1 /* $NetBSD: vfs_subr.c,v 1.149 2001/04/16 22:41:10 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998 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. 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 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Copyright (c) 1989, 1993 42 * The Regents of the University of California. All rights reserved. 43 * (c) UNIX System Laboratories, Inc. 44 * All or some portions of this file are derived from material licensed 45 * to the University of California by American Telephone and Telegraph 46 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 47 * the permission of UNIX System Laboratories, Inc. 48 * 49 * Redistribution and use in source and binary forms, with or without 50 * modification, are permitted provided that the following conditions 51 * are met: 52 * 1. Redistributions of source code must retain the above copyright 53 * notice, this list of conditions and the following disclaimer. 54 * 2. Redistributions in binary form must reproduce the above copyright 55 * notice, this list of conditions and the following disclaimer in the 56 * documentation and/or other materials provided with the distribution. 57 * 3. All advertising materials mentioning features or use of this software 58 * must display the following acknowledgement: 59 * This product includes software developed by the University of 60 * California, Berkeley and its contributors. 61 * 4. Neither the name of the University nor the names of its contributors 62 * may be used to endorse or promote products derived from this software 63 * without specific prior written permission. 64 * 65 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 66 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 68 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 69 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 70 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 71 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 72 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 73 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 74 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 75 * SUCH DAMAGE. 76 * 77 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 78 */ 79 80 /* 81 * External virtual filesystem routines 82 */ 83 84 #include "opt_ddb.h" 85 #include "opt_compat_netbsd.h" 86 #include "opt_compat_43.h" 87 88 #include <sys/param.h> 89 #include <sys/systm.h> 90 #include <sys/proc.h> 91 #include <sys/kernel.h> 92 #include <sys/mount.h> 93 #include <sys/time.h> 94 #include <sys/fcntl.h> 95 #include <sys/vnode.h> 96 #include <sys/stat.h> 97 #include <sys/namei.h> 98 #include <sys/ucred.h> 99 #include <sys/buf.h> 100 #include <sys/errno.h> 101 #include <sys/malloc.h> 102 #include <sys/domain.h> 103 #include <sys/mbuf.h> 104 #include <sys/syscallargs.h> 105 #include <sys/device.h> 106 #include <sys/dirent.h> 107 108 #include <miscfs/specfs/specdev.h> 109 #include <miscfs/genfs/genfs.h> 110 #include <miscfs/syncfs/syncfs.h> 111 112 #include <uvm/uvm.h> 113 #include <uvm/uvm_ddb.h> 114 115 #include <sys/sysctl.h> 116 117 enum vtype iftovt_tab[16] = { 118 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 119 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 120 }; 121 const int vttoif_tab[9] = { 122 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 123 S_IFSOCK, S_IFIFO, S_IFMT, 124 }; 125 126 int doforce = 1; /* 1 => permit forcible unmounting */ 127 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 128 129 extern int dovfsusermount; /* 1 => permit any user to mount filesystems */ 130 131 /* 132 * Insq/Remq for the vnode usage lists. 133 */ 134 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 135 #define bufremvn(bp) { \ 136 LIST_REMOVE(bp, b_vnbufs); \ 137 (bp)->b_vnbufs.le_next = NOLIST; \ 138 } 139 /* TAILQ_HEAD(freelst, vnode) vnode_free_list = vnode free list (in vnode.h) */ 140 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list); 141 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list); 142 143 struct mntlist mountlist = /* mounted filesystem list */ 144 CIRCLEQ_HEAD_INITIALIZER(mountlist); 145 struct vfs_list_head vfs_list = /* vfs list */ 146 LIST_HEAD_INITIALIZER(vfs_list); 147 148 struct nfs_public nfs_pub; /* publicly exported FS */ 149 150 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER; 151 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER; 152 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER; 153 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER; 154 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER; 155 156 /* 157 * These define the root filesystem and device. 158 */ 159 struct mount *rootfs; 160 struct vnode *rootvnode; 161 struct device *root_device; /* root device */ 162 163 struct pool vnode_pool; /* memory pool for vnodes */ 164 165 /* 166 * Local declarations. 167 */ 168 void insmntque __P((struct vnode *, struct mount *)); 169 int getdevvp __P((dev_t, struct vnode **, enum vtype)); 170 void vgoneall __P((struct vnode *)); 171 172 static int vfs_hang_addrlist __P((struct mount *, struct netexport *, 173 struct export_args *)); 174 static int vfs_free_netcred __P((struct radix_node *, void *)); 175 static void vfs_free_addrlist __P((struct netexport *)); 176 177 #ifdef DEBUG 178 void printlockedvnodes __P((void)); 179 #endif 180 181 /* 182 * Initialize the vnode management data structures. 183 */ 184 void 185 vntblinit() 186 { 187 188 pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl", 189 0, pool_page_alloc_nointr, pool_page_free_nointr, M_VNODE); 190 191 /* 192 * Initialize the filesystem syncer. 193 */ 194 vn_initialize_syncerd(); 195 } 196 197 /* 198 * Mark a mount point as busy. Used to synchronize access and to delay 199 * unmounting. Interlock is not released on failure. 200 */ 201 int 202 vfs_busy(mp, flags, interlkp) 203 struct mount *mp; 204 int flags; 205 struct simplelock *interlkp; 206 { 207 int lkflags; 208 209 while (mp->mnt_flag & MNT_UNMOUNT) { 210 int gone; 211 212 if (flags & LK_NOWAIT) 213 return (ENOENT); 214 if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL 215 && mp->mnt_unmounter == curproc) 216 return (EDEADLK); 217 if (interlkp) 218 simple_unlock(interlkp); 219 /* 220 * Since all busy locks are shared except the exclusive 221 * lock granted when unmounting, the only place that a 222 * wakeup needs to be done is at the release of the 223 * exclusive lock at the end of dounmount. 224 * 225 * XXX MP: add spinlock protecting mnt_wcnt here once you 226 * can atomically unlock-and-sleep. 227 */ 228 mp->mnt_wcnt++; 229 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 230 mp->mnt_wcnt--; 231 gone = mp->mnt_flag & MNT_GONE; 232 233 if (mp->mnt_wcnt == 0) 234 wakeup(&mp->mnt_wcnt); 235 if (interlkp) 236 simple_lock(interlkp); 237 if (gone) 238 return (ENOENT); 239 } 240 lkflags = LK_SHARED; 241 if (interlkp) 242 lkflags |= LK_INTERLOCK; 243 if (lockmgr(&mp->mnt_lock, lkflags, interlkp)) 244 panic("vfs_busy: unexpected lock failure"); 245 return (0); 246 } 247 248 /* 249 * Free a busy filesystem. 250 */ 251 void 252 vfs_unbusy(mp) 253 struct mount *mp; 254 { 255 256 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL); 257 } 258 259 /* 260 * Lookup a filesystem type, and if found allocate and initialize 261 * a mount structure for it. 262 * 263 * Devname is usually updated by mount(8) after booting. 264 */ 265 int 266 vfs_rootmountalloc(fstypename, devname, mpp) 267 char *fstypename; 268 char *devname; 269 struct mount **mpp; 270 { 271 struct vfsops *vfsp = NULL; 272 struct mount *mp; 273 274 for (vfsp = LIST_FIRST(&vfs_list); vfsp != NULL; 275 vfsp = LIST_NEXT(vfsp, vfs_list)) 276 if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN)) 277 break; 278 279 if (vfsp == NULL) 280 return (ENODEV); 281 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 282 memset((char *)mp, 0, (u_long)sizeof(struct mount)); 283 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 284 (void)vfs_busy(mp, LK_NOWAIT, 0); 285 LIST_INIT(&mp->mnt_vnodelist); 286 mp->mnt_op = vfsp; 287 mp->mnt_flag = MNT_RDONLY; 288 mp->mnt_vnodecovered = NULLVP; 289 vfsp->vfs_refcount++; 290 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN); 291 mp->mnt_stat.f_mntonname[0] = '/'; 292 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 293 *mpp = mp; 294 return (0); 295 } 296 297 /* 298 * Lookup a mount point by filesystem identifier. 299 */ 300 struct mount * 301 vfs_getvfs(fsid) 302 fsid_t *fsid; 303 { 304 struct mount *mp; 305 306 simple_lock(&mountlist_slock); 307 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 308 mp = mp->mnt_list.cqe_next) { 309 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 310 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 311 simple_unlock(&mountlist_slock); 312 return (mp); 313 } 314 } 315 simple_unlock(&mountlist_slock); 316 return ((struct mount *)0); 317 } 318 319 /* 320 * Get a new unique fsid 321 */ 322 void 323 vfs_getnewfsid(mp) 324 struct mount *mp; 325 { 326 static u_short xxxfs_mntid; 327 fsid_t tfsid; 328 int mtype; 329 330 simple_lock(&mntid_slock); 331 mtype = makefstype(mp->mnt_op->vfs_name); 332 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 333 mp->mnt_stat.f_fsid.val[1] = mtype; 334 if (xxxfs_mntid == 0) 335 ++xxxfs_mntid; 336 tfsid.val[0] = makedev((nblkdev + mtype) & 0xff, xxxfs_mntid); 337 tfsid.val[1] = mtype; 338 if (mountlist.cqh_first != (void *)&mountlist) { 339 while (vfs_getvfs(&tfsid)) { 340 tfsid.val[0]++; 341 xxxfs_mntid++; 342 } 343 } 344 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 345 simple_unlock(&mntid_slock); 346 } 347 348 /* 349 * Make a 'unique' number from a mount type name. 350 */ 351 long 352 makefstype(type) 353 const char *type; 354 { 355 long rv; 356 357 for (rv = 0; *type; type++) { 358 rv <<= 2; 359 rv ^= *type; 360 } 361 return rv; 362 } 363 364 365 /* 366 * Set vnode attributes to VNOVAL 367 */ 368 void 369 vattr_null(vap) 370 struct vattr *vap; 371 { 372 373 vap->va_type = VNON; 374 375 /* 376 * Assign individually so that it is safe even if size and 377 * sign of each member are varied. 