  */
1372
1373 static int chroot_allow_open_directories = 1;
1374
1375 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1376      &chroot_allow_open_directories, 0, "");
1377
1378 /*
1379  * chroot to the specified namecache entry.  We obtain the vp from the
1380  * namecache data.  The passed ncp must be locked and referenced and will
1381  * remain locked and referenced on return.
1382  */
1383 int
1384 kern_chroot(struct nchandle *nch)
1385 {
1386         struct thread *td = curthread;
1387         struct proc *p = td->td_proc;
1388         struct filedesc *fdp = p->p_fd;
1389         struct vnode *vp;
1390         int error;
1391
1392         /*
1393          * Only root can chroot
1394          */
1395         if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1396                 return (error);
1397
1398         /*
1399          * Disallow open directory descriptors (fchdir() breakouts).
1400          */
1401         if (chroot_allow_open_directories == 0 ||
1402            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1403                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1404                         return (error);
1405         }
1406         if ((vp = nch->ncp->nc_vp) == NULL)
1407                 return (ENOENT);
1408
1409         if ((error = vget(vp, LK_SHARED)) != 0)
1410                 return (error);
1411
1412         /*
1413          * Check the validity of vp as a directory to change to and
1414          * associate it with rdir/jdir.
1415          */
1416         error = checkvp_chdir(vp, td);
1417         vn_unlock(vp);                  /* leave reference intact */
1418         if (error == 0) {
1419                 vrele(fdp->fd_rdir);
1420                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1421                 cache_drop(&fdp->fd_nrdir);
1422                 cache_copy(nch, &fdp->fd_nrdir);
1423                 if (fdp->fd_jdir == NULL) {
1424                         fdp->fd_jdir = vp;
1425                         vref(fdp->fd_jdir);
1426                         cache_copy(nch, &fdp->fd_njdir);
1427                 }
1428         } else {
1429                 vrele(vp);
1430         }
1431         return (error);
1432 }
1433
1434 /*
1435  * chroot_args(char *path)
1436  *
1437  * Change notion of root (``/'') directory.
1438  */
1439 /* ARGSUSED */
1440 int
1441 sys_chroot(struct chroot_args *uap)
1442 {
1443         struct thread *td = curthread;
1444         struct nlookupdata nd;
1445         int error;
1446
1447         KKASSERT(td->td_proc);
1448         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1449         if (error) {
1450                 nlookup_done(&nd);
1451                 return(error);
1452         }
1453         error = nlookup(&nd);
1454         if (error == 0)
1455                 error = kern_chroot(&nd.nl_nch);
1456         nlookup_done(&nd);
1457         return(error);
1458 }
1459
1460 /*
1461  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1462  * determine whether it is legal to chdir to the vnode.  The vnode's state
1463  * is not changed by this call.
1464  */
1465 int
1466 checkvp_chdir(struct vnode *vp, struct thread *td)
1467 {
1468         int error;
1469
1470         if (vp->v_type != VDIR)
1471                 error = ENOTDIR;
1472         else
1473                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1474         return (error);
1475 }
1476
1477 int
1478 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1479 {
1480         struct thread *td = curthread;
1481         struct proc *p = td->td_proc;
1482         struct lwp *lp = td->td_lwp;
1483         struct filedesc *fdp = p->p_fd;
1484         int cmode, flags;
1485         struct file *nfp;
1486         struct file *fp;
1487         struct vnode *vp;
1488         int type, indx, error;
1489         struct flock lf;
1490
1491         if ((oflags & O_ACCMODE) == O_ACCMODE)
1492                 return (EINVAL);
1493         flags = FFLAGS(oflags);
1494         error = falloc(p, &nfp, NULL);
1495         if (error)
1496                 return (error);
1497         fp = nfp;
1498         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1499
1500         /*
1501          * XXX p_dupfd is a real mess.  It allows a device to return a
1502          * file descriptor to be duplicated rather then doing the open
1503          * itself.
1504          */
1505         lp->lwp_dupfd = -1;
1506
1507         /*
1508          * Call vn_open() to do the lookup and assign the vnode to the
1509          * file pointer.  vn_open() does not change the ref count on fp
1510          * and the vnode, on success, will be inherited by the file pointer
1511          * and unlocked.
1512          */
1513         nd->nl_flags |= NLC_LOCKVP;
1514         error = vn_open(nd, fp, flags, cmode);
1515         nlookup_done(nd);
1516         if (error) {
1517                 /*
1518                  * handle special fdopen() case.  bleh.  dupfdopen() is
1519                  * responsible for dropping the old contents of ofiles[indx]
1520                  * if it succeeds.
1521                  *
1522                  * Note that fsetfd() will add a ref to fp which represents
1523                  * the fd_files[] assignment.  We must still drop our
1524                  * reference.
1525                  */
1526                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1527                         if (fdalloc(p, 0, &indx) == 0) {
1528                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1529                                 if (error == 0) {
1530                                         *res = indx;
1531                                         fdrop(fp);      /* our ref */
1532                                         return (0);
1533                                 }
1534                                 fsetfd(p, NULL, indx);
1535                         }
1536                 }
1537                 fdrop(fp);      /* our ref */
1538                 if (error == ERESTART)
1539                         error = EINTR;
1540                 return (error);
1541         }
1542
1543         /*
1544          * ref the vnode for ourselves so it can't be ripped out from under
1545          * is.  XXX need an ND flag to request that the vnode be returned
1546          * anyway.
1547          *
1548          * Reserve a file descriptor but do not assign it until the open
1549          * succeeds.
1550          */
1551         vp = (struct vnode *)fp->f_data;
1552         vref(vp);
1553         if ((error = fdalloc(p, 0, &indx)) != 0) {
1554                 fdrop(fp);
1555                 vrele(vp);
1556                 return (error);
1557         }
1558
1559         /*
1560          * If no error occurs the vp will have been assigned to the file
1561          * pointer.
1562          */
1563         lp->lwp_dupfd = 0;
1564
1565         if (flags & (O_EXLOCK | O_SHLOCK)) {
1566                 lf.l_whence = SEEK_SET;
1567                 lf.l_start = 0;
1568                 lf.l_len = 0;
1569                 if (flags & O_EXLOCK)
1570                         lf.l_type = F_WRLCK;
1571                 else
1572                         lf.l_type = F_RDLCK;
1573                 if (flags & FNONBLOCK)
1574                         type = 0;
1575                 else
1576                         type = F_WAIT;
1577
1578                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1579                         /*
1580                          * lock request failed.  Clean up the reserved
1581                          * descriptor.
1582                          */
1583                         vrele(vp);
1584                         fsetfd(p, NULL, indx);
1585                         fdrop(fp);
1586                         return (error);
1587                 }
1588                 fp->f_flag |= FHASLOCK;
1589         }
1590 #if 0
1591         /*
1592          * Assert that all regular file vnodes were created with a object.
1593          */
1594         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1595                 ("open: regular file has no backing object after vn_open"));
1596 #endif
1597
1598         vrele(vp);
1599
1600         /*
1601          * release our private reference, leaving the one associated with the
1602          * descriptor table intact.
1603          */
1604         fsetfd(p, fp, indx);
1605         fdrop(fp);
1606         *res = indx;
1607         return (0);
1608 }
1609
1610 /*
1611  * open_args(char *path, int flags, int mode)
1612  *
1613  * Check permissions, allocate an open file structure,
1614  * and call the device open routine if any.
