1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 35 * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/buf.h> 41 #include <sys/conf.h> 42 #include <sys/sysent.h> 43 #include <sys/malloc.h> 44 #include <sys/mount.h> 45 #include <sys/mountctl.h> 46 #include <sys/sysmsg.h> 47 #include <sys/filedesc.h> 48 #include <sys/kernel.h> 49 #include <sys/fcntl.h> 50 #include <sys/file.h> 51 #include <sys/linker.h> 52 #include <sys/stat.h> 53 #include <sys/unistd.h> 54 #include <sys/vnode.h> 55 #include <sys/proc.h> 56 #include <sys/priv.h> 57 #include <sys/jail.h> 58 #include <sys/namei.h> 59 #include <sys/nlookup.h> 60 #include <sys/dirent.h> 61 #include <sys/extattr.h> 62 #include <sys/spinlock.h> 63 #include <sys/kern_syscall.h> 64 #include <sys/objcache.h> 65 #include <sys/sysctl.h> 66 67 #include <sys/buf2.h> 68 #include <sys/file2.h> 69 #include <sys/spinlock2.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_page.h> 74 75 #include <machine/limits.h> 76 #include <machine/stdarg.h> 77 78 static void mount_warning(struct mount *mp, const char *ctl, ...) 79 __printflike(2, 3); 80 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb); 81 static int checkvp_chdir (struct vnode *vn, struct thread *td); 82 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch); 83 static int get_fspriv(const char *); 84 static int chroot_refuse_vdir_fds (thread_t td, struct filedesc *fdp); 85 static int chroot_visible_mnt(struct mount *mp, struct proc *p); 86 static int getutimes (struct timeval *, struct timespec *); 87 static int getutimens (const struct timespec *, struct timespec *, int *); 88 static int setfown (struct mount *, struct vnode *, uid_t, gid_t); 89 static int setfmode (struct vnode *, int); 90 static int setfflags (struct vnode *, u_long); 91 static int setutimes (struct vnode *, struct vattr *, 92 const struct timespec *, int); 93 94 static int usermount = 0; /* if 1, non-root can mount fs. */ 95 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, 96 "Allow non-root users to mount filesystems"); 97 98 static int debug_unmount = 0; /* if 1 loop until unmount success */ 99 SYSCTL_INT(_vfs, OID_AUTO, debug_unmount, CTLFLAG_RW, &debug_unmount, 0, 100 "Stall failed unmounts in loop"); 101 102 static struct krate krate_rename = { 1 }; 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 sysmsg *sysmsg, const 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 int priv = 0; 126 int flags = uap->flags; 127 struct vattr va; 128 struct nlookupdata nd; 129 char fstypename[MFSNAMELEN]; 130 struct ucred *cred; 131 132 cred = td->td_ucred; 133 134 /* We do not allow user mounts inside a jail for now */ 135 if (usermount && jailed(cred)) { 136 error = EPERM; 137 goto done; 138 } 139 140 /* 141 * Extract the file system type. We need to know this early, to take 142 * appropriate actions for jails and nullfs mounts. 143 */ 144 if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) 145 goto done; 146 147 /* 148 * Select the correct priv according to the file system type. 149 */ 150 priv = get_fspriv(fstypename); 151 152 if (usermount == 0 && (error = priv_check(td, priv))) 153 goto done; 154 155 /* 156 * Do not allow NFS export by non-root users. 157 */ 158 if (flags & MNT_EXPORTED) { 159 error = priv_check(td, priv); 160 if (error) 161 goto done; 162 } 163 /* 164 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users 165 */ 166 if (priv_check(td, priv)) 167 flags |= MNT_NOSUID | MNT_NODEV; 168 169 /* 170 * Lookup the requested path and extract the nch and vnode. 171 */ 172 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 173 if (error == 0) { 174 if ((error = nlookup(&nd)) == 0) { 175 if (nd.nl_nch.ncp->nc_vp == NULL) 176 error = ENOENT; 177 } 178 } 179 if (error) { 180 nlookup_done(&nd); 181 goto done; 182 } 183 184 /* 185 * If the target filesystem is resolved via a nullfs mount, then 186 * nd.nl_nch.mount will be pointing to the nullfs mount structure 187 * instead of the target file system. We need it in case we are 188 * doing an update. 189 */ 190 nullmp = nd.nl_nch.mount; 191 192 /* 193 * Extract the locked+refd ncp and cleanup the nd structure 194 */ 195 nch = nd.nl_nch; 196 cache_zero(&nd.nl_nch); 197 nlookup_done(&nd); 198 199 if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && 200 (mp = cache_findmount(&nch)) != NULL) { 201 cache_dropmount(mp); 202 hasmount = 1; 203 } else { 204 hasmount = 0; 205 } 206 207 208 /* 209 * now we have the locked ref'd nch and unreferenced vnode. 210 */ 211 vp = nch.ncp->nc_vp; 212 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) { 213 cache_put(&nch); 214 goto done; 215 } 216 cache_unlock(&nch); 217 218 /* 219 * Now we have an unlocked ref'd nch and a locked ref'd vp 220 */ 221 if (flags & MNT_UPDATE) { 222 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 223 cache_drop(&nch); 224 vput(vp); 225 error = EINVAL; 226 goto done; 227 } 228 229 if (strncmp(fstypename, "null", 5) == 0) { 230 KKASSERT(nullmp); 231 mp = nullmp; 232 } else { 233 mp = vp->v_mount; 234 } 235 236 flag = mp->mnt_flag; 237 flag2 = mp->mnt_kern_flag; 238 /* 239 * We only allow the filesystem to be reloaded if it 240 * is currently mounted read-only. 241 */ 242 if ((flags & MNT_RELOAD) && 243 ((mp->mnt_flag & MNT_RDONLY) == 0)) { 244 cache_drop(&nch); 245 vput(vp); 246 error = EOPNOTSUPP; /* Needs translation */ 247 goto done; 248 } 249 /* 250 * Only root, or the user that did the original mount is 251 * permitted to update it. 252 */ 253 if (mp->mnt_stat.f_owner != cred->cr_uid && 254 (error = priv_check(td, priv))) { 255 cache_drop(&nch); 256 vput(vp); 257 goto done; 258 } 259 if (vfs_busy(mp, LK_NOWAIT)) { 260 cache_drop(&nch); 261 vput(vp); 262 error = EBUSY; 263 goto done; 264 } 265 if (hasmount) { 266 cache_drop(&nch); 267 vfs_unbusy(mp); 268 vput(vp); 269 error = EBUSY; 270 goto done; 271 } 272 mp->mnt_flag |= flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE); 273 lwkt_gettoken(&mp->mnt_token); 274 vn_unlock(vp); 275 vfsp = mp->mnt_vfc; 276 goto update; 277 } 278 279 /* 280 * If the user is not root, ensure that they own the directory 281 * onto which we are attempting to mount. 282 */ 283 if ((error = VOP_GETATTR(vp, &va)) || 284 (va.va_uid != cred->cr_uid && 285 (error = priv_check(td, priv)))) { 286 cache_drop(&nch); 287 vput(vp); 288 goto done; 289 } 290 if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) { 291 cache_drop(&nch); 292 vput(vp); 293 goto done; 294 } 295 if (vp->v_type != VDIR) { 296 cache_drop(&nch); 297 vput(vp); 298 error = ENOTDIR; 299 goto done; 300 } 301 if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) { 302 cache_drop(&nch); 303 vput(vp); 304 error = EPERM; 305 goto done; 306 } 307 vfsp = vfsconf_find_by_name(fstypename); 308 if (vfsp == NULL) { 309 linker_file_t lf; 310 311 /* Only load modules for root (very important!) */ 312 if ((error = priv_check(td, PRIV_ROOT)) != 0) { 313 cache_drop(&nch); 314 vput(vp); 315 goto done; 316 } 317 error = linker_load_file(fstypename, &lf); 318 if (error || lf == NULL) { 319 cache_drop(&nch); 320 vput(vp); 321 if (lf == NULL) 322 error = ENODEV; 323 goto done; 324 } 325 lf->userrefs++; 326 /* lookup again, see if the VFS was loaded */ 327 vfsp = vfsconf_find_by_name(fstypename); 328 if (vfsp == NULL) { 329 lf->userrefs--; 330 linker_file_unload(lf); 331 cache_drop(&nch); 332 vput(vp); 333 error = ENODEV; 334 goto done; 335 } 336 } 337 if (hasmount) { 338 cache_drop(&nch); 339 vput(vp); 340 error = EBUSY; 341 goto done; 342 } 343 344 /* 345 * Allocate and initialize the filesystem. 346 */ 347 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK); 348 mount_init(mp, vfsp->vfc_vfsops); 349 vfs_busy(mp, LK_NOWAIT); 350 mp->mnt_vfc = vfsp; 351 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 352 vfsp->vfc_refcount++; 353 mp->mnt_stat.f_type = vfsp->vfc_typenum; 354 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 355 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 356 mp->mnt_stat.f_owner = cred->cr_uid; 357 lwkt_gettoken(&mp->mnt_token); 358 vn_unlock(vp); 359 update: 360 /* 361 * (per-mount token acquired at this point) 362 * 363 * Set the mount level flags. 364 */ 365 if (flags & MNT_RDONLY) 366 mp->mnt_flag |= MNT_RDONLY; 367 else if (mp->mnt_flag & MNT_RDONLY) 368 mp->mnt_kern_flag |= MNTK_WANTRDWR; 369 mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV | 370 MNT_SYNCHRONOUS | MNT_ASYNC | MNT_NOATIME | 371 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM | 372 MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR | 373 MNT_AUTOMOUNTED); 374 mp->mnt_flag |= flags & (MNT_NOSUID | MNT_NOEXEC | 375 MNT_NODEV | MNT_SYNCHRONOUS | MNT_ASYNC | MNT_FORCE | 376 MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM | 377 MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR | 378 MNT_AUTOMOUNTED); 379 380 /* 381 * Pre-set the mount's ALL_MPSAFE flags if specified in the vfsconf. 382 * This way the initial VFS_MOUNT() call will also be MPSAFE. 383 */ 384 if (vfsp->vfc_flags & VFCF_MPSAFE) 385 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 386 387 /* 388 * Mount the filesystem. 389 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they 390 * get. 391 */ 392 if (mp->mnt_flag & MNT_UPDATE) { 393 error = VFS_MOUNT(mp, uap->path, uap->data, cred); 394 if (mp->mnt_kern_flag & MNTK_WANTRDWR) 395 mp->mnt_flag &= ~MNT_RDONLY; 396 mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE); 397 mp->mnt_kern_flag &=~ MNTK_WANTRDWR; 398 if (error) { 399 mp->mnt_flag = flag; 400 mp->mnt_kern_flag = flag2; 401 } 402 lwkt_reltoken(&mp->mnt_token); 403 vfs_unbusy(mp); 404 vrele(vp); 405 cache_drop(&nch); 406 goto done; 407 } 408 mp->mnt_ncmounton = nch; 409 error = VFS_MOUNT(mp, uap->path, uap->data, cred); 410 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 411 412 /* 413 * Put the new filesystem on the mount list after root. The mount 414 * point gets its own mnt_ncmountpt (unless the VFS already set one 415 * up) which represents the root of the mount. The lookup code 416 * detects the mount point going forward and checks the root of 417 * the mount going backwards. 418 * 419 * It is not necessary to invalidate or purge the vnode underneath 420 * because elements under the mount will be given their own glue 421 * namecache record. 422 */ 423 if (!error) { 424 if (mp->mnt_ncmountpt.ncp == NULL) { 425 /* 426 * Allocate, then unlock, but leave the ref intact. 427 * This is the mnt_refs (1) that we will retain 428 * through to the unmount. 429 */ 430 cache_allocroot(&mp->mnt_ncmountpt, mp, NULL); 431 cache_unlock(&mp->mnt_ncmountpt); 432 } 433 vn_unlock(vp); 434 cache_lock(&nch); 435 nch.ncp->nc_flag |= NCF_ISMOUNTPT; 436 cache_unlock(&nch); 437 cache_ismounting(mp); 438 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 439 440 mountlist_insert(mp, MNTINS_LAST); 441 vn_unlock(vp); 442 checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt); 443 error = vfs_allocate_syncvnode(mp); 444 lwkt_reltoken(&mp->mnt_token); 445 vfs_unbusy(mp); 446 error = VFS_START(mp, 0); 447 vrele(vp); 448 KNOTE(&fs_klist, VQ_MOUNT); 449 } else { 450 bzero(&mp->mnt_ncmounton, sizeof(mp->mnt_ncmounton)); 451 vn_syncer_thr_stop(mp); 452 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops); 453 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops); 454 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops); 455 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops); 456 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops); 457 if (mp->mnt_cred) { 458 crfree(mp->mnt_cred); 459 mp->mnt_cred = NULL; 460 } 461 mp->mnt_vfc->vfc_refcount--; 462 lwkt_reltoken(&mp->mnt_token); 463 vfs_unbusy(mp); 464 kfree(mp, M_MOUNT); 465 cache_drop(&nch); 466 vput(vp); 467 } 468 done: 469 return (error); 470 } 471 472 /* 473 * Scan all active processes to see if any of them have a current 474 * or root directory onto which the new filesystem has just been 475 * mounted. If so, replace them with the new mount point. 476 * 477 * Both old_nch and new_nch are ref'd on call but not locked. 478 * new_nch must be temporarily locked so it can be associated with the 479 * vnode representing the root of the mount point. 480 */ 481 struct checkdirs_info { 482 struct nchandle old_nch; 483 struct nchandle new_nch; 484 struct vnode *old_vp; 485 struct vnode *new_vp; 486 }; 487 488 static int checkdirs_callback(struct proc *p, void *data); 489 490 static void 491 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch) 492 { 493 struct checkdirs_info info; 494 struct vnode *olddp; 495 struct vnode *newdp; 496 struct mount *mp; 497 498 /* 499 * If the old mount point's vnode has a usecount of 1, it is not 500 * being held as a descriptor anywhere. 501 */ 502 olddp = old_nch->ncp->nc_vp; 503 if (olddp == NULL || VREFCNT(olddp) == 1) 504 return; 505 506 /* 507 * Force the root vnode of the new mount point to be resolved 508 * so we can update any matching processes. 509 */ 510 mp = new_nch->mount; 511 if (VFS_ROOT(mp, &newdp)) 512 panic("mount: lost mount"); 513 vn_unlock(newdp); 514 cache_lock(new_nch); 515 vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY); 516 cache_setunresolved(new_nch); 517 cache_setvp(new_nch, newdp); 518 cache_unlock(new_nch); 519 520 /* 521 * Special handling of the root node 522 */ 523 if (rootvnode == olddp) { 524 vref(newdp); 525 vfs_cache_setroot(newdp, cache_hold(new_nch)); 526 } 527 528 /* 529 * Pass newdp separately so the callback does not have to access 530 * it via new_nch->ncp->nc_vp. 531 */ 532 info.old_nch = *old_nch; 533 info.new_nch = *new_nch; 534 info.new_vp = newdp; 535 allproc_scan(checkdirs_callback, &info, 0); 536 vput(newdp); 537 } 538 539 /* 540 * NOTE: callback is not MP safe because the scanned process's filedesc 541 * structure can be ripped out from under us, amoung other things. 542 */ 543 static int 544 checkdirs_callback(struct proc *p, void *data) 545 { 546 struct checkdirs_info *info = data; 547 struct filedesc *fdp; 548 struct nchandle ncdrop1; 549 struct nchandle ncdrop2; 550 struct vnode *vprele1; 551 struct vnode *vprele2; 552 553 if ((fdp = p->p_fd) != NULL) { 554 cache_zero(&ncdrop1); 555 cache_zero(&ncdrop2); 556 vprele1 = NULL; 557 vprele2 = NULL; 558 559 /* 560 * MPUNSAFE - XXX fdp can be pulled out from under a 561 * foreign process. 562 * 563 * A shared filedesc is ok, we don't have to copy it 564 * because we are making this change globally. 565 */ 566 spin_lock(&fdp->fd_spin); 567 if (fdp->fd_ncdir.mount == info->old_nch.mount && 568 fdp->fd_ncdir.ncp == info->old_nch.ncp) { 569 vprele1 = fdp->fd_cdir; 570 vref(info->new_vp); 571 fdp->fd_cdir = info->new_vp; 572 ncdrop1 = fdp->fd_ncdir; 573 cache_copy(&info->new_nch, &fdp->fd_ncdir); 574 } 575 if (fdp->fd_nrdir.mount == info->old_nch.mount && 576 fdp->fd_nrdir.ncp == info->old_nch.ncp) { 577 vprele2 = fdp->fd_rdir; 578 vref(info->new_vp); 579 fdp->fd_rdir = info->new_vp; 580 ncdrop2 = fdp->fd_nrdir; 581 cache_copy(&info->new_nch, &fdp->fd_nrdir); 582 } 583 spin_unlock(&fdp->fd_spin); 584 if (ncdrop1.ncp) 585 cache_drop(&ncdrop1); 586 if (ncdrop2.ncp) 587 cache_drop(&ncdrop2); 588 if (vprele1) 589 vrele(vprele1); 590 if (vprele2) 591 vrele(vprele2); 592 } 593 return(0); 594 } 595 596 /* 597 * Unmount a file system. 598 * 599 * Note: unmount takes a path to the vnode mounted on as argument, 600 * not special file (as before). 601 * 602 * umount_args(char *path, int flags) 603 * 604 * MPALMOSTSAFE 605 */ 606 int 607 sys_unmount(struct sysmsg *sysmsg, const struct unmount_args *uap) 608 { 609 struct thread *td = curthread; 610 struct proc *p __debugvar = td->td_proc; 611 struct mount *mp = NULL; 612 struct nlookupdata nd; 613 char fstypename[MFSNAMELEN]; 614 int priv = 0; 615 int error; 616 struct ucred *cred; 617 618 cred = td->td_ucred; 619 620 KKASSERT(p); 621 622 /* We do not allow user umounts inside a jail for now */ 623 if (usermount && jailed(cred)) { 624 error = EPERM; 625 goto done; 626 } 627 628 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 629 NLC_FOLLOW | NLC_IGNBADDIR); 630 if (error == 0) 631 error = nlookup(&nd); 632 if (error) 633 goto out; 634 635 mp = nd.nl_nch.mount; 636 637 /* Figure out the fsname in order to select proper privs */ 638 ksnprintf(fstypename, MFSNAMELEN, "%s", mp->mnt_vfc->vfc_name); 639 priv = get_fspriv(fstypename); 640 641 if (usermount == 0 && (error = priv_check(td, priv))) { 642 nlookup_done(&nd); 643 goto done; 644 } 645 646 /* 647 * Only root, or the user that did the original mount is 648 * permitted to unmount this filesystem. 649 */ 650 if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) && 651 (error = priv_check(td, priv))) 652 goto out; 653 654 /* 655 * Don't allow unmounting the root file system. 656 */ 657 if (mp->mnt_flag & MNT_ROOTFS) { 658 error = EINVAL; 659 goto out; 660 } 661 662 /* 663 * Must be the root of the filesystem 664 */ 665 if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) { 666 error = EINVAL; 667 goto out; 668 } 669 670 /* Check if this mount belongs to this prison */ 671 if (jailed(cred) && mp->mnt_cred && (!mp->mnt_cred->cr_prison || 672 mp->mnt_cred->cr_prison != cred->cr_prison)) { 673 kprintf("mountpoint %s does not belong to this jail\n", 674 uap->path); 675 error = EPERM; 676 goto out; 677 } 678 679 /* 680 * If no error try to issue the unmount. We lose our cache 681 * ref when we call nlookup_done so we must hold the mount point 682 * to prevent use-after-free races. 