1 /* $NetBSD: vfs_mount.c,v 1.84 2020/10/13 13:15:39 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 1997-2020 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1989, 1993 35 * The Regents of the University of California. All rights reserved. 36 * (c) UNIX System Laboratories, Inc. 37 * All or some portions of this file are derived from material licensed 38 * to the University of California by American Telephone and Telegraph 39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 * the permission of UNIX System Laboratories, Inc. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: vfs_mount.c,v 1.84 2020/10/13 13:15:39 hannken Exp $"); 71 72 #include <sys/param.h> 73 #include <sys/kernel.h> 74 75 #include <sys/atomic.h> 76 #include <sys/buf.h> 77 #include <sys/conf.h> 78 #include <sys/fcntl.h> 79 #include <sys/filedesc.h> 80 #include <sys/device.h> 81 #include <sys/kauth.h> 82 #include <sys/kmem.h> 83 #include <sys/module.h> 84 #include <sys/mount.h> 85 #include <sys/fstrans.h> 86 #include <sys/namei.h> 87 #include <sys/extattr.h> 88 #include <sys/syscallargs.h> 89 #include <sys/sysctl.h> 90 #include <sys/systm.h> 91 #include <sys/vfs_syscalls.h> 92 #include <sys/vnode_impl.h> 93 94 #include <miscfs/genfs/genfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 enum mountlist_type { 98 ME_MOUNT, 99 ME_MARKER 100 }; 101 struct mountlist_entry { 102 TAILQ_ENTRY(mountlist_entry) me_list; /* Mount list. */ 103 struct mount *me_mount; /* Actual mount if ME_MOUNT, 104 current mount else. */ 105 enum mountlist_type me_type; /* Mount or marker. */ 106 }; 107 struct mount_iterator { 108 struct mountlist_entry mi_entry; 109 }; 110 111 static struct vnode *vfs_vnode_iterator_next1(struct vnode_iterator *, 112 bool (*)(void *, struct vnode *), void *, bool); 113 114 /* Root filesystem. */ 115 vnode_t * rootvnode; 116 117 /* Mounted filesystem list. */ 118 static TAILQ_HEAD(mountlist, mountlist_entry) mountlist; 119 static kmutex_t mountlist_lock __cacheline_aligned; 120 int vnode_offset_next_by_lru /* XXX: ugly hack for pstat.c */ 121 = offsetof(vnode_impl_t, vi_lrulist.tqe_next); 122 123 kmutex_t vfs_list_lock __cacheline_aligned; 124 125 static specificdata_domain_t mount_specificdata_domain; 126 static kmutex_t mntid_lock; 127 128 static kmutex_t mountgen_lock __cacheline_aligned; 129 static uint64_t mountgen; 130 131 void 132 vfs_mount_sysinit(void) 133 { 134 135 TAILQ_INIT(&mountlist); 136 mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE); 137 mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE); 138 139 mount_specificdata_domain = specificdata_domain_create(); 140 mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE); 141 mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE); 142 mountgen = 0; 143 } 144 145 struct mount * 146 vfs_mountalloc(struct vfsops *vfsops, vnode_t *vp) 147 { 148 struct mount *mp; 149 int error __diagused; 150 151 mp = kmem_zalloc(sizeof(*mp), KM_SLEEP); 152 mp->mnt_op = vfsops; 153 mp->mnt_refcnt = 1; 154 TAILQ_INIT(&mp->mnt_vnodelist); 155 mp->mnt_renamelock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 156 mp->mnt_vnodelock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 157 mp->mnt_updating = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 158 mp->mnt_vnodecovered = vp; 159 mount_initspecific(mp); 160 161 error = fstrans_mount(mp); 162 KASSERT(error == 0); 163 164 mutex_enter(&mountgen_lock); 165 mp->mnt_gen = mountgen++; 166 mutex_exit(&mountgen_lock); 167 168 return mp; 169 } 170 171 /* 172 * vfs_rootmountalloc: lookup a filesystem type, and if found allocate and 173 * initialize a mount structure for it. 174 * 175 * Devname is usually updated by mount(8) after booting. 176 */ 177 int 178 vfs_rootmountalloc(const char *fstypename, const char *devname, 179 struct mount **mpp) 180 { 181 struct vfsops *vfsp = NULL; 182 struct mount *mp; 183 int error __diagused; 184 185 mutex_enter(&vfs_list_lock); 186 LIST_FOREACH(vfsp, &vfs_list, vfs_list) 187 if (!strncmp(vfsp->vfs_name, fstypename, 188 sizeof(mp->mnt_stat.f_fstypename))) 189 break; 190 if (vfsp == NULL) { 191 mutex_exit(&vfs_list_lock); 192 return (ENODEV); 193 } 194 vfsp->vfs_refcount++; 195 mutex_exit(&vfs_list_lock); 196 197 if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL) 198 return ENOMEM; 199 error = vfs_busy(mp); 200 KASSERT(error == 0); 201 mp->mnt_flag = MNT_RDONLY; 202 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, 203 sizeof(mp->mnt_stat.f_fstypename)); 204 mp->mnt_stat.f_mntonname[0] = '/'; 205 mp->mnt_stat.f_mntonname[1] = '\0'; 206 mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] = 207 '\0'; 208 (void)copystr(devname, mp->mnt_stat.f_mntfromname, 209 sizeof(mp->mnt_stat.f_mntfromname) - 1, 0); 210 *mpp = mp; 211 return 0; 212 } 213 214 /* 215 * vfs_getnewfsid: get a new unique fsid. 216 */ 217 void 218 vfs_getnewfsid(struct mount *mp) 219 { 220 static u_short xxxfs_mntid; 221 fsid_t tfsid; 222 int mtype; 223 224 mutex_enter(&mntid_lock); 225 mtype = makefstype(mp->mnt_op->vfs_name); 226 mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0); 227 mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype; 228 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 229 if (xxxfs_mntid == 0) 230 ++xxxfs_mntid; 231 tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid); 232 tfsid.