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 1.133 2008/06/28 17:59:49 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 int
1132 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1133 {
1134         struct mount *mp;
1135         struct statvfs *sp;
1136         int error;
1137
1138         if ((error = nlookup(nd)) != 0)
1139                 return (error);
1140         mp = nd->nl_nch.mount;
1141         sp = &mp->mnt_vstat;
1142         if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1143                 return (error);
1144
1145         sp->f_flag = 0;
1146         if (mp->mnt_flag & MNT_RDONLY)
1147                 sp->f_flag |= ST_RDONLY;
1148         if (mp->mnt_flag & MNT_NOSUID)
1149                 sp->f_flag |= ST_NOSUID;
1150         bcopy(sp, buf, sizeof(*buf));
1151         return (0);
1152 }
1153
1154 /*
1155  * statfs_args(char *path, struct statfs *buf)
1156  *
1157  * Get filesystem statistics.
1158  */
1159 int
1160 sys_statvfs(struct statvfs_args *uap)
1161 {
1162         struct nlookupdata nd;
1163         struct statvfs buf;
1164         int error;
1165
1166         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1167         if (error == 0)
1168                 error = kern_statvfs(&nd, &buf);
1169         nlookup_done(&nd);
1170         if (error == 0)
1171                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1172         return (error);
1173 }
1174
1175 int
1176 kern_fstatvfs(int fd, struct statvfs *buf)
1177 {
1178         struct thread *td = curthread;
1179         struct proc *p = td->td_proc;
1180         struct file *fp;
1181         struct mount *mp;
1182         struct statvfs *sp;
1183         int error;
1184
1185         KKASSERT(p);
1186         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1187                 return (error);
1188         mp = ((struct vnode *)fp->f_data)->v_mount;
1189         if (mp == NULL) {
1190                 error = EBADF;
1191                 goto done;
1192         }
1193         if (fp->f_cred == NULL) {
1194                 error = EINVAL;
1195                 goto done;
1196         }
1197         sp = &mp->mnt_vstat;
1198         if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1199                 goto done;
1200
1201         sp->f_flag = 0;
1202         if (mp->mnt_flag & MNT_RDONLY)
1203                 sp->f_flag |= ST_RDONLY;
1204         if (mp->mnt_flag & MNT_NOSUID)
1205                 sp->f_flag |= ST_NOSUID;
1206
1207         bcopy(sp, buf, sizeof(*buf));
1208         error = 0;
1209 done:
1210         fdrop(fp);
1211         return (error);
1212 }
1213
1214 /*
1215  * fstatfs_args(int fd, struct statfs *buf)
1216  *
1217  * Get filesystem statistics.
1218  */
1219 int
1220 sys_fstatvfs(struct fstatvfs_args *uap)
1221 {
1222         struct statvfs buf;
1223         int error;
1224
1225         error = kern_fstatvfs(uap->fd, &buf);
1226
1227         if (error == 0)
1228                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1229         return (error);
1230 }
1231
1232 /*
1233  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1234  *
1235  * Get statistics on all filesystems.
1236  */
1237
1238 struct getfsstat_info {
1239         struct statfs *sfsp;
1240         long count;
1241         long maxcount;
1242         int error;
1243         int flags;
1244         struct proc *p;
1245 };
1246
1247 static int getfsstat_callback(struct mount *, void *);
1248
1249 /* ARGSUSED */
1250 int
1251 sys_getfsstat(struct getfsstat_args *uap)
1252 {
1253         struct thread *td = curthread;
1254         struct proc *p = td->td_proc;
1255         struct getfsstat_info info;
1256
1257         bzero(&info, sizeof(info));
1258
1259         info.maxcount = uap->bufsize / sizeof(struct statfs);
1260         info.sfsp = uap->buf;
1261         info.count = 0;
1262         info.flags = uap->flags;
1263         info.p = p;
1264
1265         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1266         if (info.sfsp && info.count > info.maxcount)
1267                 uap->sysmsg_result = info.maxcount;
1268         else
1269                 uap->sysmsg_result = info.count;
1270         return (info.error);
1271 }
1272
1273 static int
1274 getfsstat_callback(struct mount *mp, void *data)
1275 {
1276         struct getfsstat_info *info = data;
1277         struct statfs *sp;
1278         char *freepath;
1279         char *fullpath;
1280         int error;
1281
1282         if (info->sfsp && info->count < info->maxcount) {
1283                 if (info->p && !chroot_visible_mnt(mp, info->p))
1284                         return(0);
1285                 sp = &mp->mnt_stat;
1286
1287                 /*
1288                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1289                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1290                  * overrides MNT_WAIT.
1291                  */
1292                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1293                     (info->flags & MNT_WAIT)) &&
1294                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1295                         return(0);
1296                 }
1297                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1298
1299                 error = mount_path(info->p, mp, &fullpath, &freepath);
1300                 if (error) {
1301                         info->error = error;
1302                         return(-1);
1303                 }
1304                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1305                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1306                 kfree(freepath, M_TEMP);
1307
1308                 error = copyout(sp, info->sfsp, sizeof(*sp));
1309                 if (error) {
1310                         info->error = error;
1311                         return (-1);
1312                 }
1313                 ++info->sfsp;
1314         }
1315         info->count++;
1316         return(0);
1317 }
1318
1319 /*
1320  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1321                    long bufsize, int flags)
1322  *
1323  * Get statistics on all filesystems.
1324  */
1325
1326 struct getvfsstat_info {
1327         struct statfs *sfsp;
1328         struct statvfs *vsfsp;
1329         long count;
1330         long maxcount;
1331         int error;
1332         int flags;
1333         struct proc *p;
1334 };
1335
1336 static int getvfsstat_callback(struct mount *, void *);
1337
1338 /* ARGSUSED */
1339 int
1340 sys_getvfsstat(struct getvfsstat_args *uap)
1341 {
1342         struct thread *td = curthread;
1343         struct proc *p = td->td_proc;
1344         struct getvfsstat_info info;
1345
1346         bzero(&info, sizeof(info));
1347
1348         info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1349         info.sfsp = uap->buf;
1350         info.vsfsp = uap->vbuf;
1351         info.count = 0;
1352         info.flags = uap->flags;
1353         info.p = p;
1354
1355         mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1356         if (info.vsfsp && info.count > info.maxcount)
1357                 uap->sysmsg_result = info.maxcount;
1358         else
1359                 uap->sysmsg_result = info.count;
1360         return (info.error);
1361 }
1362
1363 static int
1364 getvfsstat_callback(struct mount *mp, void *data)
1365 {
1366         struct getvfsstat_info *info = data;
1367         struct statfs *sp;
1368         struct statvfs *vsp;
1369         char *freepath;
1370         char *fullpath;
1371         int error;
1372
1373         if (info->vsfsp && info->count < info->maxcount) {
1374                 if (info->p && !chroot_visible_mnt(mp, info->p))
1375                         return(0);
1376                 sp = &mp->mnt_stat;
1377                 vsp = &mp->mnt_vstat;
1378
1379                 /*
1380                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1381                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1382                  * overrides MNT_WAIT.
1383                  */
1384                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1385                     (info->flags & MNT_WAIT)) &&
1386                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1387                         return(0);
1388                 }
1389                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1390
1391                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1392                     (info->flags & MNT_WAIT)) &&
1393                     (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1394                         return(0);
1395                 }
1396                 vsp->f_flag = 0;
1397                 if (mp->mnt_flag & MNT_RDONLY)
1398                         vsp->f_flag |= ST_RDONLY;
1399                 if (mp->mnt_flag & MNT_NOSUID)
1400                         vsp->f_flag |= ST_NOSUID;
1401
1402                 error = mount_path(info->p, mp, &fullpath, &freepath);
1403                 if (error) {
1404                         info->error = error;
1405                         return(-1);
1406                 }
1407                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1408                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1409                 kfree(freepath, M_TEMP);
1410
1411                 error = copyout(sp, info->sfsp, sizeof(*sp));
1412                 if (error == 0)
1413                         error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1414                 if (error) {
1415                         info->error = error;
1416                         return (-1);
1417                 }
1418                 ++info->sfsp;
1419                 ++info->vsfsp;
1420         }
1421         info->count++;
1422         return(0);
1423 }
1424
1425
1426 /*
1427  * fchdir_args(int fd)
1428  *
1429  * Change current working directory to a given file descriptor.
1430  */
1431 /* ARGSUSED */
1432 int
1433 sys_fchdir(struct fchdir_args *uap)
1434 {
1435         struct thread *td = curthread;
1436         struct proc *p = td->td_proc;
1437         struct filedesc *fdp = p->p_fd;
1438         struct vnode *vp, *ovp;
1439         struct mount *mp;
1440         struct file *fp;
1441         struct nchandle nch, onch, tnch;
1442         int error;
1443
1444         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1445                 return (error);
1446         vp = (struct vnode *)fp->f_data;
1447         vref(vp);
1448         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1449         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1450                 error = ENOTDIR;
1451         else
1452                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1453         if (error) {
1454                 vput(vp);
1455                 fdrop(fp);
1456                 return (error);
1457         }
1458         cache_copy(&fp->f_nchandle, &nch);
1459
1460         /*
1461          * If the ncp has become a mount point, traverse through
1462          * the mount point.
1463          */
1464
1465         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1466                (mp = cache_findmount(&nch)) != NULL
1467         ) {
1468                 error = nlookup_mp(mp, &tnch);
1469                 if (error == 0) {
1470                         cache_unlock(&tnch);    /* leave ref intact */
1471                         vput(vp);
1472                         vp = tnch.ncp->nc_vp;
1473                         error = vget(vp, LK_SHARED);
1474                         KKASSERT(error == 0);
1475                         cache_drop(&nch);
1476                         nch = tnch;
1477                 }
1478         }
1479         if (error == 0) {
1480                 ovp = fdp->fd_cdir;
1481                 onch = fdp->fd_ncdir;
1482                 vn_unlock(vp);          /* leave ref intact */
1483                 fdp->fd_cdir = vp;
1484                 fdp->fd_ncdir = nch;
1485                 cache_drop(&onch);
1486                 vrele(ovp);
1487         } else {
1488                 cache_drop(&nch);
1489                 vput(vp);
1490         }
1491         fdrop(fp);
1492         return (error);
1493 }
1494
1495 int
1496 kern_chdir(struct nlookupdata *nd)
1497 {
1498         struct thread *td = curthread;
1499         struct proc *p = td->td_proc;
1500         struct filedesc *fdp = p->p_fd;
1501         struct vnode *vp, *ovp;
1502         struct nchandle onch;
1503         int error;
1504
1505         if ((error = nlookup(nd)) != 0)
1506                 return (error);
1507         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1508                 return (ENOENT);
1509         if ((error = vget(vp, LK_SHARED)) != 0)
1510                 return (error);
1511
1512         error = checkvp_chdir(vp, td);
1513         vn_unlock(vp);
1514         if (error == 0) {
1515                 ovp = fdp->fd_cdir;
1516                 onch = fdp->fd_ncdir;
1517                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1518                 fdp->fd_ncdir = nd->nl_nch;
1519                 fdp->fd_cdir = vp;
1520                 cache_drop(&onch);
1521                 vrele(ovp);
1522                 cache_zero(&nd->nl_nch);
1523         } else {
1524                 vrele(vp);
1525         }
1526         return (error);
1527 }
1528
1529 /*
1530  * chdir_args(char *path)
1531  *
1532  * Change current working directory (``.'').
