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 ((error = priv_check(td, PRIV_ROOT)) != 0) 912 return (error); 913 914 /* 915 * Argument length checks 916 */ 917 if (uap->ctllen < 0 || uap->ctllen > 1024) 918 return (EINVAL); 919 if (uap->buflen < 0 || uap->buflen > 16 * 1024) 920 return (EINVAL); 921 if (uap->path == NULL) 922 return (EINVAL); 923 924 /* 925 * Allocate the necessary buffers and copyin data 926 */ 927 path = objcache_get(namei_oc, M_WAITOK); 928 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 929 if (error) 930 goto done; 931 932 if (uap->ctllen) { 933 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO); 934 error = copyin(uap->ctl, ctl, uap->ctllen); 935 if (error) 936 goto done; 937 } 938 if (uap->buflen) 939 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO); 940 941 /* 942 * Validate the descriptor 943 */ 944 if (uap->fd >= 0) { 945 fp = holdfp(p->p_fd, uap->fd, -1); 946 if (fp == NULL) { 947 error = EBADF; 948 goto done; 949 } 950 } else { 951 fp = NULL; 952 } 953 954 /* 955 * Execute the internal kernel function and clean up. 956 */ 957 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result); 958 if (fp) 959 fdrop(fp); 960 if (error == 0 && uap->sysmsg_result > 0) 961 error = copyout(buf, uap->buf, uap->sysmsg_result); 962 done: 963 if (path) 964 objcache_put(namei_oc, path); 965 if (ctl) 966 kfree(ctl, M_TEMP); 967 if (buf) 968 kfree(buf, M_TEMP); 969 return (error); 970 } 971 972 /* 973 * Execute a mount control operation by resolving the path to a mount point 974 * and calling vop_mountctl(). 975 * 976 * Use the mount point from the nch instead of the vnode so nullfs mounts 977 * can properly spike the VOP. 978 */ 979 int 980 kern_mountctl(const char *path, int op, struct file *fp, 981 const void *ctl, int ctllen, 982 void *buf, int buflen, int *res) 983 { 984 struct vnode *vp; 985 struct mount *mp; 986 struct nlookupdata nd; 987 int error; 988 989 *res = 0; 990 vp = NULL; 991 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW); 992 if (error == 0) 993 error = nlookup(&nd); 994 if (error == 0) 995 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 996 mp = nd.nl_nch.mount; 997 nlookup_done(&nd); 998 if (error) 999 return (error); 1000 1001 /* 1002 * Must be the root of the filesystem 1003 */ 1004 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 1005 vput(vp); 1006 return (EINVAL); 1007 } 1008 error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen, 1009 buf, buflen, res); 1010 vput(vp); 1011 return (error); 1012 } 1013 1014 int 1015 kern_statfs(struct nlookupdata *nd, struct statfs *buf) 1016 { 1017 struct thread *td = curthread; 1018 struct proc *p = td->td_proc; 1019 struct mount *mp; 1020 struct statfs *sp; 1021 char *fullpath, *freepath; 1022 int error; 1023 1024 if ((error = nlookup(nd)) != 0) 1025 return (error); 1026 mp = nd->nl_nch.mount; 1027 sp = &mp->mnt_stat; 1028 if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0) 1029 return (error); 1030 1031 error = mount_path(p, mp, &fullpath, &freepath); 1032 if (error) 1033 return(error); 1034 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1035 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1036 kfree(freepath, M_TEMP); 1037 1038 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1039 bcopy(sp, buf, sizeof(*buf)); 1040 /* Only root should have access to the fsid's. */ 1041 if (priv_check(td, PRIV_ROOT)) 1042 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1043 return (0); 1044 } 1045 1046 /* 1047 * statfs_args(char *path, struct statfs *buf) 1048 * 1049 * Get filesystem statistics. 1050 */ 1051 int 1052 sys_statfs(struct statfs_args *uap) 1053 { 1054 struct nlookupdata nd; 1055 struct statfs buf; 1056 int error; 1057 1058 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1059 if (error == 0) 1060 error = kern_statfs(&nd, &buf); 1061 nlookup_done(&nd); 1062 if (error == 0) 1063 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1064 return (error); 1065 } 1066 1067 int 1068 kern_fstatfs(int fd, struct statfs *buf) 1069 { 1070 struct thread *td = curthread; 1071 struct proc *p = td->td_proc; 1072 struct file *fp; 1073 struct mount *mp; 1074 struct statfs *sp; 1075 char *fullpath, *freepath; 1076 int error; 1077 1078 KKASSERT(p); 1079 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 1080 return (error); 1081 mp = ((struct vnode *)fp->f_data)->v_mount; 1082 if (mp == NULL) { 1083 error = EBADF; 1084 goto done; 1085 } 1086 if (fp->f_cred == NULL) { 1087 error = EINVAL; 1088 goto done; 1089 } 1090 sp = &mp->mnt_stat; 1091 if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0) 1092 goto done; 1093 1094 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0) 1095 goto done; 1096 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1097 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1098 kfree(freepath, M_TEMP); 1099 1100 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1101 bcopy(sp, buf, sizeof(*buf)); 1102 1103 /* Only root should have access to the fsid's. */ 1104 if (priv_check(td, PRIV_ROOT)) 1105 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1106 error = 0; 1107 done: 1108 fdrop(fp); 1109 return (error); 1110 } 1111 1112 /* 1113 * fstatfs_args(int fd, struct statfs *buf) 1114 * 1115 * Get filesystem statistics. 1116 */ 1117 int 1118 sys_fstatfs(struct fstatfs_args *uap) 1119 { 1120 struct statfs buf; 1121 int error; 1122 1123 error = kern_fstatfs(uap->fd, &buf); 1124 1125 if (error == 0) 1126 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1127 return (error); 1128 } 1129 1130 int 1131 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf) 1132 { 1133 struct mount *mp; 1134 struct statvfs *sp; 1135 int error; 1136 1137 if ((error = nlookup(nd)) != 0) 1138 return (error); 1139 mp = nd->nl_nch.mount; 1140 sp = &mp->mnt_vstat; 1141 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0) 1142 return (error); 1143 1144 sp->f_flag = 0; 1145 if (mp->mnt_flag & MNT_RDONLY) 1146 sp->f_flag |= ST_RDONLY; 1147 if (mp->mnt_flag & MNT_NOSUID) 1148 sp->f_flag |= ST_NOSUID; 1149 bcopy(sp, buf, sizeof(*buf)); 1150 return (0); 1151 } 1152 1153 /* 1154 * statfs_args(char *path, struct statfs *buf) 1155 * 1156 * Get filesystem statistics. 1157 */ 1158 int 1159 sys_statvfs(struct statvfs_args *uap) 1160 { 1161 struct nlookupdata nd; 1162 struct statvfs buf; 1163 int error; 1164 1165 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1166 if (error == 0) 1167 error = kern_statvfs(&nd, &buf); 1168 nlookup_done(&nd); 1169 if (error == 0) 1170 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1171 return (error); 1172 } 1173 1174 int 1175 kern_fstatvfs(int fd, struct statvfs *buf) 1176 { 1177 struct thread *td = curthread; 1178 struct proc *p = td->td_proc; 1179 struct file *fp; 1180 struct mount *mp; 1181 struct statvfs *sp; 1182 int error; 1183 1184 KKASSERT(p); 1185 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 1186 return (error); 1187 mp = ((struct vnode *)fp->f_data)->v_mount; 1188 if (mp == NULL) { 1189 error = EBADF; 1190 goto done; 1191 } 1192 if (fp->f_cred == NULL) { 1193 error = EINVAL; 1194 goto done; 1195 } 1196 sp = &mp->mnt_vstat; 1197 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0) 1198 goto done; 1199 1200 sp->f_flag = 0; 1201 if (mp->mnt_flag & MNT_RDONLY) 1202 sp->f_flag |= ST_RDONLY; 1203 if (mp->mnt_flag & MNT_NOSUID) 1204 sp->f_flag |= ST_NOSUID; 1205 1206 bcopy(sp, buf, sizeof(*buf)); 1207 error = 0; 1208 done: 1209 fdrop(fp); 1210 return (error); 1211 } 1212 1213 /* 1214 * fstatfs_args(int fd, struct statfs *buf) 1215 * 1216 * Get filesystem statistics. 1217 */ 1218 int 1219 sys_fstatvfs(struct fstatvfs_args *uap) 1220 { 1221 struct statvfs buf; 1222 int error; 1223 1224 error = kern_fstatvfs(uap->fd, &buf); 1225 1226 if (error == 0) 1227 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1228 return (error); 1229 } 1230 1231 /* 1232 * getfsstat_args(struct statfs *buf, long bufsize, int flags) 1233 * 1234 * Get statistics on all filesystems. 1235 */ 1236 1237 struct getfsstat_info { 1238 struct statfs *sfsp; 1239 long count; 1240 long maxcount; 1241 int error; 1242 int flags; 1243 struct proc *p; 1244 }; 1245 1246 static int getfsstat_callback(struct mount *, void *); 1247 1248 /* ARGSUSED */ 1249 int 1250 sys_getfsstat(struct getfsstat_args *uap) 1251 { 1252 struct thread *td = curthread; 1253 struct proc *p = td->td_proc; 1254 struct getfsstat_info info; 1255 1256 bzero(&info, sizeof(info)); 1257 1258 info.maxcount = uap->bufsize / sizeof(struct statfs); 1259 info.sfsp = uap->buf; 1260 info.count = 0; 1261 info.flags = uap->flags; 1262 info.p = p; 1263 1264 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD); 1265 if (info.sfsp && info.count > info.maxcount) 1266 uap->sysmsg_result = info.maxcount; 1267 else 1268 uap->sysmsg_result = info.count; 1269 return (info.error); 1270 } 1271 1272 static int 1273 getfsstat_callback(struct mount *mp, void *data) 1274 { 1275 struct getfsstat_info *info = data; 1276 struct statfs *sp; 1277 char *freepath; 1278 char *fullpath; 1279 int error; 1280 1281 if (info->sfsp && info->count < info->maxcount) { 1282 if (info->p && !chroot_visible_mnt(mp, info->p)) 1283 return(0); 1284 sp = &mp->mnt_stat; 1285 1286 /* 1287 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1288 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1289 * overrides MNT_WAIT. 1290 */ 1291 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1292 (info->flags & MNT_WAIT)) && 1293 (error = VFS_STATFS(mp, sp, info->p->p_ucred))) { 1294 return(0); 1295 } 1296 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1297 1298 error = mount_path(info->p, mp, &fullpath, &freepath); 1299 if (error) { 1300 info->error = error; 1301 return(-1); 1302 } 1303 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1304 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1305 kfree(freepath, M_TEMP); 1306 1307 error = copyout(sp, info->sfsp, sizeof(*sp)); 1308 if (error) { 1309 info->error = error; 1310 return (-1); 1311 } 1312 ++info->sfsp; 1313 } 1314 info->count++; 1315 return(0); 1316 } 1317 1318 /* 1319 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf, 1320 long bufsize, int flags) 1321 * 1322 * Get statistics on all filesystems. 