378 */ 379 vap->va_mode = VNOVAL; 380 vap->va_nlink = VNOVAL; 381 vap->va_uid = VNOVAL; 382 vap->va_gid = VNOVAL; 383 vap->va_fsid = VNOVAL; 384 vap->va_fileid = VNOVAL; 385 vap->va_size = VNOVAL; 386 vap->va_blocksize = VNOVAL; 387 vap->va_atime.tv_sec = 388 vap->va_mtime.tv_sec = 389 vap->va_ctime.tv_sec = VNOVAL; 390 vap->va_atime.tv_nsec = 391 vap->va_mtime.tv_nsec = 392 vap->va_ctime.tv_nsec = VNOVAL; 393 vap->va_gen = VNOVAL; 394 vap->va_flags = VNOVAL; 395 vap->va_rdev = VNOVAL; 396 vap->va_bytes = VNOVAL; 397 vap->va_vaflags = 0; 398 } 399 400 /* 401 * Routines having to do with the management of the vnode table. 402 */ 403 extern int (**dead_vnodeop_p) __P((void *)); 404 long numvnodes; 405 406 /* 407 * Return the next vnode from the free list. 408 */ 409 int 410 getnewvnode(tag, mp, vops, vpp) 411 enum vtagtype tag; 412 struct mount *mp; 413 int (**vops) __P((void *)); 414 struct vnode **vpp; 415 { 416 extern struct uvm_pagerops uvm_vnodeops; 417 struct uvm_object *uobj; 418 struct proc *p = curproc; /* XXX */ 419 struct freelst *listhd; 420 static int toggle; 421 struct vnode *vp; 422 int error = 0; 423 #ifdef DIAGNOSTIC 424 int s; 425 #endif 426 if (mp) { 427 /* 428 * Mark filesystem busy while we're creating a vnode. 429 * If unmount is in progress, this will wait; if the 430 * unmount succeeds (only if umount -f), this will 431 * return an error. If the unmount fails, we'll keep 432 * going afterwards. 433 * (This puts the per-mount vnode list logically under 434 * the protection of the vfs_busy lock). 435 */ 436 error = vfs_busy(mp, LK_RECURSEFAIL, 0); 437 if (error && error != EDEADLK) 438 return error; 439 } 440 441 /* 442 * We must choose whether to allocate a new vnode or recycle an 443 * existing one. The criterion for allocating a new one is that 444 * the total number of vnodes is less than the number desired or 445 * there are no vnodes on either free list. Generally we only 446 * want to recycle vnodes that have no buffers associated with 447 * them, so we look first on the vnode_free_list. If it is empty, 448 * we next consider vnodes with referencing buffers on the 449 * vnode_hold_list. The toggle ensures that half the time we 450 * will use a buffer from the vnode_hold_list, and half the time 451 * we will allocate a new one unless the list has grown to twice 452 * the desired size. We are reticent to recycle vnodes from the 453 * vnode_hold_list because we will lose the identity of all its 454 * referencing buffers. 455 */ 456 457 toggle ^= 1; 458 if (numvnodes > 2 * desiredvnodes) 459 toggle = 0; 460 461 simple_lock(&vnode_free_list_slock); 462 if (numvnodes < desiredvnodes || 463 (TAILQ_FIRST(listhd = &vnode_free_list) == NULL && 464 (TAILQ_FIRST(listhd = &vnode_hold_list) == NULL || toggle))) { 465 simple_unlock(&vnode_free_list_slock); 466 vp = pool_get(&vnode_pool, PR_WAITOK); 467 memset(vp, 0, sizeof(*vp)); 468 simple_lock_init(&vp->v_interlock); 469 numvnodes++; 470 } else { 471 for (vp = TAILQ_FIRST(listhd); vp != NULLVP; 472 vp = TAILQ_NEXT(vp, v_freelist)) { 473 if (simple_lock_try(&vp->v_interlock)) { 474 if ((vp->v_flag & VLAYER) == 0) { 475 break; 476 } 477 if (VOP_ISLOCKED(vp) == 0) 478 break; 479 else 480 simple_unlock(&vp->v_interlock); 481 } 482 } 483 /* 484 * Unless this is a bad time of the month, at most 485 * the first NCPUS items on the free list are 486 * locked, so this is close enough to being empty. 487 */ 488 if (vp == NULLVP) { 489 simple_unlock(&vnode_free_list_slock); 490 if (mp && error != EDEADLK) 491 vfs_unbusy(mp); 492 tablefull("vnode", "increase kern.maxvnodes or NVNODE"); 493 *vpp = 0; 494 return (ENFILE); 495 } 496 if (vp->v_usecount) 497 panic("free vnode isn't, vp %p", vp); 498 TAILQ_REMOVE(listhd, vp, v_freelist); 499 /* see comment on why 0xdeadb is set at end of vgone (below) */ 500 vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; 501 simple_unlock(&vnode_free_list_slock); 502 vp->v_lease = NULL; 503 if (vp->v_type != VBAD) 504 vgonel(vp, p); 505 else 506 simple_unlock(&vp->v_interlock); 507 #ifdef DIAGNOSTIC 508 if (vp->v_data) 509 panic("cleaned vnode isn't, vp %p", vp); 510 s = splbio(); 511 if (vp->v_numoutput) 512 panic("clean vnode has pending I/O's, vp %p", vp); 513 splx(s); 514 #endif 515 vp->v_flag = 0; 516 vp->v_lastr = 0; 517 vp->v_ralen = 0; 518 vp->v_maxra = 0; 519 vp->v_lastw = 0; 520 vp->v_lasta = 0; 521 vp->v_cstart = 0; 522 vp->v_clen = 0; 523 vp->v_socket = 0; 524 } 525 vp->v_type = VNON; 526 vp->v_vnlock = &vp->v_lock; 527 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 528 lockinit(&vp->v_glock, PVFS, "glock", 0, 0); 529 cache_purge(vp); 530 vp->v_tag = tag; 531 vp->v_op = vops; 532 insmntque(vp, mp); 533 *vpp = vp; 534 vp->v_usecount = 1; 535 vp->v_data = 0; 536 simple_lock_init(&vp->v_uvm.u_obj.vmobjlock); 537 538 /* 539 * initialize uvm_object within vnode. 540 */ 541 542 uobj = &vp->v_uvm.u_obj; 543 uobj->pgops = &uvm_vnodeops; 544 TAILQ_INIT(&uobj->memq); 545 vp->v_uvm.u_size = VSIZENOTSET; 546 547 if (mp && error != EDEADLK) 548 vfs_unbusy(mp); 549 return (0); 550 } 551 552 /* 553 * This is really just the reverse of getnewvnode(). Needed for 554 * VFS_VGET functions who may need to push back a vnode in case 555 * of a locking race. 556 */ 557 void 558 ungetnewvnode(vp) 559 struct vnode *vp; 560 { 561 #ifdef DIAGNOSTIC 562 if (vp->v_usecount != 1) 563 panic("ungetnewvnode: busy vnode"); 564 #endif 565 vp->v_usecount--; 566 insmntque(vp, NULL); 567 vp->v_type = VBAD; 568 569 simple_lock(&vp->v_interlock); 570 /* 571 * Insert at head of LRU list 572 */ 573 simple_lock(&vnode_free_list_slock); 574 if (vp->v_holdcnt > 0) 575 TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist); 576 else 577 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 578 simple_unlock(&vnode_free_list_slock); 579 simple_unlock(&vp->v_interlock); 580 } 581 582 /* 583 * Move a vnode from one mount queue to another. 584 */ 585 void 586 insmntque(vp, mp) 587 struct vnode *vp; 588 struct mount *mp; 589 { 590 591 #ifdef DIAGNOSTIC 592 if ((mp != NULL) && 593 (mp->mnt_flag & MNT_UNMOUNT) && 594 !(mp->mnt_flag & MNT_SOFTDEP) && 595 vp->v_tag != VT_VFS) { 596 panic("insmntque into dying filesystem"); 597 } 598 #endif 599 600 simple_lock(&mntvnode_slock); 601 /* 602 * Delete from old mount point vnode list, if on one. 603 */ 604 if (vp->v_mount != NULL) 605 LIST_REMOVE(vp, v_mntvnodes); 606 /* 607 * Insert into list of vnodes for the new mount point, if available. 608 */ 609 if ((vp->v_mount = mp) != NULL) 610 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 611 simple_unlock(&mntvnode_slock); 612 } 613 614 /* 615 * Update outstanding I/O count and do wakeup if requested. 616 */ 617 void 618 vwakeup(bp) 619 struct buf *bp; 620 { 621 struct vnode *vp; 622 623 if ((vp = bp->b_vp) != NULL) { 624 if (--vp->v_numoutput < 0) 625 panic("vwakeup: neg numoutput, vp %p", vp); 626 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 627 vp->v_flag &= ~VBWAIT; 628 wakeup((caddr_t)&vp->v_numoutput); 629 } 630 } 631 } 632 633 /* 634 * Flush out and invalidate all buffers associated with a vnode. 635 * Called with the underlying vnode locked, which should prevent new dirty 636 * buffers from being queued. 