1615  */
1616 int
1617 sys_open(struct open_args *uap)
1618 {
1619         struct nlookupdata nd;
1620         int error;
1621
1622         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1623         if (error == 0) {
1624                 error = kern_open(&nd, uap->flags,
1625                                     uap->mode, &uap->sysmsg_result);
1626         }
1627         nlookup_done(&nd);
1628         return (error);
1629 }
1630
1631 int
1632 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1633 {
1634         struct thread *td = curthread;
1635         struct proc *p = td->td_proc;
1636         struct vnode *vp;
1637         struct vattr vattr;
1638         int error;
1639         int whiteout = 0;
1640
1641         KKASSERT(p);
1642
1643         switch (mode & S_IFMT) {
1644         case S_IFCHR:
1645         case S_IFBLK:
1646                 error = suser(td);
1647                 break;
1648         default:
1649                 error = suser_cred(p->p_ucred, PRISON_ROOT);
1650                 break;
1651         }
1652         if (error)
1653                 return (error);
1654
1655         bwillwrite();
1656         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1657         if ((error = nlookup(nd)) != 0)
1658                 return (error);
1659         if (nd->nl_nch.ncp->nc_vp)
1660                 return (EEXIST);
1661         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1662                 return (error);
1663
1664         VATTR_NULL(&vattr);
1665         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1666         vattr.va_rmajor = rmajor;
1667         vattr.va_rminor = rminor;
1668         whiteout = 0;
1669
1670         switch (mode & S_IFMT) {
1671         case S_IFMT:    /* used by badsect to flag bad sectors */
1672                 vattr.va_type = VBAD;
1673                 break;
1674         case S_IFCHR:
1675                 vattr.va_type = VCHR;
1676                 break;
1677         case S_IFBLK:
1678                 vattr.va_type = VBLK;
1679                 break;
1680         case S_IFWHT:
1681                 whiteout = 1;
1682                 break;
1683         default:
1684                 error = EINVAL;
1685                 break;
1686         }
1687         if (error == 0) {
1688                 if (whiteout) {
1689                         error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1690                                               nd->nl_cred, NAMEI_CREATE);
1691                 } else {
1692                         vp = NULL;
1693                         error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1694                                            &vp, nd->nl_cred, &vattr);
1695                         if (error == 0)
1696                                 vput(vp);
1697                 }
1698         }
1699         return (error);
1700 }
1701
1702 /*
1703  * mknod_args(char *path, int mode, int dev)
1704  *
1705  * Create a special file.
1706  */
1707 int
1708 sys_mknod(struct mknod_args *uap)
1709 {
1710         struct nlookupdata nd;
1711         int error;
1712
1713         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1714         if (error == 0) {
1715                 error = kern_mknod(&nd, uap->mode,
1716                                    umajor(uap->dev), uminor(uap->dev));
1717         }
1718         nlookup_done(&nd);
1719         return (error);
1720 }
1721
1722 int
1723 kern_mkfifo(struct nlookupdata *nd, int mode)
1724 {
1725         struct thread *td = curthread;
1726         struct proc *p = td->td_proc;
1727         struct vattr vattr;
1728         struct vnode *vp;
1729         int error;
1730
1731         bwillwrite();
1732
1733         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1734         if ((error = nlookup(nd)) != 0)
1735                 return (error);
1736         if (nd->nl_nch.ncp->nc_vp)
1737                 return (EEXIST);
1738         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1739                 return (error);
1740
1741         VATTR_NULL(&vattr);
1742         vattr.va_type = VFIFO;
1743         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1744         vp = NULL;
1745         error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
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         int error;
1826
1827         /*
1828          * Lookup the source and obtained a locked vnode.
1829          *
1830          * XXX relookup on vget failure / race ?
1831          */
1832         bwillwrite();
1833         if ((error = nlookup(nd)) != 0)
1834                 return (error);
1835         vp = nd->nl_nch.ncp->nc_vp;
1836         KKASSERT(vp != NULL);
1837         if (vp->v_type == VDIR)
1838                 return (EPERM);         /* POSIX */
1839         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1840                 return (error);
1841         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1842                 return (error);
1843
1844         /*
1845          * Unlock the source so we can lookup the target without deadlocking
1846          * (XXX vp is locked already, possible other deadlock?).  The target
1847          * must not exist.
1848          */
1849         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1850         nd->nl_flags &= ~NLC_NCPISLOCKED;
1851         cache_unlock(&nd->nl_nch);
1852
1853         linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1854         if ((error = nlookup(linknd)) != 0) {
1855                 vput(vp);
1856                 return (error);
1857         }
1858         if (linknd->nl_nch.ncp->nc_vp) {
1859                 vput(vp);
1860                 return (EEXIST);
1861         }
1862
1863         /*
1864          * Finally run the new API VOP.
1865          */
1866         error = can_hardlink(vp, td, td->td_proc->p_ucred);
1867         if (error == 0) {
1868                 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
1869                                   vp, linknd->nl_cred);
1870         }
1871         vput(vp);
1872         return (error);
1873 }
1874
1875 /*
1876  * link_args(char *path, char *link)
1877  *
1878  * Make a hard file link.
1879  */
1880 int
1881 sys_link(struct link_args *uap)
1882 {
1883         struct nlookupdata nd, linknd;
1884         int error;
1885
1886         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1887         if (error == 0) {
1888                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1889                 if (error == 0)
1890                         error = kern_link(&nd, &linknd);
1891                 nlookup_done(&linknd);
1892         }
1893         nlookup_done(&nd);
1894         return (error);
1895 }
1896
1897 int
1898 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1899 {
1900         struct vattr vattr;
1901         struct vnode *vp;
1902         struct vnode *dvp;
1903         int error;
1904
1905         bwillwrite();
1906         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1907         if ((error = nlookup(nd)) != 0)
1908                 return (error);
1909         if (nd->nl_nch.ncp->nc_vp)
1910                 return (EEXIST);
1911         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1912                 return (error);
1913         dvp = nd->nl_dvp;
1914         VATTR_NULL(&vattr);
1915         vattr.va_mode = mode;
1916         error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
1917         if (error == 0)
1918                 vput(vp);
1919         return (error);
1920 }
1921
1922 /*
1923  * symlink(char *path, char *link)
1924  *
1925  * Make a symbolic link.
1926  */
1927 int
1928 sys_symlink(struct symlink_args *uap)
1929 {
1930         struct thread *td = curthread;
1931         struct nlookupdata nd;
1932         char *path;
1933         int error;
1934         int mode;
1935
1936         path = objcache_get(namei_oc, M_WAITOK);
1937         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1938         if (error == 0) {
1939                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1940                 if (error == 0) {
1941                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1942                         error = kern_symlink(&nd, path, mode);
1943                 }
1944                 nlookup_done(&nd);
1945         }
1946         objcache_put(namei_oc, path);
1947         return (error);
1948 }
1949
1950 /*
1951  * undelete_args(char *path)
1952  *
1953  * Delete a whiteout from the filesystem.
1954  */
1955 /* ARGSUSED */
1956 int
1957 sys_undelete(struct undelete_args *uap)
1958 {
1959         struct nlookupdata nd;
1960         int error;
1961
1962         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1963         bwillwrite();
1964         nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
1965         if (error == 0)
1966                 error = nlookup(&nd);
1967         if (error == 0)
1968                 error = ncp_writechk(&nd.nl_nch);
1969         if (error == 0) {
1970                 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
1971                                       NAMEI_DELETE);
1972         }
1973         nlookup_done(&nd);
1974         return (error);
1975 }
1976
1977 int
1978 kern_unlink(struct nlookupdata *nd)
1979 {
1980         int error;
1981
1982         bwillwrite();
1983         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
1984         if ((error = nlookup(nd)) != 0)
1985                 return (error);
1986         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1987                 return (error);
1988         error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
1989         return (error);
1990 }
1991
1992 /*
1993  * unlink_args(char *path)
1994  *
1995  * Delete a name from the filesystem.
1996  */
1997 int
1998 sys_unlink(struct unlink_args *uap)
1999 {
2000         struct nlookupdata nd;
2001         int error;
2002
2003         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2004         if (error == 0)
2005                 error = kern_unlink(&nd);
2006         nlookup_done(&nd);
2007         return (error);
2008 }
2009
2010 int
2011 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2012 {
2013         struct thread *td = curthread;
2014         struct proc *p = td->td_proc;
2015         struct file *fp;
2016         struct vattr vattr;
2017         int error;
2018
2019         fp = holdfp(p->p_fd, fd, -1);
2020         if (fp == NULL)
2021                 return (EBADF);
2022         if (fp->f_type != DTYPE_VNODE) {
2023                 error = ESPIPE;
2024                 goto done;
2025         }
2026
2027         switch (whence) {
2028         case L_INCR:
2029                 fp->f_offset += offset;
2030                 error = 0;
2031                 break;
2032         case L_XTND:
2033                 error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
2034                 if (error == 0)
2035                         fp->f_offset = offset + vattr.va_size;
2036                 break;
2037         case L_SET:
2038                 fp->f_offset = offset;
2039                 error = 0;
2040                 break;
2041         default:
2042                 error = EINVAL;
2043                 break;
2044         }
2045         *res = fp->f_offset;
2046 done:
2047         fdrop(fp);
2048         return (error);
2049 }
2050
2051 /*
2052  * lseek_args(int fd, int pad, off_t offset, int whence)
2053  *
2054  * Reposition read/write file offset.
