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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 35 * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/buf.h> 41 #include <sys/conf.h> 42 #include <sys/sysent.h> 43 #include <sys/malloc.h> 44 #include <sys/mount.h> 45 #include <sys/mountctl.h> 46 #include <sys/sysproto.h> 47 #include <sys/filedesc.h> 48 #include <sys/kernel.h> 49 #include <sys/fcntl.h> 50 #include <sys/file.h> 51 #include <sys/linker.h> 52 #include <sys/stat.h> 53 #include <sys/unistd.h> 54 #include <sys/vnode.h> 55 #include <sys/proc.h> 56 #include <sys/priv.h> 57 #include <sys/jail.h> 58 #include <sys/namei.h> 59 #include <sys/nlookup.h> 60 #include <sys/dirent.h> 61 #include <sys/extattr.h> 62 #include <sys/spinlock.h> 63 #include <sys/kern_syscall.h> 64 #include <sys/objcache.h> 65 #include <sys/sysctl.h> 66 67 #include <sys/buf2.h> 68 #include <sys/file2.h> 69 #include <sys/spinlock2.h> 70 #include <sys/mplock2.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_object.h> 74 #include <vm/vm_page.h> 75 76 #include <machine/limits.h> 77 #include <machine/stdarg.h> 78 79 #include <vfs/union/union.h> 80 81 static void mount_warning(struct mount *mp, const char *ctl, ...) 82 __printflike(2, 3); 83 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb); 84 static int checkvp_chdir (struct vnode *vn, struct thread *td); 85 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch); 86 static int chroot_refuse_vdir_fds (struct filedesc *fdp); 87 static int chroot_visible_mnt(struct mount *mp, struct proc *p); 88 static int getutimes (const struct timeval *, struct timespec *); 89 static int setfown (struct mount *, struct vnode *, uid_t, gid_t); 90 static int setfmode (struct vnode *, int); 91 static int setfflags (struct vnode *, int); 92 static int setutimes (struct vnode *, struct vattr *, 93 const struct timespec *, int); 94 static int usermount = 0; /* if 1, non-root can mount fs. */ 95 96 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *); 97 98 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 99 "Allow non-root users to mount filesystems"); 100 101 /* 102 * Virtual File System System Calls 103 */ 104 105 /* 106 * Mount a file system. 107 * 108 * mount_args(char *type, char *path, int flags, caddr_t data) 109 * 110 * MPALMOSTSAFE 111 */ 112 int 113 sys_mount(struct mount_args *uap) 114 { 115 struct thread *td = curthread; 116 struct vnode *vp; 117 struct nchandle nch; 118 struct mount *mp, *nullmp; 119 struct vfsconf *vfsp; 120 int error, flag = 0, flag2 = 0; 121 int hasmount; 122 struct vattr va; 123 struct nlookupdata nd; 124 char fstypename[MFSNAMELEN]; 125 struct ucred *cred; 126 127 cred = td->td_ucred; 128 if (jailed(cred)) { 129 error = EPERM; 130 goto done; 131 } 132 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT))) 133 goto done; 134 135 /* 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 goto done; 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 goto done; 162 } 163 164 /* 165 * If the target filesystem is resolved via a nullfs mount, then 166 * nd.nl_nch.mount will be pointing to the nullfs mount structure 167 * instead of the target file system. We need it in case we are 168 * doing an update. 169 */ 170 nullmp = nd.nl_nch.mount; 171 172 /* 173 * Extract the locked+refd ncp and cleanup the nd structure 174 */ 175 nch = nd.nl_nch; 176 cache_zero(&nd.nl_nch); 177 nlookup_done(&nd); 178 179 if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && 180 (mp = cache_findmount(&nch)) != NULL) { 181 cache_dropmount(mp); 182 hasmount = 1; 183 } else { 184 hasmount = 0; 185 } 186 187 188 /* 189 * now we have the locked ref'd nch and unreferenced vnode. 190 */ 191 vp = nch.ncp->nc_vp; 192 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) { 193 cache_put(&nch); 194 goto done; 195 } 196 cache_unlock(&nch); 197 198 /* 199 * Extract the file system type. We need to know this early, to take 200 * appropriate actions if we are dealing with a nullfs. 201 */ 202 if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) { 203 cache_drop(&nch); 204 vput(vp); 205 goto done; 206 } 207 208 /* 209 * Now we have an unlocked ref'd nch and a locked ref'd vp 210 */ 211 if (uap->flags & MNT_UPDATE) { 212 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 213 cache_drop(&nch); 214 vput(vp); 215 error = EINVAL; 216 goto done; 217 } 218 219 if (strncmp(fstypename, "null", 5) == 0) { 220 KKASSERT(nullmp); 221 mp = nullmp; 222 } else { 223 mp = vp->v_mount; 224 } 225 226 flag = mp->mnt_flag; 227 flag2 = mp->mnt_kern_flag; 228 /* 229 * We only allow the filesystem to be reloaded if it 230 * is currently mounted read-only. 231 */ 232 if ((uap->flags & MNT_RELOAD) && 233 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 234 cache_drop(&nch); 235 vput(vp); 236 error = EOPNOTSUPP; /* Needs translation */ 237 goto done; 238 } 239 /* 240 * Only root, or the user that did the original mount is 241 * permitted to update it. 242 */ 243 if (mp->mnt_stat.f_owner != cred->cr_uid && 244 (error = priv_check(td, PRIV_ROOT))) { 245 cache_drop(&nch); 246 vput(vp); 247 goto done; 248 } 249 if (vfs_busy(mp, LK_NOWAIT)) { 250 cache_drop(&nch); 251 vput(vp); 252 error = EBUSY; 253 goto done; 254 } 255 if (hasmount) { 256 cache_drop(&nch); 257 vfs_unbusy(mp); 258 vput(vp); 259 error = EBUSY; 260 goto done; 261 } 262 mp->mnt_flag |= 263 uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE); 264 lwkt_gettoken(&mp->mnt_token); 265 vn_unlock(vp); 266 goto update; 267 } 268 269 /* 270 * If the user is not root, ensure that they own the directory 271 * onto which we are attempting to mount. 272 */ 273 if ((error = VOP_GETATTR(vp, &va)) || 274 (va.va_uid != cred->cr_uid && 275 (error = priv_check(td, PRIV_ROOT)))) { 276 cache_drop(&nch); 277 vput(vp); 278 goto done; 279 } 280 if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) { 281 cache_drop(&nch); 282 vput(vp); 283 goto done; 284 } 285 if (vp->v_type != VDIR) { 286 cache_drop(&nch); 287 vput(vp); 288 error = ENOTDIR; 289 goto done; 290 } 291 if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) { 292 cache_drop(&nch); 293 vput(vp); 294 error = EPERM; 295 goto done; 296 } 297 vfsp = vfsconf_find_by_name(fstypename); 298 if (vfsp == NULL) { 299 linker_file_t lf; 300 301 /* Only load modules for root (very important!) */ 302 if ((error = priv_check(td, PRIV_ROOT)) != 0) { 303 cache_drop(&nch); 304 vput(vp); 305 goto done; 306 } 307 error = linker_load_file(fstypename, &lf); 308 if (error || lf == NULL) { 309 cache_drop(&nch); 310 vput(vp); 311 if (lf == NULL) 312 error = ENODEV; 313 goto done; 314 } 315 lf->userrefs++; 316 /* lookup again, see if the VFS was loaded */ 317 vfsp = vfsconf_find_by_name(fstypename); 318 if (vfsp == NULL) { 319 lf->userrefs--; 320 linker_file_unload(lf); 321 cache_drop(&nch); 322 vput(vp); 323 error = ENODEV; 324 goto done; 325 } 326 } 327 if (hasmount) { 328 cache_drop(&nch); 329 vput(vp); 330 error = EBUSY; 331 goto done; 332 } 333 334 /* 335 * Allocate and initialize the filesystem. 336 */ 337 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK); 338 mount_init(mp); 339 vfs_busy(mp, LK_NOWAIT); 340 mp->mnt_op = vfsp->vfc_vfsops; 341 mp->mnt_vfc = vfsp; 342 vfsp->vfc_refcount++; 343 mp->mnt_stat.f_type = vfsp->vfc_typenum; 344 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 345 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 346 mp->mnt_stat.f_owner = cred->cr_uid; 347 lwkt_gettoken(&mp->mnt_token); 348 vn_unlock(vp); 349 update: 350 /* 351 * (per-mount token acquired at this point) 352 * 353 * Set the mount level flags. 354 */ 355 if (uap->flags & MNT_RDONLY) 356 mp->mnt_flag |= MNT_RDONLY; 357 else if (mp->mnt_flag & MNT_RDONLY) 358 mp->mnt_kern_flag |= MNTK_WANTRDWR; 359 mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV | 360 MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME | 361 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM | 362 MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR); 363 mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC | 364 MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE | 365 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM | 366 MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR); 367 /* 368 * Mount the filesystem. 369 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 370 * get. 371 */ 372 error = VFS_MOUNT(mp, uap->path, uap->data, cred); 373 if (mp->mnt_flag & MNT_UPDATE) { 374 if (mp->mnt_kern_flag & MNTK_WANTRDWR) 375 mp->mnt_flag &= ~MNT_RDONLY; 376 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE); 377 mp->mnt_kern_flag &=~ MNTK_WANTRDWR; 378 if (error) { 379 mp->mnt_flag = flag; 380 mp->mnt_kern_flag = flag2; 381 } 382 lwkt_reltoken(&mp->mnt_token); 383 vfs_unbusy(mp); 384 vrele(vp); 385 cache_drop(&nch); 386 goto done; 387 } 388 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 389 390 /* 391 * Put the new filesystem on the mount list after root. The mount 392 * point gets its own mnt_ncmountpt (unless the VFS already set one 393 * up) which represents the root of the mount. The lookup code 394 * detects the mount point going forward and checks the root of 395 * the mount going backwards. 396 * 397 * It is not necessary to invalidate or purge the vnode underneath 398 * because elements under the mount will be given their own glue 399 * namecache record. 400 */ 401 if (!error) { 402 if (mp->mnt_ncmountpt.ncp == NULL) { 403 /* 404 * allocate, then unlock, but leave the ref intact 405 */ 406 cache_allocroot(&mp->mnt_ncmountpt, mp, NULL); 407 cache_unlock(&mp->mnt_ncmountpt); 408 } 409 mp->mnt_ncmounton = nch; /* inherits ref */ 410 nch.ncp->nc_flag |= NCF_ISMOUNTPT; 411 cache_ismounting(mp); 412 413 mountlist_insert(mp, MNTINS_LAST); 414 vn_unlock(vp); 415 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt); 416 error = vfs_allocate_syncvnode(mp); 417 lwkt_reltoken(&mp->mnt_token); 418 vfs_unbusy(mp); 419 error = VFS_START(mp, 0); 420 vrele(vp); 421 } else { 422 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops); 423 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops); 424 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops); 425 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops); 426 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops); 427 mp->mnt_vfc->vfc_refcount--; 428 lwkt_reltoken(&mp->mnt_token); 429 vfs_unbusy(mp); 430 kfree(mp, M_MOUNT); 431 cache_drop(&nch); 432 vput(vp); 433 } 434 done: 435 return (error); 436 } 437 438 /* 439 * Scan all active processes to see if any of them have a current 440 * or root directory onto which the new filesystem has just been 441 * mounted. If so, replace them with the new mount point. 442 * 443 * Both old_nch and new_nch are ref'd on call but not locked. 444 * new_nch must be temporarily locked so it can be associated with the 445 * vnode representing the root of the mount point. 446 */ 447 struct checkdirs_info { 448 struct nchandle old_nch; 449 struct nchandle new_nch; 450 struct vnode *old_vp; 451 struct vnode *new_vp; 452 }; 453 454 static int checkdirs_callback(struct proc *p, void *data); 455 456 static void 457 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch) 458 { 459 struct checkdirs_info info; 460 struct vnode *olddp; 461 struct vnode *newdp; 462 struct mount *mp; 463 464 /* 465 * If the old mount point's vnode has a usecount of 1, it is not 466 * being held as a descriptor anywhere. 467 */ 468 olddp = old_nch->ncp->nc_vp; 469 if (olddp == NULL || olddp->v_sysref.refcnt == 1) 470 return; 471 472 /* 473 * Force the root vnode of the new mount point to be resolved 474 * so we can update any matching processes. 475 */ 476 mp = new_nch->mount; 477 if (VFS_ROOT(mp, &newdp)) 478 panic("mount: lost mount"); 479 vn_unlock(newdp); 480 cache_lock(new_nch); 481 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY); 482 cache_setunresolved(new_nch); 483 cache_setvp(new_nch, newdp); 484 cache_unlock(new_nch); 485 486 /* 487 * Special handling of the root node 488 */ 489 if (rootvnode == olddp) { 490 vref(newdp); 491 vfs_cache_setroot(newdp, cache_hold(new_nch)); 492 } 493 494 /* 495 * Pass newdp separately so the callback does not have to access 496 * it via new_nch->ncp->nc_vp. 497 */ 498 info.old_nch = *old_nch; 499 info.new_nch = *new_nch; 500 info.new_vp = newdp; 501 allproc_scan(checkdirs_callback, &info); 502 vput(newdp); 503 } 504 505 /* 506 * NOTE: callback is not MP safe because the scanned process's filedesc 507 * structure can be ripped out from under us, amoung other things. 508 */ 509 static int 510 checkdirs_callback(struct proc *p, void *data) 511 { 512 struct checkdirs_info *info = data; 513 struct filedesc *fdp; 514 struct nchandle ncdrop1; 515 struct nchandle ncdrop2; 516 struct vnode *vprele1; 517 struct vnode *vprele2; 518 519 if ((fdp = p->p_fd) != NULL) { 520 cache_zero(&ncdrop1); 521 cache_zero(&ncdrop2); 522 vprele1 = NULL; 523 vprele2 = NULL; 524 525 /* 526 * MPUNSAFE - XXX fdp can be pulled out from under a 527 * foreign process. 528 * 529 * A shared filedesc is ok, we don't have to copy it 530 * because we are making this change globally. 531 */ 532 spin_lock(&fdp->fd_spin); 533 if (fdp->fd_ncdir.mount == info->old_nch.mount && 534 fdp->fd_ncdir.ncp == info->old_nch.ncp) { 535 vprele1 = fdp->fd_cdir; 536 vref(info->new_vp); 537 fdp->fd_cdir = info->new_vp; 538 ncdrop1 = fdp->fd_ncdir; 539 cache_copy(&info->new_nch, &fdp->fd_ncdir); 540 } 541 if (fdp->fd_nrdir.mount == info->old_nch.mount && 542 fdp->fd_nrdir.ncp == info->old_nch.ncp) { 543 vprele2 = fdp->fd_rdir; 544 vref(info->new_vp); 545 fdp->fd_rdir = info->new_vp; 546 ncdrop2 = fdp->fd_nrdir; 547 cache_copy(&info->new_nch, &fdp->fd_nrdir); 548 } 549 spin_unlock(&fdp->fd_spin); 550 if (ncdrop1.ncp) 551 cache_drop(&ncdrop1); 552 if (ncdrop2.ncp) 553 cache_drop(&ncdrop2); 554 if (vprele1) 555 vrele(vprele1); 556 if (vprele2) 557 vrele(vprele2); 558 } 559 return(0); 560 } 561 562 /* 563 * Unmount a file system. 564 * 565 * Note: unmount takes a path to the vnode mounted on as argument, 566 * not special file (as before). 567 * 568 * umount_args(char *path, int flags) 569 * 570 * MPALMOSTSAFE 571 */ 572 int 573 sys_unmount(struct unmount_args *uap) 574 { 575 struct thread *td = curthread; 576 struct proc *p __debugvar = td->td_proc; 577 struct mount *mp = NULL; 578 struct nlookupdata nd; 579 int error; 580 581 KKASSERT(p); 582 get_mplock(); 583 if (td->td_ucred->cr_prison != NULL) { 584 error = EPERM; 585 goto done; 586 } 587 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT))) 588 goto done; 589 590 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 591 if (error == 0) 592 error = nlookup(&nd); 593 if (error) 594 goto out; 595 596 mp = nd.nl_nch.mount; 597 598 /* 599 * Only root, or the user that did the original mount is 600 * permitted to unmount this filesystem. 601 */ 602 if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) && 603 (error = priv_check(td, PRIV_ROOT))) 604 goto out; 605 606 /* 607 * Don't allow unmounting the root file system. 