637 */ 638 int 639 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 640 struct vnode *vp; 641 int flags; 642 struct ucred *cred; 643 struct proc *p; 644 int slpflag, slptimeo; 645 { 646 struct uvm_object *uobj = &vp->v_uvm.u_obj; 647 struct buf *bp, *nbp; 648 int s, error, rv; 649 int flushflags = PGO_ALLPAGES|PGO_FREE|PGO_SYNCIO| 650 (flags & V_SAVE ? PGO_CLEANIT : 0); 651 652 /* XXXUBC this doesn't look at flags or slp* */ 653 if (vp->v_type == VREG) { 654 simple_lock(&uobj->vmobjlock); 655 rv = (uobj->pgops->pgo_flush)(uobj, 0, 0, flushflags); 656 simple_unlock(&uobj->vmobjlock); 657 if (!rv) { 658 return EIO; 659 } 660 } 661 if (flags & V_SAVE) { 662 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p); 663 if (error) 664 return (error); 665 #ifdef DIAGNOSTIC 666 s = splbio(); 667 if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd)) 668 panic("vinvalbuf: dirty bufs, vp %p", vp); 669 splx(s); 670 #endif 671 } 672 673 s = splbio(); 674 675 restart: 676 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 677 nbp = LIST_NEXT(bp, b_vnbufs); 678 if (bp->b_flags & B_BUSY) { 679 bp->b_flags |= B_WANTED; 680 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 681 "vinvalbuf", slptimeo); 682 if (error) { 683 splx(s); 684 return (error); 685 } 686 goto restart; 687 } 688 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 689 brelse(bp); 690 } 691 692 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 693 nbp = LIST_NEXT(bp, b_vnbufs); 694 if (bp->b_flags & B_BUSY) { 695 bp->b_flags |= B_WANTED; 696 error = tsleep((caddr_t)bp, slpflag | (PRIBIO + 1), 697 "vinvalbuf", slptimeo); 698 if (error) { 699 splx(s); 700 return (error); 701 } 702 goto restart; 703 } 704 /* 705 * XXX Since there are no node locks for NFS, I believe 706 * there is a slight chance that a delayed write will 707 * occur while sleeping just above, so check for it. 708 */ 709 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 710 #ifdef DEBUG 711 printf("buffer still DELWRI\n"); 712 #endif 713 bp->b_flags |= B_BUSY | B_VFLUSH; 714 VOP_BWRITE(bp); 715 goto restart; 716 } 717 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 718 brelse(bp); 719 } 720 721 #ifdef DIAGNOSTIC 722 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 723 panic("vinvalbuf: flush failed, vp %p", vp); 724 #endif 725 726 splx(s); 727 728 return (0); 729 } 730 731 /* 732 * Destroy any in core blocks past the truncation length. 733 * Called with the underlying vnode locked, which should prevent new dirty 734 * buffers from being queued. 735 */ 736 int 737 vtruncbuf(vp, lbn, slpflag, slptimeo) 738 struct vnode *vp; 739 daddr_t lbn; 740 int slpflag, slptimeo; 741 { 742 struct uvm_object *uobj = &vp->v_uvm.u_obj; 743 struct buf *bp, *nbp; 744 int s, error, rv; 745 746 s = splbio(); 747 if (vp->v_type == VREG) { 748 simple_lock(&uobj->vmobjlock); 749 rv = (uobj->pgops->pgo_flush)(uobj, 750 round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift), 0, 751 PGO_FREE|PGO_SYNCIO); 752 simple_unlock(&uobj->vmobjlock); 753 if (!rv) { 754 splx(s); 755 return EIO; 756 } 757 } 758 759 restart: 760 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 761 nbp = LIST_NEXT(bp, b_vnbufs); 762 if (bp->b_lblkno < lbn) 763 continue; 764 if (bp->b_flags & B_BUSY) { 765 bp->b_flags |= B_WANTED; 766 error = tsleep(bp, slpflag | (PRIBIO + 1), 767 "vtruncbuf", slptimeo); 768 if (error) { 769 splx(s); 770 return (error); 771 } 772 goto restart; 773 } 774 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 775 brelse(bp); 776 } 777 778 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 779 nbp = LIST_NEXT(bp, b_vnbufs); 780 if (bp->b_lblkno < lbn) 781 continue; 782 if (bp->b_flags & B_BUSY) { 783 bp->b_flags |= B_WANTED; 784 error = tsleep(bp, slpflag | (PRIBIO + 1), 785 "vtruncbuf", slptimeo); 786 if (error) { 787 splx(s); 788 return (error); 789 } 790 goto restart; 791 } 792 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 793 brelse(bp); 794 } 795 796 splx(s); 797 798 return (0); 799 } 800 801 void 802 vflushbuf(vp, sync) 803 struct vnode *vp; 804 int sync; 805 { 806 struct uvm_object *uobj = &vp->v_uvm.u_obj; 807 struct buf *bp, *nbp; 808 int s; 809 810 if (vp->v_type == VREG) { 811 int flags = PGO_CLEANIT|PGO_ALLPAGES| (sync ? PGO_SYNCIO : 0); 812 813 simple_lock(&uobj->vmobjlock); 814 (uobj->pgops->pgo_flush)(uobj, 0, 0, flags); 815 simple_unlock(&uobj->vmobjlock); 816 } 817 818 loop: 819 s = splbio(); 820 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 821 nbp = LIST_NEXT(bp, b_vnbufs); 822 if ((bp->b_flags & B_BUSY)) 823 continue; 824 if ((bp->b_flags & B_DELWRI) == 0) 825 panic("vflushbuf: not dirty, bp %p", bp); 826 bp->b_flags |= B_BUSY | B_VFLUSH; 827 splx(s); 828 /* 829 * Wait for I/O associated with indirect blocks to complete, 830 * since there is no way to quickly wait for them below. 831 */ 832 if (bp->b_vp == vp || sync == 0) 833 (void) bawrite(bp); 834 else 835 (void) bwrite(bp); 836 goto loop; 837 } 838 if (sync == 0) { 839 splx(s); 840 return; 841 } 842 while (vp->v_numoutput) { 843 vp->v_flag |= VBWAIT; 844 tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0); 845 } 846 splx(s); 847 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 848 vprint("vflushbuf: dirty", vp); 849 goto loop; 850 } 851 } 852 853 /* 854 * Associate a buffer with a vnode. 855 */ 856 void 857 bgetvp(vp, bp) 858 struct vnode *vp; 859 struct buf *bp; 860 { 861 int s; 862 863 if (bp->b_vp) 864 panic("bgetvp: not free, bp %p", bp); 865 VHOLD(vp); 866 s = splbio(); 867 bp->b_vp = vp; 868 if (vp->v_type == VBLK || vp->v_type == VCHR) 869 bp->b_dev = vp->v_rdev; 870 else 871 bp->b_dev = NODEV; 872 /* 873 * Insert onto list for new vnode. 874 */ 875 bufinsvn(bp, &vp->v_cleanblkhd); 876 splx(s); 877 } 878 879 /* 880 * Disassociate a buffer from a vnode. 881 */ 882 void 883 brelvp(bp) 884 struct buf *bp; 885 { 886 struct vnode *vp; 887 int s; 888 889 if (bp->b_vp == NULL) 890 panic("brelvp: vp NULL, bp %p", bp); 891 892 s = splbio(); 893 vp = bp->b_vp; 894 /* 895 * Delete from old vnode list, if on one. 896 */ 897 if (bp->b_vnbufs.le_next != NOLIST) 898 bufremvn(bp); 899 900 if (vp->v_type != VREG && (vp->v_flag & VONWORKLST) && 901 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 902 vp->v_flag &= ~VONWORKLST; 903 LIST_REMOVE(vp, v_synclist); 904 } 905 906 bp->b_vp = NULL; 907 HOLDRELE(vp); 908 splx(s); 909 } 910 911 /* 912 * Reassign a buffer from one vnode to another. 913 * Used to assign file specific control information 914 * (indirect blocks) to the vnode to which they belong. 915 * 916 * This function must be called at splbio(). 917 */ 918 void 919 reassignbuf(bp, newvp) 920 struct buf *bp; 921 struct vnode *newvp; 922 { 923 struct buflists *listheadp; 924 int delay; 925 926 /* 927 * Delete from old vnode list, if on one. 928 */ 929 if (bp->b_vnbufs.le_next != NOLIST) 930 bufremvn(bp); 931 /* 932 * If dirty, put on list of dirty buffers; 933 * otherwise insert onto list of clean buffers. 934 */ 935 if ((bp->b_flags & B_DELWRI) == 0) { 936 listheadp = &newvp->v_cleanblkhd; 937 if (newvp->v_type != VREG && 938 (newvp->v_flag & VONWORKLST) && 939 LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) { 940 newvp->v_flag &= ~VONWORKLST; 941 LIST_REMOVE(newvp, v_synclist); 942 } 943 } else { 944 listheadp = &newvp->v_dirtyblkhd; 945 if ((newvp->v_flag & VONWORKLST) == 0) { 946 switch (newvp->v_type) { 947 case VDIR: 948 delay = dirdelay; 949 break; 950 case VBLK: 951 if (newvp->v_specmountpoint != NULL) { 952 delay = metadelay; 953 break; 954 } 955 /* fall through */ 956 default: 957 delay = filedelay; 958 break; 959 } 960 if (!newvp->v_mount || 961 (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0) 962 vn_syncer_add_to_worklist(newvp, delay); 963 } 964 } 965 bufinsvn(bp, listheadp); 966 } 967 968 /* 969 * Create a vnode for a block device. 970 * Used for root filesystem and swap areas. 971 * Also used for memory file system special devices. 972 */ 973 int 974 bdevvp(dev, vpp) 975 dev_t dev; 976 struct vnode **vpp; 977 { 978 979 return (getdevvp(dev, vpp, VBLK)); 980 } 981 982 /* 983 * Create a vnode for a character device. 984 * Used for kernfs and some console handling. 985 */ 986 int 987 cdevvp(dev, vpp) 988 dev_t dev; 989 struct vnode **vpp; 990 { 991 992 return (getdevvp(dev, vpp, VCHR)); 993 } 994 995 /* 996 * Create a vnode for a device. 997 * Used by bdevvp (block device) for root file system etc., 998 * and by cdevvp (character device) for console and kernfs. 999 */ 1000 int 1001 getdevvp(dev, vpp, type) 1002 dev_t dev; 1003 struct vnode **vpp; 1004 enum vtype type; 1005 { 1006 struct vnode *vp; 1007 struct vnode *nvp; 1008 int error; 1009 1010 if (dev == NODEV) { 1011 *vpp = NULLVP; 1012 return (0); 1013 } 1014 error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp); 1015 if (error) { 1016 *vpp = NULLVP; 1017 return (error); 1018 } 1019 vp = nvp; 1020 vp->v_type = type; 1021 if ((nvp = checkalias(vp, dev, NULL)) != 0) { 1022 vput(vp); 1023 vp = nvp; 1024 } 1025 *vpp = vp; 1026 return (0); 1027 } 1028 1029 /* 1030 * Check to see if the new vnode represents a special device 1031 * for which we already have a vnode (either because of 1032 * bdevvp() or because of a different vnode representing 1033 * the same block device). If such an alias exists, deallocate 1034 * the existing contents and return the aliased vnode. The 1035 * caller is responsible for filling it with its new contents. 