1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 39 * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $ 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/buf.h> 45 #include <sys/conf.h> 46 #include <sys/sysent.h> 47 #include <sys/malloc.h> 48 #include <sys/mount.h> 49 #include <sys/mountctl.h> 50 #include <sys/sysproto.h> 51 #include <sys/filedesc.h> 52 #include <sys/kernel.h> 53 #include <sys/fcntl.h> 54 #include <sys/file.h> 55 #include <sys/linker.h> 56 #include <sys/stat.h> 57 #include <sys/unistd.h> 58 #include <sys/vnode.h> 59 #include <sys/proc.h> 60 #include <sys/priv.h> 61 #include <sys/jail.h> 62 #include <sys/namei.h> 63 #include <sys/nlookup.h> 64 #include <sys/dirent.h> 65 #include <sys/extattr.h> 66 #include <sys/spinlock.h> 67 #include <sys/kern_syscall.h> 68 #include <sys/objcache.h> 69 #include <sys/sysctl.h> 70 71 #include <sys/buf2.h> 72 #include <sys/file2.h> 73 #include <sys/spinlock2.h> 74 #include <sys/mplock2.h> 75 76 #include <vm/vm.h> 77 #include <vm/vm_object.h> 78 #include <vm/vm_page.h> 79 80 #include <machine/limits.h> 81 #include <machine/stdarg.h> 82 83 #include <vfs/union/union.h> 84 85 static void mount_warning(struct mount *mp, const char *ctl, ...) 86 __printflike(2, 3); 87 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb); 88 static int checkvp_chdir (struct vnode *vn, struct thread *td); 89 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch); 90 static int chroot_refuse_vdir_fds (struct filedesc *fdp); 91 static int chroot_visible_mnt(struct mount *mp, struct proc *p); 92 static int getutimes (const struct timeval *, struct timespec *); 93 static int setfown (struct vnode *, uid_t, gid_t); 94 static int setfmode (struct vnode *, int); 95 static int setfflags (struct vnode *, int); 96 static int setutimes (struct vnode *, struct vattr *, 97 const struct timespec *, int); 98 static int usermount = 0; /* if 1, non-root can mount fs. */ 99 100 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *); 101 102 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, ""); 103 104 /* 105 * Virtual File System System Calls 106 */ 107 108 /* 109 * Mount a file system. 110 * 111 * mount_args(char *type, char *path, int flags, caddr_t data) 112 * 113 * MPALMOSTSAFE 114 */ 115 int 116 sys_mount(struct mount_args *uap) 117 { 118 struct thread *td = curthread; 119 struct vnode *vp; 120 struct nchandle nch; 121 struct mount *mp, *nullmp; 122 struct vfsconf *vfsp; 123 int error, flag = 0, flag2 = 0; 124 int hasmount; 125 struct vattr va; 126 struct nlookupdata nd; 127 char fstypename[MFSNAMELEN]; 128 struct ucred *cred; 129 130 get_mplock(); 131 cred = td->td_ucred; 132 if (jailed(cred)) { 133 error = EPERM; 134 goto done; 135 } 136 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT))) 137 goto done; 138 139 /* 140 * Do not allow NFS export by non-root users. 141 */ 142 if (uap->flags & MNT_EXPORTED) { 143 error = priv_check(td, PRIV_ROOT); 144 if (error) 145 goto done; 146 } 147 /* 148 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users 149 */ 150 if (priv_check(td, PRIV_ROOT)) 151 uap->flags |= MNT_NOSUID | MNT_NODEV; 152 153 /* 154 * Lookup the requested path and extract the nch and vnode. 155 */ 156 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 157 if (error == 0) { 158 if ((error = nlookup(&nd)) == 0) { 159 if (nd.nl_nch.ncp->nc_vp == NULL) 160 error = ENOENT; 161 } 162 } 163 if (error) { 164 nlookup_done(&nd); 165 goto done; 166 } 167 168 /* 169 * If the target filesystem is resolved via a nullfs mount, then 170 * nd.nl_nch.mount will be pointing to the nullfs mount structure 171 * instead of the target file system. We need it in case we are 172 * doing an update. 173 */ 174 nullmp = nd.nl_nch.mount; 175 176 /* 177 * Extract the locked+refd ncp and cleanup the nd structure 178 */ 179 nch = nd.nl_nch; 180 cache_zero(&nd.nl_nch); 181 nlookup_done(&nd); 182 183 if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch)) 184 hasmount = 1; 185 else 186 hasmount = 0; 187 188 189 /* 190 * now we have the locked ref'd nch and unreferenced vnode. 191 */ 192 vp = nch.ncp->nc_vp; 193 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) { 194 cache_put(&nch); 195 goto done; 196 } 197 cache_unlock(&nch); 198 199 /* 200 * Extract the file system type. We need to know this early, to take 201 * appropriate actions if we are dealing with a nullfs. 202 */ 203 if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) { 204 cache_drop(&nch); 205 vput(vp); 206 goto done; 207 } 208 209 /* 210 * Now we have an unlocked ref'd nch and a locked ref'd vp 211 */ 212 if (uap->flags & MNT_UPDATE) { 213 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 214 cache_drop(&nch); 215 vput(vp); 216 error = EINVAL; 217 goto done; 218 } 219 220 if (strncmp(fstypename, "null", 5) == 0) { 221 KKASSERT(nullmp); 222 mp = nullmp; 223 } else { 224 mp = vp->v_mount; 225 } 226 227 flag = mp->mnt_flag; 228 flag2 = mp->mnt_kern_flag; 229 /* 230 * We only allow the filesystem to be reloaded if it 231 * is currently mounted read-only. 232 */ 233 if ((uap->flags & MNT_RELOAD) && 234 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 235 cache_drop(&nch); 236 vput(vp); 237 error = EOPNOTSUPP; /* Needs translation */ 238 goto done; 239 } 240 /* 241 * Only root, or the user that did the original mount is 242 * permitted to update it. 243 */ 244 if (mp->mnt_stat.f_owner != cred->cr_uid && 245 (error = priv_check(td, PRIV_ROOT))) { 246 cache_drop(&nch); 247 vput(vp); 248 goto done; 249 } 250 if (vfs_busy(mp, LK_NOWAIT)) { 251 cache_drop(&nch); 252 vput(vp); 253 error = EBUSY; 254 goto done; 255 } 256 if ((vp->v_flag & VMOUNT) != 0 || hasmount) { 257 cache_drop(&nch); 258 vfs_unbusy(mp); 259 vput(vp); 260 error = EBUSY; 261 goto done; 262 } 263 vsetflags(vp, VMOUNT); 264 mp->mnt_flag |= 265 uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE); 266 vn_unlock(vp); 267 goto update; 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 && (error = priv_check(td, PRIV_ROOT)))) { 275 cache_drop(&nch); 276 vput(vp); 277 goto done; 278 } 279 if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) { 280 cache_drop(&nch); 281 vput(vp); 282 goto done; 283 } 284 if (vp->v_type != VDIR) { 285 cache_drop(&nch); 286 vput(vp); 287 error = ENOTDIR; 288 goto done; 289 } 290 if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) { 291 cache_drop(&nch); 292 vput(vp); 293 error = EPERM; 294 goto done; 295 } 296 vfsp = vfsconf_find_by_name(fstypename); 297 if (vfsp == NULL) { 298 linker_file_t lf; 299 300 /* Only load modules for root (very important!) */ 301 if ((error = priv_check(td, PRIV_ROOT)) != 0) { 302 cache_drop(&nch); 303 vput(vp); 304 goto done; 305 } 306 error = linker_load_file(fstypename, &lf); 307 if (error || lf == NULL) { 308 cache_drop(&nch); 309 vput(vp); 310 if (lf == NULL) 311 error = ENODEV; 312 goto done; 313 } 314 lf->userrefs++; 315 /* lookup again, see if the VFS was loaded */ 316 vfsp = vfsconf_find_by_name(fstypename); 317 if (vfsp == NULL) { 318 lf->userrefs--; 319 linker_file_unload(lf); 320 cache_drop(&nch); 321 vput(vp); 322 error = ENODEV; 323 goto done; 324 } 325 } 326 if ((vp->v_flag & VMOUNT) != 0 || hasmount) { 327 cache_drop(&nch); 328 vput(vp); 329 error = EBUSY; 330 goto done; 331 } 332 vsetflags(vp, VMOUNT); 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 vn_unlock(vp); 348 update: 349 /* 350 * Set the mount level flags. 351 */ 352 if (uap->flags & MNT_RDONLY) 353 mp->mnt_flag |= MNT_RDONLY; 354 else if (mp->mnt_flag & MNT_RDONLY) 355 mp->mnt_kern_flag |= MNTK_WANTRDWR; 356 mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV | 357 MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME | 358 MNT_NOSYMFOLLOW | MNT_IGNORE | 359 MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR); 360 mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC | 361 MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE | 362 MNT_NOSYMFOLLOW | MNT_IGNORE | 363 MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR); 364 /* 365 * Mount the filesystem. 366 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 367 * get. 368 */ 369 error = VFS_MOUNT(mp, uap->path, uap->data, cred); 370 if (mp->mnt_flag & MNT_UPDATE) { 371 if (mp->mnt_kern_flag & MNTK_WANTRDWR) 372 mp->mnt_flag &= ~MNT_RDONLY; 373 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE); 374 mp->mnt_kern_flag &=~ MNTK_WANTRDWR; 375 if (error) { 376 mp->mnt_flag = flag; 377 mp->mnt_kern_flag = flag2; 378 } 379 vfs_unbusy(mp); 380 vclrflags(vp, VMOUNT); 381 vrele(vp); 382 cache_drop(&nch); 383 goto done; 384 } 385 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 386 /* 387 * Put the new filesystem on the mount list after root. The mount 388 * point gets its own mnt_ncmountpt (unless the VFS already set one 389 * up) which represents the root of the mount. The lookup code 390 * detects the mount point going forward and checks the root of 391 * the mount going backwards. 392 * 393 * It is not necessary to invalidate or purge the vnode underneath 394 * because elements under the mount will be given their own glue 395 * namecache record. 396 */ 397 if (!error) { 398 if (mp->mnt_ncmountpt.ncp == NULL) { 399 /* 400 * allocate, then unlock, but leave the ref intact 401 */ 402 cache_allocroot(&mp->mnt_ncmountpt, mp, NULL); 403 cache_unlock(&mp->mnt_ncmountpt); 404 } 405 mp->mnt_ncmounton = nch; /* inherits ref */ 406 nch.ncp->nc_flag |= NCF_ISMOUNTPT; 407 408 /* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */ 409 vclrflags(vp, VMOUNT); 410 mountlist_insert(mp, MNTINS_LAST); 411 vn_unlock(vp); 412 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt); 413 error = vfs_allocate_syncvnode(mp); 414 vfs_unbusy(mp); 415 error = VFS_START(mp, 0); 416 vrele(vp); 417 } else { 418 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops); 419 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops); 420 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops); 421 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops); 422 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops); 423 vclrflags(vp, VMOUNT); 424 mp->mnt_vfc->vfc_refcount--; 425 vfs_unbusy(mp); 426 kfree(mp, M_MOUNT); 427 cache_drop(&nch); 428 vput(vp); 429 } 430 done: 431 rel_mplock(); 432 return (error); 433 } 434 435 /* 436 * Scan all active processes to see if any of them have a current 437 * or root directory onto which the new filesystem has just been 438 * mounted. If so, replace them with the new mount point. 439 * 440 * The passed ncp is ref'd and locked (from the mount code) and 441 * must be associated with the vnode representing the root of the 442 * mount point. 443 */ 444 struct checkdirs_info { 445 struct nchandle old_nch; 446 struct nchandle new_nch; 447 struct vnode *old_vp; 448 struct vnode *new_vp; 449 }; 450 451 static int checkdirs_callback(struct proc *p, void *data); 452 453 static void 454 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch) 455 { 456 struct checkdirs_info info; 457 struct vnode *olddp; 458 struct vnode *newdp; 459 struct mount *mp; 460 461 /* 462 * If the old mount point's vnode has a usecount of 1, it is not 463 * being held as a descriptor anywhere. 464 */ 465 olddp = old_nch->ncp->nc_vp; 466 if (olddp == NULL || olddp->v_sysref.refcnt == 1) 467 return; 468 469 /* 470 * Force the root vnode of the new mount point to be resolved 471 * so we can update any matching processes. 472 */ 473 mp = new_nch->mount; 474 if (VFS_ROOT(mp, &newdp)) 475 panic("mount: lost mount"); 476 cache_setunresolved(new_nch); 477 cache_setvp(new_nch, newdp); 478 479 /* 480 * Special handling of the root node 481 */ 482 if (rootvnode == olddp) { 483 vref(newdp); 484 vfs_cache_setroot(newdp, cache_hold(new_nch)); 485 } 486 487 /* 488 * Pass newdp separately so the callback does not have to access 489 * it via new_nch->ncp->nc_vp. 490 */ 491 info.old_nch = *old_nch; 492 info.new_nch = *new_nch; 493 info.new_vp = newdp; 494 allproc_scan(checkdirs_callback, &info); 495 vput(newdp); 496 } 497 498 /* 499 * NOTE: callback is not MP safe because the scanned process's filedesc 500 * structure can be ripped out from under us, amoung other things. 501 */ 502 static int 503 checkdirs_callback(struct proc *p, void *data) 504 { 505 struct checkdirs_info *info = data; 506 struct filedesc *fdp; 507 struct nchandle ncdrop1; 508 struct nchandle ncdrop2; 509 struct vnode *vprele1; 510 struct vnode *vprele2; 511 512 if ((fdp = p->p_fd) != NULL) { 513 cache_zero(&ncdrop1); 514 cache_zero(&ncdrop2); 515 vprele1 = NULL; 516 vprele2 = NULL; 517 518 /* 519 * MPUNSAFE - XXX fdp can be pulled out from under a 520 * foreign process. 521 * 522 * A shared filedesc is ok, we don't have to copy it 523 * because we are making this change globally. 524 */ 525 spin_lock(&fdp->fd_spin); 526 if (fdp->fd_ncdir.mount == info->old_nch.mount && 527 fdp->fd_ncdir.ncp == info->old_nch.ncp) { 528 vprele1 = fdp->fd_cdir; 529 vref(info->new_vp); 530 fdp->fd_cdir = info->new_vp; 531 ncdrop1 = fdp->fd_ncdir; 532 cache_copy(&info->new_nch, &fdp->fd_ncdir); 533 } 534 if (fdp->fd_nrdir.mount == info->old_nch.mount && 535 fdp->fd_nrdir.ncp == info->old_nch.ncp) { 536 vprele2 = fdp->fd_rdir; 537 vref(info->new_vp); 538 fdp->fd_rdir = info->new_vp; 539 ncdrop2 = fdp->fd_nrdir; 540 cache_copy(&info->new_nch, &fdp->fd_nrdir); 541 } 542 spin_unlock(&fdp->fd_spin); 543 if (ncdrop1.ncp) 544 cache_drop(&ncdrop1); 545 if (ncdrop2.ncp) 546 cache_drop(&ncdrop2); 547 if (vprele1) 548 vrele(vprele1); 549 if (vprele2) 550 vrele(vprele2); 551 } 552 return(0); 553 } 554 555 /* 556 * Unmount a file system. 557 * 558 * Note: unmount takes a path to the vnode mounted on as argument, 559 * not special file (as before). 560 * 561 * umount_args(char *path, int flags) 562 * 563 * MPALMOSTSAFE 564 */ 565 int 566 sys_unmount(struct unmount_args *uap) 567 { 568 struct thread *td = curthread; 569 struct proc *p __debugvar = td->td_proc; 570 struct mount *mp = NULL; 571 struct nlookupdata nd; 572 int error; 573 574 KKASSERT(p); 575 get_mplock(); 576 if (td->td_ucred->cr_prison != NULL) { 577 error = EPERM; 578 goto done; 579 } 580 if (usermount == 0 && (error = priv_check(td, PRIV_ROOT))) 581 goto done; 582 583 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 584 if (error == 0) 585 error = nlookup(&nd); 586 if (error) 587 goto out; 588 589 mp = nd.nl_nch.mount; 590 591 /* 592 * Only root, or the user that did the original mount is 593 * permitted to unmount this filesystem. 