__fsid_val[1] = mtype; 233 while (vfs_getvfs(&tfsid)) { 234 tfsid.__fsid_val[0]++; 235 xxxfs_mntid++; 236 } 237 mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0]; 238 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 239 mutex_exit(&mntid_lock); 240 } 241 242 /* 243 * Lookup a mount point by filesystem identifier. 244 * 245 * XXX Needs to add a reference to the mount point. 246 */ 247 struct mount * 248 vfs_getvfs(fsid_t *fsid) 249 { 250 mount_iterator_t *iter; 251 struct mount *mp; 252 253 mountlist_iterator_init(&iter); 254 while ((mp = mountlist_iterator_next(iter)) != NULL) { 255 if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] && 256 mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) { 257 mountlist_iterator_destroy(iter); 258 return mp; 259 } 260 } 261 mountlist_iterator_destroy(iter); 262 return NULL; 263 } 264 265 /* 266 * Take a reference to a mount structure. 267 */ 268 void 269 vfs_ref(struct mount *mp) 270 { 271 272 KASSERT(mp->mnt_refcnt > 0 || mutex_owned(&mountlist_lock)); 273 274 atomic_inc_uint(&mp->mnt_refcnt); 275 } 276 277 /* 278 * Drop a reference to a mount structure, freeing if the last reference. 279 */ 280 void 281 vfs_rele(struct mount *mp) 282 { 283 284 if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) { 285 return; 286 } 287 288 /* 289 * Nothing else has visibility of the mount: we can now 290 * free the data structures. 291 */ 292 KASSERT(mp->mnt_refcnt == 0); 293 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 294 mutex_obj_free(mp->mnt_updating); 295 mutex_obj_free(mp->mnt_renamelock); 296 mutex_obj_free(mp->mnt_vnodelock); 297 if (mp->mnt_op != NULL) { 298 vfs_delref(mp->mnt_op); 299 } 300 fstrans_unmount(mp); 301 /* 302 * Final free of mp gets done from fstrans_mount_dtor(). 303 * 304 * Prevents this memory to be reused as a mount before 305 * fstrans releases all references to it. 306 */ 307 } 308 309 /* 310 * Mark a mount point as busy, and gain a new reference to it. Used to 311 * prevent the file system from being unmounted during critical sections. 312 * 313 * vfs_busy can be called multiple times and by multiple threads 314 * and must be accompanied by the same number of vfs_unbusy calls. 315 * 316 * => The caller must hold a pre-existing reference to the mount. 317 * => Will fail if the file system is being unmounted, or is unmounted. 318 */ 319 static inline int 320 _vfs_busy(struct mount *mp, bool wait) 321 { 322 323 KASSERT(mp->mnt_refcnt > 0); 324 325 if (wait) { 326 fstrans_start(mp); 327 } else { 328 if (fstrans_start_nowait(mp)) 329 return EBUSY; 330 } 331 if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) { 332 fstrans_done(mp); 333 return ENOENT; 334 } 335 vfs_ref(mp); 336 return 0; 337 } 338 339 int 340 vfs_busy(struct mount *mp) 341 { 342 343 return _vfs_busy(mp, true); 344 } 345 346 int 347 vfs_trybusy(struct mount *mp) 348 { 349 350 return _vfs_busy(mp, false); 351 } 352 353 /* 354 * Unbusy a busy filesystem. 355 * 356 * Every successful vfs_busy() call must be undone by a vfs_unbusy() call. 357 */ 358 void 359 vfs_unbusy(struct mount *mp) 360 { 361 362 KASSERT(mp->mnt_refcnt > 0); 363 364 fstrans_done(mp); 365 vfs_rele(mp); 366 } 367 368 struct vnode_iterator { 369 vnode_impl_t vi_vnode; 370 }; 371 372 void 373 vfs_vnode_iterator_init(struct mount *mp, struct vnode_iterator **vnip) 374 { 375 vnode_t *vp; 376 vnode_impl_t *vip; 377 378 vp = vnalloc_marker(mp); 379 vip = VNODE_TO_VIMPL(vp); 380 381 mutex_enter(mp->mnt_vnodelock); 382 TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vip, vi_mntvnodes); 383 vp->v_usecount = 1; 384 mutex_exit(mp->mnt_vnodelock); 385 386 *vnip = (struct vnode_iterator *)vip; 387 } 388 389 void 390 vfs_vnode_iterator_destroy(struct vnode_iterator *vni) 391 { 392 vnode_impl_t *mvip = &vni->vi_vnode; 393 vnode_t *mvp = VIMPL_TO_VNODE(mvip); 394 kmutex_t *lock; 395 396 KASSERT(vnis_marker(mvp)); 397 if (vrefcnt(mvp) != 0) { 398 lock = mvp->v_mount->mnt_vnodelock; 399 mutex_enter(lock); 400 TAILQ_REMOVE(&mvp->v_mount->mnt_vnodelist, mvip, vi_mntvnodes); 401 mvp->v_usecount = 0; 402 mutex_exit(lock); 403 } 404 vnfree_marker(mvp); 405 } 406 407 static struct vnode * 408 vfs_vnode_iterator_next1(struct vnode_iterator *vni, 409 bool (*f)(void *, struct vnode *), void *cl, bool do_wait) 410 { 411 vnode_impl_t *mvip = &vni->vi_vnode; 412 struct mount *mp = VIMPL_TO_VNODE(mvip)->v_mount; 413 vnode_t *vp; 414 vnode_impl_t *vip; 415 kmutex_t *lock; 416 int error; 417 418 KASSERT(vnis_marker(VIMPL_TO_VNODE(mvip))); 419 420 lock = mp->mnt_vnodelock; 421 do { 422 mutex_enter(lock); 423 vip = TAILQ_NEXT(mvip, vi_mntvnodes); 424 TAILQ_REMOVE(&mp->mnt_vnodelist, mvip, vi_mntvnodes); 425 VIMPL_TO_VNODE(mvip)->v_usecount = 0; 426 again: 427 if (vip == NULL) { 428 mutex_exit(lock); 429 return NULL; 430 } 431 vp = VIMPL_TO_VNODE(vip); 432 KASSERT(vp != NULL); 433 mutex_enter(vp->v_interlock); 434 if (vnis_marker(vp) || 435 vdead_check(vp, (do_wait ? 