683 */ 684 out: 685 if (error == 0) { 686 mount_hold(mp); 687 nlookup_done(&nd); 688 error = dounmount(mp, uap->flags, 0); 689 mount_drop(mp); 690 } else { 691 nlookup_done(&nd); 692 } 693 done: 694 return (error); 695 } 696 697 /* 698 * Do the actual file system unmount (interlocked against the mountlist 699 * token and mp->mnt_token). 700 */ 701 static int 702 dounmount_interlock(struct mount *mp) 703 { 704 if (mp->mnt_kern_flag & MNTK_UNMOUNT) 705 return (EBUSY); 706 mp->mnt_kern_flag |= MNTK_UNMOUNT; 707 return(0); 708 } 709 710 static int 711 unmount_allproc_cb(struct proc *p, void *arg) 712 { 713 struct mount *mp; 714 715 if (p->p_textnch.ncp == NULL) 716 return 0; 717 718 mp = (struct mount *)arg; 719 if (p->p_textnch.mount == mp) 720 cache_drop(&p->p_textnch); 721 722 return 0; 723 } 724 725 /* 726 * The guts of the unmount code. The mount owns one ref and one hold 727 * count. If we successfully interlock the unmount, those refs are ours. 728 * (The ref is from mnt_ncmountpt). 729 * 730 * When halting we shortcut certain mount types such as devfs by not actually 731 * issuing the VFS_SYNC() or VFS_UNMOUNT(). They are still disconnected 732 * from the mountlist so higher-level filesytems can unmount cleanly. 733 * 734 * The mount types that allow QUICKHALT are: devfs, tmpfs, procfs. 735 */ 736 int 737 dounmount(struct mount *mp, int flags, int halting) 738 { 739 struct namecache *ncp; 740 struct nchandle nch; 741 struct vnode *vp; 742 int error; 743 int async_flag; 744 int lflags; 745 int freeok = 1; 746 int hadsyncer = 0; 747 int retry; 748 int quickhalt; 749 750 lwkt_gettoken(&mp->mnt_token); 751 752 /* 753 * When halting, certain mount points can essentially just 754 * be unhooked and otherwise ignored. 755 */ 756 if (halting && (mp->mnt_kern_flag & MNTK_QUICKHALT)) { 757 quickhalt = 1; 758 freeok = 0; 759 } else { 760 quickhalt = 0; 761 } 762 763 764 /* 765 * Exclusive access for unmounting purposes. 766 */ 767 if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0) 768 goto out; 769 770 /* 771 * We now 'own' the last mp->mnt_refs 772 * 773 * Allow filesystems to detect that a forced unmount is in progress. 774 */ 775 if (flags & MNT_FORCE) 776 mp->mnt_kern_flag |= MNTK_UNMOUNTF; 777 lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK); 778 error = lockmgr(&mp->mnt_lock, lflags); 779 if (error) { 780 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 781 if (mp->mnt_kern_flag & MNTK_MWAIT) { 782 mp->mnt_kern_flag &= ~MNTK_MWAIT; 783 wakeup(mp); 784 } 785 goto out; 786 } 787 788 if (mp->mnt_flag & MNT_EXPUBLIC) 789 vfs_setpublicfs(NULL, NULL, NULL); 790 791 vfs_msync(mp, MNT_WAIT); 792 async_flag = mp->mnt_flag & MNT_ASYNC; 793 mp->mnt_flag &=~ MNT_ASYNC; 794 795 /* 796 * Decomission our special mnt_syncer vnode. This also stops 797 * the vnlru code. If we are unable to unmount we recommission 798 * the vnode. 799 * 800 * Then sync the filesystem. 801 */ 802 if ((vp = mp->mnt_syncer) != NULL) { 803 mp->mnt_syncer = NULL; 804 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 805 vrele(vp); 806 hadsyncer = 1; 807 } 808 809 /* 810 * Sync normally-mounted filesystem. 811 */ 812 if (quickhalt == 0) { 813 if ((mp->mnt_flag & MNT_RDONLY) == 0) 814 VFS_SYNC(mp, MNT_WAIT); 815 } 816 817 /* 818 * nchandle records ref the mount structure. Expect a count of 1 819 * (our mount->mnt_ncmountpt). 820 * 821 * Scans can get temporary refs on a mountpoint (thought really 822 * heavy duty stuff like cache_findmount() do not). 823 */ 824 for (retry = 0; (retry < 10 || debug_unmount); ++retry) { 825 /* 826 * Invalidate the namecache topology under the mount. 827 * nullfs mounts alias a real mount's namecache topology 828 * and it should not be invalidated in that case. 829 */ 830 if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) { 831 cache_lock(&mp->mnt_ncmountpt); 832 cache_inval(&mp->mnt_ncmountpt, 833 CINV_DESTROY|CINV_CHILDREN); 834 cache_unlock(&mp->mnt_ncmountpt); 835 } 836 837 /* 838 * Clear pcpu caches 839 */ 840 cache_unmounting(mp); 841 if (mp->mnt_refs != 1) 842 cache_clearmntcache(mp); 843 844 /* 845 * Break out if we are good. Don't count ncp refs if the 846 * mount is aliased. 847 */ 848 ncp = (mp->mnt_kern_flag & MNTK_NCALIASED) ? 849 NULL : mp->mnt_ncmountpt.ncp; 850 if (mp->mnt_refs == 1 && 851 (ncp == NULL || (ncp->nc_refs == 1 && 852 TAILQ_FIRST(&ncp->nc_list) == NULL))) { 853 break; 854 } 855 856 /* 857 * If forcing the unmount, clean out any p->p_textnch 858 * nchandles that match this mount. 859 */ 860 if (flags & MNT_FORCE) 861 allproc_scan(&unmount_allproc_cb, mp, 0); 862 863 /* 864 * Sleep and retry. 865 */ 866 tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1); 867 if ((retry & 15) == 15) { 868 mount_warning(mp, 869 "(%p) debug - retry %d, " 870 "%d namecache refs, %d mount refs", 871 mp, retry, 872 (ncp ? ncp->nc_refs - 1 : 0), 873 mp->mnt_refs - 1); 874 } 875 } 876 877 error = 0; 878 ncp = (mp->mnt_kern_flag & MNTK_NCALIASED) ? 879 NULL : mp->mnt_ncmountpt.ncp; 880 if (mp->mnt_refs != 1 || 881 (ncp != NULL && (ncp->nc_refs != 1 || 882 TAILQ_FIRST(&ncp->nc_list)))) { 883 mount_warning(mp, 884 "(%p): %d namecache refs, %d mount refs " 885 "still present", 886 mp, 887 (ncp ? ncp->nc_refs - 1 : 0), 888 mp->mnt_refs - 1); 889 if (flags & MNT_FORCE) { 890 freeok = 0; 891 mount_warning(mp, "forcing unmount\n"); 892 } else { 893 error = EBUSY; 894 } 895 } 896 897 /* 898 * So far so good, sync the filesystem once more and 899 * call the VFS unmount code if the sync succeeds. 900 */ 901 if (error == 0 && quickhalt == 0) { 902 if (mp->mnt_flag & MNT_RDONLY) { 903 error = VFS_UNMOUNT(mp, flags); 904 } else { 905 error = VFS_SYNC(mp, MNT_WAIT); 906 if (error == 0 || /* no error */ 907 error == EOPNOTSUPP || /* no sync avail */ 908 (flags & MNT_FORCE)) { /* force anyway */ 909 error = VFS_UNMOUNT(mp, flags); 910 } 911 } 912 if (error) { 913 mount_warning(mp, 914 "(%p) unmount: vfs refused to unmount, " 915 "error %d", 916 mp, error); 917 } 918 } 919 920 /* 921 * If an error occurred we can still recover, restoring the 922 * syncer vnode and misc flags. 923 */ 924 if (error) { 925 if (mp->mnt_syncer == NULL && hadsyncer) 926 vfs_allocate_syncvnode(mp); 927 mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF); 928 mp->mnt_flag |= async_flag; 929 lockmgr(&mp->mnt_lock, LK_RELEASE); 930 if (mp->mnt_kern_flag & MNTK_MWAIT) { 931 mp->mnt_kern_flag &= ~MNTK_MWAIT; 932 wakeup(mp); 933 } 934 goto out; 935 } 936 /* 937 * Clean up any journals still associated with the mount after 938 * filesystem activity has ceased. 939 */ 940 journal_remove_all_journals(mp, 941 ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0)); 942 943 mountlist_remove(mp); 944 945 /* 946 * Remove any installed vnode ops here so the individual VFSs don't 947 * have to. 948 * 949 * mnt_refs should go to zero when we scrap mnt_ncmountpt. 950 * 951 * When quickhalting we have to keep these intact because the 952 * underlying vnodes have not been destroyed, and some might be 953 * dirty. 954 */ 955 if (quickhalt == 0) { 956 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops); 957 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops); 958 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops); 959 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops); 960 vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops); 961 } 962 963 if (mp->mnt_ncmountpt.ncp != NULL) { 964 nch = mp->mnt_ncmountpt; 965 cache_zero(&mp->mnt_ncmountpt); 966 cache_clrmountpt(&nch); 967 cache_drop(&nch); 968 } 969 if (mp->mnt_ncmounton.ncp != NULL) { 970 cache_unmounting(mp); 971 nch = mp->mnt_ncmounton; 972 cache_zero(&mp->mnt_ncmounton); 973 cache_clrmountpt(&nch); 974 cache_drop(&nch); 975 } 976 977 if (mp->mnt_cred) { 978 crfree(mp->mnt_cred); 979 mp->mnt_cred = NULL; 980 } 981 982 mp->mnt_vfc->vfc_refcount--; 983 984 /* 985 * If not quickhalting the mount, we expect there to be no 986 * vnodes left. 987 */ 988 if (quickhalt == 0 && !TAILQ_EMPTY(&mp->mnt_nvnodelist)) 989 panic("unmount: dangling vnode"); 990 991 /* 992 * Release the lock 993 */ 994 lockmgr(&mp->mnt_lock, LK_RELEASE); 995 if (mp->mnt_kern_flag & MNTK_MWAIT) { 996 mp->mnt_kern_flag &= ~MNTK_MWAIT; 997 wakeup(mp); 998 } 999 1000 /* 1001 * If we reach here and freeok != 0 we must free the mount. 1002 * mnt_refs should already have dropped to 0, so if it is not 1003 * zero we must cycle the caches and wait. 1004 * 1005 * When we are satisfied that the mount has disconnected we can 1006 * drop the hold on the mp that represented the mount (though the 1007 * caller might actually have another, so the caller's drop may 1008 * do the actual free). 1009 */ 1010 if (freeok) { 1011 if (mp->mnt_refs > 0) 1012 cache_clearmntcache(mp); 1013 while (mp->mnt_refs > 0) { 1014 cache_unmounting(mp); 1015 wakeup(mp); 1016 tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1); 1017 cache_clearmntcache(mp); 1018 } 1019 lwkt_reltoken(&mp->mnt_token); 1020 mount_drop(mp); 1021 mp = NULL; 1022 } else { 1023 cache_clearmntcache(mp); 1024 } 1025 error = 0; 1026 KNOTE(&fs_klist, VQ_UNMOUNT); 1027 out: 1028 if (mp) 1029 lwkt_reltoken(&mp->mnt_token); 1030 return (error); 1031 } 1032 1033 static 1034 void 1035 mount_warning(struct mount *mp, const char *ctl, ...) 1036 { 1037 char *ptr; 1038 char *buf; 1039 __va_list va; 1040 1041 __va_start(va, ctl); 1042 if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL, 1043 &ptr, &buf, 0) == 0) { 1044 kprintf("unmount(%s): ", ptr); 1045 kvprintf(ctl, va); 1046 kprintf("\n"); 1047 kfree(buf, M_TEMP); 1048 } else { 1049 kprintf("unmount(%p", mp); 1050 if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name) 1051 kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name); 1052 kprintf("): "); 1053 kvprintf(ctl, va); 1054 kprintf("\n"); 1055 } 1056 __va_end(va); 1057 } 1058 1059 /* 1060 * Shim cache_fullpath() to handle the case where a process is chrooted into 1061 * a subdirectory of a mount. In this case if the root mount matches the 1062 * process root directory's mount we have to specify the process's root 1063 * directory instead of the mount point, because the mount point might 1064 * be above the root directory. 1065 */ 1066 static 1067 int 1068 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb) 1069 { 1070 struct nchandle *nch; 1071 1072 if (p && p->p_fd->fd_nrdir.mount == mp) 1073 nch = &p->p_fd->fd_nrdir; 1074 else 1075 nch = &mp->mnt_ncmountpt; 1076 return(cache_fullpath(p, nch, NULL, rb, fb, 0)); 1077 } 1078 1079 /* 1080 * Sync each mounted filesystem. 1081 */ 1082 1083 #ifdef DEBUG 1084 static int syncprt = 0; 1085 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, ""); 1086 #endif /* DEBUG */ 1087 1088 static int sync_callback(struct mount *mp, void *data); 1089 1090 int 1091 sys_sync(struct sysmsg *sysmsg, const struct sync_args *uap) 1092 { 1093 mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD); 1094 return (0); 1095 } 1096 1097 static 1098 int 1099 sync_callback(struct mount *mp, void *data __unused) 1100 { 1101 int asyncflag; 1102 1103 if ((mp->mnt_flag & MNT_RDONLY) == 0) { 1104 lwkt_gettoken(&mp->mnt_token); 1105 asyncflag = mp->mnt_flag & MNT_ASYNC; 1106 mp->mnt_flag &= ~MNT_ASYNC; 1107 lwkt_reltoken(&mp->mnt_token); 1108 vfs_msync(mp, MNT_NOWAIT); 1109 VFS_SYNC(mp, MNT_NOWAIT); 1110 lwkt_gettoken(&mp->mnt_token); 1111 mp->mnt_flag |= asyncflag; 1112 lwkt_reltoken(&mp->mnt_token); 1113 } 1114 return(0); 1115 } 1116 1117 /* XXX PRISON: could be per prison flag */ 1118 static int prison_quotas; 1119 #if 0 1120 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, ""); 1121 #endif 1122 1123 /* 1124 * quotactl_args(char *path, int fcmd, int uid, caddr_t arg) 1125 * 1126 * Change filesystem quotas. 1127 * 1128 * MPALMOSTSAFE 1129 */ 1130 int 1131 sys_quotactl(struct sysmsg *sysmsg, const struct quotactl_args *uap) 1132 { 1133 struct nlookupdata nd; 1134 struct thread *td; 1135 struct mount *mp; 1136 int error; 1137 1138 td = curthread; 1139 if (td->td_ucred->cr_prison && !prison_quotas) { 1140 error = EPERM; 1141 goto done; 1142 } 1143 1144 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1145 if (error == 0) 1146 error = nlookup(&nd); 1147 if (error == 0) { 1148 mp = nd.nl_nch.mount; 1149 error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, 1150 uap->arg, nd.nl_cred); 1151 } 1152 nlookup_done(&nd); 1153 done: 1154 return (error); 1155 } 1156 1157 /* 1158 * mountctl(char *path, int op, int fd, const void *ctl, int ctllen, 1159 * void *buf, int buflen) 1160 * 1161 * This function operates on a mount point and executes the specified 1162 * operation using the specified control data, and possibly returns data. 1163 * 1164 * The actual number of bytes stored in the result buffer is returned, 0 1165 * if none, otherwise an error is returned. 1166 * 1167 * MPALMOSTSAFE 1168 */ 1169 int 1170 sys_mountctl(struct sysmsg *sysmsg, const struct mountctl_args *uap) 1171 { 1172 struct thread *td = curthread; 1173 struct file *fp; 1174 void *ctl = NULL; 1175 void *buf = NULL; 1176 char *path = NULL; 1177 int error; 1178 1179 /* 1180 * Sanity and permissions checks. We must be root. 1181 */ 1182 if (td->td_ucred->cr_prison != NULL) 1183 return (EPERM); 1184 if ((uap->op != MOUNTCTL_MOUNTFLAGS) && 1185 (error = priv_check(td, PRIV_ROOT)) != 0) 1186 return (error); 1187 1188 /* 1189 * Argument length checks 1190 */ 1191 if (uap->ctllen < 0 || uap->ctllen > 1024) 1192 return (EINVAL); 1193 if (uap->buflen < 0 || uap->buflen > 16 * 1024) 1194 return (EINVAL); 1195 if (uap->path == NULL) 1196 return (EINVAL); 1197 1198 /* 1199 * Allocate the necessary buffers and copyin data 1200 */ 1201 path = objcache_get(namei_oc, M_WAITOK); 1202 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 1203 if (error) 1204 goto done; 1205 1206 if (uap->ctllen) { 1207 ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO); 1208 error = copyin(uap->ctl, ctl, uap->ctllen); 1209 if (error) 1210 goto done; 1211 } 1212 if (uap->buflen) 1213 buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO); 1214 1215 /* 1216 * Validate the descriptor 1217 */ 1218 if (uap->fd >= 0) { 1219 fp = holdfp(td, uap->fd, -1); 1220 if (fp == NULL) { 1221 error = EBADF; 1222 goto done; 1223 } 1224 } else { 1225 fp = NULL; 1226 } 1227 1228 /* 1229 * Execute the internal kernel function and clean up. 1230 */ 1231 error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, 1232 buf, uap->buflen, &sysmsg->sysmsg_result); 1233 if (fp) 1234 dropfp(td, uap->fd, fp); 1235 if (error == 0 && sysmsg->sysmsg_result > 0) 1236 error = copyout(buf, uap->buf, sysmsg->sysmsg_result); 1237 done: 1238 if (path) 1239 objcache_put(namei_oc, path); 1240 if (ctl) 1241 kfree(ctl, M_TEMP); 1242 if (buf) 1243 kfree(buf, M_TEMP); 1244 return (error); 1245 } 1246 1247 /* 1248 * Execute a mount control operation by resolving the path to a mount point 1249 * and calling vop_mountctl(). 1250 * 1251 * Use the mount point from the nch instead of the vnode so nullfs mounts 1252 * can properly spike the VOP. 1253 */ 1254 int 1255 kern_mountctl(const char *path, int op, struct file *fp, 1256 const void *ctl, int ctllen, 1257 void *buf, int buflen, int *res) 1258 { 1259 struct vnode *vp; 1260 struct nlookupdata nd; 1261 struct nchandle nch; 1262 struct mount *mp; 1263 int error; 1264 1265 *res = 0; 1266 vp = NULL; 1267 error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW); 1268 if (error) 1269 return (error); 1270 error = nlookup(&nd); 1271 if (error) { 1272 nlookup_done(&nd); 1273 return (error); 1274 } 1275 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 1276 if (error) { 1277 nlookup_done(&nd); 1278 return (error); 1279 } 1280 1281 /* 1282 * Yes, all this is needed to use the nch.mount below, because 1283 * we must maintain a ref on the mount to avoid ripouts (e.g. 1284 * due to heavy mount/unmount use by synth or poudriere). 1285 */ 1286 nch = nd.nl_nch; 1287 cache_zero(&nd.nl_nch); 1288 cache_unlock(&nch); 1289 nlookup_done(&nd); 1290 vn_unlock(vp); 1291 1292 mp = nch.mount; 1293 1294 /* 1295 * Must be the root of the filesystem 1296 */ 1297 if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) { 1298 cache_drop(&nch); 1299 vrele(vp); 1300 return (EINVAL); 1301 } 1302 if (mp == NULL || mp->mnt_kern_flag & MNTK_UNMOUNT) { 1303 kprintf("kern_mountctl: Warning, \"%s\" racing unmount\n", 1304 path); 1305 cache_drop(&nch); 1306 vrele(vp); 1307 return (EINVAL); 1308 } 1309 error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen, 1310 buf, buflen, res); 1311 vrele(vp); 1312 cache_drop(&nch); 1313 1314 return (error); 1315 } 1316 1317 int 1318 kern_statfs(struct nlookupdata *nd, struct statfs *buf) 1319 { 1320 struct thread *td = curthread; 1321 struct proc *p = td->td_proc; 1322 struct mount *mp; 1323 struct statfs *sp; 1324 char *fullpath, *freepath; 1325 int error; 1326 1327 if ((error = nlookup(nd)) != 0) 1328 return (error); 1329 mp = nd->nl_nch.mount; 1330 sp = &mp->mnt_stat; 1331 1332 /* 1333 * Ignore refresh error, user should have visibility. 