608 */ 609 if (mp->mnt_flag & MNT_ROOTFS) { 610 error = EINVAL; 611 goto out; 612 } 613 614 /* 615 * Must be the root of the filesystem 616 */ 617 if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) { 618 error = EINVAL; 619 goto out; 620 } 621 622 out: 623 nlookup_done(&nd); 624 if (error == 0) 625 error = dounmount(mp, uap->flags); 626 done: 627 rel_mplock(); 628 return (error); 629 } 630 631 /* 632 * Do the actual file system unmount. 633 */ 634 static int 635 dounmount_interlock(struct mount *mp) 636 { 637 if (mp->mnt_kern_flag & MNTK_UNMOUNT) 638 return (EBUSY); 639 mp->mnt_kern_flag |= MNTK_UNMOUNT; 640 return(0); 641 } 642 643 static int 644 unmount_allproc_cb(struct proc *p, void *arg) 645 { 646 struct mount *mp; 647 648 if (p->p_textnch.ncp == NULL) 649 return 0; 650 651 mp = (struct mount *)arg; 652 if (p->p_textnch.mount == mp) 653 cache_drop(&p->p_textnch); 654 655 return 0; 656 } 657 658 int 659 dounmount(struct mount *mp, int flags) 660 { 661 struct namecache *ncp; 662 struct nchandle nch; 663 struct vnode *vp; 664 int error; 665 int async_flag; 666 int lflags; 667 int freeok = 1; 668 int retry; 669 670 lwkt_gettoken(&mp->mnt_token); 671 /* 672 * Exclusive access for unmounting purposes 673 */ 674 if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0) 675 goto out; 676 677 /* 678 * Allow filesystems to detect that a forced unmount is in progress. 679 */ 680 if (flags & MNT_FORCE) 681 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 682 lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK); 683 error = lockmgr(&mp->mnt_lock, lflags); 684 if (error) { 685 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 686 if (mp->mnt_kern_flag & MNTK_MWAIT) { 687 mp->mnt_kern_flag &= ~MNTK_MWAIT; 688 wakeup(mp); 689 } 690 goto out; 691 } 692 693 if (mp->mnt_flag & MNT_EXPUBLIC) 694 vfs_setpublicfs(NULL, NULL, NULL); 695 696 vfs_msync(mp, MNT_WAIT); 697 async_flag = mp->mnt_flag & MNT_ASYNC; 698 mp->mnt_flag &=~ MNT_ASYNC; 699 700 /* 701 * If this filesystem isn't aliasing other filesystems, 702 * try to invalidate any remaining namecache entries and 703 * check the count afterwords. 704 */ 705 if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) { 706 cache_lock(&mp->mnt_ncmountpt); 707 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN); 708 cache_unlock(&mp->mnt_ncmountpt); 709 710 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL && 711 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) { 712 allproc_scan(&unmount_allproc_cb, mp); 713 } 714 715 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL && 716 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) { 717 718 if ((flags & MNT_FORCE) == 0) { 719 error = EBUSY; 720 mount_warning(mp, "Cannot unmount: " 721 "%d namecache " 722 "references still " 723 "present", 724 ncp->nc_refs - 1); 725 } else { 726 mount_warning(mp, "Forced unmount: " 727 "%d namecache " 728 "references still " 729 "present", 730 ncp->nc_refs - 1); 731 freeok = 0; 732 } 733 } 734 } 735 736 /* 737 * Decomission our special mnt_syncer vnode. This also stops 738 * the vnlru code. If we are unable to unmount we recommission 739 * the vnode. 740 * 741 * Then sync the filesystem. 742 */ 743 if ((vp = mp->mnt_syncer) != NULL) { 744 mp->mnt_syncer = NULL; 745 vrele(vp); 746 } 747 if ((mp->mnt_flag & MNT_RDONLY) == 0) 748 VFS_SYNC(mp, MNT_WAIT); 749 750 /* 751 * nchandle records ref the mount structure. Expect a count of 1 752 * (our mount->mnt_ncmountpt). 753 * 754 * Scans can get temporary refs on a mountpoint (thought really 755 * heavy duty stuff like cache_findmount() do not). 756 */ 757 for (retry = 0; retry < 10 && mp->mnt_refs != 1; ++retry) { 758 cache_unmounting(mp); 759 tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1); 760 } 761 if (mp->mnt_refs != 1) { 762 if ((flags & MNT_FORCE) == 0) { 763 mount_warning(mp, "Cannot unmount: " 764 "%d mount refs still present", 765 mp->mnt_refs); 766 error = EBUSY; 767 } else { 768 mount_warning(mp, "Forced unmount: " 769 "%d mount refs still present", 770 mp->mnt_refs); 771 freeok = 0; 772 } 773 } 774 775 /* 776 * So far so good, sync the filesystem once more and 777 * call the VFS unmount code if the sync succeeds. 778 */ 779 if (error == 0) { 780 if (((mp->mnt_flag & MNT_RDONLY) || 781 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || 782 (flags & MNT_FORCE)) { 783 error = VFS_UNMOUNT(mp, flags); 784 } 785 } 786 787 /* 788 * If an error occurred we can still recover, restoring the 789 * syncer vnode and misc flags. 790 */ 791 if (error) { 792 if (mp->mnt_syncer == NULL) 793 vfs_allocate_syncvnode(mp); 794 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 795 mp->mnt_flag |= async_flag; 796 lockmgr(&mp->mnt_lock, LK_RELEASE); 797 if (mp->mnt_kern_flag & MNTK_MWAIT) { 798 mp->mnt_kern_flag &= ~MNTK_MWAIT; 799 wakeup(mp); 800 } 801 goto out; 802 } 803 /* 804 * Clean up any journals still associated with the mount after 805 * filesystem activity has ceased. 806 */ 807 journal_remove_all_journals(mp, 808 ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0)); 809 810 mountlist_remove(mp); 811 812 /* 813 * Remove any installed vnode ops here so the individual VFSs don't 814 * have to. 815 */ 816 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops); 817 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops); 818 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops); 819 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops); 820 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops); 821 822 if (mp->mnt_ncmountpt.ncp != NULL) { 823 nch = mp->mnt_ncmountpt; 824 cache_zero(&mp->mnt_ncmountpt); 825 cache_clrmountpt(&nch); 826 cache_drop(&nch); 827 } 828 if (mp->mnt_ncmounton.ncp != NULL) { 829 cache_unmounting(mp); 830 nch = mp->mnt_ncmounton; 831 cache_zero(&mp->mnt_ncmounton); 832 cache_clrmountpt(&nch); 833 cache_drop(&nch); 834 } 835 836 mp->mnt_vfc->vfc_refcount--; 837 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) 838 panic("unmount: dangling vnode"); 839 lockmgr(&mp->mnt_lock, LK_RELEASE); 840 if (mp->mnt_kern_flag & MNTK_MWAIT) { 841 mp->mnt_kern_flag &= ~MNTK_MWAIT; 842 wakeup(mp); 843 } 844 845 /* 846 * If we reach here and freeok != 0 we must free the mount. 847 * If refs > 1 cycle and wait, just in case someone tried 848 * to busy the mount after we decided to do the unmount. 849 */ 850 if (freeok) { 851 while (mp->mnt_refs > 1) { 852 cache_unmounting(mp); 853 wakeup(mp); 854 tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1); 855 } 856 lwkt_reltoken(&mp->mnt_token); 857 kfree(mp, M_MOUNT); 858 mp = NULL; 859 } 860 error = 0; 861 out: 862 if (mp) 863 lwkt_reltoken(&mp->mnt_token); 864 return (error); 865 } 866 867 static 868 void 869 mount_warning(struct mount *mp, const char *ctl, ...) 870 { 871 char *ptr; 872 char *buf; 873 __va_list va; 874 875 __va_start(va, ctl); 876 if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL, 877 &ptr, &buf, 0) == 0) { 878 kprintf("unmount(%s): ", ptr); 879 kvprintf(ctl, va); 880 kprintf("\n"); 881 kfree(buf, M_TEMP); 882 } else { 883 kprintf("unmount(%p", mp); 884 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name) 885 kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name); 886 kprintf("): "); 887 kvprintf(ctl, va); 888 kprintf("\n"); 889 } 890 __va_end(va); 891 } 892 893 /* 894 * Shim cache_fullpath() to handle the case where a process is chrooted into 895 * a subdirectory of a mount. In this case if the root mount matches the 896 * process root directory's mount we have to specify the process's root 897 * directory instead of the mount point, because the mount point might 898 * be above the root directory. 899 */ 900 static 901 int 902 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb) 903 { 904 struct nchandle *nch; 905 906 if (p && p->p_fd->fd_nrdir.mount == mp) 907 nch = &p->p_fd->fd_nrdir; 908 else 909 nch = &mp->mnt_ncmountpt; 910 return(cache_fullpath(p, nch, NULL, rb, fb, 0)); 911 } 912 913 /* 914 * Sync each mounted filesystem. 915 */ 916 917 #ifdef DEBUG 918 static int syncprt = 0; 919 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, ""); 920 #endif /* DEBUG */ 921 922 static int sync_callback(struct mount *mp, void *data); 923 924 int 925 sys_sync(struct sync_args *uap) 926 { 927 mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD); 928 #ifdef DEBUG 929 /* 930 * print out buffer pool stat information on each sync() call. 931 */ 932 if (syncprt) 933 vfs_bufstats(); 934 #endif /* DEBUG */ 935 return (0); 936 } 937 938 static 939 int 940 sync_callback(struct mount *mp, void *data __unused) 941 { 942 int asyncflag; 943 944 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 945 asyncflag = mp->mnt_flag & MNT_ASYNC; 946 mp->mnt_flag &= ~MNT_ASYNC; 947 vfs_msync(mp, MNT_NOWAIT); 948 VFS_SYNC(mp, MNT_NOWAIT); 949 mp->mnt_flag |= asyncflag; 950 } 951 return(0); 952 } 953 954 /* XXX PRISON: could be per prison flag */ 955 static int prison_quotas; 956 #if 0 957 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, ""); 958 #endif 959 960 /* 961 * quotactl_args(char *path, int fcmd, int uid, caddr_t arg) 962 * 963 * Change filesystem quotas. 964 * 965 * MPALMOSTSAFE 966 */ 967 int 968 sys_quotactl(struct quotactl_args *uap) 969 { 970 struct nlookupdata nd; 971 struct thread *td; 972 struct mount *mp; 973 int error; 974 975 get_mplock(); 976 td = curthread; 977 if (td->td_ucred->cr_prison && !prison_quotas) { 978 error = EPERM; 979 goto done; 980 } 981 982 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 983 if (error == 0) 984 error = nlookup(&nd); 985 if (error == 0) { 986 mp = nd.nl_nch.mount; 987 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, 988 uap->arg, nd.nl_cred); 989 } 990 nlookup_done(&nd); 991 done: 992 rel_mplock(); 993 return (error); 994 } 995 996 /* 997 * mountctl(char *path, int op, int fd, const void *ctl, int ctllen, 998 * void *buf, int buflen) 999 * 1000 * This function operates on a mount point and executes the specified 1001 * operation using the specified control data, and possibly returns data. 1002 * 1003 * The actual number of bytes stored in the result buffer is returned, 0 1004 * if none, otherwise an error is returned. 1005 * 1006 * MPALMOSTSAFE 1007 */ 1008 int 1009 sys_mountctl(struct mountctl_args *uap) 1010 { 1011 struct thread *td = curthread; 1012 struct proc *p = td->td_proc; 1013 struct file *fp; 1014 void *ctl = NULL; 1015 void *buf = NULL; 1016 char *path = NULL; 1017 int error; 1018 1019 /* 1020 * Sanity and permissions checks. We must be root. 1021 */ 1022 KKASSERT(p); 1023 if (td->td_ucred->cr_prison != NULL) 1024 return (EPERM); 1025 if ((uap->op != MOUNTCTL_MOUNTFLAGS) && 1026 (error = priv_check(td, PRIV_ROOT)) != 0) 1027 return (error); 1028 1029 /* 1030 * Argument length checks 1031 */ 1032 if (uap->ctllen < 0 || uap->ctllen > 1024) 1033 return (EINVAL); 1034 if (uap->buflen < 0 || uap->buflen > 16 * 1024) 1035 return (EINVAL); 1036 if (uap->path == NULL) 1037 return (EINVAL); 1038 1039 /* 1040 * Allocate the necessary buffers and copyin data 1041 */ 1042 path = objcache_get(namei_oc, M_WAITOK); 1043 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 1044 if (error) 1045 goto done; 1046 1047 if (uap->ctllen) { 1048 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO); 1049 error = copyin(uap->ctl, ctl, uap->ctllen); 1050 if (error) 1051 goto done; 1052 } 1053 if (uap->buflen) 1054 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO); 1055 1056 /* 1057 * Validate the descriptor 1058 */ 1059 if (uap->fd >= 0) { 1060 fp = holdfp(p->p_fd, uap->fd, -1); 1061 if (fp == NULL) { 1062 error = EBADF; 1063 goto done; 1064 } 1065 } else { 1066 fp = NULL; 1067 } 1068 1069 /* 1070 * Execute the internal kernel function and clean up. 1071 */ 1072 get_mplock(); 1073 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result); 1074 rel_mplock(); 1075 if (fp) 1076 fdrop(fp); 1077 if (error == 0 && uap->sysmsg_result > 0) 1078 error = copyout(buf, uap->buf, uap->sysmsg_result); 1079 done: 1080 if (path) 1081 objcache_put(namei_oc, path); 1082 if (ctl) 1083 kfree(ctl, M_TEMP); 1084 if (buf) 1085 kfree(buf, M_TEMP); 1086 return (error); 1087 } 1088 1089 /* 1090 * Execute a mount control operation by resolving the path to a mount point 1091 * and calling vop_mountctl(). 1092 * 1093 * Use the mount point from the nch instead of the vnode so nullfs mounts 1094 * can properly spike the VOP. 1095 */ 1096 int 1097 kern_mountctl(const char *path, int op, struct file *fp, 1098 const void *ctl, int ctllen, 1099 void *buf, int buflen, int *res) 1100 { 1101 struct vnode *vp; 1102 struct mount *mp; 1103 struct nlookupdata nd; 1104 int error; 1105 1106 *res = 0; 1107 vp = NULL; 1108 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW); 1109 if (error == 0) 1110 error = nlookup(&nd); 1111 if (error == 0) 1112 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 1113 mp = nd.nl_nch.mount; 1114 nlookup_done(&nd); 1115 if (error) 1116 return (error); 1117 vn_unlock(vp); 1118 1119 /* 1120 * Must be the root of the filesystem 1121 */ 1122 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 1123 vrele(vp); 1124 return (EINVAL); 1125 } 1126 error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen, 1127 buf, buflen, res); 1128 vrele(vp); 1129 return (error); 1130 } 1131 1132 int 1133 kern_statfs(struct nlookupdata *nd, struct statfs *buf) 1134 { 1135 struct thread *td = curthread; 1136 struct proc *p = td->td_proc; 1137 struct mount *mp; 1138 struct statfs *sp; 1139 char *fullpath, *freepath; 1140 int error; 1141 1142 if ((error = nlookup(nd)) != 0) 1143 return (error); 1144 mp = nd->nl_nch.mount; 1145 sp = &mp->mnt_stat; 1146 if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0) 1147 return (error); 1148 1149 error = mount_path(p, mp, &fullpath, &freepath); 1150 if (error) 1151 return(error); 1152 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1153 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1154 kfree(freepath, M_TEMP); 1155 1156 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1157 bcopy(sp, buf, sizeof(*buf)); 1158 /* Only root should have access to the fsid's. */ 1159 if (priv_check(td, PRIV_ROOT)) 1160 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1161 return (0); 1162 } 1163 1164 /* 1165 * statfs_args(char *path, struct statfs *buf) 1166 * 1167 * Get filesystem statistics. 1168 */ 1169 int 1170 sys_statfs(struct statfs_args *uap) 1171 { 1172 struct nlookupdata nd; 1173 struct statfs buf; 1174 int error; 1175 1176 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1177 if (error == 0) 1178 error = kern_statfs(&nd, &buf); 1179 nlookup_done(&nd); 1180 if (error == 0) 1181 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1182 return (error); 1183 } 1184 1185 int 1186 kern_fstatfs(int fd, struct statfs *buf) 1187 { 1188 struct thread *td = curthread; 1189 struct proc *p = td->td_proc; 1190 struct file *fp; 1191 struct mount *mp; 1192 struct statfs *sp; 1193 char *fullpath, *freepath; 1194 int error; 1195 1196 KKASSERT(p); 1197 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 1198 return (error); 1199 1200 /* 1201 * Try to use mount info from any overlays rather than the 1202 * mount info for the underlying vnode, otherwise we will 1203 * fail when operating on null-mounted paths inside a chroot. 