2055  */
2056 int
2057 sys_lseek(struct lseek_args *uap)
2058 {
2059         int error;
2060
2061         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2062             &uap->sysmsg_offset);
2063
2064         return (error);
2065 }
2066
2067 int
2068 kern_access(struct nlookupdata *nd, int aflags)
2069 {
2070         struct vnode *vp;
2071         int error, flags;
2072
2073         if ((error = nlookup(nd)) != 0)
2074                 return (error);
2075 retry:
2076         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2077         if (error)
2078                 return (error);
2079
2080         /* Flags == 0 means only check for existence. */
2081         if (aflags) {
2082                 flags = 0;
2083                 if (aflags & R_OK)
2084                         flags |= VREAD;
2085                 if (aflags & W_OK)
2086                         flags |= VWRITE;
2087                 if (aflags & X_OK)
2088                         flags |= VEXEC;
2089                 if ((flags & VWRITE) == 0 ||
2090                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2091                         error = VOP_ACCESS(vp, flags, nd->nl_cred);
2092
2093                 /*
2094                  * If the file handle is stale we have to re-resolve the
2095                  * entry.  This is a hack at the moment.
2096                  */
2097                 if (error == ESTALE) {
2098                         vput(vp);
2099                         cache_setunresolved(&nd->nl_nch);
2100                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2101                         if (error == 0) {
2102                                 vp = NULL;
2103                                 goto retry;
2104                         }
2105                         return(error);
2106                 }
2107         }
2108         vput(vp);
2109         return (error);
2110 }
2111
2112 /*
2113  * access_args(char *path, int flags)
2114  *
2115  * Check access permissions.
2116  */
2117 int
2118 sys_access(struct access_args *uap)
2119 {
2120         struct nlookupdata nd;
2121         int error;
2122
2123         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2124         if (error == 0)
2125                 error = kern_access(&nd, uap->flags);
2126         nlookup_done(&nd);
2127         return (error);
2128 }
2129
2130 int
2131 kern_stat(struct nlookupdata *nd, struct stat *st)
2132 {
2133         int error;
2134         struct vnode *vp;
2135         thread_t td;
2136
2137         if ((error = nlookup(nd)) != 0)
2138                 return (error);
2139 again:
2140         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2141                 return (ENOENT);
2142
2143         td = curthread;
2144         if ((error = vget(vp, LK_SHARED)) != 0)
2145                 return (error);
2146         error = vn_stat(vp, st, nd->nl_cred);
2147
2148         /*
2149          * If the file handle is stale we have to re-resolve the entry.  This
2150          * is a hack at the moment.
2151          */
2152         if (error == ESTALE) {
2153                 vput(vp);
2154                 cache_setunresolved(&nd->nl_nch);
2155                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2156                 if (error == 0)
2157                         goto again;
2158         } else {
2159                 vput(vp);
2160         }
2161         return (error);
2162 }
2163
2164 /*
2165  * stat_args(char *path, struct stat *ub)
2166  *
2167  * Get file status; this version follows links.
2168  */
2169 int
2170 sys_stat(struct stat_args *uap)
2171 {
2172         struct nlookupdata nd;
2173         struct stat st;
2174         int error;
2175
2176         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2177         if (error == 0) {
2178                 error = kern_stat(&nd, &st);
2179                 if (error == 0)
2180                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2181         }
2182         nlookup_done(&nd);
2183         return (error);
2184 }
2185
2186 /*
2187  * lstat_args(char *path, struct stat *ub)
2188  *
2189  * Get file status; this version does not follow links.
2190  */
2191 int
2192 sys_lstat(struct lstat_args *uap)
2193 {
2194         struct nlookupdata nd;
2195         struct stat st;
2196         int error;
2197
2198         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2199         if (error == 0) {
2200                 error = kern_stat(&nd, &st);
2201                 if (error == 0)
2202                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2203         }
2204         nlookup_done(&nd);
2205         return (error);
2206 }
2207
2208 /*
2209  * pathconf_Args(char *path, int name)
2210  *
2211  * Get configurable pathname variables.
2212  */
2213 /* ARGSUSED */
2214 int
2215 sys_pathconf(struct pathconf_args *uap)
2216 {
2217         struct nlookupdata nd;
2218         struct vnode *vp;
2219         int error;
2220
2221         vp = NULL;
2222         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2223         if (error == 0)
2224                 error = nlookup(&nd);
2225         if (error == 0)
2226                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2227         nlookup_done(&nd);
2228         if (error == 0) {
2229                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2230                 vput(vp);
2231         }
2232         return (error);
2233 }
2234
2235 /*
2236  * XXX: daver
2237  * kern_readlink isn't properly split yet.  There is a copyin burried
2238  * in VOP_READLINK().
2239  */
2240 int
2241 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2242 {
2243         struct thread *td = curthread;
2244         struct proc *p = td->td_proc;
2245         struct vnode *vp;
2246         struct iovec aiov;
2247         struct uio auio;
2248         int error;
2249
2250         if ((error = nlookup(nd)) != 0)
2251                 return (error);
2252         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2253         if (error)
2254                 return (error);
2255         if (vp->v_type != VLNK) {
2256                 error = EINVAL;
2257         } else {
2258                 aiov.iov_base = buf;
2259                 aiov.iov_len = count;
2260                 auio.uio_iov = &aiov;
2261                 auio.uio_iovcnt = 1;
2262                 auio.uio_offset = 0;
2263                 auio.uio_rw = UIO_READ;
2264                 auio.uio_segflg = UIO_USERSPACE;
2265                 auio.uio_td = td;
2266                 auio.uio_resid = count;
2267                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2268         }
2269         vput(vp);
2270         *res = count - auio.uio_resid;
2271         return (error);
2272 }
2273
2274 /*
2275  * readlink_args(char *path, char *buf, int count)
2276  *
2277  * Return target name of a symbolic link.
2278  */
2279 int
2280 sys_readlink(struct readlink_args *uap)
2281 {
2282         struct nlookupdata nd;
2283         int error;
2284
2285         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2286         if (error == 0) {
2287                 error = kern_readlink(&nd, uap->buf, uap->count,
2288                                         &uap->sysmsg_result);
2289         }
2290         nlookup_done(&nd);
2291         return (error);
2292 }
2293
2294 static int
2295 setfflags(struct vnode *vp, int flags)
2296 {
2297         struct thread *td = curthread;
2298         struct proc *p = td->td_proc;
2299         int error;
2300         struct vattr vattr;
2301
2302         /*
2303          * Prevent non-root users from setting flags on devices.  When
2304          * a device is reused, users can retain ownership of the device
2305          * if they are allowed to set flags and programs assume that
2306          * chown can't fail when done as root.
2307          */
2308         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2309             ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2310                 return (error);
2311
2312         /*
2313          * note: vget is required for any operation that might mod the vnode
2314          * so VINACTIVE is properly cleared.
2315          */
2316         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2317                 VATTR_NULL(&vattr);
2318                 vattr.va_flags = flags;
2319                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2320                 vput(vp);
2321         }
2322         return (error);
2323 }
2324
2325 /*
2326  * chflags(char *path, int flags)
2327  *
2328  * Change flags of a file given a path name.
2329  */
2330 /* ARGSUSED */
2331 int
2332 sys_chflags(struct chflags_args *uap)
2333 {
2334         struct nlookupdata nd;
2335         struct vnode *vp;
2336         int error;
2337
2338         vp = NULL;
2339         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2340         /* XXX Add NLC flag indicating modifying operation? */
2341         if (error == 0)
2342                 error = nlookup(&nd);
2343         if (error == 0)
2344                 error = ncp_writechk(&nd.nl_nch);
2345         if (error == 0)
2346                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2347         nlookup_done(&nd);
2348         if (error == 0) {
2349                 error = setfflags(vp, uap->flags);
2350                 vrele(vp);
2351         }
2352         return (error);
2353 }
2354
2355 /*
2356  * fchflags_args(int fd, int flags)
2357  *
2358  * Change flags of a file given a file descriptor.
