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 case S_IFIFO: 2266 return (kern_mkfifo(nd, mode)); 2267 break; 2268 default: 2269 error = EINVAL; 2270 break; 2271 } 2272 2273 if (error) 2274 return (error); 2275 2276 bwillinode(1); 2277 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2278 if ((error = nlookup(nd)) != 0) 2279 return (error); 2280 if (nd->nl_nch.ncp->nc_vp) 2281 return (EEXIST); 2282 if (nd->nl_dvp == NULL) 2283 return (EINVAL); 2284 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2285 return (error); 2286 2287 if (whiteout) { 2288 error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp, 2289 nd->nl_cred, NAMEI_CREATE); 2290 } else { 2291 vp = NULL; 2292 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, 2293 &vp, nd->nl_cred, &vattr); 2294 if (error == 0) 2295 vput(vp); 2296 } 2297 return (error); 2298 } 2299 2300 /* 2301 * mknod_args(char *path, int mode, int dev) 2302 * 2303 * Create a special file. 2304 */ 2305 int 2306 sys_mknod(struct sysmsg *sysmsg, const struct mknod_args *uap) 2307 { 2308 struct nlookupdata nd; 2309 int error; 2310 2311 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2312 if (error == 0) { 2313 error = kern_mknod(&nd, uap->mode, 2314 umajor(uap->dev), uminor(uap->dev)); 2315 } 2316 nlookup_done(&nd); 2317 return (error); 2318 } 2319 2320 /* 2321 * mknodat_args(int fd, char *path, mode_t mode, dev_t dev) 2322 * 2323 * Create a special file. The path is relative to the directory associated 2324 * with fd. 2325 */ 2326 int 2327 sys_mknodat(struct sysmsg *sysmsg, const struct mknodat_args *uap) 2328 { 2329 struct nlookupdata nd; 2330 struct file *fp; 2331 int error; 2332 2333 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2334 if (error == 0) { 2335 error = kern_mknod(&nd, uap->mode, 2336 umajor(uap->dev), uminor(uap->dev)); 2337 } 2338 nlookup_done_at(&nd, fp); 2339 return (error); 2340 } 2341 2342 int 2343 kern_mkfifo(struct nlookupdata *nd, int mode) 2344 { 2345 struct thread *td = curthread; 2346 struct proc *p = td->td_proc; 2347 struct vattr vattr; 2348 struct vnode *vp; 2349 int error; 2350 2351 bwillinode(1); 2352 2353 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2354 if ((error = nlookup(nd)) != 0) 2355 return (error); 2356 if (nd->nl_nch.ncp->nc_vp) 2357 return (EEXIST); 2358 if (nd->nl_dvp == NULL) 2359 return (EINVAL); 2360 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2361 return (error); 2362 2363 VATTR_NULL(&vattr); 2364 vattr.va_type = VFIFO; 2365 vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask; 2366 vp = NULL; 2367 error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr); 2368 if (error == 0) 2369 vput(vp); 2370 return (error); 2371 } 2372 2373 /* 2374 * mkfifo_args(char *path, int mode) 2375 * 2376 * Create a named pipe. 2377 */ 2378 int 2379 sys_mkfifo(struct sysmsg *sysmsg, const struct mkfifo_args *uap) 2380 { 2381 struct nlookupdata nd; 2382 int error; 2383 2384 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2385 if (error == 0) 2386 error = kern_mkfifo(&nd, uap->mode); 2387 nlookup_done(&nd); 2388 return (error); 2389 } 2390 2391 /* 2392 * mkfifoat_args(int fd, char *path, mode_t mode) 2393 * 2394 * Create a named pipe. The path is relative to the directory associated 2395 * with fd. 2396 */ 2397 int 2398 sys_mkfifoat(struct sysmsg *sysmsg, const struct mkfifoat_args *uap) 2399 { 2400 struct nlookupdata nd; 2401 struct file *fp; 2402 int error; 2403 2404 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2405 if (error == 0) 2406 error = kern_mkfifo(&nd, uap->mode); 2407 nlookup_done_at(&nd, fp); 2408 return (error); 2409 } 2410 2411 static int hardlink_check_uid = 0; 2412 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, 2413 &hardlink_check_uid, 0, 2414 "Unprivileged processes cannot create hard links to files owned by other " 2415 "users"); 2416 static int hardlink_check_gid = 0; 2417 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, 2418 &hardlink_check_gid, 0, 2419 "Unprivileged processes cannot create hard links to files owned by other " 2420 "groups"); 2421 2422 static int 2423 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred) 2424 { 2425 struct vattr va; 2426 int error; 2427 2428 /* 2429 * Shortcut if disabled 2430 */ 2431 if (hardlink_check_uid == 0 && hardlink_check_gid == 0) 2432 return (0); 2433 2434 /* 2435 * Privileged user can always hardlink 2436 */ 2437 if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0) 2438 return (0); 2439 2440 /* 2441 * Otherwise only if the originating file is owned by the 2442 * same user or group. Note that any group is allowed if 2443 * the file is owned by the caller. 2444 */ 2445 error = VOP_GETATTR(vp, &va); 2446 if (error != 0) 2447 return (error); 2448 2449 if (hardlink_check_uid) { 2450 if (cred->cr_uid != va.va_uid) 2451 return (EPERM); 2452 } 2453 2454 if (hardlink_check_gid) { 2455 if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred)) 2456 return (EPERM); 2457 } 2458 2459 return (0); 2460 } 2461 2462 int 2463 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd) 2464 { 2465 struct thread *td = curthread; 2466 struct vnode *vp; 2467 int error; 2468 2469 /* 2470 * Lookup the source and obtained a locked vnode. 2471 * 2472 * You may only hardlink a file which you have write permission 2473 * on or which you own. 2474 * 2475 * XXX relookup on vget failure / race ? 2476 */ 2477 bwillinode(1); 2478 nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK; 2479 if ((error = nlookup(nd)) != 0) 2480 return (error); 2481 vp = nd->nl_nch.ncp->nc_vp; 2482 KKASSERT(vp != NULL); 2483 if (vp->v_type == VDIR) 2484 return (EPERM); /* POSIX */ 2485 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2486 return (error); 2487 if ((error = vget(vp, LK_EXCLUSIVE)) != 0) 2488 return (error); 2489 2490 /* 2491 * Unlock the source so we can lookup the target without deadlocking 2492 * (XXX vp is locked already, possible other deadlock?). The target 2493 * must not exist. 2494 */ 2495 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 2496 nd->nl_flags &= ~NLC_NCPISLOCKED; 2497 cache_unlock(&nd->nl_nch); 2498 vn_unlock(vp); 2499 2500 linknd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2501 if ((error = nlookup(linknd)) != 0) { 2502 vrele(vp); 2503 return (error); 2504 } 2505 if (linknd->nl_nch.ncp->nc_vp) { 2506 vrele(vp); 2507 return (EEXIST); 2508 } 2509 if (linknd->nl_dvp == NULL) { 2510 vrele(vp); 2511 return (EINVAL); 2512 } 2513 VFS_MODIFYING(vp->v_mount); 2514 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 2515 if (error) { 2516 vrele(vp); 2517 return (error); 2518 } 2519 2520 /* 2521 * Finally run the new API VOP. 2522 */ 2523 error = can_hardlink(vp, td, td->td_ucred); 2524 if (error == 0) { 2525 error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp, 2526 vp, linknd->nl_cred); 2527 } 2528 vput(vp); 2529 return (error); 2530 } 2531 2532 /* 2533 * link_args(char *path, char *link) 2534 * 2535 * Make a hard file link. 2536 */ 2537 int 2538 sys_link(struct sysmsg *sysmsg, const struct link_args *uap) 2539 { 2540 struct nlookupdata nd, linknd; 2541 int error; 2542 2543 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2544 if (error == 0) { 2545 error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0); 2546 if (error == 0) 2547 error = kern_link(&nd, &linknd); 2548 nlookup_done(&linknd); 2549 } 2550 nlookup_done(&nd); 2551 return (error); 2552 } 2553 2554 /* 2555 * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags) 2556 * 2557 * Make a hard file link. The path1 argument is relative to the directory 2558 * associated with fd1, and similarly the path2 argument is relative to 2559 * the directory associated with fd2. 2560 */ 2561 int 2562 sys_linkat(struct sysmsg *sysmsg, const struct linkat_args *uap) 2563 { 2564 struct nlookupdata nd, linknd; 2565 struct file *fp1, *fp2; 2566 int error; 2567 2568 error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE, 2569 (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0); 2570 if (error == 0) { 2571 error = nlookup_init_at(&linknd, &fp2, uap->fd2, 2572 uap->path2, UIO_USERSPACE, 0); 2573 if (error == 0) 2574 error = kern_link(&nd, &linknd); 2575 nlookup_done_at(&linknd, fp2); 2576 } 2577 nlookup_done_at(&nd, fp1); 2578 return (error); 2579 } 2580 2581 int 2582 kern_symlink(struct nlookupdata *nd, char *path, int mode) 2583 { 2584 struct vattr vattr; 2585 struct vnode *vp; 2586 struct vnode *dvp; 2587 int error; 2588 2589 bwillinode(1); 2590 nd->nl_flags |= NLC_CREATE | NLC_REFDVP; 2591 if ((error = nlookup(nd)) != 0) 2592 return (error); 2593 if (nd->nl_nch.ncp->nc_vp) 2594 return (EEXIST); 2595 if (nd->nl_dvp == NULL) 2596 return (EINVAL); 2597 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2598 return (error); 2599 dvp = nd->nl_dvp; 2600 VATTR_NULL(&vattr); 2601 vattr.va_mode = mode; 2602 error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path); 2603 if (error == 0) 2604 vput(vp); 2605 return (error); 2606 } 2607 2608 /* 2609 * symlink(char *path, char *link) 2610 * 2611 * Make a symbolic link. 2612 */ 2613 int 2614 sys_symlink(struct sysmsg *sysmsg, const struct symlink_args *uap) 2615 { 2616 struct thread *td = curthread; 2617 struct nlookupdata nd; 2618 char *path; 2619 int error; 2620 int mode; 2621 2622 path = objcache_get(namei_oc, M_WAITOK); 2623 error = copyinstr(uap->path, path, MAXPATHLEN, NULL); 2624 if (error == 0) { 2625 error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0); 2626 if (error == 0) { 2627 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2628 error = kern_symlink(&nd, path, mode); 2629 } 2630 nlookup_done(&nd); 2631 } 2632 objcache_put(namei_oc, path); 2633 return (error); 2634 } 2635 2636 /* 2637 * symlinkat_args(char *path1, int fd, char *path2) 2638 * 2639 * Make a symbolic link. The path2 argument is relative to the directory 2640 * associated with fd. 2641 */ 2642 int 2643 sys_symlinkat(struct sysmsg *sysmsg, const struct symlinkat_args *uap) 2644 { 2645 struct thread *td = curthread; 2646 struct nlookupdata nd; 2647 struct file *fp; 2648 char *path1; 2649 int error; 2650 int mode; 2651 2652 path1 = objcache_get(namei_oc, M_WAITOK); 2653 error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL); 2654 if (error == 0) { 2655 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2, 2656 UIO_USERSPACE, 0); 2657 if (error == 0) { 2658 mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask; 2659 error = kern_symlink(&nd, path1, mode); 2660 } 2661 nlookup_done_at(&nd, fp); 2662 } 2663 objcache_put(namei_oc, path1); 2664 return (error); 2665 } 2666 2667 /* 2668 * undelete_args(char *path) 2669 * 2670 * Delete a whiteout from the filesystem. 