1204 */ 1205 if ((mp = fp->f_nchandle.mount) == NULL) 1206 mp = ((struct vnode *)fp->f_data)->v_mount; 1207 if (mp == NULL) { 1208 error = EBADF; 1209 goto done; 1210 } 1211 if (fp->f_cred == NULL) { 1212 error = EINVAL; 1213 goto done; 1214 } 1215 sp = &mp->mnt_stat; 1216 if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0) 1217 goto done; 1218 1219 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0) 1220 goto done; 1221 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1222 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1223 kfree(freepath, M_TEMP); 1224 1225 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1226 bcopy(sp, buf, sizeof(*buf)); 1227 1228 /* Only root should have access to the fsid's. */ 1229 if (priv_check(td, PRIV_ROOT)) 1230 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1231 error = 0; 1232 done: 1233 fdrop(fp); 1234 return (error); 1235 } 1236 1237 /* 1238 * fstatfs_args(int fd, struct statfs *buf) 1239 * 1240 * Get filesystem statistics. 1241 */ 1242 int 1243 sys_fstatfs(struct fstatfs_args *uap) 1244 { 1245 struct statfs buf; 1246 int error; 1247 1248 error = kern_fstatfs(uap->fd, &buf); 1249 1250 if (error == 0) 1251 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1252 return (error); 1253 } 1254 1255 int 1256 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf) 1257 { 1258 struct mount *mp; 1259 struct statvfs *sp; 1260 int error; 1261 1262 if ((error = nlookup(nd)) != 0) 1263 return (error); 1264 mp = nd->nl_nch.mount; 1265 sp = &mp->mnt_vstat; 1266 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0) 1267 return (error); 1268 1269 sp->f_flag = 0; 1270 if (mp->mnt_flag & MNT_RDONLY) 1271 sp->f_flag |= ST_RDONLY; 1272 if (mp->mnt_flag & MNT_NOSUID) 1273 sp->f_flag |= ST_NOSUID; 1274 bcopy(sp, buf, sizeof(*buf)); 1275 return (0); 1276 } 1277 1278 /* 1279 * statfs_args(char *path, struct statfs *buf) 1280 * 1281 * Get filesystem statistics. 1282 */ 1283 int 1284 sys_statvfs(struct statvfs_args *uap) 1285 { 1286 struct nlookupdata nd; 1287 struct statvfs buf; 1288 int error; 1289 1290 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1291 if (error == 0) 1292 error = kern_statvfs(&nd, &buf); 1293 nlookup_done(&nd); 1294 if (error == 0) 1295 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1296 return (error); 1297 } 1298 1299 int 1300 kern_fstatvfs(int fd, struct statvfs *buf) 1301 { 1302 struct thread *td = curthread; 1303 struct proc *p = td->td_proc; 1304 struct file *fp; 1305 struct mount *mp; 1306 struct statvfs *sp; 1307 int error; 1308 1309 KKASSERT(p); 1310 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 1311 return (error); 1312 if ((mp = fp->f_nchandle.mount) == NULL) 1313 mp = ((struct vnode *)fp->f_data)->v_mount; 1314 if (mp == NULL) { 1315 error = EBADF; 1316 goto done; 1317 } 1318 if (fp->f_cred == NULL) { 1319 error = EINVAL; 1320 goto done; 1321 } 1322 sp = &mp->mnt_vstat; 1323 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0) 1324 goto done; 1325 1326 sp->f_flag = 0; 1327 if (mp->mnt_flag & MNT_RDONLY) 1328 sp->f_flag |= ST_RDONLY; 1329 if (mp->mnt_flag & MNT_NOSUID) 1330 sp->f_flag |= ST_NOSUID; 1331 1332 bcopy(sp, buf, sizeof(*buf)); 1333 error = 0; 1334 done: 1335 fdrop(fp); 1336 return (error); 1337 } 1338 1339 /* 1340 * fstatfs_args(int fd, struct statfs *buf) 1341 * 1342 * Get filesystem statistics. 1343 */ 1344 int 1345 sys_fstatvfs(struct fstatvfs_args *uap) 1346 { 1347 struct statvfs buf; 1348 int error; 1349 1350 error = kern_fstatvfs(uap->fd, &buf); 1351 1352 if (error == 0) 1353 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1354 return (error); 1355 } 1356 1357 /* 1358 * getfsstat_args(struct statfs *buf, long bufsize, int flags) 1359 * 1360 * Get statistics on all filesystems. 1361 */ 1362 1363 struct getfsstat_info { 1364 struct statfs *sfsp; 1365 long count; 1366 long maxcount; 1367 int error; 1368 int flags; 1369 struct thread *td; 1370 }; 1371 1372 static int getfsstat_callback(struct mount *, void *); 1373 1374 int 1375 sys_getfsstat(struct getfsstat_args *uap) 1376 { 1377 struct thread *td = curthread; 1378 struct getfsstat_info info; 1379 1380 bzero(&info, sizeof(info)); 1381 1382 info.maxcount = uap->bufsize / sizeof(struct statfs); 1383 info.sfsp = uap->buf; 1384 info.count = 0; 1385 info.flags = uap->flags; 1386 info.td = td; 1387 1388 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD); 1389 if (info.sfsp && info.count > info.maxcount) 1390 uap->sysmsg_result = info.maxcount; 1391 else 1392 uap->sysmsg_result = info.count; 1393 return (info.error); 1394 } 1395 1396 static int 1397 getfsstat_callback(struct mount *mp, void *data) 1398 { 1399 struct getfsstat_info *info = data; 1400 struct statfs *sp; 1401 char *freepath; 1402 char *fullpath; 1403 int error; 1404 1405 if (info->sfsp && info->count < info->maxcount) { 1406 if (info->td->td_proc && 1407 !chroot_visible_mnt(mp, info->td->td_proc)) { 1408 return(0); 1409 } 1410 sp = &mp->mnt_stat; 1411 1412 /* 1413 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1414 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1415 * overrides MNT_WAIT. 1416 */ 1417 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1418 (info->flags & MNT_WAIT)) && 1419 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) { 1420 return(0); 1421 } 1422 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1423 1424 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath); 1425 if (error) { 1426 info->error = error; 1427 return(-1); 1428 } 1429 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1430 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1431 kfree(freepath, M_TEMP); 1432 1433 error = copyout(sp, info->sfsp, sizeof(*sp)); 1434 if (error) { 1435 info->error = error; 1436 return (-1); 1437 } 1438 ++info->sfsp; 1439 } 1440 info->count++; 1441 return(0); 1442 } 1443 1444 /* 1445 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf, 1446 long bufsize, int flags) 1447 * 1448 * Get statistics on all filesystems. 1449 */ 1450 1451 struct getvfsstat_info { 1452 struct statfs *sfsp; 1453 struct statvfs *vsfsp; 1454 long count; 1455 long maxcount; 1456 int error; 1457 int flags; 1458 struct thread *td; 1459 }; 1460 1461 static int getvfsstat_callback(struct mount *, void *); 1462 1463 int 1464 sys_getvfsstat(struct getvfsstat_args *uap) 1465 { 1466 struct thread *td = curthread; 1467 struct getvfsstat_info info; 1468 1469 bzero(&info, sizeof(info)); 1470 1471 info.maxcount = uap->vbufsize / sizeof(struct statvfs); 1472 info.sfsp = uap->buf; 1473 info.vsfsp = uap->vbuf; 1474 info.count = 0; 1475 info.flags = uap->flags; 1476 info.td = td; 1477 1478 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD); 1479 if (info.vsfsp && info.count > info.maxcount) 1480 uap->sysmsg_result = info.maxcount; 1481 else 1482 uap->sysmsg_result = info.count; 1483 return (info.error); 1484 } 1485 1486 static int 1487 getvfsstat_callback(struct mount *mp, void *data) 1488 { 1489 struct getvfsstat_info *info = data; 1490 struct statfs *sp; 1491 struct statvfs *vsp; 1492 char *freepath; 1493 char *fullpath; 1494 int error; 1495 1496 if (info->vsfsp && info->count < info->maxcount) { 1497 if (info->td->td_proc && 1498 !chroot_visible_mnt(mp, info->td->td_proc)) { 1499 return(0); 1500 } 1501 sp = &mp->mnt_stat; 1502 vsp = &mp->mnt_vstat; 1503 1504 /* 1505 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1506 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1507 * overrides MNT_WAIT. 1508 */ 1509 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1510 (info->flags & MNT_WAIT)) && 1511 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) { 1512 return(0); 1513 } 1514 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1515 1516 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1517 (info->flags & MNT_WAIT)) && 1518 (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) { 1519 return(0); 1520 } 1521 vsp->f_flag = 0; 1522 if (mp->mnt_flag & MNT_RDONLY) 1523 vsp->f_flag |= ST_RDONLY; 1524 if (mp->mnt_flag & MNT_NOSUID) 1525 vsp->f_flag |= ST_NOSUID; 1526 1527 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath); 1528 if (error) { 1529 info->error = error; 1530 return(-1); 1531 } 1532 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1533 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1534 kfree(freepath, M_TEMP); 1535 1536 error = copyout(sp, info->sfsp, sizeof(*sp)); 1537 if (error == 0) 1538 error = copyout(vsp, info->vsfsp, sizeof(*vsp)); 1539 if (error) { 1540 info->error = error; 1541 return (-1); 1542 } 1543 ++info->sfsp; 1544 ++info->vsfsp; 1545 } 1546 info->count++; 1547 return(0); 1548 } 1549 1550 1551 /* 1552 * fchdir_args(int fd) 1553 * 1554 * Change current working directory to a given file descriptor. 1555 */ 1556 int 1557 sys_fchdir(struct fchdir_args *uap) 1558 { 1559 struct thread *td = curthread; 1560 struct proc *p = td->td_proc; 1561 struct filedesc *fdp = p->p_fd; 1562 struct vnode *vp, *ovp; 1563 struct mount *mp; 1564 struct file *fp; 1565 struct nchandle nch, onch, tnch; 1566 int error; 1567 1568 if ((error = holdvnode(fdp, uap->fd, &fp)) != 0) 1569 return (error); 1570 lwkt_gettoken(&p->p_token); 1571 vp = (struct vnode *)fp->f_data; 1572 vref(vp); 1573 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1574 if (fp->f_nchandle.ncp == NULL) 1575 error = ENOTDIR; 1576 else 1577 error = checkvp_chdir(vp, td); 1578 if (error) { 1579 vput(vp); 1580 goto done; 1581 } 1582 cache_copy(&fp->f_nchandle, &nch); 1583 1584 /* 1585 * If the ncp has become a mount point, traverse through 1586 * the mount point. 1587 */ 1588 1589 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) && 1590 (mp = cache_findmount(&nch)) != NULL 1591 ) { 1592 error = nlookup_mp(mp, &tnch); 1593 if (error == 0) { 1594 cache_unlock(&tnch); /* leave ref intact */ 1595 vput(vp); 1596 vp = tnch.ncp->nc_vp; 1597 error = vget(vp, LK_SHARED); 1598 KKASSERT(error == 0); 1599 cache_drop(&nch); 1600 nch = tnch; 1601 } 1602 cache_dropmount(mp); 1603 } 1604 if (error == 0) { 1605 ovp = fdp->fd_cdir; 1606 onch = fdp->fd_ncdir; 1607 vn_unlock(vp); /* leave ref intact */ 1608 fdp->fd_cdir = vp; 1609 fdp->fd_ncdir = nch; 1610 cache_drop(&onch); 1611 vrele(ovp); 1612 } else { 1613 cache_drop(&nch); 1614 vput(vp); 1615 } 1616 fdrop(fp); 1617 done: 1618 lwkt_reltoken(&p->p_token); 1619 return (error); 1620 } 1621 1622 int 1623 kern_chdir(struct nlookupdata *nd) 1624 { 1625 struct thread *td = curthread; 1626 struct proc *p = td->td_proc; 1627 struct filedesc *fdp = p->p_fd; 1628 struct vnode *vp, *ovp; 1629 struct nchandle onch; 1630 int error; 1631 1632 if ((error = nlookup(nd)) != 0) 1633 return (error); 1634 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 1635 return (ENOENT); 1636 if ((error = vget(vp, LK_SHARED)) != 0) 1637 return (error); 1638 1639 lwkt_gettoken(&p->p_token); 1640 error = checkvp_chdir(vp, td); 1641 vn_unlock(vp); 1642 if (error == 0) { 1643 ovp = fdp->fd_cdir; 1644 onch = fdp->fd_ncdir; 1645 cache_unlock(&nd->nl_nch); /* leave reference intact */ 1646 fdp->fd_ncdir = nd->nl_nch; 1647 fdp->fd_cdir = vp; 1648 cache_drop(&onch); 1649 vrele(ovp); 1650 cache_zero(&nd->nl_nch); 1651 } else { 1652 vrele(vp); 1653 } 1654 lwkt_reltoken(&p->p_token); 1655 return (error); 1656 } 1657 1658 /* 1659 * chdir_args(char *path) 1660 * 1661 * Change current working directory (``.''). 1662 */ 1663 int 1664 sys_chdir(struct chdir_args *uap) 1665 { 1666 struct nlookupdata nd; 1667 int error; 1668 1669 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1670 if (error == 0) 1671 error = kern_chdir(&nd); 1672 nlookup_done(&nd); 1673 return (error); 1674 } 1675 1676 /* 1677 * Helper function for raised chroot(2) security function: Refuse if 1678 * any filedescriptors are open directories. 1679 */ 1680 static int 1681 chroot_refuse_vdir_fds(struct filedesc *fdp) 1682 { 1683 struct vnode *vp; 1684 struct file *fp; 1685 int error; 1686 int fd; 1687 1688 for (fd = 0; fd < fdp->fd_nfiles ; fd++) { 1689 if ((error = holdvnode(fdp, fd, &fp)) != 0) 1690 continue; 1691 vp = (struct vnode *)fp->f_data; 1692 if (vp->v_type != VDIR) { 1693 fdrop(fp); 1694 continue; 1695 } 1696 fdrop(fp); 1697 return(EPERM); 1698 } 1699 return (0); 1700 } 1701 1702 /* 1703 * This sysctl determines if we will allow a process to chroot(2) if it 1704 * has a directory open: 1705 * 0: disallowed for all processes. 1706 * 1: allowed for processes that were not already chroot(2)'ed. 1707 * 2: allowed for all processes. 1708 */ 1709 1710 static int chroot_allow_open_directories = 1; 1711 1712 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 1713 &chroot_allow_open_directories, 0, ""); 1714 1715 /* 1716 * chroot to the specified namecache entry. We obtain the vp from the 1717 * namecache data. The passed ncp must be locked and referenced and will 1718 * remain locked and referenced on return. 1719 */ 1720 int 1721 kern_chroot(struct nchandle *nch) 1722 { 1723 struct thread *td = curthread; 1724 struct proc *p = td->td_proc; 1725 struct filedesc *fdp = p->p_fd; 1726 struct vnode *vp; 1727 int error; 1728 1729 /* 1730 * Only privileged user can chroot 1731 */ 1732 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0); 1733 if (error) 1734 return (error); 1735 1736 /* 1737 * Disallow open directory descriptors (fchdir() breakouts). 1738 */ 1739 if (chroot_allow_open_directories == 0 || 1740 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 1741 if ((error = chroot_refuse_vdir_fds(fdp)) != 0) 1742 return (error); 1743 } 1744 if ((vp = nch->ncp->nc_vp) == NULL) 1745 return (ENOENT); 1746 1747 if ((error = vget(vp, LK_SHARED)) != 0) 1748 return (error); 1749 1750 /* 1751 * Check the validity of vp as a directory to change to and 1752 * associate it with rdir/jdir. 1753 */ 1754 error = checkvp_chdir(vp, td); 1755 vn_unlock(vp); /* leave reference intact */ 1756 if (error == 0) { 1757 vrele(fdp->fd_rdir); 1758 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */ 1759 cache_drop(&fdp->fd_nrdir); 1760 cache_copy(nch, &fdp->fd_nrdir); 1761 if (fdp->fd_jdir == NULL) { 1762 fdp->fd_jdir = vp; 1763 vref(fdp->fd_jdir); 1764 cache_copy(nch, &fdp->fd_njdir); 1765 } 1766 } else { 1767 vrele(vp); 1768 } 1769 return (error); 1770 } 1771 1772 /* 1773 * chroot_args(char *path) 1774 * 1775 * Change notion of root (``/'') directory. 