1036 */ 1037 struct vnode * 1038 checkalias(nvp, nvp_rdev, mp) 1039 struct vnode *nvp; 1040 dev_t nvp_rdev; 1041 struct mount *mp; 1042 { 1043 struct proc *p = curproc; /* XXX */ 1044 struct vnode *vp; 1045 struct vnode **vpp; 1046 1047 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1048 return (NULLVP); 1049 1050 vpp = &speclisth[SPECHASH(nvp_rdev)]; 1051 loop: 1052 simple_lock(&spechash_slock); 1053 for (vp = *vpp; vp; vp = vp->v_specnext) { 1054 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 1055 continue; 1056 /* 1057 * Alias, but not in use, so flush it out. 1058 */ 1059 simple_lock(&vp->v_interlock); 1060 if (vp->v_usecount == 0) { 1061 simple_unlock(&spechash_slock); 1062 vgonel(vp, p); 1063 goto loop; 1064 } 1065 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) { 1066 simple_unlock(&spechash_slock); 1067 goto loop; 1068 } 1069 break; 1070 } 1071 if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) { 1072 MALLOC(nvp->v_specinfo, struct specinfo *, 1073 sizeof(struct specinfo), M_VNODE, M_WAITOK); 1074 nvp->v_rdev = nvp_rdev; 1075 nvp->v_hashchain = vpp; 1076 nvp->v_specnext = *vpp; 1077 nvp->v_specmountpoint = NULL; 1078 simple_unlock(&spechash_slock); 1079 nvp->v_speclockf = NULL; 1080 *vpp = nvp; 1081 if (vp != NULLVP) { 1082 nvp->v_flag |= VALIASED; 1083 vp->v_flag |= VALIASED; 1084 vput(vp); 1085 } 1086 return (NULLVP); 1087 } 1088 simple_unlock(&spechash_slock); 1089 VOP_UNLOCK(vp, 0); 1090 simple_lock(&vp->v_interlock); 1091 vclean(vp, 0, p); 1092 vp->v_op = nvp->v_op; 1093 vp->v_tag = nvp->v_tag; 1094 vp->v_vnlock = &vp->v_lock; 1095 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 1096 nvp->v_type = VNON; 1097 insmntque(vp, mp); 1098 return (vp); 1099 } 1100 1101 /* 1102 * Grab a particular vnode from the free list, increment its 1103 * reference count and lock it. If the vnode lock bit is set the 1104 * vnode is being eliminated in vgone. In that case, we can not 1105 * grab the vnode, so the process is awakened when the transition is 1106 * completed, and an error returned to indicate that the vnode is no 1107 * longer usable (possibly having been changed to a new file system type). 1108 */ 1109 int 1110 vget(vp, flags) 1111 struct vnode *vp; 1112 int flags; 1113 { 1114 int error; 1115 1116 /* 1117 * If the vnode is in the process of being cleaned out for 1118 * another use, we wait for the cleaning to finish and then 1119 * return failure. Cleaning is determined by checking that 1120 * the VXLOCK flag is set. 1121 */ 1122 1123 if ((flags & LK_INTERLOCK) == 0) 1124 simple_lock(&vp->v_interlock); 1125 if (vp->v_flag & VXLOCK) { 1126 if (flags & LK_NOWAIT) { 1127 simple_unlock(&vp->v_interlock); 1128 return EBUSY; 1129 } 1130 vp->v_flag |= VXWANT; 1131 ltsleep((caddr_t)vp, PINOD|PNORELOCK, 1132 "vget", 0, &vp->v_interlock); 1133 return (ENOENT); 1134 } 1135 if (vp->v_usecount == 0) { 1136 simple_lock(&vnode_free_list_slock); 1137 if (vp->v_holdcnt > 0) 1138 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 1139 else 1140 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1141 simple_unlock(&vnode_free_list_slock); 1142 } 1143 vp->v_usecount++; 1144 #ifdef DIAGNOSTIC 1145 if (vp->v_usecount == 0) { 1146 vprint("vget", vp); 1147 panic("vget: usecount overflow, vp %p", vp); 1148 } 1149 #endif 1150 if (flags & LK_TYPE_MASK) { 1151 if ((error = vn_lock(vp, flags | LK_INTERLOCK))) { 1152 /* 1153 * must expand vrele here because we do not want 1154 * to call VOP_INACTIVE if the reference count 1155 * drops back to zero since it was never really 1156 * active. We must remove it from the free list 1157 * before sleeping so that multiple processes do 1158 * not try to recycle it. 1159 */ 1160 simple_lock(&vp->v_interlock); 1161 vp->v_usecount--; 1162 if (vp->v_usecount > 0) { 1163 simple_unlock(&vp->v_interlock); 1164 return (error); 1165 } 1166 /* 1167 * insert at tail of LRU list 1168 */ 1169 simple_lock(&vnode_free_list_slock); 1170 if (vp->v_holdcnt > 0) 1171 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, 1172 v_freelist); 1173 else 1174 TAILQ_INSERT_TAIL(&vnode_free_list, vp, 1175 v_freelist); 1176 simple_unlock(&vnode_free_list_slock); 1177 simple_unlock(&vp->v_interlock); 1178 } 1179 return (error); 1180 } 1181 simple_unlock(&vp->v_interlock); 1182 return (0); 1183 } 1184 1185 /* 1186 * vput(), just unlock and vrele() 1187 */ 1188 void 1189 vput(vp) 1190 struct vnode *vp; 1191 { 1192 struct proc *p = curproc; /* XXX */ 1193 1194 #ifdef DIAGNOSTIC 1195 if (vp == NULL) 1196 panic("vput: null vp"); 1197 #endif 1198 simple_lock(&vp->v_interlock); 1199 vp->v_usecount--; 1200 if (vp->v_usecount > 0) { 1201 simple_unlock(&vp->v_interlock); 1202 VOP_UNLOCK(vp, 0); 1203 return; 1204 } 1205 #ifdef DIAGNOSTIC 1206 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1207 vprint("vput: bad ref count", vp); 1208 panic("vput: ref cnt"); 1209 } 1210 #endif 1211 /* 1212 * Insert at tail of LRU list. 1213 */ 1214 simple_lock(&vnode_free_list_slock); 1215 if (vp->v_holdcnt > 0) 1216 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1217 else 1218 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1219 simple_unlock(&vnode_free_list_slock); 1220 if (vp->v_flag & VTEXT) { 1221 uvmexp.vtextpages -= vp->v_uvm.u_obj.uo_npages; 1222 uvmexp.vnodepages += vp->v_uvm.u_obj.uo_npages; 1223 } 1224 vp->v_flag &= ~VTEXT; 1225 simple_unlock(&vp->v_interlock); 1226 VOP_INACTIVE(vp, p); 1227 } 1228 1229 /* 1230 * Vnode release. 1231 * If count drops to zero, call inactive routine and return to freelist. 1232 */ 1233 void 1234 vrele(vp) 1235 struct vnode *vp; 1236 { 1237 struct proc *p = curproc; /* XXX */ 1238 1239 #ifdef DIAGNOSTIC 1240 if (vp == NULL) 1241 panic("vrele: null vp"); 1242 #endif 1243 simple_lock(&vp->v_interlock); 1244 vp->v_usecount--; 1245 if (vp->v_usecount > 0) { 1246 simple_unlock(&vp->v_interlock); 1247 return; 1248 } 1249 #ifdef DIAGNOSTIC 1250 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1251 vprint("vrele: bad ref count", vp); 1252 panic("vrele: ref cnt vp %p", vp); 1253 } 1254 #endif 1255 /* 1256 * Insert at tail of LRU list. 1257 */ 1258 simple_lock(&vnode_free_list_slock); 1259 if (vp->v_holdcnt > 0) 1260 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1261 else 1262 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1263 simple_unlock(&vnode_free_list_slock); 1264 if (vp->v_flag & VTEXT) { 1265 uvmexp.vtextpages -= vp->v_uvm.u_obj.uo_npages; 1266 uvmexp.vnodepages += vp->v_uvm.u_obj.uo_npages; 1267 } 1268 vp->v_flag &= ~VTEXT; 1269 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) 1270 VOP_INACTIVE(vp, p); 1271 } 1272 1273 #ifdef DIAGNOSTIC 1274 /* 1275 * Page or buffer structure gets a reference. 1276 */ 1277 void 1278 vhold(vp) 1279 struct vnode *vp; 1280 { 1281 1282 /* 1283 * If it is on the freelist and the hold count is currently 1284 * zero, move it to the hold list. The test of the back 1285 * pointer and the use reference count of zero is because 1286 * it will be removed from a free list by getnewvnode, 1287 * but will not have its reference count incremented until 1288 * after calling vgone. If the reference count were 1289 * incremented first, vgone would (incorrectly) try to 1290 * close the previous instance of the underlying object. 1291 * So, the back pointer is explicitly set to `0xdeadb' in 1292 * getnewvnode after removing it from a freelist to ensure 1293 * that we do not try to move it here. 1294 */ 1295 simple_lock(&vp->v_interlock); 1296 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1297 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1298 simple_lock(&vnode_free_list_slock); 1299 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1300 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1301 simple_unlock(&vnode_free_list_slock); 1302 } 1303 vp->v_holdcnt++; 1304 simple_unlock(&vp->v_interlock); 1305 } 1306 1307 /* 1308 * Page or buffer structure frees a reference. 1309 */ 1310 void 1311 holdrele(vp) 1312 struct vnode *vp; 1313 { 1314 1315 simple_lock(&vp->v_interlock); 1316 if (vp->v_holdcnt <= 0) 1317 panic("holdrele: holdcnt vp %p", vp); 1318 vp->v_holdcnt--; 1319 1320 /* 1321 * If it is on the holdlist and the hold count drops to 1322 * zero, move it to the free list. The test of the back 1323 * pointer and the use reference count of zero is because 1324 * it will be removed from a free list by getnewvnode, 1325 * but will not have its reference count incremented until 1326 * after calling vgone. If the reference count were 1327 * incremented first, vgone would (incorrectly) try to 1328 * close the previous instance of the underlying object. 1329 * So, the back pointer is explicitly set to `0xdeadb' in 1330 * getnewvnode after removing it from a freelist to ensure 1331 * that we do not try to move it here. 1332 */ 1333 1334 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1335 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1336 simple_lock(&vnode_free_list_slock); 1337 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 1338 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1339 simple_unlock(&vnode_free_list_slock); 1340 } 1341 simple_unlock(&vp->v_interlock); 1342 } 1343 1344 /* 1345 * Vnode reference. 1346 */ 1347 void 1348 vref(vp) 1349 struct vnode *vp; 1350 { 1351 1352 simple_lock(&vp->v_interlock); 1353 if (vp->v_usecount <= 0) 1354 panic("vref used where vget required, vp %p", vp); 1355 vp->v_usecount++; 1356 #ifdef DIAGNOSTIC 1357 if (vp->v_usecount == 0) { 1358 vprint("vref", vp); 1359 panic("vref: usecount overflow, vp %p", vp); 1360 } 1361 #endif 1362 simple_unlock(&vp->v_interlock); 1363 } 1364 #endif /* DIAGNOSTIC */ 1365 1366 /* 1367 * Remove any vnodes in the vnode table belonging to mount point mp. 1368 * 1369 * If MNT_NOFORCE is specified, there should not be any active ones, 1370 * return error if any are found (nb: this is a user error, not a 1371 * system error). If MNT_FORCE is specified, detach any active vnodes 1372 * that are found. 1373 */ 1374 #ifdef DEBUG 1375 int busyprt = 0; /* print out busy vnodes */ 1376 struct ctldebug debug1 = { "busyprt", &busyprt }; 1377 #endif 1378 1379 int 1380 vflush(mp, skipvp, flags) 1381 struct mount *mp; 1382 struct vnode *skipvp; 1383 int flags; 1384 { 1385 struct proc *p = curproc; /* XXX */ 1386 struct vnode *vp, *nvp; 1387 int busy = 0; 1388 1389 simple_lock(&mntvnode_slock); 1390 loop: 1391 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 1392 if (vp->v_mount != mp) 1393 goto loop; 1394 nvp = vp->v_mntvnodes.le_next; 1395 /* 1396 * Skip over a selected vnode. 