2671 */ 2672 int 2673 sys_undelete(struct sysmsg *sysmsg, const struct undelete_args *uap) 2674 { 2675 struct nlookupdata nd; 2676 int error; 2677 2678 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2679 bwillinode(1); 2680 nd.nl_flags |= NLC_DELETE | NLC_REFDVP; 2681 if (error == 0) 2682 error = nlookup(&nd); 2683 if (error == 0 && nd.nl_dvp == NULL) 2684 error = EINVAL; 2685 if (error == 0) 2686 error = ncp_writechk(&nd.nl_nch); 2687 if (error == 0) { 2688 error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred, 2689 NAMEI_DELETE); 2690 } 2691 nlookup_done(&nd); 2692 return (error); 2693 } 2694 2695 int 2696 kern_unlink(struct nlookupdata *nd) 2697 { 2698 int error; 2699 2700 bwillinode(1); 2701 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 2702 if ((error = nlookup(nd)) != 0) 2703 return (error); 2704 if (nd->nl_dvp == NULL) 2705 return EINVAL; 2706 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 2707 return (error); 2708 error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 2709 return (error); 2710 } 2711 2712 /* 2713 * unlink_args(char *path) 2714 * 2715 * Delete a name from the filesystem. 2716 */ 2717 int 2718 sys_unlink(struct sysmsg *sysmsg, const struct unlink_args *uap) 2719 { 2720 struct nlookupdata nd; 2721 int error; 2722 2723 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 2724 if (error == 0) 2725 error = kern_unlink(&nd); 2726 nlookup_done(&nd); 2727 return (error); 2728 } 2729 2730 2731 /* 2732 * unlinkat_args(int fd, char *path, int flags) 2733 * 2734 * Delete the file or directory entry pointed to by fd/path. 2735 */ 2736 int 2737 sys_unlinkat(struct sysmsg *sysmsg, const struct unlinkat_args *uap) 2738 { 2739 struct nlookupdata nd; 2740 struct file *fp; 2741 int error; 2742 2743 if (uap->flags & ~AT_REMOVEDIR) 2744 return (EINVAL); 2745 2746 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 2747 if (error == 0) { 2748 if (uap->flags & AT_REMOVEDIR) 2749 error = kern_rmdir(&nd); 2750 else 2751 error = kern_unlink(&nd); 2752 } 2753 nlookup_done_at(&nd, fp); 2754 return (error); 2755 } 2756 2757 int 2758 kern_lseek(int fd, off_t offset, int whence, off_t *res) 2759 { 2760 struct thread *td = curthread; 2761 struct file *fp; 2762 struct vnode *vp; 2763 struct vattr_lite lva; 2764 off_t new_offset; 2765 int error; 2766 2767 fp = holdfp(td, fd, -1); 2768 if (fp == NULL) 2769 return (EBADF); 2770 if (fp->f_type != DTYPE_VNODE) { 2771 error = ESPIPE; 2772 goto done; 2773 } 2774 vp = (struct vnode *)fp->f_data; 2775 2776 switch (whence) { 2777 case L_INCR: 2778 spin_lock(&fp->f_spin); 2779 new_offset = fp->f_offset + offset; 2780 error = 0; 2781 break; 2782 case L_XTND: 2783 error = VOP_GETATTR_LITE(vp, &lva); 2784 spin_lock(&fp->f_spin); 2785 new_offset = offset + lva.va_size; 2786 break; 2787 case L_SET: 2788 new_offset = offset; 2789 error = 0; 2790 spin_lock(&fp->f_spin); 2791 break; 2792 default: 2793 new_offset = 0; 2794 error = EINVAL; 2795 spin_lock(&fp->f_spin); 2796 break; 2797 } 2798 2799 /* 2800 * Validate the seek position. Negative offsets are not allowed 2801 * for regular files or directories. 2802 * 2803 * Normally we would also not want to allow negative offsets for 2804 * character and block-special devices. However kvm addresses 2805 * on 64 bit architectures might appear to be negative and must 2806 * be allowed. 2807 */ 2808 if (error == 0) { 2809 if (new_offset < 0 && 2810 (vp->v_type == VREG || vp->v_type == VDIR)) { 2811 error = EINVAL; 2812 } else { 2813 fp->f_offset = new_offset; 2814 } 2815 } 2816 *res = fp->f_offset; 2817 spin_unlock(&fp->f_spin); 2818 done: 2819 dropfp(td, fd, fp); 2820 2821 return (error); 2822 } 2823 2824 /* 2825 * lseek_args(int fd, int pad, off_t offset, int whence) 2826 * 2827 * Reposition read/write file offset. 2828 */ 2829 int 2830 sys_lseek(struct sysmsg *sysmsg, const struct lseek_args *uap) 2831 { 2832 int error; 2833 2834 error = kern_lseek(uap->fd, uap->offset, uap->whence, 2835 &sysmsg->sysmsg_offset); 2836 2837 return (error); 2838 } 2839 2840 /* 2841 * Check if current process can access given file. amode is a bitmask of *_OK 2842 * access bits. flags is a bitmask of AT_* flags. 2843 */ 2844 int 2845 kern_access(struct nlookupdata *nd, int amode, int flags) 2846 { 2847 struct vnode *vp; 2848 int error, mode; 2849 2850 if (flags & ~AT_EACCESS) 2851 return (EINVAL); 2852 nd->nl_flags |= NLC_SHAREDLOCK; 2853 if ((error = nlookup(nd)) != 0) 2854 return (error); 2855 if ((amode & W_OK) && (error = ncp_writechk(&nd->nl_nch)) != 0) 2856 return (error); 2857 retry: 2858 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp); 2859 if (error) 2860 return (error); 2861 2862 /* Flags == 0 means only check for existence. */ 2863 if (amode) { 2864 mode = 0; 2865 if (amode & R_OK) 2866 mode |= VREAD; 2867 if (amode & W_OK) 2868 mode |= VWRITE; 2869 if (amode & X_OK) 2870 mode |= VEXEC; 2871 if ((mode & VWRITE) == 0 || 2872 (error = vn_writechk(vp)) == 0) { 2873 error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred); 2874 } 2875 2876 /* 2877 * If the file handle is stale we have to re-resolve the 2878 * entry with the ncp held exclusively. This is a hack 2879 * at the moment. 2880 */ 2881 if (error == ESTALE) { 2882 vput(vp); 2883 cache_unlock(&nd->nl_nch); 2884 cache_lock(&nd->nl_nch); 2885 cache_setunresolved(&nd->nl_nch); 2886 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2887 if (error == 0) { 2888 vp = NULL; 2889 goto retry; 2890 } 2891 return(error); 2892 } 2893 } 2894 vput(vp); 2895 return (error); 2896 } 2897 2898 /* 2899 * access_args(char *path, int flags) 2900 * 2901 * Check access permissions. 2902 */ 2903 int 2904 sys_access(struct sysmsg *sysmsg, const struct access_args *uap) 2905 { 2906 struct nlookupdata nd; 2907 int error; 2908 2909 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2910 if (error == 0) 2911 error = kern_access(&nd, uap->flags, 0); 2912 nlookup_done(&nd); 2913 return (error); 2914 } 2915 2916 2917 /* 2918 * eaccess_args(char *path, int flags) 2919 * 2920 * Check access permissions. 2921 */ 2922 int 2923 sys_eaccess(struct sysmsg *sysmsg, const struct eaccess_args *uap) 2924 { 2925 struct nlookupdata nd; 2926 int error; 2927 2928 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 2929 if (error == 0) 2930 error = kern_access(&nd, uap->flags, AT_EACCESS); 2931 nlookup_done(&nd); 2932 return (error); 2933 } 2934 2935 2936 /* 2937 * faccessat_args(int fd, char *path, int amode, int flags) 2938 * 2939 * Check access permissions. 2940 */ 2941 int 2942 sys_faccessat(struct sysmsg *sysmsg, const struct faccessat_args *uap) 2943 { 2944 struct nlookupdata nd; 2945 struct file *fp; 2946 int error; 2947 2948 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 2949 NLC_FOLLOW); 2950 if (error == 0) 2951 error = kern_access(&nd, uap->amode, uap->flags); 2952 nlookup_done_at(&nd, fp); 2953 return (error); 2954 } 2955 2956 int 2957 kern_stat(struct nlookupdata *nd, struct stat *st) 2958 { 2959 int error; 2960 struct vnode *vp; 2961 2962 nd->nl_flags |= NLC_SHAREDLOCK; 2963 if ((error = nlookup(nd)) != 0) 2964 return (error); 2965 again: 2966 if ((vp = nd->nl_nch.ncp->nc_vp) == NULL) 2967 return (ENOENT); 2968 2969 #if 1 2970 error = cache_vref(&nd->nl_nch, NULL, &vp); 2971 #else 2972 error = vget(vp, LK_SHARED); 2973 #endif 2974 if (error) 2975 return (error); 2976 error = vn_stat(vp, st, nd->nl_cred); 2977 2978 /* 2979 * If the file handle is stale we have to re-resolve the 2980 * entry with the ncp held exclusively. This is a hack 2981 * at the moment. 2982 */ 2983 if (error == ESTALE) { 2984 #if 1 2985 vrele(vp); 2986 #else 2987 vput(vp); 2988 #endif 2989 cache_unlock(&nd->nl_nch); 2990 cache_lock(&nd->nl_nch); 2991 cache_setunresolved(&nd->nl_nch); 2992 error = cache_resolve(&nd->nl_nch, nd->nl_cred); 2993 if (error == 0) 2994 goto again; 2995 } else { 2996 #if 1 2997 vrele(vp); 2998 #else 2999 vput(vp); 3000 #endif 3001 } 3002 return (error); 3003 } 3004 3005 /* 3006 * stat_args(char *path, struct stat *ub) 3007 * 3008 * Get file status; this version follows links. 3009 */ 3010 int 3011 sys_stat(struct sysmsg *sysmsg, const struct stat_args *uap) 3012 { 3013 struct nlookupdata nd; 3014 struct stat st; 3015 int error; 3016 3017 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3018 if (error == 0) { 3019 error = kern_stat(&nd, &st); 3020 if (error == 0) 3021 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 3022 } 3023 nlookup_done(&nd); 3024 return (error); 3025 } 3026 3027 /* 3028 * lstat_args(char *path, struct stat *ub) 3029 * 3030 * Get file status; this version does not follow links. 