1776 */ 1777 int 1778 sys_chroot(struct chroot_args *uap) 1779 { 1780 struct thread *td __debugvar = curthread; 1781 struct nlookupdata nd; 1782 int error; 1783 1784 KKASSERT(td->td_proc); 1785 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1786 if (error == 0) { 1787 nd.nl_flags |= NLC_EXEC; 1788 error = nlookup(&nd); 1789 if (error == 0) 1790 error = kern_chroot(&nd.nl_nch); 1791 } 1792 nlookup_done(&nd); 1793 return(error); 1794 } 1795 1796 int 1797 sys_chroot_kernel(struct chroot_kernel_args *uap) 1798 { 1799 struct thread *td = curthread; 1800 struct nlookupdata nd; 1801 struct nchandle *nch; 1802 struct vnode *vp; 1803 int error; 1804 1805 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1806 if (error) 1807 goto error_nond; 1808 1809 error = nlookup(&nd); 1810 if (error) 1811 goto error_out; 1812 1813 nch = &nd.nl_nch; 1814 1815 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0); 1816 if (error) 1817 goto error_out; 1818 1819 if ((vp = nch->ncp->nc_vp) == NULL) { 1820 error = ENOENT; 1821 goto error_out; 1822 } 1823 1824 if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0) 1825 goto error_out; 1826 1827 kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path); 1828 get_mplock(); 1829 vfs_cache_setroot(vp, cache_hold(nch)); 1830 rel_mplock(); 1831 1832 error_out: 1833 nlookup_done(&nd); 1834 error_nond: 1835 return(error); 1836 } 1837 1838 /* 1839 * Common routine for chroot and chdir. Given a locked, referenced vnode, 1840 * determine whether it is legal to chdir to the vnode. The vnode's state 1841 * is not changed by this call. 1842 */ 1843 int 1844 checkvp_chdir(struct vnode *vp, struct thread *td) 1845 { 1846 int error; 1847 1848 if (vp->v_type != VDIR) 1849 error = ENOTDIR; 1850 else 1851 error = VOP_EACCESS(vp, VEXEC, td->td_ucred); 1852 return (error); 1853 } 1854 1855 int 1856 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res) 1857 { 1858 struct thread *td = curthread; 1859 struct proc *p = td->td_proc; 1860 struct lwp *lp = td->td_lwp; 1861 struct filedesc *fdp = p->p_fd; 1862 int cmode, flags; 1863 struct file *nfp; 1864 struct file *fp; 1865 struct vnode *vp; 1866 int type, indx, error = 0; 1867 struct flock lf; 1868 1869 if ((oflags & O_ACCMODE) == O_ACCMODE) 1870 return (EINVAL); 1871 flags = FFLAGS(oflags); 1872 error = falloc(lp, &nfp, NULL); 1873 if (error) 1874 return (error); 1875 fp = nfp; 1876 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; 1877 1878 /* 1879 * XXX p_dupfd is a real mess. It allows a device to return a 1880 * file descriptor to be duplicated rather then doing the open 1881 * itself. 1882 */ 1883 lp->lwp_dupfd = -1; 1884 1885 /* 1886 * Call vn_open() to do the lookup and assign the vnode to the 1887 * file pointer. vn_open() does not change the ref count on fp 1888 * and the vnode, on success, will be inherited by the file pointer 1889 * and unlocked. 1890 */ 1891 nd->nl_flags |= NLC_LOCKVP; 1892 error = vn_open(nd, fp, flags, cmode); 1893 nlookup_done(nd); 1894 if (error) { 1895 /* 1896 * handle special fdopen() case. bleh. dupfdopen() is 1897 * responsible for dropping the old contents of ofiles[indx] 1898 * if it succeeds. 1899 * 1900 * Note that fsetfd() will add a ref to fp which represents 1901 * the fd_files[] assignment. We must still drop our 1902 * reference. 1903 */ 1904 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) { 1905 if (fdalloc(p, 0, &indx) == 0) { 1906 error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error); 1907 if (error == 0) { 1908 *res = indx; 1909 fdrop(fp); /* our ref */ 1910 return (0); 1911 } 1912 fsetfd(fdp, NULL, indx); 1913 } 1914 } 1915 fdrop(fp); /* our ref */ 1916 if (error == ERESTART) 1917 error = EINTR; 1918 return (error); 1919 } 1920 1921 /* 1922 * ref the vnode for ourselves so it can't be ripped out from under 1923 * is. XXX need an ND flag to request that the vnode be returned 1924 * anyway. 1925 * 1926 * Reserve a file descriptor but do not assign it until the open 1927 * succeeds. 1928 */ 1929 vp = (struct vnode *)fp->f_data; 1930 vref(vp); 1931 if ((error = fdalloc(p, 0, &indx)) != 0) { 1932 fdrop(fp); 1933 vrele(vp); 1934 return (error); 1935 } 1936 1937 /* 1938 * If no error occurs the vp will have been assigned to the file 1939 * pointer. 1940 */ 1941 lp->lwp_dupfd = 0; 1942 1943 if (flags & (O_EXLOCK | O_SHLOCK)) { 1944 lf.l_whence = SEEK_SET; 1945 lf.l_start = 0; 1946 lf.l_len = 0; 1947 if (flags & O_EXLOCK) 1948 lf.l_type = F_WRLCK; 1949 else 1950 lf.l_type = F_RDLCK; 1951 if (flags & FNONBLOCK) 1952 type = 0; 1953 else 1954 type = F_WAIT; 1955 1956 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) { 1957 /* 1958 * lock request failed. Clean up the reserved 1959 * descriptor. 1960 */ 1961 vrele(vp); 1962 fsetfd(fdp, NULL, indx); 1963 fdrop(fp); 1964 return (error); 1965 } 1966 fp->f_flag |= FHASLOCK; 1967 } 1968 #if 0 1969 /* 1970 * Assert that all regular file vnodes were created with a object. 1971 */ 1972 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 1973 ("open: regular file has no backing object after vn_open")); 1974 #endif 1975 1976 vrele(vp); 1977 1978 /* 1979 * release our private reference, leaving the one associated with the 1980 * descriptor table intact. 1981 */ 1982 fsetfd(fdp, fp, indx); 1983 fdrop(fp); 1984 *res = indx; 1985 if (oflags & O_CLOEXEC) 1986 error = fsetfdflags(fdp, *res, UF_EXCLOSE); 1987 return (error); 1988 } 1989 1990 /* 1991 * open_args(char *path, int flags, int mode) 1992 * 1993 * Check permissions, allocate an open file structure, 1994 * and call the device open routine if any. 1995 */ 1996 int 1997 sys_open(struct open_args *uap) 1998 { 1999 struct nlookupdata nd; 2000 int error; 2001 2002 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2003 if (error == 0) { 2004 error = kern_open(&nd, uap->flags, 2005 uap->mode, &uap->sysmsg_result); 2006 } 2007 nlookup_done(&nd); 2008 return (error); 2009 } 2010 2011 /* 2012 * openat_args(int fd, char *path, int flags, int mode) 2013 */ 2014 int 2015 sys_openat(struct openat_args *uap) 2016 { 2017 struct nlookupdata nd; 2018 int error; 2019 struct file *fp; 2020 2021 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2022 if (error == 0) { 2023 error = kern_open(&nd, uap->flags, uap->mode, 2024 &uap->sysmsg_result); 2025 } 2026 nlookup_done_at(&nd, fp); 2027 return (error); 2028 } 2029 2030 int 2031 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor) 2032 { 2033 struct thread *td = curthread; 2034 struct proc *p = td->td_proc; 2035 struct vnode *vp; 2036 struct vattr vattr; 2037 int error; 2038 int whiteout = 0; 2039 2040 KKASSERT(p); 2041 2042 VATTR_NULL(&vattr); 2043 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2044 vattr.va_rmajor = rmajor; 2045 vattr.va_rminor = rminor; 2046 2047 switch (mode & S_IFMT) { 2048 case S_IFMT: /* used by badsect to flag bad sectors */ 2049 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0); 2050 vattr.va_type = VBAD; 2051 break; 2052 case S_IFCHR: 2053 error = priv_check(td, PRIV_VFS_MKNOD_DEV); 2054 vattr.va_type = VCHR; 2055 break; 2056 case S_IFBLK: 2057 error = priv_check(td, PRIV_VFS_MKNOD_DEV); 2058 vattr.va_type = VBLK; 2059 break; 2060 case S_IFWHT: 2061 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0); 2062 whiteout = 1; 2063 break; 2064 case S_IFDIR: /* special directories support for HAMMER */ 2065 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0); 2066 vattr.va_type = VDIR; 2067 break; 2068 default: 2069 error = EINVAL; 2070 break; 2071 } 2072 2073 if (error) 2074 return (error); 2075 2076 bwillinode(1); 2077 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2078 if ((error = nlookup(nd)) != 0) 2079 return (error); 2080 if (nd->nl_nch.ncp->nc_vp) 2081 return (EEXIST); 2082 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2083 return (error); 2084 2085 if (whiteout) { 2086 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp, 2087 nd->nl_cred, NAMEI_CREATE); 2088 } else { 2089 vp = NULL; 2090 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, 2091 &vp, nd->nl_cred, &vattr); 2092 if (error == 0) 2093 vput(vp); 2094 } 2095 return (error); 2096 } 2097 2098 /* 2099 * mknod_args(char *path, int mode, int dev) 2100 * 2101 * Create a special file. 2102 */ 2103 int 2104 sys_mknod(struct mknod_args *uap) 2105 { 2106 struct nlookupdata nd; 2107 int error; 2108 2109 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2110 if (error == 0) { 2111 error = kern_mknod(&nd, uap->mode, 2112 umajor(uap->dev), uminor(uap->dev)); 2113 } 2114 nlookup_done(&nd); 2115 return (error); 2116 } 2117 2118 /* 2119 * mknodat_args(int fd, char *path, mode_t mode, dev_t dev) 2120 * 2121 * Create a special file. The path is relative to the directory associated 2122 * with fd. 2123 */ 2124 int 2125 sys_mknodat(struct mknodat_args *uap) 2126 { 2127 struct nlookupdata nd; 2128 struct file *fp; 2129 int error; 2130 2131 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2132 if (error == 0) { 2133 error = kern_mknod(&nd, uap->mode, 2134 umajor(uap->dev), uminor(uap->dev)); 2135 } 2136 nlookup_done_at(&nd, fp); 2137 return (error); 2138 } 2139 2140 int 2141 kern_mkfifo(struct nlookupdata *nd, int mode) 2142 { 2143 struct thread *td = curthread; 2144 struct proc *p = td->td_proc; 2145 struct vattr vattr; 2146 struct vnode *vp; 2147 int error; 2148 2149 bwillinode(1); 2150 2151 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2152 if ((error = nlookup(nd)) != 0) 2153 return (error); 2154 if (nd->nl_nch.ncp->nc_vp) 2155 return (EEXIST); 2156 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2157 return (error); 2158 2159 VATTR_NULL(&vattr); 2160 vattr.va_type = VFIFO; 2161 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2162 vp = NULL; 2163 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr); 2164 if (error == 0) 2165 vput(vp); 2166 return (error); 2167 } 2168 2169 /* 2170 * mkfifo_args(char *path, int mode) 2171 * 2172 * Create a named pipe. 2173 */ 2174 int 2175 sys_mkfifo(struct mkfifo_args *uap) 2176 { 2177 struct nlookupdata nd; 2178 int error; 2179 2180 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2181 if (error == 0) 2182 error = kern_mkfifo(&nd, uap->mode); 2183 nlookup_done(&nd); 2184 return (error); 2185 } 2186 2187 /* 2188 * mkfifoat_args(int fd, char *path, mode_t mode) 2189 * 2190 * Create a named pipe. The path is relative to the directory associated 2191 * with fd. 2192 */ 2193 int 2194 sys_mkfifoat(struct mkfifoat_args *uap) 2195 { 2196 struct nlookupdata nd; 2197 struct file *fp; 2198 int error; 2199 2200 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2201 if (error == 0) 2202 error = kern_mkfifo(&nd, uap->mode); 2203 nlookup_done_at(&nd, fp); 2204 return (error); 2205 } 2206 2207 static int hardlink_check_uid = 0; 2208 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, 2209 &hardlink_check_uid, 0, 2210 "Unprivileged processes cannot create hard links to files owned by other " 2211 "users"); 2212 static int hardlink_check_gid = 0; 2213 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, 2214 &hardlink_check_gid, 0, 2215 "Unprivileged processes cannot create hard links to files owned by other " 2216 "groups"); 2217 2218 static int 2219 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred) 2220 { 2221 struct vattr va; 2222 int error; 2223 2224 /* 2225 * Shortcut if disabled 2226 */ 2227 if (hardlink_check_uid == 0 && hardlink_check_gid == 0) 2228 return (0); 2229 2230 /* 2231 * Privileged user can always hardlink 2232 */ 2233 if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0) 2234 return (0); 2235 2236 /* 2237 * Otherwise only if the originating file is owned by the 2238 * same user or group. Note that any group is allowed if 2239 * the file is owned by the caller. 2240 */ 2241 error = VOP_GETATTR(vp, &va); 2242 if (error != 0) 2243 return (error); 2244 2245 if (hardlink_check_uid) { 2246 if (cred->cr_uid != va.va_uid) 2247 return (EPERM); 2248 } 2249 2250 if (hardlink_check_gid) { 2251 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred)) 2252 return (EPERM); 2253 } 2254 2255 return (0); 2256 } 2257 2258 int 2259 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd) 2260 { 2261 struct thread *td = curthread; 2262 struct vnode *vp; 2263 int error; 2264 2265 /* 2266 * Lookup the source and obtained a locked vnode. 2267 * 2268 * You may only hardlink a file which you have write permission 2269 * on or which you own. 2270 * 2271 * XXX relookup on vget failure / race ? 2272 */ 2273 bwillinode(1); 2274 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK; 2275 if ((error = nlookup(nd)) != 0) 2276 return (error); 2277 vp = nd->nl_nch.ncp->nc_vp; 2278 KKASSERT(vp != NULL); 2279 if (vp->v_type == VDIR) 2280 return (EPERM); /* POSIX */ 2281 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2282 return (error); 2283 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) 2284 return (error); 2285 2286 /* 2287 * Unlock the source so we can lookup the target without deadlocking 2288 * (XXX vp is locked already, possible other deadlock?). The target 2289 * must not exist. 2290 */ 2291 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 2292 nd->nl_flags &= ~NLC_NCPISLOCKED; 2293 cache_unlock(&nd->nl_nch); 2294 vn_unlock(vp); 2295 2296 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2297 if ((error = nlookup(linknd)) != 0) { 2298 vrele(vp); 2299 return (error); 2300 } 2301 if (linknd->nl_nch.ncp->nc_vp) { 2302 vrele(vp); 2303 return (EEXIST); 2304 } 2305 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) { 2306 vrele(vp); 2307 return (error); 2308 } 2309 2310 /* 2311 * Finally run the new API VOP. 2312 */ 2313 error = can_hardlink(vp, td, td->td_ucred); 2314 if (error == 0) { 2315 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp, 2316 vp, linknd->nl_cred); 2317 } 2318 vput(vp); 2319 return (error); 2320 } 2321 2322 /* 2323 * link_args(char *path, char *link) 2324 * 2325 * Make a hard file link. 2326 */ 2327 int 2328 sys_link(struct link_args *uap) 2329 { 2330 struct nlookupdata nd, linknd; 2331 int error; 2332 2333 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2334 if (error == 0) { 2335 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0); 2336 if (error == 0) 2337 error = kern_link(&nd, &linknd); 2338 nlookup_done(&linknd); 2339 } 2340 nlookup_done(&nd); 2341 return (error); 2342 } 2343 2344 /* 2345 * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags) 2346 * 2347 * Make a hard file link. The path1 argument is relative to the directory 2348 * associated with fd1, and similarly the path2 argument is relative to 2349 * the directory associated with fd2. 