vfs_export.c revision 29358
1/* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 39 * $Id: vfs_subr.c,v 1.102 1997/09/13 15:02:28 peter Exp $ 40 */ 41 42/* 43 * External virtual filesystem routines 44 */ 45#include "opt_ddb.h" 46#include "opt_devfs.h" 47 48#include <sys/param.h> 49#include <sys/systm.h> 50#include <sys/kernel.h> 51#include <sys/proc.h> 52#include <sys/mount.h> 53#include <sys/vnode.h> 54#include <sys/stat.h> 55#include <sys/buf.h> 56#include <sys/malloc.h> 57#include <sys/poll.h> 58#include <sys/domain.h> 59#include <sys/dirent.h> 60 61#include <machine/limits.h> 62 63#include <vm/vm.h> 64#include <vm/vm_object.h> 65#include <vm/vm_extern.h> 66#include <vm/vnode_pager.h> 67#include <sys/sysctl.h> 68 69#include <miscfs/specfs/specdev.h> 70 71#ifdef DDB 72extern void printlockedvnodes __P((void)); 73#endif 74static void vclean __P((struct vnode *vp, int flags, struct proc *p)); 75static void vgonel __P((struct vnode *vp, struct proc *p)); 76unsigned long numvnodes; 77SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 78static void vputrele __P((struct vnode *vp, int put)); 79 80enum vtype iftovt_tab[16] = { 81 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 82 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 83}; 84int vttoif_tab[9] = { 85 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 86 S_IFSOCK, S_IFIFO, S_IFMT, 87}; 88 89/* 90 * Insq/Remq for the vnode usage lists. 91 */ 92#define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 93#define bufremvn(bp) { \ 94 LIST_REMOVE(bp, b_vnbufs); \ 95 (bp)->b_vnbufs.le_next = NOLIST; \ 96} 97TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 98static u_long freevnodes = 0; 99SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 100 101struct mntlist mountlist; /* mounted filesystem list */ 102struct simplelock mountlist_slock; 103static struct simplelock mntid_slock; 104struct simplelock mntvnode_slock; 105struct simplelock vnode_free_list_slock; 106static struct simplelock spechash_slock; 107struct nfs_public nfs_pub; /* publicly exported FS */ 108 109int desiredvnodes; 110SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, ""); 111 112static void vfs_free_addrlist __P((struct netexport *nep)); 113static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 114static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 115 struct export_args *argp)); 116 117/* 118 * Initialize the vnode management data structures. 119 */ 120void 121vntblinit() 122{ 123 124 desiredvnodes = maxproc + vm_object_cache_max; 125 simple_lock_init(&mntvnode_slock); 126 simple_lock_init(&mntid_slock); 127 simple_lock_init(&spechash_slock); 128 TAILQ_INIT(&vnode_free_list); 129 simple_lock_init(&vnode_free_list_slock); 130 CIRCLEQ_INIT(&mountlist); 131} 132 133/* 134 * Mark a mount point as busy. Used to synchronize access and to delay 135 * unmounting. Interlock is not released on failure. 136 */ 137int 138vfs_busy(mp, flags, interlkp, p) 139 struct mount *mp; 140 int flags; 141 struct simplelock *interlkp; 142 struct proc *p; 143{ 144 int lkflags; 145 146 if (mp->mnt_flag & MNT_UNMOUNT) { 147 if (flags & LK_NOWAIT) 148 return (ENOENT); 149 mp->mnt_flag |= MNT_MWAIT; 150 if (interlkp) { 151 simple_unlock(interlkp); 152 } 153 /* 154 * Since all busy locks are shared except the exclusive 155 * lock granted when unmounting, the only place that a 156 * wakeup needs to be done is at the release of the 157 * exclusive lock at the end of dounmount. 158 */ 159 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 160 if (interlkp) { 161 simple_lock(interlkp); 162 } 163 return (ENOENT); 164 } 165 lkflags = LK_SHARED; 166 if (interlkp) 167 lkflags |= LK_INTERLOCK; 168 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 169 panic("vfs_busy: unexpected lock failure"); 170 return (0); 171} 172 173/* 174 * Free a busy filesystem. 175 */ 176void 177vfs_unbusy(mp, p) 178 struct mount *mp; 179 struct proc *p; 180{ 181 182 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 183} 184 185/* 186 * Lookup a filesystem type, and if found allocate and initialize 187 * a mount structure for it. 188 * 189 * Devname is usually updated by mount(8) after booting. 