1533  */
1534 int
1535 sys_chdir(struct chdir_args *uap)
1536 {
1537         struct nlookupdata nd;
1538         int error;
1539
1540         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1541         if (error == 0)
1542                 error = kern_chdir(&nd);
1543         nlookup_done(&nd);
1544         return (error);
1545 }
1546
1547 /*
1548  * Helper function for raised chroot(2) security function:  Refuse if
1549  * any filedescriptors are open directories.
1550  */
1551 static int
1552 chroot_refuse_vdir_fds(struct filedesc *fdp)
1553 {
1554         struct vnode *vp;
1555         struct file *fp;
1556         int error;
1557         int fd;
1558
1559         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1560                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1561                         continue;
1562                 vp = (struct vnode *)fp->f_data;
1563                 if (vp->v_type != VDIR) {
1564                         fdrop(fp);
1565                         continue;
1566                 }
1567                 fdrop(fp);
1568                 return(EPERM);
1569         }
1570         return (0);
1571 }
1572
1573 /*
1574  * This sysctl determines if we will allow a process to chroot(2) if it
1575  * has a directory open:
1576  *      0: disallowed for all processes.
1577  *      1: allowed for processes that were not already chroot(2)'ed.
1578  *      2: allowed for all processes.
1579  */
1580
1581 static int chroot_allow_open_directories = 1;
1582
1583 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1584      &chroot_allow_open_directories, 0, "");
1585
1586 /*
1587  * chroot to the specified namecache entry.  We obtain the vp from the
1588  * namecache data.  The passed ncp must be locked and referenced and will
1589  * remain locked and referenced on return.
1590  */
1591 int
1592 kern_chroot(struct nchandle *nch)
1593 {
1594         struct thread *td = curthread;
1595         struct proc *p = td->td_proc;
1596         struct filedesc *fdp = p->p_fd;
1597         struct vnode *vp;
1598         int error;
1599
1600         /*
1601          * Only root can chroot
1602          */
1603         if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1604                 return (error);
1605
1606         /*
1607          * Disallow open directory descriptors (fchdir() breakouts).
1608          */
1609         if (chroot_allow_open_directories == 0 ||
1610            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1611                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1612                         return (error);
1613         }
1614         if ((vp = nch->ncp->nc_vp) == NULL)
1615                 return (ENOENT);
1616
1617         if ((error = vget(vp, LK_SHARED)) != 0)
1618                 return (error);
1619
1620         /*
1621          * Check the validity of vp as a directory to change to and
1622          * associate it with rdir/jdir.
1623          */
1624         error = checkvp_chdir(vp, td);
1625         vn_unlock(vp);                  /* leave reference intact */
1626         if (error == 0) {
1627                 vrele(fdp->fd_rdir);
1628                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1629                 cache_drop(&fdp->fd_nrdir);
1630                 cache_copy(nch, &fdp->fd_nrdir);
1631                 if (fdp->fd_jdir == NULL) {
1632                         fdp->fd_jdir = vp;
1633                         vref(fdp->fd_jdir);
1634                         cache_copy(nch, &fdp->fd_njdir);
1635                 }
1636         } else {
1637                 vrele(vp);
1638         }
1639         return (error);
1640 }
1641
1642 /*
1643  * chroot_args(char *path)
1644  *
1645  * Change notion of root (``/'') directory.
1646  */
1647 /* ARGSUSED */
1648 int
1649 sys_chroot(struct chroot_args *uap)
1650 {
1651         struct thread *td = curthread;
1652         struct nlookupdata nd;
1653         int error;
1654
1655         KKASSERT(td->td_proc);
1656         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1657         if (error) {
1658                 nlookup_done(&nd);
1659                 return(error);
1660         }
1661         error = nlookup(&nd);
1662         if (error == 0)
1663                 error = kern_chroot(&nd.nl_nch);
1664         nlookup_done(&nd);
1665         return(error);
1666 }
1667
1668 /*
1669  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1670  * determine whether it is legal to chdir to the vnode.  The vnode's state
1671  * is not changed by this call.
1672  */
1673 int
1674 checkvp_chdir(struct vnode *vp, struct thread *td)
1675 {
1676         int error;
1677
1678         if (vp->v_type != VDIR)
1679                 error = ENOTDIR;
1680         else
1681                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1682         return (error);
1683 }
1684
1685 int
1686 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1687 {
1688         struct thread *td = curthread;
1689         struct proc *p = td->td_proc;
1690         struct lwp *lp = td->td_lwp;
1691         struct filedesc *fdp = p->p_fd;
1692         int cmode, flags;
1693         struct file *nfp;
1694         struct file *fp;
1695         struct vnode *vp;
1696         int type, indx, error;
1697         struct flock lf;
1698
1699         if ((oflags & O_ACCMODE) == O_ACCMODE)
1700                 return (EINVAL);
1701         flags = FFLAGS(oflags);
1702         error = falloc(p, &nfp, NULL);
1703         if (error)
1704                 return (error);
1705         fp = nfp;
1706         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1707
1708         /*
1709          * XXX p_dupfd is a real mess.  It allows a device to return a
1710          * file descriptor to be duplicated rather then doing the open
1711          * itself.
1712          */
1713         lp->lwp_dupfd = -1;
1714
1715         /*
1716          * Call vn_open() to do the lookup and assign the vnode to the
1717          * file pointer.  vn_open() does not change the ref count on fp
1718          * and the vnode, on success, will be inherited by the file pointer
1719          * and unlocked.
1720          */
1721         nd->nl_flags |= NLC_LOCKVP;
1722         error = vn_open(nd, fp, flags, cmode);
1723         nlookup_done(nd);
1724         if (error) {
1725                 /*
1726                  * handle special fdopen() case.  bleh.  dupfdopen() is
1727                  * responsible for dropping the old contents of ofiles[indx]
1728                  * if it succeeds.
1729                  *
1730                  * Note that fsetfd() will add a ref to fp which represents
1731                  * the fd_files[] assignment.  We must still drop our
1732                  * reference.
1733                  */
1734                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1735                         if (fdalloc(p, 0, &indx) == 0) {
1736                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1737                                 if (error == 0) {
1738                                         *res = indx;
1739                                         fdrop(fp);      /* our ref */
1740                                         return (0);
1741                                 }
1742                                 fsetfd(p, NULL, indx);
1743                         }
1744                 }
1745                 fdrop(fp);      /* our ref */
1746                 if (error == ERESTART)
1747                         error = EINTR;
1748                 return (error);
1749         }
1750
1751         /*
1752          * ref the vnode for ourselves so it can't be ripped out from under
1753          * is.  XXX need an ND flag to request that the vnode be returned
1754          * anyway.
1755          *
1756          * Reserve a file descriptor but do not assign it until the open
1757          * succeeds.
1758          */
1759         vp = (struct vnode *)fp->f_data;
1760         vref(vp);
1761         if ((error = fdalloc(p, 0, &indx)) != 0) {
1762                 fdrop(fp);
1763                 vrele(vp);
1764                 return (error);
1765         }
1766
1767         /*
1768          * If no error occurs the vp will have been assigned to the file
1769          * pointer.
1770          */
1771         lp->lwp_dupfd = 0;
1772
1773         if (flags & (O_EXLOCK | O_SHLOCK)) {
1774                 lf.l_whence = SEEK_SET;
1775                 lf.l_start = 0;
1776                 lf.l_len = 0;
1777                 if (flags & O_EXLOCK)
1778                         lf.l_type = F_WRLCK;
1779                 else
1780                         lf.l_type = F_RDLCK;
1781                 if (flags & FNONBLOCK)
1782                         type = 0;
1783                 else
1784                         type = F_WAIT;
1785
1786                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1787                         /*
1788                          * lock request failed.  Clean up the reserved
1789                          * descriptor.
1790                          */
1791                         vrele(vp);
1792                         fsetfd(p, NULL, indx);
1793                         fdrop(fp);
1794                         return (error);
1795                 }
1796                 fp->f_flag |= FHASLOCK;
1797         }
1798 #if 0
1799         /*
1800          * Assert that all regular file vnodes were created with a object.
1801          */
1802         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1803                 ("open: regular file has no backing object after vn_open"));
1804 #endif
1805
1806         vrele(vp);
1807
1808         /*
1809          * release our private reference, leaving the one associated with the
1810          * descriptor table intact.
1811          */
1812         fsetfd(p, fp, indx);
1813         fdrop(fp);
1814         *res = indx;
1815         return (0);
1816 }
1817
1818 /*
1819  * open_args(char *path, int flags, int mode)
1820  *
1821  * Check permissions, allocate an open file structure,
1822  * and call the device open routine if any.