2350 */ 2351 int 2352 sys_linkat(struct linkat_args *uap) 2353 { 2354 struct nlookupdata nd, linknd; 2355 struct file *fp1, *fp2; 2356 int error; 2357 2358 error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE, 2359 (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0); 2360 if (error == 0) { 2361 error = nlookup_init_at(&linknd, &fp2, uap->fd2, 2362 uap->path2, UIO_USERSPACE, 0); 2363 if (error == 0) 2364 error = kern_link(&nd, &linknd); 2365 nlookup_done_at(&linknd, fp2); 2366 } 2367 nlookup_done_at(&nd, fp1); 2368 return (error); 2369 } 2370 2371 int 2372 kern_symlink(struct nlookupdata *nd, char *path, int mode) 2373 { 2374 struct vattr vattr; 2375 struct vnode *vp; 2376 struct vnode *dvp; 2377 int error; 2378 2379 bwillinode(1); 2380 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2381 if ((error = nlookup(nd)) != 0) 2382 return (error); 2383 if (nd->nl_nch.ncp->nc_vp) 2384 return (EEXIST); 2385 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2386 return (error); 2387 dvp = nd->nl_dvp; 2388 VATTR_NULL(&vattr); 2389 vattr.va_mode = mode; 2390 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path); 2391 if (error == 0) 2392 vput(vp); 2393 return (error); 2394 } 2395 2396 /* 2397 * symlink(char *path, char *link) 2398 * 2399 * Make a symbolic link. 2400 */ 2401 int 2402 sys_symlink(struct symlink_args *uap) 2403 { 2404 struct thread *td = curthread; 2405 struct nlookupdata nd; 2406 char *path; 2407 int error; 2408 int mode; 2409 2410 path = objcache_get(namei_oc, M_WAITOK); 2411 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 2412 if (error == 0) { 2413 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0); 2414 if (error == 0) { 2415 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2416 error = kern_symlink(&nd, path, mode); 2417 } 2418 nlookup_done(&nd); 2419 } 2420 objcache_put(namei_oc, path); 2421 return (error); 2422 } 2423 2424 /* 2425 * symlinkat_args(char *path1, int fd, char *path2) 2426 * 2427 * Make a symbolic link. The path2 argument is relative to the directory 2428 * associated with fd. 2429 */ 2430 int 2431 sys_symlinkat(struct symlinkat_args *uap) 2432 { 2433 struct thread *td = curthread; 2434 struct nlookupdata nd; 2435 struct file *fp; 2436 char *path1; 2437 int error; 2438 int mode; 2439 2440 path1 = objcache_get(namei_oc, M_WAITOK); 2441 error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL); 2442 if (error == 0) { 2443 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2, 2444 UIO_USERSPACE, 0); 2445 if (error == 0) { 2446 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2447 error = kern_symlink(&nd, path1, mode); 2448 } 2449 nlookup_done_at(&nd, fp); 2450 } 2451 objcache_put(namei_oc, path1); 2452 return (error); 2453 } 2454 2455 /* 2456 * undelete_args(char *path) 2457 * 2458 * Delete a whiteout from the filesystem. 2459 */ 2460 int 2461 sys_undelete(struct undelete_args *uap) 2462 { 2463 struct nlookupdata nd; 2464 int error; 2465 2466 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2467 bwillinode(1); 2468 nd.nl_flags |= NLC_DELETE | NLC_REFDVP; 2469 if (error == 0) 2470 error = nlookup(&nd); 2471 if (error == 0) 2472 error = ncp_writechk(&nd.nl_nch); 2473 if (error == 0) { 2474 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred, 2475 NAMEI_DELETE); 2476 } 2477 nlookup_done(&nd); 2478 return (error); 2479 } 2480 2481 int 2482 kern_unlink(struct nlookupdata *nd) 2483 { 2484 int error; 2485 2486 bwillinode(1); 2487 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 2488 if ((error = nlookup(nd)) != 0) 2489 return (error); 2490 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2491 return (error); 2492 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 2493 return (error); 2494 } 2495 2496 /* 2497 * unlink_args(char *path) 2498 * 2499 * Delete a name from the filesystem. 2500 */ 2501 int 2502 sys_unlink(struct unlink_args *uap) 2503 { 2504 struct nlookupdata nd; 2505 int error; 2506 2507 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2508 if (error == 0) 2509 error = kern_unlink(&nd); 2510 nlookup_done(&nd); 2511 return (error); 2512 } 2513 2514 2515 /* 2516 * unlinkat_args(int fd, char *path, int flags) 2517 * 2518 * Delete the file or directory entry pointed to by fd/path. 2519 */ 2520 int 2521 sys_unlinkat(struct unlinkat_args *uap) 2522 { 2523 struct nlookupdata nd; 2524 struct file *fp; 2525 int error; 2526 2527 if (uap->flags & ~AT_REMOVEDIR) 2528 return (EINVAL); 2529 2530 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2531 if (error == 0) { 2532 if (uap->flags & AT_REMOVEDIR) 2533 error = kern_rmdir(&nd); 2534 else 2535 error = kern_unlink(&nd); 2536 } 2537 nlookup_done_at(&nd, fp); 2538 return (error); 2539 } 2540 2541 int 2542 kern_lseek(int fd, off_t offset, int whence, off_t *res) 2543 { 2544 struct thread *td = curthread; 2545 struct proc *p = td->td_proc; 2546 struct file *fp; 2547 struct vnode *vp; 2548 struct vattr vattr; 2549 off_t new_offset; 2550 int error; 2551 2552 fp = holdfp(p->p_fd, fd, -1); 2553 if (fp == NULL) 2554 return (EBADF); 2555 if (fp->f_type != DTYPE_VNODE) { 2556 error = ESPIPE; 2557 goto done; 2558 } 2559 vp = (struct vnode *)fp->f_data; 2560 2561 switch (whence) { 2562 case L_INCR: 2563 spin_lock(&fp->f_spin); 2564 new_offset = fp->f_offset + offset; 2565 error = 0; 2566 break; 2567 case L_XTND: 2568 error = VOP_GETATTR(vp, &vattr); 2569 spin_lock(&fp->f_spin); 2570 new_offset = offset + vattr.va_size; 2571 break; 2572 case L_SET: 2573 new_offset = offset; 2574 error = 0; 2575 spin_lock(&fp->f_spin); 2576 break; 2577 default: 2578 new_offset = 0; 2579 error = EINVAL; 2580 spin_lock(&fp->f_spin); 2581 break; 2582 } 2583 2584 /* 2585 * Validate the seek position. Negative offsets are not allowed 2586 * for regular files or directories. 2587 * 2588 * Normally we would also not want to allow negative offsets for 2589 * character and block-special devices. However kvm addresses 2590 * on 64 bit architectures might appear to be negative and must 2591 * be allowed. 2592 */ 2593 if (error == 0) { 2594 if (new_offset < 0 && 2595 (vp->v_type == VREG || vp->v_type == VDIR)) { 2596 error = EINVAL; 2597 } else { 2598 fp->f_offset = new_offset; 2599 } 2600 } 2601 *res = fp->f_offset; 2602 spin_unlock(&fp->f_spin); 2603 done: 2604 fdrop(fp); 2605 return (error); 2606 } 2607 2608 /* 2609 * lseek_args(int fd, int pad, off_t offset, int whence) 2610 * 2611 * Reposition read/write file offset. 2612 */ 2613 int 2614 sys_lseek(struct lseek_args *uap) 2615 { 2616 int error; 2617 2618 error = kern_lseek(uap->fd, uap->offset, uap->whence, 2619 &uap->sysmsg_offset); 2620 2621 return (error); 2622 } 2623 2624 /* 2625 * Check if current process can access given file. amode is a bitmask of *_OK 2626 * access bits. flags is a bitmask of AT_* flags. 2627 */ 2628 int 2629 kern_access(struct nlookupdata *nd, int amode, int flags) 2630 { 2631 struct vnode *vp; 2632 int error, mode; 2633 2634 if (flags & ~AT_EACCESS) 2635 return (EINVAL); 2636 if ((error = nlookup(nd)) != 0) 2637 return (error); 2638 retry: 2639 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp); 2640 if (error) 2641 return (error); 2642 2643 /* Flags == 0 means only check for existence. */ 2644 if (amode) { 2645 mode = 0; 2646 if (amode & R_OK) 2647 mode |= VREAD; 2648 if (amode & W_OK) 2649 mode |= VWRITE; 2650 if (amode & X_OK) 2651 mode |= VEXEC; 2652 if ((mode & VWRITE) == 0 || 2653 (error = vn_writechk(vp, &nd->nl_nch)) == 0) 2654 error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred); 2655 2656 /* 2657 * If the file handle is stale we have to re-resolve the 2658 * entry. This is a hack at the moment. 2659 */ 2660 if (error == ESTALE) { 2661 vput(vp); 2662 cache_setunresolved(&nd->nl_nch); 2663 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2664 if (error == 0) { 2665 vp = NULL; 2666 goto retry; 2667 } 2668 return(error); 2669 } 2670 } 2671 vput(vp); 2672 return (error); 2673 } 2674 2675 /* 2676 * access_args(char *path, int flags) 2677 * 2678 * Check access permissions. 2679 */ 2680 int 2681 sys_access(struct access_args *uap) 2682 { 2683 struct nlookupdata nd; 2684 int error; 2685 2686 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2687 if (error == 0) 2688 error = kern_access(&nd, uap->flags, 0); 2689 nlookup_done(&nd); 2690 return (error); 2691 } 2692 2693 2694 /* 2695 * eaccess_args(char *path, int flags) 2696 * 2697 * Check access permissions. 2698 */ 2699 int 2700 sys_eaccess(struct eaccess_args *uap) 2701 { 2702 struct nlookupdata nd; 2703 int error; 2704 2705 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2706 if (error == 0) 2707 error = kern_access(&nd, uap->flags, AT_EACCESS); 2708 nlookup_done(&nd); 2709 return (error); 2710 } 2711 2712 2713 /* 2714 * faccessat_args(int fd, char *path, int amode, int flags) 2715 * 2716 * Check access permissions. 2717 */ 2718 int 2719 sys_faccessat(struct faccessat_args *uap) 2720 { 2721 struct nlookupdata nd; 2722 struct file *fp; 2723 int error; 2724 2725 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 2726 NLC_FOLLOW); 2727 if (error == 0) 2728 error = kern_access(&nd, uap->amode, uap->flags); 2729 nlookup_done_at(&nd, fp); 2730 return (error); 2731 } 2732 2733 2734 int 2735 kern_stat(struct nlookupdata *nd, struct stat *st) 2736 { 2737 int error; 2738 struct vnode *vp; 2739 2740 if ((error = nlookup(nd)) != 0) 2741 return (error); 2742 again: 2743 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 2744 return (ENOENT); 2745 2746 if ((error = vget(vp, LK_SHARED)) != 0) 2747 return (error); 2748 error = vn_stat(vp, st, nd->nl_cred); 2749 2750 /* 2751 * If the file handle is stale we have to re-resolve the entry. This 2752 * is a hack at the moment. 2753 */ 2754 if (error == ESTALE) { 2755 vput(vp); 2756 cache_setunresolved(&nd->nl_nch); 2757 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2758 if (error == 0) 2759 goto again; 2760 } else { 2761 vput(vp); 2762 } 2763 return (error); 2764 } 2765 2766 /* 2767 * stat_args(char *path, struct stat *ub) 2768 * 2769 * Get file status; this version follows links. 2770 */ 2771 int 2772 sys_stat(struct stat_args *uap) 2773 { 2774 struct nlookupdata nd; 2775 struct stat st; 2776 int error; 2777 2778 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2779 if (error == 0) { 2780 error = kern_stat(&nd, &st); 2781 if (error == 0) 2782 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 2783 } 2784 nlookup_done(&nd); 2785 return (error); 2786 } 2787 2788 /* 2789 * lstat_args(char *path, struct stat *ub) 2790 * 2791 * Get file status; this version does not follow links. 2792 */ 2793 int 2794 sys_lstat(struct lstat_args *uap) 2795 { 2796 struct nlookupdata nd; 2797 struct stat st; 2798 int error; 2799 2800 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2801 if (error == 0) { 2802 error = kern_stat(&nd, &st); 2803 if (error == 0) 2804 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 2805 } 2806 nlookup_done(&nd); 2807 return (error); 2808 } 2809 2810 /* 2811 * fstatat_args(int fd, char *path, struct stat *sb, int flags) 2812 * 2813 * Get status of file pointed to by fd/path. 2814 */ 2815 int 2816 sys_fstatat(struct fstatat_args *uap) 2817 { 2818 struct nlookupdata nd; 2819 struct stat st; 2820 int error; 2821 int flags; 2822 struct file *fp; 2823 2824 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 2825 return (EINVAL); 2826 2827 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 2828 2829 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 2830 UIO_USERSPACE, flags); 2831 if (error == 0) { 2832 error = kern_stat(&nd, &st); 2833 if (error == 0) 2834 error = copyout(&st, uap->sb, sizeof(*uap->sb)); 2835 } 2836 nlookup_done_at(&nd, fp); 2837 return (error); 2838 } 2839 2840 static int 2841 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp) 2842 { 2843 struct nlookupdata nd; 2844 struct vnode *vp; 2845 int error; 2846 2847 vp = NULL; 2848 error = nlookup_init(&nd, path, UIO_USERSPACE, flags); 2849 if (error == 0) 2850 error = nlookup(&nd); 2851 if (error == 0) 2852 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 2853 nlookup_done(&nd); 2854 if (error == 0) { 2855 error = VOP_PATHCONF(vp, name, sysmsg_regp); 2856 vput(vp); 2857 } 2858 return (error); 2859 } 2860 2861 /* 2862 * pathconf_Args(char *path, int name) 2863 * 2864 * Get configurable pathname variables. 2865 */ 2866 int 2867 sys_pathconf(struct pathconf_args *uap) 2868 { 2869 return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW, 2870 &uap->sysmsg_reg)); 2871 } 2872 2873 /* 2874 * lpathconf_Args(char *path, int name) 2875 * 2876 * Get configurable pathname variables, but don't follow symlinks. 2877 */ 2878 int 2879 sys_lpathconf(struct lpathconf_args *uap) 2880 { 2881 return (kern_pathconf(uap->path, uap->name, 0, &uap->sysmsg_reg)); 2882 } 2883 2884 /* 2885 * XXX: daver 2886 * kern_readlink isn't properly split yet. There is a copyin burried 2887 * in VOP_READLINK(). 2888 */ 2889 int 2890 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res) 2891 { 2892 struct thread *td = curthread; 2893 struct vnode *vp; 2894 struct iovec aiov; 2895 struct uio auio; 2896 int error; 2897 2898 if ((error = nlookup(nd)) != 0) 2899 return (error); 2900 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp); 2901 if (error) 2902 return (error); 2903 if (vp->v_type != VLNK) { 2904 error = EINVAL; 2905 } else { 2906 aiov.iov_base = buf; 2907 aiov.iov_len = count; 2908 auio.uio_iov = &aiov; 2909 auio.uio_iovcnt = 1; 2910 auio.uio_offset = 0; 2911 auio.uio_rw = UIO_READ; 2912 auio.uio_segflg = UIO_USERSPACE; 2913 auio.uio_td = td; 2914 auio.uio_resid = count; 2915 error = VOP_READLINK(vp, &auio, td->td_ucred); 2916 } 2917 vput(vp); 2918 *res = count - auio.uio_resid; 2919 return (error); 2920 } 2921 2922 /* 2923 * readlink_args(char *path, char *buf, int count) 2924 * 2925 * Return target name of a symbolic link. 2926 */ 2927 int 2928 sys_readlink(struct readlink_args *uap) 2929 { 2930 struct nlookupdata nd; 2931 int error; 2932 2933 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2934 if (error == 0) { 2935 error = kern_readlink(&nd, uap->buf, uap->count, 2936 &uap->sysmsg_result); 2937 } 2938 nlookup_done(&nd); 2939 return (error); 2940 } 2941 2942 /* 2943 * readlinkat_args(int fd, char *path, char *buf, size_t bufsize) 2944 * 2945 * Return target name of a symbolic link. The path is relative to the 2946 * directory associated with fd. 2947 */ 2948 int 2949 sys_readlinkat(struct readlinkat_args *uap) 2950 { 2951 struct nlookupdata nd; 2952 struct file *fp; 2953 int error; 2954 2955 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2956 if (error == 0) { 2957 error = kern_readlink(&nd, uap->buf, uap->bufsize, 2958 &uap->sysmsg_result); 2959 } 2960 nlookup_done_at(&nd, fp); 2961 return (error); 2962 } 2963 2964 static int 2965 setfflags(struct vnode *vp, int flags) 2966 { 2967 struct thread *td = curthread; 2968 int error; 2969 struct vattr vattr; 2970 2971 /* 2972 * Prevent non-root users from setting flags on devices. When 2973 * a device is reused, users can retain ownership of the device 2974 * if they are allowed to set flags and programs assume that 2975 * chown can't fail when done as root. 2976 */ 2977 if ((vp->v_type == VCHR || vp->v_type == VBLK) && 2978 ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0)) 2979 return (error); 2980 2981 /* 2982 * note: vget is required for any operation that might mod the vnode 2983 * so VINACTIVE is properly cleared. 2984 */ 2985 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 2986 VATTR_NULL(&vattr); 2987 vattr.va_flags = flags; 2988 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 2989 vput(vp); 2990 } 2991 return (error); 2992 } 2993 2994 /* 2995 * chflags(char *path, int flags) 2996 * 2997 * Change flags of a file given a path name. 2998 */ 2999 int 3000 sys_chflags(struct chflags_args *uap) 3001 { 3002 struct nlookupdata nd; 3003 struct vnode *vp; 3004 int error; 3005 3006 vp = NULL; 3007 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3008 if (error == 0) 3009 error = nlookup(&nd); 3010 if (error == 0) 3011 error = ncp_writechk(&nd.nl_nch); 3012 if (error == 0) 3013 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3014 nlookup_done(&nd); 3015 if (error == 0) { 3016 error = setfflags(vp, uap->flags); 3017 vrele(vp); 3018 } 3019 return (error); 3020 } 3021 3022 /* 3023 * lchflags(char *path, int flags) 3024 * 3025 * Change flags of a file given a path name, but don't follow symlinks. 