vfs_subr.c revision 1.111
1/* $OpenBSD: vfs_subr.c,v 1.111 2005/05/01 12:28:18 pedro Exp $ */ 2/* $NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $ */ 3 4/* 5 * Copyright (c) 1989, 1993 6 * The Regents of the University of California. All rights reserved. 7 * (c) UNIX System Laboratories, Inc. 8 * All or some portions of this file are derived from material licensed 9 * to the University of California by American Telephone and Telegraph 10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 11 * the permission of UNIX System Laboratories, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 38 */ 39 40/* 41 * External virtual filesystem routines 42 */ 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/proc.h> 47#include <sys/mount.h> 48#include <sys/time.h> 49#include <sys/fcntl.h> 50#include <sys/kernel.h> 51#include <sys/vnode.h> 52#include <sys/stat.h> 53#include <sys/namei.h> 54#include <sys/ucred.h> 55#include <sys/buf.h> 56#include <sys/errno.h> 57#include <sys/malloc.h> 58#include <sys/domain.h> 59#include <sys/mbuf.h> 60#include <sys/syscallargs.h> 61#include <sys/pool.h> 62 63#include <uvm/uvm_extern.h> 64#include <sys/sysctl.h> 65 66#include <miscfs/specfs/specdev.h> 67 68enum vtype iftovt_tab[16] = { 69 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 70 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 71}; 72int vttoif_tab[9] = { 73 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 74 S_IFSOCK, S_IFIFO, S_IFMT, 75}; 76 77int doforce = 1; /* 1 => permit forcible unmounting */ 78int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 79int suid_clear = 1; /* 1 => clear SUID / SGID on owner change */ 80 81/* 82 * Insq/Remq for the vnode usage lists. 83 */ 84#define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 85#define bufremvn(bp) { \ 86 LIST_REMOVE(bp, b_vnbufs); \ 87 LIST_NEXT(bp, b_vnbufs) = NOLIST; \ 88} 89 90struct freelst vnode_hold_list; /* list of vnodes referencing buffers */ 91struct freelst vnode_free_list; /* vnode free list */ 92 93struct mntlist mountlist; /* mounted filesystem list */ 94struct simplelock mountlist_slock; 95static struct simplelock mntid_slock; 96struct simplelock mntvnode_slock; 97struct simplelock vnode_free_list_slock; 98struct simplelock spechash_slock; 99 100void vclean(struct vnode *, int, struct proc *); 101 102void insmntque(struct vnode *, struct mount *); 103int getdevvp(dev_t, struct vnode **, enum vtype); 104 105int vfs_hang_addrlist(struct mount *, struct netexport *, 106 struct export_args *); 107int vfs_free_netcred(struct radix_node *, void *); 108void vfs_free_addrlist(struct netexport *); 109void vputonfreelist(struct vnode *); 110 111int vflush_vnode(struct vnode *, void *); 112 113#ifdef DEBUG 114void printlockedvnodes(void); 115#endif 116 117#define VN_KNOTE(vp, b) \ 118 KNOTE((struct klist *)&vp->v_selectinfo.vsi_selinfo.si_note, (b)) 119 120struct pool vnode_pool; 121 122/* 123 * Initialize the vnode management data structures. 124 */ 125void 126vntblinit() 127{ 128 129 pool_init(&vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodes", 130 &pool_allocator_nointr); 131 simple_lock_init(&mntvnode_slock); 132 simple_lock_init(&mntid_slock); 133 simple_lock_init(&spechash_slock); 134 TAILQ_INIT(&vnode_hold_list); 135 TAILQ_INIT(&vnode_free_list); 136 simple_lock_init(&vnode_free_list_slock); 137 CIRCLEQ_INIT(&mountlist); 138 simple_lock_init(&mountlist_slock); 139 /* 140 * Initialize the filesystem syncer. 141 */ 142 vn_initialize_syncerd(); 143} 144 145/* 146 * Mark a mount point as busy. Used to synchronize access and to delay 147 * unmounting. Interlock is not released on failure. 148 * 149 * historical behavior: 150 * - LK_NOWAIT means that we should just ignore the mount point if it's 151 * being unmounted. 152 * - no flags means that we should sleep on the mountpoint and then 153 * fail. 154 */ 155int 156vfs_busy(struct mount *mp, int flags, struct simplelock *interlkp, 157 struct proc *p) 158{ 159 int lkflags; 160 161 switch (flags) { 162 case LK_NOWAIT: 163 lkflags = LK_SHARED|LK_NOWAIT; 164 break; 165 case 0: 166 lkflags = LK_SHARED; 167 break; 168 default: 169 lkflags = flags; 170 } 171 172 /* 173 * Always sleepfail. We will only sleep for an exclusive lock 174 * and the exclusive lock will only be acquired when unmounting. 175 */ 176 lkflags |= LK_SLEEPFAIL; 177 178 if (interlkp) 179 lkflags |= LK_INTERLOCK; 180 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 181 return (ENOENT); 182 return (0); 183} 184 185 186/* 187 * Free a busy file system 188 */ 189void 190vfs_unbusy(struct mount *mp, struct proc *p) 191{ 192 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 193} 194 195int 196vfs_isbusy(struct mount *mp) 197{ 198 return (lockstatus(&mp->mnt_lock)); 199} 200 201/* 202 * Lookup a filesystem type, and if found allocate and initialize 203 * a mount structure for it. 204 * 205 * Devname is usually updated by mount(8) after booting. 206 */ 207int 208vfs_rootmountalloc(fstypename, devname, mpp) 209 char *fstypename; 210 char *devname; 211 struct mount **mpp; 212{ 213 struct proc *p = curproc; /* XXX */ 214 struct vfsconf *vfsp; 215 struct mount *mp; 216 217 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 218 if (!strcmp(vfsp->vfc_name, fstypename)) 219 break; 220 if (vfsp == NULL) 221 return (ENODEV); 222 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 223 bzero((char *)mp, (u_long)sizeof(struct mount)); 224 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 225 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 226 LIST_INIT(&mp->mnt_vnodelist); 227 mp->mnt_vfc = vfsp; 228 mp->mnt_op = vfsp->vfc_vfsops; 229 mp->mnt_flag = MNT_RDONLY; 230 mp->mnt_vnodecovered = NULLVP; 231 vfsp->vfc_refcount++; 232 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 233 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 234 mp->mnt_stat.f_mntonname[0] = '/'; 235 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 236 *mpp = mp; 237 return (0); 238 } 239 240/* 241 * Find an appropriate filesystem to use for the root. If a filesystem 242 * has not been preselected, walk through the list of known filesystems 243 * trying those that have mountroot routines, and try them until one 244 * works or we have tried them all. 245 */ 246int 247vfs_mountroot() 248{ 249 struct vfsconf *vfsp; 250 int error; 251 252 if (mountroot != NULL) 253 return ((*mountroot)()); 254 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 255 if (vfsp->vfc_mountroot == NULL) 256 continue; 257 if ((error = (*vfsp->vfc_mountroot)()) == 0) 258 return (0); 259 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 260 } 261 return (ENODEV); 262} 263 264/* 265 * Lookup a mount point by filesystem identifier. 