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 1.135 2008/11/11 00:55:49 pavalos 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/priv.h>
  62 #include <sys/jail.h>
  63 #include <sys/namei.h>
  64 #include <sys/nlookup.h>
  65 #include <sys/dirent.h>
  66 #include <sys/extattr.h>
  67 #include <sys/spinlock.h>
  68 #include <sys/kern_syscall.h>
  69 #include <sys/objcache.h>
  70 #include <sys/sysctl.h>
  71
  72 #include <sys/buf2.h>
  73 #include <sys/file2.h>
  74 #include <sys/spinlock2.h>
  75
  76 #include <vm/vm.h>
  77 #include <vm/vm_object.h>
  78 #include <vm/vm_page.h>
  79
  80 #include <machine/limits.h>
  81 #include <machine/stdarg.h>
  82
  83 #include <vfs/union/union.h>
  84
  85 static void mount_warning(struct mount *mp, const char *ctl, ...);
  86 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
  87 static int checkvp_chdir (struct vnode *vn, struct thread *td);
  88 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
  89 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
  90 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
  91 static int getutimes (const struct timeval *, struct timespec *);
  92 static int setfown (struct vnode *, uid_t, gid_t);
  93 static int setfmode (struct vnode *, int);
  94 static int setfflags (struct vnode *, int);
  95 static int setutimes (struct vnode *, struct vattr *,
  96                         const struct timespec *, int);
  97 static int      usermount = 0;  /* if 1, non-root can mount fs. */
  98
  99 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
 100
 101 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
 102
 103 /*
 104  * Virtual File System System Calls
 105  */
 106
 107 /*
 108  * Mount a file system.
 109  */
 110 /*
 111  * mount_args(char *type, char *path, int flags, caddr_t data)
 112  */
 113 /* ARGSUSED */
 114 int
 115 sys_mount(struct mount_args *uap)
 116 {
 117         struct thread *td = curthread;
 118         struct proc *p = td->td_proc;
 119         struct vnode *vp;
 120         struct nchandle nch;
 121         struct mount *mp;
 122         struct vfsconf *vfsp;
 123         int error, flag = 0, flag2 = 0;
 124         int hasmount;
 125         struct vattr va;
 126         struct nlookupdata nd;
 127         char fstypename[MFSNAMELEN];
 128         struct ucred *cred = p->p_ucred;
 129
 130         KKASSERT(p);
 131         if (jailed(cred))
 132                 return (EPERM);
 133         if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
 134                 return (error);
 135         /*
 136          * Do not allow NFS export by non-root users.
 137          */
 138         if (uap->flags & MNT_EXPORTED) {
 139                 error = priv_check(td, PRIV_ROOT);
 140                 if (error)
 141                         return (error);
 142         }
 143         /*
 144          * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
 145          */
 146         if (priv_check(td, PRIV_ROOT))
 147                 uap->flags |= MNT_NOSUID | MNT_NODEV;
 148
 149         /*
 150          * Lookup the requested path and extract the nch and vnode.
 151          */
 152         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 153         if (error == 0) {
 154                 if ((error = nlookup(&nd)) == 0) {
 155                         if (nd.nl_nch.ncp->nc_vp == NULL)
 156                                 error = ENOENT;
 157                 }
 158         }
 159         if (error) {
 160                 nlookup_done(&nd);
 161                 return (error);
 162         }
 163
 164         /*
 165          * Extract the locked+refd ncp and cleanup the nd structure
 166          */
 167         nch = nd.nl_nch;
 168         cache_zero(&nd.nl_nch);
 169         nlookup_done(&nd);
 170
 171         if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
 172                 hasmount = 1;
 173         else
 174                 hasmount = 0;
 175
 176
 177         /*
 178          * now we have the locked ref'd nch and unreferenced vnode.
 179          */
 180         vp = nch.ncp->nc_vp;
 181         if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
 182                 cache_put(&nch);
 183                 return (error);
 184         }
 185         cache_unlock(&nch);
 186
 187         /*
 188          * Now we have an unlocked ref'd nch and a locked ref'd vp
 189          */
 190         if (uap->flags & MNT_UPDATE) {
 191                 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
 192                         cache_drop(&nch);
 193                         vput(vp);
 194                         return (EINVAL);
 195                 }
 196                 mp = vp->v_mount;
 197                 flag = mp->mnt_flag;
 198                 flag2 = mp->mnt_kern_flag;
 199                 /*
 200                  * We only allow the filesystem to be reloaded if it
 201                  * is currently mounted read-only.
 202                  */
 203                 if ((uap->flags & MNT_RELOAD) &&
 204                     ((mp->mnt_flag & MNT_RDONLY) == 0)) {
 205                         cache_drop(&nch);
 206                         vput(vp);
 207                         return (EOPNOTSUPP);    /* Needs translation */
 208                 }
 209                 /*
 210                  * Only root, or the user that did the original mount is
 211                  * permitted to update it.
 212                  */
 213                 if (mp->mnt_stat.f_owner != cred->cr_uid &&
 214                     (error = priv_check(td, PRIV_ROOT))) {
 215                         cache_drop(&nch);
 216                         vput(vp);
 217                         return (error);
 218                 }
 219                 if (vfs_busy(mp, LK_NOWAIT)) {
 220                         cache_drop(&nch);
 221                         vput(vp);
 222                         return (EBUSY);
 223                 }
 224                 if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 225                         cache_drop(&nch);
 226                         vfs_unbusy(mp);
 227                         vput(vp);
 228                         return (EBUSY);
 229                 }
 230                 vp->v_flag |= VMOUNT;
 231                 mp->mnt_flag |=
 232                     uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
 233                 vn_unlock(vp);
 234                 goto update;
 235         }
 236         /*
 237          * If the user is not root, ensure that they own the directory
 238          * onto which we are attempting to mount.
 239          */
 240         if ((error = VOP_GETATTR(vp, &va)) ||
 241             (va.va_uid != cred->cr_uid && (error = priv_check(td, PRIV_ROOT)))) {
 242                 cache_drop(&nch);
 243                 vput(vp);
 244                 return (error);
 245         }
 246         if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
 247                 cache_drop(&nch);
 248                 vput(vp);
 249                 return (error);
 250         }
 251         if (vp->v_type != VDIR) {
 252                 cache_drop(&nch);
 253                 vput(vp);
 254                 return (ENOTDIR);
 255         }
 256         if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
 257                 cache_drop(&nch);
 258                 vput(vp);
 259                 return (EPERM);
 260         }
 261         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
 262                 cache_drop(&nch);
 263                 vput(vp);
 264                 return (error);
 265         }
 266         vfsp = vfsconf_find_by_name(fstypename);
 267         if (vfsp == NULL) {
 268                 linker_file_t lf;
 269
 270                 /* Only load modules for root (very important!) */
 271                 if ((error = priv_check(td, PRIV_ROOT)) != 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                 vfsp = vfsconf_find_by_name(fstypename);
 287                 if (vfsp == NULL) {
 288                         lf->userrefs--;
 289                         linker_file_unload(lf);
 290                         cache_drop(&nch);
 291                         vput(vp);
 292                         return (ENODEV);
 293                 }
 294         }
 295         if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
 296                 cache_drop(&nch);
 297                 vput(vp);
 298                 return (EBUSY);
 299         }
 300         vp->v_flag |= VMOUNT;
 301
 302         /*
 303          * Allocate and initialize the filesystem.
 304          */
 305         mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
 306         TAILQ_INIT(&mp->mnt_nvnodelist);
 307         TAILQ_INIT(&mp->mnt_reservedvnlist);
 308         TAILQ_INIT(&mp->mnt_jlist);
 309         mp->mnt_nvnodelistsize = 0;
 310         lockinit(&mp->mnt_lock, "vfslock", 0, 0);
 311         vfs_busy(mp, LK_NOWAIT);
 312         mp->mnt_op = vfsp->vfc_vfsops;
 313         mp->mnt_vfc = vfsp;
 314         vfsp->vfc_refcount++;
 315         mp->mnt_stat.f_type = vfsp->vfc_typenum;
 316         mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
 317         strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
 318         mp->mnt_stat.f_owner = cred->cr_uid;
 319         mp->mnt_iosize_max = DFLTPHYS;
 320         vn_unlock(vp);
 321 update:
 322         /*
 323          * Set the mount level flags.
 324          */
 325         if (uap->flags & MNT_RDONLY)
 326                 mp->mnt_flag |= MNT_RDONLY;
 327         else if (mp->mnt_flag & MNT_RDONLY)
 328                 mp->mnt_kern_flag |= MNTK_WANTRDWR;
 329         mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
 330             MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
 331             MNT_NOSYMFOLLOW | MNT_IGNORE |
 332             MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 333         mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
 334             MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
 335             MNT_NOSYMFOLLOW | MNT_IGNORE |
 336             MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
 337         /*
 338          * Mount the filesystem.
 339          * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
 340          * get.
 341          */
 342         error = VFS_MOUNT(mp, uap->path, uap->data, cred);
 343         if (mp->mnt_flag & MNT_UPDATE) {
 344                 if (mp->mnt_kern_flag & MNTK_WANTRDWR)
 345                         mp->mnt_flag &= ~MNT_RDONLY;
 346                 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
 347                 mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
 348                 if (error) {
 349                         mp->mnt_flag = flag;
 350                         mp->mnt_kern_flag = flag2;
 351                 }
 352                 vfs_unbusy(mp);
 353                 vp->v_flag &= ~VMOUNT;
 354                 vrele(vp);
 355                 cache_drop(&nch);
 356                 return (error);
 357         }
 358         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 359         /*
 360          * Put the new filesystem on the mount list after root.  The mount
 361          * point gets its own mnt_ncmountpt (unless the VFS already set one
 362          * up) which represents the root of the mount.  The lookup code
 363          * detects the mount point going forward and checks the root of
 364          * the mount going backwards.
 365          *
 366          * It is not necessary to invalidate or purge the vnode underneath
 367          * because elements under the mount will be given their own glue
 368          * namecache record.
 369          */
 370         if (!error) {
 371                 if (mp->mnt_ncmountpt.ncp == NULL) {
 372                         /*
 373                          * allocate, then unlock, but leave the ref intact
 374                          */
 375                         cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
 376                         cache_unlock(&mp->mnt_ncmountpt);
 377                 }
 378                 mp->mnt_ncmounton = nch;                /* inherits ref */
 379                 nch.ncp->nc_flag |= NCF_ISMOUNTPT;
 380
 381                 /* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
 382                 vp->v_flag &= ~VMOUNT;
 383                 mountlist_insert(mp, MNTINS_LAST);
 384                 vn_unlock(vp);
 385                 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
 386                 error = vfs_allocate_syncvnode(mp);
 387                 vfs_unbusy(mp);
 388                 error = VFS_START(mp, 0);
 389                 vrele(vp);
 390         } else {
 391                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 392                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 393                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 394                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 395                 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 396                 vp->v_flag &= ~VMOUNT;
 397                 mp->mnt_vfc->vfc_refcount--;
 398                 vfs_unbusy(mp);
 399                 kfree(mp, M_MOUNT);
 400                 cache_drop(&nch);
 401                 vput(vp);
 402         }
 403         return (error);
 404 }
 405
 406 /*
 407  * Scan all active processes to see if any of them have a current
 408  * or root directory onto which the new filesystem has just been
 409  * mounted. If so, replace them with the new mount point.
 410  *
 411  * The passed ncp is ref'd and locked (from the mount code) and
 412  * must be associated with the vnode representing the root of the
 413  * mount point.
 414  */
 415 struct checkdirs_info {
 416         struct nchandle old_nch;
 417         struct nchandle new_nch;
 418         struct vnode *old_vp;
 419         struct vnode *new_vp;
 420 };
 421
 422 static int checkdirs_callback(struct proc *p, void *data);
 423
 424 static void
 425 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
 426 {
 427         struct checkdirs_info info;
 428         struct vnode *olddp;
 429         struct vnode *newdp;
 430         struct mount *mp;
 431
 432         /*
 433          * If the old mount point's vnode has a usecount of 1, it is not
 434          * being held as a descriptor anywhere.
 435          */
 436         olddp = old_nch->ncp->nc_vp;
 437         if (olddp == NULL || olddp->v_sysref.refcnt == 1)
 438                 return;
 439
 440         /*
 441          * Force the root vnode of the new mount point to be resolved
 442          * so we can update any matching processes.
 443          */
 444         mp = new_nch->mount;
 445         if (VFS_ROOT(mp, &newdp))
 446                 panic("mount: lost mount");
 447         cache_setunresolved(new_nch);
 448         cache_setvp(new_nch, newdp);
 449
 450         /*
 451          * Special handling of the root node
 452          */
 453         if (rootvnode == olddp) {
 454                 vref(newdp);
 455                 vfs_cache_setroot(newdp, cache_hold(new_nch));
 456         }
 457
 458         /*
 459          * Pass newdp separately so the callback does not have to access
 460          * it via new_nch->ncp->nc_vp.
 461          */
 462         info.old_nch = *old_nch;
 463         info.new_nch = *new_nch;
 464         info.new_vp = newdp;
 465         allproc_scan(checkdirs_callback, &info);
 466         vput(newdp);
 467 }
 468
 469 /*
 470  * NOTE: callback is not MP safe because the scanned process's filedesc
 471  * structure can be ripped out from under us, amoung other things.
 472  */
 473 static int
 474 checkdirs_callback(struct proc *p, void *data)
 475 {
 476         struct checkdirs_info *info = data;
 477         struct filedesc *fdp;
 478         struct nchandle ncdrop1;
 479         struct nchandle ncdrop2;
 480         struct vnode *vprele1;
 481         struct vnode *vprele2;
 482
 483         if ((fdp = p->p_fd) != NULL) {
 484                 cache_zero(&ncdrop1);
 485                 cache_zero(&ncdrop2);
 486                 vprele1 = NULL;
 487                 vprele2 = NULL;
 488
 489                 /*
 490                  * MPUNSAFE - XXX fdp can be pulled out from under a
 491                  * foreign process.
 492                  *
 493                  * A shared filedesc is ok, we don't have to copy it
 494                  * because we are making this change globally.
 495                  */
 496                 spin_lock_wr(&fdp->fd_spin);
 497                 if (fdp->fd_ncdir.mount == info->old_nch.mount &&
 498                     fdp->fd_ncdir.ncp == info->old_nch.ncp) {
 499                         vprele1 = fdp->fd_cdir;
 500                         vref(info->new_vp);
 501                         fdp->fd_cdir = info->new_vp;
 502                         ncdrop1 = fdp->fd_ncdir;
 503                         cache_copy(&info->new_nch, &fdp->fd_ncdir);
 504                 }
 505                 if (fdp->fd_nrdir.mount == info->old_nch.mount &&
 506                     fdp->fd_nrdir.ncp == info->old_nch.ncp) {
 507                         vprele2 = fdp->fd_rdir;
 508                         vref(info->new_vp);
 509                         fdp->fd_rdir = info->new_vp;
 510                         ncdrop2 = fdp->fd_nrdir;
 511                         cache_copy(&info->new_nch, &fdp->fd_nrdir);
 512                 }
 513                 spin_unlock_wr(&fdp->fd_spin);
 514                 if (ncdrop1.ncp)
 515                         cache_drop(&ncdrop1);
 516                 if (ncdrop2.ncp)
 517                         cache_drop(&ncdrop2);
 518                 if (vprele1)
 519                         vrele(vprele1);
 520                 if (vprele2)
 521                         vrele(vprele2);
 522         }
 523         return(0);
 524 }
 525
 526 /*
 527  * Unmount a file system.