1823  */
1824 int
1825 sys_open(struct open_args *uap)
1826 {
1827         struct nlookupdata nd;
1828         int error;
1829
1830         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1831         if (error == 0) {
1832                 error = kern_open(&nd, uap->flags,
1833                                     uap->mode, &uap->sysmsg_result);
1834         }
1835         nlookup_done(&nd);
1836         return (error);
1837 }
1838
1839 int
1840 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1841 {
1842         struct thread *td = curthread;
1843         struct proc *p = td->td_proc;
1844         struct vnode *vp;
1845         struct vattr vattr;
1846         int error;
1847         int whiteout = 0;
1848
1849         KKASSERT(p);
1850
1851         switch (mode & S_IFMT) {
1852         case S_IFCHR:
1853         case S_IFBLK:
1854                 error = suser(td);
1855                 break;
1856         default:
1857                 error = suser_cred(p->p_ucred, PRISON_ROOT);
1858                 break;
1859         }
1860         if (error)
1861                 return (error);
1862
1863         bwillinode(1);
1864         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1865         if ((error = nlookup(nd)) != 0)
1866                 return (error);
1867         if (nd->nl_nch.ncp->nc_vp)
1868                 return (EEXIST);
1869         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1870                 return (error);
1871
1872         VATTR_NULL(&vattr);
1873         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1874         vattr.va_rmajor = rmajor;
1875         vattr.va_rminor = rminor;
1876         whiteout = 0;
1877
1878         switch (mode & S_IFMT) {
1879         case S_IFMT:    /* used by badsect to flag bad sectors */
1880                 vattr.va_type = VBAD;
1881                 break;
1882         case S_IFCHR:
1883                 vattr.va_type = VCHR;
1884                 break;
1885         case S_IFBLK:
1886                 vattr.va_type = VBLK;
1887                 break;
1888         case S_IFWHT:
1889                 whiteout = 1;
1890                 break;
1891         case S_IFDIR:
1892                 /* special directories support for HAMMER */
1893                 vattr.va_type = VDIR;
1894                 break;
1895         default:
1896                 error = EINVAL;
1897                 break;
1898         }
1899         if (error == 0) {
1900                 if (whiteout) {
1901                         error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1902                                               nd->nl_cred, NAMEI_CREATE);
1903                 } else {
1904                         vp = NULL;
1905                         error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1906                                            &vp, nd->nl_cred, &vattr);
1907                         if (error == 0)
1908                                 vput(vp);
1909                 }
1910         }
1911         return (error);
1912 }
1913
1914 /*
1915  * mknod_args(char *path, int mode, int dev)
1916  *
1917  * Create a special file.
1918  */
1919 int
1920 sys_mknod(struct mknod_args *uap)
1921 {
1922         struct nlookupdata nd;
1923         int error;
1924
1925         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1926         if (error == 0) {
1927                 error = kern_mknod(&nd, uap->mode,
1928                                    umajor(uap->dev), uminor(uap->dev));
1929         }
1930         nlookup_done(&nd);
1931         return (error);
1932 }
1933
1934 int
1935 kern_mkfifo(struct nlookupdata *nd, int mode)
1936 {
1937         struct thread *td = curthread;
1938         struct proc *p = td->td_proc;
1939         struct vattr vattr;
1940         struct vnode *vp;
1941         int error;
1942
1943         bwillinode(1);
1944
1945         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1946         if ((error = nlookup(nd)) != 0)
1947                 return (error);
1948         if (nd->nl_nch.ncp->nc_vp)
1949                 return (EEXIST);
1950         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1951                 return (error);
1952
1953         VATTR_NULL(&vattr);
1954         vattr.va_type = VFIFO;
1955         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1956         vp = NULL;
1957         error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1958         if (error == 0)
1959                 vput(vp);
1960         return (error);
1961 }
1962
1963 /*
1964  * mkfifo_args(char *path, int mode)
1965  *
1966  * Create a named pipe.
1967  */
1968 int
1969 sys_mkfifo(struct mkfifo_args *uap)
1970 {
1971         struct nlookupdata nd;
1972         int error;
1973
1974         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1975         if (error == 0)
1976                 error = kern_mkfifo(&nd, uap->mode);
1977         nlookup_done(&nd);
1978         return (error);
1979 }
1980
1981 static int hardlink_check_uid = 0;
1982 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1983     &hardlink_check_uid, 0,
1984     "Unprivileged processes cannot create hard links to files owned by other "
1985     "users");
1986 static int hardlink_check_gid = 0;
1987 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1988     &hardlink_check_gid, 0,
1989     "Unprivileged processes cannot create hard links to files owned by other "
1990     "groups");
1991
1992 static int
1993 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1994 {
1995         struct vattr va;
1996         int error;
1997
1998         /*
1999          * Shortcut if disabled
2000          */
2001         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2002                 return (0);
2003
2004         /*
2005          * root cred can always hardlink
2006          */
2007         if (suser_cred(cred, PRISON_ROOT) == 0)
2008                 return (0);
2009
2010         /*
2011          * Otherwise only if the originating file is owned by the
2012          * same user or group.  Note that any group is allowed if
2013          * the file is owned by the caller.
2014          */
2015         error = VOP_GETATTR(vp, &va);
2016         if (error != 0)
2017                 return (error);
2018
2019         if (hardlink_check_uid) {
2020                 if (cred->cr_uid != va.va_uid)
2021                         return (EPERM);
2022         }
2023
2024         if (hardlink_check_gid) {
2025                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2026                         return (EPERM);
2027         }
2028
2029         return (0);
2030 }
2031
2032 int
2033 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2034 {
2035         struct thread *td = curthread;
2036         struct vnode *vp;
2037         int error;
2038
2039         /*
2040          * Lookup the source and obtained a locked vnode.
2041          *
2042          * XXX relookup on vget failure / race ?
2043          */
2044         bwillinode(1);
2045         if ((error = nlookup(nd)) != 0)
2046                 return (error);
2047         vp = nd->nl_nch.ncp->nc_vp;
2048         KKASSERT(vp != NULL);
2049         if (vp->v_type == VDIR)
2050                 return (EPERM);         /* POSIX */
2051         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2052                 return (error);
2053         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2054                 return (error);
2055
2056         /*
2057          * Unlock the source so we can lookup the target without deadlocking
2058          * (XXX vp is locked already, possible other deadlock?).  The target
2059          * must not exist.
2060          */
2061         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2062         nd->nl_flags &= ~NLC_NCPISLOCKED;
2063         cache_unlock(&nd->nl_nch);
2064
2065         linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2066         if ((error = nlookup(linknd)) != 0) {
2067                 vput(vp);
2068                 return (error);
2069         }
2070         if (linknd->nl_nch.ncp->nc_vp) {
2071                 vput(vp);
2072                 return (EEXIST);
2073         }
2074
2075         /*
2076          * Finally run the new API VOP.
2077          */
2078         error = can_hardlink(vp, td, td->td_proc->p_ucred);
2079         if (error == 0) {
2080                 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2081                                   vp, linknd->nl_cred);
2082         }
2083         vput(vp);
2084         return (error);
2085 }
2086
2087 /*
2088  * link_args(char *path, char *link)
2089  *
2090  * Make a hard file link.
2091  */
2092 int
2093 sys_link(struct link_args *uap)
2094 {
2095         struct nlookupdata nd, linknd;
2096         int error;
2097
2098         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2099         if (error == 0) {
2100                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2101                 if (error == 0)
2102                         error = kern_link(&nd, &linknd);
2103                 nlookup_done(&linknd);
2104         }
2105         nlookup_done(&nd);
2106         return (error);
2107 }
2108
2109 int
2110 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2111 {
2112         struct vattr vattr;
2113         struct vnode *vp;
2114         struct vnode *dvp;
2115         int error;
2116
2117         bwillinode(1);
2118         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2119         if ((error = nlookup(nd)) != 0)
2120                 return (error);
2121         if (nd->nl_nch.ncp->nc_vp)
2122                 return (EEXIST);
2123         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2124                 return (error);
2125         dvp = nd->nl_dvp;
2126         VATTR_NULL(&vattr);
2127         vattr.va_mode = mode;
2128         error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2129         if (error == 0)
2130                 vput(vp);
2131         return (error);
2132 }
2133
2134 /*
2135  * symlink(char *path, char *link)
2136  *
2137  * Make a symbolic link.
2138  */
2139 int
2140 sys_symlink(struct symlink_args *uap)
2141 {
2142         struct thread *td = curthread;
2143         struct nlookupdata nd;
2144         char *path;
2145         int error;
2146         int mode;
2147
2148         path = objcache_get(namei_oc, M_WAITOK);
2149         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2150         if (error == 0) {
2151                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2152                 if (error == 0) {
2153                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2154                         error = kern_symlink(&nd, path, mode);
2155                 }
2156                 nlookup_done(&nd);
2157         }
2158         objcache_put(namei_oc, path);
2159         return (error);
2160 }
2161
2162 /*
2163  * undelete_args(char *path)
2164  *
2165  * Delete a whiteout from the filesystem.
2166  */
2167 /* ARGSUSED */
2168 int
2169 sys_undelete(struct undelete_args *uap)
2170 {
2171         struct nlookupdata nd;
2172         int error;
2173
2174         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2175         bwillinode(1);
2176         nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2177         if (error == 0)
2178                 error = nlookup(&nd);
2179         if (error == 0)
2180                 error = ncp_writechk(&nd.nl_nch);
2181         if (error == 0) {
2182                 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2183                                       NAMEI_DELETE);
2184         }
2185         nlookup_done(&nd);
2186         return (error);
2187 }
2188
2189 int
2190 kern_unlink(struct nlookupdata *nd)
2191 {
2192         int error;
2193
2194         bwillinode(1);
2195         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2196         if ((error = nlookup(nd)) != 0)
2197                 return (error);
2198         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2199                 return (error);
2200         error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2201         return (error);
2202 }
2203
2204 /*
2205  * unlink_args(char *path)
2206  *
2207  * Delete a name from the filesystem.
2208  */
2209 int
2210 sys_unlink(struct unlink_args *uap)
2211 {
2212         struct nlookupdata nd;
2213         int error;
2214
2215         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2216         if (error == 0)
2217                 error = kern_unlink(&nd);
2218         nlookup_done(&nd);
2219         return (error);
2220 }
2221
2222 int
2223 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2224 {
2225         struct thread *td = curthread;
2226         struct proc *p = td->td_proc;
2227         struct file *fp;
2228         struct vnode *vp;
2229         struct vattr vattr;
2230         off_t new_offset;
2231         int error;
2232
2233         fp = holdfp(p->p_fd, fd, -1);
2234         if (fp == NULL)
2235                 return (EBADF);
2236         if (fp->f_type != DTYPE_VNODE) {
2237                 error = ESPIPE;
2238                 goto done;
2239         }
2240         vp = (struct vnode *)fp->f_data;
2241
2242         switch (whence) {
2243         case L_INCR:
2244                 new_offset = fp->f_offset + offset;
2245                 error = 0;
2246                 break;
2247         case L_XTND:
2248                 error = VOP_GETATTR(vp, &vattr);
2249                 new_offset = offset + vattr.va_size;
2250                 break;
2251         case L_SET:
2252                 new_offset = offset;
2253                 error = 0;
2254                 break;
2255         default:
2256                 new_offset = 0;
2257                 error = EINVAL;
2258                 break;
2259         }
2260
2261         /*
2262          * Validate the seek position.  Negative offsets are not allowed
2263          * for regular files, block specials, or directories.