2359  */
2360 /* ARGSUSED */
2361 int
2362 sys_fchflags(struct fchflags_args *uap)
2363 {
2364         struct thread *td = curthread;
2365         struct proc *p = td->td_proc;
2366         struct file *fp;
2367         int error;
2368
2369         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2370                 return (error);
2371         if (fp->f_nchandle.ncp)
2372                 error = ncp_writechk(&fp->f_nchandle);
2373         if (error == 0)
2374                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2375         fdrop(fp);
2376         return (error);
2377 }
2378
2379 static int
2380 setfmode(struct vnode *vp, int mode)
2381 {
2382         struct thread *td = curthread;
2383         struct proc *p = td->td_proc;
2384         int error;
2385         struct vattr vattr;
2386
2387         /*
2388          * note: vget is required for any operation that might mod the vnode
2389          * so VINACTIVE is properly cleared.
2390          */
2391         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2392                 VATTR_NULL(&vattr);
2393                 vattr.va_mode = mode & ALLPERMS;
2394                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2395                 vput(vp);
2396         }
2397         return error;
2398 }
2399
2400 int
2401 kern_chmod(struct nlookupdata *nd, int mode)
2402 {
2403         struct vnode *vp;
2404         int error;
2405
2406         /* XXX Add NLC flag indicating modifying operation? */
2407         if ((error = nlookup(nd)) != 0)
2408                 return (error);
2409         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2410                 return (error);
2411         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2412                 error = setfmode(vp, mode);
2413         vrele(vp);
2414         return (error);
2415 }
2416
2417 /*
2418  * chmod_args(char *path, int mode)
2419  *
2420  * Change mode of a file given path name.
2421  */
2422 /* ARGSUSED */
2423 int
2424 sys_chmod(struct chmod_args *uap)
2425 {
2426         struct nlookupdata nd;
2427         int error;
2428
2429         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2430         if (error == 0)
2431                 error = kern_chmod(&nd, uap->mode);
2432         nlookup_done(&nd);
2433         return (error);
2434 }
2435
2436 /*
2437  * lchmod_args(char *path, int mode)
2438  *
2439  * Change mode of a file given path name (don't follow links.)
2440  */
2441 /* ARGSUSED */
2442 int
2443 sys_lchmod(struct lchmod_args *uap)
2444 {
2445         struct nlookupdata nd;
2446         int error;
2447
2448         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2449         if (error == 0)
2450                 error = kern_chmod(&nd, uap->mode);
2451         nlookup_done(&nd);
2452         return (error);
2453 }
2454
2455 /*
2456  * fchmod_args(int fd, int mode)
2457  *
2458  * Change mode of a file given a file descriptor.
2459  */
2460 /* ARGSUSED */
2461 int
2462 sys_fchmod(struct fchmod_args *uap)
2463 {
2464         struct thread *td = curthread;
2465         struct proc *p = td->td_proc;
2466         struct file *fp;
2467         int error;
2468
2469         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2470                 return (error);
2471         if (fp->f_nchandle.ncp)
2472                 error = ncp_writechk(&fp->f_nchandle);
2473         if (error == 0)
2474                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2475         fdrop(fp);
2476         return (error);
2477 }
2478
2479 static int
2480 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2481 {
2482         struct thread *td = curthread;
2483         struct proc *p = td->td_proc;
2484         int error;
2485         struct vattr vattr;
2486
2487         /*
2488          * note: vget is required for any operation that might mod the vnode
2489          * so VINACTIVE is properly cleared.
2490          */
2491         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2492                 VATTR_NULL(&vattr);
2493                 vattr.va_uid = uid;
2494                 vattr.va_gid = gid;
2495                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2496                 vput(vp);
2497         }
2498         return error;
2499 }
2500
2501 int
2502 kern_chown(struct nlookupdata *nd, int uid, int gid)
2503 {
2504         struct vnode *vp;
2505         int error;
2506
2507         /* XXX Add NLC flag indicating modifying operation? */
2508         if ((error = nlookup(nd)) != 0)
2509                 return (error);
2510         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2511                 return (error);
2512         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2513                 error = setfown(vp, uid, gid);
2514         vrele(vp);
2515         return (error);
2516 }
2517
2518 /*
2519  * chown(char *path, int uid, int gid)
2520  *
2521  * Set ownership given a path name.
2522  */
2523 int
2524 sys_chown(struct chown_args *uap)
2525 {
2526         struct nlookupdata nd;
2527         int error;
2528
2529         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2530         if (error == 0)
2531                 error = kern_chown(&nd, uap->uid, uap->gid);
2532         nlookup_done(&nd);
2533         return (error);
2534 }
2535
2536 /*
2537  * lchown_args(char *path, int uid, int gid)
2538  *
2539  * Set ownership given a path name, do not cross symlinks.
2540  */
2541 int
2542 sys_lchown(struct lchown_args *uap)
2543 {
2544         struct nlookupdata nd;
2545         int error;
2546
2547         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2548         if (error == 0)
2549                 error = kern_chown(&nd, uap->uid, uap->gid);
2550         nlookup_done(&nd);
2551         return (error);
2552 }
2553
2554 /*
2555  * fchown_args(int fd, int uid, int gid)
2556  *
2557  * Set ownership given a file descriptor.
2558  */
2559 /* ARGSUSED */
2560 int
2561 sys_fchown(struct fchown_args *uap)
2562 {
2563         struct thread *td = curthread;
2564         struct proc *p = td->td_proc;
2565         struct file *fp;
2566         int error;
2567
2568         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2569                 return (error);
2570         if (fp->f_nchandle.ncp)
2571                 error = ncp_writechk(&fp->f_nchandle);
2572         if (error == 0)
2573                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2574         fdrop(fp);
2575         return (error);
2576 }
2577
2578 static int
2579 getutimes(const struct timeval *tvp, struct timespec *tsp)
2580 {
2581         struct timeval tv[2];
2582
2583         if (tvp == NULL) {
2584                 microtime(&tv[0]);
2585                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2586                 tsp[1] = tsp[0];
2587         } else {
2588                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2589                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2590         }
2591         return 0;
2592 }
2593
2594 static int
2595 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2596 {
2597         struct thread *td = curthread;
2598         struct proc *p = td->td_proc;
2599         int error;
2600         struct vattr vattr;
2601
2602         /*
2603          * note: vget is required for any operation that might mod the vnode
2604          * so VINACTIVE is properly cleared.
2605          */
2606         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2607                 VATTR_NULL(&vattr);
2608                 vattr.va_atime = ts[0];
2609                 vattr.va_mtime = ts[1];
2610                 if (nullflag)
2611                         vattr.va_vaflags |= VA_UTIMES_NULL;
2612                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2613                 vput(vp);
2614         }
2615         return error;
2616 }
2617
2618 int
2619 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2620 {
2621         struct timespec ts[2];
2622         struct vnode *vp;
2623         int error;
2624
2625         if ((error = getutimes(tptr, ts)) != 0)
2626                 return (error);
2627         /* XXX Add NLC flag indicating modifying operation? */
2628         if ((error = nlookup(nd)) != 0)
2629                 return (error);
2630         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2631                 return (error);
2632         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2633                 return (error);
2634         error = setutimes(vp, ts, tptr == NULL);
2635         vrele(vp);
2636         return (error);
2637 }
2638
2639 /*
2640  * utimes_args(char *path, struct timeval *tptr)
2641  *
2642  * Set the access and modification times of a file.
2643  */
2644 int
2645 sys_utimes(struct utimes_args *uap)
2646 {
2647         struct timeval tv[2];
2648         struct nlookupdata nd;
2649         int error;
2650
2651         if (uap->tptr) {
2652                 error = copyin(uap->tptr, tv, sizeof(tv));
2653                 if (error)
2654                         return (error);
2655         }
2656         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2657         if (error == 0)
2658                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2659         nlookup_done(&nd);
2660         return (error);
2661 }
2662
2663 /*
2664  * lutimes_args(char *path, struct timeval *tptr)
2665  *
2666  * Set the access and modification times of a file.