 528  *
 529  * Note: unmount takes a path to the vnode mounted on as argument,
 530  * not special file (as before).
 531  */
 532 /*
 533  * umount_args(char *path, int flags)
 534  */
 535 /* ARGSUSED */
 536 int
 537 sys_unmount(struct unmount_args *uap)
 538 {
 539         struct thread *td = curthread;
 540         struct proc *p = td->td_proc;
 541         struct mount *mp = NULL;
 542         int error;
 543         struct nlookupdata nd;
 544
 545         KKASSERT(p);
 546         if (p->p_ucred->cr_prison != NULL)
 547                 return (EPERM);
 548         if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
 549                 return (error);
 550
 551         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 552         if (error == 0)
 553                 error = nlookup(&nd);
 554         if (error)
 555                 goto out;
 556
 557         mp = nd.nl_nch.mount;
 558
 559         /*
 560          * Only root, or the user that did the original mount is
 561          * permitted to unmount this filesystem.
 562          */
 563         if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
 564             (error = priv_check(td, PRIV_ROOT)))
 565                 goto out;
 566
 567         /*
 568          * Don't allow unmounting the root file system.
 569          */
 570         if (mp->mnt_flag & MNT_ROOTFS) {
 571                 error = EINVAL;
 572                 goto out;
 573         }
 574
 575         /*
 576          * Must be the root of the filesystem
 577          */
 578         if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
 579                 error = EINVAL;
 580                 goto out;
 581         }
 582
 583 out:
 584         nlookup_done(&nd);
 585         if (error)
 586                 return (error);
 587         return (dounmount(mp, uap->flags));
 588 }
 589
 590 /*
 591  * Do the actual file system unmount.
 592  */
 593 static int
 594 dounmount_interlock(struct mount *mp)
 595 {
 596         if (mp->mnt_kern_flag & MNTK_UNMOUNT)
 597                 return (EBUSY);
 598         mp->mnt_kern_flag |= MNTK_UNMOUNT;
 599         return(0);
 600 }
 601
 602 int
 603 dounmount(struct mount *mp, int flags)
 604 {
 605         struct namecache *ncp;
 606         struct nchandle nch;
 607         struct vnode *vp;
 608         int error;
 609         int async_flag;
 610         int lflags;
 611         int freeok = 1;
 612
 613         /*
 614          * Exclusive access for unmounting purposes
 615          */
 616         if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
 617                 return (error);
 618
 619         /*
 620          * Allow filesystems to detect that a forced unmount is in progress.
 621          */
 622         if (flags & MNT_FORCE)
 623                 mp->mnt_kern_flag |= MNTK_UNMOUNTF;
 624         lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
 625         error = lockmgr(&mp->mnt_lock, lflags);
 626         if (error) {
 627                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 628                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 629                         wakeup(mp);
 630                 return (error);
 631         }
 632
 633         if (mp->mnt_flag & MNT_EXPUBLIC)
 634                 vfs_setpublicfs(NULL, NULL, NULL);
 635
 636         vfs_msync(mp, MNT_WAIT);
 637         async_flag = mp->mnt_flag & MNT_ASYNC;
 638         mp->mnt_flag &=~ MNT_ASYNC;
 639
 640         /*
 641          * If this filesystem isn't aliasing other filesystems,
 642          * try to invalidate any remaining namecache entries and
 643          * check the count afterwords.
 644          */
 645         if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
 646                 cache_lock(&mp->mnt_ncmountpt);
 647                 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
 648                 cache_unlock(&mp->mnt_ncmountpt);
 649
 650                 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
 651                     (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
 652
 653                         if ((flags & MNT_FORCE) == 0) {
 654                                 error = EBUSY;
 655                                 mount_warning(mp, "Cannot unmount: "
 656                                                   "%d namecache "
 657                                                   "references still "
 658                                                   "present",
 659                                                   ncp->nc_refs - 1);
 660                         } else {
 661                                 mount_warning(mp, "Forced unmount: "
 662                                                   "%d namecache "
 663                                                   "references still "
 664                                                   "present",
 665                                                   ncp->nc_refs - 1);
 666                                 freeok = 0;
 667                         }
 668                 }
 669         }
 670
 671         /*
 672          * nchandle records ref the mount structure.  Expect a count of 1
 673          * (our mount->mnt_ncmountpt).
 674          */
 675         if (mp->mnt_refs != 1) {
 676                 if ((flags & MNT_FORCE) == 0) {
 677                         mount_warning(mp, "Cannot unmount: "
 678                                           "%d process references still "
 679                                           "present", mp->mnt_refs);
 680                         error = EBUSY;
 681                 } else {
 682                         mount_warning(mp, "Forced unmount: "
 683                                           "%d process references still "
 684                                           "present", mp->mnt_refs);
 685                         freeok = 0;
 686                 }
 687         }
 688
 689         /*
 690          * Decomission our special mnt_syncer vnode.  This also stops
 691          * the vnlru code.  If we are unable to unmount we recommission
 692          * the vnode.
 693          */
 694         if (error == 0) {
 695                 if ((vp = mp->mnt_syncer) != NULL) {
 696                         mp->mnt_syncer = NULL;
 697                         vrele(vp);
 698                 }
 699                 if (((mp->mnt_flag & MNT_RDONLY) ||
 700                      (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
 701                     (flags & MNT_FORCE)) {
 702                         error = VFS_UNMOUNT(mp, flags);
 703                 }
 704         }
 705         if (error) {
 706                 if (mp->mnt_syncer == NULL)
 707                         vfs_allocate_syncvnode(mp);
 708                 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
 709                 mp->mnt_flag |= async_flag;
 710                 lockmgr(&mp->mnt_lock, LK_RELEASE);
 711                 if (mp->mnt_kern_flag & MNTK_MWAIT)
 712                         wakeup(mp);
 713                 return (error);
 714         }
 715         /*
 716          * Clean up any journals still associated with the mount after
 717          * filesystem activity has ceased.
 718          */
 719         journal_remove_all_journals(mp,
 720             ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
 721
 722         mountlist_remove(mp);
 723
 724         /*
 725          * Remove any installed vnode ops here so the individual VFSs don't
 726          * have to.
 727          */
 728         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
 729         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
 730         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
 731         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
 732         vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
 733
 734         if (mp->mnt_ncmountpt.ncp != NULL) {
 735                 nch = mp->mnt_ncmountpt;
 736                 cache_zero(&mp->mnt_ncmountpt);
 737                 cache_clrmountpt(&nch);
 738                 cache_drop(&nch);
 739         }
 740         if (mp->mnt_ncmounton.ncp != NULL) {
 741                 nch = mp->mnt_ncmounton;
 742                 cache_zero(&mp->mnt_ncmounton);
 743                 cache_clrmountpt(&nch);
 744                 cache_drop(&nch);
 745         }
 746
 747         mp->mnt_vfc->vfc_refcount--;
 748         if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
 749                 panic("unmount: dangling vnode");
 750         lockmgr(&mp->mnt_lock, LK_RELEASE);
 751         if (mp->mnt_kern_flag & MNTK_MWAIT)
 752                 wakeup(mp);
 753         if (freeok)
 754                 kfree(mp, M_MOUNT);
 755         return (0);
 756 }
 757
 758 static
 759 void
 760 mount_warning(struct mount *mp, const char *ctl, ...)
 761 {
 762         char *ptr;
 763         char *buf;
 764         __va_list va;
 765
 766         __va_start(va, ctl);
 767         if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
 768                 kprintf("unmount(%s): ", ptr);
 769                 kvprintf(ctl, va);
 770                 kprintf("\n");
 771                 kfree(buf, M_TEMP);
 772         } else {
 773                 kprintf("unmount(%p", mp);
 774                 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
 775                         kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
 776                 kprintf("): ");
 777                 kvprintf(ctl, va);
 778                 kprintf("\n");
 779         }
 780         __va_end(va);
 781 }
 782
 783 /*
 784  * Shim cache_fullpath() to handle the case where a process is chrooted into
 785  * a subdirectory of a mount.  In this case if the root mount matches the
 786  * process root directory's mount we have to specify the process's root
 787  * directory instead of the mount point, because the mount point might
 788  * be above the root directory.
 789  */
 790 static
 791 int
 792 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
 793 {
 794         struct nchandle *nch;
 795
 796         if (p && p->p_fd->fd_nrdir.mount == mp)
 797                 nch = &p->p_fd->fd_nrdir;
 798         else
 799                 nch = &mp->mnt_ncmountpt;
 800         return(cache_fullpath(p, nch, rb, fb));
 801 }
 802
 803 /*
 804  * Sync each mounted filesystem.
 805  */
 806
 807 #ifdef DEBUG
 808 static int syncprt = 0;
 809 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
 810 #endif /* DEBUG */
 811
 812 static int sync_callback(struct mount *mp, void *data);
 813
 814 /* ARGSUSED */
 815 int
 816 sys_sync(struct sync_args *uap)
 817 {
 818         mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
 819 #ifdef DEBUG
 820         /*
 821          * print out buffer pool stat information on each sync() call.
 822          */
 823         if (syncprt)
 824                 vfs_bufstats();
 825 #endif /* DEBUG */
 826         return (0);
 827 }
 828
 829 static
 830 int
 831 sync_callback(struct mount *mp, void *data __unused)
 832 {
 833         int asyncflag;
 834
 835         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 836                 asyncflag = mp->mnt_flag & MNT_ASYNC;
 837                 mp->mnt_flag &= ~MNT_ASYNC;
 838                 vfs_msync(mp, MNT_NOWAIT);
 839                 VFS_SYNC(mp, MNT_NOWAIT);
 840                 mp->mnt_flag |= asyncflag;
 841         }
 842         return(0);
 843 }
 844
 845 /* XXX PRISON: could be per prison flag */
 846 static int prison_quotas;
 847 #if 0
 848 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
 849 #endif
 850
 851 /*
 852  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
 853  *
 854  * Change filesystem quotas.
 855  */
 856 /* ARGSUSED */
 857 int
 858 sys_quotactl(struct quotactl_args *uap)
 859 {
 860         struct nlookupdata nd;
 861         struct thread *td;
 862         struct proc *p;
 863         struct mount *mp;
 864         int error;
 865
 866         td = curthread;
 867         p = td->td_proc;
 868         if (p->p_ucred->cr_prison && !prison_quotas)
 869                 return (EPERM);
 870
 871         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
 872         if (error == 0)
 873                 error = nlookup(&nd);
 874         if (error == 0) {
 875                 mp = nd.nl_nch.mount;
 876                 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
 877                                     uap->arg, nd.nl_cred);
 878         }
 879         nlookup_done(&nd);
 880         return (error);
 881 }
 882
 883 /*
 884  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
 885  *              void *buf, int buflen)
 886  *
 887  * This function operates on a mount point and executes the specified
 888  * operation using the specified control data, and possibly returns data.
 889  *
 890  * The actual number of bytes stored in the result buffer is returned, 0
 891  * if none, otherwise an error is returned.
 892  */
 893 /* ARGSUSED */
 894 int
 895 sys_mountctl(struct mountctl_args *uap)
 896 {
 897         struct thread *td = curthread;
 898         struct proc *p = td->td_proc;
 899         struct file *fp;
 900         void *ctl = NULL;
 901         void *buf = NULL;
 902         char *path = NULL;
 903         int error;
 904
 905         /*
 906          * Sanity and permissions checks.  We must be root.
 907          */
 908         KKASSERT(p);
 909         if (p->p_ucred->cr_prison != NULL)
 910                 return (EPERM);
 911         if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
 912             (error = priv_check(td, PRIV_ROOT)) != 0)
 913                 return (error);
 914
 915         /*
 916          * Argument length checks
 917          */
 918         if (uap->ctllen < 0 || uap->ctllen > 1024)
 919                 return (EINVAL);
 920         if (uap->buflen < 0 || uap->buflen > 16 * 1024)
 921                 return (EINVAL);
 922         if (uap->path == NULL)
 923                 return (EINVAL);
 924
 925         /*
 926          * Allocate the necessary buffers and copyin data
 927          */
 928         path = objcache_get(namei_oc, M_WAITOK);
 929         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
 930         if (error)
 931                 goto done;
 932
 933         if (uap->ctllen) {
 934                 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
 935                 error = copyin(uap->ctl, ctl, uap->ctllen);
 936                 if (error)
 937                         goto done;
 938         }
 939         if (uap->buflen)
 940                 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
 941
 942         /*
 943          * Validate the descriptor
 944          */
 945         if (uap->fd >= 0) {
 946                 fp = holdfp(p->p_fd, uap->fd, -1);
 947                 if (fp == NULL) {
 948                         error = EBADF;
 949                         goto done;
 950                 }
 951         } else {
 952                 fp = NULL;
 953         }
 954
 955         /*
 956          * Execute the internal kernel function and clean up.
 957          */
 958         error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
 959         if (fp)
 960                 fdrop(fp);
 961         if (error == 0 && uap->sysmsg_result > 0)
 962                 error = copyout(buf, uap->buf, uap->sysmsg_result);
 963 done:
 964         if (path)
 965                 objcache_put(namei_oc, path);
 966         if (ctl)
 967                 kfree(ctl, M_TEMP);
 968         if (buf)
 969                 kfree(buf, M_TEMP);
 970         return (error);
 971 }
 972
 973 /*
 974  * Execute a mount control operation by resolving the path to a mount point
 975  * and calling vop_mountctl().
 976  *
 977  * Use the mount point from the nch instead of the vnode so nullfs mounts
 978  * can properly spike the VOP.
 979  */
 980 int
 981 kern_mountctl(const char *path, int op, struct file *fp,
 982                 const void *ctl, int ctllen,
 983                 void *buf, int buflen, int *res)
 984 {
 985         struct vnode *vp;
 986         struct mount *mp;
 987         struct nlookupdata nd;
 988         int error;
 989
 990         *res = 0;
 991         vp = NULL;
 992         error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
 993         if (error == 0)
 994                 error = nlookup(&nd);
 995         if (error == 0)
 996                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
 997         mp = nd.nl_nch.mount;
 998         nlookup_done(&nd);
 999         if (error)
1000                 return (error);
1001         vn_unlock(vp);
1002
1003         /*
1004          * Must be the root of the filesystem
1005          */
1006         if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1007                 vrele(vp);
1008                 return (EINVAL);
1009         }
1010         error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1011                                 buf, buflen, res);
1012         vrele(vp);
1013         return (error);
1014 }
1015
1016 int
1017 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1018 {
1019         struct thread *td = curthread;
1020         struct proc *p = td->td_proc;
1021         struct mount *mp;
1022         struct statfs *sp;
1023         char *fullpath, *freepath;
1024         int error;
1025
1026         if ((error = nlookup(nd)) != 0)
1027                 return (error);
1028         mp = nd->nl_nch.mount;
1029         sp = &mp->mnt_stat;
1030         if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1031                 return (error);
1032
1033         error = mount_path(p, mp, &fullpath, &freepath);
1034         if (error)
1035                 return(error);
1036         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1037         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1038         kfree(freepath, M_TEMP);
1039
1040         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1041         bcopy(sp, buf, sizeof(*buf));
1042         /* Only root should have access to the fsid's. */
1043         if (priv_check(td, PRIV_ROOT))
1044                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1045         return (0);
1046 }
1047
1048 /*
1049  * statfs_args(char *path, struct statfs *buf)
1050  *
1051  * Get filesystem statistics.