2264          */
2265         if (error == 0) {
2266                 if (new_offset < 0 &&
2267                     (vp->v_type == VREG || vp->v_type == VDIR ||
2268                      vp->v_type == VCHR || vp->v_type == VBLK)) {
2269                         error = EINVAL;
2270                 } else {
2271                         fp->f_offset = new_offset;
2272                 }
2273         }
2274         *res = fp->f_offset;
2275 done:
2276         fdrop(fp);
2277         return (error);
2278 }
2279
2280 /*
2281  * lseek_args(int fd, int pad, off_t offset, int whence)
2282  *
2283  * Reposition read/write file offset.
2284  */
2285 int
2286 sys_lseek(struct lseek_args *uap)
2287 {
2288         int error;
2289
2290         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2291                            &uap->sysmsg_offset);
2292
2293         return (error);
2294 }
2295
2296 int
2297 kern_access(struct nlookupdata *nd, int aflags)
2298 {
2299         struct vnode *vp;
2300         int error, flags;
2301
2302         if ((error = nlookup(nd)) != 0)
2303                 return (error);
2304 retry:
2305         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2306         if (error)
2307                 return (error);
2308
2309         /* Flags == 0 means only check for existence. */
2310         if (aflags) {
2311                 flags = 0;
2312                 if (aflags & R_OK)
2313                         flags |= VREAD;
2314                 if (aflags & W_OK)
2315                         flags |= VWRITE;
2316                 if (aflags & X_OK)
2317                         flags |= VEXEC;
2318                 if ((flags & VWRITE) == 0 ||
2319                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2320                         error = VOP_ACCESS(vp, flags, nd->nl_cred);
2321
2322                 /*
2323                  * If the file handle is stale we have to re-resolve the
2324                  * entry.  This is a hack at the moment.
2325                  */
2326                 if (error == ESTALE) {
2327                         vput(vp);
2328                         cache_setunresolved(&nd->nl_nch);
2329                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2330                         if (error == 0) {
2331                                 vp = NULL;
2332                                 goto retry;
2333                         }
2334                         return(error);
2335                 }
2336         }
2337         vput(vp);
2338         return (error);
2339 }
2340
2341 /*
2342  * access_args(char *path, int flags)
2343  *
2344  * Check access permissions.
2345  */
2346 int
2347 sys_access(struct access_args *uap)
2348 {
2349         struct nlookupdata nd;
2350         int error;
2351
2352         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2353         if (error == 0)
2354                 error = kern_access(&nd, uap->flags);
2355         nlookup_done(&nd);
2356         return (error);
2357 }
2358
2359 int
2360 kern_stat(struct nlookupdata *nd, struct stat *st)
2361 {
2362         int error;
2363         struct vnode *vp;
2364         thread_t td;
2365
2366         if ((error = nlookup(nd)) != 0)
2367                 return (error);
2368 again:
2369         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2370                 return (ENOENT);
2371
2372         td = curthread;
2373         if ((error = vget(vp, LK_SHARED)) != 0)
2374                 return (error);
2375         error = vn_stat(vp, st, nd->nl_cred);
2376
2377         /*
2378          * If the file handle is stale we have to re-resolve the entry.  This
2379          * is a hack at the moment.
2380          */
2381         if (error == ESTALE) {
2382                 vput(vp);
2383                 cache_setunresolved(&nd->nl_nch);
2384                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2385                 if (error == 0)
2386                         goto again;
2387         } else {
2388                 vput(vp);
2389         }
2390         return (error);
2391 }
2392
2393 /*
2394  * stat_args(char *path, struct stat *ub)
2395  *
2396  * Get file status; this version follows links.
2397  */
2398 int
2399 sys_stat(struct stat_args *uap)
2400 {
2401         struct nlookupdata nd;
2402         struct stat st;
2403         int error;
2404
2405         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2406         if (error == 0) {
2407                 error = kern_stat(&nd, &st);
2408                 if (error == 0)
2409                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2410         }
2411         nlookup_done(&nd);
2412         return (error);
2413 }
2414
2415 /*
2416  * lstat_args(char *path, struct stat *ub)
2417  *
2418  * Get file status; this version does not follow links.
2419  */
2420 int
2421 sys_lstat(struct lstat_args *uap)
2422 {
2423         struct nlookupdata nd;
2424         struct stat st;
2425         int error;
2426
2427         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2428         if (error == 0) {
2429                 error = kern_stat(&nd, &st);
2430                 if (error == 0)
2431                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2432         }
2433         nlookup_done(&nd);
2434         return (error);
2435 }
2436
2437 /*
2438  * pathconf_Args(char *path, int name)
2439  *
2440  * Get configurable pathname variables.
2441  */
2442 /* ARGSUSED */
2443 int
2444 sys_pathconf(struct pathconf_args *uap)
2445 {
2446         struct nlookupdata nd;
2447         struct vnode *vp;
2448         int error;
2449
2450         vp = NULL;
2451         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2452         if (error == 0)
2453                 error = nlookup(&nd);
2454         if (error == 0)
2455                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2456         nlookup_done(&nd);
2457         if (error == 0) {
2458                 error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2459                 vput(vp);
2460         }
2461         return (error);
2462 }
2463
2464 /*
2465  * XXX: daver
2466  * kern_readlink isn't properly split yet.  There is a copyin burried
2467  * in VOP_READLINK().
2468  */
2469 int
2470 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2471 {
2472         struct thread *td = curthread;
2473         struct proc *p = td->td_proc;
2474         struct vnode *vp;
2475         struct iovec aiov;
2476         struct uio auio;
2477         int error;
2478
2479         if ((error = nlookup(nd)) != 0)
2480                 return (error);
2481         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2482         if (error)
2483                 return (error);
2484         if (vp->v_type != VLNK) {
2485                 error = EINVAL;
2486         } else {
2487                 aiov.iov_base = buf;
2488                 aiov.iov_len = count;
2489                 auio.uio_iov = &aiov;
2490                 auio.uio_iovcnt = 1;
2491                 auio.uio_offset = 0;
2492                 auio.uio_rw = UIO_READ;
2493                 auio.uio_segflg = UIO_USERSPACE;
2494                 auio.uio_td = td;
2495                 auio.uio_resid = count;
2496                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2497         }
2498         vput(vp);
2499         *res = count - auio.uio_resid;
2500         return (error);
2501 }
2502
2503 /*
2504  * readlink_args(char *path, char *buf, int count)
2505  *
2506  * Return target name of a symbolic link.
2507  */
2508 int
2509 sys_readlink(struct readlink_args *uap)
2510 {
2511         struct nlookupdata nd;
2512         int error;
2513
2514         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2515         if (error == 0) {
2516                 error = kern_readlink(&nd, uap->buf, uap->count,
2517                                         &uap->sysmsg_result);
2518         }
2519         nlookup_done(&nd);
2520         return (error);
2521 }
2522
2523 static int
2524 setfflags(struct vnode *vp, int flags)
2525 {
2526         struct thread *td = curthread;
2527         struct proc *p = td->td_proc;
2528         int error;
2529         struct vattr vattr;
2530
2531         /*
2532          * Prevent non-root users from setting flags on devices.  When
2533          * a device is reused, users can retain ownership of the device
2534          * if they are allowed to set flags and programs assume that
2535          * chown can't fail when done as root.
2536          */
2537         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2538             ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2539                 return (error);
2540
2541         /*
2542          * note: vget is required for any operation that might mod the vnode
2543          * so VINACTIVE is properly cleared.
2544          */
2545         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2546                 VATTR_NULL(&vattr);
2547                 vattr.va_flags = flags;
2548                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2549                 vput(vp);
2550         }
2551         return (error);
2552 }
2553
2554 /*
2555  * chflags(char *path, int flags)
2556  *
2557  * Change flags of a file given a path name.
2558  */
2559 /* ARGSUSED */
2560 int
2561 sys_chflags(struct chflags_args *uap)
2562 {
2563         struct nlookupdata nd;
2564         struct vnode *vp;
2565         int error;
2566
2567         vp = NULL;
2568         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2569         /* XXX Add NLC flag indicating modifying operation? */
2570         if (error == 0)
2571                 error = nlookup(&nd);
2572         if (error == 0)
2573                 error = ncp_writechk(&nd.nl_nch);
2574         if (error == 0)
2575                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2576         nlookup_done(&nd);
2577         if (error == 0) {
2578                 error = setfflags(vp, uap->flags);
2579                 vrele(vp);
2580         }
2581         return (error);
2582 }
2583
2584 /*
2585  * fchflags_args(int fd, int flags)
2586  *
2587  * Change flags of a file given a file descriptor.
2588  */
2589 /* ARGSUSED */
2590 int
2591 sys_fchflags(struct fchflags_args *uap)
2592 {
2593         struct thread *td = curthread;
2594         struct proc *p = td->td_proc;
2595         struct file *fp;
2596         int error;
2597
2598         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2599                 return (error);
2600         if (fp->f_nchandle.ncp)
2601                 error = ncp_writechk(&fp->f_nchandle);
2602         if (error == 0)
2603                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2604         fdrop(fp);
2605         return (error);
2606 }
2607
2608 static int
2609 setfmode(struct vnode *vp, int mode)
2610 {
2611         struct thread *td = curthread;
2612         struct proc *p = td->td_proc;
2613         int error;
2614         struct vattr vattr;
2615
2616         /*
2617          * note: vget is required for any operation that might mod the vnode
2618          * so VINACTIVE is properly cleared.
2619          */
2620         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2621                 VATTR_NULL(&vattr);
2622                 vattr.va_mode = mode & ALLPERMS;
2623                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2624                 vput(vp);
2625         }
2626         return error;
2627 }
2628
2629 int
2630 kern_chmod(struct nlookupdata *nd, int mode)
2631 {
2632         struct vnode *vp;
2633         int error;
2634
2635         /* XXX Add NLC flag indicating modifying operation? */
2636         if ((error = nlookup(nd)) != 0)
2637                 return (error);
2638         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2639                 return (error);
2640         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2641                 error = setfmode(vp, mode);
2642         vrele(vp);
2643         return (error);
2644 }
2645
2646 /*
2647  * chmod_args(char *path, int mode)
2648  *
2649  * Change mode of a file given path name.