1397 */ 1398 if (vp == skipvp) 1399 continue; 1400 simple_lock(&vp->v_interlock); 1401 /* 1402 * Skip over a vnodes marked VSYSTEM. 1403 */ 1404 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1405 simple_unlock(&vp->v_interlock); 1406 continue; 1407 } 1408 /* 1409 * If WRITECLOSE is set, only flush out regular file 1410 * vnodes open for writing. 1411 */ 1412 if ((flags & WRITECLOSE) && 1413 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1414 simple_unlock(&vp->v_interlock); 1415 continue; 1416 } 1417 /* 1418 * With v_usecount == 0, all we need to do is clear 1419 * out the vnode data structures and we are done. 1420 */ 1421 if (vp->v_usecount == 0) { 1422 simple_unlock(&mntvnode_slock); 1423 vgonel(vp, p); 1424 simple_lock(&mntvnode_slock); 1425 continue; 1426 } 1427 /* 1428 * If FORCECLOSE is set, forcibly close the vnode. 1429 * For block or character devices, revert to an 1430 * anonymous device. For all other files, just kill them. 1431 */ 1432 if (flags & FORCECLOSE) { 1433 simple_unlock(&mntvnode_slock); 1434 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1435 vgonel(vp, p); 1436 } else { 1437 vclean(vp, 0, p); 1438 vp->v_op = spec_vnodeop_p; 1439 insmntque(vp, (struct mount *)0); 1440 } 1441 simple_lock(&mntvnode_slock); 1442 continue; 1443 } 1444 #ifdef DEBUG 1445 if (busyprt) 1446 vprint("vflush: busy vnode", vp); 1447 #endif 1448 simple_unlock(&vp->v_interlock); 1449 busy++; 1450 } 1451 simple_unlock(&mntvnode_slock); 1452 if (busy) 1453 return (EBUSY); 1454 return (0); 1455 } 1456 1457 /* 1458 * Disassociate the underlying file system from a vnode. 1459 */ 1460 void 1461 vclean(vp, flags, p) 1462 struct vnode *vp; 1463 int flags; 1464 struct proc *p; 1465 { 1466 int active; 1467 1468 /* 1469 * Check to see if the vnode is in use. 1470 * If so we have to reference it before we clean it out 1471 * so that its count cannot fall to zero and generate a 1472 * race against ourselves to recycle it. 1473 */ 1474 if ((active = vp->v_usecount) != 0) { 1475 /* We have the vnode interlock. */ 1476 vp->v_usecount++; 1477 #ifdef DIAGNOSTIC 1478 if (vp->v_usecount == 0) { 1479 vprint("vclean", vp); 1480 panic("vclean: usecount overflow"); 1481 } 1482 #endif 1483 } 1484 1485 /* 1486 * Prevent the vnode from being recycled or 1487 * brought into use while we clean it out. 1488 */ 1489 if (vp->v_flag & VXLOCK) 1490 panic("vclean: deadlock, vp %p", vp); 1491 vp->v_flag |= VXLOCK; 1492 if (vp->v_flag & VTEXT) { 1493 uvmexp.vtextpages -= vp->v_uvm.u_obj.uo_npages; 1494 uvmexp.vnodepages += vp->v_uvm.u_obj.uo_npages; 1495 } 1496 vp->v_flag &= ~VTEXT; 1497 1498 /* 1499 * Even if the count is zero, the VOP_INACTIVE routine may still 1500 * have the object locked while it cleans it out. The VOP_LOCK 1501 * ensures that the VOP_INACTIVE routine is done with its work. 1502 * For active vnodes, it ensures that no other activity can 1503 * occur while the underlying object is being cleaned out. 1504 */ 1505 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK); 1506 1507 /* 1508 * Clean out any cached data associated with the vnode. 1509 */ 1510 if (flags & DOCLOSE) 1511 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1512 1513 /* 1514 * If purging an active vnode, it must be closed and 1515 * deactivated before being reclaimed. Note that the 1516 * VOP_INACTIVE will unlock the vnode. 1517 */ 1518 if (active) { 1519 if (flags & DOCLOSE) 1520 VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL); 1521 VOP_INACTIVE(vp, p); 1522 } else { 1523 /* 1524 * Any other processes trying to obtain this lock must first 1525 * wait for VXLOCK to clear, then call the new lock operation. 1526 */ 1527 VOP_UNLOCK(vp, 0); 1528 } 1529 /* 1530 * Reclaim the vnode. 1531 */ 1532 if (VOP_RECLAIM(vp, p)) 1533 panic("vclean: cannot reclaim, vp %p", vp); 1534 if (active) { 1535 /* 1536 * Inline copy of vrele() since VOP_INACTIVE 1537 * has already been called. 1538 */ 1539 simple_lock(&vp->v_interlock); 1540 if (--vp->v_usecount <= 0) { 1541 #ifdef DIAGNOSTIC 1542 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1543 vprint("vclean: bad ref count", vp); 1544 panic("vclean: ref cnt"); 1545 } 1546 #endif 1547 /* 1548 * Insert at tail of LRU list. 1549 */ 1550 1551 simple_unlock(&vp->v_interlock); 1552 simple_lock(&vnode_free_list_slock); 1553 #ifdef DIAGNOSTIC 1554 if (vp->v_vnlock) { 1555 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0) 1556 vprint("vclean: lock not drained", vp); 1557 } 1558 if (vp->v_holdcnt > 0) 1559 panic("vclean: not clean, vp %p", vp); 1560 #endif 1561 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1562 simple_unlock(&vnode_free_list_slock); 1563 } else 1564 simple_unlock(&vp->v_interlock); 1565 } 1566 1567 cache_purge(vp); 1568 1569 /* 1570 * Done with purge, notify sleepers of the grim news. 1571 */ 1572 vp->v_op = dead_vnodeop_p; 1573 vp->v_tag = VT_NON; 1574 simple_lock(&vp->v_interlock); 1575 vp->v_flag &= ~VXLOCK; 1576 if (vp->v_flag & VXWANT) { 1577 vp->v_flag &= ~VXWANT; 1578 simple_unlock(&vp->v_interlock); 1579 wakeup((caddr_t)vp); 1580 } else 1581 simple_unlock(&vp->v_interlock); 1582 } 1583 1584 /* 1585 * Recycle an unused vnode to the front of the free list. 1586 * Release the passed interlock if the vnode will be recycled. 1587 */ 1588 int 1589 vrecycle(vp, inter_lkp, p) 1590 struct vnode *vp; 1591 struct simplelock *inter_lkp; 1592 struct proc *p; 1593 { 1594 1595 simple_lock(&vp->v_interlock); 1596 if (vp->v_usecount == 0) { 1597 if (inter_lkp) 1598 simple_unlock(inter_lkp); 1599 vgonel(vp, p); 1600 return (1); 1601 } 1602 simple_unlock(&vp->v_interlock); 1603 return (0); 1604 } 1605 1606 /* 1607 * Eliminate all activity associated with a vnode 1608 * in preparation for reuse. 1609 */ 1610 void 1611 vgone(vp) 1612 struct vnode *vp; 1613 { 1614 struct proc *p = curproc; /* XXX */ 1615 1616 simple_lock(&vp->v_interlock); 1617 vgonel(vp, p); 1618 } 1619 1620 /* 1621 * vgone, with the vp interlock held. 1622 */ 1623 void 1624 vgonel(vp, p) 1625 struct vnode *vp; 1626 struct proc *p; 1627 { 1628 struct vnode *vq; 1629 struct vnode *vx; 1630 1631 /* 1632 * If a vgone (or vclean) is already in progress, 1633 * wait until it is done and return. 1634 */ 1635 if (vp->v_flag & VXLOCK) { 1636 vp->v_flag |= VXWANT; 1637 ltsleep((caddr_t)vp, PINOD | PNORELOCK, 1638 "vgone", 0, &vp->v_interlock); 1639 return; 1640 } 1641 /* 1642 * Clean out the filesystem specific data. 1643 */ 1644 vclean(vp, DOCLOSE, p); 1645 /* 1646 * Delete from old mount point vnode list, if on one. 1647 */ 1648 if (vp->v_mount != NULL) 1649 insmntque(vp, (struct mount *)0); 1650 /* 1651 * If special device, remove it from special device alias list. 1652 * if it is on one. 1653 */ 1654 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1655 simple_lock(&spechash_slock); 1656 if (vp->v_hashchain != NULL) { 1657 if (*vp->v_hashchain == vp) { 1658 *vp->v_hashchain = vp->v_specnext; 1659 } else { 1660 for (vq = *vp->v_hashchain; vq; 1661 vq = vq->v_specnext) { 1662 if (vq->v_specnext != vp) 1663 continue; 1664 vq->v_specnext = vp->v_specnext; 1665 break; 1666 } 1667 if (vq == NULL) 1668 panic("missing bdev"); 1669 } 1670 if (vp->v_flag & VALIASED) { 1671 vx = NULL; 1672 for (vq = *vp->v_hashchain; vq; 1673 vq = vq->v_specnext) { 1674 if (vq->v_rdev != vp->v_rdev || 1675 vq->v_type != vp->v_type) 1676 continue; 1677 if (vx) 1678 break; 1679 vx = vq; 1680 } 1681 if (vx == NULL) 1682 panic("missing alias"); 1683 if (vq == NULL) 1684 vx->v_flag &= ~VALIASED; 1685 vp->v_flag &= ~VALIASED; 1686 } 1687 } 1688 simple_unlock(&spechash_slock); 1689 FREE(vp->v_specinfo, M_VNODE); 1690 vp->v_specinfo = NULL; 1691 } 1692 /* 1693 * If it is on the freelist and not already at the head, 1694 * move it to the head of the list. The test of the back 1695 * pointer and the reference count of zero is because 1696 * it will be removed from the free list by getnewvnode, 1697 * but will not have its reference count incremented until 1698 * after calling vgone. If the reference count were 1699 * incremented first, vgone would (incorrectly) try to 1700 * close the previous instance of the underlying object. 1701 * So, the back pointer is explicitly set to `0xdeadb' in 1702 * getnewvnode after removing it from the freelist to ensure 1703 * that we do not try to move it here. 1704 */ 1705 if (vp->v_usecount == 0) { 1706 simple_lock(&vnode_free_list_slock); 1707 if (vp->v_holdcnt > 0) 1708 panic("vgonel: not clean, vp %p", vp); 1709 if (vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb && 1710 TAILQ_FIRST(&vnode_free_list) != vp) { 1711 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1712 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1713 } 1714 simple_unlock(&vnode_free_list_slock); 1715 } 1716 vp->v_type = VBAD; 1717 } 1718 1719 /* 1720 * Lookup a vnode by device number. 