2667  */
2668 int
2669 sys_lutimes(struct lutimes_args *uap)
2670 {
2671         struct timeval tv[2];
2672         struct nlookupdata nd;
2673         int error;
2674
2675         if (uap->tptr) {
2676                 error = copyin(uap->tptr, tv, sizeof(tv));
2677                 if (error)
2678                         return (error);
2679         }
2680         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2681         if (error == 0)
2682                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2683         nlookup_done(&nd);
2684         return (error);
2685 }
2686
2687 int
2688 kern_futimes(int fd, struct timeval *tptr)
2689 {
2690         struct thread *td = curthread;
2691         struct proc *p = td->td_proc;
2692         struct timespec ts[2];
2693         struct file *fp;
2694         int error;
2695
2696         error = getutimes(tptr, ts);
2697         if (error)
2698                 return (error);
2699         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2700                 return (error);
2701         if (fp->f_nchandle.ncp)
2702                 error = ncp_writechk(&fp->f_nchandle);
2703         if (error == 0)
2704                 error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2705         fdrop(fp);
2706         return (error);
2707 }
2708
2709 /*
2710  * futimes_args(int fd, struct timeval *tptr)
2711  *
2712  * Set the access and modification times of a file.
2713  */
2714 int
2715 sys_futimes(struct futimes_args *uap)
2716 {
2717         struct timeval tv[2];
2718         int error;
2719
2720         if (uap->tptr) {
2721                 error = copyin(uap->tptr, tv, sizeof(tv));
2722                 if (error)
2723                         return (error);
2724         }
2725
2726         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2727
2728         return (error);
2729 }
2730
2731 int
2732 kern_truncate(struct nlookupdata *nd, off_t length)
2733 {
2734         struct vnode *vp;
2735         struct vattr vattr;
2736         int error;
2737
2738         if (length < 0)
2739                 return(EINVAL);
2740         /* XXX Add NLC flag indicating modifying operation? */
2741         if ((error = nlookup(nd)) != 0)
2742                 return (error);
2743         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2744                 return (error);
2745         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2746                 return (error);
2747         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2748                 vrele(vp);
2749                 return (error);
2750         }
2751         if (vp->v_type == VDIR) {
2752                 error = EISDIR;
2753         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2754             (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2755                 VATTR_NULL(&vattr);
2756                 vattr.va_size = length;
2757                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2758         }
2759         vput(vp);
2760         return (error);
2761 }
2762
2763 /*
2764  * truncate(char *path, int pad, off_t length)
2765  *
2766  * Truncate a file given its path name.
2767  */
2768 int
2769 sys_truncate(struct truncate_args *uap)
2770 {
2771         struct nlookupdata nd;
2772         int error;
2773
2774         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2775         if (error == 0)
2776                 error = kern_truncate(&nd, uap->length);
2777         nlookup_done(&nd);
2778         return error;
2779 }
2780
2781 int
2782 kern_ftruncate(int fd, off_t length)
2783 {
2784         struct thread *td = curthread;
2785         struct proc *p = td->td_proc;
2786         struct vattr vattr;
2787         struct vnode *vp;
2788         struct file *fp;
2789         int error;
2790
2791         if (length < 0)
2792                 return(EINVAL);
2793         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2794                 return (error);
2795         if (fp->f_nchandle.ncp) {
2796                 error = ncp_writechk(&fp->f_nchandle);
2797                 if (error)
2798                         goto done;
2799         }
2800         if ((fp->f_flag & FWRITE) == 0) {
2801                 error = EINVAL;
2802                 goto done;
2803         }
2804         vp = (struct vnode *)fp->f_data;
2805         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2806         if (vp->v_type == VDIR) {
2807                 error = EISDIR;
2808         } else if ((error = vn_writechk(vp, NULL)) == 0) {
2809                 VATTR_NULL(&vattr);
2810                 vattr.va_size = length;
2811                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2812         }
2813         vn_unlock(vp);
2814 done:
2815         fdrop(fp);
2816         return (error);
2817 }
2818
2819 /*
2820  * ftruncate_args(int fd, int pad, off_t length)
2821  *
2822  * Truncate a file given a file descriptor.
2823  */
2824 int
2825 sys_ftruncate(struct ftruncate_args *uap)
2826 {
2827         int error;
2828
2829         error = kern_ftruncate(uap->fd, uap->length);
2830
2831         return (error);
2832 }
2833
2834 /*
2835  * fsync(int fd)
2836  *
2837  * Sync an open file.
2838  */
2839 /* ARGSUSED */
2840 int
2841 sys_fsync(struct fsync_args *uap)
2842 {
2843         struct thread *td = curthread;
2844         struct proc *p = td->td_proc;
2845         struct vnode *vp;
2846         struct file *fp;
2847         vm_object_t obj;
2848         int error;
2849
2850         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2851                 return (error);
2852         vp = (struct vnode *)fp->f_data;
2853         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2854         if ((obj = vp->v_object) != NULL)
2855                 vm_object_page_clean(obj, 0, 0, 0);
2856         if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
2857                 error = buf_fsync(vp);
2858         vn_unlock(vp);
2859         fdrop(fp);
2860         return (error);
2861 }
2862
2863 int
2864 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2865 {
2866         struct nchandle fnchd;
2867         struct nchandle tnchd;
2868         struct namecache *ncp;
2869         struct vnode *fdvp;
2870         struct vnode *tdvp;
2871         struct mount *mp;
2872         int error;
2873
2874         bwillwrite();
2875         fromnd->nl_flags |= NLC_REFDVP;
2876         if ((error = nlookup(fromnd)) != 0)
2877                 return (error);
2878         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2879                 return (ENOENT);
2880         fnchd.mount = fromnd->nl_nch.mount;
2881         cache_hold(&fnchd);
2882
2883         /*
2884          * unlock the source nch so we can lookup the target nch without
2885          * deadlocking.  The target may or may not exist so we do not check
2886          * for a target vp like kern_mkdir() and other creation functions do.
2887          *
2888          * The source and target directories are ref'd and rechecked after
2889          * everything is relocked to determine if the source or target file
2890          * has been renamed.
2891          */
2892         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2893         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2894         cache_unlock(&fromnd->nl_nch);
2895
2896         tond->nl_flags |= NLC_CREATE | NLC_REFDVP;
2897         if ((error = nlookup(tond)) != 0) {
2898                 cache_drop(&fnchd);
2899                 return (error);
2900         }
2901         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2902                 cache_drop(&fnchd);
2903                 return (ENOENT);
2904         }
2905         tnchd.mount = tond->nl_nch.mount;
2906         cache_hold(&tnchd);
2907
2908         /*
2909          * If the source and target are the same there is nothing to do
2910          */
2911         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2912                 cache_drop(&fnchd);
2913                 cache_drop(&tnchd);
2914                 return (0);
2915         }
2916
2917         /*
2918          * Mount points cannot be renamed or overwritten
2919          */
2920         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2921             NCF_ISMOUNTPT
2922         ) {
2923                 cache_drop(&fnchd);
2924                 cache_drop(&tnchd);
2925                 return (EINVAL);
2926         }
2927
2928         /*
2929          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2930          * may have become invalid while it was unlocked, nc_vp and nc_mount
2931          * could be NULL.
2932          */
2933         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2934                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2935         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2936                 cache_lock(&fromnd->nl_nch);
2937                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2938         } else {
2939                 cache_unlock(&tond->nl_nch);
2940                 cache_lock(&fromnd->nl_nch);
2941                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2942                 cache_lock(&tond->nl_nch);
2943                 cache_resolve(&tond->nl_nch, tond->nl_cred);
2944         }
2945         fromnd->nl_flags |= NLC_NCPISLOCKED;
2946
2947         /*
2948          * make sure the parent directories linkages are the same
2949          */
2950         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2951             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2952                 cache_drop(&fnchd);
2953                 cache_drop(&tnchd);
2954                 return (ENOENT);
2955         }
2956
2957         /*
2958          * Both the source and target must be within the same filesystem and
2959          * in the same filesystem as their parent directories within the
2960          * namecache topology.
2961          *
2962          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2963          */
2964         mp = fnchd.mount;
2965         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2966             mp != tond->nl_nch.mount) {
2967                 cache_drop(&fnchd);
2968                 cache_drop(&tnchd);
2969                 return (EXDEV);
2970         }
2971
2972         /*
2973          * Make sure the mount point is writable
2974          */
2975         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2976                 cache_drop(&fnchd);
2977                 cache_drop(&tnchd);
2978                 return (error);
2979         }
2980
2981         /*
2982          * If the target exists and either the source or target is a directory,
2983          * then both must be directories.
2984          *
2985          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
2986          * have become NULL.
2987          */
2988         if (tond->nl_nch.ncp->nc_vp) {
2989                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
2990                         error = ENOENT;
2991                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
2992                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
2993                                 error = ENOTDIR;
2994                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
2995                         error = EISDIR;
2996                 }
2997         }
2998
2999         /*
3000          * You cannot rename a source into itself or a subdirectory of itself.