266 */ 267struct mount * 268vfs_getvfs(fsid) 269 fsid_t *fsid; 270{ 271 register struct mount *mp; 272 273 simple_lock(&mountlist_slock); 274 CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) { 275 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 276 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 277 simple_unlock(&mountlist_slock); 278 return (mp); 279 } 280 } 281 simple_unlock(&mountlist_slock); 282 return ((struct mount *)0); 283} 284 285 286/* 287 * Get a new unique fsid 288 */ 289void 290vfs_getnewfsid(mp) 291 struct mount *mp; 292{ 293 static u_short xxxfs_mntid; 294 295 fsid_t tfsid; 296 int mtype; 297 298 simple_lock(&mntid_slock); 299 mtype = mp->mnt_vfc->vfc_typenum; 300 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 301 mp->mnt_stat.f_fsid.val[1] = mtype; 302 if (xxxfs_mntid == 0) 303 ++xxxfs_mntid; 304 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 305 tfsid.val[1] = mtype; 306 if (!CIRCLEQ_EMPTY(&mountlist)) { 307 while (vfs_getvfs(&tfsid)) { 308 tfsid.val[0]++; 309 xxxfs_mntid++; 310 } 311 } 312 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 313 simple_unlock(&mntid_slock); 314} 315 316/* 317 * Make a 'unique' number from a mount type name. 318 * Note that this is no longer used for ffs which 319 * now has an on-disk filesystem id. 320 */ 321long 322makefstype(type) 323 char *type; 324{ 325 long rv; 326 327 for (rv = 0; *type; type++) { 328 rv <<= 2; 329 rv ^= *type; 330 } 331 return rv; 332} 333 334/* 335 * Set vnode attributes to VNOVAL 336 */ 337void 338vattr_null(vap) 339 register struct vattr *vap; 340{ 341 342 vap->va_type = VNON; 343 /* XXX These next two used to be one line, but for a GCC bug. */ 344 vap->va_size = VNOVAL; 345 vap->va_bytes = VNOVAL; 346 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 347 vap->va_fsid = vap->va_fileid = 348 vap->va_blocksize = vap->va_rdev = 349 vap->va_atime.tv_sec = vap->va_atime.tv_nsec = 350 vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec = 351 vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec = 352 vap->va_flags = vap->va_gen = VNOVAL; 353 vap->va_vaflags = 0; 354} 355 356/* 357 * Routines having to do with the management of the vnode table. 358 */ 359extern int (**dead_vnodeop_p)(void *); 360long numvnodes; 361 362/* 363 * Return the next vnode from the free list. 364 */ 365int 366getnewvnode(tag, mp, vops, vpp) 367 enum vtagtype tag; 368 struct mount *mp; 369 int (**vops)(void *); 370 struct vnode **vpp; 371{ 372 struct proc *p = curproc; /* XXX */ 373 struct freelst *listhd; 374 static int toggle; 375 struct vnode *vp; 376 int s; 377 378 /* 379 * We must choose whether to allocate a new vnode or recycle an 380 * existing one. The criterion for allocating a new one is that 381 * the total number of vnodes is less than the number desired or 382 * there are no vnodes on either free list. Generally we only 383 * want to recycle vnodes that have no buffers associated with 384 * them, so we look first on the vnode_free_list. If it is empty, 385 * we next consider vnodes with referencing buffers on the 386 * vnode_hold_list. The toggle ensures that half the time we 387 * will use a buffer from the vnode_hold_list, and half the time 388 * we will allocate a new one unless the list has grown to twice 389 * the desired size. We are reticent to recycle vnodes from the 390 * vnode_hold_list because we will lose the identity of all its 391 * referencing buffers. 392 */ 393 toggle ^= 1; 394 if (numvnodes > 2 * desiredvnodes) 395 toggle = 0; 396 397 simple_lock(&vnode_free_list_slock); 398 s = splbio(); 399 if ((numvnodes < desiredvnodes) || 400 ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) && 401 ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) { 402 splx(s); 403 simple_unlock(&vnode_free_list_slock); 404 vp = pool_get(&vnode_pool, PR_WAITOK); 405 bzero((char *)vp, sizeof *vp); 406 simple_lock_init(&vp->v_interlock); 407 numvnodes++; 408 } else { 409 for (vp = TAILQ_FIRST(listhd); vp != NULLVP; 410 vp = TAILQ_NEXT(vp, v_freelist)) { 411 if (simple_lock_try(&vp->v_interlock)) { 412 if ((vp->v_flag & VLAYER) == 0) 413 break; 414 if (VOP_ISLOCKED(vp) == 0) 415 break; 416 else 417 simple_unlock(&vp->v_interlock); 418 } 419 } 420 /* 421 * Unless this is a bad time of the month, at most 422 * the first NCPUS items on the free list are 423 * locked, so this is close enough to being empty. 424 */ 425 if (vp == NULL) { 426 splx(s); 427 simple_unlock(&vnode_free_list_slock); 428 tablefull("vnode"); 429 *vpp = 0; 430 return (ENFILE); 431 } 432 if (vp->v_usecount) { 433 vprint("free vnode", vp); 434 panic("free vnode isn't"); 435 } 436 437 TAILQ_REMOVE(listhd, vp, v_freelist); 438 vp->v_bioflag &= ~VBIOONFREELIST; 439 splx(s); 440 441 simple_unlock(&vnode_free_list_slock); 442 if (vp->v_type != VBAD) 443 vgonel(vp, p); 444 else 445 simple_unlock(&vp->v_interlock); 446#ifdef DIAGNOSTIC 447 if (vp->v_data) { 448 vprint("cleaned vnode", vp); 449 panic("cleaned vnode isn't"); 450 } 451 s = splbio(); 452 if (vp->v_numoutput) 453 panic("Clean vnode has pending I/O's"); 454 splx(s); 455#endif 456 vp->v_flag = 0; 457 vp->v_socket = 0; 458 } 459 vp->v_type = VNON; 460 cache_purge(vp); 461 vp->v_vnlock = NULL; 462 lockinit(&vp->v_lock, PVFS, "v_lock", 0, 0); 463 vp->v_tag = tag; 464 vp->v_op = vops; 465 insmntque(vp, mp); 466 *vpp = vp; 467 vp->v_usecount = 1; 468 vp->v_data = 0; 469 simple_lock_init(&vp->v_uvm.u_obj.vmobjlock); 470 return (0); 471} 472 473/* 474 * Move a vnode from one mount queue to another. 475 */ 476void 477insmntque(vp, mp) 478 register struct vnode *vp; 479 register struct mount *mp; 480{ 481 simple_lock(&mntvnode_slock); 482 483 /* 484 * Delete from old mount point vnode list, if on one. 485 */ 486 if (vp->v_mount != NULL) 487 LIST_REMOVE(vp, v_mntvnodes); 488 /* 489 * Insert into list of vnodes for the new mount point, if available. 490 */ 491 if ((vp->v_mount = mp) != NULL) 492 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 493 494 simple_unlock(&mntvnode_slock); 495} 496 497 498/* 499 * Create a vnode for a block device. 500 * Used for root filesystem, argdev, and swap areas. 501 * Also used for memory file system special devices. 502 */ 503int 504bdevvp(dev, vpp) 505 dev_t dev; 506 struct vnode **vpp; 507{ 508 509 return (getdevvp(dev, vpp, VBLK)); 510} 511 512/* 513 * Create a vnode for a character device. 514 * Used for kernfs and some console handling. 515 */ 516int 517cdevvp(dev, vpp) 518 dev_t dev; 519 struct vnode **vpp; 520{ 521 522 return (getdevvp(dev, vpp, VCHR)); 523} 524 525/* 526 * Create a vnode for a device. 527 * Used by bdevvp (block device) for root file system etc., 528 * and by cdevvp (character device) for console and kernfs. 529 */ 530int 531getdevvp(dev, vpp, type) 532 dev_t dev; 533 struct vnode **vpp; 534 enum vtype type; 535{ 536 register struct vnode *vp; 537 struct vnode *nvp; 538 int error; 539 540 if (dev == NODEV) { 541 *vpp = NULLVP; 542 return (0); 543 } 544 error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp); 545 if (error) { 546 *vpp = NULLVP; 547 return (error); 548 } 549 vp = nvp; 550 vp->v_type = type; 551 if ((nvp = checkalias(vp, dev, NULL)) != 0) { 552 vput(vp); 553 vp = nvp; 554 } 555 *vpp = vp; 556 return (0); 557} 558 559/* 560 * Check to see if the new vnode represents a special device 561 * for which we already have a vnode (either because of 562 * bdevvp() or because of a different vnode representing 563 * the same block device). If such an alias exists, deallocate 564 * the existing contents and return the aliased vnode. The 565 * caller is responsible for filling it with its new contents. 