1721 */ 1722 int 1723 vfinddev(dev, type, vpp) 1724 dev_t dev; 1725 enum vtype type; 1726 struct vnode **vpp; 1727 { 1728 struct vnode *vp; 1729 int rc = 0; 1730 1731 simple_lock(&spechash_slock); 1732 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1733 if (dev != vp->v_rdev || type != vp->v_type) 1734 continue; 1735 *vpp = vp; 1736 rc = 1; 1737 break; 1738 } 1739 simple_unlock(&spechash_slock); 1740 return (rc); 1741 } 1742 1743 /* 1744 * Revoke all the vnodes corresponding to the specified minor number 1745 * range (endpoints inclusive) of the specified major. 1746 */ 1747 void 1748 vdevgone(maj, minl, minh, type) 1749 int maj, minl, minh; 1750 enum vtype type; 1751 { 1752 struct vnode *vp; 1753 int mn; 1754 1755 for (mn = minl; mn <= minh; mn++) 1756 if (vfinddev(makedev(maj, mn), type, &vp)) 1757 VOP_REVOKE(vp, REVOKEALL); 1758 } 1759 1760 /* 1761 * Calculate the total number of references to a special device. 1762 */ 1763 int 1764 vcount(vp) 1765 struct vnode *vp; 1766 { 1767 struct vnode *vq, *vnext; 1768 int count; 1769 1770 loop: 1771 if ((vp->v_flag & VALIASED) == 0) 1772 return (vp->v_usecount); 1773 simple_lock(&spechash_slock); 1774 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1775 vnext = vq->v_specnext; 1776 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1777 continue; 1778 /* 1779 * Alias, but not in use, so flush it out. 1780 */ 1781 if (vq->v_usecount == 0 && vq != vp) { 1782 simple_unlock(&spechash_slock); 1783 vgone(vq); 1784 goto loop; 1785 } 1786 count += vq->v_usecount; 1787 } 1788 simple_unlock(&spechash_slock); 1789 return (count); 1790 } 1791 1792 /* 1793 * Print out a description of a vnode. 1794 */ 1795 static const char * const typename[] = 1796 { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; 1797 1798 void 1799 vprint(label, vp) 1800 char *label; 1801 struct vnode *vp; 1802 { 1803 char buf[64]; 1804 1805 if (label != NULL) 1806 printf("%s: ", label); 1807 printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,", 1808 vp->v_tag, typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1809 vp->v_holdcnt); 1810 buf[0] = '\0'; 1811 if (vp->v_flag & VROOT) 1812 strcat(buf, "|VROOT"); 1813 if (vp->v_flag & VTEXT) 1814 strcat(buf, "|VTEXT"); 1815 if (vp->v_flag & VSYSTEM) 1816 strcat(buf, "|VSYSTEM"); 1817 if (vp->v_flag & VXLOCK) 1818 strcat(buf, "|VXLOCK"); 1819 if (vp->v_flag & VXWANT) 1820 strcat(buf, "|VXWANT"); 1821 if (vp->v_flag & VBWAIT) 1822 strcat(buf, "|VBWAIT"); 1823 if (vp->v_flag & VALIASED) 1824 strcat(buf, "|VALIASED"); 1825 if (buf[0] != '\0') 1826 printf(" flags (%s)", &buf[1]); 1827 if (vp->v_data == NULL) { 1828 printf("\n"); 1829 } else { 1830 printf("\n\t"); 1831 VOP_PRINT(vp); 1832 } 1833 } 1834 1835 #ifdef DEBUG 1836 /* 1837 * List all of the locked vnodes in the system. 1838 * Called when debugging the kernel. 1839 */ 1840 void 1841 printlockedvnodes() 1842 { 1843 struct mount *mp, *nmp; 1844 struct vnode *vp; 1845 1846 printf("Locked vnodes\n"); 1847 simple_lock(&mountlist_slock); 1848 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1849 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 1850 nmp = mp->mnt_list.cqe_next; 1851 continue; 1852 } 1853 for (vp = mp->mnt_vnodelist.lh_first; 1854 vp != NULL; 1855 vp = vp->v_mntvnodes.le_next) { 1856 if (VOP_ISLOCKED(vp)) 1857 vprint((char *)0, vp); 1858 } 1859 simple_lock(&mountlist_slock); 1860 nmp = mp->mnt_list.cqe_next; 1861 vfs_unbusy(mp); 1862 } 1863 simple_unlock(&mountlist_slock); 1864 } 1865 #endif 1866 1867 extern const char *mountcompatnames[]; 1868 extern const int nmountcompatnames; 1869 1870 /* 1871 * Top level filesystem related information gathering. 1872 */ 1873 int 1874 vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) 1875 int *name; 1876 u_int namelen; 1877 void *oldp; 1878 size_t *oldlenp; 1879 void *newp; 1880 size_t newlen; 1881 struct proc *p; 1882 { 1883 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 1884 struct vfsconf vfc; 1885 #endif 1886 struct vfsops *vfsp; 1887 1888 /* all sysctl names at this level are at least name and field */ 1889 if (namelen < 2) 1890 return (ENOTDIR); /* overloaded */ 1891 1892 /* Not generic: goes to file system. */ 1893 if (name[0] != VFS_GENERIC) { 1894 if (name[0] >= nmountcompatnames || name[0] < 0 || 1895 mountcompatnames[name[0]] == NULL) 1896 return (EOPNOTSUPP); 1897 vfsp = vfs_getopsbyname(mountcompatnames[name[0]]); 1898 if (vfsp == NULL || vfsp->vfs_sysctl == NULL) 1899 return (EOPNOTSUPP); 1900 return ((*vfsp->vfs_sysctl)(&name[1], namelen - 1, 1901 oldp, oldlenp, newp, newlen, p)); 1902 } 1903 1904 /* The rest are generic vfs sysctls. */ 1905 switch (name[1]) { 1906 case VFS_USERMOUNT: 1907 return sysctl_int(oldp, oldlenp, newp, newlen, &dovfsusermount); 1908 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 1909 case VFS_MAXTYPENUM: 1910 /* 1911 * Provided for 4.4BSD-Lite2 compatibility. 1912 */ 1913 return (sysctl_rdint(oldp, oldlenp, newp, nmountcompatnames)); 1914 case VFS_CONF: 1915 /* 1916 * Special: a node, next is a file system name. 1917 * Provided for 4.4BSD-Lite2 compatibility. 1918 */ 1919 if (namelen < 3) 1920 return (ENOTDIR); /* overloaded */ 1921 if (name[2] >= nmountcompatnames || name[2] < 0 || 1922 mountcompatnames[name[2]] == NULL) 1923 return (EOPNOTSUPP); 1924 vfsp = vfs_getopsbyname(mountcompatnames[name[2]]); 1925 if (vfsp == NULL) 1926 return (EOPNOTSUPP); 1927 vfc.vfc_vfsops = vfsp; 1928 strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN); 1929 vfc.vfc_typenum = name[2]; 1930 vfc.vfc_refcount = vfsp->vfs_refcount; 1931 vfc.vfc_flags = 0; 1932 vfc.vfc_mountroot = vfsp->vfs_mountroot; 1933 vfc.vfc_next = NULL; 1934 return (sysctl_rdstruct(oldp, oldlenp, newp, &vfc, 1935 sizeof(struct vfsconf))); 1936 #endif 1937 default: 1938 break; 1939 } 1940 return (EOPNOTSUPP); 1941 } 1942 1943 int kinfo_vdebug = 1; 1944 int kinfo_vgetfailed; 1945 #define KINFO_VNODESLOP 10 1946 /* 1947 * Dump vnode list (via sysctl). 1948 * Copyout address of vnode followed by vnode. 1949 */ 1950 /* ARGSUSED */ 1951 int 1952 sysctl_vnode(where, sizep, p) 1953 char *where; 1954 size_t *sizep; 1955 struct proc *p; 1956 { 1957 struct mount *mp, *nmp; 1958 struct vnode *nvp, *vp; 1959 char *bp = where, *savebp; 1960 char *ewhere; 1961 int error; 1962 1963 #define VPTRSZ sizeof(struct vnode *) 1964 #define VNODESZ sizeof(struct vnode) 1965 if (where == NULL) { 1966 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 1967 return (0); 1968 } 1969 ewhere = where + *sizep; 1970 1971 simple_lock(&mountlist_slock); 1972 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1973 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 1974 nmp = mp->mnt_list.cqe_next; 1975 continue; 1976 } 1977 savebp = bp; 1978 again: 1979 simple_lock(&mntvnode_slock); 1980 for (vp = mp->mnt_vnodelist.lh_first; 1981 vp != NULL; 1982 vp = nvp) { 1983 /* 1984 * Check that the vp is still associated with 1985 * this filesystem. RACE: could have been 1986 * recycled onto the same filesystem. 1987 */ 1988 if (vp->v_mount != mp) { 1989 simple_unlock(&mntvnode_slock); 1990 if (kinfo_vdebug) 1991 printf("kinfo: vp changed\n"); 1992 bp = savebp; 1993 goto again; 1994 } 1995 nvp = vp->v_mntvnodes.le_next; 1996 if (bp + VPTRSZ + VNODESZ > ewhere) { 1997 simple_unlock(&mntvnode_slock); 1998 *sizep = bp - where; 1999 return (ENOMEM); 2000 } 2001 simple_unlock(&mntvnode_slock); 2002 if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || 2003 (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ))) 2004 return (error); 2005 bp += VPTRSZ + VNODESZ; 2006 simple_lock(&mntvnode_slock); 2007 } 2008 simple_unlock(&mntvnode_slock); 2009 simple_lock(&mountlist_slock); 2010 nmp = mp->mnt_list.cqe_next; 2011 vfs_unbusy(mp); 2012 } 2013 simple_unlock(&mountlist_slock); 2014 2015 *sizep = bp - where; 2016 return (0); 2017 } 2018 2019 /* 2020 * Check to see if a filesystem is mounted on a block device. 2021 */ 2022 int 2023 vfs_mountedon(vp) 2024 struct vnode *vp; 2025 { 2026 struct vnode *vq; 2027 int error = 0; 2028 2029 if (vp->v_specmountpoint != NULL) 2030 return (EBUSY); 2031 if (vp->v_flag & VALIASED) { 2032 simple_lock(&spechash_slock); 2033 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 2034 if (vq->v_rdev != vp->v_rdev || 2035 vq->v_type != vp->v_type) 2036 continue; 2037 if (vq->v_specmountpoint != NULL) { 2038 error = EBUSY; 2039 break; 2040 } 2041 } 2042 simple_unlock(&spechash_slock); 2043 } 2044 return (error); 2045 } 2046 2047 /* 2048 * Build hash lists of net addresses and hang them off the mount point. 2049 * Called by ufs_mount() to set up the lists of export addresses. 