3001          * We check this by travsersing the target directory upwards looking
3002          * for a match against the source.
3003          */
3004         if (error == 0) {
3005                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3006                         if (fromnd->nl_nch.ncp == ncp) {
3007                                 error = EINVAL;
3008                                 break;
3009                         }
3010                 }
3011         }
3012
3013         cache_drop(&fnchd);
3014         cache_drop(&tnchd);
3015
3016         /*
3017          * Even though the namespaces are different, they may still represent
3018          * hardlinks to the same file.  The filesystem might have a hard time
3019          * with this so we issue a NREMOVE of the source instead of a NRENAME
3020          * when we detect the situation.
3021          */
3022         if (error == 0) {
3023                 fdvp = fromnd->nl_dvp;
3024                 tdvp = tond->nl_dvp;
3025                 if (fdvp == NULL || tdvp == NULL) {
3026                         error = EPERM;
3027                 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3028                         error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3029                                             fromnd->nl_cred);
3030                 } else {
3031                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3032                                             fdvp, tdvp, tond->nl_cred);
3033                 }
3034         }
3035         return (error);
3036 }
3037
3038 /*
3039  * rename_args(char *from, char *to)
3040  *
3041  * Rename files.  Source and destination must either both be directories,
3042  * or both not be directories.  If target is a directory, it must be empty.
3043  */
3044 int
3045 sys_rename(struct rename_args *uap)
3046 {
3047         struct nlookupdata fromnd, tond;
3048         int error;
3049
3050         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3051         if (error == 0) {
3052                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3053                 if (error == 0)
3054                         error = kern_rename(&fromnd, &tond);
3055                 nlookup_done(&tond);
3056         }
3057         nlookup_done(&fromnd);
3058         return (error);
3059 }
3060
3061 int
3062 kern_mkdir(struct nlookupdata *nd, int mode)
3063 {
3064         struct thread *td = curthread;
3065         struct proc *p = td->td_proc;
3066         struct vnode *vp;
3067         struct vattr vattr;
3068         int error;
3069
3070         bwillwrite();
3071         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3072         if ((error = nlookup(nd)) != 0)
3073                 return (error);
3074
3075         if (nd->nl_nch.ncp->nc_vp)
3076                 return (EEXIST);
3077         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3078                 return (error);
3079         VATTR_NULL(&vattr);
3080         vattr.va_type = VDIR;
3081         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3082
3083         vp = NULL;
3084         error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3085         if (error == 0)
3086                 vput(vp);
3087         return (error);
3088 }
3089
3090 /*
3091  * mkdir_args(char *path, int mode)
3092  *
3093  * Make a directory file.
3094  */
3095 /* ARGSUSED */
3096 int
3097 sys_mkdir(struct mkdir_args *uap)
3098 {
3099         struct nlookupdata nd;
3100         int error;
3101
3102         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3103         if (error == 0)
3104                 error = kern_mkdir(&nd, uap->mode);
3105         nlookup_done(&nd);
3106         return (error);
3107 }
3108
3109 int
3110 kern_rmdir(struct nlookupdata *nd)
3111 {
3112         int error;
3113
3114         bwillwrite();
3115         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3116         if ((error = nlookup(nd)) != 0)
3117                 return (error);
3118
3119         /*
3120          * Do not allow directories representing mount points to be
3121          * deleted, even if empty.  Check write perms on mount point
3122          * in case the vnode is aliased (aka nullfs).
3123          */
3124         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3125                 return (EINVAL);
3126         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3127                 return (error);
3128         error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3129         return (error);
3130 }
3131
3132 /*
3133  * rmdir_args(char *path)
3134  *
3135  * Remove a directory file.
3136  */
3137 /* ARGSUSED */
3138 int
3139 sys_rmdir(struct rmdir_args *uap)
3140 {
3141         struct nlookupdata nd;
3142         int error;
3143
3144         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3145         if (error == 0)
3146                 error = kern_rmdir(&nd);
3147         nlookup_done(&nd);
3148         return (error);
3149 }
3150
3151 int
3152 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3153     enum uio_seg direction)
3154 {
3155         struct thread *td = curthread;
3156         struct proc *p = td->td_proc;
3157         struct vnode *vp;
3158         struct file *fp;
3159         struct uio auio;
3160         struct iovec aiov;
3161         off_t loff;
3162         int error, eofflag;
3163
3164         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3165                 return (error);
3166         if ((fp->f_flag & FREAD) == 0) {
3167                 error = EBADF;
3168                 goto done;
3169         }
3170         vp = (struct vnode *)fp->f_data;
3171 unionread:
3172         if (vp->v_type != VDIR) {
3173                 error = EINVAL;
3174                 goto done;
3175         }
3176         aiov.iov_base = buf;
3177         aiov.iov_len = count;
3178         auio.uio_iov = &aiov;
3179         auio.uio_iovcnt = 1;
3180         auio.uio_rw = UIO_READ;
3181         auio.uio_segflg = direction;
3182         auio.uio_td = td;
3183         auio.uio_resid = count;
3184         loff = auio.uio_offset = fp->f_offset;
3185         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3186         fp->f_offset = auio.uio_offset;
3187         if (error)
3188                 goto done;
3189         if (count == auio.uio_resid) {
3190                 if (union_dircheckp) {
3191                         error = union_dircheckp(td, &vp, fp);
3192                         if (error == -1)
3193                                 goto unionread;
3194                         if (error)
3195                                 goto done;
3196                 }
3197 #if 0
3198                 if ((vp->v_flag & VROOT) &&
3199                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3200                         struct vnode *tvp = vp;
3201                         vp = vp->v_mount->mnt_vnodecovered;
3202                         vref(vp);
3203                         fp->f_data = vp;
3204                         fp->f_offset = 0;
3205                         vrele(tvp);
3206                         goto unionread;
3207                 }
3208 #endif
3209         }
3210
3211         /*
3212          * WARNING!  *basep may not be wide enough to accomodate the
3213          * seek offset.   XXX should we hack this to return the upper 32 bits
3214          * for offsets greater then 4G?
3215          */
3216         if (basep) {
3217                 *basep = (long)loff;
3218         }
3219         *res = count - auio.uio_resid;
3220 done:
3221         fdrop(fp);
3222         return (error);
3223 }
3224
3225 /*
3226  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3227  *
3228  * Read a block of directory entries in a file system independent format.
3229  */
3230 int
3231 sys_getdirentries(struct getdirentries_args *uap)
3232 {
3233         long base;
3234         int error;
3235
3236         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3237                                    &uap->sysmsg_result, UIO_USERSPACE);
3238
3239         if (error == 0 && uap->basep)
3240                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3241         return (error);
3242 }
3243
3244 /*
3245  * getdents_args(int fd, char *buf, size_t count)
3246  */
3247 int
3248 sys_getdents(struct getdents_args *uap)
3249 {
3250         int error;
3251
3252         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3253             &uap->sysmsg_result, UIO_USERSPACE);
3254
3255         return (error);
3256 }
3257
3258 /*
3259  * umask(int newmask)
3260  *
3261  * Set the mode mask for creation of filesystem nodes.
3262  *
3263  * MP SAFE
3264  */
3265 int
3266 sys_umask(struct umask_args *uap)
3267 {
3268         struct thread *td = curthread;
3269         struct proc *p = td->td_proc;
3270         struct filedesc *fdp;
3271
3272         fdp = p->p_fd;
3273         uap->sysmsg_result = fdp->fd_cmask;
3274         fdp->fd_cmask = uap->newmask & ALLPERMS;
3275         return (0);
3276 }
3277
3278 /*
3279  * revoke(char *path)
3280  *
3281  * Void all references to file by ripping underlying filesystem
3282  * away from vnode.