3031 */ 3032 int 3033 sys_lstat(struct sysmsg *sysmsg, const struct lstat_args *uap) 3034 { 3035 struct nlookupdata nd; 3036 struct stat st; 3037 int error; 3038 3039 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3040 if (error == 0) { 3041 error = kern_stat(&nd, &st); 3042 if (error == 0) 3043 error = copyout(&st, uap->ub, sizeof(*uap->ub)); 3044 } 3045 nlookup_done(&nd); 3046 return (error); 3047 } 3048 3049 /* 3050 * fstatat_args(int fd, char *path, struct stat *sb, int flags) 3051 * 3052 * Get status of file pointed to by fd/path. 3053 */ 3054 int 3055 sys_fstatat(struct sysmsg *sysmsg, const struct fstatat_args *uap) 3056 { 3057 struct nlookupdata nd; 3058 struct stat st; 3059 int error; 3060 int flags; 3061 struct file *fp; 3062 3063 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3064 return (EINVAL); 3065 3066 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3067 3068 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3069 UIO_USERSPACE, flags); 3070 if (error == 0) { 3071 error = kern_stat(&nd, &st); 3072 if (error == 0) 3073 error = copyout(&st, uap->sb, sizeof(*uap->sb)); 3074 } 3075 nlookup_done_at(&nd, fp); 3076 return (error); 3077 } 3078 3079 static int 3080 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp) 3081 { 3082 struct nlookupdata nd; 3083 struct vnode *vp; 3084 int error; 3085 3086 vp = NULL; 3087 error = nlookup_init(&nd, path, UIO_USERSPACE, flags); 3088 if (error == 0) 3089 error = nlookup(&nd); 3090 if (error == 0) 3091 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 3092 nlookup_done(&nd); 3093 if (error == 0) { 3094 error = VOP_PATHCONF(vp, name, sysmsg_regp); 3095 vput(vp); 3096 } 3097 return (error); 3098 } 3099 3100 /* 3101 * pathconf_Args(char *path, int name) 3102 * 3103 * Get configurable pathname variables. 3104 */ 3105 int 3106 sys_pathconf(struct sysmsg *sysmsg, const struct pathconf_args *uap) 3107 { 3108 return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW, 3109 &sysmsg->sysmsg_reg)); 3110 } 3111 3112 /* 3113 * lpathconf_Args(char *path, int name) 3114 * 3115 * Get configurable pathname variables, but don't follow symlinks. 3116 */ 3117 int 3118 sys_lpathconf(struct sysmsg *sysmsg, const struct lpathconf_args *uap) 3119 { 3120 return (kern_pathconf(uap->path, uap->name, 0, &sysmsg->sysmsg_reg)); 3121 } 3122 3123 /* 3124 * XXX: daver 3125 * kern_readlink isn't properly split yet. There is a copyin burried 3126 * in VOP_READLINK(). 3127 */ 3128 int 3129 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res) 3130 { 3131 struct thread *td = curthread; 3132 struct vnode *vp; 3133 struct iovec aiov; 3134 struct uio auio; 3135 int error; 3136 3137 nd->nl_flags |= NLC_SHAREDLOCK; 3138 if ((error = nlookup(nd)) != 0) 3139 return (error); 3140 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp); 3141 if (error) 3142 return (error); 3143 if (vp->v_type != VLNK) { 3144 error = EINVAL; 3145 } else { 3146 aiov.iov_base = buf; 3147 aiov.iov_len = count; 3148 auio.uio_iov = &aiov; 3149 auio.uio_iovcnt = 1; 3150 auio.uio_offset = 0; 3151 auio.uio_rw = UIO_READ; 3152 auio.uio_segflg = UIO_USERSPACE; 3153 auio.uio_td = td; 3154 auio.uio_resid = count; 3155 error = VOP_READLINK(vp, &auio, td->td_ucred); 3156 } 3157 vput(vp); 3158 *res = count - auio.uio_resid; 3159 return (error); 3160 } 3161 3162 /* 3163 * readlink_args(char *path, char *buf, int count) 3164 * 3165 * Return target name of a symbolic link. 3166 */ 3167 int 3168 sys_readlink(struct sysmsg *sysmsg, const struct readlink_args *uap) 3169 { 3170 struct nlookupdata nd; 3171 int error; 3172 3173 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3174 if (error == 0) { 3175 error = kern_readlink(&nd, uap->buf, uap->count, 3176 &sysmsg->sysmsg_result); 3177 } 3178 nlookup_done(&nd); 3179 return (error); 3180 } 3181 3182 /* 3183 * readlinkat_args(int fd, char *path, char *buf, size_t bufsize) 3184 * 3185 * Return target name of a symbolic link. The path is relative to the 3186 * directory associated with fd. 3187 */ 3188 int 3189 sys_readlinkat(struct sysmsg *sysmsg, const struct readlinkat_args *uap) 3190 { 3191 struct nlookupdata nd; 3192 struct file *fp; 3193 int error; 3194 3195 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 3196 if (error == 0) { 3197 error = kern_readlink(&nd, uap->buf, uap->bufsize, 3198 &sysmsg->sysmsg_result); 3199 } 3200 nlookup_done_at(&nd, fp); 3201 return (error); 3202 } 3203 3204 static int 3205 setfflags(struct vnode *vp, u_long flags) 3206 { 3207 struct thread *td = curthread; 3208 int error; 3209 struct vattr vattr; 3210 3211 /* 3212 * Prevent non-root users from setting flags on devices. When 3213 * a device is reused, users can retain ownership of the device 3214 * if they are allowed to set flags and programs assume that 3215 * chown can't fail when done as root. 3216 */ 3217 if ((vp->v_type == VCHR || vp->v_type == VBLK) && 3218 ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0)) 3219 return (error); 3220 3221 /* 3222 * note: vget is required for any operation that might mod the vnode 3223 * so VINACTIVE is properly cleared. 3224 */ 3225 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3226 VATTR_NULL(&vattr); 3227 vattr.va_flags = flags; 3228 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3229 vput(vp); 3230 } 3231 return (error); 3232 } 3233 3234 /* 3235 * chflags(const char *path, u_long flags) 3236 * 3237 * Change flags of a file given a path name. 3238 */ 3239 int 3240 sys_chflags(struct sysmsg *sysmsg, const struct chflags_args *uap) 3241 { 3242 struct nlookupdata nd; 3243 struct vnode *vp; 3244 int error; 3245 3246 vp = NULL; 3247 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3248 if (error == 0) 3249 error = nlookup(&nd); 3250 if (error == 0) 3251 error = ncp_writechk(&nd.nl_nch); 3252 if (error == 0) 3253 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3254 nlookup_done(&nd); 3255 if (error == 0) { 3256 error = setfflags(vp, uap->flags); 3257 vrele(vp); 3258 } 3259 return (error); 3260 } 3261 3262 /* 3263 * lchflags(const char *path, u_long flags) 3264 * 3265 * Change flags of a file given a path name, but don't follow symlinks. 3266 */ 3267 int 3268 sys_lchflags(struct sysmsg *sysmsg, const struct lchflags_args *uap) 3269 { 3270 struct nlookupdata nd; 3271 struct vnode *vp; 3272 int error; 3273 3274 vp = NULL; 3275 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3276 if (error == 0) 3277 error = nlookup(&nd); 3278 if (error == 0) 3279 error = ncp_writechk(&nd.nl_nch); 3280 if (error == 0) 3281 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3282 nlookup_done(&nd); 3283 if (error == 0) { 3284 error = setfflags(vp, uap->flags); 3285 vrele(vp); 3286 } 3287 return (error); 3288 } 3289 3290 /* 3291 * fchflags_args(int fd, u_flags flags) 3292 * 3293 * Change flags of a file given a file descriptor. 3294 */ 3295 int 3296 sys_fchflags(struct sysmsg *sysmsg, const struct fchflags_args *uap) 3297 { 3298 struct thread *td = curthread; 3299 struct file *fp; 3300 int error; 3301 3302 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 3303 return (error); 3304 if (fp->f_nchandle.ncp) 3305 error = ncp_writechk(&fp->f_nchandle); 3306 if (error == 0) 3307 error = setfflags((struct vnode *) fp->f_data, uap->flags); 3308 fdrop(fp); 3309 return (error); 3310 } 3311 3312 /* 3313 * chflagsat_args(int fd, const char *path, u_long flags, int atflags) 3314 * change flags given a pathname relative to a filedescriptor 3315 */ 3316 int 3317 sys_chflagsat(struct sysmsg *sysmsg, const struct chflagsat_args *uap) 3318 { 3319 struct nlookupdata nd; 3320 struct vnode *vp; 3321 struct file *fp; 3322 int error; 3323 int lookupflags; 3324 3325 if (uap->atflags & ~AT_SYMLINK_NOFOLLOW) 3326 return (EINVAL); 3327 3328 lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3329 3330 vp = NULL; 3331 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, lookupflags); 3332 if (error == 0) 3333 error = nlookup(&nd); 3334 if (error == 0) 3335 error = ncp_writechk(&nd.nl_nch); 3336 if (error == 0) 3337 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 3338 nlookup_done_at(&nd, fp); 3339 if (error == 0) { 3340 error = setfflags(vp, uap->flags); 3341 vrele(vp); 3342 } 3343 return (error); 3344 } 3345 3346 3347 static int 3348 setfmode(struct vnode *vp, int mode) 3349 { 3350 struct thread *td = curthread; 3351 int error; 3352 struct vattr vattr; 3353 3354 /* 3355 * note: vget is required for any operation that might mod the vnode 3356 * so VINACTIVE is properly cleared. 3357 */ 3358 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3359 VATTR_NULL(&vattr); 3360 vattr.va_mode = mode & ALLPERMS; 3361 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3362 cache_inval_wxok(vp); 3363 vput(vp); 3364 } 3365 return error; 3366 } 3367 3368 int 3369 kern_chmod(struct nlookupdata *nd, int mode) 3370 { 3371 struct vnode *vp; 3372 int error; 3373 3374 if ((error = nlookup(nd)) != 0) 3375 return (error); 3376 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3377 return (error); 3378 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3379 error = setfmode(vp, mode); 3380 vrele(vp); 3381 return (error); 3382 } 3383 3384 /* 3385 * chmod_args(char *path, int mode) 3386 * 3387 * Change mode of a file given path name. 3388 */ 3389 int 3390 sys_chmod(struct sysmsg *sysmsg, const struct chmod_args *uap) 3391 { 3392 struct nlookupdata nd; 3393 int error; 3394 3395 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3396 if (error == 0) 3397 error = kern_chmod(&nd, uap->mode); 3398 nlookup_done(&nd); 3399 return (error); 3400 } 3401 3402 /* 3403 * lchmod_args(char *path, int mode) 3404 * 3405 * Change mode of a file given path name (don't follow links.) 3406 */ 3407 int 3408 sys_lchmod(struct sysmsg *sysmsg, const struct lchmod_args *uap) 3409 { 3410 struct nlookupdata nd; 3411 int error; 3412 3413 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3414 if (error == 0) 3415 error = kern_chmod(&nd, uap->mode); 3416 nlookup_done(&nd); 3417 return (error); 3418 } 3419 3420 /* 3421 * fchmod_args(int fd, int mode) 3422 * 3423 * Change mode of a file given a file descriptor. 