1052  */
1053 int
1054 sys_statfs(struct statfs_args *uap)
1055 {
1056         struct nlookupdata nd;
1057         struct statfs buf;
1058         int error;
1059
1060         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1061         if (error == 0)
1062                 error = kern_statfs(&nd, &buf);
1063         nlookup_done(&nd);
1064         if (error == 0)
1065                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1066         return (error);
1067 }
1068
1069 int
1070 kern_fstatfs(int fd, struct statfs *buf)
1071 {
1072         struct thread *td = curthread;
1073         struct proc *p = td->td_proc;
1074         struct file *fp;
1075         struct mount *mp;
1076         struct statfs *sp;
1077         char *fullpath, *freepath;
1078         int error;
1079
1080         KKASSERT(p);
1081         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1082                 return (error);
1083         mp = ((struct vnode *)fp->f_data)->v_mount;
1084         if (mp == NULL) {
1085                 error = EBADF;
1086                 goto done;
1087         }
1088         if (fp->f_cred == NULL) {
1089                 error = EINVAL;
1090                 goto done;
1091         }
1092         sp = &mp->mnt_stat;
1093         if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1094                 goto done;
1095
1096         if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1097                 goto done;
1098         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1099         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1100         kfree(freepath, M_TEMP);
1101
1102         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1103         bcopy(sp, buf, sizeof(*buf));
1104
1105         /* Only root should have access to the fsid's. */
1106         if (priv_check(td, PRIV_ROOT))
1107                 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1108         error = 0;
1109 done:
1110         fdrop(fp);
1111         return (error);
1112 }
1113
1114 /*
1115  * fstatfs_args(int fd, struct statfs *buf)
1116  *
1117  * Get filesystem statistics.
1118  */
1119 int
1120 sys_fstatfs(struct fstatfs_args *uap)
1121 {
1122         struct statfs buf;
1123         int error;
1124
1125         error = kern_fstatfs(uap->fd, &buf);
1126
1127         if (error == 0)
1128                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1129         return (error);
1130 }
1131
1132 int
1133 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1134 {
1135         struct mount *mp;
1136         struct statvfs *sp;
1137         int error;
1138
1139         if ((error = nlookup(nd)) != 0)
1140                 return (error);
1141         mp = nd->nl_nch.mount;
1142         sp = &mp->mnt_vstat;
1143         if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1144                 return (error);
1145
1146         sp->f_flag = 0;
1147         if (mp->mnt_flag & MNT_RDONLY)
1148                 sp->f_flag |= ST_RDONLY;
1149         if (mp->mnt_flag & MNT_NOSUID)
1150                 sp->f_flag |= ST_NOSUID;
1151         bcopy(sp, buf, sizeof(*buf));
1152         return (0);
1153 }
1154
1155 /*
1156  * statfs_args(char *path, struct statfs *buf)
1157  *
1158  * Get filesystem statistics.
1159  */
1160 int
1161 sys_statvfs(struct statvfs_args *uap)
1162 {
1163         struct nlookupdata nd;
1164         struct statvfs buf;
1165         int error;
1166
1167         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1168         if (error == 0)
1169                 error = kern_statvfs(&nd, &buf);
1170         nlookup_done(&nd);
1171         if (error == 0)
1172                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1173         return (error);
1174 }
1175
1176 int
1177 kern_fstatvfs(int fd, struct statvfs *buf)
1178 {
1179         struct thread *td = curthread;
1180         struct proc *p = td->td_proc;
1181         struct file *fp;
1182         struct mount *mp;
1183         struct statvfs *sp;
1184         int error;
1185
1186         KKASSERT(p);
1187         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1188                 return (error);
1189         mp = ((struct vnode *)fp->f_data)->v_mount;
1190         if (mp == NULL) {
1191                 error = EBADF;
1192                 goto done;
1193         }
1194         if (fp->f_cred == NULL) {
1195                 error = EINVAL;
1196                 goto done;
1197         }
1198         sp = &mp->mnt_vstat;
1199         if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1200                 goto done;
1201
1202         sp->f_flag = 0;
1203         if (mp->mnt_flag & MNT_RDONLY)
1204                 sp->f_flag |= ST_RDONLY;
1205         if (mp->mnt_flag & MNT_NOSUID)
1206                 sp->f_flag |= ST_NOSUID;
1207
1208         bcopy(sp, buf, sizeof(*buf));
1209         error = 0;
1210 done:
1211         fdrop(fp);
1212         return (error);
1213 }
1214
1215 /*
1216  * fstatfs_args(int fd, struct statfs *buf)
1217  *
1218  * Get filesystem statistics.
1219  */
1220 int
1221 sys_fstatvfs(struct fstatvfs_args *uap)
1222 {
1223         struct statvfs buf;
1224         int error;
1225
1226         error = kern_fstatvfs(uap->fd, &buf);
1227
1228         if (error == 0)
1229                 error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1230         return (error);
1231 }
1232
1233 /*
1234  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1235  *
1236  * Get statistics on all filesystems.
1237  */
1238
1239 struct getfsstat_info {
1240         struct statfs *sfsp;
1241         long count;
1242         long maxcount;
1243         int error;
1244         int flags;
1245         struct proc *p;
1246 };
1247
1248 static int getfsstat_callback(struct mount *, void *);
1249
1250 /* ARGSUSED */
1251 int
1252 sys_getfsstat(struct getfsstat_args *uap)
1253 {
1254         struct thread *td = curthread;
1255         struct proc *p = td->td_proc;
1256         struct getfsstat_info info;
1257
1258         bzero(&info, sizeof(info));
1259
1260         info.maxcount = uap->bufsize / sizeof(struct statfs);
1261         info.sfsp = uap->buf;
1262         info.count = 0;
1263         info.flags = uap->flags;
1264         info.p = p;
1265
1266         mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1267         if (info.sfsp && info.count > info.maxcount)
1268                 uap->sysmsg_result = info.maxcount;
1269         else
1270                 uap->sysmsg_result = info.count;
1271         return (info.error);
1272 }
1273
1274 static int
1275 getfsstat_callback(struct mount *mp, void *data)
1276 {
1277         struct getfsstat_info *info = data;
1278         struct statfs *sp;
1279         char *freepath;
1280         char *fullpath;
1281         int error;
1282
1283         if (info->sfsp && info->count < info->maxcount) {
1284                 if (info->p && !chroot_visible_mnt(mp, info->p))
1285                         return(0);
1286                 sp = &mp->mnt_stat;
1287
1288                 /*
1289                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1290                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1291                  * overrides MNT_WAIT.
1292                  */
1293                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1294                     (info->flags & MNT_WAIT)) &&
1295                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1296                         return(0);
1297                 }
1298                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1299
1300                 error = mount_path(info->p, mp, &fullpath, &freepath);
1301                 if (error) {
1302                         info->error = error;
1303                         return(-1);
1304                 }
1305                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1306                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1307                 kfree(freepath, M_TEMP);
1308
1309                 error = copyout(sp, info->sfsp, sizeof(*sp));
1310                 if (error) {
1311                         info->error = error;
1312                         return (-1);
1313                 }
1314                 ++info->sfsp;
1315         }
1316         info->count++;
1317         return(0);
1318 }
1319
1320 /*
1321  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1322                    long bufsize, int flags)
1323  *
1324  * Get statistics on all filesystems.
1325  */
1326
1327 struct getvfsstat_info {
1328         struct statfs *sfsp;
1329         struct statvfs *vsfsp;
1330         long count;
1331         long maxcount;
1332         int error;
1333         int flags;
1334         struct proc *p;
1335 };
1336
1337 static int getvfsstat_callback(struct mount *, void *);
1338
1339 /* ARGSUSED */
1340 int
1341 sys_getvfsstat(struct getvfsstat_args *uap)
1342 {
1343         struct thread *td = curthread;
1344         struct proc *p = td->td_proc;
1345         struct getvfsstat_info info;
1346
1347         bzero(&info, sizeof(info));
1348
1349         info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1350         info.sfsp = uap->buf;
1351         info.vsfsp = uap->vbuf;
1352         info.count = 0;
1353         info.flags = uap->flags;
1354         info.p = p;
1355
1356         mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1357         if (info.vsfsp && info.count > info.maxcount)
1358                 uap->sysmsg_result = info.maxcount;
1359         else
1360                 uap->sysmsg_result = info.count;
1361         return (info.error);
1362 }
1363
1364 static int
1365 getvfsstat_callback(struct mount *mp, void *data)
1366 {
1367         struct getvfsstat_info *info = data;
1368         struct statfs *sp;
1369         struct statvfs *vsp;
1370         char *freepath;
1371         char *fullpath;
1372         int error;
1373
1374         if (info->vsfsp && info->count < info->maxcount) {
1375                 if (info->p && !chroot_visible_mnt(mp, info->p))
1376                         return(0);
1377                 sp = &mp->mnt_stat;
1378                 vsp = &mp->mnt_vstat;
1379
1380                 /*
1381                  * If MNT_NOWAIT or MNT_LAZY is specified, do not
1382                  * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1383                  * overrides MNT_WAIT.
1384                  */
1385                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1386                     (info->flags & MNT_WAIT)) &&
1387                     (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1388                         return(0);
1389                 }
1390                 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1391
1392                 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1393                     (info->flags & MNT_WAIT)) &&
1394                     (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1395                         return(0);
1396                 }
1397                 vsp->f_flag = 0;
1398                 if (mp->mnt_flag & MNT_RDONLY)
1399                         vsp->f_flag |= ST_RDONLY;
1400                 if (mp->mnt_flag & MNT_NOSUID)
1401                         vsp->f_flag |= ST_NOSUID;
1402
1403                 error = mount_path(info->p, mp, &fullpath, &freepath);
1404                 if (error) {
1405                         info->error = error;
1406                         return(-1);
1407                 }
1408                 bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1409                 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1410                 kfree(freepath, M_TEMP);
1411
1412                 error = copyout(sp, info->sfsp, sizeof(*sp));
1413                 if (error == 0)
1414                         error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1415                 if (error) {
1416                         info->error = error;
1417                         return (-1);
1418                 }
1419                 ++info->sfsp;
1420                 ++info->vsfsp;
1421         }
1422         info->count++;
1423         return(0);
1424 }
1425
1426
1427 /*
1428  * fchdir_args(int fd)
1429  *
1430  * Change current working directory to a given file descriptor.
1431  */
1432 /* ARGSUSED */
1433 int
1434 sys_fchdir(struct fchdir_args *uap)
1435 {
1436         struct thread *td = curthread;
1437         struct proc *p = td->td_proc;
1438         struct filedesc *fdp = p->p_fd;
1439         struct vnode *vp, *ovp;
1440         struct mount *mp;
1441         struct file *fp;
1442         struct nchandle nch, onch, tnch;
1443         int error;
1444
1445         if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1446                 return (error);
1447         vp = (struct vnode *)fp->f_data;
1448         vref(vp);
1449         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1450         if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1451                 error = ENOTDIR;
1452         else
1453                 error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1454         if (error) {
1455                 vput(vp);
1456                 fdrop(fp);
1457                 return (error);
1458         }
1459         cache_copy(&fp->f_nchandle, &nch);
1460
1461         /*
1462          * If the ncp has become a mount point, traverse through
1463          * the mount point.
1464          */
1465
1466         while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1467                (mp = cache_findmount(&nch)) != NULL
1468         ) {
1469                 error = nlookup_mp(mp, &tnch);
1470                 if (error == 0) {
1471                         cache_unlock(&tnch);    /* leave ref intact */
1472                         vput(vp);
1473                         vp = tnch.ncp->nc_vp;
1474                         error = vget(vp, LK_SHARED);
1475                         KKASSERT(error == 0);
1476                         cache_drop(&nch);
1477                         nch = tnch;
1478                 }
1479         }
1480         if (error == 0) {
1481                 ovp = fdp->fd_cdir;
1482                 onch = fdp->fd_ncdir;
1483                 vn_unlock(vp);          /* leave ref intact */
1484                 fdp->fd_cdir = vp;
1485                 fdp->fd_ncdir = nch;
1486                 cache_drop(&onch);
1487                 vrele(ovp);
1488         } else {
1489                 cache_drop(&nch);
1490                 vput(vp);
1491         }
1492         fdrop(fp);
1493         return (error);
1494 }
1495
1496 int
1497 kern_chdir(struct nlookupdata *nd)
1498 {
1499         struct thread *td = curthread;
1500         struct proc *p = td->td_proc;
1501         struct filedesc *fdp = p->p_fd;
1502         struct vnode *vp, *ovp;
1503         struct nchandle onch;
1504         int error;
1505
1506         if ((error = nlookup(nd)) != 0)
1507                 return (error);
1508         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1509                 return (ENOENT);
1510         if ((error = vget(vp, LK_SHARED)) != 0)
1511                 return (error);
1512
1513         error = checkvp_chdir(vp, td);
1514         vn_unlock(vp);
1515         if (error == 0) {
1516                 ovp = fdp->fd_cdir;
1517                 onch = fdp->fd_ncdir;
1518                 cache_unlock(&nd->nl_nch);      /* leave reference intact */
1519                 fdp->fd_ncdir = nd->nl_nch;
1520                 fdp->fd_cdir = vp;
1521                 cache_drop(&onch);
1522                 vrele(ovp);
1523                 cache_zero(&nd->nl_nch);
1524         } else {
1525                 vrele(vp);
1526         }
1527         return (error);
1528 }
1529
1530 /*
1531  * chdir_args(char *path)
1532  *
1533  * Change current working directory (``.'').
1534  */
1535 int
1536 sys_chdir(struct chdir_args *uap)
1537 {
1538         struct nlookupdata nd;
1539         int error;
1540
1541         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1542         if (error == 0)
1543                 error = kern_chdir(&nd);
1544         nlookup_done(&nd);
1545         return (error);
1546 }
1547
1548 /*
1549  * Helper function for raised chroot(2) security function:  Refuse if
1550  * any filedescriptors are open directories.
1551  */
1552 static int
1553 chroot_refuse_vdir_fds(struct filedesc *fdp)
1554 {
1555         struct vnode *vp;
1556         struct file *fp;
1557         int error;
1558         int fd;
1559
1560         for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1561                 if ((error = holdvnode(fdp, fd, &fp)) != 0)
1562                         continue;
1563                 vp = (struct vnode *)fp->f_data;
1564                 if (vp->v_type != VDIR) {
1565                         fdrop(fp);
1566                         continue;
1567                 }
1568                 fdrop(fp);
1569                 return(EPERM);
1570         }
1571         return (0);
1572 }
1573
1574 /*
1575  * This sysctl determines if we will allow a process to chroot(2) if it
1576  * has a directory open:
1577  *      0: disallowed for all processes.
1578  *      1: allowed for processes that were not already chroot(2)'ed.
1579  *      2: allowed for all processes.
1580  */
1581
1582 static int chroot_allow_open_directories = 1;
1583
1584 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1585      &chroot_allow_open_directories, 0, "");
1586
1587 /*
1588  * chroot to the specified namecache entry.  We obtain the vp from the
1589  * namecache data.  The passed ncp must be locked and referenced and will
1590  * remain locked and referenced on return.
1591  */
1592 int
1593 kern_chroot(struct nchandle *nch)
1594 {
1595         struct thread *td = curthread;
1596         struct proc *p = td->td_proc;
1597         struct filedesc *fdp = p->p_fd;
1598         struct vnode *vp;
1599         int error;
1600
1601         /*
1602          * Only privileged user can chroot
1603          */
1604         error = priv_check_cred(p->p_ucred, PRIV_VFS_CHROOT, 0);
1605         if (error)
1606                 return (error);
1607
1608         /*
1609          * Disallow open directory descriptors (fchdir() breakouts).