2050 */ 2051 static int 2052 vfs_hang_addrlist(mp, nep, argp) 2053 struct mount *mp; 2054 struct netexport *nep; 2055 struct export_args *argp; 2056 { 2057 struct netcred *np, *enp; 2058 struct radix_node_head *rnh; 2059 int i; 2060 struct radix_node *rn; 2061 struct sockaddr *saddr, *smask = 0; 2062 struct domain *dom; 2063 int error; 2064 2065 if (argp->ex_addrlen == 0) { 2066 if (mp->mnt_flag & MNT_DEFEXPORTED) 2067 return (EPERM); 2068 np = &nep->ne_defexported; 2069 np->netc_exflags = argp->ex_flags; 2070 np->netc_anon = argp->ex_anon; 2071 np->netc_anon.cr_ref = 1; 2072 mp->mnt_flag |= MNT_DEFEXPORTED; 2073 return (0); 2074 } 2075 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2076 np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); 2077 memset((caddr_t)np, 0, i); 2078 saddr = (struct sockaddr *)(np + 1); 2079 error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen); 2080 if (error) 2081 goto out; 2082 if (saddr->sa_len > argp->ex_addrlen) 2083 saddr->sa_len = argp->ex_addrlen; 2084 if (argp->ex_masklen) { 2085 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2086 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2087 if (error) 2088 goto out; 2089 if (smask->sa_len > argp->ex_masklen) 2090 smask->sa_len = argp->ex_masklen; 2091 } 2092 i = saddr->sa_family; 2093 if ((rnh = nep->ne_rtable[i]) == 0) { 2094 /* 2095 * Seems silly to initialize every AF when most are not 2096 * used, do so on demand here 2097 */ 2098 for (dom = domains; dom; dom = dom->dom_next) 2099 if (dom->dom_family == i && dom->dom_rtattach) { 2100 dom->dom_rtattach((void **)&nep->ne_rtable[i], 2101 dom->dom_rtoffset); 2102 break; 2103 } 2104 if ((rnh = nep->ne_rtable[i]) == 0) { 2105 error = ENOBUFS; 2106 goto out; 2107 } 2108 } 2109 rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh, 2110 np->netc_rnodes); 2111 if (rn == 0 || np != (struct netcred *)rn) { /* already exists */ 2112 if (rn == 0) { 2113 enp = (struct netcred *)(*rnh->rnh_lookup)(saddr, 2114 smask, rnh); 2115 if (enp == 0) { 2116 error = EPERM; 2117 goto out; 2118 } 2119 } else 2120 enp = (struct netcred *)rn; 2121 2122 if (enp->netc_exflags != argp->ex_flags || 2123 enp->netc_anon.cr_uid != argp->ex_anon.cr_uid || 2124 enp->netc_anon.cr_gid != argp->ex_anon.cr_gid || 2125 enp->netc_anon.cr_ngroups != argp->ex_anon.cr_ngroups || 2126 memcmp(&enp->netc_anon.cr_groups, &argp->ex_anon.cr_groups, 2127 enp->netc_anon.cr_ngroups)) 2128 error = EPERM; 2129 else 2130 error = 0; 2131 goto out; 2132 } 2133 np->netc_exflags = argp->ex_flags; 2134 np->netc_anon = argp->ex_anon; 2135 np->netc_anon.cr_ref = 1; 2136 return (0); 2137 out: 2138 free(np, M_NETADDR); 2139 return (error); 2140 } 2141 2142 /* ARGSUSED */ 2143 static int 2144 vfs_free_netcred(rn, w) 2145 struct radix_node *rn; 2146 void *w; 2147 { 2148 struct radix_node_head *rnh = (struct radix_node_head *)w; 2149 2150 (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh); 2151 free((caddr_t)rn, M_NETADDR); 2152 return (0); 2153 } 2154 2155 /* 2156 * Free the net address hash lists that are hanging off the mount points. 2157 */ 2158 static void 2159 vfs_free_addrlist(nep) 2160 struct netexport *nep; 2161 { 2162 int i; 2163 struct radix_node_head *rnh; 2164 2165 for (i = 0; i <= AF_MAX; i++) 2166 if ((rnh = nep->ne_rtable[i]) != NULL) { 2167 (*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh); 2168 free((caddr_t)rnh, M_RTABLE); 2169 nep->ne_rtable[i] = 0; 2170 } 2171 } 2172 2173 int 2174 vfs_export(mp, nep, argp) 2175 struct mount *mp; 2176 struct netexport *nep; 2177 struct export_args *argp; 2178 { 2179 int error; 2180 2181 if (argp->ex_flags & MNT_DELEXPORT) { 2182 if (mp->mnt_flag & MNT_EXPUBLIC) { 2183 vfs_setpublicfs(NULL, NULL, NULL); 2184 mp->mnt_flag &= ~MNT_EXPUBLIC; 2185 } 2186 vfs_free_addrlist(nep); 2187 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2188 } 2189 if (argp->ex_flags & MNT_EXPORTED) { 2190 if (argp->ex_flags & MNT_EXPUBLIC) { 2191 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2192 return (error); 2193 mp->mnt_flag |= MNT_EXPUBLIC; 2194 } 2195 if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0) 2196 return (error); 2197 mp->mnt_flag |= MNT_EXPORTED; 2198 } 2199 return (0); 2200 } 2201 2202 /* 2203 * Set the publicly exported filesystem (WebNFS). Currently, only 2204 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2205 */ 2206 int 2207 vfs_setpublicfs(mp, nep, argp) 2208 struct mount *mp; 2209 struct netexport *nep; 2210 struct export_args *argp; 2211 { 2212 int error; 2213 struct vnode *rvp; 2214 char *cp; 2215 2216 /* 2217 * mp == NULL -> invalidate the current info, the FS is 2218 * no longer exported. May be called from either vfs_export 2219 * or unmount, so check if it hasn't already been done. 2220 */ 2221 if (mp == NULL) { 2222 if (nfs_pub.np_valid) { 2223 nfs_pub.np_valid = 0; 2224 if (nfs_pub.np_index != NULL) { 2225 FREE(nfs_pub.np_index, M_TEMP); 2226 nfs_pub.np_index = NULL; 2227 } 2228 } 2229 return (0); 2230 } 2231 2232 /* 2233 * Only one allowed at a time. 2234 */ 2235 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2236 return (EBUSY); 2237 2238 /* 2239 * Get real filehandle for root of exported FS. 2240 */ 2241 memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle)); 2242 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2243 2244 if ((error = VFS_ROOT(mp, &rvp))) 2245 return (error); 2246 2247 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2248 return (error); 2249 2250 vput(rvp); 2251 2252 /* 2253 * If an indexfile was specified, pull it in. 2254 */ 2255 if (argp->ex_indexfile != NULL) { 2256 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2257 M_WAITOK); 2258 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2259 MAXNAMLEN, (size_t *)0); 2260 if (!error) { 2261 /* 2262 * Check for illegal filenames. 2263 */ 2264 for (cp = nfs_pub.np_index; *cp; cp++) { 2265 if (*cp == '/') { 2266 error = EINVAL; 2267 break; 2268 } 2269 } 2270 } 2271 if (error) { 2272 FREE(nfs_pub.np_index, M_TEMP); 2273 return (error); 2274 } 2275 } 2276 2277 nfs_pub.np_mount = mp; 2278 nfs_pub.np_valid = 1; 2279 return (0); 2280 } 2281 2282 struct netcred * 2283 vfs_export_lookup(mp, nep, nam) 2284 struct mount *mp; 2285 struct netexport *nep; 2286 struct mbuf *nam; 2287 { 2288 struct netcred *np; 2289 struct radix_node_head *rnh; 2290 struct sockaddr *saddr; 2291 2292 np = NULL; 2293 if (mp->mnt_flag & MNT_EXPORTED) { 2294 /* 2295 * Lookup in the export list first. 2296 */ 2297 if (nam != NULL) { 2298 saddr = mtod(nam, struct sockaddr *); 2299 rnh = nep->ne_rtable[saddr->sa_family]; 2300 if (rnh != NULL) { 2301 np = (struct netcred *) 2302 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2303 rnh); 2304 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2305 np = NULL; 2306 } 2307 } 2308 /* 2309 * If no address match, use the default if it exists. 2310 */ 2311 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2312 np = &nep->ne_defexported; 2313 } 2314 return (np); 2315 } 2316 2317 /* 2318 * Do the usual access checking. 2319 * file_mode, uid and gid are from the vnode in question, 2320 * while acc_mode and cred are from the VOP_ACCESS parameter list 2321 */ 2322 int 2323 vaccess(type, file_mode, uid, gid, acc_mode, cred) 2324 enum vtype type; 2325 mode_t file_mode; 2326 uid_t uid; 2327 gid_t gid; 2328 mode_t acc_mode; 2329 struct ucred *cred; 2330 { 2331 mode_t mask; 2332 2333 /* 2334 * Super-user always gets read/write access, but execute access depends 2335 * on at least one execute bit being set. 2336 */ 2337 if (cred->cr_uid == 0) { 2338 if ((acc_mode & VEXEC) && type != VDIR && 2339 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 2340 return (EACCES); 2341 return (0); 2342 } 2343 2344 mask = 0; 2345 2346 /* Otherwise, check the owner. */ 2347 if (cred->cr_uid == uid) { 2348 if (acc_mode & VEXEC) 2349 mask |= S_IXUSR; 2350 if (acc_mode & VREAD) 2351 mask |= S_IRUSR; 2352 if (acc_mode & VWRITE) 2353 mask |= S_IWUSR; 2354 return ((file_mode & mask) == mask ? 0 : EACCES); 2355 } 2356 2357 /* Otherwise, check the groups. */ 2358 if (cred->cr_gid == gid || groupmember(gid, cred)) { 2359 if (acc_mode & VEXEC) 2360 mask |= S_IXGRP; 2361 if (acc_mode & VREAD) 2362 mask |= S_IRGRP; 2363 if (acc_mode & VWRITE) 2364 mask |= S_IWGRP; 2365 return ((file_mode & mask) == mask ? 0 : EACCES); 2366 } 2367 2368 /* Otherwise, check everyone else. */ 2369 if (acc_mode & VEXEC) 2370 mask |= S_IXOTH; 2371 if (acc_mode & VREAD) 2372 mask |= S_IROTH; 2373 if (acc_mode & VWRITE) 2374 mask |= S_IWOTH; 2375 return ((file_mode & mask) == mask ? 0 : EACCES); 2376 } 2377 2378 /* 2379 * Unmount all file systems. 2380 * We traverse the list in reverse order under the assumption that doing so 2381 * will avoid needing to worry about dependencies. 2382 */ 2383 void 2384 vfs_unmountall(p) 2385 struct proc *p; 2386 { 2387 struct mount *mp, *nmp; 2388 int allerror, error; 2389 2390 for (allerror = 0, 2391 mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 2392 nmp = mp->mnt_list.cqe_prev; 2393 #ifdef DEBUG 2394 printf("unmounting %s (%s)...\n", 2395 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname); 2396 #endif 2397 /* 2398 * XXX Freeze syncer. Must do this before locking the 2399 * mount point. See dounmount() for details. 2400 */ 2401 lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL); 2402 if (vfs_busy(mp, 0, 0)) { 2403 lockmgr(&syncer_lock, LK_RELEASE, NULL); 2404 continue; 2405 } 2406 if ((error = dounmount(mp, MNT_FORCE, p)) != 0) { 2407 printf("unmount of %s failed with error %d\n", 2408 mp->mnt_stat.f_mntonname, error); 2409 allerror = 1; 2410 } 2411 } 2412 if (allerror) 2413 printf("WARNING: some file systems would not unmount\n"); 2414 } 2415 2416 /* 2417 * Sync and unmount file systems before shutting down. 2418 */ 2419 void 2420 vfs_shutdown() 2421 { 2422 struct buf *bp; 2423 int iter, nbusy, nbusy_prev = 0, dcount, s; 2424 struct proc *p = curproc; 2425 2426 /* XXX we're certainly not running in proc0's context! */ 2427 if (p == NULL) 2428 p = &proc0; 2429 2430 printf("syncing disks... "); 2431 2432 /* remove user process from run queue */ 2433 suspendsched(); 2434 (void) spl0(); 2435 2436 /* avoid coming back this way again if we panic. */ 2437 doing_shutdown = 1; 2438 2439 sys_sync(p, NULL, NULL); 2440 2441 /* Wait for sync to finish. */ 2442 dcount = 10000; 2443 for (iter = 0; iter < 20;) { 2444 nbusy = 0; 2445 for (bp = &buf[nbuf]; --bp >= buf; ) { 2446 if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY) 2447 nbusy++; 2448 /* 2449 * With soft updates, some buffers that are 2450 * written will be remarked as dirty until other 2451 * buffers are written. 