3424 */ 3425 int 3426 sys_fchmod(struct sysmsg *sysmsg, const struct fchmod_args *uap) 3427 { 3428 struct thread *td = curthread; 3429 struct file *fp; 3430 int error; 3431 3432 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 3433 return (error); 3434 if (fp->f_nchandle.ncp) 3435 error = ncp_writechk(&fp->f_nchandle); 3436 if (error == 0) 3437 error = setfmode((struct vnode *)fp->f_data, uap->mode); 3438 fdrop(fp); 3439 return (error); 3440 } 3441 3442 /* 3443 * fchmodat_args(char *path, int mode) 3444 * 3445 * Change mode of a file pointed to by fd/path. 3446 */ 3447 int 3448 sys_fchmodat(struct sysmsg *sysmsg, const struct fchmodat_args *uap) 3449 { 3450 struct nlookupdata nd; 3451 struct file *fp; 3452 int error; 3453 int flags; 3454 3455 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3456 return (EINVAL); 3457 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3458 3459 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3460 UIO_USERSPACE, flags); 3461 if (error == 0) 3462 error = kern_chmod(&nd, uap->mode); 3463 nlookup_done_at(&nd, fp); 3464 return (error); 3465 } 3466 3467 static int 3468 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid) 3469 { 3470 struct thread *td = curthread; 3471 int error; 3472 struct vattr vattr; 3473 uid_t o_uid; 3474 gid_t o_gid; 3475 uint64_t size; 3476 3477 /* 3478 * note: vget is required for any operation that might mod the vnode 3479 * so VINACTIVE is properly cleared. 3480 */ 3481 if ((error = vget(vp, LK_EXCLUSIVE)) == 0) { 3482 if ((error = VOP_GETATTR(vp, &vattr)) != 0) 3483 return error; 3484 o_uid = vattr.va_uid; 3485 o_gid = vattr.va_gid; 3486 size = vattr.va_size; 3487 3488 VATTR_NULL(&vattr); 3489 vattr.va_uid = uid; 3490 vattr.va_gid = gid; 3491 error = VOP_SETATTR(vp, &vattr, td->td_ucred); 3492 vput(vp); 3493 } 3494 3495 if (error == 0) { 3496 if (uid == -1) 3497 uid = o_uid; 3498 if (gid == -1) 3499 gid = o_gid; 3500 VFS_ACCOUNT(mp, o_uid, o_gid, -size); 3501 VFS_ACCOUNT(mp, uid, gid, size); 3502 } 3503 3504 return error; 3505 } 3506 3507 int 3508 kern_chown(struct nlookupdata *nd, int uid, int gid) 3509 { 3510 struct vnode *vp; 3511 int error; 3512 3513 if ((error = nlookup(nd)) != 0) 3514 return (error); 3515 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3516 return (error); 3517 if ((error = ncp_writechk(&nd->nl_nch)) == 0) 3518 error = setfown(nd->nl_nch.mount, vp, uid, gid); 3519 vrele(vp); 3520 return (error); 3521 } 3522 3523 /* 3524 * chown(char *path, int uid, int gid) 3525 * 3526 * Set ownership given a path name. 3527 */ 3528 int 3529 sys_chown(struct sysmsg *sysmsg, const struct chown_args *uap) 3530 { 3531 struct nlookupdata nd; 3532 int error; 3533 3534 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3535 if (error == 0) 3536 error = kern_chown(&nd, uap->uid, uap->gid); 3537 nlookup_done(&nd); 3538 return (error); 3539 } 3540 3541 /* 3542 * lchown_args(char *path, int uid, int gid) 3543 * 3544 * Set ownership given a path name, do not cross symlinks. 3545 */ 3546 int 3547 sys_lchown(struct sysmsg *sysmsg, const struct lchown_args *uap) 3548 { 3549 struct nlookupdata nd; 3550 int error; 3551 3552 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3553 if (error == 0) 3554 error = kern_chown(&nd, uap->uid, uap->gid); 3555 nlookup_done(&nd); 3556 return (error); 3557 } 3558 3559 /* 3560 * fchown_args(int fd, int uid, int gid) 3561 * 3562 * Set ownership given a file descriptor. 3563 */ 3564 int 3565 sys_fchown(struct sysmsg *sysmsg, const struct fchown_args *uap) 3566 { 3567 struct thread *td = curthread; 3568 struct proc *p = td->td_proc; 3569 struct file *fp; 3570 int error; 3571 3572 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 3573 return (error); 3574 if (fp->f_nchandle.ncp) 3575 error = ncp_writechk(&fp->f_nchandle); 3576 if (error == 0) 3577 error = setfown(p->p_fd->fd_ncdir.mount, 3578 (struct vnode *)fp->f_data, uap->uid, uap->gid); 3579 fdrop(fp); 3580 return (error); 3581 } 3582 3583 /* 3584 * fchownat(int fd, char *path, int uid, int gid, int flags) 3585 * 3586 * Set ownership of file pointed to by fd/path. 3587 */ 3588 int 3589 sys_fchownat(struct sysmsg *sysmsg, const struct fchownat_args *uap) 3590 { 3591 struct nlookupdata nd; 3592 struct file *fp; 3593 int error; 3594 int flags; 3595 3596 if (uap->flags & ~AT_SYMLINK_NOFOLLOW) 3597 return (EINVAL); 3598 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3599 3600 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3601 UIO_USERSPACE, flags); 3602 if (error == 0) 3603 error = kern_chown(&nd, uap->uid, uap->gid); 3604 nlookup_done_at(&nd, fp); 3605 return (error); 3606 } 3607 3608 3609 static int 3610 getutimes(struct timeval *tvp, struct timespec *tsp) 3611 { 3612 struct timeval tv[2]; 3613 int error; 3614 3615 if (tvp == NULL) { 3616 microtime(&tv[0]); 3617 TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]); 3618 tsp[1] = tsp[0]; 3619 } else { 3620 if ((error = itimerfix(tvp)) != 0) 3621 return (error); 3622 TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); 3623 TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); 3624 } 3625 return 0; 3626 } 3627 3628 static int 3629 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag) 3630 { 3631 struct timespec tsnow; 3632 int error; 3633 3634 *nullflag = 0; 3635 nanotime(&tsnow); 3636 if (ts == NULL) { 3637 newts[0] = tsnow; 3638 newts[1] = tsnow; 3639 *nullflag = 1; 3640 return (0); 3641 } 3642 3643 newts[0] = ts[0]; 3644 newts[1] = ts[1]; 3645 if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT) 3646 return (0); 3647 if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW) 3648 *nullflag = 1; 3649 3650 if (newts[0].tv_nsec == UTIME_OMIT) 3651 newts[0].tv_sec = VNOVAL; 3652 else if (newts[0].tv_nsec == UTIME_NOW) 3653 newts[0] = tsnow; 3654 else if ((error = itimespecfix(&newts[0])) != 0) 3655 return (error); 3656 3657 if (newts[1].tv_nsec == UTIME_OMIT) 3658 newts[1].tv_sec = VNOVAL; 3659 else if (newts[1].tv_nsec == UTIME_NOW) 3660 newts[1] = tsnow; 3661 else if ((error = itimespecfix(&newts[1])) != 0) 3662 return (error); 3663 3664 return (0); 3665 } 3666 3667 static int 3668 setutimes(struct vnode *vp, struct vattr *vattr, 3669 const struct timespec *ts, int nullflag) 3670 { 3671 struct thread *td = curthread; 3672 int error; 3673 3674 VATTR_NULL(vattr); 3675 vattr->va_atime = ts[0]; 3676 vattr->va_mtime = ts[1]; 3677 if (nullflag) 3678 vattr->va_vaflags |= VA_UTIMES_NULL; 3679 error = VOP_SETATTR(vp, vattr, td->td_ucred); 3680 3681 return error; 3682 } 3683 3684 int 3685 kern_utimes(struct nlookupdata *nd, struct timeval *tptr) 3686 { 3687 struct timespec ts[2]; 3688 int error; 3689 3690 if (tptr) { 3691 if ((error = getutimes(tptr, ts)) != 0) 3692 return (error); 3693 } 3694 error = kern_utimensat(nd, tptr ? ts : NULL, 0); 3695 return (error); 3696 } 3697 3698 /* 3699 * utimes_args(char *path, struct timeval *tptr) 3700 * 3701 * Set the access and modification times of a file. 3702 */ 3703 int 3704 sys_utimes(struct sysmsg *sysmsg, const struct utimes_args *uap) 3705 { 3706 struct timeval tv[2]; 3707 struct nlookupdata nd; 3708 int error; 3709 3710 if (uap->tptr) { 3711 error = copyin(uap->tptr, tv, sizeof(tv)); 3712 if (error) 3713 return (error); 3714 } 3715 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3716 if (error == 0) 3717 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3718 nlookup_done(&nd); 3719 return (error); 3720 } 3721 3722 /* 3723 * lutimes_args(char *path, struct timeval *tptr) 3724 * 3725 * Set the access and modification times of a file. 3726 */ 3727 int 3728 sys_lutimes(struct sysmsg *sysmsg, const struct lutimes_args *uap) 3729 { 3730 struct timeval tv[2]; 3731 struct nlookupdata nd; 3732 int error; 3733 3734 if (uap->tptr) { 3735 error = copyin(uap->tptr, tv, sizeof(tv)); 3736 if (error) 3737 return (error); 3738 } 3739 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 3740 if (error == 0) 3741 error = kern_utimes(&nd, uap->tptr ? tv : NULL); 3742 nlookup_done(&nd); 3743 return (error); 3744 } 3745 3746 /* 3747 * Set utimes on a file descriptor. The creds used to open the 3748 * file are used to determine whether the operation is allowed 3749 * or not. 3750 */ 3751 int 3752 kern_futimens(int fd, struct timespec *ts) 3753 { 3754 struct thread *td = curthread; 3755 struct timespec newts[2]; 3756 struct file *fp; 3757 struct vnode *vp; 3758 struct vattr vattr; 3759 struct vattr_lite lva; 3760 int nullflag; 3761 int error; 3762 3763 error = getutimens(ts, newts, &nullflag); 3764 if (error) 3765 return (error); 3766 if ((error = holdvnode(td, fd, &fp)) != 0) 3767 return (error); 3768 if (fp->f_nchandle.ncp) 3769 error = ncp_writechk(&fp->f_nchandle); 3770 if (error == 0) { 3771 vp = fp->f_data; 3772 error = vget(vp, LK_EXCLUSIVE); 3773 if (error == 0) { 3774 error = VOP_GETATTR_FP(vp, &vattr, fp); 3775 if (error == 0) { 3776 lva.va_type = vattr.va_type; 3777 lva.va_nlink = vattr.va_nlink; 3778 lva.va_mode = vattr.va_mode; 3779 lva.va_uid = vattr.va_uid; 3780 lva.va_gid = vattr.va_gid; 3781 lva.va_size = vattr.va_size; 3782 lva.va_flags = vattr.va_flags; 3783 3784 error = naccess_lva(&lva, NLC_OWN | NLC_WRITE, 3785 fp->f_cred); 3786 } 3787 if (error == 0) { 3788 error = setutimes(vp, &vattr, newts, nullflag); 3789 } 3790 vput(vp); 3791 } 3792 } 3793 fdrop(fp); 3794 return (error); 3795 } 3796 3797 /* 3798 * futimens_args(int fd, struct timespec *ts) 3799 * 3800 * Set the access and modification times of a file. 3801 */ 3802 int 3803 sys_futimens(struct sysmsg *sysmsg, const struct futimens_args *uap) 3804 { 3805 struct timespec ts[2]; 3806 int error; 3807 3808 if (uap->ts) { 3809 error = copyin(uap->ts, ts, sizeof(ts)); 3810 if (error) 3811 return (error); 3812 } 3813 error = kern_futimens(uap->fd, uap->ts ? ts : NULL); 3814 return (error); 3815 } 3816 3817 int 3818 kern_futimes(int fd, struct timeval *tptr) 3819 { 3820 struct timespec ts[2]; 3821 int error; 3822 3823 if (tptr) { 3824 if ((error = getutimes(tptr, ts)) != 0) 3825 return (error); 3826 } 3827 error = kern_futimens(fd, tptr ? ts : NULL); 3828 return (error); 3829 } 3830 3831 /* 3832 * futimes_args(int fd, struct timeval *tptr) 3833 * 3834 * Set the access and modification times of a file. 3835 */ 3836 int 3837 sys_futimes(struct sysmsg *sysmsg, const struct futimes_args *uap) 3838 { 3839 struct timeval tv[2]; 3840 int error; 3841 3842 if (uap->tptr) { 3843 error = copyin(uap->tptr, tv, sizeof(tv)); 3844 if (error) 3845 return (error); 3846 } 3847 error = kern_futimes(uap->fd, uap->tptr ? tv : NULL); 3848 return (error); 3849 } 3850 