566 */ 567struct vnode * 568checkalias(nvp, nvp_rdev, mp) 569 register struct vnode *nvp; 570 dev_t nvp_rdev; 571 struct mount *mp; 572{ 573 struct proc *p = curproc; 574 register struct vnode *vp; 575 struct vnode **vpp; 576 577 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 578 return (NULLVP); 579 580 vpp = &speclisth[SPECHASH(nvp_rdev)]; 581loop: 582 simple_lock(&spechash_slock); 583 for (vp = *vpp; vp; vp = vp->v_specnext) { 584 simple_lock(&vp->v_interlock); 585 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) { 586 simple_unlock(&vp->v_interlock); 587 continue; 588 } 589 /* 590 * Alias, but not in use, so flush it out. 591 */ 592 if (vp->v_usecount == 0) { 593 simple_unlock(&spechash_slock); 594 vgonel(vp, p); 595 goto loop; 596 } 597 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { 598 simple_unlock(&spechash_slock); 599 goto loop; 600 } 601 break; 602 } 603 604 /* 605 * Common case is actually in the if statement 606 */ 607 if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) { 608 MALLOC(nvp->v_specinfo, struct specinfo *, 609 sizeof(struct specinfo), M_VNODE, M_WAITOK); 610 nvp->v_rdev = nvp_rdev; 611 nvp->v_hashchain = vpp; 612 nvp->v_specnext = *vpp; 613 nvp->v_specmountpoint = NULL; 614 nvp->v_speclockf = NULL; 615 simple_unlock(&spechash_slock); 616 *vpp = nvp; 617 if (vp != NULLVP) { 618 nvp->v_flag |= VALIASED; 619 vp->v_flag |= VALIASED; 620 vput(vp); 621 } 622 return (NULLVP); 623 } 624 625 /* 626 * This code is the uncommon case. It is called in case 627 * we found an alias that was VT_NON && vtype of VBLK 628 * This means we found a block device that was created 629 * using bdevvp. 630 * An example of such a vnode is the root partition device vnode 631 * created in ffs_mountroot. 632 * 633 * The vnodes created by bdevvp should not be aliased (why?). 634 */ 635 636 simple_unlock(&spechash_slock); 637 VOP_UNLOCK(vp, 0, p); 638 simple_lock(&vp->v_interlock); 639 vclean(vp, 0, p); 640 vp->v_vnlock = NULL; 641 lockinit(&vp->v_lock, PVFS, "v_lock", 0, 0); 642 vp->v_op = nvp->v_op; 643 vp->v_tag = nvp->v_tag; 644 nvp->v_type = VNON; 645 insmntque(vp, mp); 646 return (vp); 647} 648 649/* 650 * Grab a particular vnode from the free list, increment its 651 * reference count and lock it. If the vnode lock bit is set, 652 * the vnode is being eliminated in vgone. In that case, we 653 * cannot grab it, so the process is awakened when the 654 * transition is completed, and an error code is returned to 655 * indicate that the vnode is no longer usable, possibly 656 * having been changed to a new file system type. 657 */ 658int 659vget(vp, flags, p) 660 struct vnode *vp; 661 int flags; 662 struct proc *p; 663{ 664 int error, s, onfreelist; 665 666 /* 667 * If the vnode is in the process of being cleaned out for 668 * another use, we wait for the cleaning to finish and then 669 * return failure. Cleaning is determined by checking that 670 * the VXLOCK flag is set. 671 */ 672 if ((flags & LK_INTERLOCK) == 0) { 673 simple_lock(&vp->v_interlock); 674 flags |= LK_INTERLOCK; 675 } 676 677 if (vp->v_flag & VXLOCK) { 678 if (flags & LK_NOWAIT) { 679 simple_unlock(&vp->v_interlock); 680 return (EBUSY); 681 } 682 683 vp->v_flag |= VXWANT; 684 ltsleep(vp, PINOD | PNORELOCK, "vget", 0, &vp->v_interlock); 685 return (ENOENT); 686 } 687 688 onfreelist = vp->v_bioflag & VBIOONFREELIST; 689 if (vp->v_usecount == 0 && onfreelist) { 690 s = splbio(); 691 simple_lock(&vnode_free_list_slock); 692 if (vp->v_holdcnt > 0) 693 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 694 else 695 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 696 simple_unlock(&vnode_free_list_slock); 697 vp->v_bioflag &= ~VBIOONFREELIST; 698 splx(s); 699 } 700 701 vp->v_usecount++; 702 if (flags & LK_TYPE_MASK) { 703 if ((error = vn_lock(vp, flags, p)) != 0) { 704 vp->v_usecount--; 705 if (vp->v_usecount == 0 && onfreelist) 706 vputonfreelist(vp); 707 708 simple_unlock(&vp->v_interlock); 709 } 710 return (error); 711 } 712 713 simple_unlock(&vp->v_interlock); 714 715 return (0); 716} 717 718 719#ifdef DIAGNOSTIC 720/* 721 * Vnode reference. 722 */ 723void 724vref(vp) 725 struct vnode *vp; 726{ 727 simple_lock(&vp->v_interlock); 728 if (vp->v_usecount == 0) 729 panic("vref used where vget required"); 730 vp->v_usecount++; 731 simple_unlock(&vp->v_interlock); 732} 733#endif /* DIAGNOSTIC */ 734 735void 736vputonfreelist(struct vnode *vp) 737{ 738 int s; 739 struct freelst *lst; 740 741 s = splbio(); 742#ifdef DIAGNOSTIC 743 if (vp->v_usecount != 0) 744 panic("Use count is not zero!"); 745 746 if (vp->v_bioflag & VBIOONFREELIST) { 747 vprint("vnode already on free list: ", vp); 748 panic("vnode already on free list"); 749 } 750#endif 751 752 vp->v_bioflag |= VBIOONFREELIST; 753 754 if (vp->v_holdcnt > 0) 755 lst = &vnode_hold_list; 756 else 757 lst = &vnode_free_list; 758 759 if (vp->v_type == VBAD) 760 TAILQ_INSERT_HEAD(lst, vp, v_freelist); 761 else 762 TAILQ_INSERT_TAIL(lst, vp, v_freelist); 763 764 splx(s); 765} 766 767/* 768 * vput(), just unlock and vrele() 769 */ 770void 771vput(vp) 772 register struct vnode *vp; 773{ 774 struct proc *p = curproc; /* XXX */ 775 776#ifdef DIAGNOSTIC 777 if (vp == NULL) 778 panic("vput: null vp"); 779#endif 780 simple_lock(&vp->v_interlock); 781 782#ifdef DIAGNOSTIC 783 if (vp->v_usecount == 0) { 784 vprint("vput: bad ref count", vp); 785 panic("vput: ref cnt"); 786 } 787#endif 788 vp->v_usecount--; 789 if (vp->v_usecount > 0) { 790 simple_unlock(&vp->v_interlock); 791 VOP_UNLOCK(vp, 0, p); 792 return; 793 } 794 795#ifdef DIAGNOSTIC 796 if (vp->v_writecount != 0) { 797 vprint("vput: bad writecount", vp); 798 panic("vput: v_writecount != 0"); 799 } 800#endif 801 simple_unlock(&vp->v_interlock); 802 803 VOP_INACTIVE(vp, p); 804 805 simple_lock(&vp->v_interlock); 806 807 if (vp->v_usecount == 0) 808 vputonfreelist(vp); 809 810 simple_unlock(&vp->v_interlock); 811} 812 813/* 814 * Vnode release - use for active VNODES. 815 * If count drops to zero, call inactive routine and return to freelist. 816 */ 817void 818vrele(vp) 819 register struct vnode *vp; 820{ 821 struct proc *p = curproc; /* XXX */ 822 823#ifdef DIAGNOSTIC 824 if (vp == NULL) 825 panic("vrele: null vp"); 826#endif 827 simple_lock(&vp->v_interlock); 828#ifdef DIAGNOSTIC 829 if (vp->v_usecount == 0) { 830 vprint("vrele: bad ref count", vp); 831 panic("vrele: ref cnt"); 832 } 833#endif 834 vp->v_usecount--; 835 if (vp->v_usecount > 0) { 836 simple_unlock(&vp->v_interlock); 837 return; 838 } 839 840#ifdef DIAGNOSTIC 841 if (vp->v_writecount != 0) { 842 vprint("vrele: bad writecount", vp); 843 panic("vrele: v_writecount != 0"); 844 } 845#endif 846 if (vn_lock(vp, LK_EXCLUSIVE|LK_INTERLOCK, p)) { 847 vprint("vrele: cannot lock", vp); 848 return; 849 } 850 851 VOP_INACTIVE(vp, p); 852 853 simple_lock(&vp->v_interlock); 854 855 if (vp->v_usecount == 0) 856 vputonfreelist(vp); 857 858 simple_unlock(&vp->v_interlock); 859} 860 861void vhold(struct vnode *vp); 862 863/* 864 * Page or buffer structure gets a reference. 865 */ 866void 867vhold(vp) 868 register struct vnode *vp; 869{ 870 871 /* 872 * If it is on the freelist and the hold count is currently 873 * zero, move it to the hold list. 874 */ 875 simple_lock(&vp->v_interlock); 876 if ((vp->v_bioflag & VBIOONFREELIST) && 877 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 878 simple_lock(&vnode_free_list_slock); 879 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 880 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 881 simple_unlock(&vnode_free_list_slock); 882 } 883 vp->v_holdcnt++; 884 simple_unlock(&vp->v_interlock); 885} 886 887/* 888 * Remove any vnodes in the vnode table belonging to mount point mp. 889 * 890 * If MNT_NOFORCE is specified, there should not be any active ones, 891 * return error if any are found (nb: this is a user error, not a 892 * system error). If MNT_FORCE is specified, detach any active vnodes 893 * that are found. 