0 : VDEAD_NOWAIT)) || 436 (f && !(*f)(cl, vp))) { 437 mutex_exit(vp->v_interlock); 438 vip = TAILQ_NEXT(vip, vi_mntvnodes); 439 goto again; 440 } 441 442 TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vip, mvip, vi_mntvnodes); 443 VIMPL_TO_VNODE(mvip)->v_usecount = 1; 444 mutex_exit(lock); 445 error = vcache_vget(vp); 446 KASSERT(error == 0 || error == ENOENT); 447 } while (error != 0); 448 449 return vp; 450 } 451 452 struct vnode * 453 vfs_vnode_iterator_next(struct vnode_iterator *vni, 454 bool (*f)(void *, struct vnode *), void *cl) 455 { 456 457 return vfs_vnode_iterator_next1(vni, f, cl, false); 458 } 459 460 /* 461 * Move a vnode from one mount queue to another. 462 */ 463 void 464 vfs_insmntque(vnode_t *vp, struct mount *mp) 465 { 466 vnode_impl_t *vip = VNODE_TO_VIMPL(vp); 467 struct mount *omp; 468 kmutex_t *lock; 469 470 KASSERT(mp == NULL || (mp->mnt_iflag & IMNT_UNMOUNT) == 0 || 471 vp->v_tag == VT_VFS); 472 473 /* 474 * Delete from old mount point vnode list, if on one. 475 */ 476 if ((omp = vp->v_mount) != NULL) { 477 lock = omp->mnt_vnodelock; 478 mutex_enter(lock); 479 TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vip, vi_mntvnodes); 480 mutex_exit(lock); 481 } 482 483 /* 484 * Insert into list of vnodes for the new mount point, if 485 * available. The caller must take a reference on the mount 486 * structure and donate to the vnode. 487 */ 488 if ((vp->v_mount = mp) != NULL) { 489 lock = mp->mnt_vnodelock; 490 mutex_enter(lock); 491 TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vip, vi_mntvnodes); 492 mutex_exit(lock); 493 } 494 495 if (omp != NULL) { 496 /* Release reference to old mount. */ 497 vfs_rele(omp); 498 } 499 } 500 501 /* 502 * Remove any vnodes in the vnode table belonging to mount point mp. 503 * 504 * If FORCECLOSE is not specified, there should not be any active ones, 505 * return error if any are found (nb: this is a user error, not a 506 * system error). If FORCECLOSE is specified, detach any active vnodes 507 * that are found. 508 * 509 * If WRITECLOSE is set, only flush out regular file vnodes open for 510 * writing. 511 * 512 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 513 */ 514 #ifdef DEBUG 515 int busyprt = 0; /* print out busy vnodes */ 516 struct ctldebug debug1 = { "busyprt", &busyprt }; 517 #endif 518 519 static vnode_t * 520 vflushnext(struct vnode_iterator *marker, int *when) 521 { 522 if (getticks() > *when) { 523 yield(); 524 *when = getticks() + hz / 10; 525 } 526 return vfs_vnode_iterator_next1(marker, NULL, NULL, true); 527 } 528 529 /* 530 * Flush one vnode. Referenced on entry, unreferenced on return. 531 */ 532 static int 533 vflush_one(vnode_t *vp, vnode_t *skipvp, int flags) 534 { 535 int error; 536 struct vattr vattr; 537 538 if (vp == skipvp || 539 ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM))) { 540 vrele(vp); 541 return 0; 542 } 543 /* 544 * If WRITECLOSE is set, only flush out regular file 545 * vnodes open for writing or open and unlinked. 546 */ 547 if ((flags & WRITECLOSE)) { 548 if (vp->v_type != VREG) { 549 vrele(vp); 550 return 0; 551 } 552 error = vn_lock(vp, LK_EXCLUSIVE); 553 if (error) { 554 KASSERT(error == ENOENT); 555 vrele(vp); 556 return 0; 557 } 558 error = VOP_FSYNC(vp, curlwp->l_cred, FSYNC_WAIT, 0, 0); 559 if (error == 0) 560 error = VOP_GETATTR(vp, &vattr, curlwp->l_cred); 561 VOP_UNLOCK(vp); 562 if (error) { 563 vrele(vp); 564 return error; 565 } 566 if (vp->v_writecount == 0 && vattr.va_nlink > 0) { 567 vrele(vp); 568 return 0; 569 } 570 } 571 /* 572 * First try to recycle the vnode. 573 */ 574 if (vrecycle(vp)) 575 return 0; 576 /* 577 * If FORCECLOSE is set, forcibly close the vnode. 