594 */ 595 if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) && 596 (error = priv_check(td, PRIV_ROOT))) 597 goto out; 598 599 /* 600 * Don't allow unmounting the root file system. 601 */ 602 if (mp->mnt_flag & MNT_ROOTFS) { 603 error = EINVAL; 604 goto out; 605 } 606 607 /* 608 * Must be the root of the filesystem 609 */ 610 if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) { 611 error = EINVAL; 612 goto out; 613 } 614 615 out: 616 nlookup_done(&nd); 617 if (error == 0) 618 error = dounmount(mp, uap->flags); 619 done: 620 rel_mplock(); 621 return (error); 622 } 623 624 /* 625 * Do the actual file system unmount. 626 */ 627 static int 628 dounmount_interlock(struct mount *mp) 629 { 630 if (mp->mnt_kern_flag & MNTK_UNMOUNT) 631 return (EBUSY); 632 mp->mnt_kern_flag |= MNTK_UNMOUNT; 633 return(0); 634 } 635 636 static int 637 unmount_allproc_cb(struct proc *p, void *arg) 638 { 639 struct mount *mp; 640 641 if (p->p_textnch.ncp == NULL) 642 return 0; 643 644 mp = (struct mount *)arg; 645 if (p->p_textnch.mount == mp) 646 cache_drop(&p->p_textnch); 647 648 return 0; 649 } 650 651 int 652 dounmount(struct mount *mp, int flags) 653 { 654 struct namecache *ncp; 655 struct nchandle nch; 656 struct vnode *vp; 657 int error; 658 int async_flag; 659 int lflags; 660 int freeok = 1; 661 662 /* 663 * Exclusive access for unmounting purposes 664 */ 665 if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0) 666 return (error); 667 668 /* 669 * Allow filesystems to detect that a forced unmount is in progress. 670 */ 671 if (flags & MNT_FORCE) 672 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 673 lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT); 674 error = lockmgr(&mp->mnt_lock, lflags); 675 if (error) { 676 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 677 if (mp->mnt_kern_flag & MNTK_MWAIT) 678 wakeup(mp); 679 return (error); 680 } 681 682 if (mp->mnt_flag & MNT_EXPUBLIC) 683 vfs_setpublicfs(NULL, NULL, NULL); 684 685 vfs_msync(mp, MNT_WAIT); 686 async_flag = mp->mnt_flag & MNT_ASYNC; 687 mp->mnt_flag &=~ MNT_ASYNC; 688 689 /* 690 * If this filesystem isn't aliasing other filesystems, 691 * try to invalidate any remaining namecache entries and 692 * check the count afterwords. 693 */ 694 if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) { 695 cache_lock(&mp->mnt_ncmountpt); 696 cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN); 697 cache_unlock(&mp->mnt_ncmountpt); 698 699 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL && 700 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) { 701 allproc_scan(&unmount_allproc_cb, mp); 702 } 703 704 if ((ncp = mp->mnt_ncmountpt.ncp) != NULL && 705 (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) { 706 707 if ((flags & MNT_FORCE) == 0) { 708 error = EBUSY; 709 mount_warning(mp, "Cannot unmount: " 710 "%d namecache " 711 "references still " 712 "present", 713 ncp->nc_refs - 1); 714 } else { 715 mount_warning(mp, "Forced unmount: " 716 "%d namecache " 717 "references still " 718 "present", 719 ncp->nc_refs - 1); 720 freeok = 0; 721 } 722 } 723 } 724 725 /* 726 * nchandle records ref the mount structure. Expect a count of 1 727 * (our mount->mnt_ncmountpt). 728 */ 729 if (mp->mnt_refs != 1) { 730 if ((flags & MNT_FORCE) == 0) { 731 mount_warning(mp, "Cannot unmount: " 732 "%d process references still " 733 "present", mp->mnt_refs); 734 error = EBUSY; 735 } else { 736 mount_warning(mp, "Forced unmount: " 737 "%d process references still " 738 "present", mp->mnt_refs); 739 freeok = 0; 740 } 741 } 742 743 /* 744 * Decomission our special mnt_syncer vnode. This also stops 745 * the vnlru code. If we are unable to unmount we recommission 746 * the vnode. 747 */ 748 if (error == 0) { 749 if ((vp = mp->mnt_syncer) != NULL) { 750 mp->mnt_syncer = NULL; 751 vrele(vp); 752 } 753 if (((mp->mnt_flag & MNT_RDONLY) || 754 (error = VFS_SYNC(mp, MNT_WAIT)) == 0) || 755 (flags & MNT_FORCE)) { 756 error = VFS_UNMOUNT(mp, flags); 757 } 758 } 759 if (error) { 760 if (mp->mnt_syncer == NULL) 761 vfs_allocate_syncvnode(mp); 762 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 763 mp->mnt_flag |= async_flag; 764 lockmgr(&mp->mnt_lock, LK_RELEASE); 765 if (mp->mnt_kern_flag & MNTK_MWAIT) 766 wakeup(mp); 767 return (error); 768 } 769 /* 770 * Clean up any journals still associated with the mount after 771 * filesystem activity has ceased. 772 */ 773 journal_remove_all_journals(mp, 774 ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0)); 775 776 mountlist_remove(mp); 777 778 /* 779 * Remove any installed vnode ops here so the individual VFSs don't 780 * have to. 781 */ 782 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops); 783 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops); 784 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops); 785 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops); 786 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops); 787 788 if (mp->mnt_ncmountpt.ncp != NULL) { 789 nch = mp->mnt_ncmountpt; 790 cache_zero(&mp->mnt_ncmountpt); 791 cache_clrmountpt(&nch); 792 cache_drop(&nch); 793 } 794 if (mp->mnt_ncmounton.ncp != NULL) { 795 nch = mp->mnt_ncmounton; 796 cache_zero(&mp->mnt_ncmounton); 797 cache_clrmountpt(&nch); 798 cache_drop(&nch); 799 } 800 801 mp->mnt_vfc->vfc_refcount--; 802 if (!TAILQ_EMPTY(&mp->mnt_nvnodelist)) 803 panic("unmount: dangling vnode"); 804 lockmgr(&mp->mnt_lock, LK_RELEASE); 805 if (mp->mnt_kern_flag & MNTK_MWAIT) 806 wakeup(mp); 807 if (freeok) 808 kfree(mp, M_MOUNT); 809 return (0); 810 } 811 812 static 813 void 814 mount_warning(struct mount *mp, const char *ctl, ...) 815 { 816 char *ptr; 817 char *buf; 818 __va_list va; 819 820 __va_start(va, ctl); 821 if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf, 0) == 0) { 822 kprintf("unmount(%s): ", ptr); 823 kvprintf(ctl, va); 824 kprintf("\n"); 825 kfree(buf, M_TEMP); 826 } else { 827 kprintf("unmount(%p", mp); 828 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name) 829 kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name); 830 kprintf("): "); 831 kvprintf(ctl, va); 832 kprintf("\n"); 833 } 834 __va_end(va); 835 } 836 837 /* 838 * Shim cache_fullpath() to handle the case where a process is chrooted into 839 * a subdirectory of a mount. In this case if the root mount matches the 840 * process root directory's mount we have to specify the process's root 841 * directory instead of the mount point, because the mount point might 842 * be above the root directory. 843 */ 844 static 845 int 846 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb) 847 { 848 struct nchandle *nch; 849 850 if (p && p->p_fd->fd_nrdir.mount == mp) 851 nch = &p->p_fd->fd_nrdir; 852 else 853 nch = &mp->mnt_ncmountpt; 854 return(cache_fullpath(p, nch, rb, fb, 0)); 855 } 856 857 /* 858 * Sync each mounted filesystem. 859 */ 860 861 #ifdef DEBUG 862 static int syncprt = 0; 863 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, ""); 864 #endif /* DEBUG */ 865 866 static int sync_callback(struct mount *mp, void *data); 867 868 /* 869 * MPALMOSTSAFE 870 */ 871 int 872 sys_sync(struct sync_args *uap) 873 { 874 get_mplock(); 875 mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD); 876 #ifdef DEBUG 877 /* 878 * print out buffer pool stat information on each sync() call. 879 */ 880 if (syncprt) 881 vfs_bufstats(); 882 #endif /* DEBUG */ 883 rel_mplock(); 884 return (0); 885 } 886 887 static 888 int 889 sync_callback(struct mount *mp, void *data __unused) 890 { 891 int asyncflag; 892 893 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 894 asyncflag = mp->mnt_flag & MNT_ASYNC; 895 mp->mnt_flag &= ~MNT_ASYNC; 896 vfs_msync(mp, MNT_NOWAIT); 897 VFS_SYNC(mp, MNT_NOWAIT); 898 mp->mnt_flag |= asyncflag; 899 } 900 return(0); 901 } 902 903 /* XXX PRISON: could be per prison flag */ 904 static int prison_quotas; 905 #if 0 906 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, ""); 907 #endif 908 909 /* 910 * quotactl_args(char *path, int fcmd, int uid, caddr_t arg) 911 * 912 * Change filesystem quotas. 913 * 914 * MPALMOSTSAFE 915 */ 916 int 917 sys_quotactl(struct quotactl_args *uap) 918 { 919 struct nlookupdata nd; 920 struct thread *td; 921 struct proc *p; 922 struct mount *mp; 923 int error; 924 925 get_mplock(); 926 td = curthread; 927 p = td->td_proc; 928 if (td->td_ucred->cr_prison && !prison_quotas) { 929 error = EPERM; 930 goto done; 931 } 932 933 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 934 if (error == 0) 935 error = nlookup(&nd); 936 if (error == 0) { 937 mp = nd.nl_nch.mount; 938 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, 939 uap->arg, nd.nl_cred); 940 } 941 nlookup_done(&nd); 942 done: 943 rel_mplock(); 944 return (error); 945 } 946 947 /* 948 * mountctl(char *path, int op, int fd, const void *ctl, int ctllen, 949 * void *buf, int buflen) 950 * 951 * This function operates on a mount point and executes the specified 952 * operation using the specified control data, and possibly returns data. 953 * 954 * The actual number of bytes stored in the result buffer is returned, 0 955 * if none, otherwise an error is returned. 956 * 957 * MPALMOSTSAFE 958 */ 959 int 960 sys_mountctl(struct mountctl_args *uap) 961 { 962 struct thread *td = curthread; 963 struct proc *p = td->td_proc; 964 struct file *fp; 965 void *ctl = NULL; 966 void *buf = NULL; 967 char *path = NULL; 968 int error; 969 970 /* 971 * Sanity and permissions checks. We must be root. 972 */ 973 KKASSERT(p); 974 if (td->td_ucred->cr_prison != NULL) 975 return (EPERM); 976 if ((uap->op != MOUNTCTL_MOUNTFLAGS) && 977 (error = priv_check(td, PRIV_ROOT)) != 0) 978 return (error); 979 980 /* 981 * Argument length checks 982 */ 983 if (uap->ctllen < 0 || uap->ctllen > 1024) 984 return (EINVAL); 985 if (uap->buflen < 0 || uap->buflen > 16 * 1024) 986 return (EINVAL); 987 if (uap->path == NULL) 988 return (EINVAL); 989 990 /* 991 * Allocate the necessary buffers and copyin data 992 */ 993 path = objcache_get(namei_oc, M_WAITOK); 994 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 995 if (error) 996 goto done; 997 998 if (uap->ctllen) { 999 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO); 1000 error = copyin(uap->ctl, ctl, uap->ctllen); 1001 if (error) 1002 goto done; 1003 } 1004 if (uap->buflen) 1005 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO); 1006 1007 /* 1008 * Validate the descriptor 1009 */ 1010 if (uap->fd >= 0) { 1011 fp = holdfp(p->p_fd, uap->fd, -1); 1012 if (fp == NULL) { 1013 error = EBADF; 1014 goto done; 1015 } 1016 } else { 1017 fp = NULL; 1018 } 1019 1020 /* 1021 * Execute the internal kernel function and clean up. 1022 */ 1023 get_mplock(); 1024 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result); 1025 rel_mplock(); 1026 if (fp) 1027 fdrop(fp); 1028 if (error == 0 && uap->sysmsg_result > 0) 1029 error = copyout(buf, uap->buf, uap->sysmsg_result); 1030 done: 1031 if (path) 1032 objcache_put(namei_oc, path); 1033 if (ctl) 1034 kfree(ctl, M_TEMP); 1035 if (buf) 1036 kfree(buf, M_TEMP); 1037 return (error); 1038 } 1039 1040 /* 1041 * Execute a mount control operation by resolving the path to a mount point 1042 * and calling vop_mountctl(). 1043 * 1044 * Use the mount point from the nch instead of the vnode so nullfs mounts 1045 * can properly spike the VOP. 1046 */ 1047 int 1048 kern_mountctl(const char *path, int op, struct file *fp, 1049 const void *ctl, int ctllen, 1050 void *buf, int buflen, int *res) 1051 { 1052 struct vnode *vp; 1053 struct mount *mp; 1054 struct nlookupdata nd; 1055 int error; 1056 1057 *res = 0; 1058 vp = NULL; 1059 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW); 1060 if (error == 0) 1061 error = nlookup(&nd); 1062 if (error == 0) 1063 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 1064 mp = nd.nl_nch.mount; 1065 nlookup_done(&nd); 1066 if (error) 1067 return (error); 1068 vn_unlock(vp); 1069 1070 /* 1071 * Must be the root of the filesystem 1072 */ 1073 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 1074 vrele(vp); 1075 return (EINVAL); 1076 } 1077 error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen, 1078 buf, buflen, res); 1079 vrele(vp); 1080 return (error); 1081 } 1082 1083 int 1084 kern_statfs(struct nlookupdata *nd, struct statfs *buf) 1085 { 1086 struct thread *td = curthread; 1087 struct proc *p = td->td_proc; 1088 struct mount *mp; 1089 struct statfs *sp; 1090 char *fullpath, *freepath; 1091 int error; 1092 1093 if ((error = nlookup(nd)) != 0) 1094 return (error); 1095 mp = nd->nl_nch.mount; 1096 sp = &mp->mnt_stat; 1097 if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0) 1098 return (error); 1099 1100 error = mount_path(p, mp, &fullpath, &freepath); 1101 if (error) 1102 return(error); 1103 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1104 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1105 kfree(freepath, M_TEMP); 1106 1107 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1108 bcopy(sp, buf, sizeof(*buf)); 1109 /* Only root should have access to the fsid's. */ 1110 if (priv_check(td, PRIV_ROOT)) 1111 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1112 return (0); 1113 } 1114 1115 /* 1116 * statfs_args(char *path, struct statfs *buf) 1117 * 1118 * Get filesystem statistics. 1119 * 1120 * MPALMOSTSAFE 1121 */ 1122 int 1123 sys_statfs(struct statfs_args *uap) 1124 { 1125 struct nlookupdata nd; 1126 struct statfs buf; 1127 int error; 1128 1129 get_mplock(); 1130 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1131 if (error == 0) 1132 error = kern_statfs(&nd, &buf); 1133 nlookup_done(&nd); 1134 if (error == 0) 1135 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1136 rel_mplock(); 1137 return (error); 1138 } 1139 1140 /* 1141 * MPALMOSTSAFE 1142 */ 1143 int 1144 kern_fstatfs(int fd, struct statfs *buf) 1145 { 1146 struct thread *td = curthread; 1147 struct proc *p = td->td_proc; 1148 struct file *fp; 1149 struct mount *mp; 1150 struct statfs *sp; 1151 char *fullpath, *freepath; 1152 int error; 1153 1154 KKASSERT(p); 1155 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 1156 return (error); 1157 get_mplock(); 1158 mp = ((struct vnode *)fp->f_data)->v_mount; 1159 if (mp == NULL) { 1160 error = EBADF; 1161 goto done; 1162 } 1163 if (fp->f_cred == NULL) { 1164 error = EINVAL; 1165 goto done; 1166 } 1167 sp = &mp->mnt_stat; 1168 if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0) 1169 goto done; 1170 1171 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0) 1172 goto done; 1173 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1174 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1175 kfree(freepath, M_TEMP); 1176 1177 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1178 bcopy(sp, buf, sizeof(*buf)); 1179 1180 /* Only root should have access to the fsid's. */ 1181 if (priv_check(td, PRIV_ROOT)) 1182 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1183 error = 0; 1184 done: 1185 rel_mplock(); 1186 fdrop(fp); 1187 return (error); 1188 } 1189 1190 /* 1191 * fstatfs_args(int fd, struct statfs *buf) 1192 * 1193 * Get filesystem statistics. 