1610          */
1611         if (chroot_allow_open_directories == 0 ||
1612            (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1613                 if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1614                         return (error);
1615         }
1616         if ((vp = nch->ncp->nc_vp) == NULL)
1617                 return (ENOENT);
1618
1619         if ((error = vget(vp, LK_SHARED)) != 0)
1620                 return (error);
1621
1622         /*
1623          * Check the validity of vp as a directory to change to and
1624          * associate it with rdir/jdir.
1625          */
1626         error = checkvp_chdir(vp, td);
1627         vn_unlock(vp);                  /* leave reference intact */
1628         if (error == 0) {
1629                 vrele(fdp->fd_rdir);
1630                 fdp->fd_rdir = vp;      /* reference inherited by fd_rdir */
1631                 cache_drop(&fdp->fd_nrdir);
1632                 cache_copy(nch, &fdp->fd_nrdir);
1633                 if (fdp->fd_jdir == NULL) {
1634                         fdp->fd_jdir = vp;
1635                         vref(fdp->fd_jdir);
1636                         cache_copy(nch, &fdp->fd_njdir);
1637                 }
1638         } else {
1639                 vrele(vp);
1640         }
1641         return (error);
1642 }
1643
1644 /*
1645  * chroot_args(char *path)
1646  *
1647  * Change notion of root (``/'') directory.
1648  */
1649 /* ARGSUSED */
1650 int
1651 sys_chroot(struct chroot_args *uap)
1652 {
1653         struct thread *td = curthread;
1654         struct nlookupdata nd;
1655         int error;
1656
1657         KKASSERT(td->td_proc);
1658         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1659         if (error) {
1660                 nlookup_done(&nd);
1661                 return(error);
1662         }
1663         nd.nl_flags |= NLC_EXEC;
1664         error = nlookup(&nd);
1665         if (error == 0)
1666                 error = kern_chroot(&nd.nl_nch);
1667         nlookup_done(&nd);
1668         return(error);
1669 }
1670
1671 /*
1672  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1673  * determine whether it is legal to chdir to the vnode.  The vnode's state
1674  * is not changed by this call.
1675  */
1676 int
1677 checkvp_chdir(struct vnode *vp, struct thread *td)
1678 {
1679         int error;
1680
1681         if (vp->v_type != VDIR)
1682                 error = ENOTDIR;
1683         else
1684                 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1685         return (error);
1686 }
1687
1688 int
1689 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1690 {
1691         struct thread *td = curthread;
1692         struct proc *p = td->td_proc;
1693         struct lwp *lp = td->td_lwp;
1694         struct filedesc *fdp = p->p_fd;
1695         int cmode, flags;
1696         struct file *nfp;
1697         struct file *fp;
1698         struct vnode *vp;
1699         int type, indx, error;
1700         struct flock lf;
1701
1702         if ((oflags & O_ACCMODE) == O_ACCMODE)
1703                 return (EINVAL);
1704         flags = FFLAGS(oflags);
1705         error = falloc(p, &nfp, NULL);
1706         if (error)
1707                 return (error);
1708         fp = nfp;
1709         cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1710
1711         /*
1712          * XXX p_dupfd is a real mess.  It allows a device to return a
1713          * file descriptor to be duplicated rather then doing the open
1714          * itself.
1715          */
1716         lp->lwp_dupfd = -1;
1717
1718         /*
1719          * Call vn_open() to do the lookup and assign the vnode to the
1720          * file pointer.  vn_open() does not change the ref count on fp
1721          * and the vnode, on success, will be inherited by the file pointer
1722          * and unlocked.
1723          */
1724         nd->nl_flags |= NLC_LOCKVP;
1725         error = vn_open(nd, fp, flags, cmode);
1726         nlookup_done(nd);
1727         if (error) {
1728                 /*
1729                  * handle special fdopen() case.  bleh.  dupfdopen() is
1730                  * responsible for dropping the old contents of ofiles[indx]
1731                  * if it succeeds.
1732                  *
1733                  * Note that fsetfd() will add a ref to fp which represents
1734                  * the fd_files[] assignment.  We must still drop our
1735                  * reference.
1736                  */
1737                 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1738                         if (fdalloc(p, 0, &indx) == 0) {
1739                                 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1740                                 if (error == 0) {
1741                                         *res = indx;
1742                                         fdrop(fp);      /* our ref */
1743                                         return (0);
1744                                 }
1745                                 fsetfd(p, NULL, indx);
1746                         }
1747                 }
1748                 fdrop(fp);      /* our ref */
1749                 if (error == ERESTART)
1750                         error = EINTR;
1751                 return (error);
1752         }
1753
1754         /*
1755          * ref the vnode for ourselves so it can't be ripped out from under
1756          * is.  XXX need an ND flag to request that the vnode be returned
1757          * anyway.
1758          *
1759          * Reserve a file descriptor but do not assign it until the open
1760          * succeeds.
1761          */
1762         vp = (struct vnode *)fp->f_data;
1763         vref(vp);
1764         if ((error = fdalloc(p, 0, &indx)) != 0) {
1765                 fdrop(fp);
1766                 vrele(vp);
1767                 return (error);
1768         }
1769
1770         /*
1771          * If no error occurs the vp will have been assigned to the file
1772          * pointer.
1773          */
1774         lp->lwp_dupfd = 0;
1775
1776         if (flags & (O_EXLOCK | O_SHLOCK)) {
1777                 lf.l_whence = SEEK_SET;
1778                 lf.l_start = 0;
1779                 lf.l_len = 0;
1780                 if (flags & O_EXLOCK)
1781                         lf.l_type = F_WRLCK;
1782                 else
1783                         lf.l_type = F_RDLCK;
1784                 if (flags & FNONBLOCK)
1785                         type = 0;
1786                 else
1787                         type = F_WAIT;
1788
1789                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1790                         /*
1791                          * lock request failed.  Clean up the reserved
1792                          * descriptor.
1793                          */
1794                         vrele(vp);
1795                         fsetfd(p, NULL, indx);
1796                         fdrop(fp);
1797                         return (error);
1798                 }
1799                 fp->f_flag |= FHASLOCK;
1800         }
1801 #if 0
1802         /*
1803          * Assert that all regular file vnodes were created with a object.
1804          */
1805         KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1806                 ("open: regular file has no backing object after vn_open"));
1807 #endif
1808
1809         vrele(vp);
1810
1811         /*
1812          * release our private reference, leaving the one associated with the
1813          * descriptor table intact.
1814          */
1815         fsetfd(p, fp, indx);
1816         fdrop(fp);
1817         *res = indx;
1818         return (0);
1819 }
1820
1821 /*
1822  * open_args(char *path, int flags, int mode)
1823  *
1824  * Check permissions, allocate an open file structure,
1825  * and call the device open routine if any.
1826  */
1827 int
1828 sys_open(struct open_args *uap)
1829 {
1830         struct nlookupdata nd;
1831         int error;
1832
1833         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1834         if (error == 0) {
1835                 error = kern_open(&nd, uap->flags,
1836                                     uap->mode, &uap->sysmsg_result);
1837         }
1838         nlookup_done(&nd);
1839         return (error);
1840 }
1841
1842 /*
1843  * openat_args(int fd, char *path, int flags, int mode)
1844  */
1845 int
1846 sys_openat(struct openat_args *uap)
1847 {
1848         struct nlookupdata nd;
1849         int error;
1850         struct file *fp;
1851
1852         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
1853         if (error == 0) {
1854                 error = kern_open(&nd, uap->flags, uap->mode,
1855                                         &uap->sysmsg_result);
1856         }
1857         nlookup_done_at(&nd, fp);
1858         return (error);
1859 }
1860
1861 int
1862 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1863 {
1864         struct thread *td = curthread;
1865         struct proc *p = td->td_proc;
1866         struct vnode *vp;
1867         struct vattr vattr;
1868         int error;
1869         int whiteout = 0;
1870
1871         KKASSERT(p);
1872
1873         VATTR_NULL(&vattr);
1874         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1875         vattr.va_rmajor = rmajor;
1876         vattr.va_rminor = rminor;
1877
1878         switch (mode & S_IFMT) {
1879         case S_IFMT:    /* used by badsect to flag bad sectors */
1880                 error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_BAD, 0);
1881                 vattr.va_type = VBAD;
1882                 break;
1883         case S_IFCHR:
1884                 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1885                 vattr.va_type = VCHR;
1886                 break;
1887         case S_IFBLK:
1888                 error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1889                 vattr.va_type = VBLK;
1890                 break;
1891         case S_IFWHT:
1892                 error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_WHT, 0);
1893                 whiteout = 1;
1894                 break;
1895         case S_IFDIR:   /* special directories support for HAMMER */
1896                 error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_DIR, 0);
1897                 vattr.va_type = VDIR;
1898                 break;
1899         default:
1900                 error = EINVAL;
1901                 break;
1902         }
1903
1904         if (error)
1905                 return (error);
1906
1907         bwillinode(1);
1908         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1909         if ((error = nlookup(nd)) != 0)
1910                 return (error);
1911         if (nd->nl_nch.ncp->nc_vp)
1912                 return (EEXIST);
1913         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1914                 return (error);
1915
1916         if (whiteout) {
1917                 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1918                                       nd->nl_cred, NAMEI_CREATE);
1919         } else {
1920                 vp = NULL;
1921                 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1922                                    &vp, nd->nl_cred, &vattr);
1923                 if (error == 0)
1924                         vput(vp);
1925         }
1926         return (error);
1927 }
1928
1929 /*
1930  * mknod_args(char *path, int mode, int dev)
1931  *
1932  * Create a special file.
1933  */
1934 int
1935 sys_mknod(struct mknod_args *uap)
1936 {
1937         struct nlookupdata nd;
1938         int error;
1939
1940         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1941         if (error == 0) {
1942                 error = kern_mknod(&nd, uap->mode,
1943                                    umajor(uap->dev), uminor(uap->dev));
1944         }
1945         nlookup_done(&nd);
1946         return (error);
1947 }
1948
1949 int
1950 kern_mkfifo(struct nlookupdata *nd, int mode)
1951 {
1952         struct thread *td = curthread;
1953         struct proc *p = td->td_proc;
1954         struct vattr vattr;
1955         struct vnode *vp;
1956         int error;
1957
1958         bwillinode(1);
1959
1960         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1961         if ((error = nlookup(nd)) != 0)
1962                 return (error);
1963         if (nd->nl_nch.ncp->nc_vp)
1964                 return (EEXIST);
1965         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1966                 return (error);
1967
1968         VATTR_NULL(&vattr);
1969         vattr.va_type = VFIFO;
1970         vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1971         vp = NULL;
1972         error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1973         if (error == 0)
1974                 vput(vp);
1975         return (error);
1976 }
1977
1978 /*
1979  * mkfifo_args(char *path, int mode)
1980  *
1981  * Create a named pipe.
1982  */
1983 int
1984 sys_mkfifo(struct mkfifo_args *uap)
1985 {
1986         struct nlookupdata nd;
1987         int error;
1988
1989         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1990         if (error == 0)
1991                 error = kern_mkfifo(&nd, uap->mode);
1992         nlookup_done(&nd);
1993         return (error);
1994 }
1995
1996 static int hardlink_check_uid = 0;
1997 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1998     &hardlink_check_uid, 0,
1999     "Unprivileged processes cannot create hard links to files owned by other "
2000     "users");
2001 static int hardlink_check_gid = 0;
2002 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2003     &hardlink_check_gid, 0,
2004     "Unprivileged processes cannot create hard links to files owned by other "
2005     "groups");
2006
2007 static int
2008 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2009 {
2010         struct vattr va;
2011         int error;
2012
2013         /*
2014          * Shortcut if disabled
2015          */
2016         if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2017                 return (0);
2018
2019         /*
2020          * Privileged user can always hardlink
2021          */
2022         if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2023                 return (0);
2024
2025         /*
2026          * Otherwise only if the originating file is owned by the
2027          * same user or group.  Note that any group is allowed if
2028          * the file is owned by the caller.
2029          */
2030         error = VOP_GETATTR(vp, &va);
2031         if (error != 0)
2032                 return (error);
2033
2034         if (hardlink_check_uid) {
2035                 if (cred->cr_uid != va.va_uid)
2036                         return (EPERM);
2037         }
2038
2039         if (hardlink_check_gid) {
2040                 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2041                         return (EPERM);
2042         }
2043
2044         return (0);
2045 }
2046
2047 int
2048 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2049 {
2050         struct thread *td = curthread;
2051         struct vnode *vp;
2052         int error;
2053
2054         /*
2055          * Lookup the source and obtained a locked vnode.
2056          *
2057          * You may only hardlink a file which you have write permission
2058          * on or which you own.
2059          *
2060          * XXX relookup on vget failure / race ?
2061          */
2062         bwillinode(1);
2063         nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2064         if ((error = nlookup(nd)) != 0)
2065                 return (error);
2066         vp = nd->nl_nch.ncp->nc_vp;
2067         KKASSERT(vp != NULL);
2068         if (vp->v_type == VDIR)
2069                 return (EPERM);         /* POSIX */
2070         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2071                 return (error);
2072         if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2073                 return (error);
2074
2075         /*
2076          * Unlock the source so we can lookup the target without deadlocking
2077          * (XXX vp is locked already, possible other deadlock?).  The target
2078          * must not exist.
2079          */
2080         KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2081         nd->nl_flags &= ~NLC_NCPISLOCKED;
2082         cache_unlock(&nd->nl_nch);
2083
2084         linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2085         if ((error = nlookup(linknd)) != 0) {
2086                 vput(vp);
2087                 return (error);
2088         }
2089         if (linknd->nl_nch.ncp->nc_vp) {
2090                 vput(vp);
2091                 return (EEXIST);
2092         }
2093
2094         /*
2095          * Finally run the new API VOP.
2096          */
2097         error = can_hardlink(vp, td, td->td_proc->p_ucred);
2098         if (error == 0) {
2099                 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2100                                   vp, linknd->nl_cred);
2101         }
2102         vput(vp);
2103         return (error);
2104 }
2105
2106 /*
2107  * link_args(char *path, char *link)
2108  *
2109  * Make a hard file link.
2110  */
2111 int
2112 sys_link(struct link_args *uap)
2113 {
2114         struct nlookupdata nd, linknd;
2115         int error;
2116
2117         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2118         if (error == 0) {
2119                 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2120                 if (error == 0)
2121                         error = kern_link(&nd, &linknd);
2122                 nlookup_done(&linknd);
2123         }
2124         nlookup_done(&nd);
2125         return (error);
2126 }
2127
2128 int
2129 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2130 {
2131         struct vattr vattr;
2132         struct vnode *vp;
2133         struct vnode *dvp;
2134         int error;
2135
2136         bwillinode(1);
2137         nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2138         if ((error = nlookup(nd)) != 0)
2139                 return (error);
2140         if (nd->nl_nch.ncp->nc_vp)
2141                 return (EEXIST);
2142         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2143                 return (error);
2144         dvp = nd->nl_dvp;
2145         VATTR_NULL(&vattr);
2146         vattr.va_mode = mode;
2147         error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2148         if (error == 0)
2149                 vput(vp);
2150         return (error);
2151 }
2152
2153 /*
2154  * symlink(char *path, char *link)
2155  *
2156  * Make a symbolic link.