1323 */ 1324 1325 struct getvfsstat_info { 1326 struct statfs *sfsp; 1327 struct statvfs *vsfsp; 1328 long count; 1329 long maxcount; 1330 int error; 1331 int flags; 1332 struct proc *p; 1333 }; 1334 1335 static int getvfsstat_callback(struct mount *, void *); 1336 1337 /* ARGSUSED */ 1338 int 1339 sys_getvfsstat(struct getvfsstat_args *uap) 1340 { 1341 struct thread *td = curthread; 1342 struct proc *p = td->td_proc; 1343 struct getvfsstat_info info; 1344 1345 bzero(&info, sizeof(info)); 1346 1347 info.maxcount = uap->vbufsize / sizeof(struct statvfs); 1348 info.sfsp = uap->buf; 1349 info.vsfsp = uap->vbuf; 1350 info.count = 0; 1351 info.flags = uap->flags; 1352 info.p = p; 1353 1354 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD); 1355 if (info.vsfsp && info.count > info.maxcount) 1356 uap->sysmsg_result = info.maxcount; 1357 else 1358 uap->sysmsg_result = info.count; 1359 return (info.error); 1360 } 1361 1362 static int 1363 getvfsstat_callback(struct mount *mp, void *data) 1364 { 1365 struct getvfsstat_info *info = data; 1366 struct statfs *sp; 1367 struct statvfs *vsp; 1368 char *freepath; 1369 char *fullpath; 1370 int error; 1371 1372 if (info->vsfsp && info->count < info->maxcount) { 1373 if (info->p && !chroot_visible_mnt(mp, info->p)) 1374 return(0); 1375 sp = &mp->mnt_stat; 1376 vsp = &mp->mnt_vstat; 1377 1378 /* 1379 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1380 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1381 * overrides MNT_WAIT. 1382 */ 1383 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1384 (info->flags & MNT_WAIT)) && 1385 (error = VFS_STATFS(mp, sp, info->p->p_ucred))) { 1386 return(0); 1387 } 1388 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1389 1390 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1391 (info->flags & MNT_WAIT)) && 1392 (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) { 1393 return(0); 1394 } 1395 vsp->f_flag = 0; 1396 if (mp->mnt_flag & MNT_RDONLY) 1397 vsp->f_flag |= ST_RDONLY; 1398 if (mp->mnt_flag & MNT_NOSUID) 1399 vsp->f_flag |= ST_NOSUID; 1400 1401 error = mount_path(info->p, mp, &fullpath, &freepath); 1402 if (error) { 1403 info->error = error; 1404 return(-1); 1405 } 1406 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1407 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1408 kfree(freepath, M_TEMP); 1409 1410 error = copyout(sp, info->sfsp, sizeof(*sp)); 1411 if (error == 0) 1412 error = copyout(vsp, info->vsfsp, sizeof(*vsp)); 1413 if (error) { 1414 info->error = error; 1415 return (-1); 1416 } 1417 ++info->sfsp; 1418 ++info->vsfsp; 1419 } 1420 info->count++; 1421 return(0); 1422 } 1423 1424 1425 /* 1426 * fchdir_args(int fd) 1427 * 1428 * Change current working directory to a given file descriptor. 1429 */ 1430 /* ARGSUSED */ 1431 int 1432 sys_fchdir(struct fchdir_args *uap) 1433 { 1434 struct thread *td = curthread; 1435 struct proc *p = td->td_proc; 1436 struct filedesc *fdp = p->p_fd; 1437 struct vnode *vp, *ovp; 1438 struct mount *mp; 1439 struct file *fp; 1440 struct nchandle nch, onch, tnch; 1441 int error; 1442 1443 if ((error = holdvnode(fdp, uap->fd, &fp)) != 0) 1444 return (error); 1445 vp = (struct vnode *)fp->f_data; 1446 vref(vp); 1447 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1448 if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL) 1449 error = ENOTDIR; 1450 else 1451 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 1452 if (error) { 1453 vput(vp); 1454 fdrop(fp); 1455 return (error); 1456 } 1457 cache_copy(&fp->f_nchandle, &nch); 1458 1459 /* 1460 * If the ncp has become a mount point, traverse through 1461 * the mount point. 1462 */ 1463 1464 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) && 1465 (mp = cache_findmount(&nch)) != NULL 1466 ) { 1467 error = nlookup_mp(mp, &tnch); 1468 if (error == 0) { 1469 cache_unlock(&tnch); /* leave ref intact */ 1470 vput(vp); 1471 vp = tnch.ncp->nc_vp; 1472 error = vget(vp, LK_SHARED); 1473 KKASSERT(error == 0); 1474 cache_drop(&nch); 1475 nch = tnch; 1476 } 1477 } 1478 if (error == 0) { 1479 ovp = fdp->fd_cdir; 1480 onch = fdp->fd_ncdir; 1481 vn_unlock(vp); /* leave ref intact */ 1482 fdp->fd_cdir = vp; 1483 fdp->fd_ncdir = nch; 1484 cache_drop(&onch); 1485 vrele(ovp); 1486 } else { 1487 cache_drop(&nch); 1488 vput(vp); 1489 } 1490 fdrop(fp); 1491 return (error); 1492 } 1493 1494 int 1495 kern_chdir(struct nlookupdata *nd) 1496 { 1497 struct thread *td = curthread; 1498 struct proc *p = td->td_proc; 1499 struct filedesc *fdp = p->p_fd; 1500 struct vnode *vp, *ovp; 1501 struct nchandle onch; 1502 int error; 1503 1504 if ((error = nlookup(nd)) != 0) 1505 return (error); 1506 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 1507 return (ENOENT); 1508 if ((error = vget(vp, LK_SHARED)) != 0) 1509 return (error); 1510 1511 error = checkvp_chdir(vp, td); 1512 vn_unlock(vp); 1513 if (error == 0) { 1514 ovp = fdp->fd_cdir; 1515 onch = fdp->fd_ncdir; 1516 cache_unlock(&nd->nl_nch); /* leave reference intact */ 1517 fdp->fd_ncdir = nd->nl_nch; 1518 fdp->fd_cdir = vp; 1519 cache_drop(&onch); 1520 vrele(ovp); 1521 cache_zero(&nd->nl_nch); 1522 } else { 1523 vrele(vp); 1524 } 1525 return (error); 1526 } 1527 1528 /* 1529 * chdir_args(char *path) 1530 * 1531 * Change current working directory (``.''). 1532 */ 1533 int 1534 sys_chdir(struct chdir_args *uap) 1535 { 1536 struct nlookupdata nd; 1537 int error; 1538 1539 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1540 if (error == 0) 1541 error = kern_chdir(&nd); 1542 nlookup_done(&nd); 1543 return (error); 1544 } 1545 1546 /* 1547 * Helper function for raised chroot(2) security function: Refuse if 1548 * any filedescriptors are open directories. 1549 */ 1550 static int 1551 chroot_refuse_vdir_fds(struct filedesc *fdp) 1552 { 1553 struct vnode *vp; 1554 struct file *fp; 1555 int error; 1556 int fd; 1557 1558 for (fd = 0; fd < fdp->fd_nfiles ; fd++) { 1559 if ((error = holdvnode(fdp, fd, &fp)) != 0) 1560 continue; 1561 vp = (struct vnode *)fp->f_data; 1562 if (vp->v_type != VDIR) { 1563 fdrop(fp); 1564 continue; 1565 } 1566 fdrop(fp); 1567 return(EPERM); 1568 } 1569 return (0); 1570 } 1571 1572 /* 1573 * This sysctl determines if we will allow a process to chroot(2) if it 1574 * has a directory open: 1575 * 0: disallowed for all processes. 1576 * 1: allowed for processes that were not already chroot(2)'ed. 1577 * 2: allowed for all processes. 1578 */ 1579 1580 static int chroot_allow_open_directories = 1; 1581 1582 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 1583 &chroot_allow_open_directories, 0, ""); 1584 1585 /* 1586 * chroot to the specified namecache entry. We obtain the vp from the 1587 * namecache data. The passed ncp must be locked and referenced and will 1588 * remain locked and referenced on return. 1589 */ 1590 int 1591 kern_chroot(struct nchandle *nch) 1592 { 1593 struct thread *td = curthread; 1594 struct proc *p = td->td_proc; 1595 struct filedesc *fdp = p->p_fd; 1596 struct vnode *vp; 1597 int error; 1598 1599 /* 1600 * Only root can chroot 1601 */ 1602 if ((error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT)) != 0) 1603 return (error); 1604 1605 /* 1606 * Disallow open directory descriptors (fchdir() breakouts). 1607 */ 1608 if (chroot_allow_open_directories == 0 || 1609 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 1610 if ((error = chroot_refuse_vdir_fds(fdp)) != 0) 1611 return (error); 1612 } 1613 if ((vp = nch->ncp->nc_vp) == NULL) 1614 return (ENOENT); 1615 1616 if ((error = vget(vp, LK_SHARED)) != 0) 1617 return (error); 1618 1619 /* 1620 * Check the validity of vp as a directory to change to and 1621 * associate it with rdir/jdir. 1622 */ 1623 error = checkvp_chdir(vp, td); 1624 vn_unlock(vp); /* leave reference intact */ 1625 if (error == 0) { 1626 vrele(fdp->fd_rdir); 1627 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */ 1628 cache_drop(&fdp->fd_nrdir); 1629 cache_copy(nch, &fdp->fd_nrdir); 1630 if (fdp->fd_jdir == NULL) { 1631 fdp->fd_jdir = vp; 1632 vref(fdp->fd_jdir); 1633 cache_copy(nch, &fdp->fd_njdir); 1634 } 1635 } else { 1636 vrele(vp); 1637 } 1638 return (error); 1639 } 1640 1641 /* 1642 * chroot_args(char *path) 1643 * 1644 * Change notion of root (``/'') directory. 1645 */ 1646 /* ARGSUSED */ 1647 int 1648 sys_chroot(struct chroot_args *uap) 1649 { 1650 struct thread *td = curthread; 1651 struct nlookupdata nd; 1652 int error; 1653 1654 KKASSERT(td->td_proc); 1655 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1656 if (error) { 1657 nlookup_done(&nd); 1658 return(error); 1659 } 1660 nd.nl_flags |= NLC_EXEC; 1661 error = nlookup(&nd); 1662 if (error == 0) 1663 error = kern_chroot(&nd.nl_nch); 1664 nlookup_done(&nd); 1665 return(error); 1666 } 1667 1668 /* 1669 * Common routine for chroot and chdir. Given a locked, referenced vnode, 1670 * determine whether it is legal to chdir to the vnode. The vnode's state 1671 * is not changed by this call. 1672 */ 1673 int 1674 checkvp_chdir(struct vnode *vp, struct thread *td) 1675 { 1676 int error; 1677 1678 if (vp->v_type != VDIR) 1679 error = ENOTDIR; 1680 else 1681 error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred); 1682 return (error); 1683 } 1684 1685 int 1686 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res) 1687 { 1688 struct thread *td = curthread; 1689 struct proc *p = td->td_proc; 1690 struct lwp *lp = td->td_lwp; 1691 struct filedesc *fdp = p->p_fd; 1692 int cmode, flags; 1693 struct file *nfp; 1694 struct file *fp; 1695 struct vnode *vp; 1696 int type, indx, error; 1697 struct flock lf; 1698 1699 if ((oflags & O_ACCMODE) == O_ACCMODE) 1700 return (EINVAL); 1701 flags = FFLAGS(oflags); 1702 error = falloc(p, &nfp, NULL); 1703 if (error) 1704 return (error); 1705 fp = nfp; 1706 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; 1707 1708 /* 1709 * XXX p_dupfd is a real mess. It allows a device to return a 1710 * file descriptor to be duplicated rather then doing the open 1711 * itself. 1712 */ 1713 lp->lwp_dupfd = -1; 1714 1715 /* 1716 * Call vn_open() to do the lookup and assign the vnode to the 1717 * file pointer. vn_open() does not change the ref count on fp 1718 * and the vnode, on success, will be inherited by the file pointer 1719 * and unlocked. 1720 */ 1721 nd->nl_flags |= NLC_LOCKVP; 1722 error = vn_open(nd, fp, flags, cmode); 1723 nlookup_done(nd); 1724 if (error) { 1725 /* 1726 * handle special fdopen() case. bleh. dupfdopen() is 1727 * responsible for dropping the old contents of ofiles[indx] 1728 * if it succeeds. 1729 * 1730 * Note that fsetfd() will add a ref to fp which represents 1731 * the fd_files[] assignment. We must still drop our 1732 * reference. 