2650  */
2651 /* ARGSUSED */
2652 int
2653 sys_chmod(struct chmod_args *uap)
2654 {
2655         struct nlookupdata nd;
2656         int error;
2657
2658         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2659         if (error == 0)
2660                 error = kern_chmod(&nd, uap->mode);
2661         nlookup_done(&nd);
2662         return (error);
2663 }
2664
2665 /*
2666  * lchmod_args(char *path, int mode)
2667  *
2668  * Change mode of a file given path name (don't follow links.)
2669  */
2670 /* ARGSUSED */
2671 int
2672 sys_lchmod(struct lchmod_args *uap)
2673 {
2674         struct nlookupdata nd;
2675         int error;
2676
2677         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2678         if (error == 0)
2679                 error = kern_chmod(&nd, uap->mode);
2680         nlookup_done(&nd);
2681         return (error);
2682 }
2683
2684 /*
2685  * fchmod_args(int fd, int mode)
2686  *
2687  * Change mode of a file given a file descriptor.
2688  */
2689 /* ARGSUSED */
2690 int
2691 sys_fchmod(struct fchmod_args *uap)
2692 {
2693         struct thread *td = curthread;
2694         struct proc *p = td->td_proc;
2695         struct file *fp;
2696         int error;
2697
2698         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2699                 return (error);
2700         if (fp->f_nchandle.ncp)
2701                 error = ncp_writechk(&fp->f_nchandle);
2702         if (error == 0)
2703                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2704         fdrop(fp);
2705         return (error);
2706 }
2707
2708 static int
2709 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2710 {
2711         struct thread *td = curthread;
2712         struct proc *p = td->td_proc;
2713         int error;
2714         struct vattr vattr;
2715
2716         /*
2717          * note: vget is required for any operation that might mod the vnode
2718          * so VINACTIVE is properly cleared.
2719          */
2720         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2721                 VATTR_NULL(&vattr);
2722                 vattr.va_uid = uid;
2723                 vattr.va_gid = gid;
2724                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2725                 vput(vp);
2726         }
2727         return error;
2728 }
2729
2730 int
2731 kern_chown(struct nlookupdata *nd, int uid, int gid)
2732 {
2733         struct vnode *vp;
2734         int error;
2735
2736         /* XXX Add NLC flag indicating modifying operation? */
2737         if ((error = nlookup(nd)) != 0)
2738                 return (error);
2739         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2740                 return (error);
2741         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2742                 error = setfown(vp, uid, gid);
2743         vrele(vp);
2744         return (error);
2745 }
2746
2747 /*
2748  * chown(char *path, int uid, int gid)
2749  *
2750  * Set ownership given a path name.
2751  */
2752 int
2753 sys_chown(struct chown_args *uap)
2754 {
2755         struct nlookupdata nd;
2756         int error;
2757
2758         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2759         if (error == 0)
2760                 error = kern_chown(&nd, uap->uid, uap->gid);
2761         nlookup_done(&nd);
2762         return (error);
2763 }
2764
2765 /*
2766  * lchown_args(char *path, int uid, int gid)
2767  *
2768  * Set ownership given a path name, do not cross symlinks.
2769  */
2770 int
2771 sys_lchown(struct lchown_args *uap)
2772 {
2773         struct nlookupdata nd;
2774         int error;
2775
2776         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2777         if (error == 0)
2778                 error = kern_chown(&nd, uap->uid, uap->gid);
2779         nlookup_done(&nd);
2780         return (error);
2781 }
2782
2783 /*
2784  * fchown_args(int fd, int uid, int gid)
2785  *
2786  * Set ownership given a file descriptor.
2787  */
2788 /* ARGSUSED */
2789 int
2790 sys_fchown(struct fchown_args *uap)
2791 {
2792         struct thread *td = curthread;
2793         struct proc *p = td->td_proc;
2794         struct file *fp;
2795         int error;
2796
2797         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2798                 return (error);
2799         if (fp->f_nchandle.ncp)
2800                 error = ncp_writechk(&fp->f_nchandle);
2801         if (error == 0)
2802                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2803         fdrop(fp);
2804         return (error);
2805 }
2806
2807 static int
2808 getutimes(const struct timeval *tvp, struct timespec *tsp)
2809 {
2810         struct timeval tv[2];
2811
2812         if (tvp == NULL) {
2813                 microtime(&tv[0]);
2814                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2815                 tsp[1] = tsp[0];
2816         } else {
2817                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2818                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2819         }
2820         return 0;
2821 }
2822
2823 static int
2824 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2825 {
2826         struct thread *td = curthread;
2827         struct proc *p = td->td_proc;
2828         int error;
2829         struct vattr vattr;
2830
2831         /*
2832          * note: vget is required for any operation that might mod the vnode
2833          * so VINACTIVE is properly cleared.
2834          */
2835         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2836                 VATTR_NULL(&vattr);
2837                 vattr.va_atime = ts[0];
2838                 vattr.va_mtime = ts[1];
2839                 if (nullflag)
2840                         vattr.va_vaflags |= VA_UTIMES_NULL;
2841                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2842                 vput(vp);
2843         }
2844         return error;
2845 }
2846
2847 int
2848 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2849 {
2850         struct timespec ts[2];
2851         struct vnode *vp;
2852         int error;
2853
2854         if ((error = getutimes(tptr, ts)) != 0)
2855                 return (error);
2856         /* XXX Add NLC flag indicating modifying operation? */
2857         if ((error = nlookup(nd)) != 0)
2858                 return (error);
2859         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2860                 return (error);
2861         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2862                 return (error);
2863         error = setutimes(vp, ts, tptr == NULL);
2864         vrele(vp);
2865         return (error);
2866 }
2867
2868 /*
2869  * utimes_args(char *path, struct timeval *tptr)
2870  *
2871  * Set the access and modification times of a file.
2872  */
2873 int
2874 sys_utimes(struct utimes_args *uap)
2875 {
2876         struct timeval tv[2];
2877         struct nlookupdata nd;
2878         int error;
2879
2880         if (uap->tptr) {
2881                 error = copyin(uap->tptr, tv, sizeof(tv));
2882                 if (error)
2883                         return (error);
2884         }
2885         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2886         if (error == 0)
2887                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2888         nlookup_done(&nd);
2889         return (error);
2890 }
2891
2892 /*
2893  * lutimes_args(char *path, struct timeval *tptr)
2894  *
2895  * Set the access and modification times of a file.
2896  */
2897 int
2898 sys_lutimes(struct lutimes_args *uap)
2899 {
2900         struct timeval tv[2];
2901         struct nlookupdata nd;
2902         int error;
2903
2904         if (uap->tptr) {
2905                 error = copyin(uap->tptr, tv, sizeof(tv));
2906                 if (error)
2907                         return (error);
2908         }
2909         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2910         if (error == 0)
2911                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2912         nlookup_done(&nd);
2913         return (error);
2914 }
2915
2916 int
2917 kern_futimes(int fd, struct timeval *tptr)
2918 {
2919         struct thread *td = curthread;
2920         struct proc *p = td->td_proc;
2921         struct timespec ts[2];
2922         struct file *fp;
2923         int error;
2924
2925         error = getutimes(tptr, ts);
2926         if (error)
2927                 return (error);
2928         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2929                 return (error);
2930         if (fp->f_nchandle.ncp)
2931                 error = ncp_writechk(&fp->f_nchandle);
2932         if (error == 0)
2933                 error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2934         fdrop(fp);
2935         return (error);
2936 }
2937
2938 /*
2939  * futimes_args(int fd, struct timeval *tptr)
2940  *
2941  * Set the access and modification times of a file.
2942  */
2943 int
2944 sys_futimes(struct futimes_args *uap)
2945 {
2946         struct timeval tv[2];
2947         int error;
2948
2949         if (uap->tptr) {
2950                 error = copyin(uap->tptr, tv, sizeof(tv));
2951                 if (error)
2952                         return (error);
2953         }
2954
2955         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2956
2957         return (error);
2958 }
2959
2960 int
2961 kern_truncate(struct nlookupdata *nd, off_t length)
2962 {
2963         struct vnode *vp;
2964         struct vattr vattr;
2965         int error;
2966
2967         if (length < 0)
2968                 return(EINVAL);
2969         /* XXX Add NLC flag indicating modifying operation? */
2970         if ((error = nlookup(nd)) != 0)
2971                 return (error);
2972         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2973                 return (error);
2974         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2975                 return (error);
2976         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2977                 vrele(vp);
2978                 return (error);
2979         }
2980         if (vp->v_type == VDIR) {
2981                 error = EISDIR;
2982         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2983             (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2984                 VATTR_NULL(&vattr);
2985                 vattr.va_size = length;
2986                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2987         }
2988         vput(vp);
2989         return (error);
2990 }
2991
2992 /*
2993  * truncate(char *path, int pad, off_t length)
2994  *
2995  * Truncate a file given its path name.
2996  */
2997 int
2998 sys_truncate(struct truncate_args *uap)
2999 {
3000         struct nlookupdata nd;
3001         int error;
3002
3003         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3004         if (error == 0)
3005                 error = kern_truncate(&nd, uap->length);
3006         nlookup_done(&nd);
3007         return error;
3008 }
3009
3010 int
3011 kern_ftruncate(int fd, off_t length)
3012 {
3013         struct thread *td = curthread;
3014         struct proc *p = td->td_proc;
3015         struct vattr vattr;
3016         struct vnode *vp;
3017         struct file *fp;
3018         int error;
3019
3020         if (length < 0)
3021                 return(EINVAL);
3022         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3023                 return (error);
3024         if (fp->f_nchandle.ncp) {
3025                 error = ncp_writechk(&fp->f_nchandle);
3026                 if (error)
3027                         goto done;
3028         }
3029         if ((fp->f_flag & FWRITE) == 0) {
3030                 error = EINVAL;
3031                 goto done;
3032         }
3033         vp = (struct vnode *)fp->f_data;
3034         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3035         if (vp->v_type == VDIR) {
3036                 error = EISDIR;
3037         } else if ((error = vn_writechk(vp, NULL)) == 0) {
3038                 VATTR_NULL(&vattr);
3039                 vattr.va_size = length;
3040                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3041         }
3042         vn_unlock(vp);
3043 done:
3044         fdrop(fp);
3045         return (error);
3046 }
3047
3048 /*
3049  * ftruncate_args(int fd, int pad, off_t length)
3050  *
3051  * Truncate a file given a file descriptor.