190 */ 191int 192vfs_rootmountalloc(fstypename, devname, mpp) 193 char *fstypename; 194 char *devname; 195 struct mount **mpp; 196{ 197 struct proc *p = curproc; /* XXX */ 198 struct vfsconf *vfsp; 199 struct mount *mp; 200 201 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 202 if (!strcmp(vfsp->vfc_name, fstypename)) 203 break; 204 if (vfsp == NULL) 205 return (ENODEV); 206 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 207 bzero((char *)mp, (u_long)sizeof(struct mount)); 208 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 209 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 210 LIST_INIT(&mp->mnt_vnodelist); 211 mp->mnt_vfc = vfsp; 212 mp->mnt_op = vfsp->vfc_vfsops; 213 mp->mnt_flag = MNT_RDONLY; 214 mp->mnt_vnodecovered = NULLVP; 215 vfsp->vfc_refcount++; 216 mp->mnt_stat.f_type = vfsp->vfc_typenum; 217 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 218 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 219 mp->mnt_stat.f_mntonname[0] = '/'; 220 mp->mnt_stat.f_mntonname[1] = 0; 221 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 222 *mpp = mp; 223 return (0); 224} 225 226/* 227 * Find an appropriate filesystem to use for the root. If a filesystem 228 * has not been preselected, walk through the list of known filesystems 229 * trying those that have mountroot routines, and try them until one 230 * works or we have tried them all. 231 */ 232#ifdef notdef /* XXX JH */ 233int 234lite2_vfs_mountroot(void) 235{ 236 struct vfsconf *vfsp; 237 extern int (*lite2_mountroot)(void); 238 int error; 239 240 if (lite2_mountroot != NULL) 241 return ((*lite2_mountroot)()); 242 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 243 if (vfsp->vfc_mountroot == NULL) 244 continue; 245 if ((error = (*vfsp->vfc_mountroot)()) == 0) 246 return (0); 247 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 248 } 249 return (ENODEV); 250} 251#endif 252 253/* 254 * Lookup a mount point by filesystem identifier. 255 */ 256struct mount * 257vfs_getvfs(fsid) 258 fsid_t *fsid; 259{ 260 register struct mount *mp; 261 262 simple_lock(&mountlist_slock); 263 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 264 mp = mp->mnt_list.cqe_next) { 265 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 266 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 267 simple_unlock(&mountlist_slock); 268 return (mp); 269 } 270 } 271 simple_unlock(&mountlist_slock); 272 return ((struct mount *) 0); 273} 274 275/* 276 * Get a new unique fsid 277 */ 278void 279vfs_getnewfsid(mp) 280 struct mount *mp; 281{ 282 static u_short xxxfs_mntid; 283 284 fsid_t tfsid; 285 int mtype; 286 287 simple_lock(&mntid_slock); 288 mtype = mp->mnt_vfc->vfc_typenum; 289 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 290 mp->mnt_stat.f_fsid.val[1] = mtype; 291 if (xxxfs_mntid == 0) 292 ++xxxfs_mntid; 293 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 294 tfsid.val[1] = mtype; 295 if (mountlist.cqh_first != (void *)&mountlist) { 296 while (vfs_getvfs(&tfsid)) { 297 tfsid.val[0]++; 298 xxxfs_mntid++; 299 } 300 } 301 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 302 simple_unlock(&mntid_slock); 303} 304 305/* 306 * Set vnode attributes to VNOVAL 307 */ 308void 309vattr_null(vap) 310 register struct vattr *vap; 311{ 312 313 vap->va_type = VNON; 314 vap->va_size = VNOVAL; 315 vap->va_bytes = VNOVAL; 316 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 317 vap->va_fsid = vap->va_fileid = 318 vap->va_blocksize = vap->va_rdev = 319 vap->va_atime.tv_sec = vap->va_atime.tv_nsec = 320 vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec = 321 vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec = 322 vap->va_flags = vap->va_gen = VNOVAL; 323 vap->va_vaflags = 0; 324} 325 326/* 327 * Routines having to do with the management of the vnode table. 328 */ 329extern vop_t **dead_vnodeop_p; 330 331/* 332 * Return the next vnode from the free list. 333 */ 334int 335getnewvnode(tag, mp, vops, vpp) 336 enum vtagtype tag; 337 struct mount *mp; 338 vop_t **vops; 339 struct vnode **vpp; 340{ 341 struct proc *p = curproc; /* XXX */ 342 struct vnode *vp; 343 344 /* 345 * We take the least recently used vnode from the freelist 346 * if we can get it and it has no cached pages, and no 347 * namecache entries are relative to it. 348 * Otherwise we allocate a new vnode 349 */ 350 351 simple_lock(&vnode_free_list_slock); 352 353 if (freevnodes >= desiredvnodes) { 354 TAILQ_FOREACH(vp, &vnode_free_list, v_freelist) { 355 if (!simple_lock_try(&vp->v_interlock)) 356 continue; 357 if (vp->v_usecount) 358 panic("free vnode isn't"); 359 360 if (vp->v_object && vp->v_object->resident_page_count) { 361 /* Don't recycle if it's caching some pages */ 362 simple_unlock(&vp->v_interlock); 363 continue; 364 } else if (LIST_FIRST(&vp->v_cache_src)) { 365 /* Don't recycle if active in the namecache */ 366 simple_unlock(&vp->v_interlock); 367 continue; 368 } else { 369 break; 370 } 371 } 372 } else { 373 vp = NULL; 374 } 375 376 if (vp) { 377 vp->v_flag |= VDOOMED; 378 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 379 freevnodes--; 380 simple_unlock(&vnode_free_list_slock); 381 cache_purge(vp); 382 vp->v_lease = NULL; 383 if (vp->v_type != VBAD) 384 vgonel(vp, p); 385 else { 386 simple_unlock(&vp->v_interlock); 387 } 388 