2157  */
2158 int
2159 sys_symlink(struct symlink_args *uap)
2160 {
2161         struct thread *td = curthread;
2162         struct nlookupdata nd;
2163         char *path;
2164         int error;
2165         int mode;
2166
2167         path = objcache_get(namei_oc, M_WAITOK);
2168         error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2169         if (error == 0) {
2170                 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2171                 if (error == 0) {
2172                         mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2173                         error = kern_symlink(&nd, path, mode);
2174                 }
2175                 nlookup_done(&nd);
2176         }
2177         objcache_put(namei_oc, path);
2178         return (error);
2179 }
2180
2181 /*
2182  * undelete_args(char *path)
2183  *
2184  * Delete a whiteout from the filesystem.
2185  */
2186 /* ARGSUSED */
2187 int
2188 sys_undelete(struct undelete_args *uap)
2189 {
2190         struct nlookupdata nd;
2191         int error;
2192
2193         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2194         bwillinode(1);
2195         nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2196         if (error == 0)
2197                 error = nlookup(&nd);
2198         if (error == 0)
2199                 error = ncp_writechk(&nd.nl_nch);
2200         if (error == 0) {
2201                 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2202                                       NAMEI_DELETE);
2203         }
2204         nlookup_done(&nd);
2205         return (error);
2206 }
2207
2208 int
2209 kern_unlink(struct nlookupdata *nd)
2210 {
2211         int error;
2212
2213         bwillinode(1);
2214         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2215         if ((error = nlookup(nd)) != 0)
2216                 return (error);
2217         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2218                 return (error);
2219         error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2220         return (error);
2221 }
2222
2223 /*
2224  * unlink_args(char *path)
2225  *
2226  * Delete a name from the filesystem.
2227  */
2228 int
2229 sys_unlink(struct unlink_args *uap)
2230 {
2231         struct nlookupdata nd;
2232         int error;
2233
2234         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2235         if (error == 0)
2236                 error = kern_unlink(&nd);
2237         nlookup_done(&nd);
2238         return (error);
2239 }
2240
2241
2242 /*
2243  * unlinkat_args(int fd, char *path, int flags)
2244  *
2245  * Delete the file or directory entry pointed to by fd/path.
2246  */
2247 int
2248 sys_unlinkat(struct unlinkat_args *uap)
2249 {
2250         struct nlookupdata nd;
2251         struct file *fp;
2252         int error;
2253
2254         if (uap->flags & ~AT_REMOVEDIR)
2255                 return (EINVAL);
2256
2257         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2258         if (error == 0) {
2259                 if (uap->flags & AT_REMOVEDIR)
2260                         error = kern_rmdir(&nd);
2261                 else
2262                         error = kern_unlink(&nd);
2263         }
2264         nlookup_done_at(&nd, fp);
2265         return (error);
2266 }
2267
2268 int
2269 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2270 {
2271         struct thread *td = curthread;
2272         struct proc *p = td->td_proc;
2273         struct file *fp;
2274         struct vnode *vp;
2275         struct vattr vattr;
2276         off_t new_offset;
2277         int error;
2278
2279         fp = holdfp(p->p_fd, fd, -1);
2280         if (fp == NULL)
2281                 return (EBADF);
2282         if (fp->f_type != DTYPE_VNODE) {
2283                 error = ESPIPE;
2284                 goto done;
2285         }
2286         vp = (struct vnode *)fp->f_data;
2287
2288         switch (whence) {
2289         case L_INCR:
2290                 new_offset = fp->f_offset + offset;
2291                 error = 0;
2292                 break;
2293         case L_XTND:
2294                 error = VOP_GETATTR(vp, &vattr);
2295                 new_offset = offset + vattr.va_size;
2296                 break;
2297         case L_SET:
2298                 new_offset = offset;
2299                 error = 0;
2300                 break;
2301         default:
2302                 new_offset = 0;
2303                 error = EINVAL;
2304                 break;
2305         }
2306
2307         /*
2308          * Validate the seek position.  Negative offsets are not allowed
2309          * for regular files or directories.
2310          *
2311          * Normally we would also not want to allow negative offsets for
2312          * character and block-special devices.  However kvm addresses
2313          * on 64 bit architectures might appear to be negative and must
2314          * be allowed.
2315          */
2316         if (error == 0) {
2317                 if (new_offset < 0 &&
2318                     (vp->v_type == VREG || vp->v_type == VDIR)) {
2319                         error = EINVAL;
2320                 } else {
2321                         fp->f_offset = new_offset;
2322                 }
2323         }
2324         *res = fp->f_offset;
2325 done:
2326         fdrop(fp);
2327         return (error);
2328 }
2329
2330 /*
2331  * lseek_args(int fd, int pad, off_t offset, int whence)
2332  *
2333  * Reposition read/write file offset.
2334  */
2335 int
2336 sys_lseek(struct lseek_args *uap)
2337 {
2338         int error;
2339
2340         error = kern_lseek(uap->fd, uap->offset, uap->whence,
2341                            &uap->sysmsg_offset);
2342
2343         return (error);
2344 }
2345
2346 /*
2347  * Check if current process can access given file.  amode is a bitmask of *_OK
2348  * access bits.  flags is a bitmask of AT_* flags.
2349  */
2350 int
2351 kern_access(struct nlookupdata *nd, int amode, int flags)
2352 {
2353         struct vnode *vp;
2354         int error, mode;
2355
2356         if (flags & ~AT_EACCESS)
2357                 return (EINVAL);
2358         if ((error = nlookup(nd)) != 0)
2359                 return (error);
2360 retry:
2361         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2362         if (error)
2363                 return (error);
2364
2365         /* Flags == 0 means only check for existence. */
2366         if (amode) {
2367                 mode = 0;
2368                 if (amode & R_OK)
2369                         mode |= VREAD;
2370                 if (amode & W_OK)
2371                         mode |= VWRITE;
2372                 if (amode & X_OK)
2373                         mode |= VEXEC;
2374                 if ((mode & VWRITE) == 0 ||
2375                     (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2376                         error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2377
2378                 /*
2379                  * If the file handle is stale we have to re-resolve the
2380                  * entry.  This is a hack at the moment.
2381                  */
2382                 if (error == ESTALE) {
2383                         vput(vp);
2384                         cache_setunresolved(&nd->nl_nch);
2385                         error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2386                         if (error == 0) {
2387                                 vp = NULL;
2388                                 goto retry;
2389                         }
2390                         return(error);
2391                 }
2392         }
2393         vput(vp);
2394         return (error);
2395 }
2396
2397 /*
2398  * access_args(char *path, int flags)
2399  *
2400  * Check access permissions.
2401  */
2402 int
2403 sys_access(struct access_args *uap)
2404 {
2405         struct nlookupdata nd;
2406         int error;
2407
2408         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2409         if (error == 0)
2410                 error = kern_access(&nd, uap->flags, 0);
2411         nlookup_done(&nd);
2412         return (error);
2413 }
2414
2415
2416 /*
2417  * faccessat_args(int fd, char *path, int amode, int flags)
2418  *
2419  * Check access permissions.
2420  */
2421 int
2422 sys_faccessat(struct faccessat_args *uap)
2423 {
2424         struct nlookupdata nd;
2425         struct file *fp;
2426         int error;
2427
2428         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2429                                 NLC_FOLLOW);
2430         if (error == 0)
2431                 error = kern_access(&nd, uap->amode, uap->flags);
2432         nlookup_done_at(&nd, fp);
2433         return (error);
2434 }
2435
2436
2437 int
2438 kern_stat(struct nlookupdata *nd, struct stat *st)
2439 {
2440         int error;
2441         struct vnode *vp;
2442         thread_t td;
2443
2444         if ((error = nlookup(nd)) != 0)
2445                 return (error);
2446 again:
2447         if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2448                 return (ENOENT);
2449
2450         td = curthread;
2451         if ((error = vget(vp, LK_SHARED)) != 0)
2452                 return (error);
2453         error = vn_stat(vp, st, nd->nl_cred);
2454
2455         /*
2456          * If the file handle is stale we have to re-resolve the entry.  This
2457          * is a hack at the moment.
2458          */
2459         if (error == ESTALE) {
2460                 vput(vp);
2461                 cache_setunresolved(&nd->nl_nch);
2462                 error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2463                 if (error == 0)
2464                         goto again;
2465         } else {
2466                 vput(vp);
2467         }
2468         return (error);
2469 }
2470
2471 /*
2472  * stat_args(char *path, struct stat *ub)
2473  *
2474  * Get file status; this version follows links.
2475  */
2476 int
2477 sys_stat(struct stat_args *uap)
2478 {
2479         struct nlookupdata nd;
2480         struct stat st;
2481         int error;
2482
2483         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2484         if (error == 0) {
2485                 error = kern_stat(&nd, &st);
2486                 if (error == 0)
2487                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2488         }
2489         nlookup_done(&nd);
2490         return (error);
2491 }
2492
2493 /*
2494  * lstat_args(char *path, struct stat *ub)
2495  *
2496  * Get file status; this version does not follow links.
2497  */
2498 int
2499 sys_lstat(struct lstat_args *uap)
2500 {
2501         struct nlookupdata nd;
2502         struct stat st;
2503         int error;
2504
2505         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2506         if (error == 0) {
2507                 error = kern_stat(&nd, &st);
2508                 if (error == 0)
2509                         error = copyout(&st, uap->ub, sizeof(*uap->ub));
2510         }
2511         nlookup_done(&nd);
2512         return (error);
2513 }
2514
2515 /*
2516  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2517  *
2518  * Get status of file pointed to by fd/path.
2519  */
2520 int
2521 sys_fstatat(struct fstatat_args *uap)
2522 {
2523         struct nlookupdata nd;
2524         struct stat st;
2525         int error;
2526         int flags;
2527         struct file *fp;
2528
2529         if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2530                 return (EINVAL);
2531
2532         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2533
2534         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2535                                 UIO_USERSPACE, flags);
2536         if (error == 0) {
2537                 error = kern_stat(&nd, &st);
2538                 if (error == 0)
2539                         error = copyout(&st, uap->sb, sizeof(*uap->sb));
2540         }
2541         nlookup_done_at(&nd, fp);
2542         return (error);
2543 }
2544
2545 /*
2546  * pathconf_Args(char *path, int name)
2547  *
2548  * Get configurable pathname variables.
2549  */
2550 /* ARGSUSED */
2551 int
2552 sys_pathconf(struct pathconf_args *uap)
2553 {
2554         struct nlookupdata nd;
2555         struct vnode *vp;
2556         int error;
2557
2558         vp = NULL;
2559         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2560         if (error == 0)
2561                 error = nlookup(&nd);
2562         if (error == 0)
2563                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2564         nlookup_done(&nd);
2565         if (error == 0) {
2566                 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2567                 vput(vp);
2568         }
2569         return (error);
2570 }
2571
2572 /*
2573  * XXX: daver
2574  * kern_readlink isn't properly split yet.  There is a copyin burried
2575  * in VOP_READLINK().
2576  */
2577 int
2578 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2579 {
2580         struct thread *td = curthread;
2581         struct proc *p = td->td_proc;
2582         struct vnode *vp;
2583         struct iovec aiov;
2584         struct uio auio;
2585         int error;
2586
2587         if ((error = nlookup(nd)) != 0)
2588                 return (error);
2589         error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2590         if (error)
2591                 return (error);
2592         if (vp->v_type != VLNK) {
2593                 error = EINVAL;
2594         } else {
2595                 aiov.iov_base = buf;
2596                 aiov.iov_len = count;
2597                 auio.uio_iov = &aiov;
2598                 auio.uio_iovcnt = 1;
2599                 auio.uio_offset = 0;
2600                 auio.uio_rw = UIO_READ;
2601                 auio.uio_segflg = UIO_USERSPACE;
2602                 auio.uio_td = td;
2603                 auio.uio_resid = count;
2604                 error = VOP_READLINK(vp, &auio, p->p_ucred);
2605         }
2606         vput(vp);
2607         *res = count - auio.uio_resid;
2608         return (error);
2609 }
2610
2611 /*
2612  * readlink_args(char *path, char *buf, int count)
2613  *
2614  * Return target name of a symbolic link.
2615  */
2616 int
2617 sys_readlink(struct readlink_args *uap)
2618 {
2619         struct nlookupdata nd;
2620         int error;
2621
2622         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2623         if (error == 0) {
2624                 error = kern_readlink(&nd, uap->buf, uap->count,
2625                                         &uap->sysmsg_result);
2626         }
2627         nlookup_done(&nd);
2628         return (error);
2629 }
2630
2631 static int
2632 setfflags(struct vnode *vp, int flags)
2633 {
2634         struct thread *td = curthread;
2635         struct proc *p = td->td_proc;
2636         int error;
2637         struct vattr vattr;
2638
2639         /*
2640          * Prevent non-root users from setting flags on devices.  When
2641          * a device is reused, users can retain ownership of the device
2642          * if they are allowed to set flags and programs assume that
2643          * chown can't fail when done as root.
2644          */
2645         if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2646             ((error = priv_check_cred(p->p_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2647                 return (error);
2648
2649         /*
2650          * note: vget is required for any operation that might mod the vnode
2651          * so VINACTIVE is properly cleared.
2652          */
2653         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2654                 VATTR_NULL(&vattr);
2655                 vattr.va_flags = flags;
2656                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2657                 vput(vp);
2658         }
2659         return (error);
2660 }
2661
2662 /*
2663  * chflags(char *path, int flags)
2664  *
2665  * Change flags of a file given a path name.
2666  */
2667 /* ARGSUSED */
2668 int
2669 sys_chflags(struct chflags_args *uap)
2670 {
2671         struct nlookupdata nd;
2672         struct vnode *vp;
2673         int error;
2674
2675         vp = NULL;
2676         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2677         if (error == 0)
2678                 error = nlookup(&nd);
2679         if (error == 0)
2680                 error = ncp_writechk(&nd.nl_nch);
2681         if (error == 0)
2682                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2683         nlookup_done(&nd);
2684         if (error == 0) {
2685                 error = setfflags(vp, uap->flags);
2686                 vrele(vp);
2687         }
2688         return (error);
2689 }
2690
2691 /*
2692  * lchflags(char *path, int flags)
2693  *
2694  * Change flags of a file given a path name, but don't follow symlinks.
2695  */
2696 /* ARGSUSED */
2697 int
2698 sys_lchflags(struct lchflags_args *uap)
2699 {
2700         struct nlookupdata nd;
2701         struct vnode *vp;
2702         int error;
2703
2704         vp = NULL;
2705         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2706         if (error == 0)
2707                 error = nlookup(&nd);
2708         if (error == 0)
2709                 error = ncp_writechk(&nd.nl_nch);
2710         if (error == 0)
2711                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2712         nlookup_done(&nd);
2713         if (error == 0) {
2714                 error = setfflags(vp, uap->flags);
2715                 vrele(vp);
2716         }
2717         return (error);
2718 }
2719
2720 /*
2721  * fchflags_args(int fd, int flags)
2722  *
2723  * Change flags of a file given a file descriptor.
