1 /* $NetBSD: vfs_subr.c,v 1.395 2010/01/08 13:07:26 pooka Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008 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 /* 70 * Note on v_usecount and locking: 71 * 72 * At nearly all points it is known that v_usecount could be zero, the 73 * vnode interlock will be held. 74 * 75 * To change v_usecount away from zero, the interlock must be held. To 76 * change from a non-zero value to zero, again the interlock must be 77 * held. 78 * 79 * There's a flag bit, VC_XLOCK, embedded in v_usecount. 80 * To raise v_usecount, if the VC_XLOCK bit is set in it, the interlock 81 * must be held. 82 * To modify the VC_XLOCK bit, the interlock must be held. 83 * We always keep the usecount (v_usecount & VC_MASK) non-zero while the 84 * VC_XLOCK bit is set. 85 * 86 * Unless the VC_XLOCK bit is set, changing the usecount from a non-zero 87 * value to a non-zero value can safely be done using atomic operations, 88 * without the interlock held. 89 * Even if the VC_XLOCK bit is set, decreasing the usecount to a non-zero 90 * value can be done using atomic operations, without the interlock held. 91 */ 92 93 #include <sys/cdefs.h> 94 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.395 2010/01/08 13:07:26 pooka Exp $"); 95 96 #include "opt_ddb.h" 97 #include "opt_compat_netbsd.h" 98 #include "opt_compat_43.h" 99 100 #include <sys/param.h> 101 #include <sys/systm.h> 102 #include <sys/conf.h> 103 #include <sys/proc.h> 104 #include <sys/kernel.h> 105 #include <sys/mount.h> 106 #include <sys/fcntl.h> 107 #include <sys/vnode.h> 108 #include <sys/stat.h> 109 #include <sys/namei.h> 110 #include <sys/ucred.h> 111 #include <sys/buf.h> 112 #include <sys/errno.h> 113 #include <sys/kmem.h> 114 #include <sys/syscallargs.h> 115 #include <sys/device.h> 116 #include <sys/filedesc.h> 117 #include <sys/kauth.h> 118 #include <sys/atomic.h> 119 #include <sys/kthread.h> 120 #include <sys/wapbl.h> 121 122 #include <miscfs/genfs/genfs.h> 123 #include <miscfs/specfs/specdev.h> 124 #include <miscfs/syncfs/syncfs.h> 125 126 #include <uvm/uvm.h> 127 #include <uvm/uvm_readahead.h> 128 #include <uvm/uvm_ddb.h> 129 130 #include <sys/sysctl.h> 131 132 const enum vtype iftovt_tab[16] = { 133 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 134 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 135 }; 136 const int vttoif_tab[9] = { 137 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 138 S_IFSOCK, S_IFIFO, S_IFMT, 139 }; 140 141 /* 142 * Insq/Remq for the vnode usage lists. 143 */ 144 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 145 #define bufremvn(bp) { \ 146 LIST_REMOVE(bp, b_vnbufs); \ 147 (bp)->b_vnbufs.le_next = NOLIST; \ 148 } 149 150 int doforce = 1; /* 1 => permit forcible unmounting */ 151 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 152 153 static vnodelst_t vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list); 154 static vnodelst_t vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list); 155 static vnodelst_t vrele_list = TAILQ_HEAD_INITIALIZER(vrele_list); 156 157 struct mntlist mountlist = /* mounted filesystem list */ 158 CIRCLEQ_HEAD_INITIALIZER(mountlist); 159 160 u_int numvnodes; 161 static specificdata_domain_t mount_specificdata_domain; 162 163 static int vrele_pending; 164 static int vrele_gen; 165 static kmutex_t vrele_lock; 166 static kcondvar_t vrele_cv; 167 static lwp_t *vrele_lwp; 168 169 static uint64_t mountgen = 0; 170 static kmutex_t mountgen_lock; 171 172 kmutex_t mountlist_lock; 173 kmutex_t mntid_lock; 174 kmutex_t mntvnode_lock; 175 kmutex_t vnode_free_list_lock; 176 kmutex_t vfs_list_lock; 177 178 static pool_cache_t vnode_cache; 179 180 /* 181 * These define the root filesystem and device. 182 */ 183 struct vnode *rootvnode; 184 struct device *root_device; /* root device */ 185 186 /* 187 * Local declarations. 188 */ 189 190 static void vrele_thread(void *); 191 static void insmntque(vnode_t *, struct mount *); 192 static int getdevvp(dev_t, vnode_t **, enum vtype); 193 static vnode_t *getcleanvnode(void); 194 void vpanic(vnode_t *, const char *); 195 static void vfs_shutdown1(struct lwp *); 196 197 #ifdef DEBUG 198 void printlockedvnodes(void); 199 #endif 200 201 #ifdef DIAGNOSTIC 202 void 203 vpanic(vnode_t *vp, const char *msg) 204 { 205 206 vprint(NULL, vp); 207 panic("%s\n", msg); 208 } 209 #else 210 #define vpanic(vp, msg) /* nothing */ 211 #endif 212 213 void 214 vn_init1(void) 215 { 216 217 vnode_cache = pool_cache_init(sizeof(struct vnode), 0, 0, 0, "vnodepl", 218 NULL, IPL_NONE, NULL, NULL, NULL); 219 KASSERT(vnode_cache != NULL); 220 221 /* Create deferred release thread. */ 222 mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE); 223 cv_init(&vrele_cv, "vrele"); 224 if (kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread, 225 NULL, &vrele_lwp, "vrele")) 226 panic("fork vrele"); 227 } 228 229 /* 230 * Initialize the vnode management data structures. 231 */ 232 void 233 vntblinit(void) 234 { 235 236 mutex_init(&mountgen_lock, MUTEX_DEFAULT, IPL_NONE); 237 mutex_init(&mountlist_lock, MUTEX_DEFAULT, IPL_NONE); 238 mutex_init(&mntid_lock, MUTEX_DEFAULT, IPL_NONE); 239 mutex_init(&mntvnode_lock, MUTEX_DEFAULT, IPL_NONE); 240 mutex_init(&vnode_free_list_lock, MUTEX_DEFAULT, IPL_NONE); 241 mutex_init(&vfs_list_lock, MUTEX_DEFAULT, IPL_NONE); 242 243 mount_specificdata_domain = specificdata_domain_create(); 244 245 /* Initialize the filesystem syncer. */ 246 vn_initialize_syncerd(); 247 vn_init1(); 248 } 249 250 int 251 vfs_drainvnodes(long target, struct lwp *l) 252 { 253 254 while (numvnodes > target) { 255 vnode_t *vp; 256 257 mutex_enter(&vnode_free_list_lock); 258 vp = getcleanvnode(); 259 if (vp == NULL) 260 return EBUSY; /* give up */ 261 ungetnewvnode(vp); 262 } 263 264 return 0; 265 } 266 267 /* 268 * Lookup a mount point by filesystem identifier. 269 * 270 * XXX Needs to add a reference to the mount point. 271 */ 272 struct mount * 273 vfs_getvfs(fsid_t *fsid) 274 { 275 struct mount *mp; 276 277 mutex_enter(&mountlist_lock); 278 CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) { 279 if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] && 280 mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) { 281 mutex_exit(&mountlist_lock); 282 return (mp); 283 } 284 } 285 mutex_exit(&mountlist_lock); 286 return ((struct mount *)0); 287 } 288 289 /* 290 * Drop a reference to a mount structure, freeing if the last reference. 291 */ 292 void 293 vfs_destroy(struct mount *mp) 294 { 295 296 if (__predict_true((int)atomic_dec_uint_nv(&mp->mnt_refcnt) > 0)) { 297 return; 298 } 299 300 /* 301 * Nothing else has visibility of the mount: we can now 302 * free the data structures. 303 */ 304 KASSERT(mp->mnt_refcnt == 0); 305 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 306 rw_destroy(&mp->mnt_unmounting); 307 mutex_destroy(&mp->mnt_updating); 308 mutex_destroy(&mp->mnt_renamelock); 309 if (mp->mnt_op != NULL) { 310 vfs_delref(mp->mnt_op); 311 } 312 kmem_free(mp, sizeof(*mp)); 313 } 314 315 /* 316 * grab a vnode from freelist and clean it. 317 */ 318 vnode_t * 319 getcleanvnode(void) 320 { 321 vnode_t *vp; 322 vnodelst_t *listhd; 323 324 KASSERT(mutex_owned(&vnode_free_list_lock)); 325 326 retry: 327 listhd = &vnode_free_list; 328 try_nextlist: 329 TAILQ_FOREACH(vp, listhd, v_freelist) { 330 /* 331 * It's safe to test v_usecount and v_iflag 332 * without holding the interlock here, since 333 * these vnodes should never appear on the 334 * lists. 335 */ 336 if (vp->v_usecount != 0) { 337 vpanic(vp, "free vnode isn't"); 338 } 339 if ((vp->v_iflag & VI_CLEAN) != 0) { 340 vpanic(vp, "clean vnode on freelist"); 341 } 342 if (vp->v_freelisthd != listhd) { 343 printf("vnode sez %p, listhd %p\n", vp->v_freelisthd, listhd); 344 vpanic(vp, "list head mismatch"); 345 } 346 if (!mutex_tryenter(&vp->v_interlock)) 347 continue; 348 /* 349 * Our lwp might hold the underlying vnode 350 * locked, so don't try to reclaim a VI_LAYER 351 * node if it's locked. 352 */ 353 if ((vp->v_iflag & VI_XLOCK) == 0 && 354 ((vp->v_iflag & VI_LAYER) == 0 || VOP_ISLOCKED(vp) == 0)) { 355 break; 356 } 357 mutex_exit(&vp->v_interlock); 358 } 359 360 if (vp == NULL) { 361 if (listhd == &vnode_free_list) { 362 listhd = &vnode_hold_list; 363 goto try_nextlist; 364 } 365 mutex_exit(&vnode_free_list_lock); 366 return NULL; 367 } 368 369 /* Remove it from the freelist. */ 370 TAILQ_REMOVE(listhd, vp, v_freelist); 371 vp->v_freelisthd = NULL; 372 mutex_exit(&vnode_free_list_lock); 373 374 if (vp->v_usecount != 0) { 375 /* 376 * was referenced again before we got the interlock 377 * Don't return to freelist - the holder of the last 378 * reference will destroy it. 379 */ 380 mutex_exit(&vp->v_interlock); 381 mutex_enter(&vnode_free_list_lock); 382 goto retry; 383 } 384 385 /* 386 * The vnode is still associated with a file system, so we must 387 * clean it out before reusing it. We need to add a reference 388 * before doing this. If the vnode gains another reference while 389 * being cleaned out then we lose - retry. 390 */ 391 atomic_add_int(&vp->v_usecount, 1 + VC_XLOCK); 392 vclean(vp, DOCLOSE); 393 KASSERT(vp->v_usecount >= 1 + VC_XLOCK); 394 atomic_add_int(&vp->v_usecount, -VC_XLOCK); 395 if (vp->v_usecount == 1) { 396 /* We're about to dirty it. */ 397 vp->v_iflag &= ~VI_CLEAN; 398 mutex_exit(&vp->v_interlock); 399 if (vp->v_type == VBLK || vp->v_type == VCHR) { 400 spec_node_destroy(vp); 401 } 402 vp->v_type = VNON; 403 } else { 404 /* 405 * Don't return to freelist - the holder of the last 406 * reference will destroy it. 407 */ 408 vrelel(vp, 0); /* releases vp->v_interlock */ 409 mutex_enter(&vnode_free_list_lock); 410 goto retry; 411 } 412 413 if (vp->v_data != NULL || vp->v_uobj.uo_npages != 0 || 414 !TAILQ_EMPTY(&vp->v_uobj.memq)) { 415 vpanic(vp, "cleaned vnode isn't"); 416 } 417 if (vp->v_numoutput != 0) { 418 vpanic(vp, "clean vnode has pending I/O's"); 419 } 420 if ((vp->v_iflag & VI_ONWORKLST) != 0) { 421 vpanic(vp, "clean vnode on syncer list"); 422 } 423 424 return vp; 425 } 426 427 /* 428 * Mark a mount point as busy, and gain a new reference to it. Used to 429 * prevent the file system from being unmounted during critical sections. 430 * 431 * => The caller must hold a pre-existing reference to the mount. 432 * => Will fail if the file system is being unmounted, or is unmounted. 433 */ 434 int 435 vfs_busy(struct mount *mp, struct mount **nextp) 436 { 437 438 KASSERT(mp->mnt_refcnt > 0); 439 440 if (__predict_false(!rw_tryenter(&mp->mnt_unmounting, RW_READER))) { 441 if (nextp != NULL) { 442 KASSERT(mutex_owned(&mountlist_lock)); 443 *nextp = CIRCLEQ_NEXT(mp, mnt_list); 444 } 445 return EBUSY; 446 } 447 if (__predict_false((mp->mnt_iflag & IMNT_GONE) != 0)) { 448 rw_exit(&mp->mnt_unmounting); 449 if (nextp != NULL) { 450 KASSERT(mutex_owned(&mountlist_lock)); 451 *nextp = CIRCLEQ_NEXT(mp, mnt_list); 452 } 453 return ENOENT; 454 } 455 if (nextp != NULL) { 456 mutex_exit(&mountlist_lock); 457 } 458 atomic_inc_uint(&mp->mnt_refcnt); 459 return 0; 460 } 461 462 /* 463 * Unbusy a busy filesystem. 