3851 int 3852 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags) 3853 { 3854 struct timespec newts[2]; 3855 struct vnode *vp; 3856 struct vattr vattr; 3857 int nullflag; 3858 int error; 3859 3860 if (flags & ~AT_SYMLINK_NOFOLLOW) 3861 return (EINVAL); 3862 3863 error = getutimens(ts, newts, &nullflag); 3864 if (error) 3865 return (error); 3866 3867 nd->nl_flags |= NLC_OWN | NLC_WRITE; 3868 if ((error = nlookup(nd)) != 0) 3869 return (error); 3870 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3871 return (error); 3872 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3873 return (error); 3874 if ((error = vn_writechk(vp)) == 0) { 3875 error = vget(vp, LK_EXCLUSIVE); 3876 if (error == 0) { 3877 error = setutimes(vp, &vattr, newts, nullflag); 3878 vput(vp); 3879 } 3880 } 3881 vrele(vp); 3882 return (error); 3883 } 3884 3885 /* 3886 * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags); 3887 * 3888 * Set file access and modification times of a file. 3889 */ 3890 int 3891 sys_utimensat(struct sysmsg *sysmsg, const struct utimensat_args *uap) 3892 { 3893 struct timespec ts[2]; 3894 struct nlookupdata nd; 3895 struct file *fp; 3896 int error; 3897 int flags; 3898 3899 if (uap->ts) { 3900 error = copyin(uap->ts, ts, sizeof(ts)); 3901 if (error) 3902 return (error); 3903 } 3904 3905 flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW; 3906 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, 3907 UIO_USERSPACE, flags); 3908 if (error == 0) 3909 error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags); 3910 nlookup_done_at(&nd, fp); 3911 return (error); 3912 } 3913 3914 int 3915 kern_truncate(struct nlookupdata *nd, off_t length) 3916 { 3917 struct vnode *vp; 3918 struct vattr vattr; 3919 int error; 3920 uid_t uid = 0; 3921 gid_t gid = 0; 3922 uint64_t old_size = 0; 3923 3924 if (length < 0) 3925 return(EINVAL); 3926 nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE; 3927 if ((error = nlookup(nd)) != 0) 3928 return (error); 3929 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 3930 return (error); 3931 if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0) 3932 return (error); 3933 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 3934 if (error) { 3935 vrele(vp); 3936 return (error); 3937 } 3938 if (vp->v_type == VDIR) { 3939 error = EISDIR; 3940 goto done; 3941 } 3942 if (vfs_quota_enabled) { 3943 error = VOP_GETATTR(vp, &vattr); 3944 KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0")); 3945 uid = vattr.va_uid; 3946 gid = vattr.va_gid; 3947 old_size = vattr.va_size; 3948 } 3949 3950 if ((error = vn_writechk(vp)) == 0) { 3951 VATTR_NULL(&vattr); 3952 vattr.va_size = length; 3953 error = VOP_SETATTR(vp, &vattr, nd->nl_cred); 3954 VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size); 3955 } 3956 done: 3957 vput(vp); 3958 return (error); 3959 } 3960 3961 /* 3962 * truncate(char *path, int pad, off_t length) 3963 * 3964 * Truncate a file given its path name. 3965 */ 3966 int 3967 sys_truncate(struct sysmsg *sysmsg, const struct truncate_args *uap) 3968 { 3969 struct nlookupdata nd; 3970 int error; 3971 3972 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 3973 if (error == 0) 3974 error = kern_truncate(&nd, uap->length); 3975 nlookup_done(&nd); 3976 return error; 3977 } 3978 3979 int 3980 kern_ftruncate(int fd, off_t length) 3981 { 3982 struct thread *td = curthread; 3983 struct vattr vattr; 3984 struct vnode *vp; 3985 struct file *fp; 3986 int error; 3987 uid_t uid = 0; 3988 gid_t gid = 0; 3989 uint64_t old_size = 0; 3990 struct mount *mp; 3991 3992 if (length < 0) 3993 return(EINVAL); 3994 if ((error = holdvnode(td, fd, &fp)) != 0) 3995 return (error); 3996 if (fp->f_nchandle.ncp) { 3997 error = ncp_writechk(&fp->f_nchandle); 3998 if (error) 3999 goto done; 4000 } 4001 if ((fp->f_flag & FWRITE) == 0) { 4002 error = EINVAL; 4003 goto done; 4004 } 4005 if (fp->f_flag & FAPPENDONLY) { /* inode was set s/uapnd */ 4006 error = EINVAL; 4007 goto done; 4008 } 4009 vp = (struct vnode *)fp->f_data; 4010 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4011 if (vp->v_type == VDIR) { 4012 error = EISDIR; 4013 vn_unlock(vp); 4014 goto done; 4015 } 4016 4017 if (vfs_quota_enabled) { 4018 error = VOP_GETATTR_FP(vp, &vattr, fp); 4019 KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0")); 4020 uid = vattr.va_uid; 4021 gid = vattr.va_gid; 4022 old_size = vattr.va_size; 4023 } 4024 4025 if ((error = vn_writechk(vp)) == 0) { 4026 VATTR_NULL(&vattr); 4027 vattr.va_size = length; 4028 error = VOP_SETATTR_FP(vp, &vattr, fp->f_cred, fp); 4029 mp = vq_vptomp(vp); 4030 VFS_ACCOUNT(mp, uid, gid, length - old_size); 4031 } 4032 vn_unlock(vp); 4033 done: 4034 fdrop(fp); 4035 return (error); 4036 } 4037 4038 /* 4039 * ftruncate_args(int fd, int pad, off_t length) 4040 * 4041 * Truncate a file given a file descriptor. 4042 */ 4043 int 4044 sys_ftruncate(struct sysmsg *sysmsg, const struct ftruncate_args *uap) 4045 { 4046 int error; 4047 4048 error = kern_ftruncate(uap->fd, uap->length); 4049 4050 return (error); 4051 } 4052 4053 /* 4054 * fsync(int fd) 4055 * 4056 * Sync an open file. 4057 */ 4058 int 4059 sys_fsync(struct sysmsg *sysmsg, const struct fsync_args *uap) 4060 { 4061 struct thread *td = curthread; 4062 struct vnode *vp; 4063 struct file *fp; 4064 vm_object_t obj; 4065 int error; 4066 4067 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 4068 return (error); 4069 vp = (struct vnode *)fp->f_data; 4070 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4071 if ((obj = vp->v_object) != NULL) { 4072 if (vp->v_mount == NULL || 4073 (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) { 4074 vm_object_page_clean(obj, 0, 0, 0); 4075 } 4076 } 4077 error = VOP_FSYNC_FP(vp, MNT_WAIT, VOP_FSYNC_SYSCALL, fp); 4078 if (error == 0 && vp->v_mount) 4079 error = buf_fsync(vp); 4080 vn_unlock(vp); 4081 fdrop(fp); 4082 4083 return (error); 4084 } 4085 4086 /* 4087 * rename op. 4088 * 4089 * NOTE: error == 0 and nl_dvp is NULL indicates a mount point, operation 4090 * disallowed. e.g. /var/cache where /var/cache is a null-mount, for 4091 * example. 4092 */ 4093 int 4094 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond) 4095 { 4096 struct nchandle fnchd; 4097 struct nchandle tnchd; 4098 struct namecache *ncp; 4099 struct vnode *fdvp; 4100 struct vnode *tdvp; 4101 struct mount *mp; 4102 struct mount *userenlk; 4103 int error; 4104 u_int fncp_gen; 4105 u_int tncp_gen; 4106 4107 bwillinode(1); 4108 fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC; 4109 if ((error = nlookup(fromnd)) != 0) 4110 return (error); 4111 4112 /* 4113 * Attempt to rename a mount point (from or to) 4114 */ 4115 if (error == 0 && fromnd->nl_dvp == NULL) 4116 return (EINVAL); 4117 4118 if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL) 4119 return (ENOENT); 4120 fnchd.mount = fromnd->nl_nch.mount; 4121 cache_hold(&fnchd); 4122 4123 /* 4124 * unlock the source nch so we can lookup the target nch without 4125 * deadlocking. The target may or may not exist so we do not check 4126 * for a target vp like kern_mkdir() and other creation functions do. 4127 * 4128 * The source and target directories are ref'd and rechecked after 4129 * everything is relocked to determine if the source or target file 4130 * has been renamed. 4131 */ 4132 KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED); 4133 fromnd->nl_flags &= ~NLC_NCPISLOCKED; 4134 fncp_gen = fromnd->nl_nch.ncp->nc_generation; 4135 4136 if (fromnd->nl_nch.ncp->nc_vp && 4137 fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 4138 userenlk = fnchd.mount; 4139 cache_unlock(&fromnd->nl_nch); 4140 lockmgr(&userenlk->mnt_renlock, LK_EXCLUSIVE); 4141 } else { 4142 userenlk = NULL; 4143 cache_unlock(&fromnd->nl_nch); 4144 } 4145 4146 /* 4147 * Lookup target 4148 */ 4149 tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP; 4150 if ((error = nlookup(tond)) != 0) { 4151 cache_drop(&fnchd); 4152 goto done; 4153 } 4154 tncp_gen = tond->nl_nch.ncp->nc_generation; 4155 4156 /* 4157 * Attempt to rename a mount point (from or to) 4158 */ 4159 if (error == 0 && tond->nl_dvp == NULL) { 4160 cache_drop(&fnchd); 4161 error = ENOENT; 4162 goto done; 4163 } 4164 4165 if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) { 4166 cache_drop(&fnchd); 4167 error = ENOENT; 4168 goto done; 4169 } 4170 tnchd.mount = tond->nl_nch.mount; 4171 cache_hold(&tnchd); 4172 4173 /* 4174 * If the source and target are the same there is nothing to do 4175 */ 4176 if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) { 4177 cache_drop(&fnchd); 4178 cache_drop(&tnchd); 4179 error = 0; 4180 goto done; 4181 } 4182 4183 /* 4184 * Mount points cannot be renamed or overwritten 4185 */ 4186 if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) & 4187 NCF_ISMOUNTPT 4188 ) { 4189 cache_drop(&fnchd); 4190 cache_drop(&tnchd); 4191 error = EINVAL; 4192 goto done; 4193 } 4194 4195 /* 4196 * Lock all four namecache entries. tond is already locked. 4197 */ 4198 cache_lock4_tondlocked(&fnchd, &fromnd->nl_nch, 4199 &tnchd, &tond->nl_nch, 4200 fromnd->nl_cred, tond->nl_cred); 4201 fromnd->nl_flags |= NLC_NCPISLOCKED; 4202 4203 /* 4204 * If the namecache generation changed for either fromnd or tond, 4205 * we must retry. 4206 */ 4207 if (fromnd->nl_nch.ncp->nc_generation != fncp_gen || 4208 tond->nl_nch.ncp->nc_generation != tncp_gen) { 4209 krateprintf(&krate_rename, 4210 "kern_rename: retry due to race on: " 4211 "\"%s\" -> \"%s\"\n", 4212 fromnd->nl_nch.ncp->nc_name, 4213 tond->nl_nch.ncp->nc_name); 4214 error = EAGAIN; 4215 goto finish; 4216 } 4217 4218 /* 4219 * If either fromnd or tond are marked destroyed a ripout occured 4220 * out from under us and we must retry. 