2724  */
2725 /* ARGSUSED */
2726 int
2727 sys_fchflags(struct fchflags_args *uap)
2728 {
2729         struct thread *td = curthread;
2730         struct proc *p = td->td_proc;
2731         struct file *fp;
2732         int error;
2733
2734         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2735                 return (error);
2736         if (fp->f_nchandle.ncp)
2737                 error = ncp_writechk(&fp->f_nchandle);
2738         if (error == 0)
2739                 error = setfflags((struct vnode *) fp->f_data, uap->flags);
2740         fdrop(fp);
2741         return (error);
2742 }
2743
2744 static int
2745 setfmode(struct vnode *vp, int mode)
2746 {
2747         struct thread *td = curthread;
2748         struct proc *p = td->td_proc;
2749         int error;
2750         struct vattr vattr;
2751
2752         /*
2753          * note: vget is required for any operation that might mod the vnode
2754          * so VINACTIVE is properly cleared.
2755          */
2756         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2757                 VATTR_NULL(&vattr);
2758                 vattr.va_mode = mode & ALLPERMS;
2759                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2760                 vput(vp);
2761         }
2762         return error;
2763 }
2764
2765 int
2766 kern_chmod(struct nlookupdata *nd, int mode)
2767 {
2768         struct vnode *vp;
2769         int error;
2770
2771         if ((error = nlookup(nd)) != 0)
2772                 return (error);
2773         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2774                 return (error);
2775         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2776                 error = setfmode(vp, mode);
2777         vrele(vp);
2778         return (error);
2779 }
2780
2781 /*
2782  * chmod_args(char *path, int mode)
2783  *
2784  * Change mode of a file given path name.
2785  */
2786 /* ARGSUSED */
2787 int
2788 sys_chmod(struct chmod_args *uap)
2789 {
2790         struct nlookupdata nd;
2791         int error;
2792
2793         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2794         if (error == 0)
2795                 error = kern_chmod(&nd, uap->mode);
2796         nlookup_done(&nd);
2797         return (error);
2798 }
2799
2800 /*
2801  * lchmod_args(char *path, int mode)
2802  *
2803  * Change mode of a file given path name (don't follow links.)
2804  */
2805 /* ARGSUSED */
2806 int
2807 sys_lchmod(struct lchmod_args *uap)
2808 {
2809         struct nlookupdata nd;
2810         int error;
2811
2812         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2813         if (error == 0)
2814                 error = kern_chmod(&nd, uap->mode);
2815         nlookup_done(&nd);
2816         return (error);
2817 }
2818
2819 /*
2820  * fchmod_args(int fd, int mode)
2821  *
2822  * Change mode of a file given a file descriptor.
2823  */
2824 /* ARGSUSED */
2825 int
2826 sys_fchmod(struct fchmod_args *uap)
2827 {
2828         struct thread *td = curthread;
2829         struct proc *p = td->td_proc;
2830         struct file *fp;
2831         int error;
2832
2833         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2834                 return (error);
2835         if (fp->f_nchandle.ncp)
2836                 error = ncp_writechk(&fp->f_nchandle);
2837         if (error == 0)
2838                 error = setfmode((struct vnode *)fp->f_data, uap->mode);
2839         fdrop(fp);
2840         return (error);
2841 }
2842
2843 /*
2844  * fchmodat_args(char *path, int mode)
2845  *
2846  * Change mode of a file pointed to by fd/path.
2847  */
2848 int
2849 sys_fchmodat(struct fchmodat_args *uap)
2850 {
2851         struct nlookupdata nd;
2852         struct file *fp;
2853         int error;
2854         int flags;
2855
2856         if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2857                 return (EINVAL);
2858         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2859
2860         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2861                                 UIO_USERSPACE, flags);
2862         if (error == 0)
2863                 error = kern_chmod(&nd, uap->mode);
2864         nlookup_done_at(&nd, fp);
2865         return (error);
2866 }
2867
2868 static int
2869 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2870 {
2871         struct thread *td = curthread;
2872         struct proc *p = td->td_proc;
2873         int error;
2874         struct vattr vattr;
2875
2876         /*
2877          * note: vget is required for any operation that might mod the vnode
2878          * so VINACTIVE is properly cleared.
2879          */
2880         if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2881                 VATTR_NULL(&vattr);
2882                 vattr.va_uid = uid;
2883                 vattr.va_gid = gid;
2884                 error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2885                 vput(vp);
2886         }
2887         return error;
2888 }
2889
2890 int
2891 kern_chown(struct nlookupdata *nd, int uid, int gid)
2892 {
2893         struct vnode *vp;
2894         int error;
2895
2896         if ((error = nlookup(nd)) != 0)
2897                 return (error);
2898         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2899                 return (error);
2900         if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2901                 error = setfown(vp, uid, gid);
2902         vrele(vp);
2903         return (error);
2904 }
2905
2906 /*
2907  * chown(char *path, int uid, int gid)
2908  *
2909  * Set ownership given a path name.
2910  */
2911 int
2912 sys_chown(struct chown_args *uap)
2913 {
2914         struct nlookupdata nd;
2915         int error;
2916
2917         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2918         if (error == 0)
2919                 error = kern_chown(&nd, uap->uid, uap->gid);
2920         nlookup_done(&nd);
2921         return (error);
2922 }
2923
2924 /*
2925  * lchown_args(char *path, int uid, int gid)
2926  *
2927  * Set ownership given a path name, do not cross symlinks.
2928  */
2929 int
2930 sys_lchown(struct lchown_args *uap)
2931 {
2932         struct nlookupdata nd;
2933         int error;
2934
2935         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2936         if (error == 0)
2937                 error = kern_chown(&nd, uap->uid, uap->gid);
2938         nlookup_done(&nd);
2939         return (error);
2940 }
2941
2942 /*
2943  * fchown_args(int fd, int uid, int gid)
2944  *
2945  * Set ownership given a file descriptor.
2946  */
2947 /* ARGSUSED */
2948 int
2949 sys_fchown(struct fchown_args *uap)
2950 {
2951         struct thread *td = curthread;
2952         struct proc *p = td->td_proc;
2953         struct file *fp;
2954         int error;
2955
2956         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2957                 return (error);
2958         if (fp->f_nchandle.ncp)
2959                 error = ncp_writechk(&fp->f_nchandle);
2960         if (error == 0)
2961                 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2962         fdrop(fp);
2963         return (error);
2964 }
2965
2966 /*
2967  * fchownat(int fd, char *path, int uid, int gid, int flags)
2968  *
2969  * Set ownership of file pointed to by fd/path.
2970  */
2971 int
2972 sys_fchownat(struct fchownat_args *uap)
2973 {
2974         struct nlookupdata nd;
2975         struct file *fp;
2976         int error;
2977         int flags;
2978
2979         if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2980                 return (EINVAL);
2981         flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2982
2983         error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2984                                 UIO_USERSPACE, flags);
2985         if (error == 0)
2986                 error = kern_chown(&nd, uap->uid, uap->gid);
2987         nlookup_done_at(&nd, fp);
2988         return (error);
2989 }
2990
2991
2992 static int
2993 getutimes(const struct timeval *tvp, struct timespec *tsp)
2994 {
2995         struct timeval tv[2];
2996
2997         if (tvp == NULL) {
2998                 microtime(&tv[0]);
2999                 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3000                 tsp[1] = tsp[0];
3001         } else {
3002                 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3003                 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3004         }
3005         return 0;
3006 }
3007
3008 static int
3009 setutimes(struct vnode *vp, struct vattr *vattr,
3010           const struct timespec *ts, int nullflag)
3011 {
3012         struct thread *td = curthread;
3013         struct proc *p = td->td_proc;
3014         int error;
3015
3016         VATTR_NULL(vattr);
3017         vattr->va_atime = ts[0];
3018         vattr->va_mtime = ts[1];
3019         if (nullflag)
3020                 vattr->va_vaflags |= VA_UTIMES_NULL;
3021         error = VOP_SETATTR(vp, vattr, p->p_ucred);
3022
3023         return error;
3024 }
3025
3026 int
3027 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3028 {
3029         struct timespec ts[2];
3030         struct vnode *vp;
3031         struct vattr vattr;
3032         int error;
3033
3034         if ((error = getutimes(tptr, ts)) != 0)
3035                 return (error);
3036
3037         /*
3038          * NOTE: utimes() succeeds for the owner even if the file
3039          * is not user-writable.
3040          */
3041         nd->nl_flags |= NLC_OWN | NLC_WRITE;
3042
3043         if ((error = nlookup(nd)) != 0)
3044                 return (error);
3045         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3046                 return (error);
3047         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3048                 return (error);
3049
3050         /*
3051          * note: vget is required for any operation that might mod the vnode
3052          * so VINACTIVE is properly cleared.
3053          */
3054         if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3055                 error = vget(vp, LK_EXCLUSIVE);
3056                 if (error == 0) {
3057                         error = setutimes(vp, &vattr, ts, (tptr == NULL));
3058                         vput(vp);
3059                 }
3060         }
3061         vrele(vp);
3062         return (error);
3063 }
3064
3065 /*
3066  * utimes_args(char *path, struct timeval *tptr)
3067  *
3068  * Set the access and modification times of a file.
3069  */
3070 int
3071 sys_utimes(struct utimes_args *uap)
3072 {
3073         struct timeval tv[2];
3074         struct nlookupdata nd;
3075         int error;
3076
3077         if (uap->tptr) {
3078                 error = copyin(uap->tptr, tv, sizeof(tv));
3079                 if (error)
3080                         return (error);
3081         }
3082         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3083         if (error == 0)
3084                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3085         nlookup_done(&nd);
3086         return (error);
3087 }
3088
3089 /*
3090  * lutimes_args(char *path, struct timeval *tptr)
3091  *
3092  * Set the access and modification times of a file.
3093  */
3094 int
3095 sys_lutimes(struct lutimes_args *uap)
3096 {
3097         struct timeval tv[2];
3098         struct nlookupdata nd;
3099         int error;
3100
3101         if (uap->tptr) {
3102                 error = copyin(uap->tptr, tv, sizeof(tv));
3103                 if (error)
3104                         return (error);
3105         }
3106         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3107         if (error == 0)
3108                 error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3109         nlookup_done(&nd);
3110         return (error);
3111 }
3112
3113 /*
3114  * Set utimes on a file descriptor.  The creds used to open the
3115  * file are used to determine whether the operation is allowed
3116  * or not.
3117  */
3118 int
3119 kern_futimes(int fd, struct timeval *tptr)
3120 {
3121         struct thread *td = curthread;
3122         struct proc *p = td->td_proc;
3123         struct timespec ts[2];
3124         struct file *fp;
3125         struct vnode *vp;
3126         struct vattr vattr;
3127         int error;
3128
3129         error = getutimes(tptr, ts);
3130         if (error)
3131                 return (error);
3132         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3133                 return (error);
3134         if (fp->f_nchandle.ncp)
3135                 error = ncp_writechk(&fp->f_nchandle);
3136         if (error == 0) {
3137                 vp = fp->f_data;
3138                 error = vget(vp, LK_EXCLUSIVE);
3139                 if (error == 0) {
3140                         error = VOP_GETATTR(vp, &vattr);
3141                         if (error == 0) {
3142                                 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3143                                                    fp->f_cred);
3144                         }
3145                         if (error == 0) {
3146                                 error = setutimes(vp, &vattr, ts,
3147                                                   (tptr == NULL));
3148                         }
3149                         vput(vp);
3150                 }
3151         }
3152         fdrop(fp);
3153         return (error);
3154 }
3155
3156 /*
3157  * futimes_args(int fd, struct timeval *tptr)
3158  *
3159  * Set the access and modification times of a file.
3160  */
3161 int
3162 sys_futimes(struct futimes_args *uap)
3163 {
3164         struct timeval tv[2];
3165         int error;
3166
3167         if (uap->tptr) {
3168                 error = copyin(uap->tptr, tv, sizeof(tv));
3169                 if (error)
3170                         return (error);
3171         }
3172
3173         error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3174
3175         return (error);
3176 }
3177
3178 int
3179 kern_truncate(struct nlookupdata *nd, off_t length)
3180 {
3181         struct vnode *vp;
3182         struct vattr vattr;
3183         int error;
3184
3185         if (length < 0)
3186                 return(EINVAL);
3187         nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3188         if ((error = nlookup(nd)) != 0)
3189                 return (error);
3190         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3191                 return (error);
3192         if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3193                 return (error);
3194         if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3195                 vrele(vp);
3196                 return (error);
3197         }
3198         if (vp->v_type == VDIR) {
3199                 error = EISDIR;
3200         } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3201                 VATTR_NULL(&vattr);
3202                 vattr.va_size = length;
3203                 error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3204         }
3205         vput(vp);
3206         return (error);
3207 }
3208
3209 /*
3210  * truncate(char *path, int pad, off_t length)
3211  *
3212  * Truncate a file given its path name.
3213  */
3214 int
3215 sys_truncate(struct truncate_args *uap)
3216 {
3217         struct nlookupdata nd;
3218         int error;
3219
3220         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3221         if (error == 0)
3222                 error = kern_truncate(&nd, uap->length);
3223         nlookup_done(&nd);
3224         return error;
3225 }
3226
3227 int
3228 kern_ftruncate(int fd, off_t length)
3229 {
3230         struct thread *td = curthread;
3231         struct proc *p = td->td_proc;
3232         struct vattr vattr;
3233         struct vnode *vp;
3234         struct file *fp;
3235         int error;
3236
3237         if (length < 0)
3238                 return(EINVAL);
3239         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3240                 return (error);
3241         if (fp->f_nchandle.ncp) {
3242                 error = ncp_writechk(&fp->f_nchandle);
3243                 if (error)
3244                         goto done;
3245         }
3246         if ((fp->f_flag & FWRITE) == 0) {
3247                 error = EINVAL;
3248                 goto done;
3249         }
3250         if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */
3251                 error = EINVAL;
3252                 goto done;
3253         }
3254         vp = (struct vnode *)fp->f_data;
3255         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3256         if (vp->v_type == VDIR) {
3257                 error = EISDIR;
3258         } else if ((error = vn_writechk(vp, NULL)) == 0) {
3259                 VATTR_NULL(&vattr);
3260                 vattr.va_size = length;
3261                 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3262         }
3263         vn_unlock(vp);
3264 done:
3265         fdrop(fp);
3266         return (error);
3267 }
3268
3269 /*
3270  * ftruncate_args(int fd, int pad, off_t length)
3271  *
3272  * Truncate a file given a file descriptor.
3273  */
3274 int
3275 sys_ftruncate(struct ftruncate_args *uap)
3276 {
3277         int error;
3278
3279         error = kern_ftruncate(uap->fd, uap->length);
3280
3281         return (error);
3282 }
3283
3284 /*
3285  * fsync(int fd)
3286  *
3287  * Sync an open file.
3288  */
3289 /* ARGSUSED */
3290 int
3291 sys_fsync(struct fsync_args *uap)
3292 {
3293         struct thread *td = curthread;
3294         struct proc *p = td->td_proc;
3295         struct vnode *vp;
3296         struct file *fp;
3297         vm_object_t obj;
3298         int error;
3299
3300         if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3301                 return (error);
3302         vp = (struct vnode *)fp->f_data;
3303         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3304         if ((obj = vp->v_object) != NULL)
3305                 vm_object_page_clean(obj, 0, 0, 0);
3306         error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3307         if (error == 0 && vp->v_mount)
3308                 error = buf_fsync(vp);
3309         vn_unlock(vp);
3310         fdrop(fp);
3311         return (error);
3312 }
3313
3314 int
3315 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3316 {
3317         struct nchandle fnchd;
3318         struct nchandle tnchd;
3319         struct namecache *ncp;
3320         struct vnode *fdvp;
3321         struct vnode *tdvp;
3322         struct mount *mp;
3323         int error;
3324
3325         bwillinode(1);
3326         fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3327         if ((error = nlookup(fromnd)) != 0)
3328                 return (error);
3329         if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3330                 return (ENOENT);
3331         fnchd.mount = fromnd->nl_nch.mount;
3332         cache_hold(&fnchd);
3333
3334         /*
3335          * unlock the source nch so we can lookup the target nch without
3336          * deadlocking.  The target may or may not exist so we do not check
3337          * for a target vp like kern_mkdir() and other creation functions do.