1194 * 1195 * MPSAFE 1196 */ 1197 int 1198 sys_fstatfs(struct fstatfs_args *uap) 1199 { 1200 struct statfs buf; 1201 int error; 1202 1203 error = kern_fstatfs(uap->fd, &buf); 1204 1205 if (error == 0) 1206 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1207 return (error); 1208 } 1209 1210 int 1211 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf) 1212 { 1213 struct mount *mp; 1214 struct statvfs *sp; 1215 int error; 1216 1217 if ((error = nlookup(nd)) != 0) 1218 return (error); 1219 mp = nd->nl_nch.mount; 1220 sp = &mp->mnt_vstat; 1221 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0) 1222 return (error); 1223 1224 sp->f_flag = 0; 1225 if (mp->mnt_flag & MNT_RDONLY) 1226 sp->f_flag |= ST_RDONLY; 1227 if (mp->mnt_flag & MNT_NOSUID) 1228 sp->f_flag |= ST_NOSUID; 1229 bcopy(sp, buf, sizeof(*buf)); 1230 return (0); 1231 } 1232 1233 /* 1234 * statfs_args(char *path, struct statfs *buf) 1235 * 1236 * Get filesystem statistics. 1237 * 1238 * MPALMOSTSAFE 1239 */ 1240 int 1241 sys_statvfs(struct statvfs_args *uap) 1242 { 1243 struct nlookupdata nd; 1244 struct statvfs buf; 1245 int error; 1246 1247 get_mplock(); 1248 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1249 if (error == 0) 1250 error = kern_statvfs(&nd, &buf); 1251 nlookup_done(&nd); 1252 if (error == 0) 1253 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1254 rel_mplock(); 1255 return (error); 1256 } 1257 1258 int 1259 kern_fstatvfs(int fd, struct statvfs *buf) 1260 { 1261 struct thread *td = curthread; 1262 struct proc *p = td->td_proc; 1263 struct file *fp; 1264 struct mount *mp; 1265 struct statvfs *sp; 1266 int error; 1267 1268 KKASSERT(p); 1269 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 1270 return (error); 1271 mp = ((struct vnode *)fp->f_data)->v_mount; 1272 if (mp == NULL) { 1273 error = EBADF; 1274 goto done; 1275 } 1276 if (fp->f_cred == NULL) { 1277 error = EINVAL; 1278 goto done; 1279 } 1280 sp = &mp->mnt_vstat; 1281 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0) 1282 goto done; 1283 1284 sp->f_flag = 0; 1285 if (mp->mnt_flag & MNT_RDONLY) 1286 sp->f_flag |= ST_RDONLY; 1287 if (mp->mnt_flag & MNT_NOSUID) 1288 sp->f_flag |= ST_NOSUID; 1289 1290 bcopy(sp, buf, sizeof(*buf)); 1291 error = 0; 1292 done: 1293 fdrop(fp); 1294 return (error); 1295 } 1296 1297 /* 1298 * fstatfs_args(int fd, struct statfs *buf) 1299 * 1300 * Get filesystem statistics. 1301 * 1302 * MPALMOSTSAFE 1303 */ 1304 int 1305 sys_fstatvfs(struct fstatvfs_args *uap) 1306 { 1307 struct statvfs buf; 1308 int error; 1309 1310 get_mplock(); 1311 error = kern_fstatvfs(uap->fd, &buf); 1312 rel_mplock(); 1313 1314 if (error == 0) 1315 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1316 return (error); 1317 } 1318 1319 /* 1320 * getfsstat_args(struct statfs *buf, long bufsize, int flags) 1321 * 1322 * Get statistics on all filesystems. 1323 */ 1324 1325 struct getfsstat_info { 1326 struct statfs *sfsp; 1327 long count; 1328 long maxcount; 1329 int error; 1330 int flags; 1331 struct thread *td; 1332 }; 1333 1334 static int getfsstat_callback(struct mount *, void *); 1335 1336 /* 1337 * MPALMOSTSAFE 1338 */ 1339 int 1340 sys_getfsstat(struct getfsstat_args *uap) 1341 { 1342 struct thread *td = curthread; 1343 struct getfsstat_info info; 1344 1345 bzero(&info, sizeof(info)); 1346 1347 info.maxcount = uap->bufsize / sizeof(struct statfs); 1348 info.sfsp = uap->buf; 1349 info.count = 0; 1350 info.flags = uap->flags; 1351 info.td = td; 1352 1353 get_mplock(); 1354 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD); 1355 rel_mplock(); 1356 if (info.sfsp && info.count > info.maxcount) 1357 uap->sysmsg_result = info.maxcount; 1358 else 1359 uap->sysmsg_result = info.count; 1360 return (info.error); 1361 } 1362 1363 static int 1364 getfsstat_callback(struct mount *mp, void *data) 1365 { 1366 struct getfsstat_info *info = data; 1367 struct statfs *sp; 1368 char *freepath; 1369 char *fullpath; 1370 int error; 1371 1372 if (info->sfsp && info->count < info->maxcount) { 1373 if (info->td->td_proc && 1374 !chroot_visible_mnt(mp, info->td->td_proc)) { 1375 return(0); 1376 } 1377 sp = &mp->mnt_stat; 1378 1379 /* 1380 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1381 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1382 * overrides MNT_WAIT. 1383 */ 1384 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1385 (info->flags & MNT_WAIT)) && 1386 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) { 1387 return(0); 1388 } 1389 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1390 1391 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath); 1392 if (error) { 1393 info->error = error; 1394 return(-1); 1395 } 1396 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1397 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1398 kfree(freepath, M_TEMP); 1399 1400 error = copyout(sp, info->sfsp, sizeof(*sp)); 1401 if (error) { 1402 info->error = error; 1403 return (-1); 1404 } 1405 ++info->sfsp; 1406 } 1407 info->count++; 1408 return(0); 1409 } 1410 1411 /* 1412 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf, 1413 long bufsize, int flags) 1414 * 1415 * Get statistics on all filesystems. 1416 */ 1417 1418 struct getvfsstat_info { 1419 struct statfs *sfsp; 1420 struct statvfs *vsfsp; 1421 long count; 1422 long maxcount; 1423 int error; 1424 int flags; 1425 struct thread *td; 1426 }; 1427 1428 static int getvfsstat_callback(struct mount *, void *); 1429 1430 /* 1431 * MPALMOSTSAFE 1432 */ 1433 int 1434 sys_getvfsstat(struct getvfsstat_args *uap) 1435 { 1436 struct thread *td = curthread; 1437 struct getvfsstat_info info; 1438 1439 bzero(&info, sizeof(info)); 1440 1441 info.maxcount = uap->vbufsize / sizeof(struct statvfs); 1442 info.sfsp = uap->buf; 1443 info.vsfsp = uap->vbuf; 1444 info.count = 0; 1445 info.flags = uap->flags; 1446 info.td = td; 1447 1448 get_mplock(); 1449 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD); 1450 if (info.vsfsp && info.count > info.maxcount) 1451 uap->sysmsg_result = info.maxcount; 1452 else 1453 uap->sysmsg_result = info.count; 1454 rel_mplock(); 1455 return (info.error); 1456 } 1457 1458 static int 1459 getvfsstat_callback(struct mount *mp, void *data) 1460 { 1461 struct getvfsstat_info *info = data; 1462 struct statfs *sp; 1463 struct statvfs *vsp; 1464 char *freepath; 1465 char *fullpath; 1466 int error; 1467 1468 if (info->vsfsp && info->count < info->maxcount) { 1469 if (info->td->td_proc && 1470 !chroot_visible_mnt(mp, info->td->td_proc)) { 1471 return(0); 1472 } 1473 sp = &mp->mnt_stat; 1474 vsp = &mp->mnt_vstat; 1475 1476 /* 1477 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1478 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1479 * overrides MNT_WAIT. 1480 */ 1481 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1482 (info->flags & MNT_WAIT)) && 1483 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) { 1484 return(0); 1485 } 1486 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1487 1488 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1489 (info->flags & MNT_WAIT)) && 1490 (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) { 1491 return(0); 1492 } 1493 vsp->f_flag = 0; 1494 if (mp->mnt_flag & MNT_RDONLY) 1495 vsp->f_flag |= ST_RDONLY; 1496 if (mp->mnt_flag & MNT_NOSUID) 1497 vsp->f_flag |= ST_NOSUID; 1498 1499 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath); 1500 if (error) { 1501 info->error = error; 1502 return(-1); 1503 } 1504 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1505 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1506 kfree(freepath, M_TEMP); 1507 1508 error = copyout(sp, info->sfsp, sizeof(*sp)); 1509 if (error == 0) 1510 error = copyout(vsp, info->vsfsp, sizeof(*vsp)); 1511 if (error) { 1512 info->error = error; 1513 return (-1); 1514 } 1515 ++info->sfsp; 1516 ++info->vsfsp; 1517 } 1518 info->count++; 1519 return(0); 1520 } 1521 1522 1523 /* 1524 * fchdir_args(int fd) 1525 * 1526 * Change current working directory to a given file descriptor. 1527 * 1528 * MPALMOSTSAFE 1529 */ 1530 int 1531 sys_fchdir(struct fchdir_args *uap) 1532 { 1533 struct thread *td = curthread; 1534 struct proc *p = td->td_proc; 1535 struct filedesc *fdp = p->p_fd; 1536 struct vnode *vp, *ovp; 1537 struct mount *mp; 1538 struct file *fp; 1539 struct nchandle nch, onch, tnch; 1540 int error; 1541 1542 if ((error = holdvnode(fdp, uap->fd, &fp)) != 0) 1543 return (error); 1544 get_mplock(); 1545 vp = (struct vnode *)fp->f_data; 1546 vref(vp); 1547 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1548 if (fp->f_nchandle.ncp == NULL) 1549 error = ENOTDIR; 1550 else 1551 error = checkvp_chdir(vp, td); 1552 if (error) { 1553 vput(vp); 1554 goto done; 1555 } 1556 cache_copy(&fp->f_nchandle, &nch); 1557 1558 /* 1559 * If the ncp has become a mount point, traverse through 1560 * the mount point. 1561 */ 1562 1563 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) && 1564 (mp = cache_findmount(&nch)) != NULL 1565 ) { 1566 error = nlookup_mp(mp, &tnch); 1567 if (error == 0) { 1568 cache_unlock(&tnch); /* leave ref intact */ 1569 vput(vp); 1570 vp = tnch.ncp->nc_vp; 1571 error = vget(vp, LK_SHARED); 1572 KKASSERT(error == 0); 1573 cache_drop(&nch); 1574 nch = tnch; 1575 } 1576 } 1577 if (error == 0) { 1578 ovp = fdp->fd_cdir; 1579 onch = fdp->fd_ncdir; 1580 vn_unlock(vp); /* leave ref intact */ 1581 fdp->fd_cdir = vp; 1582 fdp->fd_ncdir = nch; 1583 cache_drop(&onch); 1584 vrele(ovp); 1585 } else { 1586 cache_drop(&nch); 1587 vput(vp); 1588 } 1589 fdrop(fp); 1590 done: 1591 rel_mplock(); 1592 return (error); 1593 } 1594 1595 int 1596 kern_chdir(struct nlookupdata *nd) 1597 { 1598 struct thread *td = curthread; 1599 struct proc *p = td->td_proc; 1600 struct filedesc *fdp = p->p_fd; 1601 struct vnode *vp, *ovp; 1602 struct nchandle onch; 1603 int error; 1604 1605 if ((error = nlookup(nd)) != 0) 1606 return (error); 1607 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 1608 return (ENOENT); 1609 if ((error = vget(vp, LK_SHARED)) != 0) 1610 return (error); 1611 1612 error = checkvp_chdir(vp, td); 1613 vn_unlock(vp); 1614 if (error == 0) { 1615 ovp = fdp->fd_cdir; 1616 onch = fdp->fd_ncdir; 1617 cache_unlock(&nd->nl_nch); /* leave reference intact */ 1618 fdp->fd_ncdir = nd->nl_nch; 1619 fdp->fd_cdir = vp; 1620 cache_drop(&onch); 1621 vrele(ovp); 1622 cache_zero(&nd->nl_nch); 1623 } else { 1624 vrele(vp); 1625 } 1626 return (error); 1627 } 1628 1629 /* 1630 * chdir_args(char *path) 1631 * 1632 * Change current working directory (``.''). 1633 * 1634 * MPALMOSTSAFE 1635 */ 1636 int 1637 sys_chdir(struct chdir_args *uap) 1638 { 1639 struct nlookupdata nd; 1640 int error; 1641 1642 get_mplock(); 1643 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1644 if (error == 0) 1645 error = kern_chdir(&nd); 1646 nlookup_done(&nd); 1647 rel_mplock(); 1648 return (error); 1649 } 1650 1651 /* 1652 * Helper function for raised chroot(2) security function: Refuse if 1653 * any filedescriptors are open directories. 1654 */ 1655 static int 1656 chroot_refuse_vdir_fds(struct filedesc *fdp) 1657 { 1658 struct vnode *vp; 1659 struct file *fp; 1660 int error; 1661 int fd; 1662 1663 for (fd = 0; fd < fdp->fd_nfiles ; fd++) { 1664 if ((error = holdvnode(fdp, fd, &fp)) != 0) 1665 continue; 1666 vp = (struct vnode *)fp->f_data; 1667 if (vp->v_type != VDIR) { 1668 fdrop(fp); 1669 continue; 1670 } 1671 fdrop(fp); 1672 return(EPERM); 1673 } 1674 return (0); 1675 } 1676 1677 /* 1678 * This sysctl determines if we will allow a process to chroot(2) if it 1679 * has a directory open: 1680 * 0: disallowed for all processes. 1681 * 1: allowed for processes that were not already chroot(2)'ed. 1682 * 2: allowed for all processes. 1683 */ 1684 1685 static int chroot_allow_open_directories = 1; 1686 1687 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 1688 &chroot_allow_open_directories, 0, ""); 1689 1690 /* 1691 * chroot to the specified namecache entry. We obtain the vp from the 1692 * namecache data. The passed ncp must be locked and referenced and will 1693 * remain locked and referenced on return. 1694 */ 1695 int 1696 kern_chroot(struct nchandle *nch) 1697 { 1698 struct thread *td = curthread; 1699 struct proc *p = td->td_proc; 1700 struct filedesc *fdp = p->p_fd; 1701 struct vnode *vp; 1702 int error; 1703 1704 /* 1705 * Only privileged user can chroot 1706 */ 1707 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0); 1708 if (error) 1709 return (error); 1710 1711 /* 1712 * Disallow open directory descriptors (fchdir() breakouts). 1713 */ 1714 if (chroot_allow_open_directories == 0 || 1715 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 1716 if ((error = chroot_refuse_vdir_fds(fdp)) != 0) 1717 return (error); 1718 } 1719 if ((vp = nch->ncp->nc_vp) == NULL) 1720 return (ENOENT); 1721 1722 if ((error = vget(vp, LK_SHARED)) != 0) 1723 return (error); 1724 1725 /* 1726 * Check the validity of vp as a directory to change to and 1727 * associate it with rdir/jdir. 1728 */ 1729 error = checkvp_chdir(vp, td); 1730 vn_unlock(vp); /* leave reference intact */ 1731 if (error == 0) { 1732 vrele(fdp->fd_rdir); 1733 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */ 1734 cache_drop(&fdp->fd_nrdir); 1735 cache_copy(nch, &fdp->fd_nrdir); 1736 if (fdp->fd_jdir == NULL) { 1737 fdp->fd_jdir = vp; 1738 vref(fdp->fd_jdir); 1739 cache_copy(nch, &fdp->fd_njdir); 1740 } 1741 } else { 1742 vrele(vp); 1743 } 1744 return (error); 1745 } 1746 1747 /* 1748 * chroot_args(char *path) 1749 * 1750 * Change notion of root (``/'') directory. 