389#ifdef DIAGNOSTIC 390 { 391 int s; 392 393 if (vp->v_data) 394 panic("cleaned vnode isn't"); 395 s = splbio(); 396 if (vp->v_numoutput) 397 panic("Clean vnode has pending I/O's"); 398 splx(s); 399 } 400#endif 401 vp->v_flag = 0; 402 vp->v_lastr = 0; 403 vp->v_lastw = 0; 404 vp->v_lasta = 0; 405 vp->v_cstart = 0; 406 vp->v_clen = 0; 407 vp->v_socket = 0; 408 vp->v_writecount = 0; /* XXX */ 409 } else { 410 simple_unlock(&vnode_free_list_slock); 411 vp = (struct vnode *) malloc((u_long) sizeof *vp, 412 M_VNODE, M_WAITOK); 413 bzero((char *) vp, sizeof *vp); 414 vp->v_dd = vp; 415 cache_purge(vp); 416 LIST_INIT(&vp->v_cache_src); 417 TAILQ_INIT(&vp->v_cache_dst); 418 numvnodes++; 419 } 420 421 vp->v_type = VNON; 422 vp->v_tag = tag; 423 vp->v_op = vops; 424 insmntque(vp, mp); 425 *vpp = vp; 426 vp->v_usecount = 1; 427 vp->v_data = 0; 428 return (0); 429} 430 431/* 432 * Move a vnode from one mount queue to another. 433 */ 434void 435insmntque(vp, mp) 436 register struct vnode *vp; 437 register struct mount *mp; 438{ 439 440 simple_lock(&mntvnode_slock); 441 /* 442 * Delete from old mount point vnode list, if on one. 443 */ 444 if (vp->v_mount != NULL) 445 LIST_REMOVE(vp, v_mntvnodes); 446 /* 447 * Insert into list of vnodes for the new mount point, if available. 448 */ 449 if ((vp->v_mount = mp) == NULL) { 450 simple_unlock(&mntvnode_slock); 451 return; 452 } 453 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 454 simple_unlock(&mntvnode_slock); 455} 456 457/* 458 * Update outstanding I/O count and do wakeup if requested. 459 */ 460void 461vwakeup(bp) 462 register struct buf *bp; 463{ 464 register struct vnode *vp; 465 466 bp->b_flags &= ~B_WRITEINPROG; 467 if ((vp = bp->b_vp)) { 468 vp->v_numoutput--; 469 if (vp->v_numoutput < 0) 470 panic("vwakeup: neg numoutput"); 471 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 472 vp->v_flag &= ~VBWAIT; 473 wakeup((caddr_t) &vp->v_numoutput); 474 } 475 } 476} 477 478/* 479 * Flush out and invalidate all buffers associated with a vnode. 480 * Called with the underlying object locked. 481 */ 482int 483vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 484 register struct vnode *vp; 485 int flags; 486 struct ucred *cred; 487 struct proc *p; 488 int slpflag, slptimeo; 489{ 490 register struct buf *bp; 491 struct buf *nbp, *blist; 492 int s, error; 493 vm_object_t object; 494 495 if (flags & V_SAVE) { 496 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p))) 497 return (error); 498 if (vp->v_dirtyblkhd.lh_first != NULL) 499 panic("vinvalbuf: dirty bufs"); 500 } 501 502 s = splbio(); 503 for (;;) { 504 if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA)) 505 while (blist && blist->b_lblkno < 0) 506 blist = blist->b_vnbufs.le_next; 507 if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 508 (flags & V_SAVEMETA)) 509 while (blist && blist->b_lblkno < 0) 510 blist = blist->b_vnbufs.le_next; 511 if (!blist) 512 break; 513 514 for (bp = blist; bp; bp = nbp) { 515 nbp = bp->b_vnbufs.le_next; 516 if ((flags & V_SAVEMETA) && bp->b_lblkno < 0) 517 continue; 518 if (bp->b_flags & B_BUSY) { 519 bp->b_flags |= B_WANTED; 520 error = tsleep((caddr_t) bp, 521 slpflag | (PRIBIO + 1), "vinvalbuf", 522 slptimeo); 523 if (error) { 524 splx(s); 525 return (error); 526 } 527 break; 528 } 529 bremfree(bp); 530 bp->b_flags |= B_BUSY; 531 /* 532 * XXX Since there are no node locks for NFS, I 533 * believe there is a slight chance that a delayed 534 * write will occur while sleeping just above, so 535 * check for it. 536 */ 537 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 538 (void) VOP_BWRITE(bp); 539 break; 540 } 541 bp->b_flags |= (B_INVAL|B_NOCACHE|B_RELBUF); 542 brelse(bp); 543 } 544 } 545 546 while (vp->v_numoutput > 0) { 547 vp->v_flag |= VBWAIT; 548 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 549 } 550 551 splx(s); 552 553 /* 554 * Destroy the copy in the VM cache, too. 555 */ 556 object = vp->v_object; 557 if (object != NULL) { 558 vm_object_page_remove(object, 0, object->size, 559 (flags & V_SAVE) ? TRUE : FALSE); 560 } 561 if (!(flags & V_SAVEMETA) && 562 (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) 563 panic("vinvalbuf: flush failed"); 564 return (0); 565} 566 567/* 568 * Associate a buffer with a vnode. 569 */ 570void 571bgetvp(vp, bp) 572 register struct vnode *vp; 573 register struct buf *bp; 574{ 575 int s; 576 577 if (bp->b_vp) 578 panic("bgetvp: not free"); 579 vhold(vp); 580 bp->b_vp = vp; 581 if (vp->v_type == VBLK || vp->v_type == VCHR) 582 bp->b_dev = vp->v_rdev; 583 else 584 bp->b_dev = NODEV; 585 /* 586 * Insert onto list for new vnode. 587 */ 588 s = splbio(); 589 bufinsvn(bp, &vp->v_cleanblkhd); 590 splx(s); 591} 592 593/* 594 * Disassociate a buffer from a vnode. 