4221 */ 4222 if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) || 4223 fromnd->nl_nch.ncp->nc_vp == NULL || 4224 (tond->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED))) { 4225 krateprintf(&krate_rename, 4226 "kern_rename: retry due to ripout on: " 4227 "\"%s\" -> \"%s\"\n", 4228 fromnd->nl_nch.ncp->nc_name, 4229 tond->nl_nch.ncp->nc_name); 4230 error = EAGAIN; 4231 goto finish; 4232 } 4233 4234 /* 4235 * Make sure the parent directories linkages are the same. We have 4236 * already checked that fromnd and tond are not mount points so this 4237 * should not loop forever on a cross-mount. 4238 */ 4239 if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent || 4240 tnchd.ncp != tond->nl_nch.ncp->nc_parent) { 4241 error = EAGAIN; 4242 goto finish; 4243 } 4244 4245 /* 4246 * Both the source and target must be within the same filesystem and 4247 * in the same filesystem as their parent directories within the 4248 * namecache topology. 4249 * 4250 * NOTE: fromnd's nc_mount or nc_vp could be NULL. 4251 */ 4252 mp = fnchd.mount; 4253 if (mp != tnchd.mount || mp != fromnd->nl_nch.mount || 4254 mp != tond->nl_nch.mount) { 4255 error = EXDEV; 4256 goto finish; 4257 } 4258 4259 /* 4260 * Make sure the mount point is writable 4261 */ 4262 if ((error = ncp_writechk(&tond->nl_nch)) != 0) { 4263 goto finish; 4264 } 4265 4266 /* 4267 * If the target exists and either the source or target is a directory, 4268 * then both must be directories. 4269 * 4270 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h 4271 * have become NULL. 4272 */ 4273 if (tond->nl_nch.ncp->nc_vp) { 4274 if (fromnd->nl_nch.ncp->nc_vp == NULL) { 4275 error = ENOENT; 4276 } else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) { 4277 if (tond->nl_nch.ncp->nc_vp->v_type != VDIR) 4278 error = ENOTDIR; 4279 } else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) { 4280 error = EISDIR; 4281 } 4282 } 4283 4284 /* 4285 * You cannot rename a source into itself or a subdirectory of itself. 4286 * We check this by travsersing the target directory upwards looking 4287 * for a match against the source. 4288 * 4289 * Only required when renaming a directory, in which case userenlk is 4290 * non-NULL. 4291 */ 4292 if (__predict_false(userenlk && error == 0)) { 4293 for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) { 4294 if (fromnd->nl_nch.ncp == ncp) { 4295 error = EINVAL; 4296 break; 4297 } 4298 } 4299 } 4300 4301 /* 4302 * Even though the namespaces are different, they may still represent 4303 * hardlinks to the same file. The filesystem might have a hard time 4304 * with this so we issue a NREMOVE of the source instead of a NRENAME 4305 * when we detect the situation. 4306 */ 4307 if (error == 0) { 4308 fdvp = fromnd->nl_dvp; 4309 tdvp = tond->nl_dvp; 4310 if (fdvp == NULL || tdvp == NULL) { 4311 error = EPERM; 4312 } else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) { 4313 error = VOP_NREMOVE(&fromnd->nl_nch, fdvp, 4314 fromnd->nl_cred); 4315 } else { 4316 error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch, 4317 fdvp, tdvp, tond->nl_cred); 4318 } 4319 } 4320 finish: 4321 cache_put(&tnchd); 4322 cache_put(&fnchd); 4323 done: 4324 if (userenlk) 4325 lockmgr(&userenlk->mnt_renlock, LK_RELEASE); 4326 return (error); 4327 } 4328 4329 /* 4330 * rename_args(char *from, char *to) 4331 * 4332 * Rename files. 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_rename(struct sysmsg *sysmsg, const struct rename_args *uap) 4337 { 4338 struct nlookupdata fromnd, tond; 4339 int error; 4340 4341 do { 4342 error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0); 4343 if (error == 0) { 4344 error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0); 4345 if (error == 0) 4346 error = kern_rename(&fromnd, &tond); 4347 nlookup_done(&tond); 4348 } 4349 nlookup_done(&fromnd); 4350 } while (error == EAGAIN); 4351 return (error); 4352 } 4353 4354 /* 4355 * renameat_args(int oldfd, char *old, int newfd, char *new) 4356 * 4357 * Rename files using paths relative to the directories associated with 4358 * oldfd and newfd. Source and destination must either both be directories, 4359 * or both not be directories. If target is a directory, it must be empty. 4360 */ 4361 int 4362 sys_renameat(struct sysmsg *sysmsg, const struct renameat_args *uap) 4363 { 4364 struct nlookupdata oldnd, newnd; 4365 struct file *oldfp, *newfp; 4366 int error; 4367 4368 do { 4369 error = nlookup_init_at(&oldnd, &oldfp, 4370 uap->oldfd, uap->old, 4371 UIO_USERSPACE, 0); 4372 if (error == 0) { 4373 error = nlookup_init_at(&newnd, &newfp, 4374 uap->newfd, uap->new, 4375 UIO_USERSPACE, 0); 4376 if (error == 0) 4377 error = kern_rename(&oldnd, &newnd); 4378 nlookup_done_at(&newnd, newfp); 4379 } 4380 nlookup_done_at(&oldnd, oldfp); 4381 } while (error == EAGAIN); 4382 return (error); 4383 } 4384 4385 int 4386 kern_mkdir(struct nlookupdata *nd, int mode) 4387 { 4388 struct thread *td = curthread; 4389 struct proc *p = td->td_proc; 4390 struct vnode *vp; 4391 struct vattr vattr; 4392 int error; 4393 4394 bwillinode(1); 4395 nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP; 4396 if ((error = nlookup(nd)) != 0) 4397 return (error); 4398 4399 if (nd->nl_nch.ncp->nc_vp) 4400 return (EEXIST); 4401 if (nd->nl_dvp == NULL) 4402 return (EINVAL); 4403 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 4404 return (error); 4405 VATTR_NULL(&vattr); 4406 vattr.va_type = VDIR; 4407 vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask; 4408 4409 vp = NULL; 4410 error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr); 4411 if (error == 0) 4412 vput(vp); 4413 return (error); 4414 } 4415 4416 /* 4417 * mkdir_args(char *path, int mode) 4418 * 4419 * Make a directory file. 4420 */ 4421 int 4422 sys_mkdir(struct sysmsg *sysmsg, const struct mkdir_args *uap) 4423 { 4424 struct nlookupdata nd; 4425 int error; 4426 4427 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 4428 if (error == 0) 4429 error = kern_mkdir(&nd, uap->mode); 4430 nlookup_done(&nd); 4431 return (error); 4432 } 4433 4434 /* 4435 * mkdirat_args(int fd, char *path, mode_t mode) 4436 * 4437 * Make a directory file. The path is relative to the directory associated 4438 * with fd. 4439 */ 4440 int 4441 sys_mkdirat(struct sysmsg *sysmsg, const struct mkdirat_args *uap) 4442 { 4443 struct nlookupdata nd; 4444 struct file *fp; 4445 int error; 4446 4447 error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0); 4448 if (error == 0) 4449 error = kern_mkdir(&nd, uap->mode); 4450 nlookup_done_at(&nd, fp); 4451 return (error); 4452 } 4453 4454 int 4455 kern_rmdir(struct nlookupdata *nd) 4456 { 4457 int error; 4458 4459 bwillinode(1); 4460 nd->nl_flags |= NLC_DELETE | NLC_REFDVP; 4461 if ((error = nlookup(nd)) != 0) 4462 return (error); 4463 4464 /* 4465 * Do not allow directories representing mount points to be 4466 * deleted, even if empty. Check write perms on mount point 4467 * in case the vnode is aliased (aka nullfs). 4468 */ 4469 if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT)) 4470 return (EBUSY); 4471 if (nd->nl_dvp == NULL) 4472 return (EINVAL); 4473 if ((error = ncp_writechk(&nd->nl_nch)) != 0) 4474 return (error); 4475 error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred); 4476 return (error); 4477 } 4478 4479 /* 4480 * rmdir_args(char *path) 4481 * 4482 * Remove a directory file. 4483 */ 4484 int 4485 sys_rmdir(struct sysmsg *sysmsg, const struct rmdir_args *uap) 4486 { 4487 struct nlookupdata nd; 4488 int error; 4489 4490 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0); 4491 if (error == 0) 4492 error = kern_rmdir(&nd); 4493 nlookup_done(&nd); 4494 return (error); 4495 } 4496 4497 int 4498 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res, 4499 enum uio_seg direction) 4500 { 4501 struct thread *td = curthread; 4502 struct vnode *vp; 4503 struct file *fp; 4504 struct uio auio; 4505 struct iovec aiov; 4506 off_t loff; 4507 int error, eofflag; 4508 4509 if ((error = holdvnode(td, fd, &fp)) != 0) 4510 return (error); 4511 if ((fp->f_flag & FREAD) == 0) { 4512 error = EBADF; 4513 goto done; 4514 } 4515 vp = (struct vnode *)fp->f_data; 4516 if (vp->v_type != VDIR) { 4517 error = EINVAL; 4518 goto done; 4519 } 4520 aiov.iov_base = buf; 4521 aiov.iov_len = count; 4522 auio.uio_iov = &aiov; 4523 auio.uio_iovcnt = 1; 4524 auio.uio_rw = UIO_READ; 4525 auio.uio_segflg = direction; 4526 auio.uio_td = td; 4527 auio.uio_resid = count; 4528 loff = auio.uio_offset = fp->f_offset; 4529 error = VOP_READDIR_FP(vp, &auio, fp->f_cred, &eofflag, NULL, NULL, fp); 4530 fp->f_offset = auio.uio_offset; 4531 if (error) 4532 goto done; 4533 4534 /* 4535 * WARNING! *basep may not be wide enough to accomodate the 4536 * seek offset. XXX should we hack this to return the upper 32 bits 4537 * for offsets greater then 4G? 4538 */ 4539 if (basep) { 4540 *basep = (long)loff; 4541 } 4542 *res = count - auio.uio_resid; 4543 done: 4544 fdrop(fp); 4545 return (error); 4546 } 4547 4548 /* 4549 * getdirentries_args(int fd, char *buf, u_int conut, long *basep) 4550 * 4551 * Read a block of directory entries in a file system independent format. 4552 */ 4553 int 4554 sys_getdirentries(struct sysmsg *sysmsg, const struct getdirentries_args *uap) 4555 { 4556 long base; 4557 int error; 4558 4559 error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base, 4560 &sysmsg->sysmsg_result, UIO_USERSPACE); 4561 4562 if (error == 0 && uap->basep) 4563 error = copyout(&base, uap->basep, sizeof(*uap->basep)); 4564 return (error); 4565 } 4566 4567 /* 4568 * getdents_args(int fd, char *buf, size_t count) 4569 */ 4570 int 4571 sys_getdents(struct sysmsg *sysmsg, const struct getdents_args *uap) 4572 { 4573 int error; 4574 4575 error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL, 4576 &sysmsg->sysmsg_result, UIO_USERSPACE); 4577 4578 return (error); 4579 } 4580 4581 /* 4582 * Set the mode mask for creation of filesystem nodes. 4583 * 4584 * umask(int newmask) 4585 */ 4586 int 4587 sys_umask(struct sysmsg *sysmsg, const struct umask_args *uap) 4588 { 4589 struct thread *td = curthread; 4590 struct proc *p = td->td_proc; 4591 struct filedesc *fdp; 4592 4593 fdp = p->p_fd; 4594 sysmsg->sysmsg_result = fdp->fd_cmask; 4595 fdp->fd_cmask = uap->newmask & ALLPERMS; 4596 return (0); 4597 } 4598 4599 /* 4600 * revoke(char *path) 4601 * 4602 * Void all references to file by ripping underlying filesystem 4603 * away from vnode. 