464 * 465 * => If keepref is true, preserve reference added by vfs_busy(). 466 * => If nextp != NULL, acquire mountlist_lock. 467 */ 468 void 469 vfs_unbusy(struct mount *mp, bool keepref, struct mount **nextp) 470 { 471 472 KASSERT(mp->mnt_refcnt > 0); 473 474 if (nextp != NULL) { 475 mutex_enter(&mountlist_lock); 476 } 477 rw_exit(&mp->mnt_unmounting); 478 if (!keepref) { 479 vfs_destroy(mp); 480 } 481 if (nextp != NULL) { 482 KASSERT(mutex_owned(&mountlist_lock)); 483 *nextp = CIRCLEQ_NEXT(mp, mnt_list); 484 } 485 } 486 487 struct mount * 488 vfs_mountalloc(struct vfsops *vfsops, struct vnode *vp) 489 { 490 int error; 491 struct mount *mp; 492 493 mp = kmem_zalloc(sizeof(*mp), KM_SLEEP); 494 if (mp == NULL) 495 return NULL; 496 497 mp->mnt_op = vfsops; 498 mp->mnt_refcnt = 1; 499 TAILQ_INIT(&mp->mnt_vnodelist); 500 rw_init(&mp->mnt_unmounting); 501 mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE); 502 mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE); 503 error = vfs_busy(mp, NULL); 504 KASSERT(error == 0); 505 mp->mnt_vnodecovered = vp; 506 mount_initspecific(mp); 507 508 mutex_enter(&mountgen_lock); 509 mp->mnt_gen = mountgen++; 510 mutex_exit(&mountgen_lock); 511 512 return mp; 513 } 514 515 /* 516 * Lookup a filesystem type, and if found allocate and initialize 517 * a mount structure for it. 518 * 519 * Devname is usually updated by mount(8) after booting. 520 */ 521 int 522 vfs_rootmountalloc(const char *fstypename, const char *devname, 523 struct mount **mpp) 524 { 525 struct vfsops *vfsp = NULL; 526 struct mount *mp; 527 528 mutex_enter(&vfs_list_lock); 529 LIST_FOREACH(vfsp, &vfs_list, vfs_list) 530 if (!strncmp(vfsp->vfs_name, fstypename, 531 sizeof(mp->mnt_stat.f_fstypename))) 532 break; 533 if (vfsp == NULL) { 534 mutex_exit(&vfs_list_lock); 535 return (ENODEV); 536 } 537 vfsp->vfs_refcount++; 538 mutex_exit(&vfs_list_lock); 539 540 if ((mp = vfs_mountalloc(vfsp, NULL)) == NULL) 541 return ENOMEM; 542 mp->mnt_flag = MNT_RDONLY; 543 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, 544 sizeof(mp->mnt_stat.f_fstypename)); 545 mp->mnt_stat.f_mntonname[0] = '/'; 546 mp->mnt_stat.f_mntonname[1] = '\0'; 547 mp->mnt_stat.f_mntfromname[sizeof(mp->mnt_stat.f_mntfromname) - 1] = 548 '\0'; 549 (void)copystr(devname, mp->mnt_stat.f_mntfromname, 550 sizeof(mp->mnt_stat.f_mntfromname) - 1, 0); 551 *mpp = mp; 552 return (0); 553 } 554 555 /* 556 * Routines having to do with the management of the vnode table. 557 */ 558 extern int (**dead_vnodeop_p)(void *); 559 560 /* 561 * Return the next vnode from the free list. 562 */ 563 int 564 getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *), 565 vnode_t **vpp) 566 { 567 struct uvm_object *uobj; 568 static int toggle; 569 vnode_t *vp; 570 int error = 0, tryalloc; 571 572 try_again: 573 if (mp != NULL) { 574 /* 575 * Mark filesystem busy while we're creating a 576 * vnode. If unmount is in progress, this will 577 * fail. 578 */ 579 error = vfs_busy(mp, NULL); 580 if (error) 581 return error; 582 } 583 584 /* 585 * We must choose whether to allocate a new vnode or recycle an 586 * existing one. The criterion for allocating a new one is that 587 * the total number of vnodes is less than the number desired or 588 * there are no vnodes on either free list. Generally we only 589 * want to recycle vnodes that have no buffers associated with 590 * them, so we look first on the vnode_free_list. If it is empty, 591 * we next consider vnodes with referencing buffers on the 592 * vnode_hold_list. The toggle ensures that half the time we 593 * will use a buffer from the vnode_hold_list, and half the time 594 * we will allocate a new one unless the list has grown to twice 595 * the desired size. We are reticent to recycle vnodes from the 596 * vnode_hold_list because we will lose the identity of all its 597 * referencing buffers. 598 */ 599 600 vp = NULL; 601 602 mutex_enter(&vnode_free_list_lock); 603 604 toggle ^= 1; 605 if (numvnodes > 2 * desiredvnodes) 606 toggle = 0; 607 608 tryalloc = numvnodes < desiredvnodes || 609 (TAILQ_FIRST(&vnode_free_list) == NULL && 610 (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle)); 611 612 if (tryalloc) { 613 numvnodes++; 614 mutex_exit(&vnode_free_list_lock); 615 if ((vp = vnalloc(NULL)) == NULL) { 616 mutex_enter(&vnode_free_list_lock); 617 numvnodes--; 618 } else 619 vp->v_usecount = 1; 620 } 621 622 if (vp == NULL) { 623 vp = getcleanvnode(); 624 if (vp == NULL) { 625 if (mp != NULL) { 626 vfs_unbusy(mp, false, NULL); 627 } 628 if (tryalloc) { 629 printf("WARNING: unable to allocate new " 630 "vnode, retrying...\n"); 631 kpause("newvn", false, hz, NULL); 632 goto try_again; 633 } 634 tablefull("vnode", "increase kern.maxvnodes or NVNODE"); 635 *vpp = 0; 636 return (ENFILE); 637 } 638 vp->v_iflag = 0; 639 vp->v_vflag = 0; 640 vp->v_uflag = 0; 641 vp->v_socket = NULL; 642 } 643 644 KASSERT(vp->v_usecount == 1); 645 KASSERT(vp->v_freelisthd == NULL); 646 KASSERT(LIST_EMPTY(&vp->v_nclist)); 647 KASSERT(LIST_EMPTY(&vp->v_dnclist)); 648 649 vp->v_type = VNON; 650 vp->v_vnlock = &vp->v_lock; 651 vp->v_tag = tag; 652 vp->v_op = vops; 653 insmntque(vp, mp); 654 *vpp = vp; 655 vp->v_data = 0; 656 657 /* 658 * initialize uvm_object within vnode. 659 */ 660 661 uobj = &vp->v_uobj; 662 KASSERT(uobj->pgops == &uvm_vnodeops); 663 KASSERT(uobj->uo_npages == 0); 664 KASSERT(TAILQ_FIRST(&uobj->memq) == NULL); 665 vp->v_size = vp->v_writesize = VSIZENOTSET; 666 667 if (mp != NULL) { 668 if ((mp->mnt_iflag & IMNT_MPSAFE) != 0) 669 vp->v_vflag |= VV_MPSAFE; 670 vfs_unbusy(mp, true, NULL); 671 } 672 673 return (0); 674 } 675 676 /* 677 * This is really just the reverse of getnewvnode(). Needed for 678 * VFS_VGET functions who may need to push back a vnode in case 679 * of a locking race. 680 */ 681 void 682 ungetnewvnode(vnode_t *vp) 683 { 684 685 KASSERT(vp->v_usecount == 1); 686 KASSERT(vp->v_data == NULL); 687 KASSERT(vp->v_freelisthd == NULL); 688 689 mutex_enter(&vp->v_interlock); 690 vp->v_iflag |= VI_CLEAN; 691 vrelel(vp, 0); 692 } 693 694 /* 695 * Allocate a new, uninitialized vnode. If 'mp' is non-NULL, this is a 696 * marker vnode and we are prepared to wait for the allocation. 697 */ 698 vnode_t * 699 vnalloc(struct mount *mp) 700 { 701 vnode_t *vp; 702 703 vp = pool_cache_get(vnode_cache, (mp != NULL ? PR_WAITOK : PR_NOWAIT)); 704 if (vp == NULL) { 705 return NULL; 706 } 707 708 memset(vp, 0, sizeof(*vp)); 709 UVM_OBJ_INIT(&vp->v_uobj, &uvm_vnodeops, 0); 710 cv_init(&vp->v_cv, "vnode"); 711 /* 712 * done by memset() above. 713 * LIST_INIT(&vp->v_nclist); 714 * LIST_INIT(&vp->v_dnclist); 715 */ 716 717 if (mp != NULL) { 718 vp->v_mount = mp; 719 vp->v_type = VBAD; 720 vp->v_iflag = VI_MARKER; 721 } else { 722 rw_init(&vp->v_lock.vl_lock); 723 } 724 725 return vp; 726 } 727 728 /* 729 * Free an unused, unreferenced vnode. 730 */ 731 void 732 vnfree(vnode_t *vp) 733 { 734 735 KASSERT(vp->v_usecount == 0); 736 737 if ((vp->v_iflag & VI_MARKER) == 0) { 738 rw_destroy(&vp->v_lock.vl_lock); 739 mutex_enter(&vnode_free_list_lock); 740 numvnodes--; 741 mutex_exit(&vnode_free_list_lock); 742 } 743 744 UVM_OBJ_DESTROY(&vp->v_uobj); 745 cv_destroy(&vp->v_cv); 746 pool_cache_put(vnode_cache, vp); 747 } 748 749 /* 750 * Remove a vnode from its freelist. 751 */ 752 static inline void 753 vremfree(vnode_t *vp) 754 { 755 756 KASSERT(mutex_owned(&vp->v_interlock)); 757 KASSERT(vp->v_usecount == 0); 758 759 /* 760 * Note that the reference count must not change until 761 * the vnode is removed. 762 */ 763 mutex_enter(&vnode_free_list_lock); 764 if (vp->v_holdcnt > 0) { 765 KASSERT(vp->v_freelisthd == &vnode_hold_list); 766 } else { 767 KASSERT(vp->v_freelisthd == &vnode_free_list); 768 } 769 TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 770 vp->v_freelisthd = NULL; 771 mutex_exit(&vnode_free_list_lock); 772 } 773 774 /* 775 * Move a vnode from one mount queue to another. 776 */ 777 static void 778 insmntque(vnode_t *vp, struct mount *mp) 779 { 780 struct mount *omp; 781 782 #ifdef DIAGNOSTIC 783 if ((mp != NULL) && 784 (mp->mnt_iflag & IMNT_UNMOUNT) && 785 vp->v_tag != VT_VFS) { 786 panic("insmntque into dying filesystem"); 787 } 788 #endif 789 790 mutex_enter(&mntvnode_lock); 791 /* 792 * Delete from old mount point vnode list, if on one. 793 */ 794 if ((omp = vp->v_mount) != NULL) 795 TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes); 796 /* 797 * Insert into list of vnodes for the new mount point, if 798 * available. The caller must take a reference on the mount 799 * structure and donate to the vnode. 800 */ 801 if ((vp->v_mount = mp) != NULL) 802 TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes); 803 mutex_exit(&mntvnode_lock); 804 805 if (omp != NULL) { 806 /* Release reference to old mount. */ 807 vfs_destroy(omp); 808 } 809 } 810 811 /* 812 * Wait for a vnode (typically with VI_XLOCK set) to be cleaned or 813 * recycled. 814 */ 815 void 816 vwait(vnode_t *vp, int flags) 817 { 818 819 KASSERT(mutex_owned(&vp->v_interlock)); 820 KASSERT(vp->v_usecount != 0); 821 822 while ((vp->v_iflag & flags) != 0) 823 cv_wait(&vp->v_cv, &vp->v_interlock); 824 } 825 826 /* 827 * Insert a marker vnode into a mount's vnode list, after the 828 * specified vnode. mntvnode_lock must be held. 829 */ 830 void 831 vmark(vnode_t *mvp, vnode_t *vp) 832 { 833 struct mount *mp; 834 835 mp = mvp->v_mount; 836 837 KASSERT(mutex_owned(&mntvnode_lock)); 838 KASSERT((mvp->v_iflag & VI_MARKER) != 0); 839 KASSERT(vp->v_mount == mp); 840 841 TAILQ_INSERT_AFTER(&mp->mnt_vnodelist, vp, mvp, v_mntvnodes); 842 } 843 844 /* 845 * Remove a marker vnode from a mount's vnode list, and return 846 * a pointer to the next vnode in the list. mntvnode_lock must 847 * be held. 848 */ 849 vnode_t * 850 vunmark(vnode_t *mvp) 851 { 852 vnode_t *vp; 853 struct mount *mp; 854 855 mp = mvp->v_mount; 856 857 KASSERT(mutex_owned(&mntvnode_lock)); 858 KASSERT((mvp->v_iflag & VI_MARKER) != 0); 859 860 vp = TAILQ_NEXT(mvp, v_mntvnodes); 861 TAILQ_REMOVE(&mp->mnt_vnodelist, mvp, v_mntvnodes); 862 863 KASSERT(vp == NULL || vp->v_mount == mp); 864 865 return vp; 866 } 867 868 /* 869 * Update outstanding I/O count and do wakeup if requested. 870 */ 871 void 872 vwakeup(struct buf *bp) 873 { 874 struct vnode *vp; 875 876 if ((vp = bp->b_vp) == NULL) 877 return; 878 879 KASSERT(bp->b_objlock == &vp->v_interlock); 880 KASSERT(mutex_owned(bp->b_objlock)); 881 882 if (--vp->v_numoutput < 0) 883 panic("vwakeup: neg numoutput, vp %p", vp); 884 if (vp->v_numoutput == 0) 885 cv_broadcast(&vp->v_cv); 886 } 887 888 /* 889 * Flush out and invalidate all buffers associated with a vnode. 890 * Called with the underlying vnode locked, which should prevent new dirty 891 * buffers from being queued. 