595 */ 596void 597brelvp(bp) 598 register struct buf *bp; 599{ 600 struct vnode *vp; 601 int s; 602 603 if (bp->b_vp == (struct vnode *) 0) 604 panic("brelvp: NULL"); 605 /* 606 * Delete from old vnode list, if on one. 607 */ 608 s = splbio(); 609 if (bp->b_vnbufs.le_next != NOLIST) 610 bufremvn(bp); 611 splx(s); 612 613 vp = bp->b_vp; 614 bp->b_vp = (struct vnode *) 0; 615 vdrop(vp); 616} 617 618/* 619 * Associate a p-buffer with a vnode. 620 */ 621void 622pbgetvp(vp, bp) 623 register struct vnode *vp; 624 register struct buf *bp; 625{ 626#if defined(DIAGNOSTIC) 627 if (bp->b_vp) 628 panic("pbgetvp: not free"); 629#endif 630 bp->b_vp = vp; 631 if (vp->v_type == VBLK || vp->v_type == VCHR) 632 bp->b_dev = vp->v_rdev; 633 else 634 bp->b_dev = NODEV; 635} 636 637/* 638 * Disassociate a p-buffer from a vnode. 639 */ 640void 641pbrelvp(bp) 642 register struct buf *bp; 643{ 644 struct vnode *vp; 645 646#if defined(DIAGNOSTIC) 647 if (bp->b_vp == (struct vnode *) 0) 648 panic("pbrelvp: NULL"); 649#endif 650 651 bp->b_vp = (struct vnode *) 0; 652} 653 654/* 655 * Reassign a buffer from one vnode to another. 656 * Used to assign file specific control information 657 * (indirect blocks) to the vnode to which they belong. 658 */ 659void 660reassignbuf(bp, newvp) 661 register struct buf *bp; 662 register struct vnode *newvp; 663{ 664 int s; 665 666 if (newvp == NULL) { 667 printf("reassignbuf: NULL"); 668 return; 669 } 670 671 s = splbio(); 672 /* 673 * Delete from old vnode list, if on one. 674 */ 675 if (bp->b_vnbufs.le_next != NOLIST) { 676 bufremvn(bp); 677 vdrop(bp->b_vp); 678 } 679 /* 680 * If dirty, put on list of dirty buffers; otherwise insert onto list 681 * of clean buffers. 682 */ 683 if (bp->b_flags & B_DELWRI) { 684 struct buf *tbp; 685 686 tbp = newvp->v_dirtyblkhd.lh_first; 687 if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) { 688 bufinsvn(bp, &newvp->v_dirtyblkhd); 689 } else { 690 while (tbp->b_vnbufs.le_next && 691 (tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) { 692 tbp = tbp->b_vnbufs.le_next; 693 } 694 LIST_INSERT_AFTER(tbp, bp, b_vnbufs); 695 } 696 } else { 697 bufinsvn(bp, &newvp->v_cleanblkhd); 698 } 699 bp->b_vp = newvp; 700 vhold(bp->b_vp); 701 splx(s); 702} 703 704#ifndef DEVFS_ROOT 705/* 706 * Create a vnode for a block device. 707 * Used for mounting the root file system. 708 */ 709int 710bdevvp(dev, vpp) 711 dev_t dev; 712 struct vnode **vpp; 713{ 714 register struct vnode *vp; 715 struct vnode *nvp; 716 int error; 717 718 if (dev == NODEV) 719 return (0); 720 error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp); 721 if (error) { 722 *vpp = 0; 723 return (error); 724 } 725 vp = nvp; 726 vp->v_type = VBLK; 727 if ((nvp = checkalias(vp, dev, (struct mount *) 0))) { 728 vput(vp); 729 vp = nvp; 730 } 731 *vpp = vp; 732 return (0); 733} 734#endif /* !DEVFS_ROOT */ 735 736/* 737 * Check to see if the new vnode represents a special device 738 * for which we already have a vnode (either because of 739 * bdevvp() or because of a different vnode representing 740 * the same block device). If such an alias exists, deallocate 741 * the existing contents and return the aliased vnode. The 742 * caller is responsible for filling it with its new contents. 743 */ 744struct vnode * 745checkalias(nvp, nvp_rdev, mp) 746 register struct vnode *nvp; 747 dev_t nvp_rdev; 748 struct mount *mp; 749{ 750 struct proc *p = curproc; /* XXX */ 751 struct vnode *vp; 752 struct vnode **vpp; 753 754 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 755 return (NULLVP); 756 757 vpp = &speclisth[SPECHASH(nvp_rdev)]; 758loop: 759 simple_lock(&spechash_slock); 760 for (vp = *vpp; vp; vp = vp->v_specnext) { 761 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 762 continue; 763 /* 764 * Alias, but not in use, so flush it out. 765 */ 766 simple_lock(&vp->v_interlock); 767 if (vp->v_usecount == 0) { 768 simple_unlock(&spechash_slock); 769 vgonel(vp, p); 770 goto loop; 771 } 772 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { 773 simple_unlock(&spechash_slock); 774 goto loop; 775 } 776 break; 777 } 778 if (vp == NULL || vp->v_tag != VT_NON) { 779 MALLOC(nvp->v_specinfo, struct specinfo *, 780 sizeof(struct specinfo), M_VNODE, M_WAITOK); 781 nvp->v_rdev = nvp_rdev; 782 nvp->v_hashchain = vpp; 783 nvp->v_specnext = *vpp; 784 nvp->v_specflags = 0; 785 simple_unlock(&spechash_slock); 786 *vpp = nvp; 787 if (vp != NULLVP) { 788 nvp->v_flag |= VALIASED; 789 vp->v_flag |= VALIASED; 790 vput(vp); 791 } 792 return (NULLVP); 793 } 794 simple_unlock(&spechash_slock); 795 VOP_UNLOCK(vp, 0, p); 796 simple_lock(&vp->v_interlock); 797 vclean(vp, 0, p); 798 vp->v_op = nvp->v_op; 799 vp->v_tag = nvp->v_tag; 800 nvp->v_type = VNON; 801 insmntque(vp, mp); 802 return (vp); 803} 804 805/* 806 * Grab a particular vnode from the free list, increment its 807 * reference count and lock it. The vnode lock bit is set the 808 * vnode is being eliminated in vgone. The process is awakened 809 * when the transition is completed, and an error returned to 810 * indicate that the vnode is no longer usable (possibly having 811 * been changed to a new file system type). 