894 */ 895#ifdef DEBUG 896int busyprt = 0; /* print out busy vnodes */ 897struct ctldebug debug1 = { "busyprt", &busyprt }; 898#endif 899 900int 901vfs_mount_foreach_vnode(struct mount *mp, 902 int (*func)(struct vnode *, void *), void *arg) { 903 struct vnode *vp, *nvp; 904 int error = 0; 905 906 simple_lock(&mntvnode_slock); 907loop: 908 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) { 909 if (vp->v_mount != mp) 910 goto loop; 911 nvp = LIST_NEXT(vp, v_mntvnodes); 912 simple_lock(&vp->v_interlock); 913 simple_unlock(&mntvnode_slock); 914 915 error = func(vp, arg); 916 917 simple_lock(&mntvnode_slock); 918 919 if (error != 0) 920 break; 921 } 922 simple_unlock(&mntvnode_slock); 923 924 return (error); 925} 926 927struct vflush_args { 928 struct vnode *skipvp; 929 int busy; 930 int flags; 931}; 932 933int 934vflush_vnode(struct vnode *vp, void *arg) { 935 struct vflush_args *va = arg; 936 struct proc *p = curproc; 937 938 if (vp == va->skipvp) { 939 simple_unlock(&vp->v_interlock); 940 return (0); 941 } 942 943 if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 944 simple_unlock(&vp->v_interlock); 945 return (0); 946 } 947 948 /* 949 * If WRITECLOSE is set, only flush out regular file 950 * vnodes open for writing. 951 */ 952 if ((va->flags & WRITECLOSE) && 953 (vp->v_writecount == 0 || vp->v_type != VREG)) { 954 simple_unlock(&vp->v_interlock); 955 return (0); 956 } 957 958 /* 959 * With v_usecount == 0, all we need to do is clear 960 * out the vnode data structures and we are done. 961 */ 962 if (vp->v_usecount == 0) { 963 vgonel(vp, p); 964 return (0); 965 } 966 967 /* 968 * If FORCECLOSE is set, forcibly close the vnode. 969 * For block or character devices, revert to an 970 * anonymous device. For all other files, just kill them. 971 */ 972 if (va->flags & FORCECLOSE) { 973 if (vp->v_type != VBLK && vp->v_type != VCHR) { 974 vgonel(vp, p); 975 } else { 976 vclean(vp, 0, p); 977 vp->v_op = spec_vnodeop_p; 978 insmntque(vp, (struct mount *)0); 979 } 980 return (0); 981 } 982 983#ifdef DEBUG 984 if (busyprt) 985 vprint("vflush: busy vnode", vp); 986#endif 987 simple_unlock(&vp->v_interlock); 988 va->busy++; 989 return (0); 990} 991 992int 993vflush(mp, skipvp, flags) 994 struct mount *mp; 995 struct vnode *skipvp; 996 int flags; 997{ 998 struct vflush_args va; 999 va.skipvp = skipvp; 1000 va.busy = 0; 1001 va.flags = flags; 1002 1003 vfs_mount_foreach_vnode(mp, vflush_vnode, &va); 1004 1005 if (va.busy) 1006 return (EBUSY); 1007 return (0); 1008} 1009 1010/* 1011 * Disassociate the underlying file system from a vnode. 1012 * The vnode interlock is held on entry. 1013 */ 1014void 1015vclean(vp, flags, p) 1016 register struct vnode *vp; 1017 int flags; 1018 struct proc *p; 1019{ 1020 int active; 1021 1022 /* 1023 * Check to see if the vnode is in use. 1024 * If so we have to reference it before we clean it out 1025 * so that its count cannot fall to zero and generate a 1026 * race against ourselves to recycle it. 1027 */ 1028 if ((active = vp->v_usecount) != 0) 1029 vp->v_usecount++; 1030 1031 /* 1032 * Prevent the vnode from being recycled or 1033 * brought into use while we clean it out. 1034 */ 1035 if (vp->v_flag & VXLOCK) 1036 panic("vclean: deadlock"); 1037 vp->v_flag |= VXLOCK; 1038 /* 1039 * Even if the count is zero, the VOP_INACTIVE routine may still 1040 * have the object locked while it cleans it out. The VOP_LOCK 1041 * ensures that the VOP_INACTIVE routine is done with its work. 1042 * For active vnodes, it ensures that no other activity can 1043 * occur while the underlying object is being cleaned out. 1044 */ 1045 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1046 1047 /* 1048 * Clean out any VM data associated with the vnode. 1049 */ 1050 uvm_vnp_terminate(vp); 1051 /* 1052 * Clean out any buffers associated with the vnode. 1053 */ 1054 if (flags & DOCLOSE) 1055 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1056 /* 1057 * If purging an active vnode, it must be closed and 1058 * deactivated before being reclaimed. Note that the 1059 * VOP_INACTIVE will unlock the vnode 1060 */ 1061 if (active) { 1062 if (flags & DOCLOSE) 1063 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p); 1064 VOP_INACTIVE(vp, p); 1065 } else { 1066 /* 1067 * Any other processes trying to obtain this lock must first 1068 * wait for VXLOCK to clear, then call the new lock operation. 1069 */ 1070 VOP_UNLOCK(vp, 0, p); 1071 } 1072 1073 /* 1074 * Reclaim the vnode. 1075 */ 1076 if (VOP_RECLAIM(vp, p)) 1077 panic("vclean: cannot reclaim"); 1078 if (active) { 1079 simple_lock(&vp->v_interlock); 1080 1081 vp->v_usecount--; 1082 if (vp->v_usecount == 0) { 1083 if (vp->v_holdcnt > 0) 1084 panic("vclean: not clean"); 1085 vputonfreelist(vp); 1086 } 1087 1088 simple_unlock(&vp->v_interlock); 1089 } 1090 cache_purge(vp); 1091 1092 /* 1093 * Done with purge, notify sleepers of the grim news. 1094 */ 1095 vp->v_op = dead_vnodeop_p; 1096 simple_lock(&vp->v_selectinfo.vsi_lock); 1097 VN_KNOTE(vp, NOTE_REVOKE); 1098 simple_unlock(&vp->v_selectinfo.vsi_lock); 1099 vp->v_tag = VT_NON; 1100 vp->v_flag &= ~VXLOCK; 1101#ifdef DIAGNOSTIC 1102 vp->v_flag &= ~VLOCKSWORK; 1103#endif 1104 if (vp->v_flag & VXWANT) { 1105 vp->v_flag &= ~VXWANT; 1106 wakeup(vp); 1107 } 1108} 1109 1110/* 1111 * Recycle an unused vnode to the front of the free list. 1112 * Release the passed interlock if the vnode will be recycled. 1113 */ 1114int 1115vrecycle(vp, inter_lkp, p) 1116 struct vnode *vp; 1117 struct simplelock *inter_lkp; 1118 struct proc *p; 1119{ 1120 1121 simple_lock(&vp->v_interlock); 1122 if (vp->v_usecount == 0) { 1123 if (inter_lkp) 1124 simple_unlock(inter_lkp); 1125 vgonel(vp, p); 1126 return (1); 1127 } 1128 simple_unlock(&vp->v_interlock); 1129 return (0); 1130} 1131 1132 1133/* 1134 * Eliminate all activity associated with a vnode 1135 * in preparation for reuse. 1136 */ 1137void 1138vgone(vp) 1139 register struct vnode *vp; 1140{ 1141 struct proc *p = curproc; 1142 1143 simple_lock (&vp->v_interlock); 1144 vgonel(vp, p); 1145} 1146 1147/* 1148 * vgone, with the vp interlock held. 1149 */ 1150void 1151vgonel(vp, p) 1152 struct vnode *vp; 1153 struct proc *p; 1154{ 1155 register struct vnode *vq; 1156 struct vnode *vx; 1157 1158 /* 1159 * If a vgone (or vclean) is already in progress, 1160 * wait until it is done and return. 