3052  */
3053 int
3054 sys_ftruncate(struct ftruncate_args *uap)
3055 {
3056         int error;
3057
3058         error = kern_ftruncate(uap->fd, uap->length);
3059
3060         return (error);
3061 }
3062
3063 /*
3064  * fsync(int fd)
3065  *
3066  * Sync an open file.
3067  */
3068 /* ARGSUSED */
3069 int
3070 sys_fsync(struct fsync_args *uap)
3071 {
3072         struct thread *td = curthread;
3073         struct proc *p = td->td_proc;
3074         struct vnode *vp;
3075         struct file *fp;
3076         vm_object_t obj;
3077         int error;
3078
3079         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3080                 return (error);
3081         vp = (struct vnode *)fp->f_data;
3082         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3083         if ((obj = vp->v_object) != NULL)
3084                 vm_object_page_clean(obj, 0, 0, 0);
3085         if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
3086                 error = buf_fsync(vp);
3087         vn_unlock(vp);
3088         fdrop(fp);
3089         return (error);
3090 }
3091
3092 int
3093 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3094 {
3095         struct nchandle fnchd;
3096         struct nchandle tnchd;
3097         struct namecache *ncp;
3098         struct vnode *fdvp;
3099         struct vnode *tdvp;
3100         struct mount *mp;
3101         int error;
3102
3103         bwillinode(1);
3104         fromnd->nl_flags |= NLC_REFDVP;
3105         if ((error = nlookup(fromnd)) != 0)
3106                 return (error);
3107         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3108                 return (ENOENT);
3109         fnchd.mount = fromnd->nl_nch.mount;
3110         cache_hold(&fnchd);
3111
3112         /*
3113          * unlock the source nch so we can lookup the target nch without
3114          * deadlocking.  The target may or may not exist so we do not check
3115          * for a target vp like kern_mkdir() and other creation functions do.
3116          *
3117          * The source and target directories are ref'd and rechecked after
3118          * everything is relocked to determine if the source or target file
3119          * has been renamed.
3120          */
3121         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3122         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3123         cache_unlock(&fromnd->nl_nch);
3124
3125         tond->nl_flags |= NLC_CREATE | NLC_REFDVP;
3126         if ((error = nlookup(tond)) != 0) {
3127                 cache_drop(&fnchd);
3128                 return (error);
3129         }
3130         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3131                 cache_drop(&fnchd);
3132                 return (ENOENT);
3133         }
3134         tnchd.mount = tond->nl_nch.mount;
3135         cache_hold(&tnchd);
3136
3137         /*
3138          * If the source and target are the same there is nothing to do
3139          */
3140         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3141                 cache_drop(&fnchd);
3142                 cache_drop(&tnchd);
3143                 return (0);
3144         }
3145
3146         /*
3147          * Mount points cannot be renamed or overwritten
3148          */
3149         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3150             NCF_ISMOUNTPT
3151         ) {
3152                 cache_drop(&fnchd);
3153                 cache_drop(&tnchd);
3154                 return (EINVAL);
3155         }
3156
3157         /*
3158          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
3159          * may have become invalid while it was unlocked, nc_vp and nc_mount
3160          * could be NULL.
3161          */
3162         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3163                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3164         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3165                 cache_lock(&fromnd->nl_nch);
3166                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3167         } else {
3168                 cache_unlock(&tond->nl_nch);
3169                 cache_lock(&fromnd->nl_nch);
3170                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3171                 cache_lock(&tond->nl_nch);
3172                 cache_resolve(&tond->nl_nch, tond->nl_cred);
3173         }
3174         fromnd->nl_flags |= NLC_NCPISLOCKED;
3175
3176         /*
3177          * make sure the parent directories linkages are the same
3178          */
3179         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3180             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3181                 cache_drop(&fnchd);
3182                 cache_drop(&tnchd);
3183                 return (ENOENT);
3184         }
3185
3186         /*
3187          * Both the source and target must be within the same filesystem and
3188          * in the same filesystem as their parent directories within the
3189          * namecache topology.
3190          *
3191          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3192          */
3193         mp = fnchd.mount;
3194         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3195             mp != tond->nl_nch.mount) {
3196                 cache_drop(&fnchd);
3197                 cache_drop(&tnchd);
3198                 return (EXDEV);
3199         }
3200
3201         /*
3202          * Make sure the mount point is writable
3203          */
3204         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3205                 cache_drop(&fnchd);
3206                 cache_drop(&tnchd);
3207                 return (error);
3208         }
3209
3210         /*
3211          * If the target exists and either the source or target is a directory,
3212          * then both must be directories.
3213          *
3214          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3215          * have become NULL.
3216          */
3217         if (tond->nl_nch.ncp->nc_vp) {
3218                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3219                         error = ENOENT;
3220                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3221                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3222                                 error = ENOTDIR;
3223                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3224                         error = EISDIR;
3225                 }
3226         }
3227
3228         /*
3229          * You cannot rename a source into itself or a subdirectory of itself.
3230          * We check this by travsersing the target directory upwards looking
3231          * for a match against the source.
3232          */
3233         if (error == 0) {
3234                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3235                         if (fromnd->nl_nch.ncp == ncp) {
3236                                 error = EINVAL;
3237                                 break;
3238                         }
3239                 }
3240         }
3241
3242         cache_drop(&fnchd);
3243         cache_drop(&tnchd);
3244
3245         /*
3246          * Even though the namespaces are different, they may still represent
3247          * hardlinks to the same file.  The filesystem might have a hard time
3248          * with this so we issue a NREMOVE of the source instead of a NRENAME
3249          * when we detect the situation.
3250          */
3251         if (error == 0) {
3252                 fdvp = fromnd->nl_dvp;
3253                 tdvp = tond->nl_dvp;
3254                 if (fdvp == NULL || tdvp == NULL) {
3255                         error = EPERM;
3256                 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3257                         error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3258                                             fromnd->nl_cred);
3259                 } else {
3260                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3261                                             fdvp, tdvp, tond->nl_cred);
3262                 }
3263         }
3264         return (error);
3265 }
3266
3267 /*
3268  * rename_args(char *from, char *to)
3269  *
3270  * Rename files.  Source and destination must either both be directories,
3271  * or both not be directories.  If target is a directory, it must be empty.
3272  */
3273 int
3274 sys_rename(struct rename_args *uap)
3275 {
3276         struct nlookupdata fromnd, tond;
3277         int error;
3278
3279         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3280         if (error == 0) {
3281                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3282                 if (error == 0)
3283                         error = kern_rename(&fromnd, &tond);
3284                 nlookup_done(&tond);
3285         }
3286         nlookup_done(&fromnd);
3287         return (error);
3288 }
3289
3290 int
3291 kern_mkdir(struct nlookupdata *nd, int mode)
3292 {
3293         struct thread *td = curthread;
3294         struct proc *p = td->td_proc;
3295         struct vnode *vp;
3296         struct vattr vattr;
3297         int error;
3298
3299         bwillinode(1);
3300         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3301         if ((error = nlookup(nd)) != 0)
3302                 return (error);
3303
3304         if (nd->nl_nch.ncp->nc_vp)
3305                 return (EEXIST);
3306         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3307                 return (error);
3308         VATTR_NULL(&vattr);
3309         vattr.va_type = VDIR;
3310         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3311
3312         vp = NULL;
3313         error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3314         if (error == 0)
3315                 vput(vp);
3316         return (error);
3317 }
3318
3319 /*
3320  * mkdir_args(char *path, int mode)
3321  *
3322  * Make a directory file.
3323  */
3324 /* ARGSUSED */
3325 int
3326 sys_mkdir(struct mkdir_args *uap)
3327 {
3328         struct nlookupdata nd;
3329         int error;
3330
3331         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3332         if (error == 0)
3333                 error = kern_mkdir(&nd, uap->mode);
3334         nlookup_done(&nd);
3335         return (error);
3336 }
3337
3338 int
3339 kern_rmdir(struct nlookupdata *nd)
3340 {
3341         int error;
3342
3343         bwillinode(1);
3344         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3345         if ((error = nlookup(nd)) != 0)
3346                 return (error);
3347
3348         /*
3349          * Do not allow directories representing mount points to be
3350          * deleted, even if empty.  Check write perms on mount point
3351          * in case the vnode is aliased (aka nullfs).
3352          */
3353         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3354                 return (EINVAL);
3355         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3356                 return (error);
3357         error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3358         return (error);
3359 }
3360
3361 /*
3362  * rmdir_args(char *path)
3363  *
3364  * Remove a directory file.
3365  */
3366 /* ARGSUSED */
3367 int
3368 sys_rmdir(struct rmdir_args *uap)
3369 {
3370         struct nlookupdata nd;
3371         int error;
3372
3373         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3374         if (error == 0)
3375                 error = kern_rmdir(&nd);
3376         nlookup_done(&nd);
3377         return (error);
3378 }
3379
3380 int
3381 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3382     enum uio_seg direction)
3383 {
3384         struct thread *td = curthread;
3385         struct proc *p = td->td_proc;
3386         struct vnode *vp;
3387         struct file *fp;
3388         struct uio auio;
3389         struct iovec aiov;
3390         off_t loff;
3391         int error, eofflag;
3392
3393         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3394                 return (error);
3395         if ((fp->f_flag & FREAD) == 0) {
3396                 error = EBADF;
3397                 goto done;
3398         }
3399         vp = (struct vnode *)fp->f_data;
3400 unionread:
3401         if (vp->v_type != VDIR) {
3402                 error = EINVAL;
3403                 goto done;
3404         }
3405         aiov.iov_base = buf;
3406         aiov.iov_len = count;
3407         auio.uio_iov = &aiov;
3408         auio.uio_iovcnt = 1;
3409         auio.uio_rw = UIO_READ;
3410         auio.uio_segflg = direction;
3411         auio.uio_td = td;
3412         auio.uio_resid = count;
3413         loff = auio.uio_offset = fp->f_offset;
3414         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3415         fp->f_offset = auio.uio_offset;
3416         if (error)
3417                 goto done;
3418         if (count == auio.uio_resid) {
3419                 if (union_dircheckp) {
3420                         error = union_dircheckp(td, &vp, fp);
3421                         if (error == -1)
3422                                 goto unionread;
3423                         if (error)
3424                                 goto done;
3425                 }
3426 #if 0
3427                 if ((vp->v_flag & VROOT) &&
3428                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3429                         struct vnode *tvp = vp;
3430                         vp = vp->v_mount->mnt_vnodecovered;
3431                         vref(vp);
3432                         fp->f_data = vp;
3433                         fp->f_offset = 0;
3434                         vrele(tvp);
3435                         goto unionread;
3436                 }
3437 #endif
3438         }
3439
3440         /*
3441          * WARNING!  *basep may not be wide enough to accomodate the
3442          * seek offset.   XXX should we hack this to return the upper 32 bits
3443          * for offsets greater then 4G?