892 */ 893 int 894 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l, 895 bool catch, int slptimeo) 896 { 897 struct buf *bp, *nbp; 898 int error; 899 int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO | 900 (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0); 901 902 /* XXXUBC this doesn't look at flags or slp* */ 903 mutex_enter(&vp->v_interlock); 904 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 905 if (error) { 906 return error; 907 } 908 909 if (flags & V_SAVE) { 910 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0); 911 if (error) 912 return (error); 913 KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd)); 914 } 915 916 mutex_enter(&bufcache_lock); 917 restart: 918 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 919 nbp = LIST_NEXT(bp, b_vnbufs); 920 error = bbusy(bp, catch, slptimeo, NULL); 921 if (error != 0) { 922 if (error == EPASSTHROUGH) 923 goto restart; 924 mutex_exit(&bufcache_lock); 925 return (error); 926 } 927 brelsel(bp, BC_INVAL | BC_VFLUSH); 928 } 929 930 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 931 nbp = LIST_NEXT(bp, b_vnbufs); 932 error = bbusy(bp, catch, slptimeo, NULL); 933 if (error != 0) { 934 if (error == EPASSTHROUGH) 935 goto restart; 936 mutex_exit(&bufcache_lock); 937 return (error); 938 } 939 /* 940 * XXX Since there are no node locks for NFS, I believe 941 * there is a slight chance that a delayed write will 942 * occur while sleeping just above, so check for it. 943 */ 944 if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) { 945 #ifdef DEBUG 946 printf("buffer still DELWRI\n"); 947 #endif 948 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 949 mutex_exit(&bufcache_lock); 950 VOP_BWRITE(bp); 951 mutex_enter(&bufcache_lock); 952 goto restart; 953 } 954 brelsel(bp, BC_INVAL | BC_VFLUSH); 955 } 956 957 #ifdef DIAGNOSTIC 958 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 959 panic("vinvalbuf: flush failed, vp %p", vp); 960 #endif 961 962 mutex_exit(&bufcache_lock); 963 964 return (0); 965 } 966 967 /* 968 * Destroy any in core blocks past the truncation length. 969 * Called with the underlying vnode locked, which should prevent new dirty 970 * buffers from being queued. 971 */ 972 int 973 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch, int slptimeo) 974 { 975 struct buf *bp, *nbp; 976 int error; 977 voff_t off; 978 979 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 980 mutex_enter(&vp->v_interlock); 981 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 982 if (error) { 983 return error; 984 } 985 986 mutex_enter(&bufcache_lock); 987 restart: 988 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 989 nbp = LIST_NEXT(bp, b_vnbufs); 990 if (bp->b_lblkno < lbn) 991 continue; 992 error = bbusy(bp, catch, slptimeo, NULL); 993 if (error != 0) { 994 if (error == EPASSTHROUGH) 995 goto restart; 996 mutex_exit(&bufcache_lock); 997 return (error); 998 } 999 brelsel(bp, BC_INVAL | BC_VFLUSH); 1000 } 1001 1002 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 1003 nbp = LIST_NEXT(bp, b_vnbufs); 1004 if (bp->b_lblkno < lbn) 1005 continue; 1006 error = bbusy(bp, catch, slptimeo, NULL); 1007 if (error != 0) { 1008 if (error == EPASSTHROUGH) 1009 goto restart; 1010 mutex_exit(&bufcache_lock); 1011 return (error); 1012 } 1013 brelsel(bp, BC_INVAL | BC_VFLUSH); 1014 } 1015 mutex_exit(&bufcache_lock); 1016 1017 return (0); 1018 } 1019 1020 /* 1021 * Flush all dirty buffers from a vnode. 1022 * Called with the underlying vnode locked, which should prevent new dirty 1023 * buffers from being queued. 1024 */ 1025 void 1026 vflushbuf(struct vnode *vp, int sync) 1027 { 1028 struct buf *bp, *nbp; 1029 int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0); 1030 bool dirty; 1031 1032 mutex_enter(&vp->v_interlock); 1033 (void) VOP_PUTPAGES(vp, 0, 0, flags); 1034 1035 loop: 1036 mutex_enter(&bufcache_lock); 1037 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1038 nbp = LIST_NEXT(bp, b_vnbufs); 1039 if ((bp->b_cflags & BC_BUSY)) 1040 continue; 1041 if ((bp->b_oflags & BO_DELWRI) == 0) 1042 panic("vflushbuf: not dirty, bp %p", bp); 1043 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 1044 mutex_exit(&bufcache_lock); 1045 /* 1046 * Wait for I/O associated with indirect blocks to complete, 1047 * since there is no way to quickly wait for them below. 1048 */ 1049 if (bp->b_vp == vp || sync == 0) 1050 (void) bawrite(bp); 1051 else 1052 (void) bwrite(bp); 1053 goto loop; 1054 } 1055 mutex_exit(&bufcache_lock); 1056 1057 if (sync == 0) 1058 return; 1059 1060 mutex_enter(&vp->v_interlock); 1061 while (vp->v_numoutput != 0) 1062 cv_wait(&vp->v_cv, &vp->v_interlock); 1063 dirty = !LIST_EMPTY(&vp->v_dirtyblkhd); 1064 mutex_exit(&vp->v_interlock); 1065 1066 if (dirty) { 1067 vprint("vflushbuf: dirty", vp); 1068 goto loop; 1069 } 1070 } 1071 1072 /* 1073 * Create a vnode for a block device. 1074 * Used for root filesystem and swap areas. 1075 * Also used for memory file system special devices. 1076 */ 1077 int 1078 bdevvp(dev_t dev, vnode_t **vpp) 1079 { 1080 1081 return (getdevvp(dev, vpp, VBLK)); 1082 } 1083 1084 /* 1085 * Create a vnode for a character device. 1086 * Used for kernfs and some console handling. 1087 */ 1088 int 1089 cdevvp(dev_t dev, vnode_t **vpp) 1090 { 1091 1092 return (getdevvp(dev, vpp, VCHR)); 1093 } 1094 1095 /* 1096 * Associate a buffer with a vnode. There must already be a hold on 1097 * the vnode. 1098 */ 1099 void 1100 bgetvp(struct vnode *vp, struct buf *bp) 1101 { 1102 1103 KASSERT(bp->b_vp == NULL); 1104 KASSERT(bp->b_objlock == &buffer_lock); 1105 KASSERT(mutex_owned(&vp->v_interlock)); 1106 KASSERT(mutex_owned(&bufcache_lock)); 1107 KASSERT((bp->b_cflags & BC_BUSY) != 0); 1108 KASSERT(!cv_has_waiters(&bp->b_done)); 1109 1110 vholdl(vp); 1111 bp->b_vp = vp; 1112 if (vp->v_type == VBLK || vp->v_type == VCHR) 1113 bp->b_dev = vp->v_rdev; 1114 else 1115 bp->b_dev = NODEV; 1116 1117 /* 1118 * Insert onto list for new vnode. 1119 */ 1120 bufinsvn(bp, &vp->v_cleanblkhd); 1121 bp->b_objlock = &vp->v_interlock; 1122 } 1123 1124 /* 1125 * Disassociate a buffer from a vnode. 1126 */ 1127 void 1128 brelvp(struct buf *bp) 1129 { 1130 struct vnode *vp = bp->b_vp; 1131 1132 KASSERT(vp != NULL); 1133 KASSERT(bp->b_objlock == &vp->v_interlock); 1134 KASSERT(mutex_owned(&vp->v_interlock)); 1135 KASSERT(mutex_owned(&bufcache_lock)); 1136 KASSERT((bp->b_cflags & BC_BUSY) != 0); 1137 KASSERT(!cv_has_waiters(&bp->b_done)); 1138 1139 /* 1140 * Delete from old vnode list, if on one. 1141 */ 1142 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 1143 bufremvn(bp); 1144 1145 if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_iflag & VI_ONWORKLST) && 1146 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1147 vp->v_iflag &= ~VI_WRMAPDIRTY; 1148 vn_syncer_remove_from_worklist(vp); 1149 } 1150 1151 bp->b_objlock = &buffer_lock; 1152 bp->b_vp = NULL; 1153 holdrelel(vp); 1154 } 1155 1156 /* 1157 * Reassign a buffer from one vnode list to another. 1158 * The list reassignment must be within the same vnode. 1159 * Used to assign file specific control information 1160 * (indirect blocks) to the list to which they belong. 1161 */ 1162 void 1163 reassignbuf(struct buf *bp, struct vnode *vp) 1164 { 1165 struct buflists *listheadp; 1166 int delayx; 1167 1168 KASSERT(mutex_owned(&bufcache_lock)); 1169 KASSERT(bp->b_objlock == &vp->v_interlock); 1170 KASSERT(mutex_owned(&vp->v_interlock)); 1171 KASSERT((bp->b_cflags & BC_BUSY) != 0); 1172 1173 /* 1174 * Delete from old vnode list, if on one. 1175 */ 1176 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 1177 bufremvn(bp); 1178 1179 /* 1180 * If dirty, put on list of dirty buffers; 1181 * otherwise insert onto list of clean buffers. 1182 */ 1183 if ((bp->b_oflags & BO_DELWRI) == 0) { 1184 listheadp = &vp->v_cleanblkhd; 1185 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 1186 (vp->v_iflag & VI_ONWORKLST) && 1187 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 1188 vp->v_iflag &= ~VI_WRMAPDIRTY; 1189 vn_syncer_remove_from_worklist(vp); 1190 } 1191 } else { 1192 listheadp = &vp->v_dirtyblkhd; 1193 if ((vp->v_iflag & VI_ONWORKLST) == 0) { 1194 switch (vp->v_type) { 1195 case VDIR: 1196 delayx = dirdelay; 1197 break; 1198 case VBLK: 1199 if (vp->v_specmountpoint != NULL) { 1200 delayx = metadelay; 1201 break; 1202 } 1203 /* fall through */ 1204 default: 1205 delayx = filedelay; 1206 break; 1207 } 1208 if (!vp->v_mount || 1209 (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) 1210 vn_syncer_add_to_worklist(vp, delayx); 1211 } 1212 } 1213 bufinsvn(bp, listheadp); 1214 } 1215 1216 /* 1217 * Create a vnode for a device. 1218 * Used by bdevvp (block device) for root file system etc., 1219 * and by cdevvp (character device) for console and kernfs. 1220 */ 1221 static int 1222 getdevvp(dev_t dev, vnode_t **vpp, enum vtype type) 1223 { 1224 vnode_t *vp; 1225 vnode_t *nvp; 1226 int error; 1227 1228 if (dev == NODEV) { 1229 *vpp = NULL; 1230 return (0); 1231 } 1232 error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp); 1233 if (error) { 1234 *vpp = NULL; 1235 return (error); 1236 } 1237 vp = nvp; 1238 vp->v_type = type; 1239 vp->v_vflag |= VV_MPSAFE; 1240 uvm_vnp_setsize(vp, 0); 1241 spec_node_init(vp, dev); 1242 *vpp = vp; 1243 return (0); 1244 } 1245 1246 /* 1247 * Try to gain a reference to a vnode, without acquiring its interlock. 1248 * The caller must hold a lock that will prevent the vnode from being 1249 * recycled or freed. 1250 */ 1251 bool 1252 vtryget(vnode_t *vp) 1253 { 1254 u_int use, next; 1255 1256 /* 1257 * If the vnode is being freed, don't make life any harder 1258 * for vclean() by adding another reference without waiting. 1259 * This is not strictly necessary, but we'll do it anyway. 1260 */ 1261 if (__predict_false((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0)) { 1262 return false; 1263 } 1264 for (use = vp->v_usecount;; use = next) { 1265 if (use == 0 || __predict_false((use & VC_XLOCK) != 0)) { 1266 /* Need interlock held if first reference. */ 1267 return false; 1268 } 1269 next = atomic_cas_uint(&vp->v_usecount, use, use + 1); 1270 if (__predict_true(next == use)) { 1271 return true; 1272 } 1273 } 1274 } 1275 1276 /* 1277 * Grab a particular vnode from the free list, increment its 1278 * reference count and lock it. If the vnode lock bit is set the 1279 * vnode is being eliminated in vgone. In that case, we can not 1280 * grab the vnode, so the process is awakened when the transition is 1281 * completed, and an error returned to indicate that the vnode is no 1282 * longer usable (possibly having been changed to a new file system type). 1283 */ 1284 int 1285 vget(vnode_t *vp, int flags) 1286 { 1287 int error; 1288 1289 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1290 1291 if ((flags & LK_INTERLOCK) == 0) 1292 mutex_enter(&vp->v_interlock); 1293 1294 /* 1295 * Before adding a reference, we must remove the vnode 1296 * from its freelist. 