812 */ 813int 814vget(vp, flags, p) 815 register struct vnode *vp; 816 int flags; 817 struct proc *p; 818{ 819 int error; 820 821 /* 822 * If the vnode is in the process of being cleaned out for 823 * another use, we wait for the cleaning to finish and then 824 * return failure. Cleaning is determined by checking that 825 * the VXLOCK flag is set. 826 */ 827 if ((flags & LK_INTERLOCK) == 0) { 828 simple_lock(&vp->v_interlock); 829 } 830 if (vp->v_flag & VXLOCK) { 831 vp->v_flag |= VXWANT; 832 simple_unlock(&vp->v_interlock); 833 tsleep((caddr_t)vp, PINOD, "vget", 0); 834 return (ENOENT); 835 } 836 vp->v_usecount++; 837 if (VSHOULDBUSY(vp)) 838 vbusy(vp); 839 /* 840 * Create the VM object, if needed 841 */ 842 if ((vp->v_type == VREG) && 843 ((vp->v_object == NULL) || 844 (vp->v_object->flags & OBJ_VFS_REF) == 0 || 845 (vp->v_object->flags & OBJ_DEAD))) { 846 /* 847 * XXX vfs_object_create probably needs the interlock. 848 */ 849 simple_unlock(&vp->v_interlock); 850 vfs_object_create(vp, curproc, curproc->p_ucred, 0); 851 simple_lock(&vp->v_interlock); 852 } 853 if (flags & LK_TYPE_MASK) { 854 if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) 855 vrele(vp); 856 return (error); 857 } 858 simple_unlock(&vp->v_interlock); 859 return (0); 860} 861 862/* 863 * Stubs to use when there is no locking to be done on the underlying object. 864 * A minimal shared lock is necessary to ensure that the underlying object 865 * is not revoked while an operation is in progress. So, an active shared 866 * count is maintained in an auxillary vnode lock structure. 867 */ 868int 869vop_sharedlock(ap) 870 struct vop_lock_args /* { 871 struct vnode *a_vp; 872 int a_flags; 873 struct proc *a_p; 874 } */ *ap; 875{ 876 /* 877 * This code cannot be used until all the non-locking filesystems 878 * (notably NFS) are converted to properly lock and release nodes. 879 * Also, certain vnode operations change the locking state within 880 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 881 * and symlink). Ideally these operations should not change the 882 * lock state, but should be changed to let the caller of the 883 * function unlock them. Otherwise all intermediate vnode layers 884 * (such as union, umapfs, etc) must catch these functions to do 885 * the necessary locking at their layer. Note that the inactive 886 * and lookup operations also change their lock state, but this 887 * cannot be avoided, so these two operations will always need 888 * to be handled in intermediate layers. 889 */ 890 struct vnode *vp = ap->a_vp; 891 int vnflags, flags = ap->a_flags; 892 893 if (vp->v_vnlock == NULL) { 894 if ((flags & LK_TYPE_MASK) == LK_DRAIN) 895 return (0); 896 MALLOC(vp->v_vnlock, struct lock *, sizeof(struct lock), 897 M_VNODE, M_WAITOK); 898 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 899 } 900 switch (flags & LK_TYPE_MASK) { 901 case LK_DRAIN: 902 vnflags = LK_DRAIN; 903 break; 904 case LK_EXCLUSIVE: 905#ifdef DEBUG_VFS_LOCKS 906 /* 907 * Normally, we use shared locks here, but that confuses 908 * the locking assertions. 909 */ 910 vnflags = LK_EXCLUSIVE; 911 break; 912#endif 913 case LK_SHARED: 914 vnflags = LK_SHARED; 915 break; 916 case LK_UPGRADE: 917 case LK_EXCLUPGRADE: 918 case LK_DOWNGRADE: 919 return (0); 920 case LK_RELEASE: 921 default: 922 panic("vop_sharedlock: bad operation %d", flags & LK_TYPE_MASK); 923 } 924 if (flags & LK_INTERLOCK) 925 vnflags |= LK_INTERLOCK; 926 return(lockmgr(vp->v_vnlock, vnflags, &vp->v_interlock, ap->a_p)); 927} 928 929/* 930 * Stubs to use when there is no locking to be done on the underlying object. 931 * A minimal shared lock is necessary to ensure that the underlying object 932 * is not revoked while an operation is in progress. So, an active shared 933 * count is maintained in an auxillary vnode lock structure. 934 */ 935int 936vop_nolock(ap) 937 struct vop_lock_args /* { 938 struct vnode *a_vp; 939 int a_flags; 940 struct proc *a_p; 941 } */ *ap; 942{ 943#ifdef notyet 944 /* 945 * This code cannot be used until all the non-locking filesystems 946 * (notably NFS) are converted to properly lock and release nodes. 