3444          */
3445         if (basep) {
3446                 *basep = (long)loff;
3447         }
3448         *res = count - auio.uio_resid;
3449 done:
3450         fdrop(fp);
3451         return (error);
3452 }
3453
3454 /*
3455  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3456  *
3457  * Read a block of directory entries in a file system independent format.
3458  */
3459 int
3460 sys_getdirentries(struct getdirentries_args *uap)
3461 {
3462         long base;
3463         int error;
3464
3465         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3466                                    &uap->sysmsg_result, UIO_USERSPACE);
3467
3468         if (error == 0 && uap->basep)
3469                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3470         return (error);
3471 }
3472
3473 /*
3474  * getdents_args(int fd, char *buf, size_t count)
3475  */
3476 int
3477 sys_getdents(struct getdents_args *uap)
3478 {
3479         int error;
3480
3481         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3482             &uap->sysmsg_result, UIO_USERSPACE);
3483
3484         return (error);
3485 }
3486
3487 /*
3488  * umask(int newmask)
3489  *
3490  * Set the mode mask for creation of filesystem nodes.
3491  *
3492  * MP SAFE
3493  */
3494 int
3495 sys_umask(struct umask_args *uap)
3496 {
3497         struct thread *td = curthread;
3498         struct proc *p = td->td_proc;
3499         struct filedesc *fdp;
3500
3501         fdp = p->p_fd;
3502         uap->sysmsg_result = fdp->fd_cmask;
3503         fdp->fd_cmask = uap->newmask & ALLPERMS;
3504         return (0);
3505 }
3506
3507 /*
3508  * revoke(char *path)
3509  *
3510  * Void all references to file by ripping underlying filesystem
3511  * away from vnode.
3512  */
3513 /* ARGSUSED */
3514 int
3515 sys_revoke(struct revoke_args *uap)
3516 {
3517         struct nlookupdata nd;
3518         struct vattr vattr;
3519         struct vnode *vp;
3520         struct ucred *cred;
3521         int error;
3522
3523         vp = NULL;
3524         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3525         if (error == 0)
3526                 error = nlookup(&nd);
3527         if (error == 0)
3528                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3529         cred = crhold(nd.nl_cred);
3530         nlookup_done(&nd);
3531         if (error == 0) {
3532                 if (vp->v_type != VCHR && vp->v_type != VBLK)
3533                         error = EINVAL;
3534                 if (error == 0)
3535                         error = VOP_GETATTR(vp, &vattr);
3536                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3537                         error = suser_cred(cred, PRISON_ROOT);
3538                 if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3539                         error = 0;
3540                         vx_lock(vp);
3541                         VOP_REVOKE(vp, REVOKEALL);
3542                         vx_unlock(vp);
3543                 }
3544                 vrele(vp);
3545         }
3546         if (cred)
3547                 crfree(cred);
3548         return (error);
3549 }
3550
3551 /*
3552  * getfh_args(char *fname, fhandle_t *fhp)
3553  *
3554  * Get (NFS) file handle
3555  */
3556 int
3557 sys_getfh(struct getfh_args *uap)
3558 {
3559         struct thread *td = curthread;
3560         struct nlookupdata nd;
3561         fhandle_t fh;
3562         struct vnode *vp;
3563         int error;
3564
3565         /*
3566          * Must be super user
3567          */
3568         if ((error = suser(td)) != 0)
3569                 return (error);
3570
3571         vp = NULL;
3572         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3573         if (error == 0)
3574                 error = nlookup(&nd);
3575         if (error == 0)
3576                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3577         nlookup_done(&nd);
3578         if (error == 0) {
3579                 bzero(&fh, sizeof(fh));
3580                 fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3581                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3582                 vput(vp);
3583                 if (error == 0)
3584                         error = copyout(&fh, uap->fhp, sizeof(fh));
3585         }
3586         return (error);
3587 }
3588
3589 /*
3590  * fhopen_args(const struct fhandle *u_fhp, int flags)
3591  *
3592  * syscall for the rpc.lockd to use to translate a NFS file handle into
3593  * an open descriptor.
3594  *
3595  * warning: do not remove the suser() call or this becomes one giant
3596  * security hole.
3597  */
3598 int
3599 sys_fhopen(struct fhopen_args *uap)
3600 {
3601         struct thread *td = curthread;
3602         struct proc *p = td->td_proc;
3603         struct mount *mp;
3604         struct vnode *vp;
3605         struct fhandle fhp;
3606         struct vattr vat;
3607         struct vattr *vap = &vat;
3608         struct flock lf;
3609         int fmode, mode, error, type;
3610         struct file *nfp;
3611         struct file *fp;
3612         int indx;
3613
3614         /*
3615          * Must be super user
3616          */
3617         error = suser(td);
3618         if (error)
3619                 return (error);
3620
3621         fmode = FFLAGS(uap->flags);
3622         /* why not allow a non-read/write open for our lockd? */
3623         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3624                 return (EINVAL);
3625         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3626         if (error)
3627                 return(error);
3628         /* find the mount point */
3629         mp = vfs_getvfs(&fhp.fh_fsid);
3630         if (mp == NULL)
3631                 return (ESTALE);
3632         /* now give me my vnode, it gets returned to me locked */
3633         error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3634         if (error)
3635                 return (error);
3636         /*
3637          * from now on we have to make sure not
3638          * to forget about the vnode
3639          * any error that causes an abort must vput(vp)
3640          * just set error = err and 'goto bad;'.
3641          */
3642
3643         /*
3644          * from vn_open
3645          */
3646         if (vp->v_type == VLNK) {
3647                 error = EMLINK;
3648                 goto bad;
3649         }
3650         if (vp->v_type == VSOCK) {
3651                 error = EOPNOTSUPP;
3652                 goto bad;
3653         }
3654         mode = 0;
3655         if (fmode & (FWRITE | O_TRUNC)) {
3656                 if (vp->v_type == VDIR) {
3657                         error = EISDIR;
3658                         goto bad;
3659                 }
3660                 error = vn_writechk(vp, NULL);
3661                 if (error)
3662                         goto bad;
3663                 mode |= VWRITE;
3664         }
3665         if (fmode & FREAD)
3666                 mode |= VREAD;
3667         if (mode) {
3668                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3669                 if (error)
3670                         goto bad;
3671         }
3672         if (fmode & O_TRUNC) {
3673                 vn_unlock(vp);                          /* XXX */
3674                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3675                 VATTR_NULL(vap);
3676                 vap->va_size = 0;
3677                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3678                 if (error)
3679                         goto bad;
3680         }
3681
3682         /*
3683          * VOP_OPEN needs the file pointer so it can potentially override
3684          * it.
3685          *
3686          * WARNING! no f_nchandle will be associated when fhopen()ing a
3687          * directory.  XXX
3688          */
3689         if ((error = falloc(p, &nfp, &indx)) != 0)
3690                 goto bad;
3691         fp = nfp;
3692
3693         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3694         if (error) {
3695                 /*
3696                  * setting f_ops this way prevents VOP_CLOSE from being
3697                  * called or fdrop() releasing the vp from v_data.   Since
3698                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3699                  */
3700                 fp->f_ops = &badfileops;
3701                 fp->f_data = NULL;
3702                 goto bad_drop;
3703         }
3704
3705         /*
3706          * The fp is given its own reference, we still have our ref and lock.
3707          *
3708          * Assert that all regular files must be created with a VM object.
3709          */
3710         if (vp->v_type == VREG && vp->v_object == NULL) {
3711                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3712                 goto bad_drop;
3713         }
3714
3715         /*
3716          * The open was successful.  Handle any locking requirements.
3717          */
3718         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3719                 lf.l_whence = SEEK_SET;
3720                 lf.l_start = 0;
3721                 lf.l_len = 0;
3722                 if (fmode & O_EXLOCK)
3723                         lf.l_type = F_WRLCK;
3724                 else
3725                         lf.l_type = F_RDLCK;
3726                 if (fmode & FNONBLOCK)
3727                         type = 0;
3728                 else
3729                         type = F_WAIT;
3730                 vn_unlock(vp);
3731                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3732                         /*
3733                          * release our private reference.
3734                          */
3735                         fsetfd(p, NULL, indx);
3736                         fdrop(fp);
3737                         vrele(vp);
3738                         return (error);
3739                 }
3740                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3741                 fp->f_flag |= FHASLOCK;
3742         }
3743
3744         /*
3745          * Clean up.  Associate the file pointer with the previously
3746          * reserved descriptor and return it.