1334 * This can happen if a NFS mount goes bad (e.g. server 1335 * revokes perms or goes down). 1336 */ 1337 error = VFS_STATFS(mp, sp, nd->nl_cred); 1338 /* ignore error */ 1339 1340 error = mount_path(p, mp, &fullpath, &freepath); 1341 if (error) 1342 return(error); 1343 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1344 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1345 kfree(freepath, M_TEMP); 1346 1347 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1348 bcopy(sp, buf, sizeof(*buf)); 1349 /* Only root should have access to the fsid's. */ 1350 if (priv_check(td, PRIV_ROOT)) 1351 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1352 return (0); 1353 } 1354 1355 /* 1356 * statfs_args(char *path, struct statfs *buf) 1357 * 1358 * Get filesystem statistics. 1359 */ 1360 int 1361 sys_statfs(struct sysmsg *sysmsg, const struct statfs_args *uap) 1362 { 1363 struct nlookupdata nd; 1364 struct statfs buf; 1365 int error; 1366 1367 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1368 if (error == 0) 1369 error = kern_statfs(&nd, &buf); 1370 nlookup_done(&nd); 1371 if (error == 0) 1372 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1373 return (error); 1374 } 1375 1376 int 1377 kern_fstatfs(int fd, struct statfs *buf) 1378 { 1379 struct thread *td = curthread; 1380 struct proc *p = td->td_proc; 1381 struct file *fp; 1382 struct mount *mp; 1383 struct statfs *sp; 1384 char *fullpath, *freepath; 1385 int error; 1386 1387 KKASSERT(p); 1388 if ((error = holdvnode(td, fd, &fp)) != 0) 1389 return (error); 1390 1391 /* 1392 * Try to use mount info from any overlays rather than the 1393 * mount info for the underlying vnode, otherwise we will 1394 * fail when operating on null-mounted paths inside a chroot. 1395 */ 1396 if ((mp = fp->f_nchandle.mount) == NULL) 1397 mp = ((struct vnode *)fp->f_data)->v_mount; 1398 if (mp == NULL) { 1399 error = EBADF; 1400 goto done; 1401 } 1402 if (fp->f_cred == NULL) { 1403 error = EINVAL; 1404 goto done; 1405 } 1406 1407 /* 1408 * Ignore refresh error, user should have visibility. 1409 * This can happen if a NFS mount goes bad (e.g. server 1410 * revokes perms or goes down). 1411 */ 1412 sp = &mp->mnt_stat; 1413 error = VFS_STATFS(mp, sp, fp->f_cred); 1414 1415 if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0) 1416 goto done; 1417 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1418 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1419 kfree(freepath, M_TEMP); 1420 1421 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1422 bcopy(sp, buf, sizeof(*buf)); 1423 1424 /* Only root should have access to the fsid's. */ 1425 if (priv_check(td, PRIV_ROOT)) 1426 buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; 1427 error = 0; 1428 done: 1429 fdrop(fp); 1430 return (error); 1431 } 1432 1433 /* 1434 * fstatfs_args(int fd, struct statfs *buf) 1435 * 1436 * Get filesystem statistics. 1437 */ 1438 int 1439 sys_fstatfs(struct sysmsg *sysmsg, const struct fstatfs_args *uap) 1440 { 1441 struct statfs buf; 1442 int error; 1443 1444 error = kern_fstatfs(uap->fd, &buf); 1445 1446 if (error == 0) 1447 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1448 return (error); 1449 } 1450 1451 int 1452 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf) 1453 { 1454 struct mount *mp; 1455 struct statvfs *sp; 1456 int error; 1457 1458 if ((error = nlookup(nd)) != 0) 1459 return (error); 1460 mp = nd->nl_nch.mount; 1461 sp = &mp->mnt_vstat; 1462 if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0) 1463 return (error); 1464 1465 sp->f_flag = 0; 1466 if (mp->mnt_flag & MNT_RDONLY) 1467 sp->f_flag |= ST_RDONLY; 1468 if (mp->mnt_flag & MNT_NOSUID) 1469 sp->f_flag |= ST_NOSUID; 1470 bcopy(sp, buf, sizeof(*buf)); 1471 return (0); 1472 } 1473 1474 /* 1475 * statfs_args(char *path, struct statfs *buf) 1476 * 1477 * Get filesystem statistics. 1478 */ 1479 int 1480 sys_statvfs(struct sysmsg *sysmsg, const struct statvfs_args *uap) 1481 { 1482 struct nlookupdata nd; 1483 struct statvfs buf; 1484 int error; 1485 1486 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1487 if (error == 0) 1488 error = kern_statvfs(&nd, &buf); 1489 nlookup_done(&nd); 1490 if (error == 0) 1491 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1492 return (error); 1493 } 1494 1495 int 1496 kern_fstatvfs(int fd, struct statvfs *buf) 1497 { 1498 struct thread *td = curthread; 1499 struct file *fp; 1500 struct mount *mp; 1501 struct statvfs *sp; 1502 int error; 1503 1504 if ((error = holdvnode(td, fd, &fp)) != 0) 1505 return (error); 1506 if ((mp = fp->f_nchandle.mount) == NULL) 1507 mp = ((struct vnode *)fp->f_data)->v_mount; 1508 if (mp == NULL) { 1509 error = EBADF; 1510 goto done; 1511 } 1512 if (fp->f_cred == NULL) { 1513 error = EINVAL; 1514 goto done; 1515 } 1516 sp = &mp->mnt_vstat; 1517 if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0) 1518 goto done; 1519 1520 sp->f_flag = 0; 1521 if (mp->mnt_flag & MNT_RDONLY) 1522 sp->f_flag |= ST_RDONLY; 1523 if (mp->mnt_flag & MNT_NOSUID) 1524 sp->f_flag |= ST_NOSUID; 1525 1526 bcopy(sp, buf, sizeof(*buf)); 1527 error = 0; 1528 done: 1529 fdrop(fp); 1530 return (error); 1531 } 1532 1533 /* 1534 * fstatfs_args(int fd, struct statfs *buf) 1535 * 1536 * Get filesystem statistics. 1537 */ 1538 int 1539 sys_fstatvfs(struct sysmsg *sysmsg, const struct fstatvfs_args *uap) 1540 { 1541 struct statvfs buf; 1542 int error; 1543 1544 error = kern_fstatvfs(uap->fd, &buf); 1545 1546 if (error == 0) 1547 error = copyout(&buf, uap->buf, sizeof(*uap->buf)); 1548 return (error); 1549 } 1550 1551 /* 1552 * getfsstat_args(struct statfs *buf, long bufsize, int flags) 1553 * 1554 * Get statistics on all filesystems. 1555 */ 1556 1557 struct getfsstat_info { 1558 struct statfs *sfsp; 1559 long count; 1560 long maxcount; 1561 int error; 1562 int flags; 1563 struct thread *td; 1564 }; 1565 1566 static int getfsstat_callback(struct mount *, void *); 1567 1568 int 1569 sys_getfsstat(struct sysmsg *sysmsg, const struct getfsstat_args *uap) 1570 { 1571 struct thread *td = curthread; 1572 struct getfsstat_info info; 1573 1574 bzero(&info, sizeof(info)); 1575 1576 info.maxcount = uap->bufsize / sizeof(struct statfs); 1577 info.sfsp = uap->buf; 1578 info.count = 0; 1579 info.flags = uap->flags; 1580 info.td = td; 1581 1582 mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD); 1583 if (info.sfsp && info.count > info.maxcount) 1584 sysmsg->sysmsg_result = info.maxcount; 1585 else 1586 sysmsg->sysmsg_result = info.count; 1587 return (info.error); 1588 } 1589 1590 static int 1591 getfsstat_callback(struct mount *mp, void *data) 1592 { 1593 struct getfsstat_info *info = data; 1594 struct statfs *sp; 1595 char *freepath; 1596 char *fullpath; 1597 int error; 1598 1599 if (info->td->td_proc && !chroot_visible_mnt(mp, info->td->td_proc)) 1600 return(0); 1601 1602 if (info->sfsp && info->count < info->maxcount) { 1603 sp = &mp->mnt_stat; 1604 1605 /* 1606 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1607 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1608 * overrides MNT_WAIT. 1609 * 1610 * Ignore refresh error, user should have visibility. 1611 * This can happen if a NFS mount goes bad (e.g. server 1612 * revokes perms or goes down). 1613 */ 1614 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1615 (info->flags & MNT_WAIT)) && 1616 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) { 1617 /* ignore error */ 1618 } 1619 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1620 1621 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath); 1622 if (error) { 1623 info->error = error; 1624 return(-1); 1625 } 1626 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1627 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1628 kfree(freepath, M_TEMP); 1629 1630 error = copyout(sp, info->sfsp, sizeof(*sp)); 1631 if (error) { 1632 info->error = error; 1633 return (-1); 1634 } 1635 ++info->sfsp; 1636 } 1637 info->count++; 1638 return(0); 1639 } 1640 1641 /* 1642 * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf, 1643 long bufsize, int flags) 1644 * 1645 * Get statistics on all filesystems. 1646 */ 1647 1648 struct getvfsstat_info { 1649 struct statfs *sfsp; 1650 struct statvfs *vsfsp; 1651 long count; 1652 long maxcount; 1653 int error; 1654 int flags; 1655 struct thread *td; 1656 }; 1657 1658 static int getvfsstat_callback(struct mount *, void *); 1659 1660 int 1661 sys_getvfsstat(struct sysmsg *sysmsg, const struct getvfsstat_args *uap) 1662 { 1663 struct thread *td = curthread; 1664 struct getvfsstat_info info; 1665 1666 bzero(&info, sizeof(info)); 1667 1668 info.maxcount = uap->vbufsize / sizeof(struct statvfs); 1669 info.sfsp = uap->buf; 1670 info.vsfsp = uap->vbuf; 1671 info.count = 0; 1672 info.flags = uap->flags; 1673 info.td = td; 1674 1675 mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD); 1676 if (info.vsfsp && info.count > info.maxcount) 1677 sysmsg->sysmsg_result = info.maxcount; 1678 else 1679 sysmsg->sysmsg_result = info.count; 1680 return (info.error); 1681 } 1682 1683 static int 1684 getvfsstat_callback(struct mount *mp, void *data) 1685 { 1686 struct getvfsstat_info *info = data; 1687 struct statfs *sp; 1688 struct statvfs *vsp; 1689 char *freepath; 1690 char *fullpath; 1691 int error; 1692 1693 if (info->td->td_proc && !chroot_visible_mnt(mp, info->td->td_proc)) 1694 return(0); 1695 1696 if (info->vsfsp && info->count < info->maxcount) { 1697 sp = &mp->mnt_stat; 1698 vsp = &mp->mnt_vstat; 1699 1700 /* 1701 * If MNT_NOWAIT or MNT_LAZY is specified, do not 1702 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY 1703 * overrides MNT_WAIT. 1704 * 1705 * Ignore refresh error, user should have visibility. 1706 * This can happen if a NFS mount goes bad (e.g. server 1707 * revokes perms or goes down). 1708 */ 1709 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1710 (info->flags & MNT_WAIT)) && 1711 (error = VFS_STATFS(mp, sp, info->td->td_ucred))) { 1712 /* ignore error */ 1713 } 1714 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 1715 1716 if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 || 1717 (info->flags & MNT_WAIT)) && 1718 (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) { 1719 /* ignore error */ 1720 } 1721 vsp->f_flag = 0; 1722 if (mp->mnt_flag & MNT_RDONLY) 1723 vsp->f_flag |= ST_RDONLY; 1724 if (mp->mnt_flag & MNT_NOSUID) 1725 vsp->f_flag |= ST_NOSUID; 1726 1727 error = mount_path(info->td->td_proc, mp, &fullpath, &freepath); 1728 if (error) { 1729 info->error = error; 1730 return(-1); 1731 } 1732 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 1733 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 1734 kfree(freepath, M_TEMP); 1735 1736 error = copyout(sp, info->sfsp, sizeof(*sp)); 1737 if (error == 0) 1738 error = copyout(vsp, info->vsfsp, sizeof(*vsp)); 1739 if (error) { 1740 info->error = error; 1741 return (-1); 1742 } 1743 ++info->sfsp; 1744 ++info->vsfsp; 1745 } 1746 info->count++; 1747 return(0); 1748 } 1749 1750 1751 /* 1752 * fchdir_args(int fd) 1753 * 1754 * Change current working directory to a given file descriptor. 1755 */ 1756 int 1757 sys_fchdir(struct sysmsg *sysmsg, const struct fchdir_args *uap) 1758 { 1759 struct thread *td = curthread; 1760 struct proc *p = td->td_proc; 1761 struct filedesc *fdp = p->p_fd; 1762 struct vnode *vp, *ovp; 1763 struct mount *mp; 1764 struct file *fp; 1765 struct nchandle nch, onch, tnch; 1766 int error; 1767 1768 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 1769 return (error); 1770 lwkt_gettoken(&p->p_token); 1771 vp = (struct vnode *)fp->f_data; 1772 vref(vp); 1773 vn_lock(vp, LK_SHARED | LK_RETRY); 1774 if (fp->f_nchandle.ncp == NULL) 1775 error = ENOTDIR; 1776 else 1777 error = checkvp_chdir(vp, td); 1778 if (error) { 1779 vput(vp); 1780 goto done; 1781 } 1782 cache_copy(&fp->f_nchandle, &nch); 1783 1784 /* 1785 * If the ncp has become a mount point, traverse through 1786 * the mount point. 1787 */ 1788 1789 while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) && 1790 (mp = cache_findmount(&nch)) != NULL 1791 ) { 1792 error = nlookup_mp(mp, &tnch); 1793 if (error == 0) { 1794 cache_unlock(&tnch); /* leave ref intact */ 1795 vput(vp); 1796 vp = tnch.ncp->nc_vp; 1797 error = vget(vp, LK_SHARED); 1798 KKASSERT(error == 0); 1799 cache_drop(&nch); 1800 nch = tnch; 1801 } 1802 cache_dropmount(mp); 1803 } 1804 if (error == 0) { 1805 spin_lock(&fdp->fd_spin); 1806 ovp = fdp->fd_cdir; 1807 onch = fdp->fd_ncdir; 1808 fdp->fd_cdir = vp; 1809 fdp->fd_ncdir = nch; 1810 spin_unlock(&fdp->fd_spin); 1811 vn_unlock(vp); /* leave ref intact */ 1812 cache_drop(&onch); 1813 vrele(ovp); 1814 } else { 1815 cache_drop(&nch); 1816 vput(vp); 1817 } 1818 fdrop(fp); 1819 done: 1820 lwkt_reltoken(&p->p_token); 1821 return (error); 1822 } 1823 1824 int 1825 kern_chdir(struct nlookupdata *nd) 1826 { 1827 struct thread *td = curthread; 1828 struct proc *p = td->td_proc; 1829 struct filedesc *fdp = p->p_fd; 1830 struct vnode *vp, *ovp; 1831 struct nchandle onch; 1832 int error; 1833 1834 nd->nl_flags |= NLC_SHAREDLOCK; 1835 if ((error = nlookup(nd)) != 0) 1836 return (error); 1837 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 1838 return (ENOENT); 1839 if ((error = vget(vp, LK_SHARED)) != 0) 1840 return (error); 1841 1842 lwkt_gettoken(&p->p_token); 1843 error = checkvp_chdir(vp, td); 1844 vn_unlock(vp); 1845 if (error == 0) { 1846 spin_lock(&fdp->fd_spin); 1847 ovp = fdp->fd_cdir; 1848 onch = fdp->fd_ncdir; 1849 fdp->fd_ncdir = nd->nl_nch; 1850 fdp->fd_cdir = vp; 1851 spin_unlock(&fdp->fd_spin); 1852 cache_unlock(&nd->nl_nch); /* leave reference intact */ 1853 cache_drop(&onch); 1854 vrele(ovp); 1855 cache_zero(&nd->nl_nch); 1856 } else { 1857 vrele(vp); 1858 } 1859 lwkt_reltoken(&p->p_token); 1860 return (error); 1861 } 1862 1863 /* 1864 * chdir_args(char *path) 1865 * 1866 * Change current working directory (``.''). 1867 */ 1868 int 1869 sys_chdir(struct sysmsg *sysmsg, const struct chdir_args *uap) 1870 { 1871 struct nlookupdata nd; 1872 int error; 1873 1874 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1875 if (error == 0) 1876 error = kern_chdir(&nd); 1877 nlookup_done(&nd); 1878 return (error); 1879 } 1880 1881 /* 1882 * Helper function for raised chroot(2) security function: Refuse if 1883 * any filedescriptors are open directories. 1884 */ 1885 static int 1886 chroot_refuse_vdir_fds(thread_t td, struct filedesc *fdp) 1887 { 1888 struct vnode *vp; 1889 struct file *fp; 1890 int error; 1891 int fd; 1892 1893 for (fd = 0; fd < fdp->fd_nfiles ; fd++) { 1894 if ((error = holdvnode(td, fd, &fp)) != 0) 1895 continue; 1896 vp = (struct vnode *)fp->f_data; 1897 if (vp->v_type != VDIR) { 1898 fdrop(fp); 1899 continue; 1900 } 1901 fdrop(fp); 1902 return(EPERM); 1903 } 1904 return (0); 1905 } 1906 1907 /* 1908 * This sysctl determines if we will allow a process to chroot(2) if it 1909 * has a directory open: 1910 * 0: disallowed for all processes. 1911 * 1: allowed for processes that were not already chroot(2)'ed. 1912 * 2: allowed for all processes. 1913 */ 1914 1915 static int chroot_allow_open_directories = 1; 1916 1917 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, 1918 &chroot_allow_open_directories, 0, ""); 1919 1920 /* 1921 * chroot to the specified namecache entry. We obtain the vp from the 1922 * namecache data. The passed ncp must be locked and referenced and will 1923 * remain locked and referenced on return. 1924 */ 1925 int 1926 kern_chroot(struct nchandle *nch) 1927 { 1928 struct thread *td = curthread; 1929 struct proc *p = td->td_proc; 1930 struct filedesc *fdp = p->p_fd; 1931 struct vnode *vp; 1932 int error; 1933 1934 /* 1935 * Only privileged user can chroot 1936 */ 1937 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0); 1938 if (error) 1939 return (error); 1940 1941 /* 1942 * Disallow open directory descriptors (fchdir() breakouts). 1943 */ 1944 if (chroot_allow_open_directories == 0 || 1945 (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) { 1946 if ((error = chroot_refuse_vdir_fds(td, fdp)) != 0) 1947 return (error); 1948 } 1949 if ((vp = nch->ncp->nc_vp) == NULL) 1950 return (ENOENT); 1951 1952 if ((error = vget(vp, LK_SHARED)) != 0) 1953 return (error); 1954 1955 /* 1956 * Check the validity of vp as a directory to change to and 1957 * associate it with rdir/jdir. 1958 */ 1959 error = checkvp_chdir(vp, td); 1960 vn_unlock(vp); /* leave reference intact */ 1961 if (error == 0) { 1962 lwkt_gettoken(&p->p_token); 1963 vrele(fdp->fd_rdir); 1964 fdp->fd_rdir = vp; /* reference inherited by fd_rdir */ 1965 cache_drop(&fdp->fd_nrdir); 1966 cache_copy(nch, &fdp->fd_nrdir); 1967 if (fdp->fd_jdir == NULL) { 1968 fdp->fd_jdir = vp; 1969 vref(fdp->fd_jdir); 1970 cache_copy(nch, &fdp->fd_njdir); 1971 } 1972 if ((p->p_flags & P_DIDCHROOT) == 0) { 1973 p->p_flags |= P_DIDCHROOT; 1974 if (p->p_depth <= 65535 - 32) 1975 p->p_depth += 32; 1976 } 1977 lwkt_reltoken(&p->p_token); 1978 } else { 1979 vrele(vp); 1980 } 1981 return (error); 1982 } 1983 1984 /* 1985 * chroot_args(char *path) 1986 * 1987 * Change notion of root (``/'') directory. 