947 * Also, certain vnode operations change the locking state within 948 * the operation (create, mknod, remove, link, rename, mkdir, rmdir, 949 * and symlink). Ideally these operations should not change the 950 * lock state, but should be changed to let the caller of the 951 * function unlock them. Otherwise all intermediate vnode layers 952 * (such as union, umapfs, etc) must catch these functions to do 953 * the necessary locking at their layer. Note that the inactive 954 * and lookup operations also change their lock state, but this 955 * cannot be avoided, so these two operations will always need 956 * to be handled in intermediate layers. 957 */ 958 struct vnode *vp = ap->a_vp; 959 int vnflags, flags = ap->a_flags; 960 961 if (vp->v_vnlock == NULL) { 962 if ((flags & LK_TYPE_MASK) == LK_DRAIN) 963 return (0); 964 MALLOC(vp->v_vnlock, struct lock *, sizeof(struct lock), 965 M_VNODE, M_WAITOK); 966 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 967 } 968 switch (flags & LK_TYPE_MASK) { 969 case LK_DRAIN: 970 vnflags = LK_DRAIN; 971 break; 972 case LK_EXCLUSIVE: 973 case LK_SHARED: 974 vnflags = LK_SHARED; 975 break; 976 case LK_UPGRADE: 977 case LK_EXCLUPGRADE: 978 case LK_DOWNGRADE: 979 return (0); 980 case LK_RELEASE: 981 default: 982 panic("vop_nolock: bad operation %d", flags & LK_TYPE_MASK); 983 } 984 if (flags & LK_INTERLOCK) 985 vnflags |= LK_INTERLOCK; 986 return(lockmgr(vp->v_vnlock, vnflags, &vp->v_interlock, ap->a_p)); 987#else /* for now */ 988 /* 989 * Since we are not using the lock manager, we must clear 990 * the interlock here. 991 */ 992 if (ap->a_flags & LK_INTERLOCK) { 993 simple_unlock(&ap->a_vp->v_interlock); 994 } 995 return (0); 996#endif 997} 998 999/* 1000 * Do the inverse of vop_nolock, handling the interlock in a compatible way. 1001 */ 1002int 1003vop_nounlock(ap) 1004 struct vop_unlock_args /* { 1005 struct vnode *a_vp; 1006 int a_flags; 1007 struct proc *a_p; 1008 } */ *ap; 1009{ 1010 struct vnode *vp = ap->a_vp; 1011 1012 if (vp->v_vnlock == NULL) { 1013 if (ap->a_flags & LK_INTERLOCK) 1014 simple_unlock(&ap->a_vp->v_interlock); 1015 return (0); 1016 } 1017 return (lockmgr(vp->v_vnlock, LK_RELEASE | ap->a_flags, 1018 &ap->a_vp->v_interlock, ap->a_p)); 1019} 1020 1021/* 1022 * Return whether or not the node is in use. 1023 */ 1024int 1025vop_noislocked(ap) 1026 struct vop_islocked_args /* { 1027 struct vnode *a_vp; 1028 } */ *ap; 1029{ 1030 struct vnode *vp = ap->a_vp; 1031 1032 if (vp->v_vnlock == NULL) 1033 return (0); 1034 return (lockstatus(vp->v_vnlock)); 1035} 1036 1037/* #ifdef DIAGNOSTIC */ 1038/* 1039 * Vnode reference, just increment the count 1040 */ 1041void 1042vref(vp) 1043 struct vnode *vp; 1044{ 1045 simple_lock(&vp->v_interlock); 1046 if (vp->v_usecount <= 0) 1047 panic("vref used where vget required"); 1048 1049 vp->v_usecount++; 1050 1051 if ((vp->v_type == VREG) && 1052 ((vp->v_object == NULL) || 1053 ((vp->v_object->flags & OBJ_VFS_REF) == 0) || 1054 (vp->v_object->flags & OBJ_DEAD))) { 1055 /* 1056 * We need to lock to VP during the time that 1057 * the object is created. This is necessary to 1058 * keep the system from re-entrantly doing it 1059 * multiple times. 1060 * XXX vfs_object_create probably needs the interlock? 1061 */ 1062 simple_unlock(&vp->v_interlock); 1063 vfs_object_create(vp, curproc, curproc->p_ucred, 0); 1064 return; 1065 } 1066 simple_unlock(&vp->v_interlock); 1067} 1068 1069/* 1070 * Vnode put/release. 1071 * If count drops to zero, call inactive routine and return to freelist. 1072 */ 1073static void 1074vputrele(vp, put) 1075 struct vnode *vp; 1076 int put; 1077{ 1078 struct proc *p = curproc; /* XXX */ 1079 1080#ifdef DIAGNOSTIC 1081 if (vp == NULL) 1082 panic("vputrele: null vp"); 1083#endif 1084 simple_lock(&vp->v_interlock); 1085 1086 if ((vp->v_usecount == 2) && 1087 vp->v_object && 1088 (vp->v_object->flags & OBJ_VFS_REF)) { 1089 vp->v_usecount--; 1090 vp->v_object->flags &= ~OBJ_VFS_REF; 1091 if (put) { 1092 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1093 } else { 1094 simple_unlock(&vp->v_interlock); 1095 } 1096 vm_object_deallocate(vp->v_object); 1097 return; 1098 } 1099 1100 if (vp->v_usecount > 1) { 1101 vp->v_usecount--; 1102 if (put) { 1103 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1104 } else { 1105 simple_unlock(&vp->v_interlock); 1106 } 1107 return; 1108 } 1109 1110 if (vp->v_usecount < 1) { 1111#ifdef DIAGNOSTIC 1112 vprint("vputrele: negative ref count", vp); 1113#endif 1114 panic("vputrele: negative ref cnt"); 1115 } 1116 1117 vp->v_usecount--; 1118 if (VSHOULDFREE(vp)) 1119 vfree(vp); 1120 /* 1121 * If we are doing a vput, the node is already locked, and we must 1122 * call VOP_INACTIVE with the node locked. So, in the case of 1123 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1124 */ 1125 if (put) { 1126 simple_unlock(&vp->v_interlock); 1127 VOP_INACTIVE(vp, p); 1128 } else if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 1129 VOP_INACTIVE(vp, p); 1130 } 1131} 1132 1133/* 1134 * vput(), just unlock and vrele() 1135 */ 1136void 1137vput(vp) 1138 struct vnode *vp; 1139{ 1140 vputrele(vp, 1); 1141} 1142 1143void 1144vrele(vp) 1145 struct vnode *vp; 1146{ 1147 vputrele(vp, 0); 1148} 1149 1150/* 1151 * Somebody doesn't want the vnode recycled. 