1297 */ 1298 if (vp->v_usecount == 0) { 1299 vremfree(vp); 1300 vp->v_usecount = 1; 1301 } else { 1302 atomic_inc_uint(&vp->v_usecount); 1303 } 1304 1305 /* 1306 * If the vnode is in the process of being cleaned out for 1307 * another use, we wait for the cleaning to finish and then 1308 * return failure. Cleaning is determined by checking if 1309 * the VI_XLOCK or VI_FREEING flags are set. 1310 */ 1311 if ((vp->v_iflag & (VI_XLOCK | VI_FREEING)) != 0) { 1312 if ((flags & LK_NOWAIT) != 0) { 1313 vrelel(vp, 0); 1314 return EBUSY; 1315 } 1316 vwait(vp, VI_XLOCK | VI_FREEING); 1317 vrelel(vp, 0); 1318 return ENOENT; 1319 } 1320 1321 if ((vp->v_iflag & VI_INACTNOW) != 0) { 1322 /* 1323 * if it's being desactived, wait for it to complete. 1324 * Make sure to not return a clean vnode. 1325 */ 1326 if ((flags & LK_NOWAIT) != 0) { 1327 vrelel(vp, 0); 1328 return EBUSY; 1329 } 1330 vwait(vp, VI_INACTNOW); 1331 if ((vp->v_iflag & VI_CLEAN) != 0) { 1332 vrelel(vp, 0); 1333 return ENOENT; 1334 } 1335 } 1336 if (flags & LK_TYPE_MASK) { 1337 error = vn_lock(vp, flags | LK_INTERLOCK); 1338 if (error != 0) { 1339 vrele(vp); 1340 } 1341 return error; 1342 } 1343 mutex_exit(&vp->v_interlock); 1344 return 0; 1345 } 1346 1347 /* 1348 * vput(), just unlock and vrele() 1349 */ 1350 void 1351 vput(vnode_t *vp) 1352 { 1353 1354 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1355 1356 VOP_UNLOCK(vp, 0); 1357 vrele(vp); 1358 } 1359 1360 /* 1361 * Try to drop reference on a vnode. Abort if we are releasing the 1362 * last reference. Note: this _must_ succeed if not the last reference. 1363 */ 1364 static inline bool 1365 vtryrele(vnode_t *vp) 1366 { 1367 u_int use, next; 1368 1369 for (use = vp->v_usecount;; use = next) { 1370 if (use == 1) { 1371 return false; 1372 } 1373 KASSERT((use & VC_MASK) > 1); 1374 next = atomic_cas_uint(&vp->v_usecount, use, use - 1); 1375 if (__predict_true(next == use)) { 1376 return true; 1377 } 1378 } 1379 } 1380 1381 /* 1382 * Vnode release. If reference count drops to zero, call inactive 1383 * routine and either return to freelist or free to the pool. 1384 */ 1385 void 1386 vrelel(vnode_t *vp, int flags) 1387 { 1388 bool recycle, defer; 1389 int error; 1390 1391 KASSERT(mutex_owned(&vp->v_interlock)); 1392 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1393 KASSERT(vp->v_freelisthd == NULL); 1394 1395 if (__predict_false(vp->v_op == dead_vnodeop_p && 1396 (vp->v_iflag & (VI_CLEAN|VI_XLOCK)) == 0)) { 1397 vpanic(vp, "dead but not clean"); 1398 } 1399 1400 /* 1401 * If not the last reference, just drop the reference count 1402 * and unlock. 1403 */ 1404 if (vtryrele(vp)) { 1405 vp->v_iflag |= VI_INACTREDO; 1406 mutex_exit(&vp->v_interlock); 1407 return; 1408 } 1409 if (vp->v_usecount <= 0 || vp->v_writecount != 0) { 1410 vpanic(vp, "vrelel: bad ref count"); 1411 } 1412 1413 KASSERT((vp->v_iflag & VI_XLOCK) == 0); 1414 1415 /* 1416 * If not clean, deactivate the vnode, but preserve 1417 * our reference across the call to VOP_INACTIVE(). 1418 */ 1419 retry: 1420 if ((vp->v_iflag & VI_CLEAN) == 0) { 1421 recycle = false; 1422 vp->v_iflag |= VI_INACTNOW; 1423 1424 /* 1425 * XXX This ugly block can be largely eliminated if 1426 * locking is pushed down into the file systems. 1427 */ 1428 if (curlwp == uvm.pagedaemon_lwp) { 1429 /* The pagedaemon can't wait around; defer. */ 1430 defer = true; 1431 } else if (curlwp == vrele_lwp) { 1432 /* 1433 * We have to try harder. But we can't sleep 1434 * with VI_INACTNOW as vget() may be waiting on it. 1435 */ 1436 vp->v_iflag &= ~(VI_INACTREDO|VI_INACTNOW); 1437 cv_broadcast(&vp->v_cv); 1438 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | 1439 LK_RETRY); 1440 if (error != 0) { 1441 /* XXX */ 1442 vpanic(vp, "vrele: unable to lock %p"); 1443 } 1444 mutex_enter(&vp->v_interlock); 1445 /* 1446 * if we did get another reference while 1447 * sleeping, don't try to inactivate it yet. 1448 */ 1449 if (__predict_false(vtryrele(vp))) { 1450 VOP_UNLOCK(vp, 0); 1451 mutex_exit(&vp->v_interlock); 1452 return; 1453 } 1454 vp->v_iflag |= VI_INACTNOW; 1455 mutex_exit(&vp->v_interlock); 1456 defer = false; 1457 } else if ((vp->v_iflag & VI_LAYER) != 0) { 1458 /* 1459 * Acquiring the stack's lock in vclean() even 1460 * for an honest vput/vrele is dangerous because 1461 * our caller may hold other vnode locks; defer. 1462 */ 1463 defer = true; 1464 } else { 1465 /* If we can't acquire the lock, then defer. */ 1466 vp->v_iflag &= ~VI_INACTREDO; 1467 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | 1468 LK_NOWAIT); 1469 if (error != 0) { 1470 defer = true; 1471 mutex_enter(&vp->v_interlock); 1472 } else { 1473 defer = false; 1474 } 1475 } 1476 1477 if (defer) { 1478 /* 1479 * Defer reclaim to the kthread; it's not safe to 1480 * clean it here. We donate it our last reference. 1481 */ 1482 KASSERT(mutex_owned(&vp->v_interlock)); 1483 KASSERT((vp->v_iflag & VI_INACTPEND) == 0); 1484 vp->v_iflag &= ~VI_INACTNOW; 1485 vp->v_iflag |= VI_INACTPEND; 1486 mutex_enter(&vrele_lock); 1487 TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist); 1488 if (++vrele_pending > (desiredvnodes >> 8)) 1489 cv_signal(&vrele_cv); 1490 mutex_exit(&vrele_lock); 1491 cv_broadcast(&vp->v_cv); 1492 mutex_exit(&vp->v_interlock); 1493 return; 1494 } 1495 1496 #ifdef DIAGNOSTIC 1497 if ((vp->v_type == VBLK || vp->v_type == VCHR) && 1498 vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) { 1499 vprint("vrelel: missing VOP_CLOSE()", vp); 1500 } 1501 #endif 1502 1503 /* 1504 * The vnode can gain another reference while being 1505 * deactivated. If VOP_INACTIVE() indicates that 1506 * the described file has been deleted, then recycle 1507 * the vnode irrespective of additional references. 1508 * Another thread may be waiting to re-use the on-disk 1509 * inode. 1510 * 1511 * Note that VOP_INACTIVE() will drop the vnode lock. 1512 */ 1513 VOP_INACTIVE(vp, &recycle); 1514 mutex_enter(&vp->v_interlock); 1515 vp->v_iflag &= ~VI_INACTNOW; 1516 cv_broadcast(&vp->v_cv); 1517 if (!recycle) { 1518 if (vtryrele(vp)) { 1519 mutex_exit(&vp->v_interlock); 1520 return; 1521 } 1522 1523 /* 1524 * If we grew another reference while 1525 * VOP_INACTIVE() was underway, retry. 1526 */ 1527 if ((vp->v_iflag & VI_INACTREDO) != 0) { 1528 goto retry; 1529 } 1530 } 1531 1532 /* Take care of space accounting. */ 1533 if (vp->v_iflag & VI_EXECMAP) { 1534 atomic_add_int(&uvmexp.execpages, 1535 -vp->v_uobj.uo_npages); 1536 atomic_add_int(&uvmexp.filepages, 1537 vp->v_uobj.uo_npages); 1538 } 1539 vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP); 1540 vp->v_vflag &= ~VV_MAPPED; 1541 1542 /* 1543 * Recycle the vnode if the file is now unused (unlinked), 1544 * otherwise just free it. 1545 */ 1546 if (recycle) { 1547 vclean(vp, DOCLOSE); 1548 } 1549 KASSERT(vp->v_usecount > 0); 1550 } 1551 1552 if (atomic_dec_uint_nv(&vp->v_usecount) != 0) { 1553 /* Gained another reference while being reclaimed. */ 1554 mutex_exit(&vp->v_interlock); 1555 return; 1556 } 1557 1558 if ((vp->v_iflag & VI_CLEAN) != 0) { 1559 /* 1560 * It's clean so destroy it. It isn't referenced 1561 * anywhere since it has been reclaimed. 1562 */ 1563 KASSERT(vp->v_holdcnt == 0); 1564 KASSERT(vp->v_writecount == 0); 1565 mutex_exit(&vp->v_interlock); 1566 insmntque(vp, NULL); 1567 if (vp->v_type == VBLK || vp->v_type == VCHR) { 1568 spec_node_destroy(vp); 1569 } 1570 vnfree(vp); 1571 } else { 1572 /* 1573 * Otherwise, put it back onto the freelist. It 1574 * can't be destroyed while still associated with 1575 * a file system. 1576 */ 1577 mutex_enter(&vnode_free_list_lock); 1578 if (vp->v_holdcnt > 0) { 1579 vp->v_freelisthd = &vnode_hold_list; 1580 } else { 1581 vp->v_freelisthd = &vnode_free_list; 1582 } 1583 TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 1584 mutex_exit(&vnode_free_list_lock); 1585 mutex_exit(&vp->v_interlock); 1586 } 1587 } 1588 1589 void 1590 vrele(vnode_t *vp) 1591 { 1592 1593 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1594 1595 if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) { 1596 return; 1597 } 1598 mutex_enter(&vp->v_interlock); 1599 vrelel(vp, 0); 1600 } 1601 1602 static void 1603 vrele_thread(void *cookie) 1604 { 1605 vnode_t *vp; 1606 1607 for (;;) { 1608 mutex_enter(&vrele_lock); 1609 while (TAILQ_EMPTY(&vrele_list)) { 1610 vrele_gen++; 1611 cv_broadcast(&vrele_cv); 1612 cv_timedwait(&vrele_cv, &vrele_lock, hz); 1613 } 1614 vp = TAILQ_FIRST(&vrele_list); 1615 TAILQ_REMOVE(&vrele_list, vp, v_freelist); 1616 vrele_pending--; 1617 mutex_exit(&vrele_lock); 1618 1619 /* 1620 * If not the last reference, then ignore the vnode 1621 * and look for more work. 1622 */ 1623 mutex_enter(&vp->v_interlock); 1624 KASSERT((vp->v_iflag & VI_INACTPEND) != 0); 1625 vp->v_iflag &= ~VI_INACTPEND; 1626 vrelel(vp, 0); 1627 } 1628 } 1629 1630 /* 1631 * Page or buffer structure gets a reference. 1632 * Called with v_interlock held. 1633 */ 1634 void 1635 vholdl(vnode_t *vp) 1636 { 1637 1638 KASSERT(mutex_owned(&vp->v_interlock)); 1639 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1640 1641 if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) { 1642 mutex_enter(&vnode_free_list_lock); 1643 KASSERT(vp->v_freelisthd == &vnode_free_list); 1644 TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 1645 vp->v_freelisthd = &vnode_hold_list; 1646 TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 1647 mutex_exit(&vnode_free_list_lock); 1648 } 1649 } 1650 1651 /* 1652 * Page or buffer structure frees a reference. 1653 * Called with v_interlock held. 1654 */ 1655 void 1656 holdrelel(vnode_t *vp) 1657 { 1658 1659 KASSERT(mutex_owned(&vp->v_interlock)); 1660 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1661 1662 if (vp->v_holdcnt <= 0) { 1663 vpanic(vp, "holdrelel: holdcnt vp %p"); 1664 } 1665 1666 vp->v_holdcnt--; 1667 if (vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1668 mutex_enter(&vnode_free_list_lock); 1669 KASSERT(vp->v_freelisthd == &vnode_hold_list); 1670 TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 1671 vp->v_freelisthd = &vnode_free_list; 1672 TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 1673 mutex_exit(&vnode_free_list_lock); 1674 } 1675 } 1676 1677 /* 1678 * Vnode reference, where a reference is already held by some other 1679 * object (for example, a file structure). 1680 */ 1681 void 1682 vref(vnode_t *vp) 1683 { 1684 1685 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1686 KASSERT(vp->v_usecount != 0); 1687 1688 atomic_inc_uint(&vp->v_usecount); 1689 } 1690 1691 /* 1692 * Remove any vnodes in the vnode table belonging to mount point mp. 1693 * 1694 * If FORCECLOSE is not specified, there should not be any active ones, 1695 * return error if any are found (nb: this is a user error, not a 1696 * system error). If FORCECLOSE is specified, detach any active vnodes 1697 * that are found. 1698 * 1699 * If WRITECLOSE is set, only flush out regular file vnodes open for 1700 * writing. 