1988 */ 1989 int 1990 sys_chroot(struct sysmsg *sysmsg, const struct chroot_args *uap) 1991 { 1992 struct thread *td __debugvar = curthread; 1993 struct nlookupdata nd; 1994 int error; 1995 1996 KKASSERT(td->td_proc); 1997 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 1998 if (error == 0) { 1999 nd.nl_flags |= NLC_EXEC; 2000 error = nlookup(&nd); 2001 if (error == 0) 2002 error = kern_chroot(&nd.nl_nch); 2003 } 2004 nlookup_done(&nd); 2005 return(error); 2006 } 2007 2008 int 2009 sys_chroot_kernel(struct sysmsg *sysmsg, const struct chroot_kernel_args *uap) 2010 { 2011 struct thread *td = curthread; 2012 struct nlookupdata nd; 2013 struct nchandle *nch; 2014 struct vnode *vp; 2015 int error; 2016 2017 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2018 if (error) 2019 goto error_nond; 2020 2021 error = nlookup(&nd); 2022 if (error) 2023 goto error_out; 2024 2025 nch = &nd.nl_nch; 2026 2027 error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0); 2028 if (error) 2029 goto error_out; 2030 2031 if ((vp = nch->ncp->nc_vp) == NULL) { 2032 error = ENOENT; 2033 goto error_out; 2034 } 2035 2036 if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0) 2037 goto error_out; 2038 2039 vfs_cache_setroot(vp, cache_hold(nch)); 2040 2041 error_out: 2042 nlookup_done(&nd); 2043 error_nond: 2044 return(error); 2045 } 2046 2047 /* 2048 * Common routine for chroot and chdir. Given a locked, referenced vnode, 2049 * determine whether it is legal to chdir to the vnode. The vnode's state 2050 * is not changed by this call. 2051 */ 2052 static int 2053 checkvp_chdir(struct vnode *vp, struct thread *td) 2054 { 2055 int error; 2056 2057 if (vp->v_type != VDIR) 2058 error = ENOTDIR; 2059 else 2060 error = VOP_EACCESS(vp, VEXEC, td->td_ucred); 2061 return (error); 2062 } 2063 2064 int 2065 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res) 2066 { 2067 struct thread *td = curthread; 2068 struct proc *p = td->td_proc; 2069 struct lwp *lp = td->td_lwp; 2070 struct filedesc *fdp = p->p_fd; 2071 int cmode, flags; 2072 struct file *nfp; 2073 struct file *fp; 2074 int type, indx, error = 0; 2075 struct flock lf; 2076 2077 if ((oflags & O_ACCMODE) == O_ACCMODE) 2078 return (EINVAL); 2079 flags = FFLAGS(oflags); 2080 error = falloc(lp, &nfp, NULL); 2081 if (error) 2082 return (error); 2083 fp = nfp; 2084 cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; 2085 2086 /* 2087 * Call vn_open() to do the lookup and assign the vnode to the 2088 * file pointer. vn_open() does not change the ref count on fp 2089 * and the vnode, on success, will be inherited by the file pointer 2090 * and unlocked. 2091 * 2092 * Request a shared lock on the vnode if possible. 2093 * 2094 * When NLC_SHAREDLOCK is set we may still need an exclusive vnode 2095 * lock for O_RDWR opens on executables in order to avoid a VTEXT 2096 * detection race. The NLC_EXCLLOCK_IFEXEC handles this case. 2097 * 2098 * NOTE: We need a flag to separate terminal vnode locking from 2099 * parent locking. O_CREAT needs parent locking, but O_TRUNC 2100 * and O_RDWR only need to lock the terminal vnode exclusively. 2101 */ 2102 nd->nl_flags |= NLC_LOCKVP; 2103 if ((flags & (O_CREAT|O_TRUNC)) == 0) { 2104 nd->nl_flags |= NLC_SHAREDLOCK; 2105 if (flags & O_RDWR) 2106 nd->nl_flags |= NLC_EXCLLOCK_IFEXEC; 2107 } 2108 2109 /* 2110 * Issue the vn_open, passing in the referenced fp. the vn_open() 2111 * is allowed to replace fp by fdrop()ing it and returning its own 2112 * referenced fp. 2113 */ 2114 nfp = fp; 2115 error = vn_open(nd, &nfp, flags, cmode); 2116 fp = nfp; 2117 nlookup_done(nd); 2118 2119 /* 2120 * Deal with any error condition 2121 */ 2122 if (error) { 2123 fdrop(fp); /* our ref */ 2124 if (error == ERESTART) 2125 error = EINTR; 2126 return (error); 2127 } 2128 2129 /* 2130 * Reserve a file descriptor. 2131 */ 2132 if ((error = fdalloc(p, 0, &indx)) != 0) { 2133 fdrop(fp); 2134 return (error); 2135 } 2136 2137 /* 2138 * Handle advisory lock flags. This is only supported with vnodes. 2139 * For things like /dev/fd/N we might not actually get a vnode. 2140 */ 2141 if ((flags & (O_EXLOCK | O_SHLOCK)) && fp->f_type == DTYPE_VNODE) { 2142 struct vnode *vp; 2143 2144 vp = (struct vnode *)fp->f_data; 2145 vref(vp); 2146 2147 lf.l_whence = SEEK_SET; 2148 lf.l_start = 0; 2149 lf.l_len = 0; 2150 if (flags & O_EXLOCK) 2151 lf.l_type = F_WRLCK; 2152 else 2153 lf.l_type = F_RDLCK; 2154 if (flags & FNONBLOCK) 2155 type = 0; 2156 else 2157 type = F_WAIT; 2158 2159 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); 2160 if (error) { 2161 /* 2162 * lock request failed. Clean up the reserved 2163 * descriptor. 2164 */ 2165 vrele(vp); 2166 fsetfd(fdp, NULL, indx); 2167 fdrop(fp); 2168 return (error); 2169 } 2170 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */ 2171 vrele(vp); 2172 } 2173 2174 /* 2175 * release our private reference, leaving the one associated with the 2176 * descriptor table intact. 2177 */ 2178 if (oflags & O_CLOEXEC) 2179 fdp->fd_files[indx].fileflags |= UF_EXCLOSE; 2180 fsetfd(fdp, fp, indx); 2181 fdrop(fp); 2182 *res = indx; 2183 2184 return (error); 2185 } 2186 2187 /* 2188 * open_args(char *path, int flags, int mode) 2189 * 2190 * Check permissions, allocate an open file structure, 2191 * and call the device open routine if any. 2192 */ 2193 int 2194 sys_open(struct sysmsg *sysmsg, const struct open_args *uap) 2195 { 2196 struct nlookupdata nd; 2197 int error; 2198 2199 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2200 if (error == 0) { 2201 error = kern_open(&nd, uap->flags, 2202 uap->mode, &sysmsg->sysmsg_result); 2203 } 2204 nlookup_done(&nd); 2205 return (error); 2206 } 2207 2208 /* 2209 * openat_args(int fd, char *path, int flags, int mode) 2210 */ 2211 int 2212 sys_openat(struct sysmsg *sysmsg, const struct openat_args *uap) 2213 { 2214 struct nlookupdata nd; 2215 int error; 2216 struct file *fp; 2217 2218 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2219 if (error == 0) { 2220 error = kern_open(&nd, uap->flags, uap->mode, 2221 &sysmsg->sysmsg_result); 2222 } 2223 nlookup_done_at(&nd, fp); 2224 return (error); 2225 } 2226 2227 int 2228 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor) 2229 { 2230 struct thread *td = curthread; 2231 struct proc *p = td->td_proc; 2232 struct vnode *vp; 2233 struct vattr vattr; 2234 int error; 2235 int whiteout = 0; 2236 2237 KKASSERT(p); 2238 2239 VATTR_NULL(&vattr); 2240 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2241 vattr.va_rmajor = rmajor; 2242 vattr.va_rminor = rminor; 2243 2244 switch (mode & S_IFMT) { 2245 case S_IFMT: /* used by badsect to flag bad sectors */ 2246 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0); 2247 vattr.va_type = VBAD; 2248 break; 2249 case S_IFCHR: 2250 error = priv_check(td, PRIV_VFS_MKNOD_DEV); 2251 vattr.va_type = VCHR; 2252 break; 2253 case S_IFBLK: 2254 error = priv_check(td, PRIV_VFS_MKNOD_DEV); 2255 vattr.va_type = VBLK; 2256 break; 2257 case S_IFWHT: 2258 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0); 2259 whiteout = 1; 2260 break; 2261 case S_IFDIR: /* special directories support for HAMMER */ 2262 error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0); 2263 vattr.va_type = VDIR; 2264 break; 2265 default: 2266 error = EINVAL; 2267 break; 2268 } 2269 2270 if (error) 2271 return (error); 2272 2273 bwillinode(1); 2274 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2275 if ((error = nlookup(nd)) != 0) 2276 return (error); 2277 if (nd->nl_nch.ncp->nc_vp) 2278 return (EEXIST); 2279 if (nd->nl_dvp == NULL) 2280 return (EINVAL); 2281 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2282 return (error); 2283 2284 if (whiteout) { 2285 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp, 2286 nd->nl_cred, NAMEI_CREATE); 2287 } else { 2288 vp = NULL; 2289 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, 2290 &vp, nd->nl_cred, &vattr); 2291 if (error == 0) 2292 vput(vp); 2293 } 2294 return (error); 2295 } 2296 2297 /* 2298 * mknod_args(char *path, int mode, int dev) 2299 * 2300 * Create a special file. 2301 */ 2302 int 2303 sys_mknod(struct sysmsg *sysmsg, const struct mknod_args *uap) 2304 { 2305 struct nlookupdata nd; 2306 int error; 2307 2308 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2309 if (error == 0) { 2310 error = kern_mknod(&nd, uap->mode, 2311 umajor(uap->dev), uminor(uap->dev)); 2312 } 2313 nlookup_done(&nd); 2314 return (error); 2315 } 2316 2317 /* 2318 * mknodat_args(int fd, char *path, mode_t mode, dev_t dev) 2319 * 2320 * Create a special file. The path is relative to the directory associated 2321 * with fd. 2322 */ 2323 int 2324 sys_mknodat(struct sysmsg *sysmsg, const struct mknodat_args *uap) 2325 { 2326 struct nlookupdata nd; 2327 struct file *fp; 2328 int error; 2329 2330 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2331 if (error == 0) { 2332 error = kern_mknod(&nd, uap->mode, 2333 umajor(uap->dev), uminor(uap->dev)); 2334 } 2335 nlookup_done_at(&nd, fp); 2336 return (error); 2337 } 2338 2339 int 2340 kern_mkfifo(struct nlookupdata *nd, int mode) 2341 { 2342 struct thread *td = curthread; 2343 struct proc *p = td->td_proc; 2344 struct vattr vattr; 2345 struct vnode *vp; 2346 int error; 2347 2348 bwillinode(1); 2349 2350 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2351 if ((error = nlookup(nd)) != 0) 2352 return (error); 2353 if (nd->nl_nch.ncp->nc_vp) 2354 return (EEXIST); 2355 if (nd->nl_dvp == NULL) 2356 return (EINVAL); 2357 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2358 return (error); 2359 2360 VATTR_NULL(&vattr); 2361 vattr.va_type = VFIFO; 2362 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2363 vp = NULL; 2364 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr); 2365 if (error == 0) 2366 vput(vp); 2367 return (error); 2368 } 2369 2370 /* 2371 * mkfifo_args(char *path, int mode) 2372 * 2373 * Create a named pipe. 2374 */ 2375 int 2376 sys_mkfifo(struct sysmsg *sysmsg, const struct mkfifo_args *uap) 2377 { 2378 struct nlookupdata nd; 2379 int error; 2380 2381 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2382 if (error == 0) 2383 error = kern_mkfifo(&nd, uap->mode); 2384 nlookup_done(&nd); 2385 return (error); 2386 } 2387 2388 /* 2389 * mkfifoat_args(int fd, char *path, mode_t mode) 2390 * 2391 * Create a named pipe. The path is relative to the directory associated 2392 * with fd. 2393 */ 2394 int 2395 sys_mkfifoat(struct sysmsg *sysmsg, const struct mkfifoat_args *uap) 2396 { 2397 struct nlookupdata nd; 2398 struct file *fp; 2399 int error; 2400 2401 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2402 if (error == 0) 2403 error = kern_mkfifo(&nd, uap->mode); 2404 nlookup_done_at(&nd, fp); 2405 return (error); 2406 } 2407 2408 static int hardlink_check_uid = 0; 2409 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, 2410 &hardlink_check_uid, 0, 2411 "Unprivileged processes cannot create hard links to files owned by other " 2412 "users"); 2413 static int hardlink_check_gid = 0; 2414 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, 2415 &hardlink_check_gid, 0, 2416 "Unprivileged processes cannot create hard links to files owned by other " 2417 "groups"); 2418 2419 static int 2420 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred) 2421 { 2422 struct vattr va; 2423 int error; 2424 2425 /* 2426 * Shortcut if disabled 2427 */ 2428 if (hardlink_check_uid == 0 && hardlink_check_gid == 0) 2429 return (0); 2430 2431 /* 2432 * Privileged user can always hardlink 2433 */ 2434 if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0) 2435 return (0); 2436 2437 /* 2438 * Otherwise only if the originating file is owned by the 2439 * same user or group. Note that any group is allowed if 2440 * the file is owned by the caller. 2441 */ 2442 error = VOP_GETATTR(vp, &va); 2443 if (error != 0) 2444 return (error); 2445 2446 if (hardlink_check_uid) { 2447 if (cred->cr_uid != va.va_uid) 2448 return (EPERM); 2449 } 2450 2451 if (hardlink_check_gid) { 2452 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred)) 2453 return (EPERM); 2454 } 2455 2456 return (0); 2457 } 2458 2459 int 2460 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd) 2461 { 2462 struct thread *td = curthread; 2463 struct vnode *vp; 2464 int error; 2465 2466 /* 2467 * Lookup the source and obtained a locked vnode. 2468 * 2469 * You may only hardlink a file which you have write permission 2470 * on or which you own. 2471 * 2472 * XXX relookup on vget failure / race ? 2473 */ 2474 bwillinode(1); 2475 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK; 2476 if ((error = nlookup(nd)) != 0) 2477 return (error); 2478 vp = nd->nl_nch.ncp->nc_vp; 2479 KKASSERT(vp != NULL); 2480 if (vp->v_type == VDIR) 2481 return (EPERM); /* POSIX */ 2482 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2483 return (error); 2484 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) 2485 return (error); 2486 2487 /* 2488 * Unlock the source so we can lookup the target without deadlocking 2489 * (XXX vp is locked already, possible other deadlock?). The target 2490 * must not exist. 2491 */ 2492 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 2493 nd->nl_flags &= ~NLC_NCPISLOCKED; 2494 cache_unlock(&nd->nl_nch); 2495 vn_unlock(vp); 2496 2497 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2498 if ((error = nlookup(linknd)) != 0) { 2499 vrele(vp); 2500 return (error); 2501 } 2502 if (linknd->nl_nch.ncp->nc_vp) { 2503 vrele(vp); 2504 return (EEXIST); 2505 } 2506 if (linknd->nl_dvp == NULL) { 2507 vrele(vp); 2508 return (EINVAL); 2509 } 2510 VFS_MODIFYING(vp->v_mount); 2511 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 2512 if (error) { 2513 vrele(vp); 2514 return (error); 2515 } 2516 2517 /* 2518 * Finally run the new API VOP. 2519 */ 2520 error = can_hardlink(vp, td, td->td_ucred); 2521 if (error == 0) { 2522 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp, 2523 vp, linknd->nl_cred); 2524 } 2525 vput(vp); 2526 return (error); 2527 } 2528 2529 /* 2530 * link_args(char *path, char *link) 2531 * 2532 * Make a hard file link. 2533 */ 2534 int 2535 sys_link(struct sysmsg *sysmsg, const struct link_args *uap) 2536 { 2537 struct nlookupdata nd, linknd; 2538 int error; 2539 2540 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2541 if (error == 0) { 2542 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0); 2543 if (error == 0) 2544 error = kern_link(&nd, &linknd); 2545 nlookup_done(&linknd); 2546 } 2547 nlookup_done(&nd); 2548 return (error); 2549 } 2550 2551 /* 2552 * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags) 2553 * 2554 * Make a hard file link. The path1 argument is relative to the directory 2555 * associated with fd1, and similarly the path2 argument is relative to 2556 * the directory associated with fd2. 2557 */ 2558 int 2559 sys_linkat(struct sysmsg *sysmsg, const struct linkat_args *uap) 2560 { 2561 struct nlookupdata nd, linknd; 2562 struct file *fp1, *fp2; 2563 int error; 2564 2565 error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE, 2566 (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0); 2567 if (error == 0) { 2568 error = nlookup_init_at(&linknd, &fp2, uap->fd2, 2569 uap->path2, UIO_USERSPACE, 0); 2570 if (error == 0) 2571 error = kern_link(&nd, &linknd); 2572 nlookup_done_at(&linknd, fp2); 2573 } 2574 nlookup_done_at(&nd, fp1); 2575 return (error); 2576 } 2577 2578 int 2579 kern_symlink(struct nlookupdata *nd, char *path, int mode) 2580 { 2581 struct vattr vattr; 2582 struct vnode *vp; 2583 struct vnode *dvp; 2584 int error; 2585 2586 bwillinode(1); 2587 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2588 if ((error = nlookup(nd)) != 0) 2589 return (error); 2590 if (nd->nl_nch.ncp->nc_vp) 2591 return (EEXIST); 2592 if (nd->nl_dvp == NULL) 2593 return (EINVAL); 2594 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2595 return (error); 2596 dvp = nd->nl_dvp; 2597 VATTR_NULL(&vattr); 2598 vattr.va_mode = mode; 2599 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path); 2600 if (error == 0) 2601 vput(vp); 2602 return (error); 2603 } 2604 2605 /* 2606 * symlink(char *path, char *link) 2607 * 2608 * Make a symbolic link. 2609 */ 2610 int 2611 sys_symlink(struct sysmsg *sysmsg, const struct symlink_args *uap) 2612 { 2613 struct thread *td = curthread; 2614 struct nlookupdata nd; 2615 char *path; 2616 int error; 2617 int mode; 2618 2619 path = objcache_get(namei_oc, M_WAITOK); 2620 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 2621 if (error == 0) { 2622 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0); 2623 if (error == 0) { 2624 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2625 error = kern_symlink(&nd, path, mode); 2626 } 2627 nlookup_done(&nd); 2628 } 2629 objcache_put(namei_oc, path); 2630 return (error); 2631 } 2632 2633 /* 2634 * symlinkat_args(char *path1, int fd, char *path2) 2635 * 2636 * Make a symbolic link. The path2 argument is relative to the directory 2637 * associated with fd. 2638 */ 2639 int 2640 sys_symlinkat(struct sysmsg *sysmsg, const struct symlinkat_args *uap) 2641 { 2642 struct thread *td = curthread; 2643 struct nlookupdata nd; 2644 struct file *fp; 2645 char *path1; 2646 int error; 2647 int mode; 2648 2649 path1 = objcache_get(namei_oc, M_WAITOK); 2650 error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL); 2651 if (error == 0) { 2652 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2, 2653 UIO_USERSPACE, 0); 2654 if (error == 0) { 2655 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2656 error = kern_symlink(&nd, path1, mode); 2657 } 2658 nlookup_done_at(&nd, fp); 2659 } 2660 objcache_put(namei_oc, path1); 2661 return (error); 2662 } 2663 2664 /* 2665 * undelete_args(char *path) 2666 * 2667 * Delete a whiteout from the filesystem. 