1751 * 1752 * MPALMOSTSAFE 1753 */ 1754 int 1755 sys_chroot(struct chroot_args *uap) 1756 { 1757 struct thread *td __debugvar = curthread; 1758 struct nlookupdata nd; 1759 int error; 1760 1761 KKASSERT(td->td_proc); 1762 get_mplock(); 1763 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1764 if (error == 0) { 1765 nd.nl_flags |= NLC_EXEC; 1766 error = nlookup(&nd); 1767 if (error == 0) 1768 error = kern_chroot(&nd.nl_nch); 1769 } 1770 nlookup_done(&nd); 1771 rel_mplock(); 1772 return(error); 1773 } 1774 1775 int 1776 sys_chroot_kernel(struct chroot_kernel_args *uap) 1777 { 1778 struct thread *td = curthread; 1779 struct nlookupdata nd; 1780 struct nchandle *nch; 1781 struct vnode *vp; 1782 int error; 1783 1784 get_mplock(); 1785 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1786 if (error) 1787 goto error_nond; 1788 1789 error = nlookup(&nd); 1790 if (error) 1791 goto error_out; 1792 1793 nch = &nd.nl_nch; 1794 1795 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0); 1796 if (error) 1797 goto error_out; 1798 1799 if ((vp = nch->ncp->nc_vp) == NULL) { 1800 error = ENOENT; 1801 goto error_out; 1802 } 1803 1804 if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0) 1805 goto error_out; 1806 1807 kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path); 1808 vfs_cache_setroot(vp, cache_hold(nch)); 1809 1810 error_out: 1811 nlookup_done(&nd); 1812 error_nond: 1813 rel_mplock(); 1814 return(error); 1815 } 1816 1817 /* 1818 * Common routine for chroot and chdir. Given a locked, referenced vnode, 1819 * determine whether it is legal to chdir to the vnode. The vnode's state 1820 * is not changed by this call. 1821 */ 1822 int 1823 checkvp_chdir(struct vnode *vp, struct thread *td) 1824 { 1825 int error; 1826 1827 if (vp->v_type != VDIR) 1828 error = ENOTDIR; 1829 else 1830 error = VOP_EACCESS(vp, VEXEC, td->td_ucred); 1831 return (error); 1832 } 1833 1834 /* 1835 * MPSAFE 1836 */ 1837 int 1838 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res) 1839 { 1840 struct thread *td = curthread; 1841 struct proc *p = td->td_proc; 1842 struct lwp *lp = td->td_lwp; 1843 struct filedesc *fdp = p->p_fd; 1844 int cmode, flags; 1845 struct file *nfp; 1846 struct file *fp; 1847 struct vnode *vp; 1848 int type, indx, error; 1849 struct flock lf; 1850 1851 if ((oflags & O_ACCMODE) == O_ACCMODE) 1852 return (EINVAL); 1853 flags = FFLAGS(oflags); 1854 error = falloc(lp, &nfp, NULL); 1855 if (error) 1856 return (error); 1857 fp = nfp; 1858 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; 1859 1860 /* 1861 * XXX p_dupfd is a real mess. It allows a device to return a 1862 * file descriptor to be duplicated rather then doing the open 1863 * itself. 1864 */ 1865 lp->lwp_dupfd = -1; 1866 1867 /* 1868 * Call vn_open() to do the lookup and assign the vnode to the 1869 * file pointer. vn_open() does not change the ref count on fp 1870 * and the vnode, on success, will be inherited by the file pointer 1871 * and unlocked. 1872 */ 1873 nd->nl_flags |= NLC_LOCKVP; 1874 error = vn_open(nd, fp, flags, cmode); 1875 nlookup_done(nd); 1876 if (error) { 1877 /* 1878 * handle special fdopen() case. bleh. dupfdopen() is 1879 * responsible for dropping the old contents of ofiles[indx] 1880 * if it succeeds. 1881 * 1882 * Note that fsetfd() will add a ref to fp which represents 1883 * the fd_files[] assignment. We must still drop our 1884 * reference. 1885 */ 1886 if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) { 1887 if (fdalloc(p, 0, &indx) == 0) { 1888 error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error); 1889 if (error == 0) { 1890 *res = indx; 1891 fdrop(fp); /* our ref */ 1892 return (0); 1893 } 1894 fsetfd(fdp, NULL, indx); 1895 } 1896 } 1897 fdrop(fp); /* our ref */ 1898 if (error == ERESTART) 1899 error = EINTR; 1900 return (error); 1901 } 1902 1903 /* 1904 * ref the vnode for ourselves so it can't be ripped out from under 1905 * is. XXX need an ND flag to request that the vnode be returned 1906 * anyway. 1907 * 1908 * Reserve a file descriptor but do not assign it until the open 1909 * succeeds. 1910 */ 1911 vp = (struct vnode *)fp->f_data; 1912 vref(vp); 1913 if ((error = fdalloc(p, 0, &indx)) != 0) { 1914 fdrop(fp); 1915 vrele(vp); 1916 return (error); 1917 } 1918 1919 /* 1920 * If no error occurs the vp will have been assigned to the file 1921 * pointer. 1922 */ 1923 lp->lwp_dupfd = 0; 1924 1925 if (flags & (O_EXLOCK | O_SHLOCK)) { 1926 lf.l_whence = SEEK_SET; 1927 lf.l_start = 0; 1928 lf.l_len = 0; 1929 if (flags & O_EXLOCK) 1930 lf.l_type = F_WRLCK; 1931 else 1932 lf.l_type = F_RDLCK; 1933 if (flags & FNONBLOCK) 1934 type = 0; 1935 else 1936 type = F_WAIT; 1937 1938 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) { 1939 /* 1940 * lock request failed. Clean up the reserved 1941 * descriptor. 1942 */ 1943 vrele(vp); 1944 fsetfd(fdp, NULL, indx); 1945 fdrop(fp); 1946 return (error); 1947 } 1948 fp->f_flag |= FHASLOCK; 1949 } 1950 #if 0 1951 /* 1952 * Assert that all regular file vnodes were created with a object. 1953 */ 1954 KASSERT(vp->v_type != VREG || vp->v_object != NULL, 1955 ("open: regular file has no backing object after vn_open")); 1956 #endif 1957 1958 vrele(vp); 1959 1960 /* 1961 * release our private reference, leaving the one associated with the 1962 * descriptor table intact. 1963 */ 1964 fsetfd(fdp, fp, indx); 1965 fdrop(fp); 1966 *res = indx; 1967 return (0); 1968 } 1969 1970 /* 1971 * open_args(char *path, int flags, int mode) 1972 * 1973 * Check permissions, allocate an open file structure, 1974 * and call the device open routine if any. 1975 * 1976 * MPALMOSTSAFE 1977 */ 1978 int 1979 sys_open(struct open_args *uap) 1980 { 1981 struct nlookupdata nd; 1982 int error; 1983 1984 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 1985 if (error == 0) { 1986 error = kern_open(&nd, uap->flags, 1987 uap->mode, &uap->sysmsg_result); 1988 } 1989 nlookup_done(&nd); 1990 return (error); 1991 } 1992 1993 /* 1994 * openat_args(int fd, char *path, int flags, int mode) 1995 * 1996 * MPALMOSTSAFE 1997 */ 1998 int 1999 sys_openat(struct openat_args *uap) 2000 { 2001 struct nlookupdata nd; 2002 int error; 2003 struct file *fp; 2004 2005 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2006 if (error == 0) { 2007 error = kern_open(&nd, uap->flags, uap->mode, 2008 &uap->sysmsg_result); 2009 } 2010 nlookup_done_at(&nd, fp); 2011 return (error); 2012 } 2013 2014 int 2015 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor) 2016 { 2017 struct thread *td = curthread; 2018 struct proc *p = td->td_proc; 2019 struct vnode *vp; 2020 struct vattr vattr; 2021 int error; 2022 int whiteout = 0; 2023 2024 KKASSERT(p); 2025 2026 VATTR_NULL(&vattr); 2027 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2028 vattr.va_rmajor = rmajor; 2029 vattr.va_rminor = rminor; 2030 2031 switch (mode & S_IFMT) { 2032 case S_IFMT: /* used by badsect to flag bad sectors */ 2033 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0); 2034 vattr.va_type = VBAD; 2035 break; 2036 case S_IFCHR: 2037 error = priv_check(td, PRIV_VFS_MKNOD_DEV); 2038 vattr.va_type = VCHR; 2039 break; 2040 case S_IFBLK: 2041 error = priv_check(td, PRIV_VFS_MKNOD_DEV); 2042 vattr.va_type = VBLK; 2043 break; 2044 case S_IFWHT: 2045 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0); 2046 whiteout = 1; 2047 break; 2048 case S_IFDIR: /* special directories support for HAMMER */ 2049 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0); 2050 vattr.va_type = VDIR; 2051 break; 2052 default: 2053 error = EINVAL; 2054 break; 2055 } 2056 2057 if (error) 2058 return (error); 2059 2060 bwillinode(1); 2061 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2062 if ((error = nlookup(nd)) != 0) 2063 return (error); 2064 if (nd->nl_nch.ncp->nc_vp) 2065 return (EEXIST); 2066 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2067 return (error); 2068 2069 if (whiteout) { 2070 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp, 2071 nd->nl_cred, NAMEI_CREATE); 2072 } else { 2073 vp = NULL; 2074 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, 2075 &vp, nd->nl_cred, &vattr); 2076 if (error == 0) 2077 vput(vp); 2078 } 2079 return (error); 2080 } 2081 2082 /* 2083 * mknod_args(char *path, int mode, int dev) 2084 * 2085 * Create a special file. 2086 * 2087 * MPALMOSTSAFE 2088 */ 2089 int 2090 sys_mknod(struct mknod_args *uap) 2091 { 2092 struct nlookupdata nd; 2093 int error; 2094 2095 get_mplock(); 2096 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2097 if (error == 0) { 2098 error = kern_mknod(&nd, uap->mode, 2099 umajor(uap->dev), uminor(uap->dev)); 2100 } 2101 nlookup_done(&nd); 2102 rel_mplock(); 2103 return (error); 2104 } 2105 2106 /* 2107 * mknodat_args(int fd, char *path, mode_t mode, dev_t dev) 2108 * 2109 * Create a special file. The path is relative to the directory associated 2110 * with fd. 2111 * 2112 * MPALMOSTSAFE 2113 */ 2114 int 2115 sys_mknodat(struct mknodat_args *uap) 2116 { 2117 struct nlookupdata nd; 2118 struct file *fp; 2119 int error; 2120 2121 get_mplock(); 2122 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2123 if (error == 0) { 2124 error = kern_mknod(&nd, uap->mode, 2125 umajor(uap->dev), uminor(uap->dev)); 2126 } 2127 nlookup_done_at(&nd, fp); 2128 rel_mplock(); 2129 return (error); 2130 } 2131 2132 int 2133 kern_mkfifo(struct nlookupdata *nd, int mode) 2134 { 2135 struct thread *td = curthread; 2136 struct proc *p = td->td_proc; 2137 struct vattr vattr; 2138 struct vnode *vp; 2139 int error; 2140 2141 bwillinode(1); 2142 2143 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2144 if ((error = nlookup(nd)) != 0) 2145 return (error); 2146 if (nd->nl_nch.ncp->nc_vp) 2147 return (EEXIST); 2148 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2149 return (error); 2150 2151 VATTR_NULL(&vattr); 2152 vattr.va_type = VFIFO; 2153 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2154 vp = NULL; 2155 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr); 2156 if (error == 0) 2157 vput(vp); 2158 return (error); 2159 } 2160 2161 /* 2162 * mkfifo_args(char *path, int mode) 2163 * 2164 * Create a named pipe. 2165 * 2166 * MPALMOSTSAFE 2167 */ 2168 int 2169 sys_mkfifo(struct mkfifo_args *uap) 2170 { 2171 struct nlookupdata nd; 2172 int error; 2173 2174 get_mplock(); 2175 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2176 if (error == 0) 2177 error = kern_mkfifo(&nd, uap->mode); 2178 nlookup_done(&nd); 2179 rel_mplock(); 2180 return (error); 2181 } 2182 2183 /* 2184 * mkfifoat_args(int fd, char *path, mode_t mode) 2185 * 2186 * Create a named pipe. The path is relative to the directory associated 2187 * with fd. 2188 * 2189 * MPALMOSTSAFE 2190 */ 2191 int 2192 sys_mkfifoat(struct mkfifoat_args *uap) 2193 { 2194 struct nlookupdata nd; 2195 struct file *fp; 2196 int error; 2197 2198 get_mplock(); 2199 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2200 if (error == 0) 2201 error = kern_mkfifo(&nd, uap->mode); 2202 nlookup_done_at(&nd, fp); 2203 rel_mplock(); 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 * MPALMOSTSAFE 2328 */ 2329 int 2330 sys_link(struct link_args *uap) 2331 { 2332 struct nlookupdata nd, linknd; 2333 int error; 2334 2335 get_mplock(); 2336 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2337 if (error == 0) { 2338 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0); 2339 if (error == 0) 2340 error = kern_link(&nd, &linknd); 2341 nlookup_done(&linknd); 2342 } 2343 nlookup_done(&nd); 2344 rel_mplock(); 2345 return (error); 2346 } 2347 2348 int 2349 kern_symlink(struct nlookupdata *nd, char *path, int mode) 2350 { 2351 struct vattr vattr; 2352 struct vnode *vp; 2353 struct vnode *dvp; 2354 int error; 2355 2356 bwillinode(1); 2357 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2358 if ((error = nlookup(nd)) != 0) 2359 return (error); 2360 if (nd->nl_nch.ncp->nc_vp) 2361 return (EEXIST); 2362 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2363 return (error); 2364 dvp = nd->nl_dvp; 2365 VATTR_NULL(&vattr); 2366 vattr.va_mode = mode; 2367 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path); 2368 if (error == 0) 2369 vput(vp); 2370 return (error); 2371 } 2372 2373 /* 2374 * symlink(char *path, char *link) 2375 * 2376 * Make a symbolic link. 2377 * 2378 * MPALMOSTSAFE 2379 */ 2380 int 2381 sys_symlink(struct symlink_args *uap) 2382 { 2383 struct thread *td = curthread; 2384 struct nlookupdata nd; 2385 char *path; 2386 int error; 2387 int mode; 2388 2389 path = objcache_get(namei_oc, M_WAITOK); 2390 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 2391 if (error == 0) { 2392 get_mplock(); 2393 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0); 2394 if (error == 0) { 2395 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2396 error = kern_symlink(&nd, path, mode); 2397 } 2398 nlookup_done(&nd); 2399 rel_mplock(); 2400 } 2401 objcache_put(namei_oc, path); 2402 return (error); 2403 } 2404 2405 /* 2406 * symlinkat_args(char *path1, int fd, char *path2) 2407 * 2408 * Make a symbolic link. The path2 argument is relative to the directory 2409 * associated with fd. 2410 * 2411 * MPALMOSTSAFE 2412 */ 2413 int 2414 sys_symlinkat(struct symlinkat_args *uap) 2415 { 2416 struct thread *td = curthread; 2417 struct nlookupdata nd; 2418 struct file *fp; 2419 char *path1; 2420 int error; 2421 int mode; 2422 2423 path1 = objcache_get(namei_oc, M_WAITOK); 2424 error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL); 2425 if (error == 0) { 2426 get_mplock(); 2427 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2, 2428 UIO_USERSPACE, 0); 2429 if (error == 0) { 2430 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2431 error = kern_symlink(&nd, path1, mode); 2432 } 2433 nlookup_done_at(&nd, fp); 2434 rel_mplock(); 2435 } 2436 objcache_put(namei_oc, path1); 2437 return (error); 2438 } 2439 2440 /* 2441 * undelete_args(char *path) 2442 * 2443 * Delete a whiteout from the filesystem. 