578 * For block or character devices, revert to an 579 * anonymous device. For all other files, just 580 * kill them. 581 */ 582 if (flags & FORCECLOSE) { 583 if (vrefcnt(vp) > 1 && 584 (vp->v_type == VBLK || vp->v_type == VCHR)) 585 vcache_make_anon(vp); 586 else 587 vgone(vp); 588 return 0; 589 } 590 vrele(vp); 591 return EBUSY; 592 } 593 594 int 595 vflush(struct mount *mp, vnode_t *skipvp, int flags) 596 { 597 vnode_t *vp; 598 struct vnode_iterator *marker; 599 int busy, error, when, retries = 2; 600 601 do { 602 busy = error = when = 0; 603 604 /* 605 * First, flush out any vnode references from the 606 * deferred vrele list. 607 */ 608 vrele_flush(mp); 609 610 vfs_vnode_iterator_init(mp, &marker); 611 612 while ((vp = vflushnext(marker, &when)) != NULL) { 613 error = vflush_one(vp, skipvp, flags); 614 if (error == EBUSY) { 615 error = 0; 616 busy++; 617 #ifdef DEBUG 618 if (busyprt && retries == 0) 619 vprint("vflush: busy vnode", vp); 620 #endif 621 } else if (error != 0) { 622 break; 623 } 624 } 625 626 vfs_vnode_iterator_destroy(marker); 627 } while (error == 0 && busy > 0 && retries-- > 0); 628 629 if (error) 630 return error; 631 if (busy) 632 return EBUSY; 633 return 0; 634 } 635 636 /* 637 * Mount a file system. 638 */ 639 640 /* 641 * Scan all active processes to see if any of them have a current or root 642 * directory onto which the new filesystem has just been mounted. If so, 643 * replace them with the new mount point. 644 */ 645 static void 646 mount_checkdirs(vnode_t *olddp) 647 { 648 vnode_t *newdp, *rele1, *rele2; 649 struct cwdinfo *cwdi; 650 struct proc *p; 651 bool retry; 652 653 if (vrefcnt(olddp) == 1) { 654 return; 655 } 656 if (VFS_ROOT(olddp->v_mountedhere, LK_EXCLUSIVE, &newdp)) 657 panic("mount: lost mount"); 658 659 do { 660 retry = false; 661 mutex_enter(&proc_lock); 662 PROCLIST_FOREACH(p, &allproc) { 663 if ((cwdi = p->p_cwdi) == NULL) 664 continue; 665 /* 666 * Cannot change to the old directory any more, 667 * so even if we see a stale value it is not a 668 * problem. 669 */ 670 if (cwdi->cwdi_cdir != olddp && 671 cwdi->cwdi_rdir != olddp) 672 continue; 673 retry = true; 674 rele1 = NULL; 675 rele2 = NULL; 676 atomic_inc_uint(&cwdi->cwdi_refcnt); 677 mutex_exit(&proc_lock); 678 rw_enter(&cwdi->cwdi_lock, RW_WRITER); 679 if (cwdi->cwdi_cdir == olddp) { 680 rele1 = cwdi->cwdi_cdir; 681 vref(newdp); 682 cwdi->cwdi_cdir = newdp; 683 } 684 if (cwdi->cwdi_rdir == olddp) { 685 rele2 = cwdi->cwdi_rdir; 686 vref(newdp); 687 cwdi->cwdi_rdir = newdp; 688 } 689 rw_exit(&cwdi->cwdi_lock); 690 cwdfree(cwdi); 691 if (rele1 != NULL) 692 vrele(rele1); 693 if (rele2 != NULL) 694 vrele(rele2); 695 mutex_enter(&proc_lock); 696 break; 697 } 698 mutex_exit(&proc_lock); 699 } while (retry); 700 701 if (rootvnode == olddp) { 702 vrele(rootvnode); 703 vref(newdp); 704 rootvnode = newdp; 705 } 706 vput(newdp); 707 } 708 709 /* 710 * Start extended attributes 711 */ 712 static int 713 start_extattr(struct mount *mp) 714 { 715 int error; 716 717 error = VFS_EXTATTRCTL(mp, EXTATTR_CMD_START, NULL, 0, NULL); 718 if (error) 719 printf("%s: failed to start extattr: error = %d\n", 720 mp->mnt_stat.f_mntonname, error); 721 722 return error; 723 } 724 725 int 726 mount_domount(struct lwp *l, vnode_t **vpp, struct vfsops *vfsops, 727 const char *path, int flags, void *data, size_t *data_len) 728 { 729 vnode_t *vp = *vpp; 730 struct mount *mp; 731 struct pathbuf *pb; 732 struct nameidata nd; 733 int error; 734 735 error = kauth_authorize_system(l->l_cred, KAUTH_SYSTEM_MOUNT, 736 KAUTH_REQ_SYSTEM_MOUNT_NEW, vp, KAUTH_ARG(flags), data); 737 if (error) { 738 vfs_delref(vfsops); 739 return error; 740 } 741 742 /* Cannot make a non-dir a mount-point (from here anyway). */ 743 if (vp->v_type != VDIR) { 744 vfs_delref(vfsops); 745 return ENOTDIR; 746 } 747 748 if (flags & MNT_EXPORTED) { 749 vfs_delref(vfsops); 750 return EINVAL; 751 } 752 753 if ((mp = vfs_mountalloc(vfsops, vp)) == NULL) { 754 vfs_delref(vfsops); 755 return ENOMEM; 756 } 757 758 mp->mnt_stat.f_owner = kauth_cred_geteuid(l->l_cred); 759 760 /* 761 * The underlying file system may refuse the mount for 762 * various reasons. Allow the user to force it to happen. 763 * 764 * Set the mount level flags. 765 */ 766 mp->mnt_flag = flags & (MNT_BASIC_FLAGS | MNT_FORCE | MNT_IGNORE); 767 768 mutex_enter(mp->mnt_updating); 769 error = VFS_MOUNT(mp, path, data, data_len); 770 mp->mnt_flag &= ~MNT_OP_FLAGS; 771 772 if (error != 0) 773 goto err_unmounted; 774 775 /* 776 * Validate and prepare the mount point. 777 */ 778 error = pathbuf_copyin(path, &pb); 779 if (error != 0) { 780 goto err_mounted; 781 } 782 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 783 error = namei(&nd); 784 pathbuf_destroy(pb); 785 if (error != 0) { 786 goto err_mounted; 787 } 788 if (nd.ni_vp != vp) { 789 vput(nd.ni_vp); 790 error = EINVAL; 791 goto err_mounted; 792 } 793 if (vp->v_mountedhere != NULL) { 794 vput(nd.ni_vp); 795 error = EBUSY; 796 goto err_mounted; 797 } 798 error = vinvalbuf(vp, V_SAVE, l->l_cred, l, 0, 0); 799 if (error != 0) { 800 vput(nd.ni_vp); 801 goto err_mounted; 802 } 803 804 /* 805 * Put the new filesystem on the mount list after root. 806 */ 807 cache_purge(vp); 808 mp->mnt_iflag &= ~IMNT_WANTRDWR; 809 810 mountlist_append(mp); 811 if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0) 812 vfs_syncer_add_to_worklist(mp); 813 vp->v_mountedhere = mp; 814 vput(nd.ni_vp); 