3338          *
3339          * The source and target directories are ref'd and rechecked after
3340          * everything is relocked to determine if the source or target file
3341          * has been renamed.
3342          */
3343         KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3344         fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3345         cache_unlock(&fromnd->nl_nch);
3346
3347         tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3348         if ((error = nlookup(tond)) != 0) {
3349                 cache_drop(&fnchd);
3350                 return (error);
3351         }
3352         if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3353                 cache_drop(&fnchd);
3354                 return (ENOENT);
3355         }
3356         tnchd.mount = tond->nl_nch.mount;
3357         cache_hold(&tnchd);
3358
3359         /*
3360          * If the source and target are the same there is nothing to do
3361          */
3362         if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3363                 cache_drop(&fnchd);
3364                 cache_drop(&tnchd);
3365                 return (0);
3366         }
3367
3368         /*
3369          * Mount points cannot be renamed or overwritten
3370          */
3371         if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3372             NCF_ISMOUNTPT
3373         ) {
3374                 cache_drop(&fnchd);
3375                 cache_drop(&tnchd);
3376                 return (EINVAL);
3377         }
3378
3379         /*
3380          * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
3381          * may have become invalid while it was unlocked, nc_vp and nc_mount
3382          * could be NULL.
3383          */
3384         if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3385                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3386         } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3387                 cache_lock(&fromnd->nl_nch);
3388                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3389         } else {
3390                 cache_unlock(&tond->nl_nch);
3391                 cache_lock(&fromnd->nl_nch);
3392                 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3393                 cache_lock(&tond->nl_nch);
3394                 cache_resolve(&tond->nl_nch, tond->nl_cred);
3395         }
3396         fromnd->nl_flags |= NLC_NCPISLOCKED;
3397
3398         /*
3399          * make sure the parent directories linkages are the same
3400          */
3401         if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3402             tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3403                 cache_drop(&fnchd);
3404                 cache_drop(&tnchd);
3405                 return (ENOENT);
3406         }
3407
3408         /*
3409          * Both the source and target must be within the same filesystem and
3410          * in the same filesystem as their parent directories within the
3411          * namecache topology.
3412          *
3413          * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3414          */
3415         mp = fnchd.mount;
3416         if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3417             mp != tond->nl_nch.mount) {
3418                 cache_drop(&fnchd);
3419                 cache_drop(&tnchd);
3420                 return (EXDEV);
3421         }
3422
3423         /*
3424          * Make sure the mount point is writable
3425          */
3426         if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3427                 cache_drop(&fnchd);
3428                 cache_drop(&tnchd);
3429                 return (error);
3430         }
3431
3432         /*
3433          * If the target exists and either the source or target is a directory,
3434          * then both must be directories.
3435          *
3436          * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3437          * have become NULL.
3438          */
3439         if (tond->nl_nch.ncp->nc_vp) {
3440                 if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3441                         error = ENOENT;
3442                 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3443                         if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3444                                 error = ENOTDIR;
3445                 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3446                         error = EISDIR;
3447                 }
3448         }
3449
3450         /*
3451          * You cannot rename a source into itself or a subdirectory of itself.
3452          * We check this by travsersing the target directory upwards looking
3453          * for a match against the source.
3454          */
3455         if (error == 0) {
3456                 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3457                         if (fromnd->nl_nch.ncp == ncp) {
3458                                 error = EINVAL;
3459                                 break;
3460                         }
3461                 }
3462         }
3463
3464         cache_drop(&fnchd);
3465         cache_drop(&tnchd);
3466
3467         /*
3468          * Even though the namespaces are different, they may still represent
3469          * hardlinks to the same file.  The filesystem might have a hard time
3470          * with this so we issue a NREMOVE of the source instead of a NRENAME
3471          * when we detect the situation.
3472          */
3473         if (error == 0) {
3474                 fdvp = fromnd->nl_dvp;
3475                 tdvp = tond->nl_dvp;
3476                 if (fdvp == NULL || tdvp == NULL) {
3477                         error = EPERM;
3478                 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3479                         error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3480                                             fromnd->nl_cred);
3481                 } else {
3482                         error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3483                                             fdvp, tdvp, tond->nl_cred);
3484                 }
3485         }
3486         return (error);
3487 }
3488
3489 /*
3490  * rename_args(char *from, char *to)
3491  *
3492  * Rename files.  Source and destination must either both be directories,
3493  * or both not be directories.  If target is a directory, it must be empty.
3494  */
3495 int
3496 sys_rename(struct rename_args *uap)
3497 {
3498         struct nlookupdata fromnd, tond;
3499         int error;
3500
3501         error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3502         if (error == 0) {
3503                 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3504                 if (error == 0)
3505                         error = kern_rename(&fromnd, &tond);
3506                 nlookup_done(&tond);
3507         }
3508         nlookup_done(&fromnd);
3509         return (error);
3510 }
3511
3512 int
3513 kern_mkdir(struct nlookupdata *nd, int mode)
3514 {
3515         struct thread *td = curthread;
3516         struct proc *p = td->td_proc;
3517         struct vnode *vp;
3518         struct vattr vattr;
3519         int error;
3520
3521         bwillinode(1);
3522         nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3523         if ((error = nlookup(nd)) != 0)
3524                 return (error);
3525
3526         if (nd->nl_nch.ncp->nc_vp)
3527                 return (EEXIST);
3528         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3529                 return (error);
3530         VATTR_NULL(&vattr);
3531         vattr.va_type = VDIR;
3532         vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3533
3534         vp = NULL;
3535         error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3536         if (error == 0)
3537                 vput(vp);
3538         return (error);
3539 }
3540
3541 /*
3542  * mkdir_args(char *path, int mode)
3543  *
3544  * Make a directory file.
3545  */
3546 /* ARGSUSED */
3547 int
3548 sys_mkdir(struct mkdir_args *uap)
3549 {
3550         struct nlookupdata nd;
3551         int error;
3552
3553         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3554         if (error == 0)
3555                 error = kern_mkdir(&nd, uap->mode);
3556         nlookup_done(&nd);
3557         return (error);
3558 }
3559
3560 int
3561 kern_rmdir(struct nlookupdata *nd)
3562 {
3563         int error;
3564
3565         bwillinode(1);
3566         nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3567         if ((error = nlookup(nd)) != 0)
3568                 return (error);
3569
3570         /*
3571          * Do not allow directories representing mount points to be
3572          * deleted, even if empty.  Check write perms on mount point
3573          * in case the vnode is aliased (aka nullfs).
3574          */
3575         if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3576                 return (EINVAL);
3577         if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3578                 return (error);
3579         error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3580         return (error);
3581 }
3582
3583 /*
3584  * rmdir_args(char *path)
3585  *
3586  * Remove a directory file.
3587  */
3588 /* ARGSUSED */
3589 int
3590 sys_rmdir(struct rmdir_args *uap)
3591 {
3592         struct nlookupdata nd;
3593         int error;
3594
3595         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3596         if (error == 0)
3597                 error = kern_rmdir(&nd);
3598         nlookup_done(&nd);
3599         return (error);
3600 }
3601
3602 int
3603 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3604     enum uio_seg direction)
3605 {
3606         struct thread *td = curthread;
3607         struct proc *p = td->td_proc;
3608         struct vnode *vp;
3609         struct file *fp;
3610         struct uio auio;
3611         struct iovec aiov;
3612         off_t loff;
3613         int error, eofflag;
3614
3615         if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3616                 return (error);
3617         if ((fp->f_flag & FREAD) == 0) {
3618                 error = EBADF;
3619                 goto done;
3620         }
3621         vp = (struct vnode *)fp->f_data;
3622 unionread:
3623         if (vp->v_type != VDIR) {
3624                 error = EINVAL;
3625                 goto done;
3626         }
3627         aiov.iov_base = buf;
3628         aiov.iov_len = count;
3629         auio.uio_iov = &aiov;
3630         auio.uio_iovcnt = 1;
3631         auio.uio_rw = UIO_READ;
3632         auio.uio_segflg = direction;
3633         auio.uio_td = td;
3634         auio.uio_resid = count;
3635         loff = auio.uio_offset = fp->f_offset;
3636         error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3637         fp->f_offset = auio.uio_offset;
3638         if (error)
3639                 goto done;
3640         if (count == auio.uio_resid) {
3641                 if (union_dircheckp) {
3642                         error = union_dircheckp(td, &vp, fp);
3643                         if (error == -1)
3644                                 goto unionread;
3645                         if (error)
3646                                 goto done;
3647                 }
3648 #if 0
3649                 if ((vp->v_flag & VROOT) &&
3650                     (vp->v_mount->mnt_flag & MNT_UNION)) {
3651                         struct vnode *tvp = vp;
3652                         vp = vp->v_mount->mnt_vnodecovered;
3653                         vref(vp);
3654                         fp->f_data = vp;
3655                         fp->f_offset = 0;
3656                         vrele(tvp);
3657                         goto unionread;
3658                 }
3659 #endif
3660         }
3661
3662         /*
3663          * WARNING!  *basep may not be wide enough to accomodate the
3664          * seek offset.   XXX should we hack this to return the upper 32 bits
3665          * for offsets greater then 4G?
3666          */
3667         if (basep) {
3668                 *basep = (long)loff;
3669         }
3670         *res = count - auio.uio_resid;
3671 done:
3672         fdrop(fp);
3673         return (error);
3674 }
3675
3676 /*
3677  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3678  *
3679  * Read a block of directory entries in a file system independent format.
3680  */
3681 int
3682 sys_getdirentries(struct getdirentries_args *uap)
3683 {
3684         long base;
3685         int error;
3686
3687         error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3688                                    &uap->sysmsg_result, UIO_USERSPACE);
3689
3690         if (error == 0 && uap->basep)
3691                 error = copyout(&base, uap->basep, sizeof(*uap->basep));
3692         return (error);
3693 }
3694
3695 /*
3696  * getdents_args(int fd, char *buf, size_t count)
3697  */
3698 int
3699 sys_getdents(struct getdents_args *uap)
3700 {
3701         int error;
3702
3703         error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3704             &uap->sysmsg_result, UIO_USERSPACE);
3705
3706         return (error);
3707 }
3708
3709 /*
3710  * umask(int newmask)
3711  *
3712  * Set the mode mask for creation of filesystem nodes.
3713  *
3714  * MP SAFE
3715  */
3716 int
3717 sys_umask(struct umask_args *uap)
3718 {
3719         struct thread *td = curthread;
3720         struct proc *p = td->td_proc;
3721         struct filedesc *fdp;
3722
3723         fdp = p->p_fd;
3724         uap->sysmsg_result = fdp->fd_cmask;
3725         fdp->fd_cmask = uap->newmask & ALLPERMS;
3726         return (0);
3727 }
3728
3729 /*
3730  * revoke(char *path)
3731  *
3732  * Void all references to file by ripping underlying filesystem
3733  * away from vnode.
3734  */
3735 /* ARGSUSED */
3736 int
3737 sys_revoke(struct revoke_args *uap)
3738 {
3739         struct nlookupdata nd;
3740         struct vattr vattr;
3741         struct vnode *vp;
3742         struct ucred *cred;
3743         int error;
3744
3745         vp = NULL;
3746         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3747         if (error == 0)
3748                 error = nlookup(&nd);
3749         if (error == 0)
3750                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3751         cred = crhold(nd.nl_cred);
3752         nlookup_done(&nd);
3753         if (error == 0) {
3754                 if (error == 0)
3755                         error = VOP_GETATTR(vp, &vattr);
3756                 if (error == 0 && cred->cr_uid != vattr.va_uid)
3757                         error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
3758                 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
3759                         if (vcount(vp) > 0)
3760                                 error = vrevoke(vp, cred);
3761                 } else if (error == 0) {
3762                         error = vrevoke(vp, cred);
3763                 }
3764                 vrele(vp);
3765         }
3766         if (cred)
3767                 crfree(cred);
3768         return (error);
3769 }
3770
3771 /*
3772  * getfh_args(char *fname, fhandle_t *fhp)
3773  *
3774  * Get (NFS) file handle
3775  *
3776  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
3777  * mount.  This allows nullfs mounts to be explicitly exported.
3778  *
3779  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
3780  *
3781  *          nullfs mounts of subdirectories are not safe.  That is, it will
3782  *          work, but you do not really have protection against access to
3783  *          the related parent directories.
3784  */
3785 int
3786 sys_getfh(struct getfh_args *uap)
3787 {
3788         struct thread *td = curthread;
3789         struct nlookupdata nd;
3790         fhandle_t fh;
3791         struct vnode *vp;
3792         struct mount *mp;
3793         int error;
3794
3795         /*
3796          * Must be super user
3797          */
3798         if ((error = priv_check(td, PRIV_ROOT)) != 0)
3799                 return (error);
3800
3801         vp = NULL;
3802         error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3803         if (error == 0)
3804                 error = nlookup(&nd);
3805         if (error == 0)
3806                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3807         mp = nd.nl_nch.mount;
3808         nlookup_done(&nd);
3809         if (error == 0) {
3810                 bzero(&fh, sizeof(fh));
3811                 fh.fh_fsid = mp->mnt_stat.f_fsid;
3812                 error = VFS_VPTOFH(vp, &fh.fh_fid);
3813                 vput(vp);
3814                 if (error == 0)
3815                         error = copyout(&fh, uap->fhp, sizeof(fh));
3816         }
3817         return (error);
3818 }
3819
3820 /*
3821  * fhopen_args(const struct fhandle *u_fhp, int flags)
3822  *
3823  * syscall for the rpc.lockd to use to translate a NFS file handle into
3824  * an open descriptor.
3825  *
3826  * warning: do not remove the priv_check() call or this becomes one giant
3827  * security hole.
3828  */
3829 int
3830 sys_fhopen(struct fhopen_args *uap)
3831 {
3832         struct thread *td = curthread;
3833         struct proc *p = td->td_proc;
3834         struct mount *mp;
3835         struct vnode *vp;
3836         struct fhandle fhp;
3837         struct vattr vat;
3838         struct vattr *vap = &vat;
3839         struct flock lf;
3840         int fmode, mode, error, type;
3841         struct file *nfp;
3842         struct file *fp;
3843         int indx;
3844
3845         /*
3846          * Must be super user
3847          */
3848         error = priv_check(td, PRIV_ROOT);
3849         if (error)
3850                 return (error);
3851
3852         fmode = FFLAGS(uap->flags);
3853         /* why not allow a non-read/write open for our lockd? */
3854         if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3855                 return (EINVAL);
3856         error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3857         if (error)
3858                 return(error);
3859         /* find the mount point */
3860         mp = vfs_getvfs(&fhp.fh_fsid);
3861         if (mp == NULL)
3862                 return (ESTALE);
3863         /* now give me my vnode, it gets returned to me locked */
3864         error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
3865         if (error)
3866                 return (error);
3867         /*
3868          * from now on we have to make sure not
3869          * to forget about the vnode
3870          * any error that causes an abort must vput(vp)
3871          * just set error = err and 'goto bad;'.