2444 * 2445 * MPALMOSTSAFE 2446 */ 2447 int 2448 sys_undelete(struct undelete_args *uap) 2449 { 2450 struct nlookupdata nd; 2451 int error; 2452 2453 get_mplock(); 2454 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2455 bwillinode(1); 2456 nd.nl_flags |= NLC_DELETE | NLC_REFDVP; 2457 if (error == 0) 2458 error = nlookup(&nd); 2459 if (error == 0) 2460 error = ncp_writechk(&nd.nl_nch); 2461 if (error == 0) { 2462 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred, 2463 NAMEI_DELETE); 2464 } 2465 nlookup_done(&nd); 2466 rel_mplock(); 2467 return (error); 2468 } 2469 2470 int 2471 kern_unlink(struct nlookupdata *nd) 2472 { 2473 int error; 2474 2475 bwillinode(1); 2476 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 2477 if ((error = nlookup(nd)) != 0) 2478 return (error); 2479 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2480 return (error); 2481 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 2482 return (error); 2483 } 2484 2485 /* 2486 * unlink_args(char *path) 2487 * 2488 * Delete a name from the filesystem. 2489 * 2490 * MPALMOSTSAFE 2491 */ 2492 int 2493 sys_unlink(struct unlink_args *uap) 2494 { 2495 struct nlookupdata nd; 2496 int error; 2497 2498 get_mplock(); 2499 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2500 if (error == 0) 2501 error = kern_unlink(&nd); 2502 nlookup_done(&nd); 2503 rel_mplock(); 2504 return (error); 2505 } 2506 2507 2508 /* 2509 * unlinkat_args(int fd, char *path, int flags) 2510 * 2511 * Delete the file or directory entry pointed to by fd/path. 2512 * 2513 * MPALMOSTSAFE 2514 */ 2515 int 2516 sys_unlinkat(struct unlinkat_args *uap) 2517 { 2518 struct nlookupdata nd; 2519 struct file *fp; 2520 int error; 2521 2522 if (uap->flags & ~AT_REMOVEDIR) 2523 return (EINVAL); 2524 2525 get_mplock(); 2526 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2527 if (error == 0) { 2528 if (uap->flags & AT_REMOVEDIR) 2529 error = kern_rmdir(&nd); 2530 else 2531 error = kern_unlink(&nd); 2532 } 2533 nlookup_done_at(&nd, fp); 2534 rel_mplock(); 2535 return (error); 2536 } 2537 2538 /* 2539 * MPALMOSTSAFE 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 get_mplock(); 2569 error = VOP_GETATTR(vp, &vattr); 2570 rel_mplock(); 2571 spin_lock(&fp->f_spin); 2572 new_offset = offset + vattr.va_size; 2573 break; 2574 case L_SET: 2575 new_offset = offset; 2576 error = 0; 2577 spin_lock(&fp->f_spin); 2578 break; 2579 default: 2580 new_offset = 0; 2581 error = EINVAL; 2582 spin_lock(&fp->f_spin); 2583 break; 2584 } 2585 2586 /* 2587 * Validate the seek position. Negative offsets are not allowed 2588 * for regular files or directories. 2589 * 2590 * Normally we would also not want to allow negative offsets for 2591 * character and block-special devices. However kvm addresses 2592 * on 64 bit architectures might appear to be negative and must 2593 * be allowed. 2594 */ 2595 if (error == 0) { 2596 if (new_offset < 0 && 2597 (vp->v_type == VREG || vp->v_type == VDIR)) { 2598 error = EINVAL; 2599 } else { 2600 fp->f_offset = new_offset; 2601 } 2602 } 2603 *res = fp->f_offset; 2604 spin_unlock(&fp->f_spin); 2605 done: 2606 fdrop(fp); 2607 return (error); 2608 } 2609 2610 /* 2611 * lseek_args(int fd, int pad, off_t offset, int whence) 2612 * 2613 * Reposition read/write file offset. 2614 * 2615 * MPSAFE 2616 */ 2617 int 2618 sys_lseek(struct lseek_args *uap) 2619 { 2620 int error; 2621 2622 error = kern_lseek(uap->fd, uap->offset, uap->whence, 2623 &uap->sysmsg_offset); 2624 2625 return (error); 2626 } 2627 2628 /* 2629 * Check if current process can access given file. amode is a bitmask of *_OK 2630 * access bits. flags is a bitmask of AT_* flags. 2631 */ 2632 int 2633 kern_access(struct nlookupdata *nd, int amode, int flags) 2634 { 2635 struct vnode *vp; 2636 int error, mode; 2637 2638 if (flags & ~AT_EACCESS) 2639 return (EINVAL); 2640 if ((error = nlookup(nd)) != 0) 2641 return (error); 2642 retry: 2643 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp); 2644 if (error) 2645 return (error); 2646 2647 /* Flags == 0 means only check for existence. */ 2648 if (amode) { 2649 mode = 0; 2650 if (amode & R_OK) 2651 mode |= VREAD; 2652 if (amode & W_OK) 2653 mode |= VWRITE; 2654 if (amode & X_OK) 2655 mode |= VEXEC; 2656 if ((mode & VWRITE) == 0 || 2657 (error = vn_writechk(vp, &nd->nl_nch)) == 0) 2658 error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred); 2659 2660 /* 2661 * If the file handle is stale we have to re-resolve the 2662 * entry. This is a hack at the moment. 2663 */ 2664 if (error == ESTALE) { 2665 vput(vp); 2666 cache_setunresolved(&nd->nl_nch); 2667 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2668 if (error == 0) { 2669 vp = NULL; 2670 goto retry; 2671 } 2672 return(error); 2673 } 2674 } 2675 vput(vp); 2676 return (error); 2677 } 2678 2679 /* 2680 * access_args(char *path, int flags) 2681 * 2682 * Check access permissions. 2683 * 2684 * MPALMOSTSAFE 2685 */ 2686 int 2687 sys_access(struct access_args *uap) 2688 { 2689 struct nlookupdata nd; 2690 int error; 2691 2692 get_mplock(); 2693 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2694 if (error == 0) 2695 error = kern_access(&nd, uap->flags, 0); 2696 nlookup_done(&nd); 2697 rel_mplock(); 2698 return (error); 2699 } 2700 2701 2702 /* 2703 * faccessat_args(int fd, char *path, int amode, int flags) 2704 * 2705 * Check access permissions. 2706 * 2707 * MPALMOSTSAFE 2708 */ 2709 int 2710 sys_faccessat(struct faccessat_args *uap) 2711 { 2712 struct nlookupdata nd; 2713 struct file *fp; 2714 int error; 2715 2716 get_mplock(); 2717 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 2718 NLC_FOLLOW); 2719 if (error == 0) 2720 error = kern_access(&nd, uap->amode, uap->flags); 2721 nlookup_done_at(&nd, fp); 2722 rel_mplock(); 2723 return (error); 2724 } 2725 2726 2727 /* 2728 * MPSAFE 2729 */ 2730 int 2731 kern_stat(struct nlookupdata *nd, struct stat *st) 2732 { 2733 int error; 2734 struct vnode *vp; 2735 thread_t td; 2736 2737 if ((error = nlookup(nd)) != 0) 2738 return (error); 2739 again: 2740 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 2741 return (ENOENT); 2742 2743 td = curthread; 2744 if ((error = vget(vp, LK_SHARED)) != 0) 2745 return (error); 2746 error = vn_stat(vp, st, nd->nl_cred); 2747 2748 /* 2749 * If the file handle is stale we have to re-resolve the entry. This 2750 * is a hack at the moment. 2751 */ 2752 if (error == ESTALE) { 2753 vput(vp); 2754 cache_setunresolved(&nd->nl_nch); 2755 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2756 if (error == 0) 2757 goto again; 2758 } else { 2759 vput(vp); 2760 } 2761 return (error); 2762 } 2763 2764 /* 2765 * stat_args(char *path, struct stat *ub) 2766 * 2767 * Get file status; this version follows links. 2768 * 2769 * MPSAFE 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 * MPALMOSTSAFE 2794 */ 2795 int 2796 sys_lstat(struct lstat_args *uap) 2797 { 2798 struct nlookupdata nd; 2799 struct stat st; 2800 int error; 2801 2802 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2803 if (error == 0) { 2804 error = kern_stat(&nd, &st); 2805 if (error == 0) 2806 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 2807 } 2808 nlookup_done(&nd); 2809 return (error); 2810 } 2811 2812 /* 2813 * fstatat_args(int fd, char *path, struct stat *sb, int flags) 2814 * 2815 * Get status of file pointed to by fd/path. 2816 * 2817 * MPALMOSTSAFE 2818 */ 2819 int 2820 sys_fstatat(struct fstatat_args *uap) 2821 { 2822 struct nlookupdata nd; 2823 struct stat st; 2824 int error; 2825 int flags; 2826 struct file *fp; 2827 2828 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 2829 return (EINVAL); 2830 2831 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 2832 2833 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 2834 UIO_USERSPACE, flags); 2835 if (error == 0) { 2836 error = kern_stat(&nd, &st); 2837 if (error == 0) 2838 error = copyout(&st, uap->sb, sizeof(*uap->sb)); 2839 } 2840 nlookup_done_at(&nd, fp); 2841 return (error); 2842 } 2843 2844 /* 2845 * pathconf_Args(char *path, int name) 2846 * 2847 * Get configurable pathname variables. 2848 * 2849 * MPALMOSTSAFE 2850 */ 2851 int 2852 sys_pathconf(struct pathconf_args *uap) 2853 { 2854 struct nlookupdata nd; 2855 struct vnode *vp; 2856 int error; 2857 2858 vp = NULL; 2859 get_mplock(); 2860 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2861 if (error == 0) 2862 error = nlookup(&nd); 2863 if (error == 0) 2864 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 2865 nlookup_done(&nd); 2866 if (error == 0) { 2867 error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg); 2868 vput(vp); 2869 } 2870 rel_mplock(); 2871 return (error); 2872 } 2873 2874 /* 2875 * XXX: daver 2876 * kern_readlink isn't properly split yet. There is a copyin burried 2877 * in VOP_READLINK(). 2878 */ 2879 int 2880 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res) 2881 { 2882 struct thread *td = curthread; 2883 struct vnode *vp; 2884 struct iovec aiov; 2885 struct uio auio; 2886 int error; 2887 2888 if ((error = nlookup(nd)) != 0) 2889 return (error); 2890 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp); 2891 if (error) 2892 return (error); 2893 if (vp->v_type != VLNK) { 2894 error = EINVAL; 2895 } else { 2896 aiov.iov_base = buf; 2897 aiov.iov_len = count; 2898 auio.uio_iov = &aiov; 2899 auio.uio_iovcnt = 1; 2900 auio.uio_offset = 0; 2901 auio.uio_rw = UIO_READ; 2902 auio.uio_segflg = UIO_USERSPACE; 2903 auio.uio_td = td; 2904 auio.uio_resid = count; 2905 error = VOP_READLINK(vp, &auio, td->td_ucred); 2906 } 2907 vput(vp); 2908 *res = count - auio.uio_resid; 2909 return (error); 2910 } 2911 2912 /* 2913 * readlink_args(char *path, char *buf, int count) 2914 * 2915 * Return target name of a symbolic link. 2916 * 2917 * MPALMOSTSAFE 2918 */ 2919 int 2920 sys_readlink(struct readlink_args *uap) 2921 { 2922 struct nlookupdata nd; 2923 int error; 2924 2925 get_mplock(); 2926 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2927 if (error == 0) { 2928 error = kern_readlink(&nd, uap->buf, uap->count, 2929 &uap->sysmsg_result); 2930 } 2931 nlookup_done(&nd); 2932 rel_mplock(); 2933 return (error); 2934 } 2935 2936 /* 2937 * readlinkat_args(int fd, char *path, char *buf, size_t bufsize) 2938 * 2939 * Return target name of a symbolic link. The path is relative to the 2940 * directory associated with fd. 2941 * 2942 * MPALMOSTSAFE 2943 */ 2944 int 2945 sys_readlinkat(struct readlinkat_args *uap) 2946 { 2947 struct nlookupdata nd; 2948 struct file *fp; 2949 int error; 2950 2951 get_mplock(); 2952 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2953 if (error == 0) { 2954 error = kern_readlink(&nd, uap->buf, uap->bufsize, 2955 &uap->sysmsg_result); 2956 } 2957 nlookup_done_at(&nd, fp); 2958 rel_mplock(); 2959 return (error); 2960 } 2961 2962 static int 2963 setfflags(struct vnode *vp, int flags) 2964 { 2965 struct thread *td = curthread; 2966 int error; 2967 struct vattr vattr; 2968 2969 /* 2970 * Prevent non-root users from setting flags on devices. When 2971 * a device is reused, users can retain ownership of the device 2972 * if they are allowed to set flags and programs assume that 2973 * chown can't fail when done as root. 2974 */ 2975 if ((vp->v_type == VCHR || vp->v_type == VBLK) && 2976 ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0)) 2977 return (error); 2978 2979 /* 2980 * note: vget is required for any operation that might mod the vnode 2981 * so VINACTIVE is properly cleared. 2982 */ 2983 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 2984 VATTR_NULL(&vattr); 2985 vattr.va_flags = flags; 2986 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 2987 vput(vp); 2988 } 2989 return (error); 2990 } 2991 2992 /* 2993 * chflags(char *path, int flags) 2994 * 2995 * Change flags of a file given a path name. 2996 * 2997 * MPALMOSTSAFE 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 get_mplock(); 3008 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3009 if (error == 0) 3010 error = nlookup(&nd); 3011 if (error == 0) 3012 error = ncp_writechk(&nd.nl_nch); 3013 if (error == 0) 3014 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3015 nlookup_done(&nd); 3016 if (error == 0) { 3017 error = setfflags(vp, uap->flags); 3018 vrele(vp); 3019 } 3020 rel_mplock(); 3021 return (error); 3022 } 3023 3024 /* 3025 * lchflags(char *path, int flags) 3026 * 3027 * Change flags of a file given a path name, but don't follow symlinks. 3028 * 3029 * MPALMOSTSAFE 3030 */ 3031 int 3032 sys_lchflags(struct lchflags_args *uap) 3033 { 3034 struct nlookupdata nd; 3035 struct vnode *vp; 3036 int error; 3037 3038 vp = NULL; 3039 get_mplock(); 3040 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3041 if (error == 0) 3042 error = nlookup(&nd); 3043 if (error == 0) 3044 error = ncp_writechk(&nd.nl_nch); 3045 if (error == 0) 3046 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3047 nlookup_done(&nd); 3048 if (error == 0) { 3049 error = setfflags(vp, uap->flags); 3050 vrele(vp); 3051 } 3052 rel_mplock(); 3053 return (error); 3054 } 3055 3056 /* 3057 * fchflags_args(int fd, int flags) 3058 * 3059 * Change flags of a file given a file descriptor. 3060 * 3061 * MPALMOSTSAFE 3062 */ 3063 int 3064 sys_fchflags(struct fchflags_args *uap) 3065 { 3066 struct thread *td = curthread; 3067 struct proc *p = td->td_proc; 3068 struct file *fp; 3069 int error; 3070 3071 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3072 return (error); 3073 get_mplock(); 3074 if (fp->f_nchandle.ncp) 3075 error = ncp_writechk(&fp->f_nchandle); 3076 if (error == 0) 3077 error = setfflags((struct vnode *) fp->f_data, uap->flags); 3078 rel_mplock(); 3079 fdrop(fp); 3080 return (error); 3081 } 3082 3083 static int 3084 setfmode(struct vnode *vp, int mode) 3085 { 3086 struct thread *td = curthread; 3087 int error; 3088 struct vattr vattr; 3089 3090 /* 3091 * note: vget is required for any operation that might mod the vnode 3092 * so VINACTIVE is properly cleared. 3093 */ 3094 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3095 VATTR_NULL(&vattr); 3096 vattr.va_mode = mode & ALLPERMS; 3097 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3098 vput(vp); 3099 } 3100 return error; 3101 } 3102 3103 int 3104 kern_chmod(struct nlookupdata *nd, int mode) 3105 { 3106 struct vnode *vp; 3107 int error; 3108 3109 if ((error = nlookup(nd)) != 0) 3110 return (error); 3111 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3112 return (error); 3113 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3114 error = setfmode(vp, mode); 3115 vrele(vp); 3116 return (error); 3117 } 3118 3119 /* 3120 * chmod_args(char *path, int mode) 3121 * 3122 * Change mode of a file given path name. 3123 * 3124 * MPALMOSTSAFE 3125 */ 3126 int 3127 sys_chmod(struct chmod_args *uap) 3128 { 3129 struct nlookupdata nd; 3130 int error; 3131 3132 get_mplock(); 3133 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3134 if (error == 0) 3135 error = kern_chmod(&nd, uap->mode); 3136 nlookup_done(&nd); 3137 rel_mplock(); 3138 return (error); 3139 } 3140 3141 /* 3142 * lchmod_args(char *path, int mode) 3143 * 3144 * Change mode of a file given path name (don't follow links.) 