2452 */ 2453 if (bp->b_vp && bp->b_vp->v_mount 2454 && (bp->b_vp->v_mount->mnt_flag & MNT_SOFTDEP) 2455 && (bp->b_flags & B_DELWRI)) { 2456 s = splbio(); 2457 bremfree(bp); 2458 bp->b_flags |= B_BUSY; 2459 splx(s); 2460 nbusy++; 2461 bawrite(bp); 2462 if (dcount-- <= 0) { 2463 printf("softdep "); 2464 goto fail; 2465 } 2466 } 2467 } 2468 if (nbusy == 0) 2469 break; 2470 if (nbusy_prev == 0) 2471 nbusy_prev = nbusy; 2472 printf("%d ", nbusy); 2473 tsleep(&nbusy, PRIBIO, "bflush", 2474 (iter == 0) ? 1 : hz / 25 * iter); 2475 if (nbusy >= nbusy_prev) /* we didn't flush anything */ 2476 iter++; 2477 else 2478 nbusy_prev = nbusy; 2479 } 2480 if (nbusy) { 2481 fail: 2482 #if defined(DEBUG) || defined(DEBUG_HALT_BUSY) 2483 printf("giving up\nPrinting vnodes for busy buffers\n"); 2484 for (bp = &buf[nbuf]; --bp >= buf; ) 2485 if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY) 2486 vprint(NULL, bp->b_vp); 2487 2488 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 2489 Debugger(); 2490 #endif 2491 2492 #else /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */ 2493 printf("giving up\n"); 2494 #endif /* defined(DEBUG) || defined(DEBUG_HALT_BUSY) */ 2495 return; 2496 } else 2497 printf("done\n"); 2498 2499 /* 2500 * If we've panic'd, don't make the situation potentially 2501 * worse by unmounting the file systems. 2502 */ 2503 if (panicstr != NULL) 2504 return; 2505 2506 /* Release inodes held by texts before update. */ 2507 #ifdef notdef 2508 vnshutdown(); 2509 #endif 2510 /* Unmount file systems. */ 2511 vfs_unmountall(p); 2512 } 2513 2514 /* 2515 * Mount the root file system. If the operator didn't specify a 2516 * file system to use, try all possible file systems until one 2517 * succeeds. 2518 */ 2519 int 2520 vfs_mountroot() 2521 { 2522 extern int (*mountroot) __P((void)); 2523 struct vfsops *v; 2524 2525 if (root_device == NULL) 2526 panic("vfs_mountroot: root device unknown"); 2527 2528 switch (root_device->dv_class) { 2529 case DV_IFNET: 2530 if (rootdev != NODEV) 2531 panic("vfs_mountroot: rootdev set for DV_IFNET"); 2532 break; 2533 2534 case DV_DISK: 2535 if (rootdev == NODEV) 2536 panic("vfs_mountroot: rootdev not set for DV_DISK"); 2537 break; 2538 2539 default: 2540 printf("%s: inappropriate for root file system\n", 2541 root_device->dv_xname); 2542 return (ENODEV); 2543 } 2544 2545 /* 2546 * If user specified a file system, use it. 2547 */ 2548 if (mountroot != NULL) 2549 return ((*mountroot)()); 2550 2551 /* 2552 * Try each file system currently configured into the kernel. 2553 */ 2554 for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) { 2555 if (v->vfs_mountroot == NULL) 2556 continue; 2557 #ifdef DEBUG 2558 printf("mountroot: trying %s...\n", v->vfs_name); 2559 #endif 2560 if ((*v->vfs_mountroot)() == 0) { 2561 printf("root file system type: %s\n", v->vfs_name); 2562 break; 2563 } 2564 } 2565 2566 if (v == NULL) { 2567 printf("no file system for %s", root_device->dv_xname); 2568 if (root_device->dv_class == DV_DISK) 2569 printf(" (dev 0x%x)", rootdev); 2570 printf("\n"); 2571 return (EFTYPE); 2572 } 2573 return (0); 2574 } 2575 2576 /* 2577 * Given a file system name, look up the vfsops for that 2578 * file system, or return NULL if file system isn't present 2579 * in the kernel. 2580 */ 2581 struct vfsops * 2582 vfs_getopsbyname(name) 2583 const char *name; 2584 { 2585 struct vfsops *v; 2586 2587 for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) { 2588 if (strcmp(v->vfs_name, name) == 0) 2589 break; 2590 } 2591 2592 return (v); 2593 } 2594 2595 /* 2596 * Establish a file system and initialize it. 2597 */ 2598 int 2599 vfs_attach(vfs) 2600 struct vfsops *vfs; 2601 { 2602 struct vfsops *v; 2603 int error = 0; 2604 2605 2606 /* 2607 * Make sure this file system doesn't already exist. 2608 */ 2609 for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) { 2610 if (strcmp(vfs->vfs_name, v->vfs_name) == 0) { 2611 error = EEXIST; 2612 goto out; 2613 } 2614 } 2615 2616 /* 2617 * Initialize the vnode operations for this file system. 2618 */ 2619 vfs_opv_init(vfs->vfs_opv_descs); 2620 2621 /* 2622 * Now initialize the file system itself. 2623 */ 2624 (*vfs->vfs_init)(); 2625 2626 /* 2627 * ...and link it into the kernel's list. 2628 */ 2629 LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list); 2630 2631 /* 2632 * Sanity: make sure the reference count is 0. 2633 */ 2634 vfs->vfs_refcount = 0; 2635 2636 out: 2637 return (error); 2638 } 2639 2640 /* 2641 * Remove a file system from the kernel. 2642 */ 2643 int 2644 vfs_detach(vfs) 2645 struct vfsops *vfs; 2646 { 2647 struct vfsops *v; 2648 2649 /* 2650 * Make sure no one is using the filesystem. 2651 */ 2652 if (vfs->vfs_refcount != 0) 2653 return (EBUSY); 2654 2655 /* 2656 * ...and remove it from the kernel's list. 2657 */ 2658 for (v = LIST_FIRST(&vfs_list); v != NULL; v = LIST_NEXT(v, vfs_list)) { 2659 if (v == vfs) { 2660 LIST_REMOVE(v, vfs_list); 2661 break; 2662 } 2663 } 2664 2665 if (v == NULL) 2666 return (ESRCH); 2667 2668 /* 2669 * Now run the file system-specific cleanups. 2670 */ 2671 (*vfs->vfs_done)(); 2672 2673 /* 2674 * Free the vnode operations vector. 2675 */ 2676 vfs_opv_free(vfs->vfs_opv_descs); 2677 return (0); 2678 } 2679 2680 #ifdef DDB 2681 const char buf_flagbits[] = 2682 "\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI" 2683 "\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE" 2684 "\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED" 2685 "\32XXX\33VFLUSH"; 2686 2687 void 2688 vfs_buf_print(bp, full, pr) 2689 struct buf *bp; 2690 int full; 2691 void (*pr) __P((const char *, ...)); 2692 { 2693 char buf[1024]; 2694 2695 (*pr)(" vp %p lblkno 0x%x blkno 0x%x dev 0x%x\n", 2696 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev); 2697 2698 bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf)); 2699 (*pr)(" error %d flags 0x%s\n", bp->b_error, buf); 2700 2701 (*pr)(" bufsize 0x%x bcount 0x%x resid 0x%x\n", 2702 bp->b_bufsize, bp->b_bcount, bp->b_resid); 2703 (*pr)(" data %p saveaddr %p dep %p\n", 2704 bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep)); 2705 (*pr)(" iodone %p\n", bp->b_iodone); 2706 } 2707 2708 2709 const char vnode_flagbits[] = 2710 "\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\11XLOCK\12XWANT\13BWAIT\14ALIASED" 2711 "\15DIROP\16LAYER\17ONWORKLIST\20DIRTY"; 2712 2713 const char *vnode_types[] = { 2714 "VNON", 2715 "VREG", 2716 "VDIR", 2717 "VBLK", 2718 "VCHR", 2719 "VLNK", 2720 "VSOCK", 2721 "VFIFO", 2722 "VBAD", 2723 }; 2724 2725 const char *vnode_tags[] = { 2726 "VT_NON", 2727 "VT_UFS", 2728 "VT_NFS", 2729 "VT_MFS", 2730 "VT_MSDOSFS", 2731 "VT_LFS", 2732 "VT_LOFS", 2733 "VT_FDESC", 2734 "VT_PORTAL", 2735 "VT_NULL", 2736 "VT_UMAP", 2737 "VT_KERNFS", 2738 "VT_PROCFS", 2739 "VT_AFS", 2740 "VT_ISOFS", 2741 "VT_UNION", 2742 "VT_ADOSFS", 2743 "VT_EXT2FS", 2744 "VT_CODA", 2745 "VT_FILECORE", 2746 "VT_NTFS", 2747 "VT_VFS", 2748 "VT_OVERLAY" 2749 }; 2750 2751 void 2752 vfs_vnode_print(vp, full, pr) 2753 struct vnode *vp; 2754 int full; 2755 void (*pr) __P((const char *, ...)); 2756 { 2757 char buf[256]; 2758 2759 const char *vtype, *vtag; 2760 2761 uvm_object_printit(&vp->v_uvm.u_obj, full, pr); 2762 bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf)); 2763 (*pr)("\nVNODE flags %s\n", buf); 2764 (*pr)("mp %p nio %d size 0x%x rwlock 0x%x glock 0x%x\n", 2765 vp->v_mount, vp->v_uvm.u_nio, (int)vp->v_uvm.u_size, 2766 vp->v_vnlock ? lockstatus(vp->v_vnlock) : 0x999, 2767 lockstatus(&vp->v_glock)); 2768 2769 (*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n", 2770 vp->v_data, vp->v_usecount, vp->v_writecount, 2771 vp->v_holdcnt, vp->v_numoutput); 2772 2773 vtype = (vp->v_type >= 0 && 2774 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ? 2775 vnode_types[vp->v_type] : "UNKNOWN"; 2776 vtag = (vp->v_tag >= 0 && 2777 vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ? 2778 vnode_tags[vp->v_tag] : "UNKNOWN"; 2779 2780 (*pr)("type %s(%d) tag %s(%d) id 0x%x mount %p typedata %p\n", 2781 vtype, vp->v_type, vtag, vp->v_tag, 2782 vp->v_id, vp->v_mount, vp->v_mountedhere); 2783 (*pr)("lastr 0x%x lastw 0x%x lasta 0x%x\n", 2784 vp->v_lastr, vp->v_lastw, vp->v_lasta); 2785 (*pr)("cstart 0x%x clen 0x%x ralen 0x%x maxra 0x%x\n", 2786 vp->v_cstart, vp->v_clen, vp->v_ralen, vp->v_maxra); 2787 2788 if (full) { 2789 struct buf *bp; 2790 2791 (*pr)("clean bufs:\n"); 2792 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 2793 (*pr)(" bp %p\n", bp); 2794 vfs_buf_print(bp, full, pr); 2795 } 2796 2797 (*pr)("dirty bufs:\n"); 2798 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 2799 (*pr)(" bp %p\n", bp); 2800 vfs_buf_print(bp, full, pr); 2801 } 2802 } 2803 } 2804 #endif 2805