1701 * 1702 * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped. 1703 */ 1704 #ifdef DEBUG 1705 int busyprt = 0; /* print out busy vnodes */ 1706 struct ctldebug debug1 = { "busyprt", &busyprt }; 1707 #endif 1708 1709 static vnode_t * 1710 vflushnext(vnode_t *mvp, int *when) 1711 { 1712 1713 if (hardclock_ticks > *when) { 1714 mutex_exit(&mntvnode_lock); 1715 yield(); 1716 mutex_enter(&mntvnode_lock); 1717 *when = hardclock_ticks + hz / 10; 1718 } 1719 1720 return vunmark(mvp); 1721 } 1722 1723 int 1724 vflush(struct mount *mp, vnode_t *skipvp, int flags) 1725 { 1726 vnode_t *vp, *mvp; 1727 int busy = 0, when = 0, gen; 1728 1729 /* 1730 * First, flush out any vnode references from vrele_list. 1731 */ 1732 mutex_enter(&vrele_lock); 1733 gen = vrele_gen; 1734 while (vrele_pending && gen == vrele_gen) { 1735 cv_broadcast(&vrele_cv); 1736 cv_wait(&vrele_cv, &vrele_lock); 1737 } 1738 mutex_exit(&vrele_lock); 1739 1740 /* Allocate a marker vnode. */ 1741 if ((mvp = vnalloc(mp)) == NULL) 1742 return (ENOMEM); 1743 1744 /* 1745 * NOTE: not using the TAILQ_FOREACH here since in this loop vgone() 1746 * and vclean() are called 1747 */ 1748 mutex_enter(&mntvnode_lock); 1749 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp != NULL; 1750 vp = vflushnext(mvp, &when)) { 1751 vmark(mvp, vp); 1752 if (vp->v_mount != mp || vismarker(vp)) 1753 continue; 1754 /* 1755 * Skip over a selected vnode. 1756 */ 1757 if (vp == skipvp) 1758 continue; 1759 mutex_enter(&vp->v_interlock); 1760 /* 1761 * Ignore clean but still referenced vnodes. 1762 */ 1763 if ((vp->v_iflag & VI_CLEAN) != 0) { 1764 mutex_exit(&vp->v_interlock); 1765 continue; 1766 } 1767 /* 1768 * Skip over a vnodes marked VSYSTEM. 1769 */ 1770 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 1771 mutex_exit(&vp->v_interlock); 1772 continue; 1773 } 1774 /* 1775 * If WRITECLOSE is set, only flush out regular file 1776 * vnodes open for writing. 1777 */ 1778 if ((flags & WRITECLOSE) && 1779 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1780 mutex_exit(&vp->v_interlock); 1781 continue; 1782 } 1783 /* 1784 * With v_usecount == 0, all we need to do is clear 1785 * out the vnode data structures and we are done. 1786 */ 1787 if (vp->v_usecount == 0) { 1788 mutex_exit(&mntvnode_lock); 1789 vremfree(vp); 1790 vp->v_usecount = 1; 1791 vclean(vp, DOCLOSE); 1792 vrelel(vp, 0); 1793 mutex_enter(&mntvnode_lock); 1794 continue; 1795 } 1796 /* 1797 * If FORCECLOSE is set, forcibly close the vnode. 1798 * For block or character devices, revert to an 1799 * anonymous device. For all other files, just 1800 * kill them. 1801 */ 1802 if (flags & FORCECLOSE) { 1803 mutex_exit(&mntvnode_lock); 1804 atomic_inc_uint(&vp->v_usecount); 1805 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1806 vclean(vp, DOCLOSE); 1807 vrelel(vp, 0); 1808 } else { 1809 vclean(vp, 0); 1810 vp->v_op = spec_vnodeop_p; /* XXXSMP */ 1811 mutex_exit(&vp->v_interlock); 1812 /* 1813 * The vnode isn't clean, but still resides 1814 * on the mount list. Remove it. XXX This 1815 * is a bit dodgy. 1816 */ 1817 insmntque(vp, NULL); 1818 vrele(vp); 1819 } 1820 mutex_enter(&mntvnode_lock); 1821 continue; 1822 } 1823 #ifdef DEBUG 1824 if (busyprt) 1825 vprint("vflush: busy vnode", vp); 1826 #endif 1827 mutex_exit(&vp->v_interlock); 1828 busy++; 1829 } 1830 mutex_exit(&mntvnode_lock); 1831 vnfree(mvp); 1832 if (busy) 1833 return (EBUSY); 1834 return (0); 1835 } 1836 1837 /* 1838 * Disassociate the underlying file system from a vnode. 1839 * 1840 * Must be called with the interlock held, and will return with it held. 1841 */ 1842 void 1843 vclean(vnode_t *vp, int flags) 1844 { 1845 lwp_t *l = curlwp; 1846 bool recycle, active; 1847 int error; 1848 1849 KASSERT(mutex_owned(&vp->v_interlock)); 1850 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1851 KASSERT(vp->v_usecount != 0); 1852 1853 /* If cleaning is already in progress wait until done and return. */ 1854 if (vp->v_iflag & VI_XLOCK) { 1855 vwait(vp, VI_XLOCK); 1856 return; 1857 } 1858 1859 /* If already clean, nothing to do. */ 1860 if ((vp->v_iflag & VI_CLEAN) != 0) { 1861 return; 1862 } 1863 1864 /* 1865 * Prevent the vnode from being recycled or brought into use 1866 * while we clean it out. 1867 */ 1868 vp->v_iflag |= VI_XLOCK; 1869 if (vp->v_iflag & VI_EXECMAP) { 1870 atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages); 1871 atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages); 1872 } 1873 vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP); 1874 active = (vp->v_usecount > 1); 1875 1876 /* XXXAD should not lock vnode under layer */ 1877 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1878 1879 /* 1880 * Clean out any cached data associated with the vnode. 1881 * If purging an active vnode, it must be closed and 1882 * deactivated before being reclaimed. Note that the 1883 * VOP_INACTIVE will unlock the vnode. 1884 */ 1885 if (flags & DOCLOSE) { 1886 error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0); 1887 if (error != 0) { 1888 /* XXX, fix vn_start_write's grab of mp and use that. */ 1889 1890 if (wapbl_vphaswapbl(vp)) 1891 WAPBL_DISCARD(wapbl_vptomp(vp)); 1892 error = vinvalbuf(vp, 0, NOCRED, l, 0, 0); 1893 } 1894 KASSERT(error == 0); 1895 KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 1896 if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) { 1897 spec_node_revoke(vp); 1898 } 1899 } 1900 if (active) { 1901 VOP_INACTIVE(vp, &recycle); 1902 } else { 1903 /* 1904 * Any other processes trying to obtain this lock must first 1905 * wait for VI_XLOCK to clear, then call the new lock operation. 1906 */ 1907 VOP_UNLOCK(vp, 0); 1908 } 1909 1910 /* Disassociate the underlying file system from the vnode. */ 1911 if (VOP_RECLAIM(vp)) { 1912 vpanic(vp, "vclean: cannot reclaim"); 1913 } 1914 1915 KASSERT(vp->v_uobj.uo_npages == 0); 1916 if (vp->v_type == VREG && vp->v_ractx != NULL) { 1917 uvm_ra_freectx(vp->v_ractx); 1918 vp->v_ractx = NULL; 1919 } 1920 cache_purge(vp); 1921 1922 /* Done with purge, notify sleepers of the grim news. */ 1923 mutex_enter(&vp->v_interlock); 1924 vp->v_op = dead_vnodeop_p; 1925 vp->v_tag = VT_NON; 1926 vp->v_vnlock = &vp->v_lock; 1927 KNOTE(&vp->v_klist, NOTE_REVOKE); 1928 vp->v_iflag &= ~(VI_XLOCK | VI_FREEING); 1929 vp->v_vflag &= ~VV_LOCKSWORK; 1930 if ((flags & DOCLOSE) != 0) { 1931 vp->v_iflag |= VI_CLEAN; 1932 } 1933 cv_broadcast(&vp->v_cv); 1934 1935 KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 1936 } 1937 1938 /* 1939 * Recycle an unused vnode to the front of the free list. 1940 * Release the passed interlock if the vnode will be recycled. 1941 */ 1942 int 1943 vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l) 1944 { 1945 1946 KASSERT((vp->v_iflag & VI_MARKER) == 0); 1947 1948 mutex_enter(&vp->v_interlock); 1949 if (vp->v_usecount != 0) { 1950 mutex_exit(&vp->v_interlock); 1951 return (0); 1952 } 1953 if (inter_lkp) 1954 mutex_exit(inter_lkp); 1955 vremfree(vp); 1956 vp->v_usecount = 1; 1957 vclean(vp, DOCLOSE); 1958 vrelel(vp, 0); 1959 return (1); 1960 } 1961 1962 /* 1963 * Eliminate all activity associated with a vnode in preparation for 1964 * reuse. Drops a reference from the vnode. 1965 */ 1966 void 1967 vgone(vnode_t *vp) 1968 { 1969 1970 mutex_enter(&vp->v_interlock); 1971 vclean(vp, DOCLOSE); 1972 vrelel(vp, 0); 1973 } 1974 1975 /* 1976 * Lookup a vnode by device number. 1977 */ 1978 int 1979 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp) 1980 { 1981 vnode_t *vp; 1982 int rc = 0; 1983 1984 mutex_enter(&device_lock); 1985 for (vp = specfs_hash[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1986 if (dev != vp->v_rdev || type != vp->v_type) 1987 continue; 1988 *vpp = vp; 1989 rc = 1; 1990 break; 1991 } 1992 mutex_exit(&device_lock); 1993 return (rc); 1994 } 1995 1996 /* 1997 * Revoke all the vnodes corresponding to the specified minor number 1998 * range (endpoints inclusive) of the specified major. 1999 */ 2000 void 2001 vdevgone(int maj, int minl, int minh, enum vtype type) 2002 { 2003 vnode_t *vp, **vpp; 2004 dev_t dev; 2005 int mn; 2006 2007 vp = NULL; /* XXX gcc */ 2008 2009 mutex_enter(&device_lock); 2010 for (mn = minl; mn <= minh; mn++) { 2011 dev = makedev(maj, mn); 2012 vpp = &specfs_hash[SPECHASH(dev)]; 2013 for (vp = *vpp; vp != NULL;) { 2014 mutex_enter(&vp->v_interlock); 2015 if ((vp->v_iflag & VI_CLEAN) != 0 || 2016 dev != vp->v_rdev || type != vp->v_type) { 2017 mutex_exit(&vp->v_interlock); 2018 vp = vp->v_specnext; 2019 continue; 2020 } 2021 mutex_exit(&device_lock); 2022 if (vget(vp, LK_INTERLOCK) == 0) { 2023 VOP_REVOKE(vp, REVOKEALL); 2024 vrele(vp); 2025 } 2026 mutex_enter(&device_lock); 2027 vp = *vpp; 2028 } 2029 } 2030 mutex_exit(&device_lock); 2031 } 2032 2033 /* 2034 * Eliminate all activity associated with the requested vnode 2035 * and with all vnodes aliased to the requested vnode. 2036 */ 2037 void 2038 vrevoke(vnode_t *vp) 2039 { 2040 vnode_t *vq, **vpp; 2041 enum vtype type; 2042 dev_t dev; 2043 2044 KASSERT(vp->v_usecount > 0); 2045 2046 mutex_enter(&vp->v_interlock); 2047 if ((vp->v_iflag & VI_CLEAN) != 0) { 2048 mutex_exit(&vp->v_interlock); 2049 return; 2050 } else if (vp->v_type != VBLK && vp->v_type != VCHR) { 2051 atomic_inc_uint(&vp->v_usecount); 2052 vclean(vp, DOCLOSE); 2053 vrelel(vp, 0); 2054 return; 2055 } else { 2056 dev = vp->v_rdev; 2057 type = vp->v_type; 2058 mutex_exit(&vp->v_interlock); 2059 } 2060 2061 vpp = &specfs_hash[SPECHASH(dev)]; 2062 mutex_enter(&device_lock); 2063 for (vq = *vpp; vq != NULL;) { 2064 /* If clean or being cleaned, then ignore it. */ 2065 mutex_enter(&vq->v_interlock); 2066 if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 || 2067 vq->v_rdev != dev || vq->v_type != type) { 2068 mutex_exit(&vq->v_interlock); 2069 vq = vq->v_specnext; 2070 continue; 2071 } 2072 mutex_exit(&device_lock); 2073 if (vq->v_usecount == 0) { 2074 vremfree(vq); 2075 vq->v_usecount = 1; 2076 } else { 2077 atomic_inc_uint(&vq->v_usecount); 2078 } 2079 vclean(vq, DOCLOSE); 2080 vrelel(vq, 0); 2081 mutex_enter(&device_lock); 2082 vq = *vpp; 2083 } 2084 mutex_exit(&device_lock); 2085 } 2086 2087 /* 2088 * sysctl helper routine to return list of supported fstypes 2089 */ 2090 int 2091 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS) 2092 { 2093 char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)]; 2094 char *where = oldp; 2095 struct vfsops *v; 2096 size_t needed, left, slen; 2097 int error, first; 2098 2099 if (newp != NULL) 2100 return (EPERM); 2101 if (namelen != 0) 2102 return (EINVAL); 2103 2104 first = 1; 2105 error = 0; 2106 needed = 0; 2107 left = *oldlenp; 2108 2109 sysctl_unlock(); 2110 mutex_enter(&vfs_list_lock); 2111 LIST_FOREACH(v, &vfs_list, vfs_list) { 2112 if (where == NULL) 2113 needed += strlen(v->vfs_name) + 1; 2114 else { 2115 memset(bf, 0, sizeof(bf)); 2116 if (first) { 2117 strncpy(bf, v->vfs_name, sizeof(bf)); 2118 first = 0; 2119 } else { 2120 bf[0] = ' '; 2121 strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1); 2122 } 2123 bf[sizeof(bf)-1] = '\0'; 2124 slen = strlen(bf); 2125 if (left < slen + 1) 2126 break; 2127 v->vfs_refcount++; 2128 mutex_exit(&vfs_list_lock); 2129 /* +1 to copy out the trailing NUL byte */ 2130 error = copyout(bf, where, slen + 1); 2131 mutex_enter(&vfs_list_lock); 2132 v->vfs_refcount--; 2133 if (error) 2134 break; 2135 where += slen; 2136 needed += slen; 2137 left -= slen; 2138 } 2139 } 2140 mutex_exit(&vfs_list_lock); 2141 sysctl_relock(); 2142 *oldlenp = needed; 2143 return (error); 2144 } 2145 2146 2147 int kinfo_vdebug = 1; 2148 int kinfo_vgetfailed; 2149 #define KINFO_VNODESLOP 10 2150 /* 2151 * Dump vnode list (via sysctl). 