815 816 mount_checkdirs(vp); 817 mutex_exit(mp->mnt_updating); 818 819 /* Hold an additional reference to the mount across VFS_START(). */ 820 vfs_ref(mp); 821 (void) VFS_STATVFS(mp, &mp->mnt_stat); 822 error = VFS_START(mp, 0); 823 if (error) { 824 vrele(vp); 825 } else if (flags & MNT_EXTATTR) { 826 if (start_extattr(mp) != 0) 827 mp->mnt_flag &= ~MNT_EXTATTR; 828 } 829 /* Drop reference held for VFS_START(). */ 830 vfs_rele(mp); 831 *vpp = NULL; 832 return error; 833 834 err_mounted: 835 if (vfs_suspend(mp, 0)) 836 panic("Suspending fresh file system failed"); 837 if (VFS_UNMOUNT(mp, MNT_FORCE) != 0) 838 panic("Unmounting fresh file system failed"); 839 vfs_resume(mp); 840 841 err_unmounted: 842 vp->v_mountedhere = NULL; 843 mutex_exit(mp->mnt_updating); 844 vfs_rele(mp); 845 846 return error; 847 } 848 849 /* 850 * Do the actual file system unmount. File system is assumed to have 851 * been locked by the caller. 852 * 853 * => Caller hold reference to the mount, explicitly for dounmount(). 854 */ 855 int 856 dounmount(struct mount *mp, int flags, struct lwp *l) 857 { 858 vnode_t *coveredvp; 859 int error, async, used_syncer, used_extattr; 860 const bool was_suspended = fstrans_is_owner(mp); 861 862 #if NVERIEXEC > 0 863 error = veriexec_unmountchk(mp); 864 if (error) 865 return (error); 866 #endif /* NVERIEXEC > 0 */ 867 868 if (!was_suspended) { 869 error = vfs_suspend(mp, 0); 870 if (error) { 871 return error; 872 } 873 } 874 875 KASSERT((mp->mnt_iflag & IMNT_GONE) == 0); 876 877 used_syncer = (mp->mnt_iflag & IMNT_ONWORKLIST) != 0; 878 used_extattr = mp->mnt_flag & MNT_EXTATTR; 879 880 mp->mnt_iflag |= IMNT_UNMOUNT; 881 mutex_enter(mp->mnt_updating); 882 async = mp->mnt_flag & MNT_ASYNC; 883 mp->mnt_flag &= ~MNT_ASYNC; 884 cache_purgevfs(mp); /* remove cache entries for this file sys */ 885 if (used_syncer) 886 vfs_syncer_remove_from_worklist(mp); 887 error = 0; 888 if (((mp->mnt_flag & MNT_RDONLY) == 0) && ((flags & MNT_FORCE) == 0)) { 889 error = VFS_SYNC(mp, MNT_WAIT, l->l_cred); 890 } 891 if (error == 0 || (flags & MNT_FORCE)) { 892 error = VFS_UNMOUNT(mp, flags); 893 } 894 if (error) { 895 mp->mnt_iflag &= ~IMNT_UNMOUNT; 896 if ((mp->mnt_flag & (MNT_RDONLY | MNT_ASYNC)) == 0) 897 vfs_syncer_add_to_worklist(mp); 898 mp->mnt_flag |= async; 899 mutex_exit(mp->mnt_updating); 900 if (!was_suspended) 901 vfs_resume(mp); 902 if (used_extattr) { 903 if (start_extattr(mp) != 0) 904 mp->mnt_flag &= ~MNT_EXTATTR; 905 else 906 mp->mnt_flag |= MNT_EXTATTR; 907 } 908 return (error); 909 } 910 mutex_exit(mp->mnt_updating); 911 912 /* 913 * mark filesystem as gone to prevent further umounts 914 * after mnt_umounting lock is gone, this also prevents 915 * vfs_busy() from succeeding. 916 */ 917 mp->mnt_iflag |= IMNT_GONE; 918 if (!was_suspended) 919 vfs_resume(mp); 920 921 if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) { 922 vn_lock(coveredvp, LK_EXCLUSIVE | LK_RETRY); 923 coveredvp->v_mountedhere = NULL; 924 VOP_UNLOCK(coveredvp); 925 } 926 mountlist_remove(mp); 927 if (TAILQ_FIRST(&mp->mnt_vnodelist) != NULL) 928 panic("unmount: dangling vnode"); 929 vfs_hooks_unmount(mp); 930 931 vfs_rele(mp); /* reference from mount() */ 932 if (coveredvp != NULLVP) { 933 vrele(coveredvp); 934 } 935 return (0); 936 } 937 938 /* 939 * Unmount all file systems. 940 * We traverse the list in reverse order under the assumption that doing so 941 * will avoid needing to worry about dependencies. 942 */ 943 bool 944 vfs_unmountall(struct lwp *l) 945 { 946 947 printf("unmounting file systems...\n"); 948 return vfs_unmountall1(l, true, true); 949 } 950 951 static void 952 vfs_unmount_print(struct mount *mp, const char *pfx) 953 { 954 955 aprint_verbose("%sunmounted %s on %s type %s\n", pfx, 956 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, 957 mp->mnt_stat.f_fstypename); 958 } 959 960 /* 961 * Return the mount with the highest generation less than "gen". 962 */ 963 static struct mount * 964 vfs_unmount_next(uint64_t gen) 965 { 966 mount_iterator_t *iter; 967 struct mount *mp, *nmp; 968 969 nmp = NULL; 970 971 mountlist_iterator_init(&iter); 972 while ((mp = mountlist_iterator_next(iter)) != NULL) { 973 if ((nmp == NULL || mp->mnt_gen > nmp->mnt_gen) && 974 mp->mnt_gen < gen) { 975 if (nmp != NULL) 976 vfs_rele(nmp); 977 nmp = mp; 978 vfs_ref(nmp); 979 } 980 } 981 mountlist_iterator_destroy(iter); 982 983 return nmp; 984 } 985 986 bool 987 vfs_unmount_forceone(struct lwp *l) 988 { 989 struct mount *mp; 990 int error; 991 992 mp = vfs_unmount_next(mountgen); 993 if (mp == NULL) { 994 return false; 995 } 996 997 #ifdef DEBUG 998 printf("forcefully unmounting %s (%s)...