3747          */
3748         vput(vp);
3749         fsetfd(p, fp, indx);
3750         fdrop(fp);
3751         uap->sysmsg_result = indx;
3752         return (0);
3753
3754 bad_drop:
3755         fsetfd(p, NULL, indx);
3756         fdrop(fp);
3757 bad:
3758         vput(vp);
3759         return (error);
3760 }
3761
3762 /*
3763  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3764  */
3765 int
3766 sys_fhstat(struct fhstat_args *uap)
3767 {
3768         struct thread *td = curthread;
3769         struct stat sb;
3770         fhandle_t fh;
3771         struct mount *mp;
3772         struct vnode *vp;
3773         int error;
3774
3775         /*
3776          * Must be super user
3777          */
3778         error = suser(td);
3779         if (error)
3780                 return (error);
3781
3782         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3783         if (error)
3784                 return (error);
3785
3786         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3787                 return (ESTALE);
3788         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3789                 return (error);
3790         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3791         vput(vp);
3792         if (error)
3793                 return (error);
3794         error = copyout(&sb, uap->sb, sizeof(sb));
3795         return (error);
3796 }
3797
3798 /*
3799  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3800  */
3801 int
3802 sys_fhstatfs(struct fhstatfs_args *uap)
3803 {
3804         struct thread *td = curthread;
3805         struct proc *p = td->td_proc;
3806         struct statfs *sp;
3807         struct mount *mp;
3808         struct vnode *vp;
3809         struct statfs sb;
3810         char *fullpath, *freepath;
3811         fhandle_t fh;
3812         int error;
3813
3814         /*
3815          * Must be super user
3816          */
3817         if ((error = suser(td)))
3818                 return (error);
3819
3820         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3821                 return (error);
3822
3823         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3824                 return (ESTALE);
3825
3826         if (p != NULL && !chroot_visible_mnt(mp, p))
3827                 return (ESTALE);
3828
3829         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3830                 return (error);
3831         mp = vp->v_mount;
3832         sp = &mp->mnt_stat;
3833         vput(vp);
3834         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3835                 return (error);
3836
3837         error = mount_path(p, mp, &fullpath, &freepath);
3838         if (error)
3839                 return(error);
3840         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3841         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3842         kfree(freepath, M_TEMP);
3843
3844         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3845         if (suser(td)) {
3846                 bcopy(sp, &sb, sizeof(sb));
3847                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3848                 sp = &sb;
3849         }
3850         return (copyout(sp, uap->buf, sizeof(*sp)));
3851 }
3852
3853 /*
3854  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
3855  */
3856 int
3857 sys_fhstatvfs(struct fhstatvfs_args *uap)
3858 {
3859         struct thread *td = curthread;
3860         struct proc *p = td->td_proc;
3861         struct statvfs *sp;
3862         struct mount *mp;
3863         struct vnode *vp;
3864         fhandle_t fh;
3865         int error;
3866
3867         /*
3868          * Must be super user
3869          */
3870         if ((error = suser(td)))
3871                 return (error);
3872
3873         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3874                 return (error);
3875
3876         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3877                 return (ESTALE);
3878
3879         if (p != NULL && !chroot_visible_mnt(mp, p))
3880                 return (ESTALE);
3881
3882         if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3883                 return (error);
3884         mp = vp->v_mount;
3885         sp = &mp->mnt_vstat;
3886         vput(vp);
3887         if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
3888                 return (error);
3889
3890         sp->f_flag = 0;
3891         if (mp->mnt_flag & MNT_RDONLY)
3892                 sp->f_flag |= ST_RDONLY;
3893         if (mp->mnt_flag & MNT_NOSUID)
3894                 sp->f_flag |= ST_NOSUID;
3895
3896         return (copyout(sp, uap->buf, sizeof(*sp)));
3897 }
3898
3899
3900 /*
3901  * Syscall to push extended attribute configuration information into the
3902  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3903  * a command (int cmd), and attribute name and misc data.  For now, the
3904  * attribute name is left in userspace for consumption by the VFS_op.
3905  * It will probably be changed to be copied into sysspace by the
3906  * syscall in the future, once issues with various consumers of the
3907  * attribute code have raised their hands.
3908  *
3909  * Currently this is used only by UFS Extended Attributes.
3910  */
3911 int
3912 sys_extattrctl(struct extattrctl_args *uap)
3913 {
3914         struct nlookupdata nd;
3915         struct mount *mp;
3916         struct vnode *vp;
3917         int error;
3918
3919         vp = NULL;
3920         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3921         if (error == 0)
3922                 error = nlookup(&nd);
3923         if (error == 0) {
3924                 mp = nd.nl_nch.mount;
3925                 error = VFS_EXTATTRCTL(mp, uap->cmd,
3926                                 uap->attrname, uap->arg,
3927                                 nd.nl_cred);
3928         }
3929         nlookup_done(&nd);
3930         return (error);
3931 }
3932
3933 /*
3934  * Syscall to set a named extended attribute on a file or directory.
3935  * Accepts attribute name, and a uio structure pointing to the data to set.
3936  * The uio is consumed in the style of writev().  The real work happens
3937  * in VOP_SETEXTATTR().
3938  */
3939 int
3940 sys_extattr_set_file(struct extattr_set_file_args *uap)
3941 {
3942         char attrname[EXTATTR_MAXNAMELEN];
3943         struct iovec aiov[UIO_SMALLIOV];
3944         struct iovec *needfree;
3945         struct nlookupdata nd;
3946         struct iovec *iov;
3947         struct vnode *vp;
3948         struct uio auio;
3949         u_int iovlen;
3950         u_int cnt;
3951         int error;
3952         int i;
3953
3954         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3955         if (error)
3956                 return (error);
3957
3958         vp = NULL;
3959         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3960         if (error == 0)
3961                 error = nlookup(&nd);
3962         if (error == 0)
3963                 error = ncp_writechk(&nd.nl_nch);
3964         if (error == 0)
3965                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3966         if (error) {
3967                 nlookup_done(&nd);
3968                 return (error);
3969         }
3970
3971         needfree = NULL;
3972         iovlen = uap->iovcnt * sizeof(struct iovec);
3973         if (uap->iovcnt > UIO_SMALLIOV) {
3974                 if (uap->iovcnt > UIO_MAXIOV) {
3975                         error = EINVAL;
3976                         goto done;
3977                 }
3978                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3979                 needfree = iov;
3980         } else {
3981                 iov = aiov;
3982         }
3983         auio.uio_iov = iov;
3984         auio.uio_iovcnt = uap->iovcnt;
3985         auio.uio_rw = UIO_WRITE;
3986         auio.uio_segflg = UIO_USERSPACE;
3987         auio.uio_td = nd.nl_td;
3988         auio.uio_offset = 0;
3989         if ((error = copyin(uap->iovp, iov, iovlen)))
3990                 goto done;
3991         auio.uio_resid = 0;
3992         for (i = 0; i < uap->iovcnt; i++) {
3993                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
3994                         error = EINVAL;
3995                         goto done;
3996                 }
3997                 auio.uio_resid += iov->iov_len;
3998                 iov++;
3999         }
4000         cnt = auio.uio_resid;
4001         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4002         cnt -= auio.uio_resid;
4003         uap->sysmsg_result = cnt;
4004 done:
4005         vput(vp);
4006         nlookup_done(&nd);
4007         if (needfree)
4008                 FREE(needfree, M_IOV);
4009         return (error);
4010 }
4011
4012 /*
4013  * Syscall to get a named extended attribute on a file or directory.
4014  * Accepts attribute name, and a uio structure pointing to a buffer for the
4015  * data.  The uio is consumed in the style of readv().  The real work
4016  * happens in VOP_GETEXTATTR();
4017  */
4018 int
4019 sys_extattr_get_file(struct extattr_get_file_args *uap)
4020 {
4021         char attrname[EXTATTR_MAXNAMELEN];
4022         struct iovec aiov[UIO_SMALLIOV];
4023         struct iovec *needfree;
4024         struct nlookupdata nd;
4025         struct iovec *iov;
4026         struct vnode *vp;
4027         struct uio auio;
4028         u_int iovlen;
4029         u_int cnt;
4030         int error;
4031         int i;
4032
4033         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4034         if (error)
4035                 return (error);
4036
4037         vp = NULL;
4038         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4039         if (error == 0)
4040                 error = nlookup(&nd);
4041         if (error == 0)
4042                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4043         if (error) {
4044                 nlookup_done(&nd);
4045                 return (error);
4046         }
4047
4048         iovlen = uap->iovcnt * sizeof (struct iovec);
4049         needfree = NULL;
4050         if (uap->iovcnt > UIO_SMALLIOV) {
4051                 if (uap->iovcnt > UIO_MAXIOV) {
4052                         error = EINVAL;
4053                         goto done;
4054                 }
4055                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4056                 needfree = iov;
4057         } else {
4058                 iov = aiov;
4059         }
4060         auio.uio_iov = iov;
4061         auio.uio_iovcnt = uap->iovcnt;
4062         auio.uio_rw = UIO_READ;
4063         auio.uio_segflg = UIO_USERSPACE;
4064         auio.uio_td = nd.nl_td;
4065         auio.uio_offset = 0;
4066         if ((error = copyin(uap->iovp, iov, iovlen)))
4067                 goto done;
4068         auio.uio_resid = 0;
4069         for (i = 0; i < uap->iovcnt; i++) {
4070                 if (iov->iov_len > INT_MAX - auio.uio_resid) {
4071                         error = EINVAL;
4072                         goto done;
4073                 }
4074                 auio.uio_resid += iov->iov_len;
4075                 iov++;
4076         }
4077         cnt = auio.uio_resid;
4078         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4079         cnt -= auio.uio_resid;
4080         uap->sysmsg_result = cnt;
4081 done:
4082         vput(vp);
4083         nlookup_done(&nd);
4084         if (needfree)
4085                 FREE(needfree, M_IOV);
4086         return(error);
4087 }
4088
4089 /*
4090  * Syscall to delete a named extended attribute from a file or directory.
4091  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4092  */
4093 int
4094 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4095 {
4096         char attrname[EXTATTR_MAXNAMELEN];
4097         struct nlookupdata nd;
4098         struct vnode *vp;
4099         int error;
4100
4101         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4102         if (error)
4103                 return(error);
4104
4105         vp = NULL;
4106         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4107         if (error == 0)
4108                 error = nlookup(&nd);
4109         if (error == 0)
4110                 error = ncp_writechk(&nd.nl_nch);
4111         if (error == 0)
4112                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4113         if (error) {
4114                 nlookup_done(&nd);
4115                 return (error);
4116         }
4117
4118         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4119         vput(vp);
4120         nlookup_done(&nd);
4121         return(error);
4122 }
4123
4124 /*
4125  * Determine if the mount is visible to the process.
4126  */
4127 static int
4128 chroot_visible_mnt(struct mount *mp, struct proc *p)
4129 {
4130         struct nchandle nch;
4131
4132         /*
4133          * Traverse from the mount point upwards.  If we hit the process
4134          * root then the mount point is visible to the process.
4135          */
4136         nch = mp->mnt_ncmountpt;
4137         while (nch.ncp) {
4138                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
4139                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
4140                         return(1);
4141                 }
4142                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4143                         nch = nch.mount->mnt_ncmounton;
4144                 } else {
4145                         nch.ncp = nch.ncp->nc_parent;
4146                 }
4147         }
4148
4149         /*
4150          * If the mount point is not visible to the process, but the
4151          * process root is in a subdirectory of the mount, return
4152          * TRUE anyway.
4153          */
4154         if (p->p_fd->fd_nrdir.mount == mp)
4155                 return(1);
4156
4157         return(0);
4158 }
4159