3026 */ 3027 int 3028 sys_lchflags(struct lchflags_args *uap) 3029 { 3030 struct nlookupdata nd; 3031 struct vnode *vp; 3032 int error; 3033 3034 vp = NULL; 3035 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3036 if (error == 0) 3037 error = nlookup(&nd); 3038 if (error == 0) 3039 error = ncp_writechk(&nd.nl_nch); 3040 if (error == 0) 3041 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3042 nlookup_done(&nd); 3043 if (error == 0) { 3044 error = setfflags(vp, uap->flags); 3045 vrele(vp); 3046 } 3047 return (error); 3048 } 3049 3050 /* 3051 * fchflags_args(int fd, int flags) 3052 * 3053 * Change flags of a file given a file descriptor. 3054 */ 3055 int 3056 sys_fchflags(struct fchflags_args *uap) 3057 { 3058 struct thread *td = curthread; 3059 struct proc *p = td->td_proc; 3060 struct file *fp; 3061 int error; 3062 3063 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3064 return (error); 3065 if (fp->f_nchandle.ncp) 3066 error = ncp_writechk(&fp->f_nchandle); 3067 if (error == 0) 3068 error = setfflags((struct vnode *) fp->f_data, uap->flags); 3069 fdrop(fp); 3070 return (error); 3071 } 3072 3073 static int 3074 setfmode(struct vnode *vp, int mode) 3075 { 3076 struct thread *td = curthread; 3077 int error; 3078 struct vattr vattr; 3079 3080 /* 3081 * note: vget is required for any operation that might mod the vnode 3082 * so VINACTIVE is properly cleared. 3083 */ 3084 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3085 VATTR_NULL(&vattr); 3086 vattr.va_mode = mode & ALLPERMS; 3087 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3088 vput(vp); 3089 } 3090 return error; 3091 } 3092 3093 int 3094 kern_chmod(struct nlookupdata *nd, int mode) 3095 { 3096 struct vnode *vp; 3097 int error; 3098 3099 if ((error = nlookup(nd)) != 0) 3100 return (error); 3101 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3102 return (error); 3103 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3104 error = setfmode(vp, mode); 3105 vrele(vp); 3106 return (error); 3107 } 3108 3109 /* 3110 * chmod_args(char *path, int mode) 3111 * 3112 * Change mode of a file given path name. 3113 */ 3114 int 3115 sys_chmod(struct chmod_args *uap) 3116 { 3117 struct nlookupdata nd; 3118 int error; 3119 3120 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3121 if (error == 0) 3122 error = kern_chmod(&nd, uap->mode); 3123 nlookup_done(&nd); 3124 return (error); 3125 } 3126 3127 /* 3128 * lchmod_args(char *path, int mode) 3129 * 3130 * Change mode of a file given path name (don't follow links.) 3131 */ 3132 int 3133 sys_lchmod(struct lchmod_args *uap) 3134 { 3135 struct nlookupdata nd; 3136 int error; 3137 3138 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3139 if (error == 0) 3140 error = kern_chmod(&nd, uap->mode); 3141 nlookup_done(&nd); 3142 return (error); 3143 } 3144 3145 /* 3146 * fchmod_args(int fd, int mode) 3147 * 3148 * Change mode of a file given a file descriptor. 3149 */ 3150 int 3151 sys_fchmod(struct fchmod_args *uap) 3152 { 3153 struct thread *td = curthread; 3154 struct proc *p = td->td_proc; 3155 struct file *fp; 3156 int error; 3157 3158 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3159 return (error); 3160 if (fp->f_nchandle.ncp) 3161 error = ncp_writechk(&fp->f_nchandle); 3162 if (error == 0) 3163 error = setfmode((struct vnode *)fp->f_data, uap->mode); 3164 fdrop(fp); 3165 return (error); 3166 } 3167 3168 /* 3169 * fchmodat_args(char *path, int mode) 3170 * 3171 * Change mode of a file pointed to by fd/path. 3172 */ 3173 int 3174 sys_fchmodat(struct fchmodat_args *uap) 3175 { 3176 struct nlookupdata nd; 3177 struct file *fp; 3178 int error; 3179 int flags; 3180 3181 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3182 return (EINVAL); 3183 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3184 3185 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3186 UIO_USERSPACE, flags); 3187 if (error == 0) 3188 error = kern_chmod(&nd, uap->mode); 3189 nlookup_done_at(&nd, fp); 3190 return (error); 3191 } 3192 3193 static int 3194 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid) 3195 { 3196 struct thread *td = curthread; 3197 int error; 3198 struct vattr vattr; 3199 uid_t o_uid; 3200 gid_t o_gid; 3201 uint64_t size; 3202 3203 /* 3204 * note: vget is required for any operation that might mod the vnode 3205 * so VINACTIVE is properly cleared. 3206 */ 3207 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3208 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 3209 return error; 3210 o_uid = vattr.va_uid; 3211 o_gid = vattr.va_gid; 3212 size = vattr.va_size; 3213 3214 VATTR_NULL(&vattr); 3215 vattr.va_uid = uid; 3216 vattr.va_gid = gid; 3217 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3218 vput(vp); 3219 } 3220 3221 if (error == 0) { 3222 if (uid == -1) 3223 uid = o_uid; 3224 if (gid == -1) 3225 gid = o_gid; 3226 VFS_ACCOUNT(mp, o_uid, o_gid, -size); 3227 VFS_ACCOUNT(mp, uid, gid, size); 3228 } 3229 3230 return error; 3231 } 3232 3233 int 3234 kern_chown(struct nlookupdata *nd, int uid, int gid) 3235 { 3236 struct vnode *vp; 3237 int error; 3238 3239 if ((error = nlookup(nd)) != 0) 3240 return (error); 3241 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3242 return (error); 3243 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3244 error = setfown(nd->nl_nch.mount, vp, uid, gid); 3245 vrele(vp); 3246 return (error); 3247 } 3248 3249 /* 3250 * chown(char *path, int uid, int gid) 3251 * 3252 * Set ownership given a path name. 3253 */ 3254 int 3255 sys_chown(struct chown_args *uap) 3256 { 3257 struct nlookupdata nd; 3258 int error; 3259 3260 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3261 if (error == 0) 3262 error = kern_chown(&nd, uap->uid, uap->gid); 3263 nlookup_done(&nd); 3264 return (error); 3265 } 3266 3267 /* 3268 * lchown_args(char *path, int uid, int gid) 3269 * 3270 * Set ownership given a path name, do not cross symlinks. 3271 */ 3272 int 3273 sys_lchown(struct lchown_args *uap) 3274 { 3275 struct nlookupdata nd; 3276 int error; 3277 3278 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3279 if (error == 0) 3280 error = kern_chown(&nd, uap->uid, uap->gid); 3281 nlookup_done(&nd); 3282 return (error); 3283 } 3284 3285 /* 3286 * fchown_args(int fd, int uid, int gid) 3287 * 3288 * Set ownership given a file descriptor. 3289 */ 3290 int 3291 sys_fchown(struct fchown_args *uap) 3292 { 3293 struct thread *td = curthread; 3294 struct proc *p = td->td_proc; 3295 struct file *fp; 3296 int error; 3297 3298 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3299 return (error); 3300 if (fp->f_nchandle.ncp) 3301 error = ncp_writechk(&fp->f_nchandle); 3302 if (error == 0) 3303 error = setfown(p->p_fd->fd_ncdir.mount, 3304 (struct vnode *)fp->f_data, uap->uid, uap->gid); 3305 fdrop(fp); 3306 return (error); 3307 } 3308 3309 /* 3310 * fchownat(int fd, char *path, int uid, int gid, int flags) 3311 * 3312 * Set ownership of file pointed to by fd/path. 3313 */ 3314 int 3315 sys_fchownat(struct fchownat_args *uap) 3316 { 3317 struct nlookupdata nd; 3318 struct file *fp; 3319 int error; 3320 int flags; 3321 3322 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3323 return (EINVAL); 3324 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3325 3326 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3327 UIO_USERSPACE, flags); 3328 if (error == 0) 3329 error = kern_chown(&nd, uap->uid, uap->gid); 3330 nlookup_done_at(&nd, fp); 3331 return (error); 3332 } 3333 3334 3335 static int 3336 getutimes(const struct timeval *tvp, struct timespec *tsp) 3337 { 3338 struct timeval tv[2]; 3339 3340 if (tvp == NULL) { 3341 microtime(&tv[0]); 3342 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]); 3343 tsp[1] = tsp[0]; 3344 } else { 3345 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); 3346 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); 3347 } 3348 return 0; 3349 } 3350 3351 static int 3352 setutimes(struct vnode *vp, struct vattr *vattr, 3353 const struct timespec *ts, int nullflag) 3354 { 3355 struct thread *td = curthread; 3356 int error; 3357 3358 VATTR_NULL(vattr); 3359 vattr->va_atime = ts[0]; 3360 vattr->va_mtime = ts[1]; 3361 if (nullflag) 3362 vattr->va_vaflags |= VA_UTIMES_NULL; 3363 error = VOP_SETATTR(vp, vattr, td->td_ucred); 3364 3365 return error; 3366 } 3367 3368 int 3369 kern_utimes(struct nlookupdata *nd, struct timeval *tptr) 3370 { 3371 struct timespec ts[2]; 3372 struct vnode *vp; 3373 struct vattr vattr; 3374 int error; 3375 3376 if ((error = getutimes(tptr, ts)) != 0) 3377 return (error); 3378 3379 /* 3380 * NOTE: utimes() succeeds for the owner even if the file 3381 * is not user-writable. 3382 */ 3383 nd->nl_flags |= NLC_OWN | NLC_WRITE; 3384 3385 if ((error = nlookup(nd)) != 0) 3386 return (error); 3387 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3388 return (error); 3389 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3390 return (error); 3391 3392 /* 3393 * note: vget is required for any operation that might mod the vnode 3394 * so VINACTIVE is properly cleared. 3395 */ 3396 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) { 3397 error = vget(vp, LK_EXCLUSIVE); 3398 if (error == 0) { 3399 error = setutimes(vp, &vattr, ts, (tptr == NULL)); 3400 vput(vp); 3401 } 3402 } 3403 vrele(vp); 3404 return (error); 3405 } 3406 3407 /* 3408 * utimes_args(char *path, struct timeval *tptr) 3409 * 3410 * Set the access and modification times of a file. 3411 */ 3412 int 3413 sys_utimes(struct utimes_args *uap) 3414 { 3415 struct timeval tv[2]; 3416 struct nlookupdata nd; 3417 int error; 3418 3419 if (uap->tptr) { 3420 error = copyin(uap->tptr, tv, sizeof(tv)); 3421 if (error) 3422 return (error); 3423 } 3424 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3425 if (error == 0) 3426 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3427 nlookup_done(&nd); 3428 return (error); 3429 } 3430 3431 /* 3432 * lutimes_args(char *path, struct timeval *tptr) 3433 * 3434 * Set the access and modification times of a file. 3435 */ 3436 int 3437 sys_lutimes(struct lutimes_args *uap) 3438 { 3439 struct timeval tv[2]; 3440 struct nlookupdata nd; 3441 int error; 3442 3443 if (uap->tptr) { 3444 error = copyin(uap->tptr, tv, sizeof(tv)); 3445 if (error) 3446 return (error); 3447 } 3448 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3449 if (error == 0) 3450 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3451 nlookup_done(&nd); 3452 return (error); 3453 } 3454 3455 /* 3456 * Set utimes on a file descriptor. The creds used to open the 3457 * file are used to determine whether the operation is allowed 3458 * or not. 3459 */ 3460 int 3461 kern_futimes(int fd, struct timeval *tptr) 3462 { 3463 struct thread *td = curthread; 3464 struct proc *p = td->td_proc; 3465 struct timespec ts[2]; 3466 struct file *fp; 3467 struct vnode *vp; 3468 struct vattr vattr; 3469 int error; 3470 3471 error = getutimes(tptr, ts); 3472 if (error) 3473 return (error); 3474 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 3475 return (error); 3476 if (fp->f_nchandle.ncp) 3477 error = ncp_writechk(&fp->f_nchandle); 3478 if (error == 0) { 3479 vp = fp->f_data; 3480 error = vget(vp, LK_EXCLUSIVE); 3481 if (error == 0) { 3482 error = VOP_GETATTR(vp, &vattr); 3483 if (error == 0) { 3484 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE, 3485 fp->f_cred); 3486 } 3487 if (error == 0) { 3488 error = setutimes(vp, &vattr, ts, 3489 (tptr == NULL)); 3490 } 3491 vput(vp); 3492 } 3493 } 3494 fdrop(fp); 3495 return (error); 3496 } 3497 3498 /* 3499 * futimes_args(int fd, struct timeval *tptr) 3500 * 3501 * Set the access and modification times of a file. 3502 */ 3503 int 3504 sys_futimes(struct futimes_args *uap) 3505 { 3506 struct timeval tv[2]; 3507 int error; 3508 3509 if (uap->tptr) { 3510 error = copyin(uap->tptr, tv, sizeof(tv)); 3511 if (error) 3512 return (error); 3513 } 3514 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL); 3515 3516 return (error); 3517 } 3518 3519 int 3520 kern_truncate(struct nlookupdata *nd, off_t length) 3521 { 3522 struct vnode *vp; 3523 struct vattr vattr; 3524 int error; 3525 uid_t uid = 0; 3526 gid_t gid = 0; 3527 uint64_t old_size = 0; 3528 3529 if (length < 0) 3530 return(EINVAL); 3531 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE; 3532 if ((error = nlookup(nd)) != 0) 3533 return (error); 3534 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3535 return (error); 3536 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3537 return (error); 3538 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) { 3539 vrele(vp); 3540 return (error); 3541 } 3542 if (vp->v_type == VDIR) { 3543 error = EISDIR; 3544 goto done; 3545 } 3546 if (vfs_quota_enabled) { 3547 error = VOP_GETATTR(vp, &vattr); 3548 KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0")); 3549 uid = vattr.va_uid; 3550 gid = vattr.va_gid; 3551 old_size = vattr.va_size; 3552 } 3553 3554 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) { 3555 VATTR_NULL(&vattr); 3556 vattr.va_size = length; 3557 error = VOP_SETATTR(vp, &vattr, nd->nl_cred); 3558 VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size); 3559 } 3560 done: 3561 vput(vp); 3562 return (error); 3563 } 3564 3565 /* 3566 * truncate(char *path, int pad, off_t length) 3567 * 3568 * Truncate a file given its path name. 3569 */ 3570 int 3571 sys_truncate(struct truncate_args *uap) 3572 { 3573 struct nlookupdata nd; 3574 int error; 3575 3576 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3577 if (error == 0) 3578 error = kern_truncate(&nd, uap->length); 3579 nlookup_done(&nd); 3580 return error; 3581 } 3582 3583 int 3584 kern_ftruncate(int fd, off_t length) 3585 { 3586 struct thread *td = curthread; 3587 struct proc *p = td->td_proc; 3588 struct vattr vattr; 3589 struct vnode *vp; 3590 struct file *fp; 3591 int error; 3592 uid_t uid = 0; 3593 gid_t gid = 0; 3594 uint64_t old_size = 0; 3595 struct mount *mp; 3596 3597 if (length < 0) 3598 return(EINVAL); 3599 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 3600 return (error); 3601 if (fp->f_nchandle.ncp) { 3602 error = ncp_writechk(&fp->f_nchandle); 3603 if (error) 3604 goto done; 3605 } 3606 if ((fp->f_flag & FWRITE) == 0) { 3607 error = EINVAL; 3608 goto done; 3609 } 3610 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */ 3611 error = EINVAL; 3612 goto done; 3613 } 3614 vp = (struct vnode *)fp->f_data; 3615 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3616 if (vp->v_type == VDIR) { 3617 error = EISDIR; 3618 goto done; 3619 } 3620 3621 if (vfs_quota_enabled) { 3622 error = VOP_GETATTR(vp, &vattr); 3623 KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0")); 3624 uid = vattr.va_uid; 3625 gid = vattr.va_gid; 3626 old_size = vattr.va_size; 3627 } 3628 3629 if ((error = vn_writechk(vp, NULL)) == 0) { 3630 VATTR_NULL(&vattr); 3631 vattr.va_size = length; 3632 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 3633 mp = vq_vptomp(vp); 3634 VFS_ACCOUNT(mp, uid, gid, length - old_size); 3635 } 3636 vn_unlock(vp); 3637 done: 3638 fdrop(fp); 3639 return (error); 3640 } 3641 3642 /* 3643 * ftruncate_args(int fd, int pad, off_t length) 3644 * 3645 * Truncate a file given a file descriptor. 