1152 */ 1153void 1154vhold(vp) 1155 register struct vnode *vp; 1156{ 1157 1158 simple_lock(&vp->v_interlock); 1159 vp->v_holdcnt++; 1160 if (VSHOULDBUSY(vp)) 1161 vbusy(vp); 1162 simple_unlock(&vp->v_interlock); 1163} 1164 1165/* 1166 * One less who cares about this vnode. 1167 */ 1168void 1169vdrop(vp) 1170 register struct vnode *vp; 1171{ 1172 1173 simple_lock(&vp->v_interlock); 1174 if (vp->v_holdcnt <= 0) 1175 panic("holdrele: holdcnt"); 1176 vp->v_holdcnt--; 1177 if (VSHOULDFREE(vp)) 1178 vfree(vp); 1179 simple_unlock(&vp->v_interlock); 1180} 1181 1182/* 1183 * Remove any vnodes in the vnode table belonging to mount point mp. 1184 * 1185 * If MNT_NOFORCE is specified, there should not be any active ones, 1186 * return error if any are found (nb: this is a user error, not a 1187 * system error). If MNT_FORCE is specified, detach any active vnodes 1188 * that are found. 1189 */ 1190#ifdef DIAGNOSTIC 1191static int busyprt = 0; /* print out busy vnodes */ 1192SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1193#endif 1194 1195int 1196vflush(mp, skipvp, flags) 1197 struct mount *mp; 1198 struct vnode *skipvp; 1199 int flags; 1200{ 1201 struct proc *p = curproc; /* XXX */ 1202 struct vnode *vp, *nvp; 1203 int busy = 0; 1204 1205 simple_lock(&mntvnode_slock); 1206loop: 1207 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 1208 /* 1209 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1210 * Start over if it has (it won't be on the list anymore). 1211 */ 1212 if (vp->v_mount != mp) 1213 goto loop; 1214 nvp = vp->v_mntvnodes.le_next; 1215 /* 1216 * Skip over a selected vnode. 1217 */ 1218 if (vp == skipvp) 1219 continue; 1220 1221 simple_lock(&vp->v_interlock); 1222 /* 1223 * Skip over a vnodes marked VSYSTEM. 1224 */ 1225 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1226 simple_unlock(&vp->v_interlock); 1227 continue; 1228 } 1229 /* 1230 * If WRITECLOSE is set, only flush out regular file vnodes 1231 * open for writing. 1232 */ 1233 if ((flags & WRITECLOSE) && 1234 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1235 simple_unlock(&vp->v_interlock); 1236 continue; 1237 } 1238 1239 /* 1240 * With v_usecount == 0, all we need to do is clear out the 1241 * vnode data structures and we are done. 1242 */ 1243 if (vp->v_usecount == 0) { 1244 simple_unlock(&mntvnode_slock); 1245 vgonel(vp, p); 1246 simple_lock(&mntvnode_slock); 1247 continue; 1248 } 1249 1250 /* 1251 * If FORCECLOSE is set, forcibly close the vnode. For block 1252 * or character devices, revert to an anonymous device. For 1253 * all other files, just kill them. 1254 */ 1255 if (flags & FORCECLOSE) { 1256 simple_unlock(&mntvnode_slock); 1257 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1258 vgonel(vp, p); 1259 } else { 1260 vclean(vp, 0, p); 1261 vp->v_op = spec_vnodeop_p; 1262 insmntque(vp, (struct mount *) 0); 1263 } 1264 simple_lock(&mntvnode_slock); 1265 continue; 1266 } 1267#ifdef DIAGNOSTIC 1268 if (busyprt) 1269 vprint("vflush: busy vnode", vp); 1270#endif 1271 simple_unlock(&vp->v_interlock); 1272 busy++; 1273 } 1274 simple_unlock(&mntvnode_slock); 1275 if (busy) 1276 return (EBUSY); 1277 return (0); 1278} 1279 1280/* 1281 * Disassociate the underlying file system from a vnode. 1282 */ 1283static void 1284vclean(struct vnode *vp, int flags, struct proc *p) 1285{ 1286 int active, irefed; 1287 vm_object_t object; 1288 1289 /* 1290 * Check to see if the vnode is in use. If so we have to reference it 1291 * before we clean it out so that its count cannot fall to zero and 1292 * generate a race against ourselves to recycle it. 1293 */ 1294 if ((active = vp->v_usecount)) 1295 vp->v_usecount++; 1296 /* 1297 * Prevent the vnode from being recycled or brought into use while we 1298 * clean it out. 1299 */ 1300 if (vp->v_flag & VXLOCK) 1301 panic("vclean: deadlock"); 1302 vp->v_flag |= VXLOCK; 1303 /* 1304 * Even if the count is zero, the VOP_INACTIVE routine may still 1305 * have the object locked while it cleans it out. The VOP_LOCK 1306 * ensures that the VOP_INACTIVE routine is done with its work. 1307 * For active vnodes, it ensures that no other activity can 1308 * occur while the underlying object is being cleaned out. 1309 */ 1310 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1311 1312 object = vp->v_object; 1313 irefed = 0; 1314 if (object && ((object->flags & OBJ_DEAD) == 0)) { 1315 if (object->ref_count == 0) { 1316 vm_object_reference(object); 1317 irefed = 1; 1318 } 1319 ++object->ref_count; 1320 pager_cache(object, FALSE); 1321 } 1322 1323 /* 1324 * Clean out any buffers associated with the vnode. 