2668 */ 2669 int 2670 sys_undelete(struct sysmsg *sysmsg, const struct undelete_args *uap) 2671 { 2672 struct nlookupdata nd; 2673 int error; 2674 2675 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2676 bwillinode(1); 2677 nd.nl_flags |= NLC_DELETE | NLC_REFDVP; 2678 if (error == 0) 2679 error = nlookup(&nd); 2680 if (error == 0 && nd.nl_dvp == NULL) 2681 error = EINVAL; 2682 if (error == 0) 2683 error = ncp_writechk(&nd.nl_nch); 2684 if (error == 0) { 2685 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred, 2686 NAMEI_DELETE); 2687 } 2688 nlookup_done(&nd); 2689 return (error); 2690 } 2691 2692 int 2693 kern_unlink(struct nlookupdata *nd) 2694 { 2695 int error; 2696 2697 bwillinode(1); 2698 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 2699 if ((error = nlookup(nd)) != 0) 2700 return (error); 2701 if (nd->nl_dvp == NULL) 2702 return EINVAL; 2703 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2704 return (error); 2705 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 2706 return (error); 2707 } 2708 2709 /* 2710 * unlink_args(char *path) 2711 * 2712 * Delete a name from the filesystem. 2713 */ 2714 int 2715 sys_unlink(struct sysmsg *sysmsg, const struct unlink_args *uap) 2716 { 2717 struct nlookupdata nd; 2718 int error; 2719 2720 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2721 if (error == 0) 2722 error = kern_unlink(&nd); 2723 nlookup_done(&nd); 2724 return (error); 2725 } 2726 2727 2728 /* 2729 * unlinkat_args(int fd, char *path, int flags) 2730 * 2731 * Delete the file or directory entry pointed to by fd/path. 2732 */ 2733 int 2734 sys_unlinkat(struct sysmsg *sysmsg, const struct unlinkat_args *uap) 2735 { 2736 struct nlookupdata nd; 2737 struct file *fp; 2738 int error; 2739 2740 if (uap->flags & ~AT_REMOVEDIR) 2741 return (EINVAL); 2742 2743 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2744 if (error == 0) { 2745 if (uap->flags & AT_REMOVEDIR) 2746 error = kern_rmdir(&nd); 2747 else 2748 error = kern_unlink(&nd); 2749 } 2750 nlookup_done_at(&nd, fp); 2751 return (error); 2752 } 2753 2754 int 2755 kern_lseek(int fd, off_t offset, int whence, off_t *res) 2756 { 2757 struct thread *td = curthread; 2758 struct file *fp; 2759 struct vnode *vp; 2760 struct vattr_lite lva; 2761 off_t new_offset; 2762 int error; 2763 2764 fp = holdfp(td, fd, -1); 2765 if (fp == NULL) 2766 return (EBADF); 2767 if (fp->f_type != DTYPE_VNODE) { 2768 error = ESPIPE; 2769 goto done; 2770 } 2771 vp = (struct vnode *)fp->f_data; 2772 2773 switch (whence) { 2774 case L_INCR: 2775 spin_lock(&fp->f_spin); 2776 new_offset = fp->f_offset + offset; 2777 error = 0; 2778 break; 2779 case L_XTND: 2780 error = VOP_GETATTR_LITE(vp, &lva); 2781 spin_lock(&fp->f_spin); 2782 new_offset = offset + lva.va_size; 2783 break; 2784 case L_SET: 2785 new_offset = offset; 2786 error = 0; 2787 spin_lock(&fp->f_spin); 2788 break; 2789 default: 2790 new_offset = 0; 2791 error = EINVAL; 2792 spin_lock(&fp->f_spin); 2793 break; 2794 } 2795 2796 /* 2797 * Validate the seek position. Negative offsets are not allowed 2798 * for regular files or directories. 2799 * 2800 * Normally we would also not want to allow negative offsets for 2801 * character and block-special devices. However kvm addresses 2802 * on 64 bit architectures might appear to be negative and must 2803 * be allowed. 2804 */ 2805 if (error == 0) { 2806 if (new_offset < 0 && 2807 (vp->v_type == VREG || vp->v_type == VDIR)) { 2808 error = EINVAL; 2809 } else { 2810 fp->f_offset = new_offset; 2811 } 2812 } 2813 *res = fp->f_offset; 2814 spin_unlock(&fp->f_spin); 2815 done: 2816 dropfp(td, fd, fp); 2817 2818 return (error); 2819 } 2820 2821 /* 2822 * lseek_args(int fd, int pad, off_t offset, int whence) 2823 * 2824 * Reposition read/write file offset. 2825 */ 2826 int 2827 sys_lseek(struct sysmsg *sysmsg, const struct lseek_args *uap) 2828 { 2829 int error; 2830 2831 error = kern_lseek(uap->fd, uap->offset, uap->whence, 2832 &sysmsg->sysmsg_offset); 2833 2834 return (error); 2835 } 2836 2837 /* 2838 * Check if current process can access given file. amode is a bitmask of *_OK 2839 * access bits. flags is a bitmask of AT_* flags. 2840 */ 2841 int 2842 kern_access(struct nlookupdata *nd, int amode, int flags) 2843 { 2844 struct vnode *vp; 2845 int error, mode; 2846 2847 if (flags & ~AT_EACCESS) 2848 return (EINVAL); 2849 nd->nl_flags |= NLC_SHAREDLOCK; 2850 if ((error = nlookup(nd)) != 0) 2851 return (error); 2852 if ((amode & W_OK) && (error = ncp_writechk(&nd->nl_nch)) != 0) 2853 return (error); 2854 retry: 2855 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp); 2856 if (error) 2857 return (error); 2858 2859 /* Flags == 0 means only check for existence. */ 2860 if (amode) { 2861 mode = 0; 2862 if (amode & R_OK) 2863 mode |= VREAD; 2864 if (amode & W_OK) 2865 mode |= VWRITE; 2866 if (amode & X_OK) 2867 mode |= VEXEC; 2868 if ((mode & VWRITE) == 0 || 2869 (error = vn_writechk(vp)) == 0) { 2870 error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred); 2871 } 2872 2873 /* 2874 * If the file handle is stale we have to re-resolve the 2875 * entry with the ncp held exclusively. This is a hack 2876 * at the moment. 2877 */ 2878 if (error == ESTALE) { 2879 vput(vp); 2880 cache_unlock(&nd->nl_nch); 2881 cache_lock(&nd->nl_nch); 2882 cache_setunresolved(&nd->nl_nch); 2883 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2884 if (error == 0) { 2885 vp = NULL; 2886 goto retry; 2887 } 2888 return(error); 2889 } 2890 } 2891 vput(vp); 2892 return (error); 2893 } 2894 2895 /* 2896 * access_args(char *path, int flags) 2897 * 2898 * Check access permissions. 2899 */ 2900 int 2901 sys_access(struct sysmsg *sysmsg, const struct access_args *uap) 2902 { 2903 struct nlookupdata nd; 2904 int error; 2905 2906 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2907 if (error == 0) 2908 error = kern_access(&nd, uap->flags, 0); 2909 nlookup_done(&nd); 2910 return (error); 2911 } 2912 2913 2914 /* 2915 * eaccess_args(char *path, int flags) 2916 * 2917 * Check access permissions. 2918 */ 2919 int 2920 sys_eaccess(struct sysmsg *sysmsg, const struct eaccess_args *uap) 2921 { 2922 struct nlookupdata nd; 2923 int error; 2924 2925 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2926 if (error == 0) 2927 error = kern_access(&nd, uap->flags, AT_EACCESS); 2928 nlookup_done(&nd); 2929 return (error); 2930 } 2931 2932 2933 /* 2934 * faccessat_args(int fd, char *path, int amode, int flags) 2935 * 2936 * Check access permissions. 2937 */ 2938 int 2939 sys_faccessat(struct sysmsg *sysmsg, const struct faccessat_args *uap) 2940 { 2941 struct nlookupdata nd; 2942 struct file *fp; 2943 int error; 2944 2945 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 2946 NLC_FOLLOW); 2947 if (error == 0) 2948 error = kern_access(&nd, uap->amode, uap->flags); 2949 nlookup_done_at(&nd, fp); 2950 return (error); 2951 } 2952 2953 int 2954 kern_stat(struct nlookupdata *nd, struct stat *st) 2955 { 2956 int error; 2957 struct vnode *vp; 2958 2959 nd->nl_flags |= NLC_SHAREDLOCK; 2960 if ((error = nlookup(nd)) != 0) 2961 return (error); 2962 again: 2963 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 2964 return (ENOENT); 2965 2966 #if 1 2967 error = cache_vref(&nd->nl_nch, NULL, &vp); 2968 #else 2969 error = vget(vp, LK_SHARED); 2970 #endif 2971 if (error) 2972 return (error); 2973 error = vn_stat(vp, st, nd->nl_cred); 2974 2975 /* 2976 * If the file handle is stale we have to re-resolve the 2977 * entry with the ncp held exclusively. This is a hack 2978 * at the moment. 2979 */ 2980 if (error == ESTALE) { 2981 #if 1 2982 vrele(vp); 2983 #else 2984 vput(vp); 2985 #endif 2986 cache_unlock(&nd->nl_nch); 2987 cache_lock(&nd->nl_nch); 2988 cache_setunresolved(&nd->nl_nch); 2989 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2990 if (error == 0) 2991 goto again; 2992 } else { 2993 #if 1 2994 vrele(vp); 2995 #else 2996 vput(vp); 2997 #endif 2998 } 2999 return (error); 3000 } 3001 3002 /* 3003 * stat_args(char *path, struct stat *ub) 3004 * 3005 * Get file status; this version follows links. 3006 */ 3007 int 3008 sys_stat(struct sysmsg *sysmsg, const struct stat_args *uap) 3009 { 3010 struct nlookupdata nd; 3011 struct stat st; 3012 int error; 3013 3014 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3015 if (error == 0) { 3016 error = kern_stat(&nd, &st); 3017 if (error == 0) 3018 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 3019 } 3020 nlookup_done(&nd); 3021 return (error); 3022 } 3023 3024 /* 3025 * lstat_args(char *path, struct stat *ub) 3026 * 3027 * Get file status; this version does not follow links. 3028 */ 3029 int 3030 sys_lstat(struct sysmsg *sysmsg, const struct lstat_args *uap) 3031 { 3032 struct nlookupdata nd; 3033 struct stat st; 3034 int error; 3035 3036 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3037 if (error == 0) { 3038 error = kern_stat(&nd, &st); 3039 if (error == 0) 3040 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 3041 } 3042 nlookup_done(&nd); 3043 return (error); 3044 } 3045 3046 /* 3047 * fstatat_args(int fd, char *path, struct stat *sb, int flags) 3048 * 3049 * Get status of file pointed to by fd/path. 3050 */ 3051 int 3052 sys_fstatat(struct sysmsg *sysmsg, const struct fstatat_args *uap) 3053 { 3054 struct nlookupdata nd; 3055 struct stat st; 3056 int error; 3057 int flags; 3058 struct file *fp; 3059 3060 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3061 return (EINVAL); 3062 3063 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3064 3065 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3066 UIO_USERSPACE, flags); 3067 if (error == 0) { 3068 error = kern_stat(&nd, &st); 3069 if (error == 0) 3070 error = copyout(&st, uap->sb, sizeof(*uap->sb)); 3071 } 3072 nlookup_done_at(&nd, fp); 3073 return (error); 3074 } 3075 3076 static int 3077 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp) 3078 { 3079 struct nlookupdata nd; 3080 struct vnode *vp; 3081 int error; 3082 3083 vp = NULL; 3084 error = nlookup_init(&nd, path, UIO_USERSPACE, flags); 3085 if (error == 0) 3086 error = nlookup(&nd); 3087 if (error == 0) 3088 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 3089 nlookup_done(&nd); 3090 if (error == 0) { 3091 error = VOP_PATHCONF(vp, name, sysmsg_regp); 3092 vput(vp); 3093 } 3094 return (error); 3095 } 3096 3097 /* 3098 * pathconf_Args(char *path, int name) 3099 * 3100 * Get configurable pathname variables. 3101 */ 3102 int 3103 sys_pathconf(struct sysmsg *sysmsg, const struct pathconf_args *uap) 3104 { 3105 return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW, 3106 &sysmsg->sysmsg_reg)); 3107 } 3108 3109 /* 3110 * lpathconf_Args(char *path, int name) 3111 * 3112 * Get configurable pathname variables, but don't follow symlinks. 3113 */ 3114 int 3115 sys_lpathconf(struct sysmsg *sysmsg, const struct lpathconf_args *uap) 3116 { 3117 return (kern_pathconf(uap->path, uap->name, 0, &sysmsg->sysmsg_reg)); 3118 } 3119 3120 /* 3121 * XXX: daver 3122 * kern_readlink isn't properly split yet. There is a copyin burried 3123 * in VOP_READLINK(). 3124 */ 3125 int 3126 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res) 3127 { 3128 struct thread *td = curthread; 3129 struct vnode *vp; 3130 struct iovec aiov; 3131 struct uio auio; 3132 int error; 3133 3134 nd->nl_flags |= NLC_SHAREDLOCK; 3135 if ((error = nlookup(nd)) != 0) 3136 return (error); 3137 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp); 3138 if (error) 3139 return (error); 3140 if (vp->v_type != VLNK) { 3141 error = EINVAL; 3142 } else { 3143 aiov.iov_base = buf; 3144 aiov.iov_len = count; 3145 auio.uio_iov = &aiov; 3146 auio.uio_iovcnt = 1; 3147 auio.uio_offset = 0; 3148 auio.uio_rw = UIO_READ; 3149 auio.uio_segflg = UIO_USERSPACE; 3150 auio.uio_td = td; 3151 auio.uio_resid = count; 3152 error = VOP_READLINK(vp, &auio, td->td_ucred); 3153 } 3154 vput(vp); 3155 *res = count - auio.uio_resid; 3156 return (error); 3157 } 3158 3159 /* 3160 * readlink_args(char *path, char *buf, int count) 3161 * 3162 * Return target name of a symbolic link. 3163 */ 3164 int 3165 sys_readlink(struct sysmsg *sysmsg, const struct readlink_args *uap) 3166 { 3167 struct nlookupdata nd; 3168 int error; 3169 3170 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3171 if (error == 0) { 3172 error = kern_readlink(&nd, uap->buf, uap->count, 3173 &sysmsg->sysmsg_result); 3174 } 3175 nlookup_done(&nd); 3176 return (error); 3177 } 3178 3179 /* 3180 * readlinkat_args(int fd, char *path, char *buf, size_t bufsize) 3181 * 3182 * Return target name of a symbolic link. The path is relative to the 3183 * directory associated with fd. 3184 */ 3185 int 3186 sys_readlinkat(struct sysmsg *sysmsg, const struct readlinkat_args *uap) 3187 { 3188 struct nlookupdata nd; 3189 struct file *fp; 3190 int error; 3191 3192 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 3193 if (error == 0) { 3194 error = kern_readlink(&nd, uap->buf, uap->bufsize, 3195 &sysmsg->sysmsg_result); 3196 } 3197 nlookup_done_at(&nd, fp); 3198 return (error); 3199 } 3200 3201 static int 3202 setfflags(struct vnode *vp, u_long flags) 3203 { 3204 struct thread *td = curthread; 3205 int error; 3206 struct vattr vattr; 3207 3208 /* 3209 * Prevent non-root users from setting flags on devices. When 3210 * a device is reused, users can retain ownership of the device 3211 * if they are allowed to set flags and programs assume that 3212 * chown can't fail when done as root. 3213 */ 3214 if ((vp->v_type == VCHR || vp->v_type == VBLK) && 3215 ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0)) 3216 return (error); 3217 3218 /* 3219 * note: vget is required for any operation that might mod the vnode 3220 * so VINACTIVE is properly cleared. 3221 */ 3222 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3223 VATTR_NULL(&vattr); 3224 vattr.va_flags = flags; 3225 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3226 vput(vp); 3227 } 3228 return (error); 3229 } 3230 3231 /* 3232 * chflags(const char *path, u_long flags) 3233 * 3234 * Change flags of a file given a path name. 3235 */ 3236 int 3237 sys_chflags(struct sysmsg *sysmsg, const struct chflags_args *uap) 3238 { 3239 struct nlookupdata nd; 3240 struct vnode *vp; 3241 int error; 3242 3243 vp = NULL; 3244 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3245 if (error == 0) 3246 error = nlookup(&nd); 3247 if (error == 0) 3248 error = ncp_writechk(&nd.nl_nch); 3249 if (error == 0) 3250 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3251 nlookup_done(&nd); 3252 if (error == 0) { 3253 error = setfflags(vp, uap->flags); 3254 vrele(vp); 3255 } 3256 return (error); 3257 } 3258 3259 /* 3260 * lchflags(const char *path, u_long flags) 3261 * 3262 * Change flags of a file given a path name, but don't follow symlinks. 3263 */ 3264 int 3265 sys_lchflags(struct sysmsg *sysmsg, const struct lchflags_args *uap) 3266 { 3267 struct nlookupdata nd; 3268 struct vnode *vp; 3269 int error; 3270 3271 vp = NULL; 3272 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3273 if (error == 0) 3274 error = nlookup(&nd); 3275 if (error == 0) 3276 error = ncp_writechk(&nd.nl_nch); 3277 if (error == 0) 3278 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3279 nlookup_done(&nd); 3280 if (error == 0) { 3281 error = setfflags(vp, uap->flags); 3282 vrele(vp); 3283 } 3284 return (error); 3285 } 3286 3287 /* 3288 * fchflags_args(int fd, u_flags flags) 3289 * 3290 * Change flags of a file given a file descriptor. 