3145 * 3146 * MPALMOSTSAFE 3147 */ 3148 int 3149 sys_lchmod(struct lchmod_args *uap) 3150 { 3151 struct nlookupdata nd; 3152 int error; 3153 3154 get_mplock(); 3155 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3156 if (error == 0) 3157 error = kern_chmod(&nd, uap->mode); 3158 nlookup_done(&nd); 3159 rel_mplock(); 3160 return (error); 3161 } 3162 3163 /* 3164 * fchmod_args(int fd, int mode) 3165 * 3166 * Change mode of a file given a file descriptor. 3167 * 3168 * MPALMOSTSAFE 3169 */ 3170 int 3171 sys_fchmod(struct fchmod_args *uap) 3172 { 3173 struct thread *td = curthread; 3174 struct proc *p = td->td_proc; 3175 struct file *fp; 3176 int error; 3177 3178 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3179 return (error); 3180 get_mplock(); 3181 if (fp->f_nchandle.ncp) 3182 error = ncp_writechk(&fp->f_nchandle); 3183 if (error == 0) 3184 error = setfmode((struct vnode *)fp->f_data, uap->mode); 3185 rel_mplock(); 3186 fdrop(fp); 3187 return (error); 3188 } 3189 3190 /* 3191 * fchmodat_args(char *path, int mode) 3192 * 3193 * Change mode of a file pointed to by fd/path. 3194 * 3195 * MPALMOSTSAFE 3196 */ 3197 int 3198 sys_fchmodat(struct fchmodat_args *uap) 3199 { 3200 struct nlookupdata nd; 3201 struct file *fp; 3202 int error; 3203 int flags; 3204 3205 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3206 return (EINVAL); 3207 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3208 3209 get_mplock(); 3210 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3211 UIO_USERSPACE, flags); 3212 if (error == 0) 3213 error = kern_chmod(&nd, uap->mode); 3214 nlookup_done_at(&nd, fp); 3215 rel_mplock(); 3216 return (error); 3217 } 3218 3219 static int 3220 setfown(struct vnode *vp, uid_t uid, gid_t gid) 3221 { 3222 struct thread *td = curthread; 3223 int error; 3224 struct vattr vattr; 3225 3226 /* 3227 * note: vget is required for any operation that might mod the vnode 3228 * so VINACTIVE is properly cleared. 3229 */ 3230 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3231 VATTR_NULL(&vattr); 3232 vattr.va_uid = uid; 3233 vattr.va_gid = gid; 3234 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3235 vput(vp); 3236 } 3237 return error; 3238 } 3239 3240 int 3241 kern_chown(struct nlookupdata *nd, int uid, int gid) 3242 { 3243 struct vnode *vp; 3244 int error; 3245 3246 if ((error = nlookup(nd)) != 0) 3247 return (error); 3248 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3249 return (error); 3250 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3251 error = setfown(vp, uid, gid); 3252 vrele(vp); 3253 return (error); 3254 } 3255 3256 /* 3257 * chown(char *path, int uid, int gid) 3258 * 3259 * Set ownership given a path name. 3260 * 3261 * MPALMOSTSAFE 3262 */ 3263 int 3264 sys_chown(struct chown_args *uap) 3265 { 3266 struct nlookupdata nd; 3267 int error; 3268 3269 get_mplock(); 3270 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3271 if (error == 0) 3272 error = kern_chown(&nd, uap->uid, uap->gid); 3273 nlookup_done(&nd); 3274 rel_mplock(); 3275 return (error); 3276 } 3277 3278 /* 3279 * lchown_args(char *path, int uid, int gid) 3280 * 3281 * Set ownership given a path name, do not cross symlinks. 3282 * 3283 * MPALMOSTSAFE 3284 */ 3285 int 3286 sys_lchown(struct lchown_args *uap) 3287 { 3288 struct nlookupdata nd; 3289 int error; 3290 3291 get_mplock(); 3292 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3293 if (error == 0) 3294 error = kern_chown(&nd, uap->uid, uap->gid); 3295 nlookup_done(&nd); 3296 rel_mplock(); 3297 return (error); 3298 } 3299 3300 /* 3301 * fchown_args(int fd, int uid, int gid) 3302 * 3303 * Set ownership given a file descriptor. 3304 * 3305 * MPALMOSTSAFE 3306 */ 3307 int 3308 sys_fchown(struct fchown_args *uap) 3309 { 3310 struct thread *td = curthread; 3311 struct proc *p = td->td_proc; 3312 struct file *fp; 3313 int error; 3314 3315 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3316 return (error); 3317 get_mplock(); 3318 if (fp->f_nchandle.ncp) 3319 error = ncp_writechk(&fp->f_nchandle); 3320 if (error == 0) 3321 error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid); 3322 rel_mplock(); 3323 fdrop(fp); 3324 return (error); 3325 } 3326 3327 /* 3328 * fchownat(int fd, char *path, int uid, int gid, int flags) 3329 * 3330 * Set ownership of file pointed to by fd/path. 3331 * 3332 * MPALMOSTSAFE 3333 */ 3334 int 3335 sys_fchownat(struct fchownat_args *uap) 3336 { 3337 struct nlookupdata nd; 3338 struct file *fp; 3339 int error; 3340 int flags; 3341 3342 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3343 return (EINVAL); 3344 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3345 3346 get_mplock(); 3347 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3348 UIO_USERSPACE, flags); 3349 if (error == 0) 3350 error = kern_chown(&nd, uap->uid, uap->gid); 3351 nlookup_done_at(&nd, fp); 3352 rel_mplock(); 3353 return (error); 3354 } 3355 3356 3357 static int 3358 getutimes(const struct timeval *tvp, struct timespec *tsp) 3359 { 3360 struct timeval tv[2]; 3361 3362 if (tvp == NULL) { 3363 microtime(&tv[0]); 3364 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]); 3365 tsp[1] = tsp[0]; 3366 } else { 3367 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); 3368 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); 3369 } 3370 return 0; 3371 } 3372 3373 static int 3374 setutimes(struct vnode *vp, struct vattr *vattr, 3375 const struct timespec *ts, int nullflag) 3376 { 3377 struct thread *td = curthread; 3378 int error; 3379 3380 VATTR_NULL(vattr); 3381 vattr->va_atime = ts[0]; 3382 vattr->va_mtime = ts[1]; 3383 if (nullflag) 3384 vattr->va_vaflags |= VA_UTIMES_NULL; 3385 error = VOP_SETATTR(vp, vattr, td->td_ucred); 3386 3387 return error; 3388 } 3389 3390 int 3391 kern_utimes(struct nlookupdata *nd, struct timeval *tptr) 3392 { 3393 struct timespec ts[2]; 3394 struct vnode *vp; 3395 struct vattr vattr; 3396 int error; 3397 3398 if ((error = getutimes(tptr, ts)) != 0) 3399 return (error); 3400 3401 /* 3402 * NOTE: utimes() succeeds for the owner even if the file 3403 * is not user-writable. 3404 */ 3405 nd->nl_flags |= NLC_OWN | NLC_WRITE; 3406 3407 if ((error = nlookup(nd)) != 0) 3408 return (error); 3409 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3410 return (error); 3411 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3412 return (error); 3413 3414 /* 3415 * note: vget is required for any operation that might mod the vnode 3416 * so VINACTIVE is properly cleared. 3417 */ 3418 if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) { 3419 error = vget(vp, LK_EXCLUSIVE); 3420 if (error == 0) { 3421 error = setutimes(vp, &vattr, ts, (tptr == NULL)); 3422 vput(vp); 3423 } 3424 } 3425 vrele(vp); 3426 return (error); 3427 } 3428 3429 /* 3430 * utimes_args(char *path, struct timeval *tptr) 3431 * 3432 * Set the access and modification times of a file. 3433 * 3434 * MPALMOSTSAFE 3435 */ 3436 int 3437 sys_utimes(struct utimes_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 get_mplock(); 3449 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3450 if (error == 0) 3451 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3452 nlookup_done(&nd); 3453 rel_mplock(); 3454 return (error); 3455 } 3456 3457 /* 3458 * lutimes_args(char *path, struct timeval *tptr) 3459 * 3460 * Set the access and modification times of a file. 3461 * 3462 * MPALMOSTSAFE 3463 */ 3464 int 3465 sys_lutimes(struct lutimes_args *uap) 3466 { 3467 struct timeval tv[2]; 3468 struct nlookupdata nd; 3469 int error; 3470 3471 if (uap->tptr) { 3472 error = copyin(uap->tptr, tv, sizeof(tv)); 3473 if (error) 3474 return (error); 3475 } 3476 get_mplock(); 3477 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3478 if (error == 0) 3479 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3480 nlookup_done(&nd); 3481 rel_mplock(); 3482 return (error); 3483 } 3484 3485 /* 3486 * Set utimes on a file descriptor. The creds used to open the 3487 * file are used to determine whether the operation is allowed 3488 * or not. 3489 */ 3490 int 3491 kern_futimes(int fd, struct timeval *tptr) 3492 { 3493 struct thread *td = curthread; 3494 struct proc *p = td->td_proc; 3495 struct timespec ts[2]; 3496 struct file *fp; 3497 struct vnode *vp; 3498 struct vattr vattr; 3499 int error; 3500 3501 error = getutimes(tptr, ts); 3502 if (error) 3503 return (error); 3504 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 3505 return (error); 3506 if (fp->f_nchandle.ncp) 3507 error = ncp_writechk(&fp->f_nchandle); 3508 if (error == 0) { 3509 vp = fp->f_data; 3510 error = vget(vp, LK_EXCLUSIVE); 3511 if (error == 0) { 3512 error = VOP_GETATTR(vp, &vattr); 3513 if (error == 0) { 3514 error = naccess_va(&vattr, NLC_OWN | NLC_WRITE, 3515 fp->f_cred); 3516 } 3517 if (error == 0) { 3518 error = setutimes(vp, &vattr, ts, 3519 (tptr == NULL)); 3520 } 3521 vput(vp); 3522 } 3523 } 3524 fdrop(fp); 3525 return (error); 3526 } 3527 3528 /* 3529 * futimes_args(int fd, struct timeval *tptr) 3530 * 3531 * Set the access and modification times of a file. 3532 * 3533 * MPALMOSTSAFE 3534 */ 3535 int 3536 sys_futimes(struct futimes_args *uap) 3537 { 3538 struct timeval tv[2]; 3539 int error; 3540 3541 if (uap->tptr) { 3542 error = copyin(uap->tptr, tv, sizeof(tv)); 3543 if (error) 3544 return (error); 3545 } 3546 get_mplock(); 3547 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL); 3548 rel_mplock(); 3549 3550 return (error); 3551 } 3552 3553 int 3554 kern_truncate(struct nlookupdata *nd, off_t length) 3555 { 3556 struct vnode *vp; 3557 struct vattr vattr; 3558 int error; 3559 3560 if (length < 0) 3561 return(EINVAL); 3562 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE; 3563 if ((error = nlookup(nd)) != 0) 3564 return (error); 3565 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3566 return (error); 3567 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3568 return (error); 3569 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) { 3570 vrele(vp); 3571 return (error); 3572 } 3573 if (vp->v_type == VDIR) { 3574 error = EISDIR; 3575 } else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) { 3576 VATTR_NULL(&vattr); 3577 vattr.va_size = length; 3578 error = VOP_SETATTR(vp, &vattr, nd->nl_cred); 3579 } 3580 vput(vp); 3581 return (error); 3582 } 3583 3584 /* 3585 * truncate(char *path, int pad, off_t length) 3586 * 3587 * Truncate a file given its path name. 3588 * 3589 * MPALMOSTSAFE 3590 */ 3591 int 3592 sys_truncate(struct truncate_args *uap) 3593 { 3594 struct nlookupdata nd; 3595 int error; 3596 3597 get_mplock(); 3598 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3599 if (error == 0) 3600 error = kern_truncate(&nd, uap->length); 3601 nlookup_done(&nd); 3602 rel_mplock(); 3603 return error; 3604 } 3605 3606 int 3607 kern_ftruncate(int fd, off_t length) 3608 { 3609 struct thread *td = curthread; 3610 struct proc *p = td->td_proc; 3611 struct vattr vattr; 3612 struct vnode *vp; 3613 struct file *fp; 3614 int error; 3615 3616 if (length < 0) 3617 return(EINVAL); 3618 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 3619 return (error); 3620 if (fp->f_nchandle.ncp) { 3621 error = ncp_writechk(&fp->f_nchandle); 3622 if (error) 3623 goto done; 3624 } 3625 if ((fp->f_flag & FWRITE) == 0) { 3626 error = EINVAL; 3627 goto done; 3628 } 3629 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */ 3630 error = EINVAL; 3631 goto done; 3632 } 3633 vp = (struct vnode *)fp->f_data; 3634 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3635 if (vp->v_type == VDIR) { 3636 error = EISDIR; 3637 } else if ((error = vn_writechk(vp, NULL)) == 0) { 3638 VATTR_NULL(&vattr); 3639 vattr.va_size = length; 3640 error = VOP_SETATTR(vp, &vattr, fp->f_cred); 3641 } 3642 vn_unlock(vp); 3643 done: 3644 fdrop(fp); 3645 return (error); 3646 } 3647 3648 /* 3649 * ftruncate_args(int fd, int pad, off_t length) 3650 * 3651 * Truncate a file given a file descriptor. 3652 * 3653 * MPALMOSTSAFE 3654 */ 3655 int 3656 sys_ftruncate(struct ftruncate_args *uap) 3657 { 3658 int error; 3659 3660 get_mplock(); 3661 error = kern_ftruncate(uap->fd, uap->length); 3662 rel_mplock(); 3663 3664 return (error); 3665 } 3666 3667 /* 3668 * fsync(int fd) 3669 * 3670 * Sync an open file. 3671 * 3672 * MPALMOSTSAFE 3673 */ 3674 int 3675 sys_fsync(struct fsync_args *uap) 3676 { 3677 struct thread *td = curthread; 3678 struct proc *p = td->td_proc; 3679 struct vnode *vp; 3680 struct file *fp; 3681 vm_object_t obj; 3682 int error; 3683 3684 if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0) 3685 return (error); 3686 get_mplock(); 3687 vp = (struct vnode *)fp->f_data; 3688 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3689 if ((obj = vp->v_object) != NULL) 3690 vm_object_page_clean(obj, 0, 0, 0); 3691 error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL); 3692 if (error == 0 && vp->v_mount) 3693 error = buf_fsync(vp); 3694 vn_unlock(vp); 3695 rel_mplock(); 3696 fdrop(fp); 3697 3698 return (error); 3699 } 3700 3701 int 3702 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond) 3703 { 3704 struct nchandle fnchd; 3705 struct nchandle tnchd; 3706 struct namecache *ncp; 3707 struct vnode *fdvp; 3708 struct vnode *tdvp; 3709 struct mount *mp; 3710 int error; 3711 3712 bwillinode(1); 3713 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC; 3714 if ((error = nlookup(fromnd)) != 0) 3715 return (error); 3716 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL) 3717 return (ENOENT); 3718 fnchd.mount = fromnd->nl_nch.mount; 3719 cache_hold(&fnchd); 3720 3721 /* 3722 * unlock the source nch so we can lookup the target nch without 3723 * deadlocking. The target may or may not exist so we do not check 3724 * for a target vp like kern_mkdir() and other creation functions do. 3725 * 3726 * The source and target directories are ref'd and rechecked after 3727 * everything is relocked to determine if the source or target file 3728 * has been renamed. 