1733 */ 1734 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) { 1735 if (fdalloc(p, 0, &indx) == 0) { 1736 error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error); 1737 if (error == 0) { 1738 *res = indx; 1739 fdrop(fp); /* our ref */ 1740 return (0); 1741 } 1742 fsetfd(p, NULL, indx); 1743 } 1744 } 1745 fdrop(fp); /* our ref */ 1746 if (error == ERESTART) 1747 error = EINTR; 1748 return (error); 1749 } 1750 1751 /* 1752 * ref the vnode for ourselves so it can't be ripped out from under 1753 * is. XXX need an ND flag to request that the vnode be returned 1754 * anyway. 1755 * 1756 * Reserve a file descriptor but do not assign it until the open 1757 * succeeds. 1758 */ 1759 vp = (struct vnode *)fp->f_data; 1760 vref(vp); 1761 if ((error = fdalloc(p, 0, &indx)) != 0) { 1762 fdrop(fp); 1763 vrele(vp); 1764 return (error); 1765 } 1766 1767 /* 1768 * If no error occurs the vp will have been assigned to the file 1769 * pointer. 1770 */ 1771 lp->lwp_dupfd = 0; 1772 1773 if (flags & (O_EXLOCK | O_SHLOCK)) { 1774 lf.l_whence = SEEK_SET; 1775 lf.l_start = 0; 1776 lf.l_len = 0; 1777 if (flags & O_EXLOCK) 1778 lf.l_type = F_WRLCK; 1779 else 1780 lf.l_type = F_RDLCK; 1781 if (flags & FNONBLOCK) 1782 type = 0; 1783 else 1784 type = F_WAIT; 1785 1786 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) { 1787 /* 1788 * lock request failed. Clean up the reserved 1789 * descriptor. 1790 */ 1791 vrele(vp); 1792 fsetfd(p, NULL, indx); 1793 fdrop(fp); 1794 return (error); 1795 } 1796 fp->f_flag |= FHASLOCK; 1797 } 1798 #if 0 1799 /* 1800 * Assert that all regular file vnodes were created with a object. 1801 */ 1802 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 1803 ("open: regular file has no backing object after vn_open")); 1804 #endif 1805 1806 vrele(vp); 1807 1808 /* 1809 * release our private reference, leaving the one associated with the 1810 * descriptor table intact. 1811 */ 1812 fsetfd(p, fp, indx); 1813 fdrop(fp); 1814 *res = indx; 1815 return (0); 1816 } 1817 1818 /* 1819 * open_args(char *path, int flags, int mode) 1820 * 1821 * Check permissions, allocate an open file structure, 1822 * and call the device open routine if any. 1823 */ 1824 int 1825 sys_open(struct open_args *uap) 1826 { 1827 struct nlookupdata nd; 1828 int error; 1829 1830 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 1831 if (error == 0) { 1832 error = kern_open(&nd, uap->flags, 1833 uap->mode, &uap->sysmsg_result); 1834 } 1835 nlookup_done(&nd); 1836 return (error); 1837 } 1838 1839 int 1840 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor) 1841 { 1842 struct thread *td = curthread; 1843 struct proc *p = td->td_proc; 1844 struct vnode *vp; 1845 struct vattr vattr; 1846 int error; 1847 int whiteout = 0; 1848 1849 KKASSERT(p); 1850 1851 switch (mode & S_IFMT) { 1852 case S_IFCHR: 1853 case S_IFBLK: 1854 error = priv_check(td, PRIV_ROOT); 1855 break; 1856 default: 1857 error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT); 1858 break; 1859 } 1860 if (error) 1861 return (error); 1862 1863 bwillinode(1); 1864 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 1865 if ((error = nlookup(nd)) != 0) 1866 return (error); 1867 if (nd->nl_nch.ncp->nc_vp) 1868 return (EEXIST); 1869 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 1870 return (error); 1871 1872 VATTR_NULL(&vattr); 1873 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 1874 vattr.va_rmajor = rmajor; 1875 vattr.va_rminor = rminor; 1876 whiteout = 0; 1877 1878 switch (mode & S_IFMT) { 1879 case S_IFMT: /* used by badsect to flag bad sectors */ 1880 vattr.va_type = VBAD; 1881 break; 1882 case S_IFCHR: 1883 vattr.va_type = VCHR; 1884 break; 1885 case S_IFBLK: 1886 vattr.va_type = VBLK; 1887 break; 1888 case S_IFWHT: 1889 whiteout = 1; 1890 break; 1891 case S_IFDIR: 1892 /* special directories support for HAMMER */ 1893 vattr.va_type = VDIR; 1894 break; 1895 default: 1896 error = EINVAL; 1897 break; 1898 } 1899 if (error == 0) { 1900 if (whiteout) { 1901 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp, 1902 nd->nl_cred, NAMEI_CREATE); 1903 } else { 1904 vp = NULL; 1905 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, 1906 &vp, nd->nl_cred, &vattr); 1907 if (error == 0) 1908 vput(vp); 1909 } 1910 } 1911 return (error); 1912 } 1913 1914 /* 1915 * mknod_args(char *path, int mode, int dev) 1916 * 1917 * Create a special file. 1918 */ 1919 int 1920 sys_mknod(struct mknod_args *uap) 1921 { 1922 struct nlookupdata nd; 1923 int error; 1924 1925 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 1926 if (error == 0) { 1927 error = kern_mknod(&nd, uap->mode, 1928 umajor(uap->dev), uminor(uap->dev)); 1929 } 1930 nlookup_done(&nd); 1931 return (error); 1932 } 1933 1934 int 1935 kern_mkfifo(struct nlookupdata *nd, int mode) 1936 { 1937 struct thread *td = curthread; 1938 struct proc *p = td->td_proc; 1939 struct vattr vattr; 1940 struct vnode *vp; 1941 int error; 1942 1943 bwillinode(1); 1944 1945 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 1946 if ((error = nlookup(nd)) != 0) 1947 return (error); 1948 if (nd->nl_nch.ncp->nc_vp) 1949 return (EEXIST); 1950 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 1951 return (error); 1952 1953 VATTR_NULL(&vattr); 1954 vattr.va_type = VFIFO; 1955 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 1956 vp = NULL; 1957 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr); 1958 if (error == 0) 1959 vput(vp); 1960 return (error); 1961 } 1962 1963 /* 1964 * mkfifo_args(char *path, int mode) 1965 * 1966 * Create a named pipe. 1967 */ 1968 int 1969 sys_mkfifo(struct mkfifo_args *uap) 1970 { 1971 struct nlookupdata nd; 1972 int error; 1973 1974 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 1975 if (error == 0) 1976 error = kern_mkfifo(&nd, uap->mode); 1977 nlookup_done(&nd); 1978 return (error); 1979 } 1980 1981 static int hardlink_check_uid = 0; 1982 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, 1983 &hardlink_check_uid, 0, 1984 "Unprivileged processes cannot create hard links to files owned by other " 1985 "users"); 1986 static int hardlink_check_gid = 0; 1987 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, 1988 &hardlink_check_gid, 0, 1989 "Unprivileged processes cannot create hard links to files owned by other " 1990 "groups"); 1991 1992 static int 1993 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred) 1994 { 1995 struct vattr va; 1996 int error; 1997 1998 /* 1999 * Shortcut if disabled 2000 */ 2001 if (hardlink_check_uid == 0 && hardlink_check_gid == 0) 2002 return (0); 2003 2004 /* 2005 * root cred can always hardlink 2006 */ 2007 if (priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT) == 0) 2008 return (0); 2009 2010 /* 2011 * Otherwise only if the originating file is owned by the 2012 * same user or group. Note that any group is allowed if 2013 * the file is owned by the caller. 2014 */ 2015 error = VOP_GETATTR(vp, &va); 2016 if (error != 0) 2017 return (error); 2018 2019 if (hardlink_check_uid) { 2020 if (cred->cr_uid != va.va_uid) 2021 return (EPERM); 2022 } 2023 2024 if (hardlink_check_gid) { 2025 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred)) 2026 return (EPERM); 2027 } 2028 2029 return (0); 2030 } 2031 2032 int 2033 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd) 2034 { 2035 struct thread *td = curthread; 2036 struct vnode *vp; 2037 int error; 2038 2039 /* 2040 * Lookup the source and obtained a locked vnode. 2041 * 2042 * You may only hardlink a file which you have write permission 2043 * on or which you own. 2044 * 2045 * XXX relookup on vget failure / race ? 2046 */ 2047 bwillinode(1); 2048 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK; 2049 if ((error = nlookup(nd)) != 0) 2050 return (error); 2051 vp = nd->nl_nch.ncp->nc_vp; 2052 KKASSERT(vp != NULL); 2053 if (vp->v_type == VDIR) 2054 return (EPERM); /* POSIX */ 2055 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2056 return (error); 2057 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) 2058 return (error); 2059 2060 /* 2061 * Unlock the source so we can lookup the target without deadlocking 2062 * (XXX vp is locked already, possible other deadlock?). The target 2063 * must not exist. 2064 */ 2065 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 2066 nd->nl_flags &= ~NLC_NCPISLOCKED; 2067 cache_unlock(&nd->nl_nch); 2068 2069 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2070 if ((error = nlookup(linknd)) != 0) { 2071 vput(vp); 2072 return (error); 2073 } 2074 if (linknd->nl_nch.ncp->nc_vp) { 2075 vput(vp); 2076 return (EEXIST); 2077 } 2078 2079 /* 2080 * Finally run the new API VOP. 2081 */ 2082 error = can_hardlink(vp, td, td->td_proc->p_ucred); 2083 if (error == 0) { 2084 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp, 2085 vp, linknd->nl_cred); 2086 } 2087 vput(vp); 2088 return (error); 2089 } 2090 2091 /* 2092 * link_args(char *path, char *link) 2093 * 2094 * Make a hard file link. 2095 */ 2096 int 2097 sys_link(struct link_args *uap) 2098 { 2099 struct nlookupdata nd, linknd; 2100 int error; 2101 2102 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2103 if (error == 0) { 2104 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0); 2105 if (error == 0) 2106 error = kern_link(&nd, &linknd); 2107 nlookup_done(&linknd); 2108 } 2109 nlookup_done(&nd); 2110 return (error); 2111 } 2112 2113 int 2114 kern_symlink(struct nlookupdata *nd, char *path, int mode) 2115 { 2116 struct vattr vattr; 2117 struct vnode *vp; 2118 struct vnode *dvp; 2119 int error; 2120 2121 bwillinode(1); 2122 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2123 if ((error = nlookup(nd)) != 0) 2124 return (error); 2125 if (nd->nl_nch.ncp->nc_vp) 2126 return (EEXIST); 2127 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2128 return (error); 2129 dvp = nd->nl_dvp; 2130 VATTR_NULL(&vattr); 2131 vattr.va_mode = mode; 2132 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path); 2133 if (error == 0) 2134 vput(vp); 2135 return (error); 2136 } 2137 2138 /* 2139 * symlink(char *path, char *link) 2140 * 2141 * Make a symbolic link. 2142 */ 2143 int 2144 sys_symlink(struct symlink_args *uap) 2145 { 2146 struct thread *td = curthread; 2147 struct nlookupdata nd; 2148 char *path; 2149 int error; 2150 int mode; 2151 2152 path = objcache_get(namei_oc, M_WAITOK); 2153 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 2154 if (error == 0) { 2155 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0); 2156 if (error == 0) { 2157 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2158 error = kern_symlink(&nd, path, mode); 2159 } 2160 nlookup_done(&nd); 2161 } 2162 objcache_put(namei_oc, path); 2163 return (error); 2164 } 2165 2166 /* 2167 * undelete_args(char *path) 2168 * 2169 * Delete a whiteout from the filesystem. 