4604 */ 4605 int 4606 sys_revoke(struct sysmsg *sysmsg, const struct revoke_args *uap) 4607 { 4608 struct nlookupdata nd; 4609 struct vattr vattr; 4610 struct vnode *vp; 4611 struct ucred *cred; 4612 int error; 4613 4614 vp = NULL; 4615 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 4616 if (error == 0) 4617 error = nlookup(&nd); 4618 if (error == 0) 4619 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 4620 cred = crhold(nd.nl_cred); 4621 nlookup_done(&nd); 4622 if (error == 0) { 4623 if (error == 0) 4624 error = VOP_GETATTR(vp, &vattr); 4625 if (error == 0 && cred->cr_uid != vattr.va_uid) 4626 error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0); 4627 if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) { 4628 if (vcount(vp) > 0) 4629 error = vrevoke(vp, cred); 4630 } else if (error == 0) { 4631 error = vrevoke(vp, cred); 4632 } 4633 vrele(vp); 4634 } 4635 if (cred) 4636 crfree(cred); 4637 return (error); 4638 } 4639 4640 /* 4641 * getfh_args(char *fname, fhandle_t *fhp) 4642 * 4643 * Get (NFS) file handle 4644 * 4645 * NOTE: We use the fsid of the covering mount, even if it is a nullfs 4646 * mount. This allows nullfs mounts to be explicitly exported. 4647 * 4648 * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe. 4649 * 4650 * nullfs mounts of subdirectories are not safe. That is, it will 4651 * work, but you do not really have protection against access to 4652 * the related parent directories. 4653 */ 4654 int 4655 sys_getfh(struct sysmsg *sysmsg, const struct getfh_args *uap) 4656 { 4657 struct thread *td = curthread; 4658 struct nlookupdata nd; 4659 fhandle_t fh; 4660 struct vnode *vp; 4661 struct mount *mp; 4662 int error; 4663 4664 /* 4665 * Must be super user 4666 */ 4667 if ((error = priv_check(td, PRIV_ROOT)) != 0) 4668 return (error); 4669 4670 vp = NULL; 4671 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 4672 if (error == 0) 4673 error = nlookup(&nd); 4674 if (error == 0) 4675 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 4676 mp = nd.nl_nch.mount; 4677 nlookup_done(&nd); 4678 if (error == 0) { 4679 bzero(&fh, sizeof(fh)); 4680 fh.fh_fsid = mp->mnt_stat.f_fsid; 4681 error = VFS_VPTOFH(vp, &fh.fh_fid); 4682 vput(vp); 4683 if (error == 0) 4684 error = copyout(&fh, uap->fhp, sizeof(fh)); 4685 } 4686 return (error); 4687 } 4688 4689 /* 4690 * fhopen_args(const struct fhandle *u_fhp, int flags) 4691 * 4692 * syscall for the rpc.lockd to use to translate a NFS file handle into 4693 * an open descriptor. 4694 * 4695 * warning: do not remove the priv_check() call or this becomes one giant 4696 * security hole. 4697 */ 4698 int 4699 sys_fhopen(struct sysmsg *sysmsg, const struct fhopen_args *uap) 4700 { 4701 struct thread *td = curthread; 4702 struct filedesc *fdp = td->td_proc->p_fd; 4703 struct mount *mp; 4704 struct vnode *vp; 4705 struct fhandle fhp; 4706 struct vattr vat; 4707 struct vattr *vap = &vat; 4708 struct flock lf; 4709 int fmode, mode, error = 0, type; 4710 struct file *nfp; 4711 struct file *fp; 4712 int indx; 4713 4714 /* 4715 * Must be super user 4716 */ 4717 error = priv_check(td, PRIV_ROOT); 4718 if (error) 4719 return (error); 4720 4721 fmode = FFLAGS(uap->flags); 4722 4723 /* 4724 * Why not allow a non-read/write open for our lockd? 4725 */ 4726 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) 4727 return (EINVAL); 4728 error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); 4729 if (error) 4730 return(error); 4731 4732 /* 4733 * Find the mount point 4734 */ 4735 mp = vfs_getvfs(&fhp.fh_fsid); 4736 if (mp == NULL) { 4737 error = ESTALE; 4738 goto done2; 4739 } 4740 /* now give me my vnode, it gets returned to me locked */ 4741 error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp); 4742 if (error) 4743 goto done; 4744 /* 4745 * from now on we have to make sure not 4746 * to forget about the vnode 4747 * any error that causes an abort must vput(vp) 4748 * just set error = err and 'goto bad;'. 4749 */ 4750 4751 /* 4752 * from vn_open 4753 */ 4754 if (vp->v_type == VLNK) { 4755 error = EMLINK; 4756 goto bad; 4757 } 4758 if (vp->v_type == VSOCK) { 4759 error = EOPNOTSUPP; 4760 goto bad; 4761 } 4762 mode = 0; 4763 if (fmode & (FWRITE | O_TRUNC)) { 4764 if (vp->v_type == VDIR) { 4765 error = EISDIR; 4766 goto bad; 4767 } 4768 error = vn_writechk(vp); 4769 if (error) 4770 goto bad; 4771 mode |= VWRITE; 4772 } 4773 if (fmode & FREAD) 4774 mode |= VREAD; 4775 if (mode) { 4776 error = VOP_ACCESS(vp, mode, td->td_ucred); 4777 if (error) 4778 goto bad; 4779 } 4780 if (fmode & O_TRUNC) { 4781 vn_unlock(vp); /* XXX */ 4782 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* XXX */ 4783 VATTR_NULL(vap); 4784 vap->va_size = 0; 4785 error = VOP_SETATTR(vp, vap, td->td_ucred); 4786 if (error) 4787 goto bad; 4788 } 4789 4790 /* 4791 * VOP_OPEN needs the file pointer so it can potentially override 4792 * it. 4793 * 4794 * WARNING! no f_nchandle will be associated when fhopen()ing a 4795 * directory. XXX 4796 */ 4797 if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0) 4798 goto bad; 4799 error = VOP_OPEN(vp, fmode, td->td_ucred, &nfp); 4800 fp = nfp; 4801 4802 if (error) { 4803 /* 4804 * setting f_ops this way prevents VOP_CLOSE from being 4805 * called or fdrop() releasing the vp from v_data. Since 4806 * the VOP_OPEN failed we don't want to VOP_CLOSE. 4807 */ 4808 fp->f_ops = &badfileops; 4809 fp->f_data = NULL; 4810 goto bad_drop; 4811 } 4812 4813 /* 4814 * The fp is given its own reference, we still have our ref and lock. 4815 * 4816 * Assert that all regular files must be created with a VM object. 4817 */ 4818 if (vp->v_type == VREG && vp->v_object == NULL) { 4819 kprintf("fhopen: regular file did not " 4820 "have VM object: %p\n", 4821 vp); 4822 goto bad_drop; 4823 } 4824 4825 /* 4826 * The open was successful. Handle any locking requirements. 4827 */ 4828 if (fmode & (O_EXLOCK | O_SHLOCK)) { 4829 lf.l_whence = SEEK_SET; 4830 lf.l_start = 0; 4831 lf.l_len = 0; 4832 if (fmode & O_EXLOCK) 4833 lf.l_type = F_WRLCK; 4834 else 4835 lf.l_type = F_RDLCK; 4836 if (fmode & FNONBLOCK) 4837 type = 0; 4838 else 4839 type = F_WAIT; 4840 vn_unlock(vp); 4841 if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, 4842 &lf, type)) != 0) { 4843 /* 4844 * release our private reference. 4845 */ 4846 fsetfd(fdp, NULL, indx); 4847 fdrop(fp); 4848 vrele(vp); 4849 goto done; 4850 } 4851 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4852 atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */ 4853 } 4854 4855 /* 4856 * Clean up. Associate the file pointer with the previously 4857 * reserved descriptor and return it. 4858 */ 4859 vput(vp); 4860 if (uap->flags & O_CLOEXEC) 4861 fdp->fd_files[indx].fileflags |= UF_EXCLOSE; 4862 fsetfd(fdp, fp, indx); 4863 fdrop(fp); 4864 sysmsg->sysmsg_result = indx; 4865 mount_drop(mp); 4866 4867 return (error); 4868 4869 bad_drop: 4870 fsetfd(fdp, NULL, indx); 4871 fdrop(fp); 4872 bad: 4873 vput(vp); 4874 done: 4875 mount_drop(mp); 4876 done2: 4877 return (error); 4878 } 4879 4880 /* 4881 * fhstat_args(struct fhandle *u_fhp, struct stat *sb) 4882 */ 4883 int 4884 sys_fhstat(struct sysmsg *sysmsg, const struct fhstat_args *uap) 4885 { 4886 struct thread *td = curthread; 4887 struct stat sb; 4888 fhandle_t fh; 4889 struct mount *mp; 4890 struct vnode *vp; 4891 int error; 4892 4893 /* 4894 * Must be super user 4895 */ 4896 error = priv_check(td, PRIV_ROOT); 4897 if (error) 4898 return (error); 4899 4900 error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); 4901 if (error) 4902 return (error); 4903 4904 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) 4905 error = ESTALE; 4906 if (error == 0) { 4907 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) { 4908 error = vn_stat(vp, &sb, td->td_ucred); 4909 vput(vp); 4910 } 4911 } 4912 if (error == 0) 4913 error = copyout(&sb, uap->sb, sizeof(sb)); 4914 if (mp) 4915 mount_drop(mp); 4916 4917 return (error); 4918 } 4919 4920 /* 4921 * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf) 4922 */ 4923 int 4924 sys_fhstatfs(struct sysmsg *sysmsg, const struct fhstatfs_args *uap) 4925 { 4926 struct thread *td = curthread; 4927 struct proc *p = td->td_proc; 4928 struct statfs *sp; 4929 struct mount *mp; 4930 struct vnode *vp; 4931 struct statfs sb; 4932 char *fullpath, *freepath; 4933 fhandle_t fh; 4934 int error; 4935 4936 /* 4937 * Must be super user 4938 */ 4939 if ((error = priv_check(td, PRIV_ROOT))) 4940 return (error); 4941 4942 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 4943 return (error); 4944 4945 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 4946 error = ESTALE; 4947 goto done; 4948 } 4949 if (p != NULL && !chroot_visible_mnt(mp, p)) { 4950 error = ESTALE; 4951 goto done; 4952 } 4953 4954 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0) 4955 goto done; 4956 mp = vp->v_mount; 4957 sp = &mp->mnt_stat; 4958 vput(vp); 4959 if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0) 4960 goto done; 4961 4962 error = mount_path(p, mp, &fullpath, &freepath); 4963 if (error) 4964 goto done; 4965 bzero(sp->f_mntonname, sizeof(sp->f_mntonname)); 4966 strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname)); 4967 kfree(freepath, M_TEMP); 4968 4969 sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK; 4970 if (priv_check(td, PRIV_ROOT)) { 4971 bcopy(sp, &sb, sizeof(sb)); 4972 sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0; 4973 sp = &sb; 4974 } 4975 error = copyout(sp, uap->buf, sizeof(*sp)); 4976 done: 4977 if (mp) 4978 mount_drop(mp); 4979 4980 return (error); 4981 } 4982 4983 /* 4984 * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf) 4985 */ 4986 int 4987 sys_fhstatvfs(struct sysmsg *sysmsg, const struct fhstatvfs_args *uap) 4988 { 4989 struct thread *td = curthread; 4990 struct proc *p = td->td_proc; 4991 struct statvfs *sp; 4992 struct mount *mp; 4993 struct vnode *vp; 4994 fhandle_t fh; 4995 int error; 4996 4997 /* 4998 * Must be super user 4999 */ 5000 if ((error = priv_check(td, PRIV_ROOT))) 5001 return (error); 5002 5003 if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0) 5004 return (error); 5005 5006 if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) { 5007 error = ESTALE; 5008 goto done; 5009 } 5010 if (p != NULL && !chroot_visible_mnt(mp, p)) { 5011 error = ESTALE; 5012 goto done; 5013 } 5014 5015 if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp))) 5016 goto done; 5017 mp = vp->v_mount; 5018 sp = &mp->mnt_vstat; 5019 vput(vp); 5020 if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0) 5021 goto done; 5022 5023 sp->f_flag = 0; 5024 if (mp->mnt_flag & MNT_RDONLY) 5025 sp->f_flag |= ST_RDONLY; 5026 if (mp->mnt_flag & MNT_NOSUID) 5027 sp->f_flag |= ST_NOSUID; 5028 error = copyout(sp, uap->buf, sizeof(*sp)); 5029 done: 5030 if (mp) 5031 mount_drop(mp); 5032 return (error); 5033 } 5034 5035 5036 /* 5037 * Syscall to push extended attribute configuration information into the 5038 * VFS. Accepts a path, which it converts to a mountpoint, as well as 5039 * a command (int cmd), and attribute name and misc data. For now, the 5040 * attribute name is left in userspace for consumption by the VFS_op. 5041 * It will probably be changed to be copied into sysspace by the 5042 * syscall in the future, once issues with various consumers of the 5043 * attribute code have raised their hands. 