3729 */ 3730 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED); 3731 fromnd->nl_flags &= ~NLC_NCPISLOCKED; 3732 cache_unlock(&fromnd->nl_nch); 3733 3734 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP; 3735 if ((error = nlookup(tond)) != 0) { 3736 cache_drop(&fnchd); 3737 return (error); 3738 } 3739 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) { 3740 cache_drop(&fnchd); 3741 return (ENOENT); 3742 } 3743 tnchd.mount = tond->nl_nch.mount; 3744 cache_hold(&tnchd); 3745 3746 /* 3747 * If the source and target are the same there is nothing to do 3748 */ 3749 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) { 3750 cache_drop(&fnchd); 3751 cache_drop(&tnchd); 3752 return (0); 3753 } 3754 3755 /* 3756 * Mount points cannot be renamed or overwritten 3757 */ 3758 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) & 3759 NCF_ISMOUNTPT 3760 ) { 3761 cache_drop(&fnchd); 3762 cache_drop(&tnchd); 3763 return (EINVAL); 3764 } 3765 3766 /* 3767 * Relock the source ncp. cache_relock() will deal with any 3768 * deadlocks against the already-locked tond and will also 3769 * make sure both are resolved. 3770 * 3771 * NOTE AFTER RELOCKING: The source or target ncp may have become 3772 * invalid while they were unlocked, nc_vp and nc_mount could 3773 * be NULL. 3774 */ 3775 cache_relock(&fromnd->nl_nch, fromnd->nl_cred, 3776 &tond->nl_nch, tond->nl_cred); 3777 fromnd->nl_flags |= NLC_NCPISLOCKED; 3778 3779 /* 3780 * make sure the parent directories linkages are the same 3781 */ 3782 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent || 3783 tnchd.ncp != tond->nl_nch.ncp->nc_parent) { 3784 cache_drop(&fnchd); 3785 cache_drop(&tnchd); 3786 return (ENOENT); 3787 } 3788 3789 /* 3790 * Both the source and target must be within the same filesystem and 3791 * in the same filesystem as their parent directories within the 3792 * namecache topology. 3793 * 3794 * NOTE: fromnd's nc_mount or nc_vp could be NULL. 3795 */ 3796 mp = fnchd.mount; 3797 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount || 3798 mp != tond->nl_nch.mount) { 3799 cache_drop(&fnchd); 3800 cache_drop(&tnchd); 3801 return (EXDEV); 3802 } 3803 3804 /* 3805 * Make sure the mount point is writable 3806 */ 3807 if ((error = ncp_writechk(&tond->nl_nch)) != 0) { 3808 cache_drop(&fnchd); 3809 cache_drop(&tnchd); 3810 return (error); 3811 } 3812 3813 /* 3814 * If the target exists and either the source or target is a directory, 3815 * then both must be directories. 3816 * 3817 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h 3818 * have become NULL. 3819 */ 3820 if (tond->nl_nch.ncp->nc_vp) { 3821 if (fromnd->nl_nch.ncp->nc_vp == NULL) { 3822 error = ENOENT; 3823 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 3824 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR) 3825 error = ENOTDIR; 3826 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) { 3827 error = EISDIR; 3828 } 3829 } 3830 3831 /* 3832 * You cannot rename a source into itself or a subdirectory of itself. 3833 * We check this by travsersing the target directory upwards looking 3834 * for a match against the source. 3835 * 3836 * XXX MPSAFE 3837 */ 3838 if (error == 0) { 3839 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) { 3840 if (fromnd->nl_nch.ncp == ncp) { 3841 error = EINVAL; 3842 break; 3843 } 3844 } 3845 } 3846 3847 cache_drop(&fnchd); 3848 cache_drop(&tnchd); 3849 3850 /* 3851 * Even though the namespaces are different, they may still represent 3852 * hardlinks to the same file. The filesystem might have a hard time 3853 * with this so we issue a NREMOVE of the source instead of a NRENAME 3854 * when we detect the situation. 3855 */ 3856 if (error == 0) { 3857 fdvp = fromnd->nl_dvp; 3858 tdvp = tond->nl_dvp; 3859 if (fdvp == NULL || tdvp == NULL) { 3860 error = EPERM; 3861 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) { 3862 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp, 3863 fromnd->nl_cred); 3864 } else { 3865 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch, 3866 fdvp, tdvp, tond->nl_cred); 3867 } 3868 } 3869 return (error); 3870 } 3871 3872 /* 3873 * rename_args(char *from, char *to) 3874 * 3875 * Rename files. Source and destination must either both be directories, 3876 * or both not be directories. If target is a directory, it must be empty. 3877 * 3878 * MPALMOSTSAFE 3879 */ 3880 int 3881 sys_rename(struct rename_args *uap) 3882 { 3883 struct nlookupdata fromnd, tond; 3884 int error; 3885 3886 get_mplock(); 3887 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0); 3888 if (error == 0) { 3889 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0); 3890 if (error == 0) 3891 error = kern_rename(&fromnd, &tond); 3892 nlookup_done(&tond); 3893 } 3894 nlookup_done(&fromnd); 3895 rel_mplock(); 3896 return (error); 3897 } 3898 3899 /* 3900 * renameat_args(int oldfd, char *old, int newfd, char *new) 3901 * 3902 * Rename files using paths relative to the directories associated with 3903 * oldfd and newfd. Source and destination must either both be directories, 3904 * or both not be directories. If target is a directory, it must be empty. 3905 * 3906 * MPALMOSTSAFE 3907 */ 3908 int 3909 sys_renameat(struct renameat_args *uap) 3910 { 3911 struct nlookupdata oldnd, newnd; 3912 struct file *oldfp, *newfp; 3913 int error; 3914 3915 get_mplock(); 3916 error = nlookup_init_at(&oldnd, &oldfp, uap->oldfd, uap->old, 3917 UIO_USERSPACE, 0); 3918 if (error == 0) { 3919 error = nlookup_init_at(&newnd, &newfp, uap->newfd, uap->new, 3920 UIO_USERSPACE, 0); 3921 if (error == 0) 3922 error = kern_rename(&oldnd, &newnd); 3923 nlookup_done_at(&newnd, newfp); 3924 } 3925 nlookup_done_at(&oldnd, oldfp); 3926 rel_mplock(); 3927 return (error); 3928 } 3929 3930 int 3931 kern_mkdir(struct nlookupdata *nd, int mode) 3932 { 3933 struct thread *td = curthread; 3934 struct proc *p = td->td_proc; 3935 struct vnode *vp; 3936 struct vattr vattr; 3937 int error; 3938 3939 bwillinode(1); 3940 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP; 3941 if ((error = nlookup(nd)) != 0) 3942 return (error); 3943 3944 if (nd->nl_nch.ncp->nc_vp) 3945 return (EEXIST); 3946 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3947 return (error); 3948 VATTR_NULL(&vattr); 3949 vattr.va_type = VDIR; 3950 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask; 3951 3952 vp = NULL; 3953 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr); 3954 if (error == 0) 3955 vput(vp); 3956 return (error); 3957 } 3958 3959 /* 3960 * mkdir_args(char *path, int mode) 3961 * 3962 * Make a directory file. 3963 * 3964 * MPALMOSTSAFE 3965 */ 3966 int 3967 sys_mkdir(struct mkdir_args *uap) 3968 { 3969 struct nlookupdata nd; 3970 int error; 3971 3972 get_mplock(); 3973 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3974 if (error == 0) 3975 error = kern_mkdir(&nd, uap->mode); 3976 nlookup_done(&nd); 3977 rel_mplock(); 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 * MPALMOSTSAFE 3988 */ 3989 int 3990 sys_mkdirat(struct mkdirat_args *uap) 3991 { 3992 struct nlookupdata nd; 3993 struct file *fp; 3994 int error; 3995 3996 get_mplock(); 3997 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 3998 if (error == 0) 3999 error = kern_mkdir(&nd, uap->mode); 4000 nlookup_done_at(&nd, fp); 4001 rel_mplock(); 4002 return (error); 4003 } 4004 4005 int 4006 kern_rmdir(struct nlookupdata *nd) 4007 { 4008 int error; 4009 4010 bwillinode(1); 4011 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 4012 if ((error = nlookup(nd)) != 0) 4013 return (error); 4014 4015 /* 4016 * Do not allow directories representing mount points to be 4017 * deleted, even if empty. Check write perms on mount point 4018 * in case the vnode is aliased (aka nullfs). 4019 */ 4020 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT)) 4021 return (EINVAL); 4022 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 4023 return (error); 4024 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 4025 return (error); 4026 } 4027 4028 /* 4029 * rmdir_args(char *path) 4030 * 4031 * Remove a directory file. 4032 * 4033 * MPALMOSTSAFE 4034 */ 4035 int 4036 sys_rmdir(struct rmdir_args *uap) 4037 { 4038 struct nlookupdata nd; 4039 int error; 4040 4041 get_mplock(); 4042 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 4043 if (error == 0) 4044 error = kern_rmdir(&nd); 4045 nlookup_done(&nd); 4046 rel_mplock(); 4047 return (error); 4048 } 4049 4050 int 4051 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res, 4052 enum uio_seg direction) 4053 { 4054 struct thread *td = curthread; 4055 struct proc *p = td->td_proc; 4056 struct vnode *vp; 4057 struct file *fp; 4058 struct uio auio; 4059 struct iovec aiov; 4060 off_t loff; 4061 int error, eofflag; 4062 4063 if ((error = holdvnode(p->p_fd, fd, &fp)) != 0) 4064 return (error); 4065 if ((fp->f_flag & FREAD) == 0) { 4066 error = EBADF; 4067 goto done; 4068 } 4069 vp = (struct vnode *)fp->f_data; 4070 unionread: 4071 if (vp->v_type != VDIR) { 4072 error = EINVAL; 4073 goto done; 4074 } 4075 aiov.iov_base = buf; 4076 aiov.iov_len = count; 4077 auio.uio_iov = &aiov; 4078 auio.uio_iovcnt = 1; 4079 auio.uio_rw = UIO_READ; 4080 auio.uio_segflg = direction; 4081 auio.uio_td = td; 4082 auio.uio_resid = count; 4083 loff = auio.uio_offset = fp->f_offset; 4084 error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); 4085 fp->f_offset = auio.uio_offset; 4086 if (error) 4087 goto done; 4088 if (count == auio.uio_resid) { 4089 if (union_dircheckp) { 4090 error = union_dircheckp(td, &vp, fp); 4091 if (error == -1) 4092 goto unionread; 4093 if (error) 4094 goto done; 4095 } 4096 #if 0 4097 if ((vp->v_flag & VROOT) && 4098 (vp->v_mount->mnt_flag & MNT_UNION)) { 4099 struct vnode *tvp = vp; 4100 vp = vp->v_mount->mnt_vnodecovered; 4101 vref(vp); 4102 fp->f_data = vp; 4103 fp->f_offset = 0; 4104 vrele(tvp); 4105 goto unionread; 4106 } 4107 #endif 4108 } 4109 4110 /* 4111 * WARNING! *basep may not be wide enough to accomodate the 4112 * seek offset. XXX should we hack this to return the upper 32 bits 4113 * for offsets greater then 4G? 4114 */ 4115 if (basep) { 4116 *basep = (long)loff; 4117 } 4118 *res = count - auio.uio_resid; 4119 done: 4120 fdrop(fp); 4121 return (error); 4122 } 4123 4124 /* 4125 * getdirentries_args(int fd, char *buf, u_int conut, long *basep) 4126 * 4127 * Read a block of directory entries in a file system independent format. 4128 * 4129 * MPALMOSTSAFE 4130 */ 4131 int 4132 sys_getdirentries(struct getdirentries_args *uap) 4133 { 4134 long base; 4135 int error; 4136 4137 get_mplock(); 4138 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base, 4139 &uap->sysmsg_result, UIO_USERSPACE); 4140 rel_mplock(); 4141 4142 if (error == 0 && uap->basep) 4143 error = copyout(&base, uap->basep, sizeof(*uap->basep)); 4144 return (error); 4145 } 4146 4147 /* 4148 * getdents_args(int fd, char *buf, size_t count) 4149 * 4150 * MPALMOSTSAFE 4151 */ 4152 int 4153 sys_getdents(struct getdents_args *uap) 4154 { 4155 int error; 4156 4157 get_mplock(); 4158 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL, 4159 &uap->sysmsg_result, UIO_USERSPACE); 4160 rel_mplock(); 4161 4162 return (error); 4163 } 4164 4165 /* 4166 * Set the mode mask for creation of filesystem nodes. 4167 * 4168 * umask(int newmask) 4169 * 4170 * MPSAFE 4171 */ 4172 int 4173 sys_umask(struct umask_args *uap) 4174 { 4175 struct thread *td = curthread; 4176 struct proc *p = td->td_proc; 4177 struct filedesc *fdp; 4178 4179 fdp = p->p_fd; 4180 uap->sysmsg_result = fdp->fd_cmask; 4181 fdp->fd_cmask = uap->newmask & ALLPERMS; 4182 return (0); 4183 } 4184 4185 /* 4186 * revoke(char *path) 4187 * 4188 * Void all references to file by ripping underlying filesystem 4189 * away from vnode. 4190 * 4191 * MPALMOSTSAFE 4192 */ 4193 int 4194 sys_revoke(struct revoke_args *uap) 4195 { 4196 struct nlookupdata nd; 4197 struct vattr vattr; 4198 struct vnode *vp; 4199 struct ucred *cred; 4200 int error; 4201 4202 vp = NULL; 4203 get_mplock(); 4204 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4205 if (error == 0) 4206 error = nlookup(&nd); 4207 if (error == 0) 4208 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 4209 cred = crhold(nd.nl_cred); 4210 nlookup_done(&nd); 4211 if (error == 0) { 4212 if (error == 0) 4213 error = VOP_GETATTR(vp, &vattr); 4214 if (error == 0 && cred->cr_uid != vattr.va_uid) 4215 error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0); 4216 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) { 4217 if (vcount(vp) > 0) 4218 error = vrevoke(vp, cred); 4219 } else if (error == 0) { 4220 error = vrevoke(vp, cred); 4221 } 4222 vrele(vp); 4223 } 4224 if (cred) 4225 crfree(cred); 4226 rel_mplock(); 4227 return (error); 4228 } 4229 4230 /* 4231 * getfh_args(char *fname, fhandle_t *fhp) 4232 * 4233 * Get (NFS) file handle 4234 * 4235 * NOTE: We use the fsid of the covering mount, even if it is a nullfs 4236 * mount. This allows nullfs mounts to be explicitly exported. 4237 * 4238 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe. 4239 * 4240 * nullfs mounts of subdirectories are not safe. That is, it will 4241 * work, but you do not really have protection against access to 4242 * the related parent directories. 4243 * 4244 * MPALMOSTSAFE 4245 */ 4246 int 4247 sys_getfh(struct getfh_args *uap) 4248 { 4249 struct thread *td = curthread; 4250 struct nlookupdata nd; 4251 fhandle_t fh; 4252 struct vnode *vp; 4253 struct mount *mp; 4254 int error; 4255 4256 /* 4257 * Must be super user 4258 */ 4259 if ((error = priv_check(td, PRIV_ROOT)) != 0) 4260 return (error); 4261 4262 vp = NULL; 4263 get_mplock(); 4264 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 4265 if (error == 0) 4266 error = nlookup(&nd); 4267 if (error == 0) 4268 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4269 mp = nd.nl_nch.mount; 4270 nlookup_done(&nd); 4271 if (error == 0) { 4272 bzero(&fh, sizeof(fh)); 4273 fh.fh_fsid = mp->mnt_stat.f_fsid; 4274 error = VFS_VPTOFH(vp, &fh.fh_fid); 4275 vput(vp); 4276 if (error == 0) 4277 error = copyout(&fh, uap->fhp, sizeof(fh)); 4278 } 4279 rel_mplock(); 4280 return (error); 4281 } 4282 4283 /* 4284 * fhopen_args(const struct fhandle *u_fhp, int flags) 4285 * 4286 * syscall for the rpc.lockd to use to translate a NFS file handle into 4287 * an open descriptor. 