2170 */ 2171 /* ARGSUSED */ 2172 int 2173 sys_undelete(struct undelete_args *uap) 2174 { 2175 struct nlookupdata nd; 2176 int error; 2177 2178 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2179 bwillinode(1); 2180 nd.nl_flags |= NLC_DELETE | NLC_REFDVP; 2181 if (error == 0) 2182 error = nlookup(&nd); 2183 if (error == 0) 2184 error = ncp_writechk(&nd.nl_nch); 2185 if (error == 0) { 2186 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred, 2187 NAMEI_DELETE); 2188 } 2189 nlookup_done(&nd); 2190 return (error); 2191 } 2192 2193 int 2194 kern_unlink(struct nlookupdata *nd) 2195 { 2196 int error; 2197 2198 bwillinode(1); 2199 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 2200 if ((error = nlookup(nd)) != 0) 2201 return (error); 2202 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2203 return (error); 2204 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 2205 return (error); 2206 } 2207 2208 /* 2209 * unlink_args(char *path) 2210 * 2211 * Delete a name from the filesystem. 2212 */ 2213 int 2214 sys_unlink(struct unlink_args *uap) 2215 { 2216 struct nlookupdata nd; 2217 int error; 2218 2219 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2220 if (error == 0) 2221 error = kern_unlink(&nd); 2222 nlookup_done(&nd); 2223 return (error); 2224 } 2225 2226 int 2227 kern_lseek(int fd, off_t offset, int whence, off_t *res) 2228 { 2229 struct thread *td = curthread; 2230 struct proc *p = td->td_proc; 2231 struct file *fp; 2232 struct vnode *vp; 2233 struct vattr vattr; 2234 off_t new_offset; 2235 int error; 2236 2237 fp = holdfp(p->p_fd, fd, -1); 2238 if (fp == NULL) 2239 return (EBADF); 2240 if (fp->f_type != DTYPE_VNODE) { 2241 error = ESPIPE; 2242 goto done; 2243 } 2244 vp = (struct vnode *)fp->f_data; 2245 2246 switch (whence) { 2247 case L_INCR: 2248 new_offset = fp->f_offset + offset; 2249 error = 0; 2250 break; 2251 case L_XTND: 2252 error = VOP_GETATTR(vp, &vattr); 2253 new_offset = offset + vattr.va_size; 2254 break; 2255 case L_SET: 2256 new_offset = offset; 2257 error = 0; 2258 break; 2259 default: 2260 new_offset = 0; 2261 error = EINVAL; 2262 break; 2263 } 2264 2265 /* 2266 * Validate the seek position. Negative offsets are not allowed 2267 * for regular files or directories. 2268 * 2269 * Normally we would also not want to allow negative offsets for 2270 * character and block-special devices. However kvm addresses 2271 * on 64 bit architectures might appear to be negative and must 2272 * be allowed. 2273 */ 2274 if (error == 0) { 2275 if (new_offset < 0 && 2276 (vp->v_type == VREG || vp->v_type == VDIR)) { 2277 error = EINVAL; 2278 } else { 2279 fp->f_offset = new_offset; 2280 } 2281 } 2282 *res = fp->f_offset; 2283 done: 2284 fdrop(fp); 2285 return (error); 2286 } 2287 2288 /* 2289 * lseek_args(int fd, int pad, off_t offset, int whence) 2290 * 2291 * Reposition read/write file offset. 2292 */ 2293 int 2294 sys_lseek(struct lseek_args *uap) 2295 { 2296 int error; 2297 2298 error = kern_lseek(uap->fd, uap->offset, uap->whence, 2299 &uap->sysmsg_offset); 2300 2301 return (error); 2302 } 2303 2304 int 2305 kern_access(struct nlookupdata *nd, int aflags) 2306 { 2307 struct vnode *vp; 2308 int error, flags; 2309 2310 if ((error = nlookup(nd)) != 0) 2311 return (error); 2312 retry: 2313 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp); 2314 if (error) 2315 return (error); 2316 2317 /* Flags == 0 means only check for existence. */ 2318 if (aflags) { 2319 flags = 0; 2320 if (aflags & R_OK) 2321 flags |= VREAD; 2322 if (aflags & W_OK) 2323 flags |= VWRITE; 2324 if (aflags & X_OK) 2325 flags |= VEXEC; 2326 if ((flags & VWRITE) == 0 || 2327 (error = vn_writechk(vp, &nd->nl_nch)) == 0) 2328 error = VOP_ACCESS(vp, flags, nd->nl_cred); 2329 2330 /* 2331 * If the file handle is stale we have to re-resolve the 2332 * entry. This is a hack at the moment. 2333 */ 2334 if (error == ESTALE) { 2335 vput(vp); 2336 cache_setunresolved(&nd->nl_nch); 2337 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2338 if (error == 0) { 2339 vp = NULL; 2340 goto retry; 2341 } 2342 return(error); 2343 } 2344 } 2345 vput(vp); 2346 return (error); 2347 } 2348 2349 /* 2350 * access_args(char *path, int flags) 2351 * 2352 * Check access permissions. 2353 */ 2354 int 2355 sys_access(struct access_args *uap) 2356 { 2357 struct nlookupdata nd; 2358 int error; 2359 2360 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2361 if (error == 0) 2362 error = kern_access(&nd, uap->flags); 2363 nlookup_done(&nd); 2364 return (error); 2365 } 2366 2367 int 2368 kern_stat(struct nlookupdata *nd, struct stat *st) 2369 { 2370 int error; 2371 struct vnode *vp; 2372 thread_t td; 2373 2374 if ((error = nlookup(nd)) != 0) 2375 return (error); 2376 again: 2377 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 2378 return (ENOENT); 2379 2380 td = curthread; 2381 if ((error = vget(vp, LK_SHARED)) != 0) 2382 return (error); 2383 error = vn_stat(vp, st, nd->nl_cred); 2384 2385 /* 2386 * If the file handle is stale we have to re-resolve the entry. This 2387 * is a hack at the moment. 2388 */ 2389 if (error == ESTALE) { 2390 vput(vp); 2391 cache_setunresolved(&nd->nl_nch); 2392 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2393 if (error == 0) 2394 goto again; 2395 } else { 2396 vput(vp); 2397 } 2398 return (error); 2399 } 2400 2401 /* 2402 * stat_args(char *path, struct stat *ub) 2403 * 2404 * Get file status; this version follows links. 2405 */ 2406 int 2407 sys_stat(struct stat_args *uap) 2408 { 2409 struct nlookupdata nd; 2410 struct stat st; 2411 int error; 2412 2413 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2414 if (error == 0) { 2415 error = kern_stat(&nd, &st); 2416 if (error == 0) 2417 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 2418 } 2419 nlookup_done(&nd); 2420 return (error); 2421 } 2422 2423 /* 2424 * lstat_args(char *path, struct stat *ub) 2425 * 2426 * Get file status; this version does not follow links. 2427 */ 2428 int 2429 sys_lstat(struct lstat_args *uap) 2430 { 2431 struct nlookupdata nd; 2432 struct stat st; 2433 int error; 2434 2435 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2436 if (error == 0) { 2437 error = kern_stat(&nd, &st); 2438 if (error == 0) 2439 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 2440 } 2441 nlookup_done(&nd); 2442 return (error); 2443 } 2444 2445 /* 2446 * pathconf_Args(char *path, int name) 2447 * 2448 * Get configurable pathname variables. 2449 */ 2450 /* ARGSUSED */ 2451 int 2452 sys_pathconf(struct pathconf_args *uap) 2453 { 2454 struct nlookupdata nd; 2455 struct vnode *vp; 2456 int error; 2457 2458 vp = NULL; 2459 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2460 if (error == 0) 2461 error = nlookup(&nd); 2462 if (error == 0) 2463 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 2464 nlookup_done(&nd); 2465 if (error == 0) { 2466 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 2467 vput(vp); 2468 } 2469 return (error); 2470 } 2471 2472 /* 2473 * XXX: daver 2474 * kern_readlink isn't properly split yet. There is a copyin burried 2475 * in VOP_READLINK(). 2476 */ 2477 int 2478 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res) 2479 { 2480 struct thread *td = curthread; 2481 struct proc *p = td->td_proc; 2482 struct vnode *vp; 2483 struct iovec aiov; 2484 struct uio auio; 2485 int error; 2486 2487 if ((error = nlookup(nd)) != 0) 2488 return (error); 2489 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp); 2490 if (error) 2491 return (error); 2492 if (vp->v_type != VLNK) { 2493 error = EINVAL; 2494 } else { 2495 aiov.iov_base = buf; 2496 aiov.iov_len = count; 2497 auio.uio_iov = &aiov; 2498 auio.uio_iovcnt = 1; 2499 auio.uio_offset = 0; 2500 auio.uio_rw = UIO_READ; 2501 auio.uio_segflg = UIO_USERSPACE; 2502 auio.uio_td = td; 2503 auio.uio_resid = count; 2504 error = VOP_READLINK(vp, &auio, p->p_ucred); 2505 } 2506 vput(vp); 2507 *res = count - auio.uio_resid; 2508 return (error); 2509 } 2510 2511 /* 2512 * readlink_args(char *path, char *buf, int count) 2513 * 2514 * Return target name of a symbolic link. 2515 */ 2516 int 2517 sys_readlink(struct readlink_args *uap) 2518 { 2519 struct nlookupdata nd; 2520 int error; 2521 2522 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2523 if (error == 0) { 2524 error = kern_readlink(&nd, uap->buf, uap->count, 2525 &uap->sysmsg_result); 2526 } 2527 nlookup_done(&nd); 2528 return (error); 2529 } 2530 2531 static int 2532 setfflags(struct vnode *vp, int flags) 2533 { 2534 struct thread *td = curthread; 2535 struct proc *p = td->td_proc; 2536 int error; 2537 struct vattr vattr; 2538 2539 /* 2540 * Prevent non-root users from setting flags on devices. When 2541 * a device is reused, users can retain ownership of the device 2542 * if they are allowed to set flags and programs assume that 2543 * chown can't fail when done as root. 2544 */ 2545 if ((vp->v_type == VCHR || vp->v_type == VBLK) && 2546 ((error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT)) != 0)) 2547 return (error); 2548 2549 /* 2550 * note: vget is required for any operation that might mod the vnode 2551 * so VINACTIVE is properly cleared. 2552 */ 2553 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 2554 VATTR_NULL(&vattr); 2555 vattr.va_flags = flags; 2556 error = VOP_SETATTR(vp, &vattr, p->p_ucred); 2557 vput(vp); 2558 } 2559 return (error); 2560 } 2561 2562 /* 2563 * chflags(char *path, int flags) 2564 * 2565 * Change flags of a file given a path name. 2566 */ 2567 /* ARGSUSED */ 2568 int 2569 sys_chflags(struct chflags_args *uap) 2570 { 2571 struct nlookupdata nd; 2572 struct vnode *vp; 2573 int error; 2574 2575 vp = NULL; 2576 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2577 if (error == 0) 2578 error = nlookup(&nd); 2579 if (error == 0) 2580 error = ncp_writechk(&nd.nl_nch); 2581 if (error == 0) 2582 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 2583 nlookup_done(&nd); 2584 if (error == 0) { 2585 error = setfflags(vp, uap->flags); 2586 vrele(vp); 2587 } 2588 return (error); 2589 } 2590 2591 /* 2592 * lchflags(char *path, int flags) 2593 * 2594 * Change flags of a file given a path name, but don't follow symlinks. 