1161 */ 1162 if (vp->v_flag & VXLOCK) { 1163 vp->v_flag |= VXWANT; 1164 ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock); 1165 return; 1166 } 1167 1168 /* 1169 * Clean out the filesystem specific data. 1170 */ 1171 vclean(vp, DOCLOSE, p); 1172 /* 1173 * Delete from old mount point vnode list, if on one. 1174 */ 1175 if (vp->v_mount != NULL) 1176 insmntque(vp, (struct mount *)0); 1177 /* 1178 * If special device, remove it from special device alias list 1179 * if it is on one. 1180 */ 1181 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1182 simple_lock(&spechash_slock); 1183 if (*vp->v_hashchain == vp) { 1184 *vp->v_hashchain = vp->v_specnext; 1185 } else { 1186 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1187 if (vq->v_specnext != vp) 1188 continue; 1189 vq->v_specnext = vp->v_specnext; 1190 break; 1191 } 1192 if (vq == NULL) 1193 panic("missing bdev"); 1194 } 1195 if (vp->v_flag & VALIASED) { 1196 vx = NULL; 1197 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1198 if (vq->v_rdev != vp->v_rdev || 1199 vq->v_type != vp->v_type) 1200 continue; 1201 if (vx) 1202 break; 1203 vx = vq; 1204 } 1205 if (vx == NULL) 1206 panic("missing alias"); 1207 if (vq == NULL) 1208 vx->v_flag &= ~VALIASED; 1209 vp->v_flag &= ~VALIASED; 1210 } 1211 simple_unlock(&spechash_slock); 1212 FREE(vp->v_specinfo, M_VNODE); 1213 vp->v_specinfo = NULL; 1214 } 1215 /* 1216 * If it is on the freelist and not already at the head, 1217 * move it to the head of the list. 1218 */ 1219 vp->v_type = VBAD; 1220 1221 /* 1222 * Move onto the free list, unless we were called from 1223 * getnewvnode and we're not on any free list 1224 */ 1225 if (vp->v_usecount == 0 && 1226 (vp->v_bioflag & VBIOONFREELIST)) { 1227 int s; 1228 1229 simple_lock(&vnode_free_list_slock); 1230 s = splbio(); 1231 1232 if (vp->v_holdcnt > 0) 1233 panic("vgonel: not clean"); 1234 1235 if (TAILQ_FIRST(&vnode_free_list) != vp) { 1236 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1237 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1238 } 1239 splx(s); 1240 simple_unlock(&vnode_free_list_slock); 1241 } 1242} 1243 1244/* 1245 * Lookup a vnode by device number. 1246 */ 1247int 1248vfinddev(dev, type, vpp) 1249 dev_t dev; 1250 enum vtype type; 1251 struct vnode **vpp; 1252{ 1253 register struct vnode *vp; 1254 int rc =0; 1255 1256 simple_lock(&spechash_slock); 1257 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1258 if (dev != vp->v_rdev || type != vp->v_type) 1259 continue; 1260 *vpp = vp; 1261 rc = 1; 1262 break; 1263 } 1264 simple_unlock(&spechash_slock); 1265 return (rc); 1266} 1267 1268/* 1269 * Revoke all the vnodes corresponding to the specified minor number 1270 * range (endpoints inclusive) of the specified major. 1271 */ 1272void 1273vdevgone(maj, minl, minh, type) 1274 int maj, minl, minh; 1275 enum vtype type; 1276{ 1277 struct vnode *vp; 1278 int mn; 1279 1280 for (mn = minl; mn <= minh; mn++) 1281 if (vfinddev(makedev(maj, mn), type, &vp)) 1282 VOP_REVOKE(vp, REVOKEALL); 1283} 1284 1285/* 1286 * Calculate the total number of references to a special device. 1287 */ 1288int 1289vcount(vp) 1290 struct vnode *vp; 1291{ 1292 struct vnode *vq, *vnext; 1293 int count; 1294 1295loop: 1296 if ((vp->v_flag & VALIASED) == 0) 1297 return (vp->v_usecount); 1298 simple_lock(&spechash_slock); 1299 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1300 vnext = vq->v_specnext; 1301 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1302 continue; 1303 /* 1304 * Alias, but not in use, so flush it out. 1305 */ 1306 if (vq->v_usecount == 0 && vq != vp) { 1307 simple_unlock(&spechash_slock); 1308 vgone(vq); 1309 goto loop; 1310 } 1311 count += vq->v_usecount; 1312 } 1313 simple_unlock(&spechash_slock); 1314 return (count); 1315} 1316 1317/* 1318 * Print out a description of a vnode. 1319 */ 1320static char *typename[] = 1321 { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; 1322 1323void 1324vprint(label, vp) 1325 char *label; 1326 register struct vnode *vp; 1327{ 1328 char buf[64]; 1329 1330 if (label != NULL) 1331 printf("%s: ", label); 1332 printf("type %s, usecount %u, writecount %u, holdcount %u,", 1333 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1334 vp->v_holdcnt); 1335 buf[0] = '\0'; 1336 if (vp->v_flag & VROOT) 1337 strlcat(buf, "|VROOT", sizeof buf); 1338 if (vp->v_flag & VTEXT) 1339 strlcat(buf, "|VTEXT", sizeof buf); 1340 if (vp->v_flag & VSYSTEM) 1341 strlcat(buf, "|VSYSTEM", sizeof buf); 1342 if (vp->v_flag & VXLOCK) 1343 strlcat(buf, "|VXLOCK", sizeof buf); 1344 if (vp->v_flag & VXWANT) 1345 strlcat(buf, "|VXWANT", sizeof buf); 1346 if (vp->v_bioflag & VBIOWAIT) 1347 strlcat(buf, "|VBIOWAIT", sizeof buf); 1348 if (vp->v_bioflag & VBIOONFREELIST) 1349 strlcat(buf, "|VBIOONFREELIST", sizeof buf); 1350 if (vp->v_bioflag & VBIOONSYNCLIST) 1351 strlcat(buf, "|VBIOONSYNCLIST", sizeof buf); 1352 if (vp->v_flag & VALIASED) 1353 strlcat(buf, "|VALIASED", sizeof buf); 1354 if (buf[0] != '\0') 1355 printf(" flags (%s)", &buf[1]); 1356 if (vp->v_data == NULL) { 1357 printf("\n"); 1358 } else { 1359 printf("\n\t"); 1360 VOP_PRINT(vp); 1361 } 1362} 1363 1364#ifdef DEBUG 1365/* 1366 * List all of the locked vnodes in the system. 1367 * Called when debugging the kernel. 1368 */ 1369void 1370printlockedvnodes() 1371{ 1372 struct proc *p = curproc; 1373 register struct mount *mp, *nmp; 1374 register struct vnode *vp; 1375 1376 printf("Locked vnodes\n"); 1377 simple_lock(&mountlist_slock); 1378 for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist); 1379 mp = nmp) { 1380 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1381 nmp = CIRCLEQ_NEXT(mp, mnt_list); 1382 continue; 1383 } 1384 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 1385 if (VOP_ISLOCKED(vp)) 1386 vprint((char *)0, vp); 1387 } 1388 simple_lock(&mountlist_slock); 1389 nmp = CIRCLEQ_NEXT(mp, mnt_list); 1390 vfs_unbusy(mp, p); 1391 } 1392 simple_unlock(&mountlist_slock); 1393 1394} 1395#endif 1396 1397/* 1398 * Top level filesystem related information gathering. 1399 */ 1400int 1401vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) 1402 int *name; 1403 u_int namelen; 1404 void *oldp; 1405 size_t *oldlenp; 1406 void *newp; 1407 size_t newlen; 1408 struct proc *p; 1409{ 1410 struct vfsconf *vfsp; 1411 1412 /* all sysctl names at this level are at least name and field */ 1413 if (namelen < 2) 1414 return (ENOTDIR); /* overloaded */ 1415 if (name[0] != VFS_GENERIC) { 1416 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1417 if (vfsp->vfc_typenum == name[0]) 1418 break; 1419 if (vfsp == NULL) 1420 return (EOPNOTSUPP); 1421 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1422 oldp, oldlenp, newp, newlen, p)); 1423 } 1424 switch (name[1]) { 1425 case VFS_MAXTYPENUM: 1426 return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf)); 1427 case VFS_CONF: 1428 if (namelen < 3) 1429 return (ENOTDIR); /* overloaded */ 1430 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1431 if (vfsp->vfc_typenum == name[2]) 1432 break; 1433 if (vfsp == NULL) 1434 return (EOPNOTSUPP); 1435 return (sysctl_rdstruct(oldp, oldlenp, newp, vfsp, 1436 sizeof(struct vfsconf))); 1437 } 1438 return (EOPNOTSUPP); 1439} 1440 1441int kinfo_vdebug = 1; 1442int kinfo_vgetfailed; 1443#define KINFO_VNODESLOP 10 1444/* 1445 * Dump vnode list (via sysctl). 1446 * Copyout address of vnode followed by vnode. 1447 */ 1448/* ARGSUSED */ 1449int 1450sysctl_vnode(where, sizep, p) 1451 char *where; 1452 size_t *sizep; 1453 struct proc *p; 1454{ 1455 register struct mount *mp, *nmp; 1456 struct vnode *vp, *nvp; 1457 register char *bp = where, *savebp; 1458 char *ewhere; 1459 int error; 1460 1461 if (where == NULL) { 1462 *sizep = (numvnodes + KINFO_VNODESLOP) * sizeof(struct e_vnode); 1463 return (0); 1464 } 1465 ewhere = where + *sizep; 1466 1467 simple_lock(&mountlist_slock); 1468 for (mp = CIRCLEQ_FIRST(&mountlist); mp != CIRCLEQ_END(&mountlist); 1469 mp = nmp) { 1470 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1471 nmp = CIRCLEQ_NEXT(mp, mnt_list); 1472 continue; 1473 } 1474 savebp = bp; 1475again: 1476 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; 1477 vp = nvp) { 1478 /* 1479 * Check that the vp is still associated with 1480 * this filesystem. RACE: could have been 1481 * recycled onto the same filesystem. 