3291 */ 3292 int 3293 sys_fchflags(struct sysmsg *sysmsg, const struct fchflags_args *uap) 3294 { 3295 struct thread *td = curthread; 3296 struct file *fp; 3297 int error; 3298 3299 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 3300 return (error); 3301 if (fp->f_nchandle.ncp) 3302 error = ncp_writechk(&fp->f_nchandle); 3303 if (error == 0) 3304 error = setfflags((struct vnode *) fp->f_data, uap->flags); 3305 fdrop(fp); 3306 return (error); 3307 } 3308 3309 /* 3310 * chflagsat_args(int fd, const char *path, u_long flags, int atflags) 3311 * change flags given a pathname relative to a filedescriptor 3312 */ 3313 int 3314 sys_chflagsat(struct sysmsg *sysmsg, const struct chflagsat_args *uap) 3315 { 3316 struct nlookupdata nd; 3317 struct vnode *vp; 3318 struct file *fp; 3319 int error; 3320 int lookupflags; 3321 3322 if (uap->atflags & ~AT_SYMLINK_NOFOLLOW) 3323 return (EINVAL); 3324 3325 lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3326 3327 vp = NULL; 3328 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, lookupflags); 3329 if (error == 0) 3330 error = nlookup(&nd); 3331 if (error == 0) 3332 error = ncp_writechk(&nd.nl_nch); 3333 if (error == 0) 3334 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3335 nlookup_done_at(&nd, fp); 3336 if (error == 0) { 3337 error = setfflags(vp, uap->flags); 3338 vrele(vp); 3339 } 3340 return (error); 3341 } 3342 3343 3344 static int 3345 setfmode(struct vnode *vp, int mode) 3346 { 3347 struct thread *td = curthread; 3348 int error; 3349 struct vattr vattr; 3350 3351 /* 3352 * note: vget is required for any operation that might mod the vnode 3353 * so VINACTIVE is properly cleared. 3354 */ 3355 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3356 VATTR_NULL(&vattr); 3357 vattr.va_mode = mode & ALLPERMS; 3358 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3359 cache_inval_wxok(vp); 3360 vput(vp); 3361 } 3362 return error; 3363 } 3364 3365 int 3366 kern_chmod(struct nlookupdata *nd, int mode) 3367 { 3368 struct vnode *vp; 3369 int error; 3370 3371 if ((error = nlookup(nd)) != 0) 3372 return (error); 3373 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3374 return (error); 3375 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3376 error = setfmode(vp, mode); 3377 vrele(vp); 3378 return (error); 3379 } 3380 3381 /* 3382 * chmod_args(char *path, int mode) 3383 * 3384 * Change mode of a file given path name. 3385 */ 3386 int 3387 sys_chmod(struct sysmsg *sysmsg, const struct chmod_args *uap) 3388 { 3389 struct nlookupdata nd; 3390 int error; 3391 3392 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3393 if (error == 0) 3394 error = kern_chmod(&nd, uap->mode); 3395 nlookup_done(&nd); 3396 return (error); 3397 } 3398 3399 /* 3400 * lchmod_args(char *path, int mode) 3401 * 3402 * Change mode of a file given path name (don't follow links.) 3403 */ 3404 int 3405 sys_lchmod(struct sysmsg *sysmsg, const struct lchmod_args *uap) 3406 { 3407 struct nlookupdata nd; 3408 int error; 3409 3410 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3411 if (error == 0) 3412 error = kern_chmod(&nd, uap->mode); 3413 nlookup_done(&nd); 3414 return (error); 3415 } 3416 3417 /* 3418 * fchmod_args(int fd, int mode) 3419 * 3420 * Change mode of a file given a file descriptor. 3421 */ 3422 int 3423 sys_fchmod(struct sysmsg *sysmsg, const struct fchmod_args *uap) 3424 { 3425 struct thread *td = curthread; 3426 struct file *fp; 3427 int error; 3428 3429 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 3430 return (error); 3431 if (fp->f_nchandle.ncp) 3432 error = ncp_writechk(&fp->f_nchandle); 3433 if (error == 0) 3434 error = setfmode((struct vnode *)fp->f_data, uap->mode); 3435 fdrop(fp); 3436 return (error); 3437 } 3438 3439 /* 3440 * fchmodat_args(char *path, int mode) 3441 * 3442 * Change mode of a file pointed to by fd/path. 3443 */ 3444 int 3445 sys_fchmodat(struct sysmsg *sysmsg, const struct fchmodat_args *uap) 3446 { 3447 struct nlookupdata nd; 3448 struct file *fp; 3449 int error; 3450 int flags; 3451 3452 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3453 return (EINVAL); 3454 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3455 3456 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3457 UIO_USERSPACE, flags); 3458 if (error == 0) 3459 error = kern_chmod(&nd, uap->mode); 3460 nlookup_done_at(&nd, fp); 3461 return (error); 3462 } 3463 3464 static int 3465 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid) 3466 { 3467 struct thread *td = curthread; 3468 int error; 3469 struct vattr vattr; 3470 uid_t o_uid; 3471 gid_t o_gid; 3472 uint64_t size; 3473 3474 /* 3475 * note: vget is required for any operation that might mod the vnode 3476 * so VINACTIVE is properly cleared. 3477 */ 3478 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3479 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 3480 return error; 3481 o_uid = vattr.va_uid; 3482 o_gid = vattr.va_gid; 3483 size = vattr.va_size; 3484 3485 VATTR_NULL(&vattr); 3486 vattr.va_uid = uid; 3487 vattr.va_gid = gid; 3488 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3489 vput(vp); 3490 } 3491 3492 if (error == 0) { 3493 if (uid == -1) 3494 uid = o_uid; 3495 if (gid == -1) 3496 gid = o_gid; 3497 VFS_ACCOUNT(mp, o_uid, o_gid, -size); 3498 VFS_ACCOUNT(mp, uid, gid, size); 3499 } 3500 3501 return error; 3502 } 3503 3504 int 3505 kern_chown(struct nlookupdata *nd, int uid, int gid) 3506 { 3507 struct vnode *vp; 3508 int error; 3509 3510 if ((error = nlookup(nd)) != 0) 3511 return (error); 3512 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3513 return (error); 3514 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3515 error = setfown(nd->nl_nch.mount, vp, uid, gid); 3516 vrele(vp); 3517 return (error); 3518 } 3519 3520 /* 3521 * chown(char *path, int uid, int gid) 3522 * 3523 * Set ownership given a path name. 3524 */ 3525 int 3526 sys_chown(struct sysmsg *sysmsg, const struct chown_args *uap) 3527 { 3528 struct nlookupdata nd; 3529 int error; 3530 3531 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3532 if (error == 0) 3533 error = kern_chown(&nd, uap->uid, uap->gid); 3534 nlookup_done(&nd); 3535 return (error); 3536 } 3537 3538 /* 3539 * lchown_args(char *path, int uid, int gid) 3540 * 3541 * Set ownership given a path name, do not cross symlinks. 3542 */ 3543 int 3544 sys_lchown(struct sysmsg *sysmsg, const struct lchown_args *uap) 3545 { 3546 struct nlookupdata nd; 3547 int error; 3548 3549 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3550 if (error == 0) 3551 error = kern_chown(&nd, uap->uid, uap->gid); 3552 nlookup_done(&nd); 3553 return (error); 3554 } 3555 3556 /* 3557 * fchown_args(int fd, int uid, int gid) 3558 * 3559 * Set ownership given a file descriptor. 3560 */ 3561 int 3562 sys_fchown(struct sysmsg *sysmsg, const struct fchown_args *uap) 3563 { 3564 struct thread *td = curthread; 3565 struct proc *p = td->td_proc; 3566 struct file *fp; 3567 int error; 3568 3569 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 3570 return (error); 3571 if (fp->f_nchandle.ncp) 3572 error = ncp_writechk(&fp->f_nchandle); 3573 if (error == 0) 3574 error = setfown(p->p_fd->fd_ncdir.mount, 3575 (struct vnode *)fp->f_data, uap->uid, uap->gid); 3576 fdrop(fp); 3577 return (error); 3578 } 3579 3580 /* 3581 * fchownat(int fd, char *path, int uid, int gid, int flags) 3582 * 3583 * Set ownership of file pointed to by fd/path. 3584 */ 3585 int 3586 sys_fchownat(struct sysmsg *sysmsg, const struct fchownat_args *uap) 3587 { 3588 struct nlookupdata nd; 3589 struct file *fp; 3590 int error; 3591 int flags; 3592 3593 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3594 return (EINVAL); 3595 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3596 3597 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3598 UIO_USERSPACE, flags); 3599 if (error == 0) 3600 error = kern_chown(&nd, uap->uid, uap->gid); 3601 nlookup_done_at(&nd, fp); 3602 return (error); 3603 } 3604 3605 3606 static int 3607 getutimes(struct timeval *tvp, struct timespec *tsp) 3608 { 3609 struct timeval tv[2]; 3610 int error; 3611 3612 if (tvp == NULL) { 3613 microtime(&tv[0]); 3614 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]); 3615 tsp[1] = tsp[0]; 3616 } else { 3617 if ((error = itimerfix(tvp)) != 0) 3618 return (error); 3619 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); 3620 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); 3621 } 3622 return 0; 3623 } 3624 3625 static int 3626 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag) 3627 { 3628 struct timespec tsnow; 3629 int error; 3630 3631 *nullflag = 0; 3632 nanotime(&tsnow); 3633 if (ts == NULL) { 3634 newts[0] = tsnow; 3635 newts[1] = tsnow; 3636 *nullflag = 1; 3637 return (0); 3638 } 3639 3640 newts[0] = ts[0]; 3641 newts[1] = ts[1]; 3642 if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT) 3643 return (0); 3644 if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW) 3645 *nullflag = 1; 3646 3647 if (newts[0].tv_nsec == UTIME_OMIT) 3648 newts[0].tv_sec = VNOVAL; 3649 else if (newts[0].tv_nsec == UTIME_NOW) 3650 newts[0] = tsnow; 3651 else if ((error = itimespecfix(&newts[0])) != 0) 3652 return (error); 3653 3654 if (newts[1].tv_nsec == UTIME_OMIT) 3655 newts[1].tv_sec = VNOVAL; 3656 else if (newts[1].tv_nsec == UTIME_NOW) 3657 newts[1] = tsnow; 3658 else if ((error = itimespecfix(&newts[1])) != 0) 3659 return (error); 3660 3661 return (0); 3662 } 3663 3664 static int 3665 setutimes(struct vnode *vp, struct vattr *vattr, 3666 const struct timespec *ts, int nullflag) 3667 { 3668 struct thread *td = curthread; 3669 int error; 3670 3671 VATTR_NULL(vattr); 3672 vattr->va_atime = ts[0]; 3673 vattr->va_mtime = ts[1]; 3674 if (nullflag) 3675 vattr->va_vaflags |= VA_UTIMES_NULL; 3676 error = VOP_SETATTR(vp, vattr, td->td_ucred); 3677 3678 return error; 3679 } 3680 3681 int 3682 kern_utimes(struct nlookupdata *nd, struct timeval *tptr) 3683 { 3684 struct timespec ts[2]; 3685 int error; 3686 3687 if (tptr) { 3688 if ((error = getutimes(tptr, ts)) != 0) 3689 return (error); 3690 } 3691 error = kern_utimensat(nd, tptr ? ts : NULL, 0); 3692 return (error); 3693 } 3694 3695 /* 3696 * utimes_args(char *path, struct timeval *tptr) 3697 * 3698 * Set the access and modification times of a file. 3699 */ 3700 int 3701 sys_utimes(struct sysmsg *sysmsg, const struct utimes_args *uap) 3702 { 3703 struct timeval tv[2]; 3704 struct nlookupdata nd; 3705 int error; 3706 3707 if (uap->tptr) { 3708 error = copyin(uap->tptr, tv, sizeof(tv)); 3709 if (error) 3710 return (error); 3711 } 3712 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3713 if (error == 0) 3714 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3715 nlookup_done(&nd); 3716 return (error); 3717 } 3718 3719 /* 3720 * lutimes_args(char *path, struct timeval *tptr) 3721 * 3722 * Set the access and modification times of a file. 3723 */ 3724 int 3725 sys_lutimes(struct sysmsg *sysmsg, const struct lutimes_args *uap) 3726 { 3727 struct timeval tv[2]; 3728 struct nlookupdata nd; 3729 int error; 3730 3731 if (uap->tptr) { 3732 error = copyin(uap->tptr, tv, sizeof(tv)); 3733 if (error) 3734 return (error); 3735 } 3736 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3737 if (error == 0) 3738 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3739 nlookup_done(&nd); 3740 return (error); 3741 } 3742 3743 /* 3744 * Set utimes on a file descriptor. The creds used to open the 3745 * file are used to determine whether the operation is allowed 3746 * or not. 3747 */ 3748 int 3749 kern_futimens(int fd, struct timespec *ts) 3750 { 3751 struct thread *td = curthread; 3752 struct timespec newts[2]; 3753 struct file *fp; 3754 struct vnode *vp; 3755 struct vattr vattr; 3756 struct vattr_lite lva; 3757 int nullflag; 3758 int error; 3759 3760 error = getutimens(ts, newts, &nullflag); 3761 if (error) 3762 return (error); 3763 if ((error = holdvnode(td, fd, &fp)) != 0) 3764 return (error); 3765 if (fp->f_nchandle.ncp) 3766 error = ncp_writechk(&fp->f_nchandle); 3767 if (error == 0) { 3768 vp = fp->f_data; 3769 error = vget(vp, LK_EXCLUSIVE); 3770 if (error == 0) { 3771 error = VOP_GETATTR_FP(vp, &vattr, fp); 3772 if (error == 0) { 3773 lva.va_type = vattr.va_type; 3774 lva.va_nlink = vattr.va_nlink; 3775 lva.va_mode = vattr.va_mode; 3776 lva.va_uid = vattr.va_uid; 3777 lva.va_gid = vattr.va_gid; 3778 lva.va_size = vattr.va_size; 3779 lva.va_flags = vattr.va_flags; 3780 3781 error = naccess_lva(&lva, NLC_OWN | NLC_WRITE, 3782 fp->f_cred); 3783 } 3784 if (error == 0) { 3785 error = setutimes(vp, &vattr, newts, nullflag); 3786 } 3787 vput(vp); 3788 } 3789 } 3790 fdrop(fp); 3791 return (error); 3792 } 3793 3794 /* 3795 * futimens_args(int fd, struct timespec *ts) 3796 * 3797 * Set the access and modification times of a file. 3798 */ 3799 int 3800 sys_futimens(struct sysmsg *sysmsg, const struct futimens_args *uap) 3801 { 3802 struct timespec ts[2]; 3803 int error; 3804 3805 if (uap->ts) { 3806 error = copyin(uap->ts, ts, sizeof(ts)); 3807 if (error) 3808 return (error); 3809 } 3810 error = kern_futimens(uap->fd, uap->ts ? ts : NULL); 3811 return (error); 3812 } 3813 3814 int 3815 kern_futimes(int fd, struct timeval *tptr) 3816 { 3817 struct timespec ts[2]; 3818 int error; 3819 3820 if (tptr) { 3821 if ((error = getutimes(tptr, ts)) != 0) 3822 return (error); 3823 } 3824 error = kern_futimens(fd, tptr ? ts : NULL); 3825 return (error); 3826 } 3827 3828 /* 3829 * futimes_args(int fd, struct timeval *tptr) 3830 * 3831 * Set the access and modification times of a file. 3832 */ 3833 int 3834 sys_futimes(struct sysmsg *sysmsg, const struct futimes_args *uap) 3835 { 3836 struct timeval tv[2]; 3837 int error; 3838 3839 if (uap->tptr) { 3840 error = copyin(uap->tptr, tv, sizeof(tv)); 3841 if (error) 3842 return (error); 3843 } 3844 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL); 3845 return (error); 3846 } 3847 3848 int 3849 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags) 3850 { 3851 struct timespec newts[2]; 3852 struct vnode *vp; 3853 struct vattr vattr; 3854 int nullflag; 3855 int error; 3856 3857 if (flags & ~AT_SYMLINK_NOFOLLOW) 3858 return (EINVAL); 3859 3860 error = getutimens(ts, newts, &nullflag); 3861 if (error) 3862 return (error); 3863 3864 nd->nl_flags |= NLC_OWN | NLC_WRITE; 3865 if ((error = nlookup(nd)) != 0) 3866 return (error); 3867 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3868 return (error); 3869 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3870 return (error); 3871 if ((error = vn_writechk(vp)) == 0) { 3872 error = vget(vp, LK_EXCLUSIVE); 3873 if (error == 0) { 3874 error = setutimes(vp, &vattr, newts, nullflag); 3875 vput(vp); 3876 } 3877 } 3878 vrele(vp); 3879 return (error); 3880 } 3881 3882 /* 3883 * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags); 3884 * 3885 * Set file access and modification times of a file. 3886 */ 3887 int 3888 sys_utimensat(struct sysmsg *sysmsg, const struct utimensat_args *uap) 3889 { 3890 struct timespec ts[2]; 3891 struct nlookupdata nd; 3892 struct file *fp; 3893 int error; 3894 int flags; 3895 3896 if (uap->ts) { 3897 error = copyin(uap->ts, ts, sizeof(ts)); 3898 if (error) 3899 return (error); 3900 } 3901 3902 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3903 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3904 UIO_USERSPACE, flags); 3905 if (error == 0) 3906 error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags); 3907 nlookup_done_at(&nd, fp); 3908 return (error); 3909 } 3910 3911 int 3912 kern_truncate(struct nlookupdata *nd, off_t length) 3913 { 3914 struct vnode *vp; 3915 struct vattr vattr; 3916 int error; 3917 uid_t uid = 0; 3918 gid_t gid = 0; 3919 uint64_t old_size = 0; 3920 3921 if (length < 0) 3922 return(EINVAL); 3923 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE; 3924 if ((error = nlookup(nd)) != 0) 3925 return (error); 3926 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3927 return (error); 3928 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3929 return (error); 3930 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 3931 if (error) { 3932 vrele(vp); 3933 return (error); 3934 } 3935 if (vp->v_type == VDIR) { 3936 error = EISDIR; 3937 goto done; 3938 } 3939 if (vfs_quota_enabled) { 3940 error = VOP_GETATTR(vp, &vattr); 3941 KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0")); 3942 uid = vattr.va_uid; 3943 gid = vattr.va_gid; 3944 old_size = vattr.va_size; 3945 } 3946 3947 if ((error = vn_writechk(vp)) == 0) { 3948 VATTR_NULL(&vattr); 3949 vattr.va_size = length; 3950 error = VOP_SETATTR(vp, &vattr, nd->nl_cred); 3951 VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size); 3952 } 3953 done: 3954 vput(vp); 3955 return (error); 3956 } 3957 3958 /* 3959 * truncate(char *path, int pad, off_t length) 3960 * 3961 * Truncate a file given its path name. 3962 */ 3963 int 3964 sys_truncate(struct sysmsg *sysmsg, const struct truncate_args *uap) 3965 { 3966 struct nlookupdata nd; 3967 int error; 3968 3969 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3970 if (error == 0) 3971 error = kern_truncate(&nd, uap->length); 3972 nlookup_done(&nd); 3973 return error; 3974 } 3975 3976 int 3977 kern_ftruncate(int fd, off_t length) 3978 { 3979 struct thread *td = curthread; 3980 struct vattr vattr; 3981 struct vnode *vp; 3982 struct file *fp; 3983 int error; 3984 uid_t uid = 0; 3985 gid_t gid = 0; 3986 uint64_t old_size = 0; 3987 struct mount *mp; 3988 3989 if (length < 0) 3990 return(EINVAL); 3991 if ((error = holdvnode(td, fd, &fp)) != 0) 3992 return (error); 3993 if (fp->f_nchandle.ncp) { 3994 error = ncp_writechk(&fp->f_nchandle); 3995 if (error) 3996 goto done; 3997 } 3998 if ((fp->f_flag & FWRITE) == 0) { 3999 error = EINVAL; 4000 goto done; 4001 } 4002 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */ 4003 error = EINVAL; 4004 goto done; 4005 } 4006 vp = (struct vnode *)fp->f_data; 4007 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4008 if (vp->v_type == VDIR) { 4009 error = EISDIR; 4010 vn_unlock(vp); 4011 goto done; 4012 } 4013 4014 if (vfs_quota_enabled) { 4015 error = VOP_GETATTR_FP(vp, &vattr, fp); 4016 KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0")); 4017 uid = vattr.va_uid; 4018 gid = vattr.va_gid; 4019 old_size = vattr.va_size; 4020 } 4021 4022 if ((error = vn_writechk(vp)) == 0) { 4023 VATTR_NULL(&vattr); 4024 vattr.va_size = length; 4025 error = VOP_SETATTR_FP(vp, &vattr, fp->f_cred, fp); 4026 mp = vq_vptomp(vp); 4027 VFS_ACCOUNT(mp, uid, gid, length - old_size); 4028 } 4029 vn_unlock(vp); 4030 done: 4031 fdrop(fp); 4032 return (error); 4033 } 4034 4035 /* 4036 * ftruncate_args(int fd, int pad, off_t length) 4037 * 4038 * Truncate a file given a file descriptor. 