\n", 999 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname); 1000 #endif 1001 if ((error = dounmount(mp, MNT_FORCE, l)) == 0) { 1002 vfs_unmount_print(mp, "forcefully "); 1003 return true; 1004 } else { 1005 vfs_rele(mp); 1006 } 1007 1008 #ifdef DEBUG 1009 printf("forceful unmount of %s failed with error %d\n", 1010 mp->mnt_stat.f_mntonname, error); 1011 #endif 1012 1013 return false; 1014 } 1015 1016 bool 1017 vfs_unmountall1(struct lwp *l, bool force, bool verbose) 1018 { 1019 struct mount *mp; 1020 bool any_error = false, progress = false; 1021 uint64_t gen; 1022 int error; 1023 1024 gen = mountgen; 1025 for (;;) { 1026 mp = vfs_unmount_next(gen); 1027 if (mp == NULL) 1028 break; 1029 gen = mp->mnt_gen; 1030 1031 #ifdef DEBUG 1032 printf("unmounting %p %s (%s)...\n", 1033 (void *)mp, mp->mnt_stat.f_mntonname, 1034 mp->mnt_stat.f_mntfromname); 1035 #endif 1036 if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) { 1037 vfs_unmount_print(mp, ""); 1038 progress = true; 1039 } else { 1040 vfs_rele(mp); 1041 if (verbose) { 1042 printf("unmount of %s failed with error %d\n", 1043 mp->mnt_stat.f_mntonname, error); 1044 } 1045 any_error = true; 1046 } 1047 } 1048 if (verbose) { 1049 printf("unmounting done\n"); 1050 } 1051 if (any_error && verbose) { 1052 printf("WARNING: some file systems would not unmount\n"); 1053 } 1054 return progress; 1055 } 1056 1057 void 1058 vfs_sync_all(struct lwp *l) 1059 { 1060 printf("syncing disks... "); 1061 1062 /* remove user processes from run queue */ 1063 suspendsched(); 1064 (void)spl0(); 1065 1066 /* avoid coming back this way again if we panic. */ 1067 doing_shutdown = 1; 1068 1069 do_sys_sync(l); 1070 1071 /* Wait for sync to finish. */ 1072 if (vfs_syncwait() != 0) { 1073 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 1074 Debugger(); 1075 #endif 1076 printf("giving up\n"); 1077 return; 1078 } else 1079 printf("done\n"); 1080 } 1081 1082 /* 1083 * Sync and unmount file systems before shutting down. 1084 */ 1085 void 1086 vfs_shutdown(void) 1087 { 1088 lwp_t *l = curlwp; 1089 1090 vfs_sync_all(l); 1091 1092 /* 1093 * If we have paniced - do not make the situation potentially 1094 * worse by unmounting the file systems. 1095 */ 1096 if (panicstr != NULL) { 1097 return; 1098 } 1099 1100 /* Unmount file systems. */ 1101 vfs_unmountall(l); 1102 } 1103 1104 /* 1105 * Print a list of supported file system types (used by vfs_mountroot) 1106 */ 1107 static void 1108 vfs_print_fstypes(void) 1109 { 1110 struct vfsops *v; 1111 int cnt = 0; 1112 1113 mutex_enter(&vfs_list_lock); 1114 LIST_FOREACH(v, &vfs_list, vfs_list) 1115 ++cnt; 1116 mutex_exit(&vfs_list_lock); 1117 1118 if (cnt == 0) { 1119 printf("WARNING: No file system modules have been loaded.\n"); 1120 return; 1121 } 1122 1123 printf("Supported file systems:"); 1124 mutex_enter(&vfs_list_lock); 1125 LIST_FOREACH(v, &vfs_list, vfs_list) { 1126 printf(" %s", v->vfs_name); 1127 } 1128 mutex_exit(&vfs_list_lock); 1129 printf("\n"); 1130 } 1131 1132 /* 1133 * Mount the root file system. If the operator didn't specify a 1134 * file system to use, try all possible file systems until one 1135 * succeeds. 1136 */ 1137 int 1138 vfs_mountroot(void) 1139 { 1140 struct vfsops *v; 1141 int error = ENODEV; 1142 1143 if (root_device == NULL) 1144 panic("vfs_mountroot: root device unknown"); 1145 1146 switch (device_class(root_device)) { 1147 case DV_IFNET: 1148 if (rootdev != NODEV) 1149 panic("vfs_mountroot: rootdev set for DV_IFNET " 1150 "(0x%llx -> %llu,%llu)", 1151 (unsigned long long)rootdev, 1152 (unsigned long long)major(rootdev), 1153 (unsigned long long)minor(rootdev)); 1154 break; 1155 1156 case DV_DISK: 1157 if (rootdev == NODEV) 1158 panic("vfs_mountroot: rootdev not set for DV_DISK"); 1159 if (bdevvp(rootdev, &rootvp)) 1160 panic("vfs_mountroot: can't get vnode for rootdev"); 1161 error = VOP_OPEN(rootvp, FREAD, FSCRED); 1162 if (error) { 1163 printf("vfs_mountroot: can't open root device\n"); 1164 return (error); 1165 } 1166 break; 1167 1168 case DV_VIRTUAL: 1169 break; 1170 1171 default: 1172 printf("%s: inappropriate for root file system\n", 1173 device_xname(root_device)); 1174 return (ENODEV); 1175 } 1176 1177 /* 1178 * If user specified a root fs type, use it. Make sure the 1179 * specified type exists and has a mount_root() 1180 */ 1181 if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) { 1182 v = vfs_getopsbyname(rootfstype); 1183 error = EFTYPE; 1184 if (v != NULL) { 1185 if (v->vfs_mountroot != NULL) { 1186 error = (v->vfs_mountroot)(); 1187 } 1188 v->vfs_refcount--; 1189 } 1190 goto done; 1191 } 1192 1193 /* 1194 * Try each file system currently configured into the kernel. 1195 */ 1196 mutex_enter(&vfs_list_lock); 1197 LIST_FOREACH(v, &vfs_list, vfs_list) { 1198 if (v->vfs_mountroot == NULL) 1199 continue; 1200 #ifdef DEBUG 1201 aprint_normal("mountroot: trying %s...