5044 * 5045 * Currently this is used only by UFS Extended Attributes. 5046 */ 5047 int 5048 sys_extattrctl(struct sysmsg *sysmsg, const struct extattrctl_args *uap) 5049 { 5050 struct nlookupdata nd; 5051 struct vnode *vp; 5052 char attrname[EXTATTR_MAXNAMELEN]; 5053 int error; 5054 size_t size; 5055 5056 attrname[0] = 0; 5057 vp = NULL; 5058 error = 0; 5059 5060 if (error == 0 && uap->filename) { 5061 error = nlookup_init(&nd, uap->filename, UIO_USERSPACE, 5062 NLC_FOLLOW); 5063 if (error == 0) 5064 error = nlookup(&nd); 5065 if (error == 0) 5066 error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp); 5067 nlookup_done(&nd); 5068 } 5069 5070 if (error == 0 && uap->attrname) { 5071 error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN, 5072 &size); 5073 } 5074 5075 if (error == 0) { 5076 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5077 if (error == 0) 5078 error = nlookup(&nd); 5079 if (error == 0) 5080 error = ncp_writechk(&nd.nl_nch); 5081 if (error == 0) { 5082 error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp, 5083 uap->attrnamespace, 5084 uap->attrname, nd.nl_cred); 5085 } 5086 nlookup_done(&nd); 5087 } 5088 5089 return (error); 5090 } 5091 5092 /* 5093 * Syscall to get a named extended attribute on a file or directory. 5094 */ 5095 int 5096 sys_extattr_set_file(struct sysmsg *sysmsg, 5097 const struct extattr_set_file_args *uap) 5098 { 5099 char attrname[EXTATTR_MAXNAMELEN]; 5100 struct nlookupdata nd; 5101 struct vnode *vp; 5102 struct uio auio; 5103 struct iovec aiov; 5104 int error; 5105 5106 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 5107 if (error) 5108 return (error); 5109 5110 vp = NULL; 5111 5112 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5113 if (error == 0) 5114 error = nlookup(&nd); 5115 if (error == 0) 5116 error = ncp_writechk(&nd.nl_nch); 5117 if (error == 0) 5118 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 5119 if (error) { 5120 nlookup_done(&nd); 5121 return (error); 5122 } 5123 5124 bzero(&auio, sizeof(auio)); 5125 aiov.iov_base = uap->data; 5126 aiov.iov_len = uap->nbytes; 5127 auio.uio_iov = &aiov; 5128 auio.uio_iovcnt = 1; 5129 auio.uio_offset = 0; 5130 auio.uio_resid = uap->nbytes; 5131 auio.uio_rw = UIO_WRITE; 5132 auio.uio_td = curthread; 5133 5134 error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname, 5135 &auio, nd.nl_cred); 5136 5137 vput(vp); 5138 nlookup_done(&nd); 5139 return (error); 5140 } 5141 5142 /* 5143 * Syscall to get a named extended attribute on a file or directory. 5144 */ 5145 int 5146 sys_extattr_get_file(struct sysmsg *sysmsg, 5147 const struct extattr_get_file_args *uap) 5148 { 5149 char attrname[EXTATTR_MAXNAMELEN]; 5150 struct nlookupdata nd; 5151 struct uio auio; 5152 struct iovec aiov; 5153 struct vnode *vp; 5154 int error; 5155 5156 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 5157 if (error) 5158 return (error); 5159 5160 vp = NULL; 5161 5162 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5163 if (error == 0) 5164 error = nlookup(&nd); 5165 if (error == 0) 5166 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp); 5167 if (error) { 5168 nlookup_done(&nd); 5169 return (error); 5170 } 5171 5172 bzero(&auio, sizeof(auio)); 5173 aiov.iov_base = uap->data; 5174 aiov.iov_len = uap->nbytes; 5175 auio.uio_iov = &aiov; 5176 auio.uio_iovcnt = 1; 5177 auio.uio_offset = 0; 5178 auio.uio_resid = uap->nbytes; 5179 auio.uio_rw = UIO_READ; 5180 auio.uio_td = curthread; 5181 5182 error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname, 5183 &auio, nd.nl_cred); 5184 sysmsg->sysmsg_result = uap->nbytes - auio.uio_resid; 5185 5186 vput(vp); 5187 nlookup_done(&nd); 5188 return(error); 5189 } 5190 5191 /* 5192 * Syscall to delete a named extended attribute from a file or directory. 5193 * Accepts attribute name. The real work happens in VOP_SETEXTATTR(). 5194 */ 5195 int 5196 sys_extattr_delete_file(struct sysmsg *sysmsg, 5197 const struct extattr_delete_file_args *uap) 5198 { 5199 char attrname[EXTATTR_MAXNAMELEN]; 5200 struct nlookupdata nd; 5201 struct vnode *vp; 5202 int error; 5203 5204 error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN); 5205 if (error) 5206 return(error); 5207 5208 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5209 if (error == 0) 5210 error = nlookup(&nd); 5211 if (error == 0) 5212 error = ncp_writechk(&nd.nl_nch); 5213 if (error == 0) { 5214 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp); 5215 if (error == 0) { 5216 error = VOP_SETEXTATTR(vp, uap->attrnamespace, 5217 attrname, NULL, nd.nl_cred); 5218 vput(vp); 5219 } 5220 } 5221 nlookup_done(&nd); 5222 return(error); 5223 } 5224 5225 /* 5226 * Determine if the mount is visible to the process. 5227 */ 5228 static int 5229 chroot_visible_mnt(struct mount *mp, struct proc *p) 5230 { 5231 struct nchandle nch; 5232 5233 /* 5234 * Traverse from the mount point upwards. If we hit the process 5235 * root then the mount point is visible to the process. 5236 */ 5237 nch = mp->mnt_ncmountpt; 5238 while (nch.ncp) { 5239 if (nch.mount == p->p_fd->fd_nrdir.mount && 5240 nch.ncp == p->p_fd->fd_nrdir.ncp) { 5241 return(1); 5242 } 5243 if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) { 5244 nch = nch.mount->mnt_ncmounton; 5245 } else { 5246 nch.ncp = nch.ncp->nc_parent; 5247 } 5248 } 5249 5250 /* 5251 * If the mount point is not visible to the process, but the 5252 * process root is in a subdirectory of the mount, return 5253 * TRUE anyway. 5254 */ 5255 if (p->p_fd->fd_nrdir.mount == mp) 5256 return(1); 5257 5258 return(0); 5259 } 5260 5261 /* Sets priv to PRIV_ROOT in case no matching fs */ 5262 static int 5263 get_fspriv(const char *fsname) 5264 { 5265 5266 if (strncmp("null", fsname, 5) == 0) { 5267 return PRIV_VFS_MOUNT_NULLFS; 5268 } else if (strncmp(fsname, "tmpfs", 6) == 0) { 5269 return PRIV_VFS_MOUNT_TMPFS; 5270 } 5271 5272 return PRIV_ROOT; 5273 } 5274 5275 int 5276 sys___realpath(struct sysmsg *sysmsg, const struct __realpath_args *uap) 5277 { 5278 struct nlookupdata nd; 5279 char *rbuf; 5280 char *fbuf; 5281 ssize_t rlen; 5282 int error; 5283 5284 /* 5285 * Invalid length if less than 0. 0 is allowed 5286 */ 5287 if ((ssize_t)uap->len < 0) 5288 return EINVAL; 5289 5290 rbuf = NULL; 5291 fbuf = NULL; 5292 error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW); 5293 if (error) 5294 goto done; 5295 5296 nd.nl_flags |= NLC_SHAREDLOCK; 5297 error = nlookup(&nd); 5298 if (error) 5299 goto done; 5300 5301 if (nd.nl_nch.ncp->nc_vp == NULL) { 5302 error = ENOENT; 5303 goto done; 5304 } 5305 5306 /* 5307 * Shortcut test for existence. 5308 */ 5309 if (uap->len == 0) { 5310 error = ENAMETOOLONG; 5311 goto done; 5312 } 5313 5314 /* 5315 * Obtain the path relative to the process root. The nch must not 5316 * be locked for the cache_fullpath() call. 5317 */ 5318 if (nd.nl_flags & NLC_NCPISLOCKED) { 5319 nd.nl_flags &= ~NLC_NCPISLOCKED; 5320 cache_unlock(&nd.nl_nch); 5321 } 5322 error = cache_fullpath(curproc, &nd.nl_nch, NULL, &rbuf, &fbuf, 0); 5323 if (error) 5324 goto done; 5325 5326 rlen = (ssize_t)strlen(rbuf); 5327 if (rlen >= uap->len) { 5328 error = ENAMETOOLONG; 5329 goto done; 5330 } 5331 error = copyout(rbuf, uap->buf, rlen + 1); 5332 if (error == 0) 5333 sysmsg->sysmsg_szresult = rlen; 5334 done: 5335 nlookup_done(&nd); 5336 if (fbuf) 5337 kfree(fbuf, M_TEMP); 5338 5339 return error; 5340 } 5341 5342 int 5343 sys_posix_fallocate(struct sysmsg *sysmsg, const struct posix_fallocate_args *uap) 5344 { 5345 return (kern_posix_fallocate(uap->fd, uap->offset, uap->len)); 5346 } 5347 5348 int 5349 kern_posix_fallocate(int fd, off_t offset, off_t len) 5350 { 5351 struct thread *td = curthread; 5352 struct vnode *vp; 5353 struct file *fp; 5354 int error; 5355 5356 if (offset < 0 || len <= 0) 5357 return (EINVAL); 5358 /* Check for wrap. */ 5359 if (offset > OFF_MAX - len) 5360 return (EFBIG); 5361 5362 fp = holdfp(td, fd, -1); 5363 if (fp == NULL) 5364 return (EBADF); 5365 5366 switch (fp->f_type) { 5367 case DTYPE_VNODE: 5368 break; 5369 case DTYPE_PIPE: 5370 case DTYPE_FIFO: 5371 error = ESPIPE; 5372 goto out; 5373 default: 5374 error = ENODEV; 5375 goto out; 5376 } 5377 5378 if ((fp->f_flag & FWRITE) == 0) { 5379 error = EBADF; 5380 goto out; 5381 } 5382 5383 vp = fp->f_data; 5384 if (vp->v_type != VREG) { 5385 error = ENODEV; 5386 goto out; 5387 } 5388 5389 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 5390 error = VOP_ALLOCATE(vp, offset, len); 5391 vn_unlock(vp); 5392 out: 5393 dropfp(td, fd, fp); 5394 return (error); 5395 } 5396