3646 */ 3647 int 3648 sys_ftruncate(struct ftruncate_args *uap) 3649 { 3650 int error; 3651 3652 error = kern_ftruncate(uap->fd, uap->length); 3653 3654 return (error); 3655 } 3656 3657 /* 3658 * fsync(int fd) 3659 * 3660 * Sync an open file. 3661 */ 3662 int 3663 sys_fsync(struct fsync_args *uap) 3664 { 3665 struct thread *td = curthread; 3666 struct proc *p = td->td_proc; 3667 struct vnode *vp; 3668 struct file *fp; 3669 vm_object_t obj; 3670 int error; 3671 3672 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3673 return (error); 3674 vp = (struct vnode *)fp->f_data; 3675 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3676 if ((obj = vp->v_object) != NULL) { 3677 if (vp->v_mount == NULL || 3678 (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) { 3679 vm_object_page_clean(obj, 0, 0, 0); 3680 } 3681 } 3682 error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL); 3683 if (error == 0 && vp->v_mount) 3684 error = buf_fsync(vp); 3685 vn_unlock(vp); 3686 fdrop(fp); 3687 3688 return (error); 3689 } 3690 3691 int 3692 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond) 3693 { 3694 struct nchandle fnchd; 3695 struct nchandle tnchd; 3696 struct namecache *ncp; 3697 struct vnode *fdvp; 3698 struct vnode *tdvp; 3699 struct mount *mp; 3700 int error; 3701 3702 bwillinode(1); 3703 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC; 3704 if ((error = nlookup(fromnd)) != 0) 3705 return (error); 3706 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL) 3707 return (ENOENT); 3708 fnchd.mount = fromnd->nl_nch.mount; 3709 cache_hold(&fnchd); 3710 3711 /* 3712 * unlock the source nch so we can lookup the target nch without 3713 * deadlocking. The target may or may not exist so we do not check 3714 * for a target vp like kern_mkdir() and other creation functions do. 3715 * 3716 * The source and target directories are ref'd and rechecked after 3717 * everything is relocked to determine if the source or target file 3718 * has been renamed. 3719 */ 3720 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED); 3721 fromnd->nl_flags &= ~NLC_NCPISLOCKED; 3722 cache_unlock(&fromnd->nl_nch); 3723 3724 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP; 3725 if ((error = nlookup(tond)) != 0) { 3726 cache_drop(&fnchd); 3727 return (error); 3728 } 3729 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) { 3730 cache_drop(&fnchd); 3731 return (ENOENT); 3732 } 3733 tnchd.mount = tond->nl_nch.mount; 3734 cache_hold(&tnchd); 3735 3736 /* 3737 * If the source and target are the same there is nothing to do 3738 */ 3739 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) { 3740 cache_drop(&fnchd); 3741 cache_drop(&tnchd); 3742 return (0); 3743 } 3744 3745 /* 3746 * Mount points cannot be renamed or overwritten 3747 */ 3748 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) & 3749 NCF_ISMOUNTPT 3750 ) { 3751 cache_drop(&fnchd); 3752 cache_drop(&tnchd); 3753 return (EINVAL); 3754 } 3755 3756 /* 3757 * Relock the source ncp. cache_relock() will deal with any 3758 * deadlocks against the already-locked tond and will also 3759 * make sure both are resolved. 3760 * 3761 * NOTE AFTER RELOCKING: The source or target ncp may have become 3762 * invalid while they were unlocked, nc_vp and nc_mount could 3763 * be NULL. 3764 */ 3765 cache_relock(&fromnd->nl_nch, fromnd->nl_cred, 3766 &tond->nl_nch, tond->nl_cred); 3767 fromnd->nl_flags |= NLC_NCPISLOCKED; 3768 3769 /* 3770 * If either fromnd or tond are marked destroyed a ripout occured 3771 * out from under us and we must retry. 3772 */ 3773 if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) || 3774 fromnd->nl_nch.ncp->nc_vp == NULL || 3775 (tond->nl_nch.ncp->nc_flag & NCF_DESTROYED)) { 3776 kprintf("kern_rename: retry due to ripout on: " 3777 "\"%s\" -> \"%s\"\n", 3778 fromnd->nl_nch.ncp->nc_name, 3779 tond->nl_nch.ncp->nc_name); 3780 cache_drop(&fnchd); 3781 cache_drop(&tnchd); 3782 return (EAGAIN); 3783 } 3784 3785 /* 3786 * make sure the parent directories linkages are the same 3787 */ 3788 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent || 3789 tnchd.ncp != tond->nl_nch.ncp->nc_parent) { 3790 cache_drop(&fnchd); 3791 cache_drop(&tnchd); 3792 return (ENOENT); 3793 } 3794 3795 /* 3796 * Both the source and target must be within the same filesystem and 3797 * in the same filesystem as their parent directories within the 3798 * namecache topology. 3799 * 3800 * NOTE: fromnd's nc_mount or nc_vp could be NULL. 3801 */ 3802 mp = fnchd.mount; 3803 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount || 3804 mp != tond->nl_nch.mount) { 3805 cache_drop(&fnchd); 3806 cache_drop(&tnchd); 3807 return (EXDEV); 3808 } 3809 3810 /* 3811 * Make sure the mount point is writable 3812 */ 3813 if ((error = ncp_writechk(&tond->nl_nch)) != 0) { 3814 cache_drop(&fnchd); 3815 cache_drop(&tnchd); 3816 return (error); 3817 } 3818 3819 /* 3820 * If the target exists and either the source or target is a directory, 3821 * then both must be directories. 3822 * 3823 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h 3824 * have become NULL. 3825 */ 3826 if (tond->nl_nch.ncp->nc_vp) { 3827 if (fromnd->nl_nch.ncp->nc_vp == NULL) { 3828 error = ENOENT; 3829 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 3830 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR) 3831 error = ENOTDIR; 3832 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) { 3833 error = EISDIR; 3834 } 3835 } 3836 3837 /* 3838 * You cannot rename a source into itself or a subdirectory of itself. 3839 * We check this by travsersing the target directory upwards looking 3840 * for a match against the source. 3841 * 3842 * XXX MPSAFE 3843 */ 3844 if (error == 0) { 3845 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) { 3846 if (fromnd->nl_nch.ncp == ncp) { 3847 error = EINVAL; 3848 break; 3849 } 3850 } 3851 } 3852 3853 cache_drop(&fnchd); 3854 cache_drop(&tnchd); 3855 3856 /* 3857 * Even though the namespaces are different, they may still represent 3858 * hardlinks to the same file. The filesystem might have a hard time 3859 * with this so we issue a NREMOVE of the source instead of a NRENAME 3860 * when we detect the situation. 3861 */ 3862 if (error == 0) { 3863 fdvp = fromnd->nl_dvp; 3864 tdvp = tond->nl_dvp; 3865 if (fdvp == NULL || tdvp == NULL) { 3866 error = EPERM; 3867 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) { 3868 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp, 3869 fromnd->nl_cred); 3870 } else { 3871 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch, 3872 fdvp, tdvp, tond->nl_cred); 3873 } 3874 } 3875 return (error); 3876 } 3877 3878 /* 3879 * rename_args(char *from, char *to) 3880 * 3881 * Rename files. Source and destination must either both be directories, 3882 * or both not be directories. If target is a directory, it must be empty. 3883 */ 3884 int 3885 sys_rename(struct rename_args *uap) 3886 { 3887 struct nlookupdata fromnd, tond; 3888 int error; 3889 3890 do { 3891 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0); 3892 if (error == 0) { 3893 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0); 3894 if (error == 0) 3895 error = kern_rename(&fromnd, &tond); 3896 nlookup_done(&tond); 3897 } 3898 nlookup_done(&fromnd); 3899 } while (error == EAGAIN); 3900 return (error); 3901 } 3902 3903 /* 3904 * renameat_args(int oldfd, char *old, int newfd, char *new) 3905 * 3906 * Rename files using paths relative to the directories associated with 3907 * oldfd and newfd. Source and destination must either both be directories, 3908 * or both not be directories. If target is a directory, it must be empty. 3909 */ 3910 int 3911 sys_renameat(struct renameat_args *uap) 3912 { 3913 struct nlookupdata oldnd, newnd; 3914 struct file *oldfp, *newfp; 3915 int error; 3916 3917 do { 3918 error = nlookup_init_at(&oldnd, &oldfp, 3919 uap->oldfd, uap->old, 3920 UIO_USERSPACE, 0); 3921 if (error == 0) { 3922 error = nlookup_init_at(&newnd, &newfp, 3923 uap->newfd, uap->new, 3924 UIO_USERSPACE, 0); 3925 if (error == 0) 3926 error = kern_rename(&oldnd, &newnd); 3927 nlookup_done_at(&newnd, newfp); 3928 } 3929 nlookup_done_at(&oldnd, oldfp); 3930 } while (error == EAGAIN); 3931 return (error); 3932 } 3933 3934 int 3935 kern_mkdir(struct nlookupdata *nd, int mode) 3936 { 3937 struct thread *td = curthread; 3938 struct proc *p = td->td_proc; 3939 struct vnode *vp; 3940 struct vattr vattr; 3941 int error; 3942 3943 bwillinode(1); 3944 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP; 3945 if ((error = nlookup(nd)) != 0) 3946 return (error); 3947 3948 if (nd->nl_nch.ncp->nc_vp) 3949 return (EEXIST); 3950 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3951 return (error); 3952 VATTR_NULL(&vattr); 3953 vattr.va_type = VDIR; 3954 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask; 3955 3956 vp = NULL; 3957 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr); 3958 if (error == 0) 3959 vput(vp); 3960 return (error); 3961 } 3962 3963 /* 3964 * mkdir_args(char *path, int mode) 3965 * 3966 * Make a directory file. 3967 */ 3968 int 3969 sys_mkdir(struct mkdir_args *uap) 3970 { 3971 struct nlookupdata nd; 3972 int error; 3973 3974 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3975 if (error == 0) 3976 error = kern_mkdir(&nd, uap->mode); 3977 nlookup_done(&nd); 3978 return (error); 3979 } 3980 3981 /* 3982 * mkdirat_args(int fd, char *path, mode_t mode) 3983 * 3984 * Make a directory file. The path is relative to the directory associated 3985 * with fd. 3986 */ 3987 int 3988 sys_mkdirat(struct mkdirat_args *uap) 3989 { 3990 struct nlookupdata nd; 3991 struct file *fp; 3992 int error; 3993 3994 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 3995 if (error == 0) 3996 error = kern_mkdir(&nd, uap->mode); 3997 nlookup_done_at(&nd, fp); 3998 return (error); 3999 } 4000 4001 int 4002 kern_rmdir(struct nlookupdata *nd) 4003 { 4004 int error; 4005 4006 bwillinode(1); 4007 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 4008 if ((error = nlookup(nd)) != 0) 4009 return (error); 4010 4011 /* 4012 * Do not allow directories representing mount points to be 4013 * deleted, even if empty. Check write perms on mount point 4014 * in case the vnode is aliased (aka nullfs). 4015 */ 4016 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT)) 4017 return (EBUSY); 4018 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 4019 return (error); 4020 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 4021 return (error); 4022 } 4023 4024 /* 4025 * rmdir_args(char *path) 4026 * 4027 * Remove a directory file. 4028 */ 4029 int 4030 sys_rmdir(struct rmdir_args *uap) 4031 { 4032 struct nlookupdata nd; 4033 int error; 4034 4035 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 4036 if (error == 0) 4037 error = kern_rmdir(&nd); 4038 nlookup_done(&nd); 4039 return (error); 4040 } 4041 4042 int 4043 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res, 4044 enum uio_seg direction) 4045 { 4046 struct thread *td = curthread; 4047 struct proc *p = td->td_proc; 4048 struct vnode *vp; 4049 struct file *fp; 4050 struct uio auio; 4051 struct iovec aiov; 4052 off_t loff; 4053 int error, eofflag; 4054 4055 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 4056 return (error); 4057 if ((fp->f_flag & FREAD) == 0) { 4058 error = EBADF; 4059 goto done; 4060 } 4061 vp = (struct vnode *)fp->f_data; 4062 unionread: 4063 if (vp->v_type != VDIR) { 4064 error = EINVAL; 4065 goto done; 4066 } 4067 aiov.iov_base = buf; 4068 aiov.iov_len = count; 4069 auio.uio_iov = &aiov; 4070 auio.uio_iovcnt = 1; 4071 auio.uio_rw = UIO_READ; 4072 auio.uio_segflg = direction; 4073 auio.uio_td = td; 4074 auio.uio_resid = count; 4075 loff = auio.uio_offset = fp->f_offset; 4076 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); 4077 fp->f_offset = auio.uio_offset; 4078 if (error) 4079 goto done; 4080 if (count == auio.uio_resid) { 4081 if (union_dircheckp) { 4082 error = union_dircheckp(td, &vp, fp); 4083 if (error == -1) 4084 goto unionread; 4085 if (error) 4086 goto done; 4087 } 4088 #if 0 4089 if ((vp->v_flag & VROOT) && 4090 (vp->v_mount->mnt_flag & MNT_UNION)) { 4091 struct vnode *tvp = vp; 4092 vp = vp->v_mount->mnt_vnodecovered; 4093 vref(vp); 4094 fp->f_data = vp; 4095 fp->f_offset = 0; 4096 vrele(tvp); 4097 goto unionread; 4098 } 4099 #endif 4100 } 4101 4102 /* 4103 * WARNING! *basep may not be wide enough to accomodate the 4104 * seek offset. XXX should we hack this to return the upper 32 bits 4105 * for offsets greater then 4G? 4106 */ 4107 if (basep) { 4108 *basep = (long)loff; 4109 } 4110 *res = count - auio.uio_resid; 4111 done: 4112 fdrop(fp); 4113 return (error); 4114 } 4115 4116 /* 4117 * getdirentries_args(int fd, char *buf, u_int conut, long *basep) 4118 * 4119 * Read a block of directory entries in a file system independent format. 4120 */ 4121 int 4122 sys_getdirentries(struct getdirentries_args *uap) 4123 { 4124 long base; 4125 int error; 4126 4127 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base, 4128 &uap->sysmsg_result, UIO_USERSPACE); 4129 4130 if (error == 0 && uap->basep) 4131 error = copyout(&base, uap->basep, sizeof(*uap->basep)); 4132 return (error); 4133 } 4134 4135 /* 4136 * getdents_args(int fd, char *buf, size_t count) 4137 */ 4138 int 4139 sys_getdents(struct getdents_args *uap) 4140 { 4141 int error; 4142 4143 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL, 4144 &uap->sysmsg_result, UIO_USERSPACE); 4145 4146 return (error); 4147 } 4148 4149 /* 4150 * Set the mode mask for creation of filesystem nodes. 4151 * 4152 * umask(int newmask) 4153 */ 4154 int 4155 sys_umask(struct umask_args *uap) 4156 { 4157 struct thread *td = curthread; 4158 struct proc *p = td->td_proc; 4159 struct filedesc *fdp; 4160 4161 fdp = p->p_fd; 4162 uap->sysmsg_result = fdp->fd_cmask; 4163 fdp->fd_cmask = uap->newmask & ALLPERMS; 4164 return (0); 4165 } 4166 4167 /* 4168 * revoke(char *path) 4169 * 4170 * Void all references to file by ripping underlying filesystem 4171 * away from vnode. 4172 */ 4173 int 4174 sys_revoke(struct revoke_args *uap) 4175 { 4176 struct nlookupdata nd; 4177 struct vattr vattr; 4178 struct vnode *vp; 4179 struct ucred *cred; 4180 int error; 4181 4182 vp = NULL; 4183 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4184 if (error == 0) 4185 error = nlookup(&nd); 4186 if (error == 0) 4187 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 4188 cred = crhold(nd.nl_cred); 4189 nlookup_done(&nd); 4190 if (error == 0) { 4191 if (error == 0) 4192 error = VOP_GETATTR(vp, &vattr); 4193 if (error == 0 && cred->cr_uid != vattr.va_uid) 4194 error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0); 4195 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) { 4196 if (vcount(vp) > 0) 4197 error = vrevoke(vp, cred); 4198 } else if (error == 0) { 4199 error = vrevoke(vp, cred); 4200 } 4201 vrele(vp); 4202 } 4203 if (cred) 4204 crfree(cred); 4205 return (error); 4206 } 4207 4208 /* 4209 * getfh_args(char *fname, fhandle_t *fhp) 4210 * 4211 * Get (NFS) file handle 4212 * 4213 * NOTE: We use the fsid of the covering mount, even if it is a nullfs 4214 * mount. This allows nullfs mounts to be explicitly exported. 