3283  */
3284 /* ARGSUSED */
3285 int
3286 sys_revoke(struct revoke_args *uap)
3287 {
3288         struct nlookupdata nd;
3289         struct vattr vattr;
3290         struct vnode *vp;
3291         struct ucred *cred;
3292         int error;
3293
3294         vp = NULL;
3295         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3296         if (error == 0)
3297                 error = nlookup(&nd);
3298         if (error == 0)
3299                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3300         cred = crhold(nd.nl_cred);
3301         nlookup_done(&nd);
3302         if (error == 0) {
3303                 if (vp->v_type != VCHR && vp->v_type != VBLK)
3304                         error = EINVAL;
3305                 if (error == 0)
3306                         error = VOP_GETATTR(vp, &vattr);
3307                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3308                         error = suser_cred(cred, PRISON_ROOT);
3309                 if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3310                         error = 0;
3311                         vx_lock(vp);
3312                         VOP_REVOKE(vp, REVOKEALL);
3313                         vx_unlock(vp);
3314                 }
3315                 vrele(vp);
3316         }
3317         if (cred)
3318                 crfree(cred);
3319         return (error);
3320 }
3321
3322 /*
3323  * getfh_args(char *fname, fhandle_t *fhp)
3324  *
3325  * Get (NFS) file handle
3326  */
3327 int
3328 sys_getfh(struct getfh_args *uap)
3329 {
3330         struct thread *td = curthread;
3331         struct nlookupdata nd;
3332         fhandle_t fh;
3333         struct vnode *vp;
3334         int error;
3335
3336         /*
3337          * Must be super user
3338          */
3339         if ((error = suser(td)) != 0)
3340                 return (error);
3341
3342         vp = NULL;
3343         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3344         if (error == 0)
3345                 error = nlookup(&nd);
3346         if (error == 0)
3347                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3348         nlookup_done(&nd);
3349         if (error == 0) {
3350                 bzero(&fh, sizeof(fh));
3351                 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3352                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3353                 vput(vp);
3354                 if (error == 0)
3355                         error = copyout(&fh, uap->fhp, sizeof(fh));
3356         }
3357         return (error);
3358 }
3359
3360 /*
3361  * fhopen_args(const struct fhandle *u_fhp, int flags)
3362  *
3363  * syscall for the rpc.lockd to use to translate a NFS file handle into
3364  * an open descriptor.
3365  *
3366  * warning: do not remove the suser() call or this becomes one giant
3367  * security hole.
3368  */
3369 int
3370 sys_fhopen(struct fhopen_args *uap)
3371 {
3372         struct thread *td = curthread;
3373         struct proc *p = td->td_proc;
3374         struct mount *mp;
3375         struct vnode *vp;
3376         struct fhandle fhp;
3377         struct vattr vat;
3378         struct vattr *vap = &vat;
3379         struct flock lf;
3380         int fmode, mode, error, type;
3381         struct file *nfp;
3382         struct file *fp;
3383         int indx;
3384
3385         /*
3386          * Must be super user
3387          */
3388         error = suser(td);
3389         if (error)
3390                 return (error);
3391
3392         fmode = FFLAGS(uap->flags);
3393         /* why not allow a non-read/write open for our lockd? */
3394         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3395                 return (EINVAL);
3396         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3397         if (error)
3398                 return(error);
3399         /* find the mount point */
3400         mp = vfs_getvfs(&fhp.fh_fsid);
3401         if (mp == NULL)
3402                 return (ESTALE);
3403         /* now give me my vnode, it gets returned to me locked */
3404         error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3405         if (error)
3406                 return (error);
3407         /*
3408          * from now on we have to make sure not
3409          * to forget about the vnode
3410          * any error that causes an abort must vput(vp)
3411          * just set error = err and 'goto bad;'.
3412          */
3413
3414         /*
3415          * from vn_open
3416          */
3417         if (vp->v_type == VLNK) {
3418                 error = EMLINK;
3419                 goto bad;
3420         }
3421         if (vp->v_type == VSOCK) {
3422                 error = EOPNOTSUPP;
3423                 goto bad;
3424         }
3425         mode = 0;
3426         if (fmode & (FWRITE | O_TRUNC)) {
3427                 if (vp->v_type == VDIR) {
3428                         error = EISDIR;
3429                         goto bad;
3430                 }
3431                 error = vn_writechk(vp, NULL);
3432                 if (error)
3433                         goto bad;
3434                 mode |= VWRITE;
3435         }
3436         if (fmode & FREAD)
3437                 mode |= VREAD;
3438         if (mode) {
3439                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3440                 if (error)
3441                         goto bad;
3442         }
3443         if (fmode & O_TRUNC) {
3444                 vn_unlock(vp);                          /* XXX */
3445                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3446                 VATTR_NULL(vap);
3447                 vap->va_size = 0;
3448                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3449                 if (error)
3450                         goto bad;
3451         }
3452
3453         /*
3454          * VOP_OPEN needs the file pointer so it can potentially override
3455          * it.
3456          *
3457          * WARNING! no f_nchandle will be associated when fhopen()ing a
3458          * directory.  XXX
3459          */
3460         if ((error = falloc(p, &nfp, &indx)) != 0)
3461                 goto bad;
3462         fp = nfp;
3463
3464         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3465         if (error) {
3466                 /*
3467                  * setting f_ops this way prevents VOP_CLOSE from being
3468                  * called or fdrop() releasing the vp from v_data.   Since
3469                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3470                  */
3471                 fp->f_ops = &badfileops;
3472                 fp->f_data = NULL;
3473                 goto bad_drop;
3474         }
3475
3476         /*
3477          * The fp is given its own reference, we still have our ref and lock.
3478          *
3479          * Assert that all regular files must be created with a VM object.
3480          */
3481         if (vp->v_type == VREG && vp->v_object == NULL) {
3482                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3483                 goto bad_drop;
3484         }
3485
3486         /*
3487          * The open was successful.  Handle any locking requirements.
3488          */
3489         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3490                 lf.l_whence = SEEK_SET;
3491                 lf.l_start = 0;
3492                 lf.l_len = 0;
3493                 if (fmode & O_EXLOCK)
3494                         lf.l_type = F_WRLCK;
3495                 else
3496                         lf.l_type = F_RDLCK;
3497                 if (fmode & FNONBLOCK)
3498                         type = 0;
3499                 else
3500                         type = F_WAIT;
3501                 vn_unlock(vp);
3502                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3503                         /*
3504                          * release our private reference.
3505                          */
3506                         fsetfd(p, NULL, indx);
3507                         fdrop(fp);
3508                         vrele(vp);
3509                         return (error);
3510                 }
3511                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3512                 fp->f_flag |= FHASLOCK;
3513         }
3514
3515         /*
3516          * Clean up.  Associate the file pointer with the previously
3517          * reserved descriptor and return it.
3518          */
3519         vput(vp);
3520         fsetfd(p, fp, indx);
3521         fdrop(fp);
3522         uap->sysmsg_result = indx;
3523         return (0);
3524
3525 bad_drop:
3526         fsetfd(p, NULL, indx);
3527         fdrop(fp);
3528 bad:
3529         vput(vp);
3530         return (error);
3531 }
3532
3533 /*
3534  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3535  */
3536 int
3537 sys_fhstat(struct fhstat_args *uap)
3538 {
3539         struct thread *td = curthread;
3540         struct stat sb;
3541         fhandle_t fh;
3542         struct mount *mp;
3543         struct vnode *vp;
3544         int error;
3545
3546         /*
3547          * Must be super user
3548          */
3549         error = suser(td);
3550         if (error)
3551                 return (error);
3552
3553         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3554         if (error)
3555                 return (error);
3556
3557         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3558                 return (ESTALE);
3559         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3560                 return (error);
3561         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3562         vput(vp);
3563         if (error)
3564                 return (error);
3565         error = copyout(&sb, uap->sb, sizeof(sb));
3566         return (error);
3567 }
3568
3569 /*
3570  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3571  */
3572 int
3573 sys_fhstatfs(struct fhstatfs_args *uap)
3574 {
3575         struct thread *td = curthread;
3576         struct proc *p = td->td_proc;
3577         struct statfs *sp;
3578         struct mount *mp;
3579         struct vnode *vp;
3580         struct statfs sb;
3581         char *fullpath, *freepath;
3582         fhandle_t fh;
3583         int error;
3584
3585         /*
3586          * Must be super user
3587          */
3588         if ((error = suser(td)))
3589                 return (error);
3590
3591         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3592                 return (error);
3593
3594         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3595                 return (ESTALE);
3596
3597         if (p != NULL && !chroot_visible_mnt(mp, p))
3598                 return (ESTALE);
3599
3600         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3601                 return (error);
3602         mp = vp->v_mount;
3603         sp = &mp->mnt_stat;
3604         vput(vp);
3605         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3606                 return (error);
3607
3608         error = mount_path(p, mp, &fullpath, &freepath);
3609         if (error)
3610                 return(error);
3611         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3612         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3613         kfree(freepath, M_TEMP);
3614
3615         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3616         if (suser(td)) {
3617                 bcopy(sp, &sb, sizeof(sb));
3618                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3619                 sp = &sb;
3620         }
3621         return (copyout(sp, uap->buf, sizeof(*sp)));
3622 }
3623
3624 /*
3625  * Syscall to push extended attribute configuration information into the
3626  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3627  * a command (int cmd), and attribute name and misc data.  For now, the
3628  * attribute name is left in userspace for consumption by the VFS_op.