2152 * Copyout address of vnode followed by vnode. 2153 */ 2154 /* ARGSUSED */ 2155 int 2156 sysctl_kern_vnode(SYSCTLFN_ARGS) 2157 { 2158 char *where = oldp; 2159 size_t *sizep = oldlenp; 2160 struct mount *mp, *nmp; 2161 vnode_t *vp, *mvp, vbuf; 2162 char *bp = where; 2163 char *ewhere; 2164 int error; 2165 2166 if (namelen != 0) 2167 return (EOPNOTSUPP); 2168 if (newp != NULL) 2169 return (EPERM); 2170 2171 #define VPTRSZ sizeof(vnode_t *) 2172 #define VNODESZ sizeof(vnode_t) 2173 if (where == NULL) { 2174 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 2175 return (0); 2176 } 2177 ewhere = where + *sizep; 2178 2179 sysctl_unlock(); 2180 mutex_enter(&mountlist_lock); 2181 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 2182 mp = nmp) { 2183 if (vfs_busy(mp, &nmp)) { 2184 continue; 2185 } 2186 /* Allocate a marker vnode. */ 2187 mvp = vnalloc(mp); 2188 /* Should never fail for mp != NULL */ 2189 KASSERT(mvp != NULL); 2190 mutex_enter(&mntvnode_lock); 2191 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; 2192 vp = vunmark(mvp)) { 2193 vmark(mvp, vp); 2194 /* 2195 * Check that the vp is still associated with 2196 * this filesystem. RACE: could have been 2197 * recycled onto the same filesystem. 2198 */ 2199 if (vp->v_mount != mp || vismarker(vp)) 2200 continue; 2201 if (bp + VPTRSZ + VNODESZ > ewhere) { 2202 (void)vunmark(mvp); 2203 mutex_exit(&mntvnode_lock); 2204 vnfree(mvp); 2205 vfs_unbusy(mp, false, NULL); 2206 sysctl_relock(); 2207 *sizep = bp - where; 2208 return (ENOMEM); 2209 } 2210 memcpy(&vbuf, vp, VNODESZ); 2211 mutex_exit(&mntvnode_lock); 2212 if ((error = copyout(&vp, bp, VPTRSZ)) || 2213 (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) { 2214 mutex_enter(&mntvnode_lock); 2215 (void)vunmark(mvp); 2216 mutex_exit(&mntvnode_lock); 2217 vnfree(mvp); 2218 vfs_unbusy(mp, false, NULL); 2219 sysctl_relock(); 2220 return (error); 2221 } 2222 bp += VPTRSZ + VNODESZ; 2223 mutex_enter(&mntvnode_lock); 2224 } 2225 mutex_exit(&mntvnode_lock); 2226 vnfree(mvp); 2227 vfs_unbusy(mp, false, &nmp); 2228 } 2229 mutex_exit(&mountlist_lock); 2230 sysctl_relock(); 2231 2232 *sizep = bp - where; 2233 return (0); 2234 } 2235 2236 /* 2237 * Remove clean vnodes from a mountpoint's vnode list. 2238 */ 2239 void 2240 vfs_scrubvnlist(struct mount *mp) 2241 { 2242 vnode_t *vp, *nvp; 2243 2244 retry: 2245 mutex_enter(&mntvnode_lock); 2246 for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) { 2247 nvp = TAILQ_NEXT(vp, v_mntvnodes); 2248 mutex_enter(&vp->v_interlock); 2249 if ((vp->v_iflag & VI_CLEAN) != 0) { 2250 TAILQ_REMOVE(&mp->mnt_vnodelist, vp, v_mntvnodes); 2251 vp->v_mount = NULL; 2252 mutex_exit(&mntvnode_lock); 2253 mutex_exit(&vp->v_interlock); 2254 vfs_destroy(mp); 2255 goto retry; 2256 } 2257 mutex_exit(&vp->v_interlock); 2258 } 2259 mutex_exit(&mntvnode_lock); 2260 } 2261 2262 /* 2263 * Check to see if a filesystem is mounted on a block device. 2264 */ 2265 int 2266 vfs_mountedon(vnode_t *vp) 2267 { 2268 vnode_t *vq; 2269 int error = 0; 2270 2271 if (vp->v_type != VBLK) 2272 return ENOTBLK; 2273 if (vp->v_specmountpoint != NULL) 2274 return (EBUSY); 2275 mutex_enter(&device_lock); 2276 for (vq = specfs_hash[SPECHASH(vp->v_rdev)]; vq != NULL; 2277 vq = vq->v_specnext) { 2278 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 2279 continue; 2280 if (vq->v_specmountpoint != NULL) { 2281 error = EBUSY; 2282 break; 2283 } 2284 } 2285 mutex_exit(&device_lock); 2286 return (error); 2287 } 2288 2289 /* 2290 * Unmount all file systems. 2291 * We traverse the list in reverse order under the assumption that doing so 2292 * will avoid needing to worry about dependencies. 2293 */ 2294 bool 2295 vfs_unmountall(struct lwp *l) 2296 { 2297 printf("unmounting file systems..."); 2298 return vfs_unmountall1(l, true, true); 2299 } 2300 2301 static void 2302 vfs_unmount_print(struct mount *mp, const char *pfx) 2303 { 2304 printf("%sunmounted %s on %s type %s\n", pfx, 2305 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, 2306 mp->mnt_stat.f_fstypename); 2307 } 2308 2309 bool 2310 vfs_unmount_forceone(struct lwp *l) 2311 { 2312 struct mount *mp, *nmp = NULL; 2313 int error; 2314 2315 CIRCLEQ_FOREACH_REVERSE(mp, &mountlist, mnt_list) { 2316 if (nmp == NULL || mp->mnt_gen > nmp->mnt_gen) 2317 nmp = mp; 2318 } 2319 2320 if (nmp == NULL) 2321 return false; 2322 2323 #ifdef DEBUG 2324 printf("\nforcefully unmounting %s (%s)...", 2325 nmp->mnt_stat.f_mntonname, nmp->mnt_stat.f_mntfromname); 2326 #endif 2327 atomic_inc_uint(&nmp->mnt_refcnt); 2328 if ((error = dounmount(nmp, MNT_FORCE, l)) == 0) { 2329 vfs_unmount_print(nmp, "forcefully "); 2330 return true; 2331 } else 2332 atomic_dec_uint(&nmp->mnt_refcnt); 2333 2334 #ifdef DEBUG 2335 printf("forceful unmount of %s failed with error %d\n", 2336 nmp->mnt_stat.f_mntonname, error); 2337 #endif 2338 2339 return false; 2340 } 2341 2342 bool 2343 vfs_unmountall1(struct lwp *l, bool force, bool verbose) 2344 { 2345 struct mount *mp, *nmp; 2346 bool any_error = false, progress = false; 2347 int error; 2348 2349 for (mp = CIRCLEQ_LAST(&mountlist); 2350 mp != (void *)&mountlist; 2351 mp = nmp) { 2352 nmp = CIRCLEQ_PREV(mp, mnt_list); 2353 #ifdef DEBUG 2354 printf("\nunmounting %p %s (%s)...", 2355 (void *)mp, mp->mnt_stat.f_mntonname, 2356 mp->mnt_stat.f_mntfromname); 2357 #endif 2358 atomic_inc_uint(&mp->mnt_refcnt); 2359 if ((error = dounmount(mp, force ? MNT_FORCE : 0, l)) == 0) { 2360 vfs_unmount_print(mp, ""); 2361 progress = true; 2362 } else { 2363 atomic_dec_uint(&mp->mnt_refcnt); 2364 if (verbose) { 2365 printf("unmount of %s failed with error %d\n", 2366 mp->mnt_stat.f_mntonname, error); 2367 } 2368 any_error = true; 2369 } 2370 } 2371 if (verbose) 2372 printf(" done\n"); 2373 if (any_error && verbose) 2374 printf("WARNING: some file systems would not unmount\n"); 2375 return progress; 2376 } 2377 2378 /* 2379 * Sync and unmount file systems before shutting down. 2380 */ 2381 void 2382 vfs_shutdown(void) 2383 { 2384 struct lwp *l; 2385 2386 /* XXX we're certainly not running in lwp0's context! */ 2387 l = (curlwp == NULL) ? &lwp0 : curlwp; 2388 2389 vfs_shutdown1(l); 2390 } 2391 2392 void 2393 vfs_sync_all(struct lwp *l) 2394 { 2395 printf("syncing disks... "); 2396 2397 /* remove user processes from run queue */ 2398 suspendsched(); 2399 (void) spl0(); 2400 2401 /* avoid coming back this way again if we panic. */ 2402 doing_shutdown = 1; 2403 2404 sys_sync(l, NULL, NULL); 2405 2406 /* Wait for sync to finish. */ 2407 if (buf_syncwait() != 0) { 2408 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 2409 Debugger(); 2410 #endif 2411 printf("giving up\n"); 2412 return; 2413 } else 2414 printf("done\n"); 2415 } 2416 2417 static void 2418 vfs_shutdown1(struct lwp *l) 2419 { 2420 2421 vfs_sync_all(l); 2422 2423 /* 2424 * If we've panic'd, don't make the situation potentially 2425 * worse by unmounting the file systems. 2426 */ 2427 if (panicstr != NULL) 2428 return; 2429 2430 /* Release inodes held by texts before update. */ 2431 #ifdef notdef 2432 vnshutdown(); 2433 #endif 2434 /* Unmount file systems. */ 2435 vfs_unmountall(l); 2436 } 2437 2438 /* 2439 * Print a list of supported file system types (used by vfs_mountroot) 2440 */ 2441 static void 2442 vfs_print_fstypes(void) 2443 { 2444 struct vfsops *v; 2445 int cnt = 0; 2446 2447 mutex_enter(&vfs_list_lock); 2448 LIST_FOREACH(v, &vfs_list, vfs_list) 2449 ++cnt; 2450 mutex_exit(&vfs_list_lock); 2451 2452 if (cnt == 0) { 2453 printf("WARNING: No file system modules have been loaded.\n"); 2454 return; 2455 } 2456 2457 printf("Supported file systems:"); 2458 mutex_enter(&vfs_list_lock); 2459 LIST_FOREACH(v, &vfs_list, vfs_list) { 2460 printf(" %s", v->vfs_name); 2461 } 2462 mutex_exit(&vfs_list_lock); 2463 printf("\n"); 2464 } 2465 2466 /* 2467 * Mount the root file system. If the operator didn't specify a 2468 * file system to use, try all possible file systems until one 2469 * succeeds. 2470 */ 2471 int 2472 vfs_mountroot(void) 2473 { 2474 struct vfsops *v; 2475 int error = ENODEV; 2476 2477 if (root_device == NULL) 2478 panic("vfs_mountroot: root device unknown"); 2479 2480 switch (device_class(root_device)) { 2481 case DV_IFNET: 2482 if (rootdev != NODEV) 2483 panic("vfs_mountroot: rootdev set for DV_IFNET " 2484 "(0x%llx -> %llu,%llu)", 2485 (unsigned long long)rootdev, 2486 (unsigned long long)major(rootdev), 2487 (unsigned long long)minor(rootdev)); 2488 break; 2489 2490 case DV_DISK: 2491 if (rootdev == NODEV) 2492 panic("vfs_mountroot: rootdev not set for DV_DISK"); 2493 if (bdevvp(rootdev, &rootvp)) 2494 panic("vfs_mountroot: can't get vnode for rootdev"); 2495 error = VOP_OPEN(rootvp, FREAD, FSCRED); 2496 if (error) { 2497 printf("vfs_mountroot: can't open root device\n"); 2498 return (error); 2499 } 2500 break; 2501 2502 case DV_VIRTUAL: 2503 break; 2504 2505 default: 2506 printf("%s: inappropriate for root file system\n", 2507 device_xname(root_device)); 2508 return (ENODEV); 2509 } 2510 2511 /* 2512 * If user specified a root fs type, use it. Make sure the 2513 * specified type exists and has a mount_root() 2514 */ 2515 if (strcmp(rootfstype, ROOT_FSTYPE_ANY) != 0) { 2516 v = vfs_getopsbyname(rootfstype); 2517 error = EFTYPE; 2518 if (v != NULL) { 2519 if (v->vfs_mountroot != NULL) { 2520 error = (v->vfs_mountroot)(); 2521 } 2522 v->vfs_refcount--; 2523 } 2524 goto done; 2525 } 2526 2527 /* 2528 * Try each file system currently configured into the kernel. 2529 */ 2530 mutex_enter(&vfs_list_lock); 2531 LIST_FOREACH(v, &vfs_list, vfs_list) { 2532 if (v->vfs_mountroot == NULL) 2533 continue; 2534 #ifdef DEBUG 2535 aprint_normal("mountroot: trying %s...\n", v->vfs_name); 2536 #endif 2537 v->vfs_refcount++; 2538 mutex_exit(&vfs_list_lock); 2539 error = (*v->vfs_mountroot)(); 2540 mutex_enter(&vfs_list_lock); 2541 v->vfs_refcount--; 2542 if (!error) { 2543 aprint_normal("root file system type: %s\n", 2544 v->vfs_name); 2545 break; 2546 } 2547 } 2548 mutex_exit(&vfs_list_lock); 2549 2550 if (v == NULL) { 2551 vfs_print_fstypes(); 2552 printf("no file system for %s", device_xname(root_device)); 2553 if (device_class(root_device) == DV_DISK) 2554 printf(" (dev 0x%llx)", (unsigned long long)rootdev); 2555 printf("\n"); 2556 error = EFTYPE; 2557 } 2558 2559 done: 2560 if (error && device_class(root_device) == DV_DISK) { 2561 VOP_CLOSE(rootvp, FREAD, FSCRED); 2562 vrele(rootvp); 2563 } 2564 if (error == 0) { 2565 extern struct cwdinfo cwdi0; 2566 2567 CIRCLEQ_FIRST(&mountlist)->mnt_flag |= MNT_ROOTFS; 2568 CIRCLEQ_FIRST(&mountlist)->mnt_op->vfs_refcount++; 2569 2570 /* 2571 * Get the vnode for '/'. Set cwdi0.cwdi_cdir to 2572 * reference it. 2573 */ 2574 error = VFS_ROOT(CIRCLEQ_FIRST(&mountlist), &rootvnode); 2575 if (error) 2576 panic("cannot find root vnode, error=%d", error); 2577 cwdi0.cwdi_cdir = rootvnode; 2578 vref(cwdi0.cwdi_cdir); 2579 VOP_UNLOCK(rootvnode, 0); 2580 cwdi0.cwdi_rdir = NULL; 2581 2582 /* 2583 * Now that root is mounted, we can fixup initproc's CWD 2584 * info. All other processes are kthreads, which merely 2585 * share proc0's CWD info. 2586 */ 2587 initproc->p_cwdi->cwdi_cdir = rootvnode; 2588 vref(initproc->p_cwdi->cwdi_cdir); 2589 initproc->p_cwdi->cwdi_rdir = NULL; 2590 } 2591 return (error); 2592 } 2593 2594 /* 2595 * Get a new unique fsid 2596 */ 2597 void 2598 vfs_getnewfsid(struct mount *mp) 2599 { 2600 static u_short xxxfs_mntid; 2601 fsid_t tfsid; 2602 int mtype; 2603 2604 mutex_enter(&mntid_lock); 2605 mtype = makefstype(mp->mnt_op->vfs_name); 2606 mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0); 2607 mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype; 2608 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 2609 if (xxxfs_mntid == 0) 2610 ++xxxfs_mntid; 2611 tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid); 2612 tfsid.