4215 * 4216 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe. 4217 * 4218 * nullfs mounts of subdirectories are not safe. That is, it will 4219 * work, but you do not really have protection against access to 4220 * the related parent directories. 4221 */ 4222 int 4223 sys_getfh(struct getfh_args *uap) 4224 { 4225 struct thread *td = curthread; 4226 struct nlookupdata nd; 4227 fhandle_t fh; 4228 struct vnode *vp; 4229 struct mount *mp; 4230 int error; 4231 4232 /* 4233 * Must be super user 4234 */ 4235 if ((error = priv_check(td, PRIV_ROOT)) != 0) 4236 return (error); 4237 4238 vp = NULL; 4239 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 4240 if (error == 0) 4241 error = nlookup(&nd); 4242 if (error == 0) 4243 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4244 mp = nd.nl_nch.mount; 4245 nlookup_done(&nd); 4246 if (error == 0) { 4247 bzero(&fh, sizeof(fh)); 4248 fh.fh_fsid = mp->mnt_stat.f_fsid; 4249 error = VFS_VPTOFH(vp, &fh.fh_fid); 4250 vput(vp); 4251 if (error == 0) 4252 error = copyout(&fh, uap->fhp, sizeof(fh)); 4253 } 4254 return (error); 4255 } 4256 4257 /* 4258 * fhopen_args(const struct fhandle *u_fhp, int flags) 4259 * 4260 * syscall for the rpc.lockd to use to translate a NFS file handle into 4261 * an open descriptor. 4262 * 4263 * warning: do not remove the priv_check() call or this becomes one giant 4264 * security hole. 4265 */ 4266 int 4267 sys_fhopen(struct fhopen_args *uap) 4268 { 4269 struct thread *td = curthread; 4270 struct filedesc *fdp = td->td_proc->p_fd; 4271 struct mount *mp; 4272 struct vnode *vp; 4273 struct fhandle fhp; 4274 struct vattr vat; 4275 struct vattr *vap = &vat; 4276 struct flock lf; 4277 int fmode, mode, error = 0, type; 4278 struct file *nfp; 4279 struct file *fp; 4280 int indx; 4281 4282 /* 4283 * Must be super user 4284 */ 4285 error = priv_check(td, PRIV_ROOT); 4286 if (error) 4287 return (error); 4288 4289 fmode = FFLAGS(uap->flags); 4290 4291 /* 4292 * Why not allow a non-read/write open for our lockd? 4293 */ 4294 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) 4295 return (EINVAL); 4296 error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); 4297 if (error) 4298 return(error); 4299 4300 /* 4301 * Find the mount point 4302 */ 4303 mp = vfs_getvfs(&fhp.fh_fsid); 4304 if (mp == NULL) { 4305 error = ESTALE; 4306 goto done; 4307 } 4308 /* now give me my vnode, it gets returned to me locked */ 4309 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp); 4310 if (error) 4311 goto done; 4312 /* 4313 * from now on we have to make sure not 4314 * to forget about the vnode 4315 * any error that causes an abort must vput(vp) 4316 * just set error = err and 'goto bad;'. 4317 */ 4318 4319 /* 4320 * from vn_open 4321 */ 4322 if (vp->v_type == VLNK) { 4323 error = EMLINK; 4324 goto bad; 4325 } 4326 if (vp->v_type == VSOCK) { 4327 error = EOPNOTSUPP; 4328 goto bad; 4329 } 4330 mode = 0; 4331 if (fmode & (FWRITE | O_TRUNC)) { 4332 if (vp->v_type == VDIR) { 4333 error = EISDIR; 4334 goto bad; 4335 } 4336 error = vn_writechk(vp, NULL); 4337 if (error) 4338 goto bad; 4339 mode |= VWRITE; 4340 } 4341 if (fmode & FREAD) 4342 mode |= VREAD; 4343 if (mode) { 4344 error = VOP_ACCESS(vp, mode, td->td_ucred); 4345 if (error) 4346 goto bad; 4347 } 4348 if (fmode & O_TRUNC) { 4349 vn_unlock(vp); /* XXX */ 4350 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 4351 VATTR_NULL(vap); 4352 vap->va_size = 0; 4353 error = VOP_SETATTR(vp, vap, td->td_ucred); 4354 if (error) 4355 goto bad; 4356 } 4357 4358 /* 4359 * VOP_OPEN needs the file pointer so it can potentially override 4360 * it. 4361 * 4362 * WARNING! no f_nchandle will be associated when fhopen()ing a 4363 * directory. XXX 4364 */ 4365 if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0) 4366 goto bad; 4367 fp = nfp; 4368 4369 error = VOP_OPEN(vp, fmode, td->td_ucred, fp); 4370 if (error) { 4371 /* 4372 * setting f_ops this way prevents VOP_CLOSE from being 4373 * called or fdrop() releasing the vp from v_data. Since 4374 * the VOP_OPEN failed we don't want to VOP_CLOSE. 4375 */ 4376 fp->f_ops = &badfileops; 4377 fp->f_data = NULL; 4378 goto bad_drop; 4379 } 4380 4381 /* 4382 * The fp is given its own reference, we still have our ref and lock. 4383 * 4384 * Assert that all regular files must be created with a VM object. 4385 */ 4386 if (vp->v_type == VREG && vp->v_object == NULL) { 4387 kprintf("fhopen: regular file did not have VM object: %p\n", vp); 4388 goto bad_drop; 4389 } 4390 4391 /* 4392 * The open was successful. Handle any locking requirements. 4393 */ 4394 if (fmode & (O_EXLOCK | O_SHLOCK)) { 4395 lf.l_whence = SEEK_SET; 4396 lf.l_start = 0; 4397 lf.l_len = 0; 4398 if (fmode & O_EXLOCK) 4399 lf.l_type = F_WRLCK; 4400 else 4401 lf.l_type = F_RDLCK; 4402 if (fmode & FNONBLOCK) 4403 type = 0; 4404 else 4405 type = F_WAIT; 4406 vn_unlock(vp); 4407 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) { 4408 /* 4409 * release our private reference. 4410 */ 4411 fsetfd(fdp, NULL, indx); 4412 fdrop(fp); 4413 vrele(vp); 4414 goto done; 4415 } 4416 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4417 fp->f_flag |= FHASLOCK; 4418 } 4419 4420 /* 4421 * Clean up. Associate the file pointer with the previously 4422 * reserved descriptor and return it. 4423 */ 4424 vput(vp); 4425 fsetfd(fdp, fp, indx); 4426 fdrop(fp); 4427 uap->sysmsg_result = indx; 4428 if (uap->flags & O_CLOEXEC) 4429 error = fsetfdflags(fdp, indx, UF_EXCLOSE); 4430 return (error); 4431 4432 bad_drop: 4433 fsetfd(fdp, NULL, indx); 4434 fdrop(fp); 4435 bad: 4436 vput(vp); 4437 done: 4438 return (error); 4439 } 4440 4441 /* 4442 * fhstat_args(struct fhandle *u_fhp, struct stat *sb) 4443 */ 4444 int 4445 sys_fhstat(struct fhstat_args *uap) 4446 { 4447 struct thread *td = curthread; 4448 struct stat sb; 4449 fhandle_t fh; 4450 struct mount *mp; 4451 struct vnode *vp; 4452 int error; 4453 4454 /* 4455 * Must be super user 4456 */ 4457 error = priv_check(td, PRIV_ROOT); 4458 if (error) 4459 return (error); 4460 4461 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); 4462 if (error) 4463 return (error); 4464 4465 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 4466 error = ESTALE; 4467 if (error == 0) { 4468 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) { 4469 error = vn_stat(vp, &sb, td->td_ucred); 4470 vput(vp); 4471 } 4472 } 4473 if (error == 0) 4474 error = copyout(&sb, uap->sb, sizeof(sb)); 4475 return (error); 4476 } 4477 4478 /* 4479 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf) 4480 */ 4481 int 4482 sys_fhstatfs(struct fhstatfs_args *uap) 4483 { 4484 struct thread *td = curthread; 4485 struct proc *p = td->td_proc; 4486 struct statfs *sp; 4487 struct mount *mp; 4488 struct vnode *vp; 4489 struct statfs sb; 4490 char *fullpath, *freepath; 4491 fhandle_t fh; 4492 int error; 4493 4494 /* 4495 * Must be super user 4496 */ 4497 if ((error = priv_check(td, PRIV_ROOT))) 4498 return (error); 4499 4500 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4501 return (error); 4502 4503 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4504 error = ESTALE; 4505 goto done; 4506 } 4507 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4508 error = ESTALE; 4509 goto done; 4510 } 4511 4512 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0) 4513 goto done; 4514 mp = vp->v_mount; 4515 sp = &mp->mnt_stat; 4516 vput(vp); 4517 if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0) 4518 goto done; 4519 4520 error = mount_path(p, mp, &fullpath, &freepath); 4521 if (error) 4522 goto done; 4523 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 4524 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 4525 kfree(freepath, M_TEMP); 4526 4527 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 4528 if (priv_check(td, PRIV_ROOT)) { 4529 bcopy(sp, &sb, sizeof(sb)); 4530 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; 4531 sp = &sb; 4532 } 4533 error = copyout(sp, uap->buf, sizeof(*sp)); 4534 done: 4535 return (error); 4536 } 4537 4538 /* 4539 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf) 4540 */ 4541 int 4542 sys_fhstatvfs(struct fhstatvfs_args *uap) 4543 { 4544 struct thread *td = curthread; 4545 struct proc *p = td->td_proc; 4546 struct statvfs *sp; 4547 struct mount *mp; 4548 struct vnode *vp; 4549 fhandle_t fh; 4550 int error; 4551 4552 /* 4553 * Must be super user 4554 */ 4555 if ((error = priv_check(td, PRIV_ROOT))) 4556 return (error); 4557 4558 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4559 return (error); 4560 4561 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4562 error = ESTALE; 4563 goto done; 4564 } 4565 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4566 error = ESTALE; 4567 goto done; 4568 } 4569 4570 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 4571 goto done; 4572 mp = vp->v_mount; 4573 sp = &mp->mnt_vstat; 4574 vput(vp); 4575 if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0) 4576 goto done; 4577 4578 sp->f_flag = 0; 4579 if (mp->mnt_flag & MNT_RDONLY) 4580 sp->f_flag |= ST_RDONLY; 4581 if (mp->mnt_flag & MNT_NOSUID) 4582 sp->f_flag |= ST_NOSUID; 4583 error = copyout(sp, uap->buf, sizeof(*sp)); 4584 done: 4585 return (error); 4586 } 4587 4588 4589 /* 4590 * Syscall to push extended attribute configuration information into the 4591 * VFS. Accepts a path, which it converts to a mountpoint, as well as 4592 * a command (int cmd), and attribute name and misc data. For now, the 4593 * attribute name is left in userspace for consumption by the VFS_op. 4594 * It will probably be changed to be copied into sysspace by the 4595 * syscall in the future, once issues with various consumers of the 4596 * attribute code have raised their hands. 4597 * 4598 * Currently this is used only by UFS Extended Attributes. 4599 */ 4600 int 4601 sys_extattrctl(struct extattrctl_args *uap) 4602 { 4603 struct nlookupdata nd; 4604 struct vnode *vp; 4605 char attrname[EXTATTR_MAXNAMELEN]; 4606 int error; 4607 size_t size; 4608 4609 attrname[0] = 0; 4610 vp = NULL; 4611 error = 0; 4612 4613 if (error == 0 && uap->filename) { 4614 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE, 4615 NLC_FOLLOW); 4616 if (error == 0) 4617 error = nlookup(&nd); 4618 if (error == 0) 4619 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 4620 nlookup_done(&nd); 4621 } 4622 4623 if (error == 0 && uap->attrname) { 4624 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN, 4625 &size); 4626 } 4627 4628 if (error == 0) { 4629 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4630 if (error == 0) 4631 error = nlookup(&nd); 4632 if (error == 0) 4633 error = ncp_writechk(&nd.nl_nch); 4634 if (error == 0) { 4635 error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp, 4636 uap->attrnamespace, 4637 uap->attrname, nd.nl_cred); 4638 } 4639 nlookup_done(&nd); 4640 } 4641 4642 return (error); 4643 } 4644 4645 /* 4646 * Syscall to get a named extended attribute on a file or directory. 4647 */ 4648 int 4649 sys_extattr_set_file(struct extattr_set_file_args *uap) 4650 { 4651 char attrname[EXTATTR_MAXNAMELEN]; 4652 struct nlookupdata nd; 4653 struct vnode *vp; 4654 struct uio auio; 4655 struct iovec aiov; 4656 int error; 4657 4658 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4659 if (error) 4660 return (error); 4661 4662 vp = NULL; 4663 4664 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4665 if (error == 0) 4666 error = nlookup(&nd); 4667 if (error == 0) 4668 error = ncp_writechk(&nd.nl_nch); 4669 if (error == 0) 4670 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4671 if (error) { 4672 nlookup_done(&nd); 4673 return (error); 4674 } 4675 4676 bzero(&auio, sizeof(auio)); 4677 aiov.iov_base = uap->data; 4678 aiov.iov_len = uap->nbytes; 4679 auio.uio_iov = &aiov; 4680 auio.uio_iovcnt = 1; 4681 auio.uio_offset = 0; 4682 auio.uio_resid = uap->nbytes; 4683 auio.uio_rw = UIO_WRITE; 4684 auio.uio_td = curthread; 4685 4686 error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname, 4687 &auio, nd.nl_cred); 4688 4689 vput(vp); 4690 nlookup_done(&nd); 4691 return (error); 4692 } 4693 4694 /* 4695 * Syscall to get a named extended attribute on a file or directory. 4696 */ 4697 int 4698 sys_extattr_get_file(struct extattr_get_file_args *uap) 4699 { 4700 char attrname[EXTATTR_MAXNAMELEN]; 4701 struct nlookupdata nd; 4702 struct uio auio; 4703 struct iovec aiov; 4704 struct vnode *vp; 4705 int error; 4706 4707 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4708 if (error) 4709 return (error); 4710 4711 vp = NULL; 4712 4713 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4714 if (error == 0) 4715 error = nlookup(&nd); 4716 if (error == 0) 4717 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4718 if (error) { 4719 nlookup_done(&nd); 4720 return (error); 4721 } 4722 4723 bzero(&auio, sizeof(auio)); 4724 aiov.iov_base = uap->data; 4725 aiov.iov_len = uap->nbytes; 4726 auio.uio_iov = &aiov; 4727 auio.uio_iovcnt = 1; 4728 auio.uio_offset = 0; 4729 auio.uio_resid = uap->nbytes; 4730 auio.uio_rw = UIO_READ; 4731 auio.uio_td = curthread; 4732 4733 error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname, 4734 &auio, nd.nl_cred); 4735 uap->sysmsg_result = uap->nbytes - auio.uio_resid; 4736 4737 vput(vp); 4738 nlookup_done(&nd); 4739 return(error); 4740 } 4741 4742 /* 4743 * Syscall to delete a named extended attribute from a file or directory. 4744 * Accepts attribute name. The real work happens in VOP_SETEXTATTR(). 4745 */ 4746 int 4747 sys_extattr_delete_file(struct extattr_delete_file_args *uap) 4748 { 4749 char attrname[EXTATTR_MAXNAMELEN]; 4750 struct nlookupdata nd; 4751 struct vnode *vp; 4752 int error; 4753 4754 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4755 if (error) 4756 return(error); 4757 4758 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4759 if (error == 0) 4760 error = nlookup(&nd); 4761 if (error == 0) 4762 error = ncp_writechk(&nd.nl_nch); 4763 if (error == 0) { 4764 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4765 if (error == 0) { 4766 error = VOP_SETEXTATTR(vp, uap->attrnamespace, 4767 attrname, NULL, nd.nl_cred); 4768 vput(vp); 4769 } 4770 } 4771 nlookup_done(&nd); 4772 return(error); 4773 } 4774 4775 /* 4776 * Determine if the mount is visible to the process. 4777 */ 4778 static int 4779 chroot_visible_mnt(struct mount *mp, struct proc *p) 4780 { 4781 struct nchandle nch; 4782 4783 /* 4784 * Traverse from the mount point upwards. If we hit the process 4785 * root then the mount point is visible to the process. 4786 */ 4787 nch = mp->mnt_ncmountpt; 4788 while (nch.ncp) { 4789 if (nch.mount == p->p_fd->fd_nrdir.mount && 4790 nch.ncp == p->p_fd->fd_nrdir.ncp) { 4791 return(1); 4792 } 4793 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) { 4794 nch = nch.mount->mnt_ncmounton; 4795 } else { 4796 nch.ncp = nch.ncp->nc_parent; 4797 } 4798 } 4799 4800 /* 4801 * If the mount point is not visible to the process, but the 4802 * process root is in a subdirectory of the mount, return 4803 * TRUE anyway. 4804 */ 4805 if (p->p_fd->fd_nrdir.mount == mp) 4806 return(1); 4807 4808 return(0); 4809 } 4810 4811