4288 * 4289 * warning: do not remove the priv_check() call or this becomes one giant 4290 * security hole. 4291 * 4292 * MPALMOSTSAFE 4293 */ 4294 int 4295 sys_fhopen(struct fhopen_args *uap) 4296 { 4297 struct thread *td = curthread; 4298 struct filedesc *fdp = td->td_proc->p_fd; 4299 struct mount *mp; 4300 struct vnode *vp; 4301 struct fhandle fhp; 4302 struct vattr vat; 4303 struct vattr *vap = &vat; 4304 struct flock lf; 4305 int fmode, mode, error, type; 4306 struct file *nfp; 4307 struct file *fp; 4308 int indx; 4309 4310 /* 4311 * Must be super user 4312 */ 4313 error = priv_check(td, PRIV_ROOT); 4314 if (error) 4315 return (error); 4316 4317 fmode = FFLAGS(uap->flags); 4318 4319 /* 4320 * Why not allow a non-read/write open for our lockd? 4321 */ 4322 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) 4323 return (EINVAL); 4324 error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); 4325 if (error) 4326 return(error); 4327 4328 /* 4329 * Find the mount point 4330 */ 4331 get_mplock(); 4332 mp = vfs_getvfs(&fhp.fh_fsid); 4333 if (mp == NULL) { 4334 error = ESTALE; 4335 goto done; 4336 } 4337 /* now give me my vnode, it gets returned to me locked */ 4338 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp); 4339 if (error) 4340 goto done; 4341 /* 4342 * from now on we have to make sure not 4343 * to forget about the vnode 4344 * any error that causes an abort must vput(vp) 4345 * just set error = err and 'goto bad;'. 4346 */ 4347 4348 /* 4349 * from vn_open 4350 */ 4351 if (vp->v_type == VLNK) { 4352 error = EMLINK; 4353 goto bad; 4354 } 4355 if (vp->v_type == VSOCK) { 4356 error = EOPNOTSUPP; 4357 goto bad; 4358 } 4359 mode = 0; 4360 if (fmode & (FWRITE | O_TRUNC)) { 4361 if (vp->v_type == VDIR) { 4362 error = EISDIR; 4363 goto bad; 4364 } 4365 error = vn_writechk(vp, NULL); 4366 if (error) 4367 goto bad; 4368 mode |= VWRITE; 4369 } 4370 if (fmode & FREAD) 4371 mode |= VREAD; 4372 if (mode) { 4373 error = VOP_ACCESS(vp, mode, td->td_ucred); 4374 if (error) 4375 goto bad; 4376 } 4377 if (fmode & O_TRUNC) { 4378 vn_unlock(vp); /* XXX */ 4379 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 4380 VATTR_NULL(vap); 4381 vap->va_size = 0; 4382 error = VOP_SETATTR(vp, vap, td->td_ucred); 4383 if (error) 4384 goto bad; 4385 } 4386 4387 /* 4388 * VOP_OPEN needs the file pointer so it can potentially override 4389 * it. 4390 * 4391 * WARNING! no f_nchandle will be associated when fhopen()ing a 4392 * directory. XXX 4393 */ 4394 if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0) 4395 goto bad; 4396 fp = nfp; 4397 4398 error = VOP_OPEN(vp, fmode, td->td_ucred, fp); 4399 if (error) { 4400 /* 4401 * setting f_ops this way prevents VOP_CLOSE from being 4402 * called or fdrop() releasing the vp from v_data. Since 4403 * the VOP_OPEN failed we don't want to VOP_CLOSE. 4404 */ 4405 fp->f_ops = &badfileops; 4406 fp->f_data = NULL; 4407 goto bad_drop; 4408 } 4409 4410 /* 4411 * The fp is given its own reference, we still have our ref and lock. 4412 * 4413 * Assert that all regular files must be created with a VM object. 4414 */ 4415 if (vp->v_type == VREG && vp->v_object == NULL) { 4416 kprintf("fhopen: regular file did not have VM object: %p\n", vp); 4417 goto bad_drop; 4418 } 4419 4420 /* 4421 * The open was successful. Handle any locking requirements. 4422 */ 4423 if (fmode & (O_EXLOCK | O_SHLOCK)) { 4424 lf.l_whence = SEEK_SET; 4425 lf.l_start = 0; 4426 lf.l_len = 0; 4427 if (fmode & O_EXLOCK) 4428 lf.l_type = F_WRLCK; 4429 else 4430 lf.l_type = F_RDLCK; 4431 if (fmode & FNONBLOCK) 4432 type = 0; 4433 else 4434 type = F_WAIT; 4435 vn_unlock(vp); 4436 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) { 4437 /* 4438 * release our private reference. 4439 */ 4440 fsetfd(fdp, NULL, indx); 4441 fdrop(fp); 4442 vrele(vp); 4443 goto done; 4444 } 4445 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4446 fp->f_flag |= FHASLOCK; 4447 } 4448 4449 /* 4450 * Clean up. Associate the file pointer with the previously 4451 * reserved descriptor and return it. 4452 */ 4453 vput(vp); 4454 rel_mplock(); 4455 fsetfd(fdp, fp, indx); 4456 fdrop(fp); 4457 uap->sysmsg_result = indx; 4458 return (0); 4459 4460 bad_drop: 4461 fsetfd(fdp, NULL, indx); 4462 fdrop(fp); 4463 bad: 4464 vput(vp); 4465 done: 4466 rel_mplock(); 4467 return (error); 4468 } 4469 4470 /* 4471 * fhstat_args(struct fhandle *u_fhp, struct stat *sb) 4472 * 4473 * MPALMOSTSAFE 4474 */ 4475 int 4476 sys_fhstat(struct fhstat_args *uap) 4477 { 4478 struct thread *td = curthread; 4479 struct stat sb; 4480 fhandle_t fh; 4481 struct mount *mp; 4482 struct vnode *vp; 4483 int error; 4484 4485 /* 4486 * Must be super user 4487 */ 4488 error = priv_check(td, PRIV_ROOT); 4489 if (error) 4490 return (error); 4491 4492 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); 4493 if (error) 4494 return (error); 4495 4496 get_mplock(); 4497 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 4498 error = ESTALE; 4499 if (error == 0) { 4500 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) { 4501 error = vn_stat(vp, &sb, td->td_ucred); 4502 vput(vp); 4503 } 4504 } 4505 rel_mplock(); 4506 if (error == 0) 4507 error = copyout(&sb, uap->sb, sizeof(sb)); 4508 return (error); 4509 } 4510 4511 /* 4512 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf) 4513 * 4514 * MPALMOSTSAFE 4515 */ 4516 int 4517 sys_fhstatfs(struct fhstatfs_args *uap) 4518 { 4519 struct thread *td = curthread; 4520 struct proc *p = td->td_proc; 4521 struct statfs *sp; 4522 struct mount *mp; 4523 struct vnode *vp; 4524 struct statfs sb; 4525 char *fullpath, *freepath; 4526 fhandle_t fh; 4527 int error; 4528 4529 /* 4530 * Must be super user 4531 */ 4532 if ((error = priv_check(td, PRIV_ROOT))) 4533 return (error); 4534 4535 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4536 return (error); 4537 4538 get_mplock(); 4539 4540 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4541 error = ESTALE; 4542 goto done; 4543 } 4544 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4545 error = ESTALE; 4546 goto done; 4547 } 4548 4549 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0) 4550 goto done; 4551 mp = vp->v_mount; 4552 sp = &mp->mnt_stat; 4553 vput(vp); 4554 if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0) 4555 goto done; 4556 4557 error = mount_path(p, mp, &fullpath, &freepath); 4558 if (error) 4559 goto done; 4560 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 4561 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 4562 kfree(freepath, M_TEMP); 4563 4564 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 4565 if (priv_check(td, PRIV_ROOT)) { 4566 bcopy(sp, &sb, sizeof(sb)); 4567 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; 4568 sp = &sb; 4569 } 4570 error = copyout(sp, uap->buf, sizeof(*sp)); 4571 done: 4572 rel_mplock(); 4573 return (error); 4574 } 4575 4576 /* 4577 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf) 4578 * 4579 * MPALMOSTSAFE 4580 */ 4581 int 4582 sys_fhstatvfs(struct fhstatvfs_args *uap) 4583 { 4584 struct thread *td = curthread; 4585 struct proc *p = td->td_proc; 4586 struct statvfs *sp; 4587 struct mount *mp; 4588 struct vnode *vp; 4589 fhandle_t fh; 4590 int error; 4591 4592 /* 4593 * Must be super user 4594 */ 4595 if ((error = priv_check(td, PRIV_ROOT))) 4596 return (error); 4597 4598 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4599 return (error); 4600 4601 get_mplock(); 4602 4603 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4604 error = ESTALE; 4605 goto done; 4606 } 4607 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4608 error = ESTALE; 4609 goto done; 4610 } 4611 4612 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 4613 goto done; 4614 mp = vp->v_mount; 4615 sp = &mp->mnt_vstat; 4616 vput(vp); 4617 if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0) 4618 goto done; 4619 4620 sp->f_flag = 0; 4621 if (mp->mnt_flag & MNT_RDONLY) 4622 sp->f_flag |= ST_RDONLY; 4623 if (mp->mnt_flag & MNT_NOSUID) 4624 sp->f_flag |= ST_NOSUID; 4625 error = copyout(sp, uap->buf, sizeof(*sp)); 4626 done: 4627 rel_mplock(); 4628 return (error); 4629 } 4630 4631 4632 /* 4633 * Syscall to push extended attribute configuration information into the 4634 * VFS. Accepts a path, which it converts to a mountpoint, as well as 4635 * a command (int cmd), and attribute name and misc data. For now, the 4636 * attribute name is left in userspace for consumption by the VFS_op. 4637 * It will probably be changed to be copied into sysspace by the 4638 * syscall in the future, once issues with various consumers of the 4639 * attribute code have raised their hands. 4640 * 4641 * Currently this is used only by UFS Extended Attributes. 4642 * 4643 * MPALMOSTSAFE 4644 */ 4645 int 4646 sys_extattrctl(struct extattrctl_args *uap) 4647 { 4648 struct nlookupdata nd; 4649 struct vnode *vp; 4650 char attrname[EXTATTR_MAXNAMELEN]; 4651 int error; 4652 size_t size; 4653 4654 get_mplock(); 4655 4656 attrname[0] = 0; 4657 vp = NULL; 4658 error = 0; 4659 4660 if (error == 0 && uap->filename) { 4661 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE, 4662 NLC_FOLLOW); 4663 if (error == 0) 4664 error = nlookup(&nd); 4665 if (error == 0) 4666 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 4667 nlookup_done(&nd); 4668 } 4669 4670 if (error == 0 && uap->attrname) { 4671 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN, 4672 &size); 4673 } 4674 4675 if (error == 0) { 4676 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4677 if (error == 0) 4678 error = nlookup(&nd); 4679 if (error == 0) 4680 error = ncp_writechk(&nd.nl_nch); 4681 if (error == 0) { 4682 error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp, 4683 uap->attrnamespace, 4684 uap->attrname, nd.nl_cred); 4685 } 4686 nlookup_done(&nd); 4687 } 4688 4689 rel_mplock(); 4690 4691 return (error); 4692 } 4693 4694 /* 4695 * Syscall to get a named extended attribute on a file or directory. 4696 * 4697 * MPALMOSTSAFE 4698 */ 4699 int 4700 sys_extattr_set_file(struct extattr_set_file_args *uap) 4701 { 4702 char attrname[EXTATTR_MAXNAMELEN]; 4703 struct nlookupdata nd; 4704 struct vnode *vp; 4705 struct uio auio; 4706 struct iovec aiov; 4707 int error; 4708 4709 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4710 if (error) 4711 return (error); 4712 4713 vp = NULL; 4714 get_mplock(); 4715 4716 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4717 if (error == 0) 4718 error = nlookup(&nd); 4719 if (error == 0) 4720 error = ncp_writechk(&nd.nl_nch); 4721 if (error == 0) 4722 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4723 if (error) { 4724 nlookup_done(&nd); 4725 rel_mplock(); 4726 return (error); 4727 } 4728 4729 bzero(&auio, sizeof(auio)); 4730 aiov.iov_base = uap->data; 4731 aiov.iov_len = uap->nbytes; 4732 auio.uio_iov = &aiov; 4733 auio.uio_iovcnt = 1; 4734 auio.uio_offset = 0; 4735 auio.uio_resid = uap->nbytes; 4736 auio.uio_rw = UIO_WRITE; 4737 auio.uio_td = curthread; 4738 4739 error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname, 4740 &auio, nd.nl_cred); 4741 4742 vput(vp); 4743 nlookup_done(&nd); 4744 rel_mplock(); 4745 return (error); 4746 } 4747 4748 /* 4749 * Syscall to get a named extended attribute on a file or directory. 4750 * 4751 * MPALMOSTSAFE 4752 */ 4753 int 4754 sys_extattr_get_file(struct extattr_get_file_args *uap) 4755 { 4756 char attrname[EXTATTR_MAXNAMELEN]; 4757 struct nlookupdata nd; 4758 struct uio auio; 4759 struct iovec aiov; 4760 struct vnode *vp; 4761 int error; 4762 4763 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4764 if (error) 4765 return (error); 4766 4767 vp = NULL; 4768 get_mplock(); 4769 4770 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4771 if (error == 0) 4772 error = nlookup(&nd); 4773 if (error == 0) 4774 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4775 if (error) { 4776 nlookup_done(&nd); 4777 rel_mplock(); 4778 return (error); 4779 } 4780 4781 bzero(&auio, sizeof(auio)); 4782 aiov.iov_base = uap->data; 4783 aiov.iov_len = uap->nbytes; 4784 auio.uio_iov = &aiov; 4785 auio.uio_iovcnt = 1; 4786 auio.uio_offset = 0; 4787 auio.uio_resid = uap->nbytes; 4788 auio.uio_rw = UIO_READ; 4789 auio.uio_td = curthread; 4790 4791 error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname, 4792 &auio, nd.nl_cred); 4793 uap->sysmsg_result = uap->nbytes - auio.uio_resid; 4794 4795 vput(vp); 4796 nlookup_done(&nd); 4797 rel_mplock(); 4798 return(error); 4799 } 4800 4801 /* 4802 * Syscall to delete a named extended attribute from a file or directory. 4803 * Accepts attribute name. The real work happens in VOP_SETEXTATTR(). 4804 * 4805 * MPALMOSTSAFE 4806 */ 4807 int 4808 sys_extattr_delete_file(struct extattr_delete_file_args *uap) 4809 { 4810 char attrname[EXTATTR_MAXNAMELEN]; 4811 struct nlookupdata nd; 4812 struct vnode *vp; 4813 int error; 4814 4815 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 4816 if (error) 4817 return(error); 4818 4819 get_mplock(); 4820 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4821 if (error == 0) 4822 error = nlookup(&nd); 4823 if (error == 0) 4824 error = ncp_writechk(&nd.nl_nch); 4825 if (error == 0) { 4826 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4827 if (error == 0) { 4828 error = VOP_SETEXTATTR(vp, uap->attrnamespace, 4829 attrname, NULL, nd.nl_cred); 4830 vput(vp); 4831 } 4832 } 4833 nlookup_done(&nd); 4834 rel_mplock(); 4835 return(error); 4836 } 4837 4838 /* 4839 * Determine if the mount is visible to the process. 4840 */ 4841 static int 4842 chroot_visible_mnt(struct mount *mp, struct proc *p) 4843 { 4844 struct nchandle nch; 4845 4846 /* 4847 * Traverse from the mount point upwards. If we hit the process 4848 * root then the mount point is visible to the process. 4849 */ 4850 nch = mp->mnt_ncmountpt; 4851 while (nch.ncp) { 4852 if (nch.mount == p->p_fd->fd_nrdir.mount && 4853 nch.ncp == p->p_fd->fd_nrdir.ncp) { 4854 return(1); 4855 } 4856 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) { 4857 nch = nch.mount->mnt_ncmounton; 4858 } else { 4859 nch.ncp = nch.ncp->nc_parent; 4860 } 4861 } 4862 4863 /* 4864 * If the mount point is not visible to the process, but the 4865 * process root is in a subdirectory of the mount, return 4866 * TRUE anyway. 4867 */ 4868 if (p->p_fd->fd_nrdir.mount == mp) 4869 return(1); 4870 4871 return(0); 4872 } 4873 4874