4039 */ 4040 int 4041 sys_ftruncate(struct sysmsg *sysmsg, const struct ftruncate_args *uap) 4042 { 4043 int error; 4044 4045 error = kern_ftruncate(uap->fd, uap->length); 4046 4047 return (error); 4048 } 4049 4050 /* 4051 * fsync(int fd) 4052 * 4053 * Sync an open file. 4054 */ 4055 int 4056 sys_fsync(struct sysmsg *sysmsg, const struct fsync_args *uap) 4057 { 4058 struct thread *td = curthread; 4059 struct vnode *vp; 4060 struct file *fp; 4061 vm_object_t obj; 4062 int error; 4063 4064 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 4065 return (error); 4066 vp = (struct vnode *)fp->f_data; 4067 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4068 if ((obj = vp->v_object) != NULL) { 4069 if (vp->v_mount == NULL || 4070 (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) { 4071 vm_object_page_clean(obj, 0, 0, 0); 4072 } 4073 } 4074 error = VOP_FSYNC_FP(vp, MNT_WAIT, VOP_FSYNC_SYSCALL, fp); 4075 if (error == 0 && vp->v_mount) 4076 error = buf_fsync(vp); 4077 vn_unlock(vp); 4078 fdrop(fp); 4079 4080 return (error); 4081 } 4082 4083 int 4084 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond) 4085 { 4086 struct nchandle fnchd; 4087 struct nchandle tnchd; 4088 struct namecache *ncp; 4089 struct vnode *fdvp; 4090 struct vnode *tdvp; 4091 struct mount *mp; 4092 struct mount *userenlk; 4093 int error; 4094 u_int fncp_gen; 4095 u_int tncp_gen; 4096 4097 bwillinode(1); 4098 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC; 4099 if ((error = nlookup(fromnd)) != 0) 4100 return (error); 4101 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL) 4102 return (ENOENT); 4103 fnchd.mount = fromnd->nl_nch.mount; 4104 cache_hold(&fnchd); 4105 4106 /* 4107 * unlock the source nch so we can lookup the target nch without 4108 * deadlocking. The target may or may not exist so we do not check 4109 * for a target vp like kern_mkdir() and other creation functions do. 4110 * 4111 * The source and target directories are ref'd and rechecked after 4112 * everything is relocked to determine if the source or target file 4113 * has been renamed. 4114 */ 4115 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED); 4116 fromnd->nl_flags &= ~NLC_NCPISLOCKED; 4117 fncp_gen = fromnd->nl_nch.ncp->nc_generation; 4118 4119 if (fromnd->nl_nch.ncp->nc_vp && 4120 fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 4121 userenlk = fnchd.mount; 4122 cache_unlock(&fromnd->nl_nch); 4123 lockmgr(&userenlk->mnt_renlock, LK_EXCLUSIVE); 4124 } else { 4125 userenlk = NULL; 4126 cache_unlock(&fromnd->nl_nch); 4127 } 4128 4129 /* 4130 * Lookup target 4131 */ 4132 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP; 4133 if ((error = nlookup(tond)) != 0) { 4134 cache_drop(&fnchd); 4135 goto done; 4136 } 4137 tncp_gen = tond->nl_nch.ncp->nc_generation; 4138 4139 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) { 4140 cache_drop(&fnchd); 4141 error = ENOENT; 4142 goto done; 4143 } 4144 tnchd.mount = tond->nl_nch.mount; 4145 cache_hold(&tnchd); 4146 4147 /* 4148 * If the source and target are the same there is nothing to do 4149 */ 4150 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) { 4151 cache_drop(&fnchd); 4152 cache_drop(&tnchd); 4153 error = 0; 4154 goto done; 4155 } 4156 4157 /* 4158 * Mount points cannot be renamed or overwritten 4159 */ 4160 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) & 4161 NCF_ISMOUNTPT 4162 ) { 4163 cache_drop(&fnchd); 4164 cache_drop(&tnchd); 4165 error = EINVAL; 4166 goto done; 4167 } 4168 4169 /* 4170 * Lock all four namecache entries. tond is already locked. 4171 */ 4172 cache_lock4_tondlocked(&fnchd, &fromnd->nl_nch, 4173 &tnchd, &tond->nl_nch, 4174 fromnd->nl_cred, tond->nl_cred); 4175 fromnd->nl_flags |= NLC_NCPISLOCKED; 4176 4177 /* 4178 * If the namecache generation changed for either fromnd or tond, 4179 * we must retry. 4180 */ 4181 if (fromnd->nl_nch.ncp->nc_generation != fncp_gen || 4182 tond->nl_nch.ncp->nc_generation != tncp_gen) { 4183 krateprintf(&krate_rename, 4184 "kern_rename: retry due to race on: " 4185 "\"%s\" -> \"%s\"\n", 4186 fromnd->nl_nch.ncp->nc_name, 4187 tond->nl_nch.ncp->nc_name); 4188 error = EAGAIN; 4189 goto finish; 4190 } 4191 4192 /* 4193 * If either fromnd or tond are marked destroyed a ripout occured 4194 * out from under us and we must retry. 4195 */ 4196 if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) || 4197 fromnd->nl_nch.ncp->nc_vp == NULL || 4198 (tond->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED))) { 4199 krateprintf(&krate_rename, 4200 "kern_rename: retry due to ripout on: " 4201 "\"%s\" -> \"%s\"\n", 4202 fromnd->nl_nch.ncp->nc_name, 4203 tond->nl_nch.ncp->nc_name); 4204 error = EAGAIN; 4205 goto finish; 4206 } 4207 4208 /* 4209 * Make sure the parent directories linkages are the same. We have 4210 * already checked that fromnd and tond are not mount points so this 4211 * should not loop forever on a cross-mount. 4212 */ 4213 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent || 4214 tnchd.ncp != tond->nl_nch.ncp->nc_parent) { 4215 error = EAGAIN; 4216 goto finish; 4217 } 4218 4219 /* 4220 * Both the source and target must be within the same filesystem and 4221 * in the same filesystem as their parent directories within the 4222 * namecache topology. 4223 * 4224 * NOTE: fromnd's nc_mount or nc_vp could be NULL. 4225 */ 4226 mp = fnchd.mount; 4227 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount || 4228 mp != tond->nl_nch.mount) { 4229 error = EXDEV; 4230 goto finish; 4231 } 4232 4233 /* 4234 * Make sure the mount point is writable 4235 */ 4236 if ((error = ncp_writechk(&tond->nl_nch)) != 0) { 4237 goto finish; 4238 } 4239 4240 /* 4241 * If the target exists and either the source or target is a directory, 4242 * then both must be directories. 4243 * 4244 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h 4245 * have become NULL. 4246 */ 4247 if (tond->nl_nch.ncp->nc_vp) { 4248 if (fromnd->nl_nch.ncp->nc_vp == NULL) { 4249 error = ENOENT; 4250 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 4251 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR) 4252 error = ENOTDIR; 4253 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) { 4254 error = EISDIR; 4255 } 4256 } 4257 4258 /* 4259 * You cannot rename a source into itself or a subdirectory of itself. 4260 * We check this by travsersing the target directory upwards looking 4261 * for a match against the source. 4262 * 4263 * Only required when renaming a directory, in which case userenlk is 4264 * non-NULL. 4265 */ 4266 if (__predict_false(userenlk && error == 0)) { 4267 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) { 4268 if (fromnd->nl_nch.ncp == ncp) { 4269 error = EINVAL; 4270 break; 4271 } 4272 } 4273 } 4274 4275 /* 4276 * Even though the namespaces are different, they may still represent 4277 * hardlinks to the same file. The filesystem might have a hard time 4278 * with this so we issue a NREMOVE of the source instead of a NRENAME 4279 * when we detect the situation. 4280 */ 4281 if (error == 0) { 4282 fdvp = fromnd->nl_dvp; 4283 tdvp = tond->nl_dvp; 4284 if (fdvp == NULL || tdvp == NULL) { 4285 error = EPERM; 4286 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) { 4287 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp, 4288 fromnd->nl_cred); 4289 } else { 4290 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch, 4291 fdvp, tdvp, tond->nl_cred); 4292 } 4293 } 4294 finish: 4295 cache_put(&tnchd); 4296 cache_put(&fnchd); 4297 done: 4298 if (userenlk) 4299 lockmgr(&userenlk->mnt_renlock, LK_RELEASE); 4300 return (error); 4301 } 4302 4303 /* 4304 * rename_args(char *from, char *to) 4305 * 4306 * Rename files. Source and destination must either both be directories, 4307 * or both not be directories. If target is a directory, it must be empty. 4308 */ 4309 int 4310 sys_rename(struct sysmsg *sysmsg, const struct rename_args *uap) 4311 { 4312 struct nlookupdata fromnd, tond; 4313 int error; 4314 4315 do { 4316 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0); 4317 if (error == 0) { 4318 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0); 4319 if (error == 0) 4320 error = kern_rename(&fromnd, &tond); 4321 nlookup_done(&tond); 4322 } 4323 nlookup_done(&fromnd); 4324 } while (error == EAGAIN); 4325 return (error); 4326 } 4327 4328 /* 4329 * renameat_args(int oldfd, char *old, int newfd, char *new) 4330 * 4331 * Rename files using paths relative to the directories associated with 4332 * oldfd and newfd. Source and destination must either both be directories, 4333 * or both not be directories. If target is a directory, it must be empty. 4334 */ 4335 int 4336 sys_renameat(struct sysmsg *sysmsg, const struct renameat_args *uap) 4337 { 4338 struct nlookupdata oldnd, newnd; 4339 struct file *oldfp, *newfp; 4340 int error; 4341 4342 do { 4343 error = nlookup_init_at(&oldnd, &oldfp, 4344 uap->oldfd, uap->old, 4345 UIO_USERSPACE, 0); 4346 if (error == 0) { 4347 error = nlookup_init_at(&newnd, &newfp, 4348 uap->newfd, uap->new, 4349 UIO_USERSPACE, 0); 4350 if (error == 0) 4351 error = kern_rename(&oldnd, &newnd); 4352 nlookup_done_at(&newnd, newfp); 4353 } 4354 nlookup_done_at(&oldnd, oldfp); 4355 } while (error == EAGAIN); 4356 return (error); 4357 } 4358 4359 int 4360 kern_mkdir(struct nlookupdata *nd, int mode) 4361 { 4362 struct thread *td = curthread; 4363 struct proc *p = td->td_proc; 4364 struct vnode *vp; 4365 struct vattr vattr; 4366 int error; 4367 4368 bwillinode(1); 4369 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP; 4370 if ((error = nlookup(nd)) != 0) 4371 return (error); 4372 4373 if (nd->nl_nch.ncp->nc_vp) 4374 return (EEXIST); 4375 if (nd->nl_dvp == NULL) 4376 return (EINVAL); 4377 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 4378 return (error); 4379 VATTR_NULL(&vattr); 4380 vattr.va_type = VDIR; 4381 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask; 4382 4383 vp = NULL; 4384 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr); 4385 if (error == 0) 4386 vput(vp); 4387 return (error); 4388 } 4389 4390 /* 4391 * mkdir_args(char *path, int mode) 4392 * 4393 * Make a directory file. 4394 */ 4395 int 4396 sys_mkdir(struct sysmsg *sysmsg, const struct mkdir_args *uap) 4397 { 4398 struct nlookupdata nd; 4399 int error; 4400 4401 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 4402 if (error == 0) 4403 error = kern_mkdir(&nd, uap->mode); 4404 nlookup_done(&nd); 4405 return (error); 4406 } 4407 4408 /* 4409 * mkdirat_args(int fd, char *path, mode_t mode) 4410 * 4411 * Make a directory file. The path is relative to the directory associated 4412 * with fd. 4413 */ 4414 int 4415 sys_mkdirat(struct sysmsg *sysmsg, const struct mkdirat_args *uap) 4416 { 4417 struct nlookupdata nd; 4418 struct file *fp; 4419 int error; 4420 4421 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 4422 if (error == 0) 4423 error = kern_mkdir(&nd, uap->mode); 4424 nlookup_done_at(&nd, fp); 4425 return (error); 4426 } 4427 4428 int 4429 kern_rmdir(struct nlookupdata *nd) 4430 { 4431 int error; 4432 4433 bwillinode(1); 4434 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 4435 if ((error = nlookup(nd)) != 0) 4436 return (error); 4437 4438 /* 4439 * Do not allow directories representing mount points to be 4440 * deleted, even if empty. Check write perms on mount point 4441 * in case the vnode is aliased (aka nullfs). 4442 */ 4443 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT)) 4444 return (EBUSY); 4445 if (nd->nl_dvp == NULL) 4446 return (EINVAL); 4447 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 4448 return (error); 4449 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 4450 return (error); 4451 } 4452 4453 /* 4454 * rmdir_args(char *path) 4455 * 4456 * Remove a directory file. 4457 */ 4458 int 4459 sys_rmdir(struct sysmsg *sysmsg, const struct rmdir_args *uap) 4460 { 4461 struct nlookupdata nd; 4462 int error; 4463 4464 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 4465 if (error == 0) 4466 error = kern_rmdir(&nd); 4467 nlookup_done(&nd); 4468 return (error); 4469 } 4470 4471 int 4472 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res, 4473 enum uio_seg direction) 4474 { 4475 struct thread *td = curthread; 4476 struct vnode *vp; 4477 struct file *fp; 4478 struct uio auio; 4479 struct iovec aiov; 4480 off_t loff; 4481 int error, eofflag; 4482 4483 if ((error = holdvnode(td, fd, &fp)) != 0) 4484 return (error); 4485 if ((fp->f_flag & FREAD) == 0) { 4486 error = EBADF; 4487 goto done; 4488 } 4489 vp = (struct vnode *)fp->f_data; 4490 if (vp->v_type != VDIR) { 4491 error = EINVAL; 4492 goto done; 4493 } 4494 aiov.iov_base = buf; 4495 aiov.iov_len = count; 4496 auio.uio_iov = &aiov; 4497 auio.uio_iovcnt = 1; 4498 auio.uio_rw = UIO_READ; 4499 auio.uio_segflg = direction; 4500 auio.uio_td = td; 4501 auio.uio_resid = count; 4502 loff = auio.uio_offset = fp->f_offset; 4503 error = VOP_READDIR_FP(vp, &auio, fp->f_cred, &eofflag, NULL, NULL, fp); 4504 fp->f_offset = auio.uio_offset; 4505 if (error) 4506 goto done; 4507 4508 /* 4509 * WARNING! *basep may not be wide enough to accomodate the 4510 * seek offset. XXX should we hack this to return the upper 32 bits 4511 * for offsets greater then 4G? 4512 */ 4513 if (basep) { 4514 *basep = (long)loff; 4515 } 4516 *res = count - auio.uio_resid; 4517 done: 4518 fdrop(fp); 4519 return (error); 4520 } 4521 4522 /* 4523 * getdirentries_args(int fd, char *buf, u_int conut, long *basep) 4524 * 4525 * Read a block of directory entries in a file system independent format. 4526 */ 4527 int 4528 sys_getdirentries(struct sysmsg *sysmsg, const struct getdirentries_args *uap) 4529 { 4530 long base; 4531 int error; 4532 4533 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base, 4534 &sysmsg->sysmsg_result, UIO_USERSPACE); 4535 4536 if (error == 0 && uap->basep) 4537 error = copyout(&base, uap->basep, sizeof(*uap->basep)); 4538 return (error); 4539 } 4540 4541 /* 4542 * getdents_args(int fd, char *buf, size_t count) 4543 */ 4544 int 4545 sys_getdents(struct sysmsg *sysmsg, const struct getdents_args *uap) 4546 { 4547 int error; 4548 4549 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL, 4550 &sysmsg->sysmsg_result, UIO_USERSPACE); 4551 4552 return (error); 4553 } 4554 4555 /* 4556 * Set the mode mask for creation of filesystem nodes. 4557 * 4558 * umask(int newmask) 4559 */ 4560 int 4561 sys_umask(struct sysmsg *sysmsg, const struct umask_args *uap) 4562 { 4563 struct thread *td = curthread; 4564 struct proc *p = td->td_proc; 4565 struct filedesc *fdp; 4566 4567 fdp = p->p_fd; 4568 sysmsg->sysmsg_result = fdp->fd_cmask; 4569 fdp->fd_cmask = uap->newmask & ALLPERMS; 4570 return (0); 4571 } 4572 4573 /* 4574 * revoke(char *path) 4575 * 4576 * Void all references to file by ripping underlying filesystem 4577 * away from vnode. 4578 */ 4579 int 4580 sys_revoke(struct sysmsg *sysmsg, const struct revoke_args *uap) 4581 { 4582 struct nlookupdata nd; 4583 struct vattr vattr; 4584 struct vnode *vp; 4585 struct ucred *cred; 4586 int error; 4587 4588 vp = NULL; 4589 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4590 if (error == 0) 4591 error = nlookup(&nd); 4592 if (error == 0) 4593 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 4594 cred = crhold(nd.nl_cred); 4595 nlookup_done(&nd); 4596 if (error == 0) { 4597 if (error == 0) 4598 error = VOP_GETATTR(vp, &vattr); 4599 if (error == 0 && cred->cr_uid != vattr.va_uid) 4600 error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0); 4601 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) { 4602 if (vcount(vp) > 0) 4603 error = vrevoke(vp, cred); 4604 } else if (error == 0) { 4605 error = vrevoke(vp, cred); 4606 } 4607 vrele(vp); 4608 } 4609 if (cred) 4610 crfree(cred); 4611 return (error); 4612 } 4613 4614 /* 4615 * getfh_args(char *fname, fhandle_t *fhp) 4616 * 4617 * Get (NFS) file handle 4618 * 4619 * NOTE: We use the fsid of the covering mount, even if it is a nullfs 4620 * mount. This allows nullfs mounts to be explicitly exported. 4621 * 4622 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe. 4623 * 4624 * nullfs mounts of subdirectories are not safe. That is, it will 4625 * work, but you do not really have protection against access to 4626 * the related parent directories. 