3629  * It will probably be changed to be copied into sysspace by the
3630  * syscall in the future, once issues with various consumers of the
3631  * attribute code have raised their hands.
3632  *
3633  * Currently this is used only by UFS Extended Attributes.
3634  */
3635 int
3636 sys_extattrctl(struct extattrctl_args *uap)
3637 {
3638         struct nlookupdata nd;
3639         struct mount *mp;
3640         struct vnode *vp;
3641         int error;
3642
3643         vp = NULL;
3644         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3645         if (error == 0)
3646                 error = nlookup(&nd);
3647         if (error == 0) {
3648                 mp = nd.nl_nch.mount;
3649                 error = VFS_EXTATTRCTL(mp, uap->cmd,
3650                                 uap->attrname, uap->arg,
3651                                 nd.nl_cred);
3652         }
3653         nlookup_done(&nd);
3654         return (error);
3655 }
3656
3657 /*
3658  * Syscall to set a named extended attribute on a file or directory.
3659  * Accepts attribute name, and a uio structure pointing to the data to set.
3660  * The uio is consumed in the style of writev().  The real work happens
3661  * in VOP_SETEXTATTR().
3662  */
3663 int
3664 sys_extattr_set_file(struct extattr_set_file_args *uap)
3665 {
3666         char attrname[EXTATTR_MAXNAMELEN];
3667         struct iovec aiov[UIO_SMALLIOV];
3668         struct iovec *needfree;
3669         struct nlookupdata nd;
3670         struct iovec *iov;
3671         struct vnode *vp;
3672         struct uio auio;
3673         u_int iovlen;
3674         u_int cnt;
3675         int error;
3676         int i;
3677
3678         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3679         if (error)
3680                 return (error);
3681
3682         vp = NULL;
3683         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3684         if (error == 0)
3685                 error = nlookup(&nd);
3686         if (error == 0)
3687                 error = ncp_writechk(&nd.nl_nch);
3688         if (error == 0)
3689                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3690         if (error) {
3691                 nlookup_done(&nd);
3692                 return (error);
3693         }
3694
3695         needfree = NULL;
3696         iovlen = uap->iovcnt * sizeof(struct iovec);
3697         if (uap->iovcnt > UIO_SMALLIOV) {
3698                 if (uap->iovcnt > UIO_MAXIOV) {
3699                         error = EINVAL;
3700                         goto done;
3701                 }
3702                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3703                 needfree = iov;
3704         } else {
3705                 iov = aiov;
3706         }
3707         auio.uio_iov = iov;
3708         auio.uio_iovcnt = uap->iovcnt;
3709         auio.uio_rw = UIO_WRITE;
3710         auio.uio_segflg = UIO_USERSPACE;
3711         auio.uio_td = nd.nl_td;
3712         auio.uio_offset = 0;
3713         if ((error = copyin(uap->iovp, iov, iovlen)))
3714                 goto done;
3715         auio.uio_resid = 0;
3716         for (i = 0; i < uap->iovcnt; i++) {
3717                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3718                         error = EINVAL;
3719                         goto done;
3720                 }
3721                 auio.uio_resid += iov->iov_len;
3722                 iov++;
3723         }
3724         cnt = auio.uio_resid;
3725         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3726         cnt -= auio.uio_resid;
3727         uap->sysmsg_result = cnt;
3728 done:
3729         vput(vp);
3730         nlookup_done(&nd);
3731         if (needfree)
3732                 FREE(needfree, M_IOV);
3733         return (error);
3734 }
3735
3736 /*
3737  * Syscall to get a named extended attribute on a file or directory.
3738  * Accepts attribute name, and a uio structure pointing to a buffer for the
3739  * data.  The uio is consumed in the style of readv().  The real work
3740  * happens in VOP_GETEXTATTR();
3741  */
3742 int
3743 sys_extattr_get_file(struct extattr_get_file_args *uap)
3744 {
3745         char attrname[EXTATTR_MAXNAMELEN];
3746         struct iovec aiov[UIO_SMALLIOV];
3747         struct iovec *needfree;
3748         struct nlookupdata nd;
3749         struct iovec *iov;
3750         struct vnode *vp;
3751         struct uio auio;
3752         u_int iovlen;
3753         u_int cnt;
3754         int error;
3755         int i;
3756
3757         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3758         if (error)
3759                 return (error);
3760
3761         vp = NULL;
3762         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3763         if (error == 0)
3764                 error = nlookup(&nd);
3765         if (error == 0)
3766                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3767         if (error) {
3768                 nlookup_done(&nd);
3769                 return (error);
3770         }
3771
3772         iovlen = uap->iovcnt * sizeof (struct iovec);
3773         needfree = NULL;
3774         if (uap->iovcnt > UIO_SMALLIOV) {
3775                 if (uap->iovcnt > UIO_MAXIOV) {
3776                         error = EINVAL;
3777                         goto done;
3778                 }
3779                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3780                 needfree = iov;
3781         } else {
3782                 iov = aiov;
3783         }
3784         auio.uio_iov = iov;
3785         auio.uio_iovcnt = uap->iovcnt;
3786         auio.uio_rw = UIO_READ;
3787         auio.uio_segflg = UIO_USERSPACE;
3788         auio.uio_td = nd.nl_td;
3789         auio.uio_offset = 0;
3790         if ((error = copyin(uap->iovp, iov, iovlen)))
3791                 goto done;
3792         auio.uio_resid = 0;
3793         for (i = 0; i < uap->iovcnt; i++) {
3794                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3795                         error = EINVAL;
3796                         goto done;
3797                 }
3798                 auio.uio_resid += iov->iov_len;
3799                 iov++;
3800         }
3801         cnt = auio.uio_resid;
3802         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3803         cnt -= auio.uio_resid;
3804         uap->sysmsg_result = cnt;
3805 done:
3806         vput(vp);
3807         nlookup_done(&nd);
3808         if (needfree)
3809                 FREE(needfree, M_IOV);
3810         return(error);
3811 }
3812
3813 /*
3814  * Syscall to delete a named extended attribute from a file or directory.
3815  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3816  */
3817 int
3818 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3819 {
3820         char attrname[EXTATTR_MAXNAMELEN];
3821         struct nlookupdata nd;
3822         struct vnode *vp;
3823         int error;
3824
3825         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3826         if (error)
3827                 return(error);
3828
3829         vp = NULL;
3830         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3831         if (error == 0)
3832                 error = nlookup(&nd);
3833         if (error == 0)
3834                 error = ncp_writechk(&nd.nl_nch);
3835         if (error == 0)
3836                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3837         if (error) {
3838                 nlookup_done(&nd);
3839                 return (error);
3840         }
3841
3842         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3843         vput(vp);
3844         nlookup_done(&nd);
3845         return(error);
3846 }
3847
3848 /*
3849  * Determine if the mount is visible to the process.
3850  */
3851 static int
3852 chroot_visible_mnt(struct mount *mp, struct proc *p)
3853 {
3854         struct nchandle nch;
3855
3856         /*
3857          * Traverse from the mount point upwards.  If we hit the process
3858          * root then the mount point is visible to the process.
3859          */
3860         nch = mp->mnt_ncmountpt;
3861         while (nch.ncp) {
3862                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
3863                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
3864                         return(1);
3865                 }
3866                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3867                         nch = nch.mount->mnt_ncmounton;
3868                 } else {
3869                         nch.ncp = nch.ncp->nc_parent;
3870                 }
3871         }
3872
3873         /*
3874          * If the mount point is not visible to the process, but the
3875          * process root is in a subdirectory of the mount, return
3876          * TRUE anyway.
3877          */
3878         if (p->p_fd->fd_nrdir.mount == mp)
3879                 return(1);
3880
3881         return(0);
3882 }
3883