2595 */ 2596 /* ARGSUSED */ 2597 int 2598 sys_lchflags(struct lchflags_args *uap) 2599 { 2600 struct nlookupdata nd; 2601 struct vnode *vp; 2602 int error; 2603 2604 vp = NULL; 2605 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2606 if (error == 0) 2607 error = nlookup(&nd); 2608 if (error == 0) 2609 error = ncp_writechk(&nd.nl_nch); 2610 if (error == 0) 2611 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 2612 nlookup_done(&nd); 2613 if (error == 0) { 2614 error = setfflags(vp, uap->flags); 2615 vrele(vp); 2616 } 2617 return (error); 2618 } 2619 2620 /* 2621 * fchflags_args(int fd, int flags) 2622 * 2623 * Change flags of a file given a file descriptor. 2624 */ 2625 /* ARGSUSED */ 2626 int 2627 sys_fchflags(struct fchflags_args *uap) 2628 { 2629 struct thread *td = curthread; 2630 struct proc *p = td->td_proc; 2631 struct file *fp; 2632 int error; 2633 2634 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 2635 return (error); 2636 if (fp->f_nchandle.ncp) 2637 error = ncp_writechk(&fp->f_nchandle); 2638 if (error == 0) 2639 error = setfflags((struct vnode *) fp->f_data, uap->flags); 2640 fdrop(fp); 2641 return (error); 2642 } 2643 2644 static int 2645 setfmode(struct vnode *vp, int mode) 2646 { 2647 struct thread *td = curthread; 2648 struct proc *p = td->td_proc; 2649 int error; 2650 struct vattr vattr; 2651 2652 /* 2653 * note: vget is required for any operation that might mod the vnode 2654 * so VINACTIVE is properly cleared. 2655 */ 2656 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 2657 VATTR_NULL(&vattr); 2658 vattr.va_mode = mode & ALLPERMS; 2659 error = VOP_SETATTR(vp, &vattr, p->p_ucred); 2660 vput(vp); 2661 } 2662 return error; 2663 } 2664 2665 int 2666 kern_chmod(struct nlookupdata *nd, int mode) 2667 { 2668 struct vnode *vp; 2669 int error; 2670 2671 if ((error = nlookup(nd)) != 0) 2672 return (error); 2673 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 2674 return (error); 2675 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 2676 error = setfmode(vp, mode); 2677 vrele(vp); 2678 return (error); 2679 } 2680 2681 /* 2682 * chmod_args(char *path, int mode) 2683 * 2684 * Change mode of a file given path name. 2685 */ 2686 /* ARGSUSED */ 2687 int 2688 sys_chmod(struct chmod_args *uap) 2689 { 2690 struct nlookupdata nd; 2691 int error; 2692 2693 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2694 if (error == 0) 2695 error = kern_chmod(&nd, uap->mode); 2696 nlookup_done(&nd); 2697 return (error); 2698 } 2699 2700 /* 2701 * lchmod_args(char *path, int mode) 2702 * 2703 * Change mode of a file given path name (don't follow links.) 2704 */ 2705 /* ARGSUSED */ 2706 int 2707 sys_lchmod(struct lchmod_args *uap) 2708 { 2709 struct nlookupdata nd; 2710 int error; 2711 2712 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2713 if (error == 0) 2714 error = kern_chmod(&nd, uap->mode); 2715 nlookup_done(&nd); 2716 return (error); 2717 } 2718 2719 /* 2720 * fchmod_args(int fd, int mode) 2721 * 2722 * Change mode of a file given a file descriptor. 2723 */ 2724 /* ARGSUSED */ 2725 int 2726 sys_fchmod(struct fchmod_args *uap) 2727 { 2728 struct thread *td = curthread; 2729 struct proc *p = td->td_proc; 2730 struct file *fp; 2731 int error; 2732 2733 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 2734 return (error); 2735 if (fp->f_nchandle.ncp) 2736 error = ncp_writechk(&fp->f_nchandle); 2737 if (error == 0) 2738 error = setfmode((struct vnode *)fp->f_data, uap->mode); 2739 fdrop(fp); 2740 return (error); 2741 } 2742 2743 static int 2744 setfown(struct vnode *vp, uid_t uid, gid_t gid) 2745 { 2746 struct thread *td = curthread; 2747 struct proc *p = td->td_proc; 2748 int error; 2749 struct vattr vattr; 2750 2751 /* 2752 * note: vget is required for any operation that might mod the vnode 2753 * so VINACTIVE is properly cleared. 2754 */ 2755 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 2756 VATTR_NULL(&vattr); 2757 vattr.va_uid = uid; 2758 vattr.va_gid = gid; 2759 error = VOP_SETATTR(vp, &vattr, p->p_ucred); 2760 vput(vp); 2761 } 2762 return error; 2763 } 2764 2765 int 2766 kern_chown(struct nlookupdata *nd, int uid, int gid) 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 = setfown(vp, uid, gid); 2777 vrele(vp); 2778 return (error); 2779 } 2780 2781 /* 2782 * chown(char *path, int uid, int gid) 2783 * 2784 * Set ownership given a path name. 2785 */ 2786 int 2787 sys_chown(struct chown_args *uap) 2788 { 2789 struct nlookupdata nd; 2790 int error; 2791 2792 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2793 if (error == 0) 2794 error = kern_chown(&nd, uap->uid, uap->gid); 2795 nlookup_done(&nd); 2796 return (error); 2797 } 2798 2799 /* 2800 * lchown_args(char *path, int uid, int gid) 2801 * 2802 * Set ownership given a path name, do not cross symlinks. 2803 */ 2804 int 2805 sys_lchown(struct lchown_args *uap) 2806 { 2807 struct nlookupdata nd; 2808 int error; 2809 2810 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2811 if (error == 0) 2812 error = kern_chown(&nd, uap->uid, uap->gid); 2813 nlookup_done(&nd); 2814 return (error); 2815 } 2816 2817 /* 2818 * fchown_args(int fd, int uid, int gid) 2819 * 2820 * Set ownership given a file descriptor. 2821 */ 2822 /* ARGSUSED */ 2823 int 2824 sys_fchown(struct fchown_args *uap) 2825 { 2826 struct thread *td = curthread; 2827 struct proc *p = td->td_proc; 2828 struct file *fp; 2829 int error; 2830 2831 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 2832 return (error); 2833 if (fp->f_nchandle.ncp) 2834 error = ncp_writechk(&fp->f_nchandle); 2835 if (error == 0) 2836 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid); 2837 fdrop(fp); 2838 return (error); 2839 } 2840 2841 static int 2842 getutimes(const struct timeval *tvp, struct timespec *tsp) 2843 { 2844 struct timeval tv[2]; 2845 2846 if (tvp == NULL) { 2847 microtime(&tv[0]); 2848 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]); 2849 tsp[1] = tsp[0]; 2850 } else { 2851 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); 2852 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); 2853 } 2854 return 0; 2855 } 2856 2857 static int 2858 setutimes(struct vnode *vp, struct vattr *vattr, 2859 const struct timespec *ts, int nullflag) 2860 { 2861 struct thread *td = curthread; 2862 struct proc *p = td->td_proc; 2863 int error; 2864 2865 VATTR_NULL(vattr); 2866 vattr->va_atime = ts[0]; 2867 vattr->va_mtime = ts[1]; 2868 if (nullflag) 2869 vattr->va_vaflags |= VA_UTIMES_NULL; 2870 error = VOP_SETATTR(vp, vattr, p->p_ucred); 2871 2872 return error; 2873 } 2874 2875 int 2876 kern_utimes(struct nlookupdata *nd, struct timeval *tptr) 2877 { 2878 struct timespec ts[2]; 2879 struct vnode *vp; 2880 struct vattr vattr; 2881 int error; 2882 2883 if ((error = getutimes(tptr, ts)) != 0) 2884 return (error); 2885 2886 /* 2887 * NOTE: utimes() succeeds for the owner even if the file 2888 * is not user-writable. 2889 */ 2890 nd->nl_flags |= NLC_OWN | NLC_WRITE; 2891 2892 if ((error = nlookup(nd)) != 0) 2893 return (error); 2894 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2895 return (error); 2896 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 2897 return (error); 2898 2899 /* 2900 * note: vget is required for any operation that might mod the vnode 2901 * so VINACTIVE is properly cleared. 2902 */ 2903 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) { 2904 error = vget(vp, LK_EXCLUSIVE); 2905 if (error == 0) { 2906 error = setutimes(vp, &vattr, ts, (tptr == NULL)); 2907 vput(vp); 2908 } 2909 } 2910 vrele(vp); 2911 return (error); 2912 } 2913 2914 /* 2915 * utimes_args(char *path, struct timeval *tptr) 2916 * 2917 * Set the access and modification times of a file. 2918 */ 2919 int 2920 sys_utimes(struct utimes_args *uap) 2921 { 2922 struct timeval tv[2]; 2923 struct nlookupdata nd; 2924 int error; 2925 2926 if (uap->tptr) { 2927 error = copyin(uap->tptr, tv, sizeof(tv)); 2928 if (error) 2929 return (error); 2930 } 2931 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2932 if (error == 0) 2933 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 2934 nlookup_done(&nd); 2935 return (error); 2936 } 2937 2938 /* 2939 * lutimes_args(char *path, struct timeval *tptr) 2940 * 2941 * Set the access and modification times of a file. 2942 */ 2943 int 2944 sys_lutimes(struct lutimes_args *uap) 2945 { 2946 struct timeval tv[2]; 2947 struct nlookupdata nd; 2948 int error; 2949 2950 if (uap->tptr) { 2951 error = copyin(uap->tptr, tv, sizeof(tv)); 2952 if (error) 2953 return (error); 2954 } 2955 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2956 if (error == 0) 2957 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 2958 nlookup_done(&nd); 2959 return (error); 2960 } 2961 2962 /* 2963 * Set utimes on a file descriptor. The creds used to open the 2964 * file are used to determine whether the operation is allowed 2965 * or not. 2966 */ 2967 int 2968 kern_futimes(int fd, struct timeval *tptr) 2969 { 2970 struct thread *td = curthread; 2971 struct proc *p = td->td_proc; 2972 struct timespec ts[2]; 2973 struct file *fp; 2974 struct vnode *vp; 2975 struct vattr vattr; 2976 int error; 2977 2978 error = getutimes(tptr, ts); 2979 if (error) 2980 return (error); 2981 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 2982 return (error); 2983 if (fp->f_nchandle.ncp) 2984 error = ncp_writechk(&fp->f_nchandle); 2985 if (error == 0) { 2986 vp = fp->f_data; 2987 error = vget(vp, LK_EXCLUSIVE); 2988 if (error == 0) { 2989 error = VOP_GETATTR(vp, &vattr); 2990 if (error == 0) { 2991 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE, 2992 fp->f_cred); 2993 } 2994 if (error == 0) { 2995 error = setutimes(vp, &vattr, ts, 2996 (tptr == NULL)); 2997 } 2998 vput(vp); 2999 } 3000 } 3001 fdrop(fp); 3002 return (error); 3003 } 3004 3005 /* 3006 * futimes_args(int fd, struct timeval *tptr) 3007 * 3008 * Set the access and modification times of a file. 3009 */ 3010 int 3011 sys_futimes(struct futimes_args *uap) 3012 { 3013 struct timeval tv[2]; 3014 int error; 3015 3016 if (uap->tptr) { 3017 error = copyin(uap->tptr, tv, sizeof(tv)); 3018 if (error) 3019 return (error); 3020 } 3021 3022 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL); 3023 3024 return (error); 3025 } 3026 3027 int 3028 kern_truncate(struct nlookupdata *nd, off_t length) 3029 { 3030 struct vnode *vp; 3031 struct vattr vattr; 3032 int error; 3033 3034 if (length < 0) 3035 return(EINVAL); 3036 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE; 3037 if ((error = nlookup(nd)) != 0) 3038 return (error); 3039 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3040 return (error); 3041 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3042 return (error); 3043 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) { 3044 vrele(vp); 3045 return (error); 3046 } 3047 if (vp->v_type == VDIR) { 3048 error = EISDIR; 3049 } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) { 3050 VATTR_NULL(&vattr); 3051 vattr.va_size = length; 3052 error = VOP_SETATTR(vp, &vattr, nd->nl_cred); 3053 } 3054 vput(vp); 3055 return (error); 3056 } 3057 3058 /* 3059 * truncate(char *path, int pad, off_t length) 3060 * 3061 * Truncate a file given its path name. 3062 */ 3063 int 3064 sys_truncate(struct truncate_args *uap) 3065 { 3066 struct nlookupdata nd; 3067 int error; 3068 3069 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3070 if (error == 0) 3071 error = kern_truncate(&nd, uap->length); 3072 nlookup_done(&nd); 3073 return error; 3074 } 3075 3076 int 3077 kern_ftruncate(int fd, off_t length) 3078 { 3079 struct thread *td = curthread; 3080 struct proc *p = td->td_proc; 3081 struct vattr vattr; 3082 struct vnode *vp; 3083 struct file *fp; 3084 int error; 3085 3086 if (length < 0) 3087 return(EINVAL); 3088 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 3089 return (error); 3090 if (fp->f_nchandle.ncp) { 3091 error = ncp_writechk(&fp->f_nchandle); 3092 if (error) 3093 goto done; 3094 } 3095 if ((fp->f_flag & FWRITE) == 0) { 3096 error = EINVAL; 3097 goto done; 3098 } 3099 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */ 3100 error = EINVAL; 3101 goto done; 3102 } 3103 vp = (struct vnode *)fp->f_data; 3104 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3105 if (vp->v_type == VDIR) { 3106 error = EISDIR; 3107 } else if ((error = vn_writechk(vp, NULL)) == 0) { 3108 VATTR_NULL(&vattr); 3109 vattr.va_size = length; 3110 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 3111 } 3112 vn_unlock(vp); 3113 done: 3114 fdrop(fp); 3115 return (error); 3116 } 3117 3118 /* 3119 * ftruncate_args(int fd, int pad, off_t length) 3120 * 3121 * Truncate a file given a file descriptor. 3122 */ 3123 int 3124 sys_ftruncate(struct ftruncate_args *uap) 3125 { 3126 int error; 3127 3128 error = kern_ftruncate(uap->fd, uap->length); 3129 3130 return (error); 3131 } 3132 3133 /* 3134 * fsync(int fd) 3135 * 3136 * Sync an open file. 3137 */ 3138 /* ARGSUSED */ 3139 int 3140 sys_fsync(struct fsync_args *uap) 3141 { 3142 struct thread *td = curthread; 3143 struct proc *p = td->td_proc; 3144 struct vnode *vp; 3145 struct file *fp; 3146 vm_object_t obj; 3147 int error; 3148 3149 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3150 return (error); 3151 vp = (struct vnode *)fp->f_data; 3152 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3153 if ((obj = vp->v_object) != NULL) 3154 vm_object_page_clean(obj, 0, 0, 0); 3155 if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount) 3156 error = buf_fsync(vp); 3157 vn_unlock(vp); 3158 fdrop(fp); 3159 return (error); 3160 } 3161 3162 int 3163 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond) 3164 { 3165 struct nchandle fnchd; 3166 struct nchandle tnchd; 3167 struct namecache *ncp; 3168 struct vnode *fdvp; 3169 struct vnode *tdvp; 3170 struct mount *mp; 3171 int error; 3172 3173 bwillinode(1); 3174 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC; 3175 if ((error = nlookup(fromnd)) != 0) 3176 return (error); 3177 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL) 3178 return (ENOENT); 3179 fnchd.mount = fromnd->nl_nch.mount; 3180 cache_hold(&fnchd); 3181 3182 /* 3183 * unlock the source nch so we can lookup the target nch without 3184 * deadlocking. The target may or may not exist so we do not check 3185 * for a target vp like kern_mkdir() and other creation functions do. 3186 * 3187 * The source and target directories are ref'd and rechecked after 3188 * everything is relocked to determine if the source or target file 3189 * has been renamed. 3190 */ 3191 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED); 3192 fromnd->nl_flags &= ~NLC_NCPISLOCKED; 3193 cache_unlock(&fromnd->nl_nch); 3194 3195 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP; 3196 if ((error = nlookup(tond)) != 0) { 3197 cache_drop(&fnchd); 3198 return (error); 3199 } 3200 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) { 3201 cache_drop(&fnchd); 3202 return (ENOENT); 3203 } 3204 tnchd.mount = tond->nl_nch.mount; 3205 cache_hold(&tnchd); 3206 3207 /* 3208 * If the source and target are the same there is nothing to do 3209 */ 3210 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) { 3211 cache_drop(&fnchd); 3212 cache_drop(&tnchd); 3213 return (0); 3214 } 3215 3216 /* 3217 * Mount points cannot be renamed or overwritten 3218 */ 3219 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) & 3220 NCF_ISMOUNTPT 3221 ) { 3222 cache_drop(&fnchd); 3223 cache_drop(&tnchd); 3224 return (EINVAL); 3225 } 3226 3227 /* 3228 * relock the source ncp. NOTE AFTER RELOCKING: the source ncp 3229 * may have become invalid while it was unlocked, nc_vp and nc_mount 3230 * could be NULL. 3231 */ 3232 if (cache_lock_nonblock(&fromnd->nl_nch) == 0) { 3233 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred); 3234 } else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) { 3235 cache_lock(&fromnd->nl_nch); 3236 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred); 3237 } else { 3238 cache_unlock(&tond->nl_nch); 3239 cache_lock(&fromnd->nl_nch); 3240 cache_resolve(&fromnd->nl_nch, fromnd->nl_cred); 3241 cache_lock(&tond->nl_nch); 3242 cache_resolve(&tond->nl_nch, tond->nl_cred); 3243 } 3244 fromnd->nl_flags |= NLC_NCPISLOCKED; 3245 3246 /* 3247 * make sure the parent directories linkages are the same 3248 */ 3249 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent || 3250 tnchd.ncp != tond->nl_nch.ncp->nc_parent) { 3251 cache_drop(&fnchd); 3252 cache_drop(&tnchd); 3253 return (ENOENT); 3254 } 3255 3256 /* 3257 * Both the source and target must be within the same filesystem and 3258 * in the same filesystem as their parent directories within the 3259 * namecache topology. 3260 * 3261 * NOTE: fromnd's nc_mount or nc_vp could be NULL. 3262 */ 3263 mp = fnchd.mount; 3264 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount || 3265 mp != tond->nl_nch.mount) { 3266 cache_drop(&fnchd); 3267 cache_drop(&tnchd); 3268 return (EXDEV); 3269 } 3270 3271 /* 3272 * Make sure the mount point is writable 3273 */ 3274 if ((error = ncp_writechk(&tond->nl_nch)) != 0) { 3275 cache_drop(&fnchd); 3276 cache_drop(&tnchd); 3277 return (error); 3278 } 3279 3280 /* 3281 * If the target exists and either the source or target is a directory, 3282 * then both must be directories. 3283 * 3284 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h 3285 * have become NULL. 3286 */ 3287 if (tond->nl_nch.ncp->nc_vp) { 3288 if (fromnd->nl_nch.ncp->nc_vp == NULL) { 3289 error = ENOENT; 3290 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 3291 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR) 3292 error = ENOTDIR; 3293 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) { 3294 error = EISDIR; 3295 } 3296 } 3297 3298 /* 3299 * You cannot rename a source into itself or a subdirectory of itself. 3300 * We check this by travsersing the target directory upwards looking 3301 * for a match against the source. 3302 */ 3303 if (error == 0) { 3304 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) { 3305 if (fromnd->nl_nch.ncp == ncp) { 3306 error = EINVAL; 3307 break; 3308 } 3309 } 3310 } 3311 3312 cache_drop(&fnchd); 3313 cache_drop(&tnchd); 3314 3315 /* 3316 * Even though the namespaces are different, they may still represent 3317 * hardlinks to the same file. The filesystem might have a hard time 3318 * with this so we issue a NREMOVE of the source instead of a NRENAME 3319 * when we detect the situation. 3320 */ 3321 if (error == 0) { 3322 fdvp = fromnd->nl_dvp; 3323 tdvp = tond->nl_dvp; 3324 if (fdvp == NULL || tdvp == NULL) { 3325 error = EPERM; 3326 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) { 3327 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp, 3328 fromnd->nl_cred); 3329 } else { 3330 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch, 3331 fdvp, tdvp, tond->nl_cred); 3332 } 3333 } 3334 return (error); 3335 } 3336 3337 /* 3338 * rename_args(char *from, char *to) 3339 * 3340 * Rename files. Source and destination must either both be directories, 3341 * or both not be directories. If target is a directory, it must be empty. 3342 */ 3343 int 3344 sys_rename(struct rename_args *uap) 3345 { 3346 struct nlookupdata fromnd, tond; 3347 int error; 3348 3349 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0); 3350 if (error == 0) { 3351 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0); 3352 if (error == 0) 3353 error = kern_rename(&fromnd, &tond); 3354 nlookup_done(&tond); 3355 } 3356 nlookup_done(&fromnd); 3357 return (error); 3358 } 3359 3360 int 3361 kern_mkdir(struct nlookupdata *nd, int mode) 3362 { 3363 struct thread *td = curthread; 3364 struct proc *p = td->td_proc; 3365 struct vnode *vp; 3366 struct vattr vattr; 3367 int error; 3368 3369 bwillinode(1); 3370 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP; 3371 if ((error = nlookup(nd)) != 0) 3372 return (error); 3373 3374 if (nd->nl_nch.ncp->nc_vp) 3375 return (EEXIST); 3376 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3377 return (error); 3378 VATTR_NULL(&vattr); 3379 vattr.va_type = VDIR; 3380 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask; 3381 3382 vp = NULL; 3383 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr); 3384 if (error == 0) 3385 vput(vp); 3386 return (error); 3387 } 3388 3389 /* 3390 * mkdir_args(char *path, int mode) 3391 * 3392 * Make a directory file. 3393 */ 3394 /* ARGSUSED */ 3395 int 3396 sys_mkdir(struct mkdir_args *uap) 3397 { 3398 struct nlookupdata nd; 3399 int error; 3400 3401 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3402 if (error == 0) 3403 error = kern_mkdir(&nd, uap->mode); 3404 nlookup_done(&nd); 3405 return (error); 3406 } 3407 3408 int 3409 kern_rmdir(struct nlookupdata *nd) 3410 { 3411 int error; 3412 3413 bwillinode(1); 3414 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 3415 if ((error = nlookup(nd)) != 0) 3416 return (error); 3417 3418 /* 3419 * Do not allow directories representing mount points to be 3420 * deleted, even if empty. Check write perms on mount point 3421 * in case the vnode is aliased (aka nullfs). 