1482 */ 1483 if (vp->v_mount != mp) { 1484 simple_unlock(&mntvnode_slock); 1485 if (kinfo_vdebug) 1486 printf("kinfo: vp changed\n"); 1487 bp = savebp; 1488 goto again; 1489 } 1490 nvp = LIST_NEXT(vp, v_mntvnodes); 1491 if (bp + sizeof(struct e_vnode) > ewhere) { 1492 simple_unlock(&mntvnode_slock); 1493 *sizep = bp - where; 1494 vfs_unbusy(mp, p); 1495 return (ENOMEM); 1496 } 1497 if ((error = copyout(&vp, 1498 &((struct e_vnode *)bp)->vptr, 1499 sizeof(struct vnode *))) || 1500 (error = copyout(vp, 1501 &((struct e_vnode *)bp)->vnode, 1502 sizeof(struct vnode)))) { 1503 vfs_unbusy(mp, p); 1504 return (error); 1505 } 1506 bp += sizeof(struct e_vnode); 1507 simple_lock(&mntvnode_slock); 1508 } 1509 1510 simple_unlock(&mntvnode_slock); 1511 simple_lock(&mountlist_slock); 1512 nmp = CIRCLEQ_NEXT(mp, mnt_list); 1513 vfs_unbusy(mp, p); 1514 } 1515 1516 simple_unlock(&mountlist_slock); 1517 1518 *sizep = bp - where; 1519 return (0); 1520} 1521 1522/* 1523 * Check to see if a filesystem is mounted on a block device. 1524 */ 1525int 1526vfs_mountedon(vp) 1527 register struct vnode *vp; 1528{ 1529 register struct vnode *vq; 1530 int error = 0; 1531 1532 if (vp->v_specmountpoint != NULL) 1533 return (EBUSY); 1534 if (vp->v_flag & VALIASED) { 1535 simple_lock(&spechash_slock); 1536 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1537 if (vq->v_rdev != vp->v_rdev || 1538 vq->v_type != vp->v_type) 1539 continue; 1540 if (vq->v_specmountpoint != NULL) { 1541 error = EBUSY; 1542 break; 1543 } 1544 } 1545 simple_unlock(&spechash_slock); 1546 } 1547 return (error); 1548} 1549 1550/* 1551 * Build hash lists of net addresses and hang them off the mount point. 1552 * Called by ufs_mount() to set up the lists of export addresses. 1553 */ 1554int 1555vfs_hang_addrlist(mp, nep, argp) 1556 struct mount *mp; 1557 struct netexport *nep; 1558 struct export_args *argp; 1559{ 1560 register struct netcred *np; 1561 register struct radix_node_head *rnh; 1562 register int i; 1563 struct radix_node *rn; 1564 struct sockaddr *saddr, *smask = 0; 1565 struct domain *dom; 1566 int error; 1567 1568 if (argp->ex_addrlen == 0) { 1569 if (mp->mnt_flag & MNT_DEFEXPORTED) 1570 return (EPERM); 1571 np = &nep->ne_defexported; 1572 np->netc_exflags = argp->ex_flags; 1573 np->netc_anon = argp->ex_anon; 1574 np->netc_anon.cr_ref = 1; 1575 mp->mnt_flag |= MNT_DEFEXPORTED; 1576 return (0); 1577 } 1578 if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN || 1579 argp->ex_addrlen < 0 || argp->ex_masklen < 0) 1580 return (EINVAL); 1581 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1582 np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); 1583 bzero(np, i); 1584 saddr = (struct sockaddr *)(np + 1); 1585 error = copyin(argp->ex_addr, saddr, argp->ex_addrlen); 1586 if (error) 1587 goto out; 1588 if (saddr->sa_len > argp->ex_addrlen) 1589 saddr->sa_len = argp->ex_addrlen; 1590 if (argp->ex_masklen) { 1591 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1592 error = copyin(argp->ex_mask, smask, argp->ex_masklen); 1593 if (error) 1594 goto out; 1595 if (smask->sa_len > argp->ex_masklen) 1596 smask->sa_len = argp->ex_masklen; 1597 } 1598 i = saddr->sa_family; 1599 if (i < 0 || i > AF_MAX) { 1600 error = EINVAL; 1601 goto out; 1602 } 1603 if ((rnh = nep->ne_rtable[i]) == 0) { 1604 /* 1605 * Seems silly to initialize every AF when most are not 1606 * used, do so on demand here 1607 */ 1608 for (dom = domains; dom; dom = dom->dom_next) 1609 if (dom->dom_family == i && dom->dom_rtattach) { 1610 dom->dom_rtattach((void **)&nep->ne_rtable[i], 1611 dom->dom_rtoffset); 1612 break; 1613 } 1614 if ((rnh = nep->ne_rtable[i]) == 0) { 1615 error = ENOBUFS; 1616 goto out; 1617 } 1618 } 1619 rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh, 1620 np->netc_rnodes); 1621 if (rn == 0 || np != (struct netcred *)rn) { /* already exists */ 1622 error = EPERM; 1623 goto out; 1624 } 1625 np->netc_exflags = argp->ex_flags; 1626 np->netc_anon = argp->ex_anon; 1627 np->netc_anon.cr_ref = 1; 1628 return (0); 1629out: 1630 free(np, M_NETADDR); 1631 return (error); 1632} 1633 1634/* ARGSUSED */ 1635int 1636vfs_free_netcred(rn, w) 1637 struct radix_node *rn; 1638 void *w; 1639{ 1640 register struct radix_node_head *rnh = (struct radix_node_head *)w; 1641 1642 (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh, NULL); 1643 free(rn, M_NETADDR); 1644 return (0); 1645} 1646 1647/* 1648 * Free the net address hash lists that are hanging off the mount points. 1649 */ 1650void 1651vfs_free_addrlist(nep) 1652 struct netexport *nep; 1653{ 1654 register int i; 1655 register struct radix_node_head *rnh; 1656 1657 for (i = 0; i <= AF_MAX; i++) 1658 if ((rnh = nep->ne_rtable[i]) != NULL) { 1659 (*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh); 1660 free(rnh, M_RTABLE); 1661 nep->ne_rtable[i] = 0; 1662 } 1663} 1664 1665int 1666vfs_export(mp, nep, argp) 1667 struct mount *mp; 1668 struct netexport *nep; 1669 struct export_args *argp; 1670{ 1671 int error; 1672 1673 if (argp->ex_flags & MNT_DELEXPORT) { 1674 vfs_free_addrlist(nep); 1675 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1676 } 1677 if (argp->ex_flags & MNT_EXPORTED) { 1678 if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0) 1679 return (error); 1680 mp->mnt_flag |= MNT_EXPORTED; 1681 } 1682 return (0); 1683} 1684 1685struct netcred * 1686vfs_export_lookup(mp, nep, nam) 1687 register struct mount *mp; 1688 struct netexport *nep; 1689 struct mbuf *nam; 1690{ 1691 register struct netcred *np; 1692 register struct radix_node_head *rnh; 1693 struct sockaddr *saddr; 1694 1695 np = NULL; 1696 if (mp->mnt_flag & MNT_EXPORTED) { 1697 /* 1698 * Lookup in the export list first. 1699 */ 1700 if (nam != NULL) { 1701 saddr = mtod(nam, struct sockaddr *); 1702 rnh = nep->ne_rtable[saddr->sa_family]; 1703 if (rnh != NULL) { 1704 np = (struct netcred *) 1705 (*rnh->rnh_matchaddr)((caddr_t)saddr, 1706 rnh); 1707 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1708 np = NULL; 1709 } 1710 } 1711 /* 1712 * If no address match, use the default if it exists. 1713 */ 1714 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1715 np = &nep->ne_defexported; 1716 } 1717 return (np); 1718} 1719 1720/* 1721 * Do the usual access checking. 1722 * file_mode, uid and gid are from the vnode in question, 1723 * while acc_mode and cred are from the VOP_ACCESS parameter list 1724 */ 1725int 1726vaccess(file_mode, uid, gid, acc_mode, cred) 1727 mode_t file_mode; 1728 uid_t uid; 1729 gid_t gid; 1730 mode_t acc_mode; 1731 struct ucred *cred; 1732{ 1733 mode_t mask; 1734 1735 /* User id 0 always gets access. */ 1736 if (cred->cr_uid == 0) 1737 return 0; 1738 1739 mask = 0; 1740 1741 /* Otherwise, check the owner. */ 1742 if (cred->cr_uid == uid) { 1743 if (acc_mode & VEXEC) 1744 mask |= S_IXUSR; 1745 if (acc_mode & VREAD) 1746 mask |= S_IRUSR; 1747 if (acc_mode & VWRITE) 1748 mask |= S_IWUSR; 1749 return (file_mode & mask) == mask ? 