\n", v->vfs_name); 1202 #endif 1203 v->vfs_refcount++; 1204 mutex_exit(&vfs_list_lock); 1205 error = (*v->vfs_mountroot)(); 1206 mutex_enter(&vfs_list_lock); 1207 v->vfs_refcount--; 1208 if (!error) { 1209 aprint_normal("root file system type: %s\n", 1210 v->vfs_name); 1211 break; 1212 } 1213 } 1214 mutex_exit(&vfs_list_lock); 1215 1216 if (v == NULL) { 1217 vfs_print_fstypes(); 1218 printf("no file system for %s", device_xname(root_device)); 1219 if (device_class(root_device) == DV_DISK) 1220 printf(" (dev 0x%llx)", (unsigned long long)rootdev); 1221 printf("\n"); 1222 error = EFTYPE; 1223 } 1224 1225 done: 1226 if (error && device_class(root_device) == DV_DISK) { 1227 VOP_CLOSE(rootvp, FREAD, FSCRED); 1228 vrele(rootvp); 1229 } 1230 if (error == 0) { 1231 mount_iterator_t *iter; 1232 struct mount *mp; 1233 extern struct cwdinfo cwdi0; 1234 1235 mountlist_iterator_init(&iter); 1236 mp = mountlist_iterator_next(iter); 1237 KASSERT(mp != NULL); 1238 mountlist_iterator_destroy(iter); 1239 1240 mp->mnt_flag |= MNT_ROOTFS; 1241 mp->mnt_op->vfs_refcount++; 1242 1243 /* 1244 * Get the vnode for '/'. Set cwdi0.cwdi_cdir to 1245 * reference it, and donate it the reference grabbed 1246 * with VFS_ROOT(). 1247 */ 1248 error = VFS_ROOT(mp, LK_NONE, &rootvnode); 1249 if (error) 1250 panic("cannot find root vnode, error=%d", error); 1251 cwdi0.cwdi_cdir = rootvnode; 1252 cwdi0.cwdi_rdir = NULL; 1253 1254 /* 1255 * Now that root is mounted, we can fixup initproc's CWD 1256 * info. All other processes are kthreads, which merely 1257 * share proc0's CWD info. 1258 */ 1259 initproc->p_cwdi->cwdi_cdir = rootvnode; 1260 vref(initproc->p_cwdi->cwdi_cdir); 1261 initproc->p_cwdi->cwdi_rdir = NULL; 1262 /* 1263 * Enable loading of modules from the filesystem 1264 */ 1265 module_load_vfs_init(); 1266 1267 } 1268 return (error); 1269 } 1270 1271 /* 1272 * mount_specific_key_create -- 1273 * Create a key for subsystem mount-specific data. 1274 */ 1275 int 1276 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor) 1277 { 1278 1279 return specificdata_key_create(mount_specificdata_domain, keyp, dtor); 1280 } 1281 1282 /* 1283 * mount_specific_key_delete -- 1284 * Delete a key for subsystem mount-specific data. 1285 */ 1286 void 1287 mount_specific_key_delete(specificdata_key_t key) 1288 { 1289 1290 specificdata_key_delete(mount_specificdata_domain, key); 1291 } 1292 1293 /* 1294 * mount_initspecific -- 1295 * Initialize a mount's specificdata container. 1296 */ 1297 void 1298 mount_initspecific(struct mount *mp) 1299 { 1300 int error __diagused; 1301 1302 error = specificdata_init(mount_specificdata_domain, 1303 &mp->mnt_specdataref); 1304 KASSERT(error == 0); 1305 } 1306 1307 /* 1308 * mount_finispecific -- 1309 * Finalize a mount's specificdata container. 1310 */ 1311 void 1312 mount_finispecific(struct mount *mp) 1313 { 1314 1315 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 1316 } 1317 1318 /* 1319 * mount_getspecific -- 1320 * Return mount-specific data corresponding to the specified key. 1321 */ 1322 void * 1323 mount_getspecific(struct mount *mp, specificdata_key_t key) 1324 { 1325 1326 return specificdata_getspecific(mount_specificdata_domain, 1327 &mp->mnt_specdataref, key); 1328 } 1329 1330 /* 1331 * mount_setspecific -- 1332 * Set mount-specific data corresponding to the specified key. 1333 */ 1334 void 1335 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data) 1336 { 1337 1338 specificdata_setspecific(mount_specificdata_domain, 1339 &mp->mnt_specdataref, key, data); 1340 } 1341 1342 /* 1343 * Check to see if a filesystem is mounted on a block device. 1344 */ 1345 int 1346 vfs_mountedon(vnode_t *vp) 1347 { 1348 vnode_t *vq; 1349 int error = 0; 1350 1351 if (vp->v_type != VBLK) 1352 return ENOTBLK; 1353 if (spec_node_getmountedfs(vp) != NULL) 1354 return EBUSY; 1355 if (spec_node_lookup_by_dev(vp->v_type, vp->v_rdev, &vq) == 0) { 1356 if (spec_node_getmountedfs(vq) != NULL) 1357 error = EBUSY; 1358 vrele(vq); 1359 } 1360 1361 return error; 1362 } 1363 1364 /* 1365 * Check if a device pointed to by vp is mounted. 1366 * 1367 * Returns: 1368 * EINVAL if it's not a disk 1369 * EBUSY if it's a disk and mounted 1370 * 0 if it's a disk and not mounted 1371 */ 1372 int 1373 rawdev_mounted(vnode_t *vp, vnode_t **bvpp) 1374 { 1375 vnode_t *bvp; 1376 dev_t dev; 1377 int d_type; 1378 1379 bvp = NULL; 1380 d_type = D_OTHER; 1381 1382 if (iskmemvp(vp)) 1383 return EINVAL; 1384 1385 switch (vp->v_type) { 1386 case VCHR: { 1387 const struct cdevsw *cdev; 1388 1389 dev = vp->v_rdev; 1390 cdev = cdevsw_lookup(dev); 1391 if (cdev != NULL) { 1392 dev_t blkdev; 1393 1394 blkdev = devsw_chr2blk(dev); 1395 if (blkdev != NODEV) { 1396 if (vfinddev(blkdev, VBLK, &bvp) != 0) { 1397 d_type = (cdev->d_flag & D_TYPEMASK); 1398 /* XXX: what if bvp disappears? */ 1399 vrele(bvp); 1400 } 1401 } 1402 } 1403 1404 break; 1405 } 1406 1407 case VBLK: { 1408 const struct bdevsw *bdev; 1409 1410 dev = vp->v_rdev; 1411 bdev = bdevsw_lookup(dev); 1412 if (bdev != NULL) 1413 d_type = (bdev->d_flag & D_TYPEMASK); 1414 1415 bvp = vp; 1416 1417 break; 1418 } 1419 1420 default: 1421 break; 1422 } 1423 1424 if (d_type != D_DISK) 1425 return EINVAL; 1426 1427 if (bvpp != NULL) 1428 *bvpp = bvp; 1429 1430 /* 1431 * XXX: This is bogus. We should be failing the request 1432 * XXX: not only if this specific slice is mounted, but 1433 * XXX: if it's on a disk with any other mounted slice. 