3422 */ 3423 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT)) 3424 return (EINVAL); 3425 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3426 return (error); 3427 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 3428 return (error); 3429 } 3430 3431 /* 3432 * rmdir_args(char *path) 3433 * 3434 * Remove a directory file. 3435 */ 3436 /* ARGSUSED */ 3437 int 3438 sys_rmdir(struct rmdir_args *uap) 3439 { 3440 struct nlookupdata nd; 3441 int error; 3442 3443 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3444 if (error == 0) 3445 error = kern_rmdir(&nd); 3446 nlookup_done(&nd); 3447 return (error); 3448 } 3449 3450 int 3451 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res, 3452 enum uio_seg direction) 3453 { 3454 struct thread *td = curthread; 3455 struct proc *p = td->td_proc; 3456 struct vnode *vp; 3457 struct file *fp; 3458 struct uio auio; 3459 struct iovec aiov; 3460 off_t loff; 3461 int error, eofflag; 3462 3463 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 3464 return (error); 3465 if ((fp->f_flag & FREAD) == 0) { 3466 error = EBADF; 3467 goto done; 3468 } 3469 vp = (struct vnode *)fp->f_data; 3470 unionread: 3471 if (vp->v_type != VDIR) { 3472 error = EINVAL; 3473 goto done; 3474 } 3475 aiov.iov_base = buf; 3476 aiov.iov_len = count; 3477 auio.uio_iov = &aiov; 3478 auio.uio_iovcnt = 1; 3479 auio.uio_rw = UIO_READ; 3480 auio.uio_segflg = direction; 3481 auio.uio_td = td; 3482 auio.uio_resid = count; 3483 loff = auio.uio_offset = fp->f_offset; 3484 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); 3485 fp->f_offset = auio.uio_offset; 3486 if (error) 3487 goto done; 3488 if (count == auio.uio_resid) { 3489 if (union_dircheckp) { 3490 error = union_dircheckp(td, &vp, fp); 3491 if (error == -1) 3492 goto unionread; 3493 if (error) 3494 goto done; 3495 } 3496 #if 0 3497 if ((vp->v_flag & VROOT) && 3498 (vp->v_mount->mnt_flag & MNT_UNION)) { 3499 struct vnode *tvp = vp; 3500 vp = vp->v_mount->mnt_vnodecovered; 3501 vref(vp); 3502 fp->f_data = vp; 3503 fp->f_offset = 0; 3504 vrele(tvp); 3505 goto unionread; 3506 } 3507 #endif 3508 } 3509 3510 /* 3511 * WARNING! *basep may not be wide enough to accomodate the 3512 * seek offset. XXX should we hack this to return the upper 32 bits 3513 * for offsets greater then 4G? 3514 */ 3515 if (basep) { 3516 *basep = (long)loff; 3517 } 3518 *res = count - auio.uio_resid; 3519 done: 3520 fdrop(fp); 3521 return (error); 3522 } 3523 3524 /* 3525 * getdirentries_args(int fd, char *buf, u_int conut, long *basep) 3526 * 3527 * Read a block of directory entries in a file system independent format. 3528 */ 3529 int 3530 sys_getdirentries(struct getdirentries_args *uap) 3531 { 3532 long base; 3533 int error; 3534 3535 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base, 3536 &uap->sysmsg_result, UIO_USERSPACE); 3537 3538 if (error == 0 && uap->basep) 3539 error = copyout(&base, uap->basep, sizeof(*uap->basep)); 3540 return (error); 3541 } 3542 3543 /* 3544 * getdents_args(int fd, char *buf, size_t count) 3545 */ 3546 int 3547 sys_getdents(struct getdents_args *uap) 3548 { 3549 int error; 3550 3551 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL, 3552 &uap->sysmsg_result, UIO_USERSPACE); 3553 3554 return (error); 3555 } 3556 3557 /* 3558 * umask(int newmask) 3559 * 3560 * Set the mode mask for creation of filesystem nodes. 3561 * 3562 * MP SAFE 3563 */ 3564 int 3565 sys_umask(struct umask_args *uap) 3566 { 3567 struct thread *td = curthread; 3568 struct proc *p = td->td_proc; 3569 struct filedesc *fdp; 3570 3571 fdp = p->p_fd; 3572 uap->sysmsg_result = fdp->fd_cmask; 3573 fdp->fd_cmask = uap->newmask & ALLPERMS; 3574 return (0); 3575 } 3576 3577 /* 3578 * revoke(char *path) 3579 * 3580 * Void all references to file by ripping underlying filesystem 3581 * away from vnode. 3582 */ 3583 /* ARGSUSED */ 3584 int 3585 sys_revoke(struct revoke_args *uap) 3586 { 3587 struct nlookupdata nd; 3588 struct vattr vattr; 3589 struct vnode *vp; 3590 struct ucred *cred; 3591 int error; 3592 3593 vp = NULL; 3594 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3595 if (error == 0) 3596 error = nlookup(&nd); 3597 if (error == 0) 3598 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3599 cred = crhold(nd.nl_cred); 3600 nlookup_done(&nd); 3601 if (error == 0) { 3602 if (error == 0) 3603 error = VOP_GETATTR(vp, &vattr); 3604 if (error == 0 && cred->cr_uid != vattr.va_uid) 3605 error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT); 3606 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) { 3607 if (count_udev(vp->v_umajor, vp->v_uminor) > 0) 3608 error = vrevoke(vp, cred); 3609 } else if (error == 0) { 3610 error = vrevoke(vp, cred); 3611 } 3612 vrele(vp); 3613 } 3614 if (cred) 3615 crfree(cred); 3616 return (error); 3617 } 3618 3619 /* 3620 * getfh_args(char *fname, fhandle_t *fhp) 3621 * 3622 * Get (NFS) file handle 3623 * 3624 * NOTE: We use the fsid of the covering mount, even if it is a nullfs 3625 * mount. This allows nullfs mounts to be explicitly exported. 3626 * 3627 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe. 3628 * 3629 * nullfs mounts of subdirectories are not safe. That is, it will 3630 * work, but you do not really have protection against access to 3631 * the related parent directories. 3632 */ 3633 int 3634 sys_getfh(struct getfh_args *uap) 3635 { 3636 struct thread *td = curthread; 3637 struct nlookupdata nd; 3638 fhandle_t fh; 3639 struct vnode *vp; 3640 struct mount *mp; 3641 int error; 3642 3643 /* 3644 * Must be super user 3645 */ 3646 if ((error = priv_check(td, PRIV_ROOT)) != 0) 3647 return (error); 3648 3649 vp = NULL; 3650 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 3651 if (error == 0) 3652 error = nlookup(&nd); 3653 if (error == 0) 3654 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 3655 mp = nd.nl_nch.mount; 3656 nlookup_done(&nd); 3657 if (error == 0) { 3658 bzero(&fh, sizeof(fh)); 3659 fh.fh_fsid = mp->mnt_stat.f_fsid; 3660 error = VFS_VPTOFH(vp, &fh.fh_fid); 3661 vput(vp); 3662 if (error == 0) 3663 error = copyout(&fh, uap->fhp, sizeof(fh)); 3664 } 3665 return (error); 3666 } 3667 3668 /* 3669 * fhopen_args(const struct fhandle *u_fhp, int flags) 3670 * 3671 * syscall for the rpc.lockd to use to translate a NFS file handle into 3672 * an open descriptor. 3673 * 3674 * warning: do not remove the priv_check() call or this becomes one giant 3675 * security hole. 3676 */ 3677 int 3678 sys_fhopen(struct fhopen_args *uap) 3679 { 3680 struct thread *td = curthread; 3681 struct proc *p = td->td_proc; 3682 struct mount *mp; 3683 struct vnode *vp; 3684 struct fhandle fhp; 3685 struct vattr vat; 3686 struct vattr *vap = &vat; 3687 struct flock lf; 3688 int fmode, mode, error, type; 3689 struct file *nfp; 3690 struct file *fp; 3691 int indx; 3692 3693 /* 3694 * Must be super user 3695 */ 3696 error = priv_check(td, PRIV_ROOT); 3697 if (error) 3698 return (error); 3699 3700 fmode = FFLAGS(uap->flags); 3701 /* why not allow a non-read/write open for our lockd? */ 3702 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) 3703 return (EINVAL); 3704 error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); 3705 if (error) 3706 return(error); 3707 /* find the mount point */ 3708 mp = vfs_getvfs(&fhp.fh_fsid); 3709 if (mp == NULL) 3710 return (ESTALE); 3711 /* now give me my vnode, it gets returned to me locked */ 3712 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp); 3713 if (error) 3714 return (error); 3715 /* 3716 * from now on we have to make sure not 3717 * to forget about the vnode 3718 * any error that causes an abort must vput(vp) 3719 * just set error = err and 'goto bad;'. 3720 */ 3721 3722 /* 3723 * from vn_open 3724 */ 3725 if (vp->v_type == VLNK) { 3726 error = EMLINK; 3727 goto bad; 3728 } 3729 if (vp->v_type == VSOCK) { 3730 error = EOPNOTSUPP; 3731 goto bad; 3732 } 3733 mode = 0; 3734 if (fmode & (FWRITE | O_TRUNC)) { 3735 if (vp->v_type == VDIR) { 3736 error = EISDIR; 3737 goto bad; 3738 } 3739 error = vn_writechk(vp, NULL); 3740 if (error) 3741 goto bad; 3742 mode |= VWRITE; 3743 } 3744 if (fmode & FREAD) 3745 mode |= VREAD; 3746 if (mode) { 3747 error = VOP_ACCESS(vp, mode, p->p_ucred); 3748 if (error) 3749 goto bad; 3750 } 3751 if (fmode & O_TRUNC) { 3752 vn_unlock(vp); /* XXX */ 3753 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 3754 VATTR_NULL(vap); 3755 vap->va_size = 0; 3756 error = VOP_SETATTR(vp, vap, p->p_ucred); 3757 if (error) 3758 goto bad; 3759 } 3760 3761 /* 3762 * VOP_OPEN needs the file pointer so it can potentially override 3763 * it. 3764 * 3765 * WARNING! no f_nchandle will be associated when fhopen()ing a 3766 * directory. XXX 3767 */ 3768 if ((error = falloc(p, &nfp, &indx)) != 0) 3769 goto bad; 3770 fp = nfp; 3771 3772 error = VOP_OPEN(vp, fmode, p->p_ucred, fp); 3773 if (error) { 3774 /* 3775 * setting f_ops this way prevents VOP_CLOSE from being 3776 * called or fdrop() releasing the vp from v_data. Since 3777 * the VOP_OPEN failed we don't want to VOP_CLOSE. 3778 */ 3779 fp->f_ops = &badfileops; 3780 fp->f_data = NULL; 3781 goto bad_drop; 3782 } 3783 3784 /* 3785 * The fp is given its own reference, we still have our ref and lock. 3786 * 3787 * Assert that all regular files must be created with a VM object. 3788 */ 3789 if (vp->v_type == VREG && vp->v_object == NULL) { 3790 kprintf("fhopen: regular file did not have VM object: %p\n", vp); 3791 goto bad_drop; 3792 } 3793 3794 /* 3795 * The open was successful. Handle any locking requirements. 3796 */ 3797 if (fmode & (O_EXLOCK | O_SHLOCK)) { 3798 lf.l_whence = SEEK_SET; 3799 lf.l_start = 0; 3800 lf.l_len = 0; 3801 if (fmode & O_EXLOCK) 3802 lf.l_type = F_WRLCK; 3803 else 3804 lf.l_type = F_RDLCK; 3805 if (fmode & FNONBLOCK) 3806 type = 0; 3807 else 3808 type = F_WAIT; 3809 vn_unlock(vp); 3810 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) { 3811 /* 3812 * release our private reference. 