0 : EACCES; 1750 } 1751 1752 /* Otherwise, check the groups. */ 1753 if (cred->cr_gid == gid || groupmember(gid, cred)) { 1754 if (acc_mode & VEXEC) 1755 mask |= S_IXGRP; 1756 if (acc_mode & VREAD) 1757 mask |= S_IRGRP; 1758 if (acc_mode & VWRITE) 1759 mask |= S_IWGRP; 1760 return (file_mode & mask) == mask ? 0 : EACCES; 1761 } 1762 1763 /* Otherwise, check everyone else. */ 1764 if (acc_mode & VEXEC) 1765 mask |= S_IXOTH; 1766 if (acc_mode & VREAD) 1767 mask |= S_IROTH; 1768 if (acc_mode & VWRITE) 1769 mask |= S_IWOTH; 1770 return (file_mode & mask) == mask ? 0 : EACCES; 1771} 1772 1773/* 1774 * Unmount all file systems. 1775 * We traverse the list in reverse order under the assumption that doing so 1776 * will avoid needing to worry about dependencies. 1777 */ 1778void 1779vfs_unmountall(void) 1780{ 1781 struct mount *mp, *nmp; 1782 int allerror, error, again = 1; 1783 struct proc *p = curproc; 1784 1785 retry: 1786 allerror = 0; 1787 for (mp = CIRCLEQ_LAST(&mountlist); mp != CIRCLEQ_END(&mountlist); 1788 mp = nmp) { 1789 nmp = CIRCLEQ_PREV(mp, mnt_list); 1790 if ((vfs_busy(mp, LK_EXCLUSIVE|LK_NOWAIT, NULL, p)) != 0) 1791 continue; 1792 if ((error = dounmount(mp, MNT_FORCE, curproc, NULL)) != 0) { 1793 printf("unmount of %s failed with error %d\n", 1794 mp->mnt_stat.f_mntonname, error); 1795 allerror = 1; 1796 } 1797 } 1798 1799 if (allerror) { 1800 printf("WARNING: some file systems would not unmount\n"); 1801 if (again) { 1802 printf("retrying\n"); 1803 again = 0; 1804 goto retry; 1805 } 1806 } 1807} 1808 1809/* 1810 * Sync and unmount file systems before shutting down. 1811 */ 1812void 1813vfs_shutdown() 1814{ 1815#ifdef ACCOUNTING 1816 extern void acct_shutdown(void); 1817 1818 acct_shutdown(); 1819#endif 1820 1821 /* XXX Should suspend scheduling. */ 1822 (void) spl0(); 1823 1824 printf("syncing disks... "); 1825 1826 if (panicstr == 0) { 1827 /* Sync before unmount, in case we hang on something. */ 1828 sys_sync(&proc0, (void *)0, (register_t *)0); 1829 1830 /* Unmount file systems. */ 1831 vfs_unmountall(); 1832 } 1833 1834 if (vfs_syncwait(1)) 1835 printf("giving up\n"); 1836 else 1837 printf("done\n"); 1838} 1839 1840/* 1841 * perform sync() operation and wait for buffers to flush. 1842 * assumtions: called w/ scheduler disabled and physical io enabled 1843 * for now called at spl0() XXX 1844 */ 1845int 1846vfs_syncwait(verbose) 1847 int verbose; 1848{ 1849 register struct buf *bp; 1850 int iter, nbusy, dcount, s; 1851 struct proc *p; 1852 1853 p = curproc? curproc : &proc0; 1854 sys_sync(p, (void *)0, (register_t *)0); 1855 1856 /* Wait for sync to finish. */ 1857 dcount = 10000; 1858 for (iter = 0; iter < 20; iter++) { 1859 nbusy = 0; 1860 for (bp = &buf[nbuf]; --bp >= buf; ) { 1861 if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY) 1862 nbusy++; 1863 /* 1864 * With soft updates, some buffers that are 1865 * written will be remarked as dirty until other 1866 * buffers are written. 1867 */ 1868 if (bp->b_flags & B_DELWRI) { 1869 s = splbio(); 1870 bremfree(bp); 1871 bp->b_flags |= B_BUSY; 1872 splx(s); 1873 nbusy++; 1874 bawrite(bp); 1875 if (dcount-- <= 0) { 1876 if (verbose) 1877 printf("softdep "); 1878 return 1; 1879 } 1880 } 1881 } 1882 if (nbusy == 0) 1883 break; 1884 if (verbose) 1885 printf("%d ", nbusy); 1886 DELAY(40000 * iter); 1887 } 1888 1889 return nbusy; 1890} 1891 1892/* 1893 * posix file system related system variables. 1894 */ 1895int 1896fs_posix_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) 1897 int *name; 1898 u_int namelen; 1899 void *oldp; 1900 size_t *oldlenp; 1901 void *newp; 1902 size_t newlen; 1903 struct proc *p; 1904{ 1905 /* all sysctl names at this level are terminal */ 1906 if (namelen != 1) 1907 return (ENOTDIR); 1908 1909 switch (name[0]) { 1910 case FS_POSIX_SETUID: 1911 if (newp && securelevel > 0) 1912 return (EPERM); 1913 return(sysctl_int(oldp, oldlenp, newp, newlen, &suid_clear)); 1914 default: 1915 return (EOPNOTSUPP); 1916 } 1917 /* NOTREACHED */ 1918} 1919 1920/* 1921 * file system related system variables. 1922 */ 1923int 1924fs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) 1925 int *name; 1926 u_int namelen; 1927 void *oldp; 1928 size_t *oldlenp; 1929 void *newp; 1930 size_t newlen; 1931 struct proc *p; 1932{ 1933 sysctlfn *fn; 1934 1935 switch (name[0]) { 1936 case FS_POSIX: 1937 fn = fs_posix_sysctl; 1938 break; 1939 default: 1940 return (EOPNOTSUPP); 1941 } 1942 return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p); 1943} 1944 1945 1946/* 1947 * Routines dealing with vnodes and buffers 1948 */ 1949 1950/* 1951 * Wait for all outstanding I/Os to complete 1952 * 1953 * Manipulates v_numoutput. Must be called at splbio() 1954 */ 1955int 1956vwaitforio(vp, slpflag, wmesg, timeo) 1957 struct vnode *vp; 1958 int slpflag, timeo; 1959 char *wmesg; 1960{ 1961 int error = 0; 1962 1963 splassert(IPL_BIO); 1964 1965 while (vp->v_numoutput) { 1966 vp->v_bioflag |= VBIOWAIT; 1967 error = tsleep(&vp->v_numoutput, 1968 slpflag | (PRIBIO + 1), wmesg, timeo); 1969 if (error) 1970 break; 1971 } 1972 1973 return (error); 1974} 1975 1976/* 1977 * Update outstanding I/O count and do wakeup if requested. 1978 * 1979 * Manipulates v_numoutput. Must be called at splbio() 1980 */ 1981void 1982vwakeup(vp) 1983 struct vnode *vp; 1984{ 1985 splassert(IPL_BIO); 1986 1987 if (vp != NULL) { 1988 if (vp->v_numoutput-- == 0) 1989 panic("vwakeup: neg numoutput"); 1990 if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) { 1991 vp->v_bioflag &= ~VBIOWAIT; 1992 wakeup(&vp->v_numoutput); 1993 } 1994 } 1995} 1996 1997/* 1998 * Flush out and invalidate all buffers associated with a vnode. 1999 * Called with the underlying object locked. 2000 */ 2001int 2002vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 2003 register struct vnode *vp; 2004 int flags; 2005 struct ucred *cred; 2006 struct proc *p; 2007 int slpflag, slptimeo; 2008{ 2009 register struct buf *bp; 2010 struct buf *nbp, *blist; 2011 int s, error; 2012 2013 if (flags & V_SAVE) { 2014 s = splbio(); 2015 vwaitforio(vp, 0, "vinvalbuf", 0); 2016 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 2017 splx(s); 2018 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0) 2019 return (error); 2020 s = splbio(); 2021 if (vp->v_numoutput > 0 || 2022 !LIST_EMPTY(&vp->v_dirtyblkhd)) 2023 panic("vinvalbuf: dirty bufs"); 2024 } 2025 splx(s); 2026 } 2027loop: 2028 s = splbio(); 2029 for (;;) { 2030 if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) && 2031 (flags & V_SAVEMETA)) 2032 while (blist && blist->b_lblkno < 0) 2033 blist = LIST_NEXT(blist, b_vnbufs); 2034 if (blist == NULL && 2035 (blist = LIST_FIRST(&vp->v_dirtyblkhd)) && 2036 (flags & V_SAVEMETA)) 2037 while (blist && blist->b_lblkno < 0) 2038 blist = LIST_NEXT(blist, b_vnbufs); 2039 if (!blist) 2040 break; 2041 2042 for (bp = blist; bp; bp = nbp) { 2043 nbp = LIST_NEXT(bp, b_vnbufs); 2044 if (flags & V_SAVEMETA && bp->b_lblkno < 0) 2045 continue; 2046 if (bp->b_flags & B_BUSY) { 2047 bp->b_flags |= B_WANTED; 2048 error = tsleep(bp, slpflag | (PRIBIO + 1), 2049 "vinvalbuf", slptimeo); 2050 if (error) { 2051 splx(s); 2052 return (error); 2053 } 2054 break; 2055 } 2056 bremfree(bp); 2057 bp->b_flags |= B_BUSY; 2058 /* 2059 * XXX Since there are no node locks for NFS, I believe 2060 * there is a slight chance that a delayed write will 2061 * occur while sleeping just above, so check for it. 2062 */ 2063 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 2064 splx(s); 2065 (void) VOP_BWRITE(bp); 2066 goto loop; 2067 } 2068 bp->b_flags |= B_INVAL; 2069 brelse(bp); 2070 } 2071 } 2072 if (!