4627 */ 4628 int 4629 sys_getfh(struct sysmsg *sysmsg, const struct getfh_args *uap) 4630 { 4631 struct thread *td = curthread; 4632 struct nlookupdata nd; 4633 fhandle_t fh; 4634 struct vnode *vp; 4635 struct mount *mp; 4636 int error; 4637 4638 /* 4639 * Must be super user 4640 */ 4641 if ((error = priv_check(td, PRIV_ROOT)) != 0) 4642 return (error); 4643 4644 vp = NULL; 4645 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 4646 if (error == 0) 4647 error = nlookup(&nd); 4648 if (error == 0) 4649 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4650 mp = nd.nl_nch.mount; 4651 nlookup_done(&nd); 4652 if (error == 0) { 4653 bzero(&fh, sizeof(fh)); 4654 fh.fh_fsid = mp->mnt_stat.f_fsid; 4655 error = VFS_VPTOFH(vp, &fh.fh_fid); 4656 vput(vp); 4657 if (error == 0) 4658 error = copyout(&fh, uap->fhp, sizeof(fh)); 4659 } 4660 return (error); 4661 } 4662 4663 /* 4664 * fhopen_args(const struct fhandle *u_fhp, int flags) 4665 * 4666 * syscall for the rpc.lockd to use to translate a NFS file handle into 4667 * an open descriptor. 4668 * 4669 * warning: do not remove the priv_check() call or this becomes one giant 4670 * security hole. 4671 */ 4672 int 4673 sys_fhopen(struct sysmsg *sysmsg, const struct fhopen_args *uap) 4674 { 4675 struct thread *td = curthread; 4676 struct filedesc *fdp = td->td_proc->p_fd; 4677 struct mount *mp; 4678 struct vnode *vp; 4679 struct fhandle fhp; 4680 struct vattr vat; 4681 struct vattr *vap = &vat; 4682 struct flock lf; 4683 int fmode, mode, error = 0, type; 4684 struct file *nfp; 4685 struct file *fp; 4686 int indx; 4687 4688 /* 4689 * Must be super user 4690 */ 4691 error = priv_check(td, PRIV_ROOT); 4692 if (error) 4693 return (error); 4694 4695 fmode = FFLAGS(uap->flags); 4696 4697 /* 4698 * Why not allow a non-read/write open for our lockd? 4699 */ 4700 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) 4701 return (EINVAL); 4702 error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); 4703 if (error) 4704 return(error); 4705 4706 /* 4707 * Find the mount point 4708 */ 4709 mp = vfs_getvfs(&fhp.fh_fsid); 4710 if (mp == NULL) { 4711 error = ESTALE; 4712 goto done2; 4713 } 4714 /* now give me my vnode, it gets returned to me locked */ 4715 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp); 4716 if (error) 4717 goto done; 4718 /* 4719 * from now on we have to make sure not 4720 * to forget about the vnode 4721 * any error that causes an abort must vput(vp) 4722 * just set error = err and 'goto bad;'. 4723 */ 4724 4725 /* 4726 * from vn_open 4727 */ 4728 if (vp->v_type == VLNK) { 4729 error = EMLINK; 4730 goto bad; 4731 } 4732 if (vp->v_type == VSOCK) { 4733 error = EOPNOTSUPP; 4734 goto bad; 4735 } 4736 mode = 0; 4737 if (fmode & (FWRITE | O_TRUNC)) { 4738 if (vp->v_type == VDIR) { 4739 error = EISDIR; 4740 goto bad; 4741 } 4742 error = vn_writechk(vp); 4743 if (error) 4744 goto bad; 4745 mode |= VWRITE; 4746 } 4747 if (fmode & FREAD) 4748 mode |= VREAD; 4749 if (mode) { 4750 error = VOP_ACCESS(vp, mode, td->td_ucred); 4751 if (error) 4752 goto bad; 4753 } 4754 if (fmode & O_TRUNC) { 4755 vn_unlock(vp); /* XXX */ 4756 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 4757 VATTR_NULL(vap); 4758 vap->va_size = 0; 4759 error = VOP_SETATTR(vp, vap, td->td_ucred); 4760 if (error) 4761 goto bad; 4762 } 4763 4764 /* 4765 * VOP_OPEN needs the file pointer so it can potentially override 4766 * it. 4767 * 4768 * WARNING! no f_nchandle will be associated when fhopen()ing a 4769 * directory. XXX 4770 */ 4771 if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0) 4772 goto bad; 4773 error = VOP_OPEN(vp, fmode, td->td_ucred, &nfp); 4774 fp = nfp; 4775 4776 if (error) { 4777 /* 4778 * setting f_ops this way prevents VOP_CLOSE from being 4779 * called or fdrop() releasing the vp from v_data. Since 4780 * the VOP_OPEN failed we don't want to VOP_CLOSE. 4781 */ 4782 fp->f_ops = &badfileops; 4783 fp->f_data = NULL; 4784 goto bad_drop; 4785 } 4786 4787 /* 4788 * The fp is given its own reference, we still have our ref and lock. 4789 * 4790 * Assert that all regular files must be created with a VM object. 4791 */ 4792 if (vp->v_type == VREG && vp->v_object == NULL) { 4793 kprintf("fhopen: regular file did not " 4794 "have VM object: %p\n", 4795 vp); 4796 goto bad_drop; 4797 } 4798 4799 /* 4800 * The open was successful. Handle any locking requirements. 4801 */ 4802 if (fmode & (O_EXLOCK | O_SHLOCK)) { 4803 lf.l_whence = SEEK_SET; 4804 lf.l_start = 0; 4805 lf.l_len = 0; 4806 if (fmode & O_EXLOCK) 4807 lf.l_type = F_WRLCK; 4808 else 4809 lf.l_type = F_RDLCK; 4810 if (fmode & FNONBLOCK) 4811 type = 0; 4812 else 4813 type = F_WAIT; 4814 vn_unlock(vp); 4815 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, 4816 &lf, type)) != 0) { 4817 /* 4818 * release our private reference. 4819 */ 4820 fsetfd(fdp, NULL, indx); 4821 fdrop(fp); 4822 vrele(vp); 4823 goto done; 4824 } 4825 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4826 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */ 4827 } 4828 4829 /* 4830 * Clean up. Associate the file pointer with the previously 4831 * reserved descriptor and return it. 4832 */ 4833 vput(vp); 4834 if (uap->flags & O_CLOEXEC) 4835 fdp->fd_files[indx].fileflags |= UF_EXCLOSE; 4836 fsetfd(fdp, fp, indx); 4837 fdrop(fp); 4838 sysmsg->sysmsg_result = indx; 4839 mount_drop(mp); 4840 4841 return (error); 4842 4843 bad_drop: 4844 fsetfd(fdp, NULL, indx); 4845 fdrop(fp); 4846 bad: 4847 vput(vp); 4848 done: 4849 mount_drop(mp); 4850 done2: 4851 return (error); 4852 } 4853 4854 /* 4855 * fhstat_args(struct fhandle *u_fhp, struct stat *sb) 4856 */ 4857 int 4858 sys_fhstat(struct sysmsg *sysmsg, const struct fhstat_args *uap) 4859 { 4860 struct thread *td = curthread; 4861 struct stat sb; 4862 fhandle_t fh; 4863 struct mount *mp; 4864 struct vnode *vp; 4865 int error; 4866 4867 /* 4868 * Must be super user 4869 */ 4870 error = priv_check(td, PRIV_ROOT); 4871 if (error) 4872 return (error); 4873 4874 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); 4875 if (error) 4876 return (error); 4877 4878 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 4879 error = ESTALE; 4880 if (error == 0) { 4881 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) { 4882 error = vn_stat(vp, &sb, td->td_ucred); 4883 vput(vp); 4884 } 4885 } 4886 if (error == 0) 4887 error = copyout(&sb, uap->sb, sizeof(sb)); 4888 if (mp) 4889 mount_drop(mp); 4890 4891 return (error); 4892 } 4893 4894 /* 4895 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf) 4896 */ 4897 int 4898 sys_fhstatfs(struct sysmsg *sysmsg, const struct fhstatfs_args *uap) 4899 { 4900 struct thread *td = curthread; 4901 struct proc *p = td->td_proc; 4902 struct statfs *sp; 4903 struct mount *mp; 4904 struct vnode *vp; 4905 struct statfs sb; 4906 char *fullpath, *freepath; 4907 fhandle_t fh; 4908 int error; 4909 4910 /* 4911 * Must be super user 4912 */ 4913 if ((error = priv_check(td, PRIV_ROOT))) 4914 return (error); 4915 4916 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4917 return (error); 4918 4919 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4920 error = ESTALE; 4921 goto done; 4922 } 4923 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4924 error = ESTALE; 4925 goto done; 4926 } 4927 4928 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0) 4929 goto done; 4930 mp = vp->v_mount; 4931 sp = &mp->mnt_stat; 4932 vput(vp); 4933 if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0) 4934 goto done; 4935 4936 error = mount_path(p, mp, &fullpath, &freepath); 4937 if (error) 4938 goto done; 4939 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 4940 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 4941 kfree(freepath, M_TEMP); 4942 4943 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 4944 if (priv_check(td, PRIV_ROOT)) { 4945 bcopy(sp, &sb, sizeof(sb)); 4946 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; 4947 sp = &sb; 4948 } 4949 error = copyout(sp, uap->buf, sizeof(*sp)); 4950 done: 4951 if (mp) 4952 mount_drop(mp); 4953 4954 return (error); 4955 } 4956 4957 /* 4958 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf) 4959 */ 4960 int 4961 sys_fhstatvfs(struct sysmsg *sysmsg, const struct fhstatvfs_args *uap) 4962 { 4963 struct thread *td = curthread; 4964 struct proc *p = td->td_proc; 4965 struct statvfs *sp; 4966 struct mount *mp; 4967 struct vnode *vp; 4968 fhandle_t fh; 4969 int error; 4970 4971 /* 4972 * Must be super user 4973 */ 4974 if ((error = priv_check(td, PRIV_ROOT))) 4975 return (error); 4976 4977 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4978 return (error); 4979 4980 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4981 error = ESTALE; 4982 goto done; 4983 } 4984 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4985 error = ESTALE; 4986 goto done; 4987 } 4988 4989 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 4990 goto done; 4991 mp = vp->v_mount; 4992 sp = &mp->mnt_vstat; 4993 vput(vp); 4994 if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0) 4995 goto done; 4996 4997 sp->f_flag = 0; 4998 if (mp->mnt_flag & MNT_RDONLY) 4999 sp->f_flag |= ST_RDONLY; 5000 if (mp->mnt_flag & MNT_NOSUID) 5001 sp->f_flag |= ST_NOSUID; 5002 error = copyout(sp, uap->buf, sizeof(*sp)); 5003 done: 5004 if (mp) 5005 mount_drop(mp); 5006 return (error); 5007 } 5008 5009 5010 /* 5011 * Syscall to push extended attribute configuration information into the 5012 * VFS. Accepts a path, which it converts to a mountpoint, as well as 5013 * a command (int cmd), and attribute name and misc data. For now, the 5014 * attribute name is left in userspace for consumption by the VFS_op. 5015 * It will probably be changed to be copied into sysspace by the 5016 * syscall in the future, once issues with various consumers of the 5017 * attribute code have raised their hands. 5018 * 5019 * Currently this is used only by UFS Extended Attributes. 5020 */ 5021 int 5022 sys_extattrctl(struct sysmsg *sysmsg, const struct extattrctl_args *uap) 5023 { 5024 struct nlookupdata nd; 5025 struct vnode *vp; 5026 char attrname[EXTATTR_MAXNAMELEN]; 5027 int error; 5028 size_t size; 5029 5030 attrname[0] = 0; 5031 vp = NULL; 5032 error = 0; 5033 5034 if (error == 0 && uap->filename) { 5035 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE, 5036 NLC_FOLLOW); 5037 if (error == 0) 5038 error = nlookup(&nd); 5039 if (error == 0) 5040 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 5041 nlookup_done(&nd); 5042 } 5043 5044 if (error == 0 && uap->attrname) { 5045 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN, 5046 &size); 5047 } 5048 5049 if (error == 0) { 5050 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5051 if (error == 0) 5052 error = nlookup(&nd); 5053 if (error == 0) 5054 error = ncp_writechk(&nd.nl_nch); 5055 if (error == 0) { 5056 error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp, 5057 uap->attrnamespace, 5058 uap->attrname, nd.nl_cred); 5059 } 5060 nlookup_done(&nd); 5061 } 5062 5063 return (error); 5064 } 5065 5066 /* 5067 * Syscall to get a named extended attribute on a file or directory. 5068 */ 5069 int 5070 sys_extattr_set_file(struct sysmsg *sysmsg, 5071 const struct extattr_set_file_args *uap) 5072 { 5073 char attrname[EXTATTR_MAXNAMELEN]; 5074 struct nlookupdata nd; 5075 struct vnode *vp; 5076 struct uio auio; 5077 struct iovec aiov; 5078 int error; 5079 5080 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 5081 if (error) 5082 return (error); 5083 5084 vp = NULL; 5085 5086 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5087 if (error == 0) 5088 error = nlookup(&nd); 5089 if (error == 0) 5090 error = ncp_writechk(&nd.nl_nch); 5091 if (error == 0) 5092 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 5093 if (error) { 5094 nlookup_done(&nd); 5095 return (error); 5096 } 5097 5098 bzero(&auio, sizeof(auio)); 5099 aiov.iov_base = uap->data; 5100 aiov.iov_len = uap->nbytes; 5101 auio.uio_iov = &aiov; 5102 auio.uio_iovcnt = 1; 5103 auio.uio_offset = 0; 5104 auio.uio_resid = uap->nbytes; 5105 auio.uio_rw = UIO_WRITE; 5106 auio.uio_td = curthread; 5107 5108 error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname, 5109 &auio, nd.nl_cred); 5110 5111 vput(vp); 5112 nlookup_done(&nd); 5113 return (error); 5114 } 5115 5116 /* 5117 * Syscall to get a named extended attribute on a file or directory. 5118 */ 5119 int 5120 sys_extattr_get_file(struct sysmsg *sysmsg, 5121 const struct extattr_get_file_args *uap) 5122 { 5123 char attrname[EXTATTR_MAXNAMELEN]; 5124 struct nlookupdata nd; 5125 struct uio auio; 5126 struct iovec aiov; 5127 struct vnode *vp; 5128 int error; 5129 5130 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 5131 if (error) 5132 return (error); 5133 5134 vp = NULL; 5135 5136 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5137 if (error == 0) 5138 error = nlookup(&nd); 5139 if (error == 0) 5140 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp); 5141 if (error) { 5142 nlookup_done(&nd); 5143 return (error); 5144 } 5145 5146 bzero(&auio, sizeof(auio)); 5147 aiov.iov_base = uap->data; 5148 aiov.iov_len = uap->nbytes; 5149 auio.uio_iov = &aiov; 5150 auio.uio_iovcnt = 1; 5151 auio.uio_offset = 0; 5152 auio.uio_resid = uap->nbytes; 5153 auio.uio_rw = UIO_READ; 5154 auio.uio_td = curthread; 5155 5156 error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname, 5157 &auio, nd.nl_cred); 5158 sysmsg->sysmsg_result = uap->nbytes - auio.uio_resid; 5159 5160 vput(vp); 5161 nlookup_done(&nd); 5162 return(error); 5163 } 5164 5165 /* 5166 * Syscall to delete a named extended attribute from a file or directory. 5167 * Accepts attribute name. The real work happens in VOP_SETEXTATTR(). 5168 */ 5169 int 5170 sys_extattr_delete_file(struct sysmsg *sysmsg, 5171 const struct extattr_delete_file_args *uap) 5172 { 5173 char attrname[EXTATTR_MAXNAMELEN]; 5174 struct nlookupdata nd; 5175 struct vnode *vp; 5176 int error; 5177 5178 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 5179 if (error) 5180 return(error); 5181 5182 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5183 if (error == 0) 5184 error = nlookup(&nd); 5185 if (error == 0) 5186 error = ncp_writechk(&nd.nl_nch); 5187 if (error == 0) { 5188 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 5189 if (error == 0) { 5190 error = VOP_SETEXTATTR(vp, uap->attrnamespace, 5191 attrname, NULL, nd.nl_cred); 5192 vput(vp); 5193 } 5194 } 5195 nlookup_done(&nd); 5196 return(error); 5197 } 5198 5199 /* 5200 * Determine if the mount is visible to the process. 5201 */ 5202 static int 5203 chroot_visible_mnt(struct mount *mp, struct proc *p) 5204 { 5205 struct nchandle nch; 5206 5207 /* 5208 * Traverse from the mount point upwards. If we hit the process 5209 * root then the mount point is visible to the process. 5210 */ 5211 nch = mp->mnt_ncmountpt; 5212 while (nch.ncp) { 5213 if (nch.mount == p->p_fd->fd_nrdir.mount && 5214 nch.ncp == p->p_fd->fd_nrdir.ncp) { 5215 return(1); 5216 } 5217 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) { 5218 nch = nch.mount->mnt_ncmounton; 5219 } else { 5220 nch.ncp = nch.ncp->nc_parent; 5221 } 5222 } 5223 5224 /* 5225 * If the mount point is not visible to the process, but the 5226 * process root is in a subdirectory of the mount, return 5227 * TRUE anyway. 5228 */ 5229 if (p->p_fd->fd_nrdir.mount == mp) 5230 return(1); 5231 5232 return(0); 5233 } 5234 5235 /* Sets priv to PRIV_ROOT in case no matching fs */ 5236 static int 5237 get_fspriv(const char *fsname) 5238 { 5239 5240 if (strncmp("null", fsname, 5) == 0) { 5241 return PRIV_VFS_MOUNT_NULLFS; 5242 } else if (strncmp(fsname, "tmpfs", 6) == 0) { 5243 return PRIV_VFS_MOUNT_TMPFS; 5244 } 5245 5246 return PRIV_ROOT; 5247 } 5248 5249 int 5250 sys___realpath(struct sysmsg *sysmsg, const struct __realpath_args *uap) 5251 { 5252 struct nlookupdata nd; 5253 char *rbuf; 5254 char *fbuf; 5255 ssize_t rlen; 5256 int error; 5257 5258 /* 5259 * Invalid length if less than 0. 0 is allowed 5260 */ 5261 if ((ssize_t)uap->len < 0) 5262 return EINVAL; 5263 5264 rbuf = NULL; 5265 fbuf = NULL; 5266 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5267 if (error) 5268 goto done; 5269 5270 nd.nl_flags |= NLC_SHAREDLOCK; 5271 error = nlookup(&nd); 5272 if (error) 5273 goto done; 5274 5275 if (nd.nl_nch.ncp->nc_vp == NULL) { 5276 error = ENOENT; 5277 goto done; 5278 } 5279 5280 /* 5281 * Shortcut test for existence. 5282 */ 5283 if (uap->len == 0) { 5284 error = ENAMETOOLONG; 5285 goto done; 5286 } 5287 5288 /* 5289 * Obtain the path relative to the process root. The nch must not 5290 * be locked for the cache_fullpath() call. 5291 */ 5292 if (nd.nl_flags & NLC_NCPISLOCKED) { 5293 nd.nl_flags &= ~NLC_NCPISLOCKED; 5294 cache_unlock(&nd.nl_nch); 5295 } 5296 error = cache_fullpath(curproc, &nd.nl_nch, NULL, &rbuf, &fbuf, 0); 5297 if (error) 5298 goto done; 5299 5300 rlen = (ssize_t)strlen(rbuf); 5301 if (rlen >= uap->len) { 5302 error = ENAMETOOLONG; 5303 goto done; 5304 } 5305 error = copyout(rbuf, uap->buf, rlen + 1); 5306 if (error == 0) 5307 sysmsg->sysmsg_szresult = rlen; 5308 done: 5309 nlookup_done(&nd); 5310 if (fbuf) 5311 kfree(fbuf, M_TEMP); 5312 5313 return error; 5314 } 5315