3813 */ 3814 fsetfd(p, NULL, indx); 3815 fdrop(fp); 3816 vrele(vp); 3817 return (error); 3818 } 3819 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3820 fp->f_flag |= FHASLOCK; 3821 } 3822 3823 /* 3824 * Clean up. Associate the file pointer with the previously 3825 * reserved descriptor and return it. 3826 */ 3827 vput(vp); 3828 fsetfd(p, fp, indx); 3829 fdrop(fp); 3830 uap->sysmsg_result = indx; 3831 return (0); 3832 3833 bad_drop: 3834 fsetfd(p, NULL, indx); 3835 fdrop(fp); 3836 bad: 3837 vput(vp); 3838 return (error); 3839 } 3840 3841 /* 3842 * fhstat_args(struct fhandle *u_fhp, struct stat *sb) 3843 */ 3844 int 3845 sys_fhstat(struct fhstat_args *uap) 3846 { 3847 struct thread *td = curthread; 3848 struct stat sb; 3849 fhandle_t fh; 3850 struct mount *mp; 3851 struct vnode *vp; 3852 int error; 3853 3854 /* 3855 * Must be super user 3856 */ 3857 error = priv_check(td, PRIV_ROOT); 3858 if (error) 3859 return (error); 3860 3861 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); 3862 if (error) 3863 return (error); 3864 3865 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 3866 return (ESTALE); 3867 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 3868 return (error); 3869 error = vn_stat(vp, &sb, td->td_proc->p_ucred); 3870 vput(vp); 3871 if (error) 3872 return (error); 3873 error = copyout(&sb, uap->sb, sizeof(sb)); 3874 return (error); 3875 } 3876 3877 /* 3878 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf) 3879 */ 3880 int 3881 sys_fhstatfs(struct fhstatfs_args *uap) 3882 { 3883 struct thread *td = curthread; 3884 struct proc *p = td->td_proc; 3885 struct statfs *sp; 3886 struct mount *mp; 3887 struct vnode *vp; 3888 struct statfs sb; 3889 char *fullpath, *freepath; 3890 fhandle_t fh; 3891 int error; 3892 3893 /* 3894 * Must be super user 3895 */ 3896 if ((error = priv_check(td, PRIV_ROOT))) 3897 return (error); 3898 3899 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 3900 return (error); 3901 3902 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 3903 return (ESTALE); 3904 3905 if (p != NULL && !chroot_visible_mnt(mp, p)) 3906 return (ESTALE); 3907 3908 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 3909 return (error); 3910 mp = vp->v_mount; 3911 sp = &mp->mnt_stat; 3912 vput(vp); 3913 if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0) 3914 return (error); 3915 3916 error = mount_path(p, mp, &fullpath, &freepath); 3917 if (error) 3918 return(error); 3919 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 3920 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 3921 kfree(freepath, M_TEMP); 3922 3923 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 3924 if (priv_check(td, PRIV_ROOT)) { 3925 bcopy(sp, &sb, sizeof(sb)); 3926 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; 3927 sp = &sb; 3928 } 3929 return (copyout(sp, uap->buf, sizeof(*sp))); 3930 } 3931 3932 /* 3933 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf) 3934 */ 3935 int 3936 sys_fhstatvfs(struct fhstatvfs_args *uap) 3937 { 3938 struct thread *td = curthread; 3939 struct proc *p = td->td_proc; 3940 struct statvfs *sp; 3941 struct mount *mp; 3942 struct vnode *vp; 3943 fhandle_t fh; 3944 int error; 3945 3946 /* 3947 * Must be super user 3948 */ 3949 if ((error = priv_check(td, PRIV_ROOT))) 3950 return (error); 3951 3952 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 3953 return (error); 3954 3955 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 3956 return (ESTALE); 3957 3958 if (p != NULL && !chroot_visible_mnt(mp, p)) 3959 return (ESTALE); 3960 3961 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 3962 return (error); 3963 mp = vp->v_mount; 3964 sp = &mp->mnt_vstat; 3965 vput(vp); 3966 if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0) 3967 return (error); 3968 3969 sp->f_flag = 0; 3970 if (mp->mnt_flag & MNT_RDONLY) 3971 sp->f_flag |= ST_RDONLY; 3972 if (mp->mnt_flag & MNT_NOSUID) 3973 sp->f_flag |= ST_NOSUID; 3974 3975 return (copyout(sp, uap->buf, sizeof(*sp))); 3976 } 3977 3978 3979 /* 3980 * Syscall to push extended attribute configuration information into the 3981 * VFS. Accepts a path, which it converts to a mountpoint, as well as 3982 * a command (int cmd), and attribute name and misc data. For now, the 3983 * attribute name is left in userspace for consumption by the VFS_op. 3984 * It will probably be changed to be copied into sysspace by the 3985 * syscall in the future, once issues with various consumers of the 3986 * attribute code have raised their hands. 3987 * 3988 * Currently this is used only by UFS Extended Attributes. 3989 */ 3990 int 3991 sys_extattrctl(struct extattrctl_args *uap) 3992 { 3993 struct nlookupdata nd; 3994 struct mount *mp; 3995 struct vnode *vp; 3996 int error; 3997 3998 vp = NULL; 3999 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4000 if (error == 0) 4001 error = nlookup(&nd); 4002 if (error == 0) { 4003 mp = nd.nl_nch.mount; 4004 error = VFS_EXTATTRCTL(mp, uap->cmd, 4005 uap->attrname, uap->arg, 4006 nd.nl_cred); 4007 } 4008 nlookup_done(&nd); 4009 return (error); 4010 } 4011 4012 /* 4013 * Syscall to set a named extended attribute on a file or directory. 4014 * Accepts attribute name, and a uio structure pointing to the data to set. 4015 * The uio is consumed in the style of writev(). The real work happens 4016 * in VOP_SETEXTATTR(). 4017 */ 4018 int 4019 sys_extattr_set_file(struct extattr_set_file_args *uap) 4020 { 4021 char attrname[EXTATTR_MAXNAMELEN]; 4022 struct iovec aiov[UIO_SMALLIOV]; 4023 struct iovec *needfree; 4024 struct nlookupdata nd; 4025 struct iovec *iov; 4026 struct vnode *vp; 4027 struct uio auio; 4028 u_int iovlen; 4029 u_int cnt; 4030 int error; 4031 int i; 4032 4033 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4034 if (error) 4035 return (error); 4036 4037 vp = NULL; 4038 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4039 if (error == 0) 4040 error = nlookup(&nd); 4041 if (error == 0) 4042 error = ncp_writechk(&nd.nl_nch); 4043 if (error == 0) 4044 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4045 if (error) { 4046 nlookup_done(&nd); 4047 return (error); 4048 } 4049 4050 needfree = NULL; 4051 iovlen = uap->iovcnt * sizeof(struct iovec); 4052 if (uap->iovcnt > UIO_SMALLIOV) { 4053 if (uap->iovcnt > UIO_MAXIOV) { 4054 error = EINVAL; 4055 goto done; 4056 } 4057 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK); 4058 needfree = iov; 4059 } else { 4060 iov = aiov; 4061 } 4062 auio.uio_iov = iov; 4063 auio.uio_iovcnt = uap->iovcnt; 4064 auio.uio_rw = UIO_WRITE; 4065 auio.uio_segflg = UIO_USERSPACE; 4066 auio.uio_td = nd.nl_td; 4067 auio.uio_offset = 0; 4068 if ((error = copyin(uap->iovp, iov, iovlen))) 4069 goto done; 4070 auio.uio_resid = 0; 4071 for (i = 0; i < uap->iovcnt; i++) { 4072 if (iov->iov_len > INT_MAX - auio.uio_resid) { 4073 error = EINVAL; 4074 goto done; 4075 } 4076 auio.uio_resid += iov->iov_len; 4077 iov++; 4078 } 4079 cnt = auio.uio_resid; 4080 error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred); 4081 cnt -= auio.uio_resid; 4082 uap->sysmsg_result = cnt; 4083 done: 4084 vput(vp); 4085 nlookup_done(&nd); 4086 if (needfree) 4087 FREE(needfree, M_IOV); 4088 return (error); 4089 } 4090 4091 /* 4092 * Syscall to get a named extended attribute on a file or directory. 4093 * Accepts attribute name, and a uio structure pointing to a buffer for the 4094 * data. The uio is consumed in the style of readv(). The real work 4095 * happens in VOP_GETEXTATTR(); 4096 */ 4097 int 4098 sys_extattr_get_file(struct extattr_get_file_args *uap) 4099 { 4100 char attrname[EXTATTR_MAXNAMELEN]; 4101 struct iovec aiov[UIO_SMALLIOV]; 4102 struct iovec *needfree; 4103 struct nlookupdata nd; 4104 struct iovec *iov; 4105 struct vnode *vp; 4106 struct uio auio; 4107 u_int iovlen; 4108 u_int cnt; 4109 int error; 4110 int i; 4111 4112 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4113 if (error) 4114 return (error); 4115 4116 vp = NULL; 4117 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4118 if (error == 0) 4119 error = nlookup(&nd); 4120 if (error == 0) 4121 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4122 if (error) { 4123 nlookup_done(&nd); 4124 return (error); 4125 } 4126 4127 iovlen = uap->iovcnt * sizeof (struct iovec); 4128 needfree = NULL; 4129 if (uap->iovcnt > UIO_SMALLIOV) { 4130 if (uap->iovcnt > UIO_MAXIOV) { 4131 error = EINVAL; 4132 goto done; 4133 } 4134 MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK); 4135 needfree = iov; 4136 } else { 4137 iov = aiov; 4138 } 4139 auio.uio_iov = iov; 4140 auio.uio_iovcnt = uap->iovcnt; 4141 auio.uio_rw = UIO_READ; 4142 auio.uio_segflg = UIO_USERSPACE; 4143 auio.uio_td = nd.nl_td; 4144 auio.uio_offset = 0; 4145 if ((error = copyin(uap->iovp, iov, iovlen))) 4146 goto done; 4147 auio.uio_resid = 0; 4148 for (i = 0; i < uap->iovcnt; i++) { 4149 if (iov->iov_len > INT_MAX - auio.uio_resid) { 4150 error = EINVAL; 4151 goto done; 4152 } 4153 auio.uio_resid += iov->iov_len; 4154 iov++; 4155 } 4156 cnt = auio.uio_resid; 4157 error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred); 4158 cnt -= auio.uio_resid; 4159 uap->sysmsg_result = cnt; 4160 done: 4161 vput(vp); 4162 nlookup_done(&nd); 4163 if (needfree) 4164 FREE(needfree, M_IOV); 4165 return(error); 4166 } 4167 4168 /* 4169 * Syscall to delete a named extended attribute from a file or directory. 4170 * Accepts attribute name. The real work happens in VOP_SETEXTATTR(). 4171 */ 4172 int 4173 sys_extattr_delete_file(struct extattr_delete_file_args *uap) 4174 { 4175 char attrname[EXTATTR_MAXNAMELEN]; 4176 struct nlookupdata nd; 4177 struct vnode *vp; 4178 int error; 4179 4180 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4181 if (error) 4182 return(error); 4183 4184 vp = NULL; 4185 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4186 if (error == 0) 4187 error = nlookup(&nd); 4188 if (error == 0) 4189 error = ncp_writechk(&nd.nl_nch); 4190 if (error == 0) 4191 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4192 if (error) { 4193 nlookup_done(&nd); 4194 return (error); 4195 } 4196 4197 error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred); 4198 vput(vp); 4199 nlookup_done(&nd); 4200 return(error); 4201 } 4202 4203 /* 4204 * Determine if the mount is visible to the process. 4205 */ 4206 static int 4207 chroot_visible_mnt(struct mount *mp, struct proc *p) 4208 { 4209 struct nchandle nch; 4210 4211 /* 4212 * Traverse from the mount point upwards. If we hit the process 4213 * root then the mount point is visible to the process. 4214 */ 4215 nch = mp->mnt_ncmountpt; 4216 while (nch.ncp) { 4217 if (nch.mount == p->p_fd->fd_nrdir.mount && 4218 nch.ncp == p->p_fd->fd_nrdir.ncp) { 4219 return(1); 4220 } 4221 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) { 4222 nch = nch.mount->mnt_ncmounton; 4223 } else { 4224 nch.ncp = nch.ncp->nc_parent; 4225 } 4226 } 4227 4228 /* 4229 * If the mount point is not visible to the process, but the 4230 * process root is in a subdirectory of the mount, return 4231 * TRUE anyway. 4232 */ 4233 if (p->p_fd->fd_nrdir.mount == mp) 4234 return(1); 4235 4236 return(0); 4237 } 4238 4239