(flags & V_SAVEMETA) && 2073 (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd))) 2074 panic("vinvalbuf: flush failed"); 2075 splx(s); 2076 return (0); 2077} 2078 2079void 2080vflushbuf(vp, sync) 2081 register struct vnode *vp; 2082 int sync; 2083{ 2084 register struct buf *bp, *nbp; 2085 int s; 2086 2087loop: 2088 s = splbio(); 2089 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); 2090 bp != LIST_END(&vp->v_dirtyblkhd); bp = nbp) { 2091 nbp = LIST_NEXT(bp, b_vnbufs); 2092 if ((bp->b_flags & B_BUSY)) 2093 continue; 2094 if ((bp->b_flags & B_DELWRI) == 0) 2095 panic("vflushbuf: not dirty"); 2096 bremfree(bp); 2097 bp->b_flags |= B_BUSY; 2098 splx(s); 2099 /* 2100 * Wait for I/O associated with indirect blocks to complete, 2101 * since there is no way to quickly wait for them below. 2102 */ 2103 if (bp->b_vp == vp || sync == 0) 2104 (void) bawrite(bp); 2105 else 2106 (void) bwrite(bp); 2107 goto loop; 2108 } 2109 if (sync == 0) { 2110 splx(s); 2111 return; 2112 } 2113 vwaitforio(vp, 0, "vflushbuf", 0); 2114 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 2115 splx(s); 2116 vprint("vflushbuf: dirty", vp); 2117 goto loop; 2118 } 2119 splx(s); 2120} 2121 2122/* 2123 * Associate a buffer with a vnode. 2124 * 2125 * Manipulates buffer vnode queues. Must be called at splbio(). 2126 */ 2127void 2128bgetvp(vp, bp) 2129 register struct vnode *vp; 2130 register struct buf *bp; 2131{ 2132 splassert(IPL_BIO); 2133 2134 2135 if (bp->b_vp) 2136 panic("bgetvp: not free"); 2137 vhold(vp); 2138 bp->b_vp = vp; 2139 if (vp->v_type == VBLK || vp->v_type == VCHR) 2140 bp->b_dev = vp->v_rdev; 2141 else 2142 bp->b_dev = NODEV; 2143 /* 2144 * Insert onto list for new vnode. 2145 */ 2146 bufinsvn(bp, &vp->v_cleanblkhd); 2147} 2148 2149/* 2150 * Disassociate a buffer from a vnode. 2151 * 2152 * Manipulates vnode buffer queues. Must be called at splbio(). 2153 */ 2154void 2155brelvp(bp) 2156 register struct buf *bp; 2157{ 2158 struct vnode *vp; 2159 2160 splassert(IPL_BIO); 2161 2162 if ((vp = bp->b_vp) == (struct vnode *) 0) 2163 panic("brelvp: NULL"); 2164 /* 2165 * Delete from old vnode list, if on one. 2166 */ 2167 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 2168 bufremvn(bp); 2169 if ((vp->v_bioflag & VBIOONSYNCLIST) && 2170 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2171 vp->v_bioflag &= ~VBIOONSYNCLIST; 2172 LIST_REMOVE(vp, v_synclist); 2173 } 2174 bp->b_vp = (struct vnode *) 0; 2175 2176 simple_lock(&vp->v_interlock); 2177#ifdef DIAGNOSTIC 2178 if (vp->v_holdcnt == 0) 2179 panic("brelvp: holdcnt"); 2180#endif 2181 vp->v_holdcnt--; 2182 2183 /* 2184 * If it is on the holdlist and the hold count drops to 2185 * zero, move it to the free list. 2186 */ 2187 if ((vp->v_bioflag & VBIOONFREELIST) && 2188 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 2189 simple_lock(&vnode_free_list_slock); 2190 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 2191 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2192 simple_unlock(&vnode_free_list_slock); 2193 } 2194 simple_unlock(&vp->v_interlock); 2195} 2196 2197/* 2198 * Replaces the current vnode associated with the buffer, if any, 2199 * with a new vnode. 2200 * 2201 * If an output I/O is pending on the buffer, the old vnode 2202 * I/O count is adjusted. 2203 * 2204 * Ignores vnode buffer queues. Must be called at splbio(). 2205 */ 2206void 2207buf_replacevnode(bp, newvp) 2208 struct buf *bp; 2209 struct vnode *newvp; 2210{ 2211 struct vnode *oldvp = bp->b_vp; 2212 2213 splassert(IPL_BIO); 2214 2215 if (oldvp) 2216 brelvp(bp); 2217 2218 if ((bp->b_flags & (B_READ | B_DONE)) == 0) { 2219 newvp->v_numoutput++; /* put it on swapdev */ 2220 vwakeup(oldvp); 2221 } 2222 2223 bgetvp(newvp, bp); 2224 bufremvn(bp); 2225} 2226 2227/* 2228 * Used to assign buffers to the appropriate clean or dirty list on 2229 * the vnode and to add newly dirty vnodes to the appropriate 2230 * filesystem syncer list. 2231 * 2232 * Manipulates vnode buffer queues. Must be called at splbio(). 2233 */ 2234void 2235reassignbuf(bp) 2236 struct buf *bp; 2237{ 2238 struct buflists *listheadp; 2239 int delay; 2240 struct vnode *vp = bp->b_vp; 2241 2242 splassert(IPL_BIO); 2243 2244 /* 2245 * Delete from old vnode list, if on one. 2246 */ 2247 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 2248 bufremvn(bp); 2249 2250 /* 2251 * If dirty, put on list of dirty buffers; 2252 * otherwise insert onto list of clean buffers. 2253 */ 2254 if ((bp->b_flags & B_DELWRI) == 0) { 2255 listheadp = &vp->v_cleanblkhd; 2256 if ((vp->v_bioflag & VBIOONSYNCLIST) && 2257 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2258 vp->v_bioflag &= ~VBIOONSYNCLIST; 2259 LIST_REMOVE(vp, v_synclist); 2260 } 2261 } else { 2262 listheadp = &vp->v_dirtyblkhd; 2263 if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) { 2264 switch (vp->v_type) { 2265 case VDIR: 2266 delay = syncdelay / 2; 2267 break; 2268 case VBLK: 2269 if (vp->v_specmountpoint != NULL) { 2270 delay = syncdelay / 3; 2271 break; 2272 } 2273 /* fall through */ 2274 default: 2275 delay = syncdelay; 2276 } 2277 vn_syncer_add_to_worklist(vp, delay); 2278 } 2279 } 2280 bufinsvn(bp, listheadp); 2281} 2282 2283int 2284vfs_register(vfs) 2285 struct vfsconf *vfs; 2286{ 2287 struct vfsconf *vfsp; 2288 struct vfsconf **vfspp; 2289 2290#ifdef DIAGNOSTIC 2291 /* Paranoia? */ 2292 if (vfs->vfc_refcount != 0) 2293 printf("vfs_register called with vfc_refcount > 0\n"); 2294#endif 2295 2296 /* Check if filesystem already known */ 2297 for (vfspp = &vfsconf, vfsp = vfsconf; vfsp; 2298 vfspp = &vfsp->vfc_next, vfsp = vfsp->vfc_next) 2299 if (strcmp(vfsp->vfc_name, vfs->vfc_name) == 0) 2300 return (EEXIST); 2301 2302 if (vfs->vfc_typenum > maxvfsconf) 2303 maxvfsconf = vfs->vfc_typenum; 2304 2305 vfs->vfc_next = NULL; 2306 2307 /* Add to the end of the list */ 2308 *vfspp = vfs; 2309 2310 /* Call vfs_init() */ 2311 if (vfs->vfc_vfsops->vfs_init) 2312 (*(vfs->vfc_vfsops->vfs_init))(vfs); 2313 2314 return 0; 2315} 2316 2317int 2318vfs_unregister(vfs) 2319 struct vfsconf *vfs; 2320{ 2321 struct vfsconf *vfsp; 2322 struct vfsconf **vfspp; 2323 int maxtypenum; 2324 2325 /* Find our vfsconf struct */ 2326 for (vfspp = &vfsconf, vfsp = vfsconf; vfsp; 2327 vfspp = &vfsp->vfc_next, vfsp = vfsp->vfc_next) { 2328 if (strcmp(vfsp->vfc_name, vfs->vfc_name) == 0) 2329 break; 2330 } 2331 2332 if (!vfsp) /* Not found */ 2333 return (ENOENT); 2334 2335 if (vfsp->vfc_refcount) /* In use */ 2336 return (EBUSY); 2337 2338 /* Remove from list and free */ 2339 *vfspp = vfsp->vfc_next; 2340 2341 maxtypenum = 0; 2342 2343 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2344 if (vfsp->vfc_typenum > maxtypenum) 2345 maxtypenum = vfsp->vfc_typenum; 2346 2347 maxvfsconf = maxtypenum; 2348 return 0; 2349} 2350 2351/* 2352 * Check if vnode represents a disk device 2353 */ 2354int 2355vn_isdisk(vp, errp) 2356 struct vnode *vp; 2357 int *errp; 2358{ 2359 if (vp->v_type != VBLK && vp->v_type != VCHR) 2360 return (0); 2361 2362 return (1); 2363} 2364