__fsid_val[1] = mtype; 2613 if (!CIRCLEQ_EMPTY(&mountlist)) { 2614 while (vfs_getvfs(&tfsid)) { 2615 tfsid.__fsid_val[0]++; 2616 xxxfs_mntid++; 2617 } 2618 } 2619 mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0]; 2620 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 2621 mutex_exit(&mntid_lock); 2622 } 2623 2624 /* 2625 * Make a 'unique' number from a mount type name. 2626 */ 2627 long 2628 makefstype(const char *type) 2629 { 2630 long rv; 2631 2632 for (rv = 0; *type; type++) { 2633 rv <<= 2; 2634 rv ^= *type; 2635 } 2636 return rv; 2637 } 2638 2639 /* 2640 * Set vnode attributes to VNOVAL 2641 */ 2642 void 2643 vattr_null(struct vattr *vap) 2644 { 2645 2646 memset(vap, 0, sizeof(*vap)); 2647 2648 vap->va_type = VNON; 2649 2650 /* 2651 * Assign individually so that it is safe even if size and 2652 * sign of each member are varied. 2653 */ 2654 vap->va_mode = VNOVAL; 2655 vap->va_nlink = VNOVAL; 2656 vap->va_uid = VNOVAL; 2657 vap->va_gid = VNOVAL; 2658 vap->va_fsid = VNOVAL; 2659 vap->va_fileid = VNOVAL; 2660 vap->va_size = VNOVAL; 2661 vap->va_blocksize = VNOVAL; 2662 vap->va_atime.tv_sec = 2663 vap->va_mtime.tv_sec = 2664 vap->va_ctime.tv_sec = 2665 vap->va_birthtime.tv_sec = VNOVAL; 2666 vap->va_atime.tv_nsec = 2667 vap->va_mtime.tv_nsec = 2668 vap->va_ctime.tv_nsec = 2669 vap->va_birthtime.tv_nsec = VNOVAL; 2670 vap->va_gen = VNOVAL; 2671 vap->va_flags = VNOVAL; 2672 vap->va_rdev = VNOVAL; 2673 vap->va_bytes = VNOVAL; 2674 } 2675 2676 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 2677 #define ARRAY_PRINT(idx, arr) \ 2678 ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN") 2679 2680 const char * const vnode_tags[] = { VNODE_TAGS }; 2681 const char * const vnode_types[] = { VNODE_TYPES }; 2682 const char vnode_flagbits[] = VNODE_FLAGBITS; 2683 2684 /* 2685 * Print out a description of a vnode. 2686 */ 2687 void 2688 vprint(const char *label, struct vnode *vp) 2689 { 2690 struct vnlock *vl; 2691 char bf[96]; 2692 int flag; 2693 2694 vl = (vp->v_vnlock != NULL ? vp->v_vnlock : &vp->v_lock); 2695 flag = vp->v_iflag | vp->v_vflag | vp->v_uflag; 2696 snprintb(bf, sizeof(bf), vnode_flagbits, flag); 2697 2698 if (label != NULL) 2699 printf("%s: ", label); 2700 printf("vnode @ %p, flags (%s)\n\ttag %s(%d), type %s(%d), " 2701 "usecount %d, writecount %d, holdcount %d\n" 2702 "\tfreelisthd %p, mount %p, data %p lock %p recursecnt %d\n", 2703 vp, bf, ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag, 2704 ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type, 2705 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, 2706 vp->v_freelisthd, vp->v_mount, vp->v_data, vl, vl->vl_recursecnt); 2707 if (vp->v_data != NULL) { 2708 printf("\t"); 2709 VOP_PRINT(vp); 2710 } 2711 } 2712 2713 #ifdef DEBUG 2714 /* 2715 * List all of the locked vnodes in the system. 2716 * Called when debugging the kernel. 2717 */ 2718 void 2719 printlockedvnodes(void) 2720 { 2721 struct mount *mp, *nmp; 2722 struct vnode *vp; 2723 2724 printf("Locked vnodes\n"); 2725 mutex_enter(&mountlist_lock); 2726 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 2727 mp = nmp) { 2728 if (vfs_busy(mp, &nmp)) { 2729 continue; 2730 } 2731 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2732 if (VOP_ISLOCKED(vp)) 2733 vprint(NULL, vp); 2734 } 2735 mutex_enter(&mountlist_lock); 2736 vfs_unbusy(mp, false, &nmp); 2737 } 2738 mutex_exit(&mountlist_lock); 2739 } 2740 #endif 2741 2742 /* Deprecated. Kept for KPI compatibility. */ 2743 int 2744 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 2745 mode_t acc_mode, kauth_cred_t cred) 2746 { 2747 2748 #ifdef DIAGNOSTIC 2749 printf("vaccess: deprecated interface used.\n"); 2750 #endif /* DIAGNOSTIC */ 2751 2752 return genfs_can_access(type, file_mode, uid, gid, acc_mode, cred); 2753 } 2754 2755 /* 2756 * Given a file system name, look up the vfsops for that 2757 * file system, or return NULL if file system isn't present 2758 * in the kernel. 2759 */ 2760 struct vfsops * 2761 vfs_getopsbyname(const char *name) 2762 { 2763 struct vfsops *v; 2764 2765 mutex_enter(&vfs_list_lock); 2766 LIST_FOREACH(v, &vfs_list, vfs_list) { 2767 if (strcmp(v->vfs_name, name) == 0) 2768 break; 2769 } 2770 if (v != NULL) 2771 v->vfs_refcount++; 2772 mutex_exit(&vfs_list_lock); 2773 2774 return (v); 2775 } 2776 2777 void 2778 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp) 2779 { 2780 const struct statvfs *mbp; 2781 2782 if (sbp == (mbp = &mp->mnt_stat)) 2783 return; 2784 2785 (void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx)); 2786 sbp->f_fsid = mbp->f_fsid; 2787 sbp->f_owner = mbp->f_owner; 2788 sbp->f_flag = mbp->f_flag; 2789 sbp->f_syncwrites = mbp->f_syncwrites; 2790 sbp->f_asyncwrites = mbp->f_asyncwrites; 2791 sbp->f_syncreads = mbp->f_syncreads; 2792 sbp->f_asyncreads = mbp->f_asyncreads; 2793 (void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare)); 2794 (void)memcpy(sbp->f_fstypename, mbp->f_fstypename, 2795 sizeof(sbp->f_fstypename)); 2796 (void)memcpy(sbp->f_mntonname, mbp->f_mntonname, 2797 sizeof(sbp->f_mntonname)); 2798 (void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, 2799 sizeof(sbp->f_mntfromname)); 2800 sbp->f_namemax = mbp->f_namemax; 2801 } 2802 2803 int 2804 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom, 2805 const char *vfsname, struct mount *mp, struct lwp *l) 2806 { 2807 int error; 2808 size_t size; 2809 struct statvfs *sfs = &mp->mnt_stat; 2810 int (*fun)(const void *, void *, size_t, size_t *); 2811 2812 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsname, 2813 sizeof(mp->mnt_stat.f_fstypename)); 2814 2815 if (onp) { 2816 struct cwdinfo *cwdi = l->l_proc->p_cwdi; 2817 fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr; 2818 if (cwdi->cwdi_rdir != NULL) { 2819 size_t len; 2820 char *bp; 2821 char *path = PNBUF_GET(); 2822 2823 bp = path + MAXPATHLEN; 2824 *--bp = '\0'; 2825 rw_enter(&cwdi->cwdi_lock, RW_READER); 2826 error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp, 2827 path, MAXPATHLEN / 2, 0, l); 2828 rw_exit(&cwdi->cwdi_lock); 2829 if (error) { 2830 PNBUF_PUT(path); 2831 return error; 2832 } 2833 2834 len = strlen(bp); 2835 if (len > sizeof(sfs->f_mntonname) - 1) 2836 len = sizeof(sfs->f_mntonname) - 1; 2837 (void)strncpy(sfs->f_mntonname, bp, len); 2838 PNBUF_PUT(path); 2839 2840 if (len < sizeof(sfs->f_mntonname) - 1) { 2841 error = (*fun)(onp, &sfs->f_mntonname[len], 2842 sizeof(sfs->f_mntonname) - len - 1, &size); 2843 if (error) 2844 return error; 2845 size += len; 2846 } else { 2847 size = len; 2848 } 2849 } else { 2850 error = (*fun)(onp, &sfs->f_mntonname, 2851 sizeof(sfs->f_mntonname) - 1, &size); 2852 if (error) 2853 return error; 2854 } 2855 (void)memset(sfs->f_mntonname + size, 0, 2856 sizeof(sfs->f_mntonname) - size); 2857 } 2858 2859 if (fromp) { 2860 fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr; 2861 error = (*fun)(fromp, sfs->f_mntfromname, 2862 sizeof(sfs->f_mntfromname) - 1, &size); 2863 if (error) 2864 return error; 2865 (void)memset(sfs->f_mntfromname + size, 0, 2866 sizeof(sfs->f_mntfromname) - size); 2867 } 2868 return 0; 2869 } 2870 2871 void 2872 vfs_timestamp(struct timespec *ts) 2873 { 2874 2875 nanotime(ts); 2876 } 2877 2878 time_t rootfstime; /* recorded root fs time, if known */ 2879 void 2880 setrootfstime(time_t t) 2881 { 2882 rootfstime = t; 2883 } 2884 2885 /* 2886 * Sham lock manager for vnodes. This is a temporary measure. 2887 */ 2888 int 2889 vlockmgr(struct vnlock *vl, int flags) 2890 { 2891 2892 KASSERT((flags & ~(LK_CANRECURSE | LK_NOWAIT | LK_TYPE_MASK)) == 0); 2893 2894 switch (flags & LK_TYPE_MASK) { 2895 case LK_SHARED: 2896 if (rw_tryenter(&vl->vl_lock, RW_READER)) { 2897 return 0; 2898 } 2899 if ((flags & LK_NOWAIT) != 0) { 2900 return EBUSY; 2901 } 2902 rw_enter(&vl->vl_lock, RW_READER); 2903 return 0; 2904 2905 case LK_EXCLUSIVE: 2906 if (rw_tryenter(&vl->vl_lock, RW_WRITER)) { 2907 return 0; 2908 } 2909 if ((vl->vl_canrecurse || (flags & LK_CANRECURSE) != 0) && 2910 rw_write_held(&vl->vl_lock)) { 2911 vl->vl_recursecnt++; 2912 return 0; 2913 } 2914 if ((flags & LK_NOWAIT) != 0) { 2915 return EBUSY; 2916 } 2917 rw_enter(&vl->vl_lock, RW_WRITER); 2918 return 0; 2919 2920 case LK_RELEASE: 2921 if (vl->vl_recursecnt != 0) { 2922 KASSERT(rw_write_held(&vl->vl_lock)); 2923 vl->vl_recursecnt--; 2924 return 0; 2925 } 2926 rw_exit(&vl->vl_lock); 2927 return 0; 2928 2929 default: 2930 panic("vlockmgr: flags %x", flags); 2931 } 2932 } 2933 2934 int 2935 vlockstatus(struct vnlock *vl) 2936 { 2937 2938 if (rw_write_held(&vl->vl_lock)) { 2939 return LK_EXCLUSIVE; 2940 } 2941 if (rw_read_held(&vl->vl_lock)) { 2942 return LK_SHARED; 2943 } 2944 return 0; 2945 } 2946 2947 /* 2948 * mount_specific_key_create -- 2949 * Create a key for subsystem mount-specific data. 2950 */ 2951 int 2952 mount_specific_key_create(specificdata_key_t *keyp, specificdata_dtor_t dtor) 2953 { 2954 2955 return (specificdata_key_create(mount_specificdata_domain, keyp, dtor)); 2956 } 2957 2958 /* 2959 * mount_specific_key_delete -- 2960 * Delete a key for subsystem mount-specific data. 2961 */ 2962 void 2963 mount_specific_key_delete(specificdata_key_t key) 2964 { 2965 2966 specificdata_key_delete(mount_specificdata_domain, key); 2967 } 2968 2969 /* 2970 * mount_initspecific -- 2971 * Initialize a mount's specificdata container. 2972 */ 2973 void 2974 mount_initspecific(struct mount *mp) 2975 { 2976 int error; 2977 2978 error = specificdata_init(mount_specificdata_domain, 2979 &mp->mnt_specdataref); 2980 KASSERT(error == 0); 2981 } 2982 2983 /* 2984 * mount_finispecific -- 2985 * Finalize a mount's specificdata container. 