1434 */ 1435 if (vfs_mountedon(bvp)) 1436 return EBUSY; 1437 1438 return 0; 1439 } 1440 1441 /* 1442 * Make a 'unique' number from a mount type name. 1443 */ 1444 long 1445 makefstype(const char *type) 1446 { 1447 long rv; 1448 1449 for (rv = 0; *type; type++) { 1450 rv <<= 2; 1451 rv ^= *type; 1452 } 1453 return rv; 1454 } 1455 1456 static struct mountlist_entry * 1457 mountlist_alloc(enum mountlist_type type, struct mount *mp) 1458 { 1459 struct mountlist_entry *me; 1460 1461 me = kmem_zalloc(sizeof(*me), KM_SLEEP); 1462 me->me_mount = mp; 1463 me->me_type = type; 1464 1465 return me; 1466 } 1467 1468 static void 1469 mountlist_free(struct mountlist_entry *me) 1470 { 1471 1472 kmem_free(me, sizeof(*me)); 1473 } 1474 1475 void 1476 mountlist_iterator_init(mount_iterator_t **mip) 1477 { 1478 struct mountlist_entry *me; 1479 1480 me = mountlist_alloc(ME_MARKER, NULL); 1481 mutex_enter(&mountlist_lock); 1482 TAILQ_INSERT_HEAD(&mountlist, me, me_list); 1483 mutex_exit(&mountlist_lock); 1484 *mip = (mount_iterator_t *)me; 1485 } 1486 1487 void 1488 mountlist_iterator_destroy(mount_iterator_t *mi) 1489 { 1490 struct mountlist_entry *marker = &mi->mi_entry; 1491 1492 if (marker->me_mount != NULL) 1493 vfs_unbusy(marker->me_mount); 1494 1495 mutex_enter(&mountlist_lock); 1496 TAILQ_REMOVE(&mountlist, marker, me_list); 1497 mutex_exit(&mountlist_lock); 1498 1499 mountlist_free(marker); 1500 1501 } 1502 1503 /* 1504 * Return the next mount or NULL for this iterator. 1505 * Mark it busy on success. 1506 */ 1507 static inline struct mount * 1508 _mountlist_iterator_next(mount_iterator_t *mi, bool wait) 1509 { 1510 struct mountlist_entry *me, *marker = &mi->mi_entry; 1511 struct mount *mp; 1512 int error; 1513 1514 if (marker->me_mount != NULL) { 1515 vfs_unbusy(marker->me_mount); 1516 marker->me_mount = NULL; 1517 } 1518 1519 mutex_enter(&mountlist_lock); 1520 for (;;) { 1521 KASSERT(marker->me_type == ME_MARKER); 1522 1523 me = TAILQ_NEXT(marker, me_list); 1524 if (me == NULL) { 1525 /* End of list: keep marker and return. */ 1526 mutex_exit(&mountlist_lock); 1527 return NULL; 1528 } 1529 TAILQ_REMOVE(&mountlist, marker, me_list); 1530 TAILQ_INSERT_AFTER(&mountlist, me, marker, me_list); 1531 1532 /* Skip other markers. */ 1533 if (me->me_type != ME_MOUNT) 1534 continue; 1535 1536 /* Take an initial reference for vfs_busy() below. */ 1537 mp = me->me_mount; 1538 KASSERT(mp != NULL); 1539 vfs_ref(mp); 1540 mutex_exit(&mountlist_lock); 1541 1542 /* Try to mark this mount busy and return on success. */ 1543 if (wait) 1544 error = vfs_busy(mp); 1545 else 1546 error = vfs_trybusy(mp); 1547 if (error == 0) { 1548 vfs_rele(mp); 1549 marker->me_mount = mp; 1550 return mp; 1551 } 1552 vfs_rele(mp); 1553 mutex_enter(&mountlist_lock); 1554 } 1555 } 1556 1557 struct mount * 1558 mountlist_iterator_next(mount_iterator_t *mi) 1559 { 1560 1561 return _mountlist_iterator_next(mi, true); 1562 } 1563 1564 struct mount * 1565 mountlist_iterator_trynext(mount_iterator_t *mi) 1566 { 1567 1568 return _mountlist_iterator_next(mi, false); 1569 } 1570 1571 /* 1572 * Attach new mount to the end of the mount list. 1573 */ 1574 void 1575 mountlist_append(struct mount *mp) 1576 { 1577 struct mountlist_entry *me; 1578 1579 me = mountlist_alloc(ME_MOUNT, mp); 1580 mutex_enter(&mountlist_lock); 1581 TAILQ_INSERT_TAIL(&mountlist, me, me_list); 1582 mutex_exit(&mountlist_lock); 1583 } 1584 1585 /* 1586 * Remove mount from mount list. 1587 */void 1588 mountlist_remove(struct mount *mp) 1589 { 1590 struct mountlist_entry *me; 1591 1592 mutex_enter(&mountlist_lock); 1593 TAILQ_FOREACH(me, &mountlist, me_list) 1594 if (me->me_type == ME_MOUNT && me->me_mount == mp) 1595 break; 1596 KASSERT(me != NULL); 1597 TAILQ_REMOVE(&mountlist, me, me_list); 1598 mutex_exit(&mountlist_lock); 1599 mountlist_free(me); 1600 } 1601 1602 /* 1603 * Unlocked variant to traverse the mountlist. 1604 * To be used from DDB only. 1605 */ 1606 struct mount * 1607 _mountlist_next(struct mount *mp) 1608 { 1609 struct mountlist_entry *me; 1610 1611 if (mp == NULL) { 1612 me = TAILQ_FIRST(&mountlist); 1613 } else { 1614 TAILQ_FOREACH(me, &mountlist, me_list) 1615 if (me->me_type == ME_MOUNT && me->me_mount == mp) 1616 break; 1617 if (me != NULL) 1618 me = TAILQ_NEXT(me, me_list); 1619 } 1620 1621 while (me != NULL && me->me_type != ME_MOUNT) 1622 me = TAILQ_NEXT(me, me_list); 1623 1624 return (me ? me->me_mount : NULL); 1625 } 1626