3872          */
3873
3874         /*
3875          * from vn_open
3876          */
3877         if (vp->v_type == VLNK) {
3878                 error = EMLINK;
3879                 goto bad;
3880         }
3881         if (vp->v_type == VSOCK) {
3882                 error = EOPNOTSUPP;
3883                 goto bad;
3884         }
3885         mode = 0;
3886         if (fmode & (FWRITE | O_TRUNC)) {
3887                 if (vp->v_type == VDIR) {
3888                         error = EISDIR;
3889                         goto bad;
3890                 }
3891                 error = vn_writechk(vp, NULL);
3892                 if (error)
3893                         goto bad;
3894                 mode |= VWRITE;
3895         }
3896         if (fmode & FREAD)
3897                 mode |= VREAD;
3898         if (mode) {
3899                 error = VOP_ACCESS(vp, mode, p->p_ucred);
3900                 if (error)
3901                         goto bad;
3902         }
3903         if (fmode & O_TRUNC) {
3904                 vn_unlock(vp);                          /* XXX */
3905                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);   /* XXX */
3906                 VATTR_NULL(vap);
3907                 vap->va_size = 0;
3908                 error = VOP_SETATTR(vp, vap, p->p_ucred);
3909                 if (error)
3910                         goto bad;
3911         }
3912
3913         /*
3914          * VOP_OPEN needs the file pointer so it can potentially override
3915          * it.
3916          *
3917          * WARNING! no f_nchandle will be associated when fhopen()ing a
3918          * directory.  XXX
3919          */
3920         if ((error = falloc(p, &nfp, &indx)) != 0)
3921                 goto bad;
3922         fp = nfp;
3923
3924         error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3925         if (error) {
3926                 /*
3927                  * setting f_ops this way prevents VOP_CLOSE from being
3928                  * called or fdrop() releasing the vp from v_data.   Since
3929                  * the VOP_OPEN failed we don't want to VOP_CLOSE.
3930                  */
3931                 fp->f_ops = &badfileops;
3932                 fp->f_data = NULL;
3933                 goto bad_drop;
3934         }
3935
3936         /*
3937          * The fp is given its own reference, we still have our ref and lock.
3938          *
3939          * Assert that all regular files must be created with a VM object.
3940          */
3941         if (vp->v_type == VREG && vp->v_object == NULL) {
3942                 kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3943                 goto bad_drop;
3944         }
3945
3946         /*
3947          * The open was successful.  Handle any locking requirements.
3948          */
3949         if (fmode & (O_EXLOCK | O_SHLOCK)) {
3950                 lf.l_whence = SEEK_SET;
3951                 lf.l_start = 0;
3952                 lf.l_len = 0;
3953                 if (fmode & O_EXLOCK)
3954                         lf.l_type = F_WRLCK;
3955                 else
3956                         lf.l_type = F_RDLCK;
3957                 if (fmode & FNONBLOCK)
3958                         type = 0;
3959                 else
3960                         type = F_WAIT;
3961                 vn_unlock(vp);
3962                 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3963                         /*
3964                          * release our private reference.
3965                          */
3966                         fsetfd(p, NULL, indx);
3967                         fdrop(fp);
3968                         vrele(vp);
3969                         return (error);
3970                 }
3971                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3972                 fp->f_flag |= FHASLOCK;
3973         }
3974
3975         /*
3976          * Clean up.  Associate the file pointer with the previously
3977          * reserved descriptor and return it.
3978          */
3979         vput(vp);
3980         fsetfd(p, fp, indx);
3981         fdrop(fp);
3982         uap->sysmsg_result = indx;
3983         return (0);
3984
3985 bad_drop:
3986         fsetfd(p, NULL, indx);
3987         fdrop(fp);
3988 bad:
3989         vput(vp);
3990         return (error);
3991 }
3992
3993 /*
3994  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3995  */
3996 int
3997 sys_fhstat(struct fhstat_args *uap)
3998 {
3999         struct thread *td = curthread;
4000         struct stat sb;
4001         fhandle_t fh;
4002         struct mount *mp;
4003         struct vnode *vp;
4004         int error;
4005
4006         /*
4007          * Must be super user
4008          */
4009         error = priv_check(td, PRIV_ROOT);
4010         if (error)
4011                 return (error);
4012
4013         error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4014         if (error)
4015                 return (error);
4016
4017         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4018                 return (ESTALE);
4019         if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4020                 return (error);
4021         error = vn_stat(vp, &sb, td->td_proc->p_ucred);
4022         vput(vp);
4023         if (error)
4024                 return (error);
4025         error = copyout(&sb, uap->sb, sizeof(sb));
4026         return (error);
4027 }
4028
4029 /*
4030  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4031  */
4032 int
4033 sys_fhstatfs(struct fhstatfs_args *uap)
4034 {
4035         struct thread *td = curthread;
4036         struct proc *p = td->td_proc;
4037         struct statfs *sp;
4038         struct mount *mp;
4039         struct vnode *vp;
4040         struct statfs sb;
4041         char *fullpath, *freepath;
4042         fhandle_t fh;
4043         int error;
4044
4045         /*
4046          * Must be super user
4047          */
4048         if ((error = priv_check(td, PRIV_ROOT)))
4049                 return (error);
4050
4051         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4052                 return (error);
4053
4054         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4055                 return (ESTALE);
4056
4057         if (p != NULL && !chroot_visible_mnt(mp, p))
4058                 return (ESTALE);
4059
4060         if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4061                 return (error);
4062         mp = vp->v_mount;
4063         sp = &mp->mnt_stat;
4064         vput(vp);
4065         if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
4066                 return (error);
4067
4068         error = mount_path(p, mp, &fullpath, &freepath);
4069         if (error)
4070                 return(error);
4071         bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4072         strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4073         kfree(freepath, M_TEMP);
4074
4075         sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4076         if (priv_check(td, PRIV_ROOT)) {
4077                 bcopy(sp, &sb, sizeof(sb));
4078                 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4079                 sp = &sb;
4080         }
4081         return (copyout(sp, uap->buf, sizeof(*sp)));
4082 }
4083
4084 /*
4085  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4086  */
4087 int
4088 sys_fhstatvfs(struct fhstatvfs_args *uap)
4089 {
4090         struct thread *td = curthread;
4091         struct proc *p = td->td_proc;
4092         struct statvfs *sp;
4093         struct mount *mp;
4094         struct vnode *vp;
4095         fhandle_t fh;
4096         int error;
4097
4098         /*
4099          * Must be super user
4100          */
4101         if ((error = priv_check(td, PRIV_ROOT)))
4102                 return (error);
4103
4104         if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4105                 return (error);
4106
4107         if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4108                 return (ESTALE);
4109
4110         if (p != NULL && !chroot_visible_mnt(mp, p))
4111                 return (ESTALE);
4112
4113         if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4114                 return (error);
4115         mp = vp->v_mount;
4116         sp = &mp->mnt_vstat;
4117         vput(vp);
4118         if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
4119                 return (error);
4120
4121         sp->f_flag = 0;
4122         if (mp->mnt_flag & MNT_RDONLY)
4123                 sp->f_flag |= ST_RDONLY;
4124         if (mp->mnt_flag & MNT_NOSUID)
4125                 sp->f_flag |= ST_NOSUID;
4126
4127         return (copyout(sp, uap->buf, sizeof(*sp)));
4128 }
4129
4130
4131 /*
4132  * Syscall to push extended attribute configuration information into the
4133  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4134  * a command (int cmd), and attribute name and misc data.  For now, the
4135  * attribute name is left in userspace for consumption by the VFS_op.
4136  * It will probably be changed to be copied into sysspace by the
4137  * syscall in the future, once issues with various consumers of the
4138  * attribute code have raised their hands.
4139  *
4140  * Currently this is used only by UFS Extended Attributes.
4141  */
4142 int
4143 sys_extattrctl(struct extattrctl_args *uap)
4144 {
4145         struct nlookupdata nd;
4146         struct mount *mp;
4147         struct vnode *vp;
4148         int error;
4149
4150         vp = NULL;
4151         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4152         if (error == 0)
4153                 error = nlookup(&nd);
4154         if (error == 0) {
4155                 mp = nd.nl_nch.mount;
4156                 error = VFS_EXTATTRCTL(mp, uap->cmd,
4157                                 uap->attrname, uap->arg,
4158                                 nd.nl_cred);
4159         }
4160         nlookup_done(&nd);
4161         return (error);
4162 }
4163
4164 /*
4165  * Syscall to set a named extended attribute on a file or directory.
4166  * Accepts attribute name, and a uio structure pointing to the data to set.
4167  * The uio is consumed in the style of writev().  The real work happens
4168  * in VOP_SETEXTATTR().
4169  */
4170 int
4171 sys_extattr_set_file(struct extattr_set_file_args *uap)
4172 {
4173         char attrname[EXTATTR_MAXNAMELEN];
4174         struct iovec aiov[UIO_SMALLIOV];
4175         struct iovec *needfree;
4176         struct nlookupdata nd;
4177         struct iovec *iov;
4178         struct vnode *vp;
4179         struct uio auio;
4180         u_int iovlen;
4181         u_int cnt;
4182         int error;
4183         int i;
4184
4185         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4186         if (error)
4187                 return (error);
4188
4189         vp = NULL;
4190         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4191         if (error == 0)
4192                 error = nlookup(&nd);
4193         if (error == 0)
4194                 error = ncp_writechk(&nd.nl_nch);
4195         if (error == 0)
4196                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4197         if (error) {
4198                 nlookup_done(&nd);
4199                 return (error);
4200         }
4201
4202         needfree = NULL;
4203         iovlen = uap->iovcnt * sizeof(struct iovec);
4204         if (uap->iovcnt > UIO_SMALLIOV) {
4205                 if (uap->iovcnt > UIO_MAXIOV) {
4206                         error = EINVAL;
4207                         goto done;
4208                 }
4209                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4210                 needfree = iov;
4211         } else {
4212                 iov = aiov;
4213         }
4214         auio.uio_iov = iov;
4215         auio.uio_iovcnt = uap->iovcnt;
4216         auio.uio_rw = UIO_WRITE;
4217         auio.uio_segflg = UIO_USERSPACE;
4218         auio.uio_td = nd.nl_td;
4219         auio.uio_offset = 0;
4220         if ((error = copyin(uap->iovp, iov, iovlen)))
4221                 goto done;
4222         auio.uio_resid = 0;
4223         for (i = 0; i < uap->iovcnt; i++) {
4224                 if (iov->iov_len > LONG_MAX - auio.uio_resid) {
4225                         error = EINVAL;
4226                         goto done;
4227                 }
4228                 auio.uio_resid += iov->iov_len;
4229                 iov++;
4230         }
4231         cnt = auio.uio_resid;
4232         error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4233         cnt -= auio.uio_resid;
4234         uap->sysmsg_result = cnt;
4235 done:
4236         vput(vp);
4237         nlookup_done(&nd);
4238         if (needfree)
4239                 FREE(needfree, M_IOV);
4240         return (error);
4241 }
4242
4243 /*
4244  * Syscall to get a named extended attribute on a file or directory.
4245  * Accepts attribute name, and a uio structure pointing to a buffer for the
4246  * data.  The uio is consumed in the style of readv().  The real work
4247  * happens in VOP_GETEXTATTR();
4248  */
4249 int
4250 sys_extattr_get_file(struct extattr_get_file_args *uap)
4251 {
4252         char attrname[EXTATTR_MAXNAMELEN];
4253         struct iovec aiov[UIO_SMALLIOV];
4254         struct iovec *needfree;
4255         struct nlookupdata nd;
4256         struct iovec *iov;
4257         struct vnode *vp;
4258         struct uio auio;
4259         u_int iovlen;
4260         u_int cnt;
4261         int error;
4262         int i;
4263
4264         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4265         if (error)
4266                 return (error);
4267
4268         vp = NULL;
4269         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4270         if (error == 0)
4271                 error = nlookup(&nd);
4272         if (error == 0)
4273                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4274         if (error) {
4275                 nlookup_done(&nd);
4276                 return (error);
4277         }
4278
4279         iovlen = uap->iovcnt * sizeof (struct iovec);
4280         needfree = NULL;
4281         if (uap->iovcnt > UIO_SMALLIOV) {
4282                 if (uap->iovcnt > UIO_MAXIOV) {
4283                         error = EINVAL;
4284                         goto done;
4285                 }
4286                 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4287                 needfree = iov;
4288         } else {
4289                 iov = aiov;
4290         }
4291         auio.uio_iov = iov;
4292         auio.uio_iovcnt = uap->iovcnt;
4293         auio.uio_rw = UIO_READ;
4294         auio.uio_segflg = UIO_USERSPACE;
4295         auio.uio_td = nd.nl_td;
4296         auio.uio_offset = 0;
4297         if ((error = copyin(uap->iovp, iov, iovlen)))
4298                 goto done;
4299         auio.uio_resid = 0;
4300         for (i = 0; i < uap->iovcnt; i++) {
4301                 if (iov->iov_len > LONG_MAX - auio.uio_resid) {
4302                         error = EINVAL;
4303                         goto done;
4304                 }
4305                 auio.uio_resid += iov->iov_len;
4306                 iov++;
4307         }
4308         cnt = auio.uio_resid;
4309         error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4310         cnt -= auio.uio_resid;
4311         uap->sysmsg_result = cnt;
4312 done:
4313         vput(vp);
4314         nlookup_done(&nd);
4315         if (needfree)
4316                 FREE(needfree, M_IOV);
4317         return(error);
4318 }
4319
4320 /*
4321  * Syscall to delete a named extended attribute from a file or directory.
4322  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4323  */
4324 int
4325 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4326 {
4327         char attrname[EXTATTR_MAXNAMELEN];
4328         struct nlookupdata nd;
4329         struct vnode *vp;
4330         int error;
4331
4332         error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4333         if (error)
4334                 return(error);
4335
4336         vp = NULL;
4337         error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4338         if (error == 0)
4339                 error = nlookup(&nd);
4340         if (error == 0)
4341                 error = ncp_writechk(&nd.nl_nch);
4342         if (error == 0)
4343                 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4344         if (error) {
4345                 nlookup_done(&nd);
4346                 return (error);
4347         }
4348
4349         error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4350         vput(vp);
4351         nlookup_done(&nd);
4352         return(error);
4353 }
4354
4355 /*
4356  * Determine if the mount is visible to the process.
4357  */
4358 static int
4359 chroot_visible_mnt(struct mount *mp, struct proc *p)
4360 {
4361         struct nchandle nch;
4362
4363         /*
4364          * Traverse from the mount point upwards.  If we hit the process
4365          * root then the mount point is visible to the process.
4366          */
4367         nch = mp->mnt_ncmountpt;
4368         while (nch.ncp) {
4369                 if (nch.mount == p->p_fd->fd_nrdir.mount &&
4370                     nch.ncp == p->p_fd->fd_nrdir.ncp) {
4371                         return(1);
4372                 }
4373                 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4374                         nch = nch.mount->mnt_ncmounton;
4375                 } else {
4376                         nch.ncp = nch.ncp->nc_parent;
4377                 }
4378         }
4379
4380         /*
4381          * If the mount point is not visible to the process, but the
4382          * process root is in a subdirectory of the mount, return
4383          * TRUE anyway.
4384          */
4385         if (p->p_fd->fd_nrdir.mount == mp)
4386                 return(1);
4387
4388         return(0);
4389 }
4390