2986 */ 2987 void 2988 mount_finispecific(struct mount *mp) 2989 { 2990 2991 specificdata_fini(mount_specificdata_domain, &mp->mnt_specdataref); 2992 } 2993 2994 /* 2995 * mount_getspecific -- 2996 * Return mount-specific data corresponding to the specified key. 2997 */ 2998 void * 2999 mount_getspecific(struct mount *mp, specificdata_key_t key) 3000 { 3001 3002 return (specificdata_getspecific(mount_specificdata_domain, 3003 &mp->mnt_specdataref, key)); 3004 } 3005 3006 /* 3007 * mount_setspecific -- 3008 * Set mount-specific data corresponding to the specified key. 3009 */ 3010 void 3011 mount_setspecific(struct mount *mp, specificdata_key_t key, void *data) 3012 { 3013 3014 specificdata_setspecific(mount_specificdata_domain, 3015 &mp->mnt_specdataref, key, data); 3016 } 3017 3018 int 3019 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c) 3020 { 3021 int error; 3022 3023 KERNEL_LOCK(1, NULL); 3024 error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c); 3025 KERNEL_UNLOCK_ONE(NULL); 3026 3027 return error; 3028 } 3029 3030 int 3031 VFS_START(struct mount *mp, int a) 3032 { 3033 int error; 3034 3035 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3036 KERNEL_LOCK(1, NULL); 3037 } 3038 error = (*(mp->mnt_op->vfs_start))(mp, a); 3039 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3040 KERNEL_UNLOCK_ONE(NULL); 3041 } 3042 3043 return error; 3044 } 3045 3046 int 3047 VFS_UNMOUNT(struct mount *mp, int a) 3048 { 3049 int error; 3050 3051 KERNEL_LOCK(1, NULL); 3052 error = (*(mp->mnt_op->vfs_unmount))(mp, a); 3053 KERNEL_UNLOCK_ONE(NULL); 3054 3055 return error; 3056 } 3057 3058 int 3059 VFS_ROOT(struct mount *mp, struct vnode **a) 3060 { 3061 int error; 3062 3063 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3064 KERNEL_LOCK(1, NULL); 3065 } 3066 error = (*(mp->mnt_op->vfs_root))(mp, a); 3067 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3068 KERNEL_UNLOCK_ONE(NULL); 3069 } 3070 3071 return error; 3072 } 3073 3074 int 3075 VFS_QUOTACTL(struct mount *mp, int a, uid_t b, void *c) 3076 { 3077 int error; 3078 3079 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3080 KERNEL_LOCK(1, NULL); 3081 } 3082 error = (*(mp->mnt_op->vfs_quotactl))(mp, a, b, c); 3083 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3084 KERNEL_UNLOCK_ONE(NULL); 3085 } 3086 3087 return error; 3088 } 3089 3090 int 3091 VFS_STATVFS(struct mount *mp, struct statvfs *a) 3092 { 3093 int error; 3094 3095 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3096 KERNEL_LOCK(1, NULL); 3097 } 3098 error = (*(mp->mnt_op->vfs_statvfs))(mp, a); 3099 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3100 KERNEL_UNLOCK_ONE(NULL); 3101 } 3102 3103 return error; 3104 } 3105 3106 int 3107 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b) 3108 { 3109 int error; 3110 3111 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3112 KERNEL_LOCK(1, NULL); 3113 } 3114 error = (*(mp->mnt_op->vfs_sync))(mp, a, b); 3115 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3116 KERNEL_UNLOCK_ONE(NULL); 3117 } 3118 3119 return error; 3120 } 3121 3122 int 3123 VFS_FHTOVP(struct mount *mp, struct fid *a, struct vnode **b) 3124 { 3125 int error; 3126 3127 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3128 KERNEL_LOCK(1, NULL); 3129 } 3130 error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b); 3131 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3132 KERNEL_UNLOCK_ONE(NULL); 3133 } 3134 3135 return error; 3136 } 3137 3138 int 3139 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b) 3140 { 3141 int error; 3142 3143 if ((vp->v_vflag & VV_MPSAFE) == 0) { 3144 KERNEL_LOCK(1, NULL); 3145 } 3146 error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b); 3147 if ((vp->v_vflag & VV_MPSAFE) == 0) { 3148 KERNEL_UNLOCK_ONE(NULL); 3149 } 3150 3151 return error; 3152 } 3153 3154 int 3155 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b) 3156 { 3157 int error; 3158 3159 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3160 KERNEL_LOCK(1, NULL); 3161 } 3162 error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b); 3163 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3164 KERNEL_UNLOCK_ONE(NULL); 3165 } 3166 3167 return error; 3168 } 3169 3170 int 3171 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d) 3172 { 3173 int error; 3174 3175 KERNEL_LOCK(1, NULL); /* XXXSMP check ffs */ 3176 error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d); 3177 KERNEL_UNLOCK_ONE(NULL); /* XXX */ 3178 3179 return error; 3180 } 3181 3182 int 3183 VFS_SUSPENDCTL(struct mount *mp, int a) 3184 { 3185 int error; 3186 3187 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3188 KERNEL_LOCK(1, NULL); 3189 } 3190 error = (*(mp->mnt_op->vfs_suspendctl))(mp, a); 3191 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 3192 KERNEL_UNLOCK_ONE(NULL); 3193 } 3194 3195 return error; 3196 } 3197 3198 #if defined(DDB) || defined(DEBUGPRINT) 3199 static const char buf_flagbits[] = BUF_FLAGBITS; 3200 3201 void 3202 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...)) 3203 { 3204 char bf[1024]; 3205 3206 (*pr)(" vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%" 3207 PRIx64 " dev 0x%x\n", 3208 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev); 3209 3210 snprintb(bf, sizeof(bf), 3211 buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags); 3212 (*pr)(" error %d flags 0x%s\n", bp->b_error, bf); 3213 3214 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n", 3215 bp->b_bufsize, bp->b_bcount, bp->b_resid); 3216 (*pr)(" data %p saveaddr %p\n", 3217 bp->b_data, bp->b_saveaddr); 3218 (*pr)(" iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock); 3219 } 3220 3221 3222 void 3223 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...)) 3224 { 3225 char bf[256]; 3226 3227 uvm_object_printit(&vp->v_uobj, full, pr); 3228 snprintb(bf, sizeof(bf), 3229 vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag); 3230 (*pr)("\nVNODE flags %s\n", bf); 3231 (*pr)("mp %p numoutput %d size 0x%llx writesize 0x%llx\n", 3232 vp->v_mount, vp->v_numoutput, vp->v_size, vp->v_writesize); 3233 3234 (*pr)("data %p writecount %ld holdcnt %ld\n", 3235 vp->v_data, vp->v_writecount, vp->v_holdcnt); 3236 3237 (*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n", 3238 ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag, 3239 ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type, 3240 vp->v_mount, vp->v_mountedhere); 3241 3242 (*pr)("v_lock %p v_vnlock %p\n", &vp->v_lock, vp->v_vnlock); 3243 3244 if (full) { 3245 struct buf *bp; 3246 3247 (*pr)("clean bufs:\n"); 3248 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 3249 (*pr)(" bp %p\n", bp); 3250 vfs_buf_print(bp, full, pr); 3251 } 3252 3253 (*pr)("dirty bufs:\n"); 3254 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 3255 (*pr)(" bp %p\n", bp); 3256 vfs_buf_print(bp, full, pr); 3257 } 3258 } 3259 } 3260 3261 void 3262 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...)) 3263 { 3264 char sbuf[256]; 3265 3266 (*pr)("vnodecovered = %p syncer = %p data = %p\n", 3267 mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data); 3268 3269 (*pr)("fs_bshift %d dev_bshift = %d\n", 3270 mp->mnt_fs_bshift,mp->mnt_dev_bshift); 3271 3272 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag); 3273 (*pr)("flag = %s\n", sbuf); 3274 3275 snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag); 3276 (*pr)("iflag = %s\n", sbuf); 3277 3278 (*pr)("refcnt = %d unmounting @ %p updating @ %p\n", mp->mnt_refcnt, 3279 &mp->mnt_unmounting, &mp->mnt_updating); 3280 3281 (*pr)("statvfs cache:\n"); 3282 (*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize); 3283 (*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize); 3284 (*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize); 3285 3286 (*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks); 3287 (*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree); 3288 (*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail); 3289 (*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd); 3290 3291 (*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files); 3292 (*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree); 3293 (*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail); 3294 (*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd); 3295 3296 (*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n", 3297 mp->mnt_stat.f_fsidx.__fsid_val[0], 3298 mp->mnt_stat.f_fsidx.__fsid_val[1]); 3299 3300 (*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner); 3301 (*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax); 3302 3303 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag); 3304 3305 (*pr)("\tflag = %s\n",sbuf); 3306 (*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites); 3307 (*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites); 3308 (*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads); 3309 (*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads); 3310 (*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename); 3311 (*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname); 3312 (*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname); 3313 3314 { 3315 int cnt = 0; 3316 struct vnode *vp; 3317 (*pr)("locked vnodes ="); 3318 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3319 if (VOP_ISLOCKED(vp)) { 3320 if ((++cnt % 6) == 0) { 3321 (*pr)(" %p,\n\t", vp); 3322 } else { 3323 (*pr)(" %p,", vp); 3324 } 3325 } 3326 } 3327 (*pr)("\n"); 3328 } 3329 3330 if (full) { 3331 int cnt = 0; 3332 struct vnode *vp; 3333 (*pr)("all vnodes ="); 3334 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3335 if (!TAILQ_NEXT(vp, v_mntvnodes)) { 3336 (*pr)(" %p", vp); 3337 } else if ((++cnt % 6) == 0) { 3338 (*pr)(" %p,\n\t", vp); 3339 } else { 3340 (*pr)(" %p,", vp); 3341 } 3342 } 3343 (*pr)("\n", vp); 3344 } 3345 } 3346 #endif /* DDB || DEBUGPRINT */ 3347 3348 /* 3349 * Check if a device pointed to by vp is mounted. 3350 * 3351 * Returns: 3352 * EINVAL if it's not a disk 3353 * EBUSY if it's a disk and mounted 3354 * 0 if it's a disk and not mounted 3355 */ 3356 int 3357 rawdev_mounted(struct vnode *vp, struct vnode **bvpp) 3358 { 3359 struct vnode *bvp; 3360 dev_t dev; 3361 int d_type; 3362 3363 bvp = NULL; 3364 dev = vp->v_rdev; 3365 d_type = D_OTHER; 3366 3367 if (iskmemvp(vp)) 3368 return EINVAL; 3369 3370 switch (vp->v_type) { 3371 case VCHR: { 3372 const struct cdevsw *cdev; 3373 3374 cdev = cdevsw_lookup(dev); 3375 if (cdev != NULL) { 3376 dev_t blkdev; 3377 3378 blkdev = devsw_chr2blk(dev); 3379 if (blkdev != NODEV) { 3380 vfinddev(blkdev, VBLK, &bvp); 3381 if (bvp != NULL) 3382 d_type = (cdev->d_flag & D_TYPEMASK); 3383 } 3384 } 3385 3386 break; 3387 } 3388 3389 case VBLK: { 3390 const struct bdevsw *bdev; 3391 3392 bdev = bdevsw_lookup(dev); 3393 if (bdev != NULL) 3394 d_type = (bdev->d_flag & D_TYPEMASK); 3395 3396 bvp = vp; 3397 3398 break; 3399 } 3400 3401 default: 3402 break; 3403 } 3404 3405 if (d_type != D_DISK) 3406 return EINVAL; 3407 3408 if (bvpp != NULL) 3409 *bvpp = bvp; 3410 3411 /* 3412 * XXX: This is bogus. We should be failing the request 3413 * XXX: not only if this specific slice is mounted, but 3414 * XXX: if it's on a disk with any other mounted slice. 3415 */ 3416 if (vfs_mountedon(bvp)) 3417 return EBUSY; 3418 3419 return 0; 3420 } 3421