1325 */ 1326 if (flags & DOCLOSE) 1327 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1328 1329 if (irefed) { 1330 vm_object_deallocate(object); 1331 } 1332 1333 /* 1334 * If purging an active vnode, it must be closed and 1335 * deactivated before being reclaimed. Note that the 1336 * VOP_INACTIVE will unlock the vnode. 1337 */ 1338 if (active) { 1339 if (flags & DOCLOSE) 1340 VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); 1341 VOP_INACTIVE(vp, p); 1342 } else { 1343 /* 1344 * Any other processes trying to obtain this lock must first 1345 * wait for VXLOCK to clear, then call the new lock operation. 1346 */ 1347 VOP_UNLOCK(vp, 0, p); 1348 } 1349 /* 1350 * Reclaim the vnode. 1351 */ 1352 if (VOP_RECLAIM(vp, p)) 1353 panic("vclean: cannot reclaim"); 1354 if (active) 1355 vrele(vp); 1356 cache_purge(vp); 1357 if (vp->v_vnlock) { 1358#ifdef DIAGNOSTIC 1359 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0) 1360 vprint("vclean: lock not drained", vp); 1361#endif 1362 FREE(vp->v_vnlock, M_VNODE); 1363 vp->v_vnlock = NULL; 1364 } 1365 1366 /* 1367 * Done with purge, notify sleepers of the grim news. 1368 */ 1369 vp->v_op = dead_vnodeop_p; 1370 vp->v_tag = VT_NON; 1371 vp->v_flag &= ~VXLOCK; 1372 if (vp->v_flag & VXWANT) { 1373 vp->v_flag &= ~VXWANT; 1374 wakeup((caddr_t) vp); 1375 } 1376} 1377 1378/* 1379 * Eliminate all activity associated with the requested vnode 1380 * and with all vnodes aliased to the requested vnode. 1381 */ 1382int 1383vop_revoke(ap) 1384 struct vop_revoke_args /* { 1385 struct vnode *a_vp; 1386 int a_flags; 1387 } */ *ap; 1388{ 1389 struct vnode *vp, *vq; 1390 struct proc *p = curproc; /* XXX */ 1391 1392#ifdef DIAGNOSTIC 1393 if ((ap->a_flags & REVOKEALL) == 0) 1394 panic("vop_revoke"); 1395#endif 1396 1397 vp = ap->a_vp; 1398 simple_lock(&vp->v_interlock); 1399 1400 if (vp->v_flag & VALIASED) { 1401 /* 1402 * If a vgone (or vclean) is already in progress, 1403 * wait until it is done and return. 1404 */ 1405 if (vp->v_flag & VXLOCK) { 1406 vp->v_flag |= VXWANT; 1407 simple_unlock(&vp->v_interlock); 1408 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1409 return (0); 1410 } 1411 /* 1412 * Ensure that vp will not be vgone'd while we 1413 * are eliminating its aliases. 1414 */ 1415 vp->v_flag |= VXLOCK; 1416 simple_unlock(&vp->v_interlock); 1417 while (vp->v_flag & VALIASED) { 1418 simple_lock(&spechash_slock); 1419 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1420 if (vq->v_rdev != vp->v_rdev || 1421 vq->v_type != vp->v_type || vp == vq) 1422 continue; 1423 simple_unlock(&spechash_slock); 1424 vgone(vq); 1425 break; 1426 } 1427 if (vq == NULLVP) { 1428 simple_unlock(&spechash_slock); 1429 } 1430 } 1431 /* 1432 * Remove the lock so that vgone below will 1433 * really eliminate the vnode after which time 1434 * vgone will awaken any sleepers. 1435 */ 1436 simple_lock(&vp->v_interlock); 1437 vp->v_flag &= ~VXLOCK; 1438 } 1439 vgonel(vp, p); 1440 return (0); 1441} 1442 1443/* 1444 * Recycle an unused vnode to the front of the free list. 1445 * Release the passed interlock if the vnode will be recycled. 1446 */ 1447int 1448vrecycle(vp, inter_lkp, p) 1449 struct vnode *vp; 1450 struct simplelock *inter_lkp; 1451 struct proc *p; 1452{ 1453 1454 simple_lock(&vp->v_interlock); 1455 if (vp->v_usecount == 0) { 1456 if (inter_lkp) { 1457 simple_unlock(inter_lkp); 1458 } 1459 vgonel(vp, p); 1460 return (1); 1461 } 1462 simple_unlock(&vp->v_interlock); 1463 return (0); 1464} 1465 1466/* 1467 * Eliminate all activity associated with a vnode 1468 * in preparation for reuse. 1469 */ 1470void 1471vgone(vp) 1472 register struct vnode *vp; 1473{ 1474 struct proc *p = curproc; /* XXX */ 1475 1476 simple_lock(&vp->v_interlock); 1477 vgonel(vp, p); 1478} 1479 1480/* 1481 * vgone, with the vp interlock held. 1482 */ 1483static void 1484vgonel(vp, p) 1485 struct vnode *vp; 1486 struct proc *p; 1487{ 1488 struct vnode *vq; 1489 struct vnode *vx; 1490 1491 /* 1492 * If a vgone (or vclean) is already in progress, 1493 * wait until it is done and return. 1494 */ 1495 if (vp->v_flag & VXLOCK) { 1496 vp->v_flag |= VXWANT; 1497 simple_unlock(&vp->v_interlock); 1498 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1499 return; 1500 } 1501 1502 if (vp->v_object) { 1503 vp->v_object->flags |= OBJ_VNODE_GONE; 1504 } 1505 1506 /* 1507 * Clean out the filesystem specific data. 1508 */ 1509 vclean(vp, DOCLOSE, p); 1510 /* 1511 * Delete from old mount point vnode list, if on one. 1512 */ 1513 if (vp->v_mount != NULL) 1514 insmntque(vp, (struct mount *)0); 1515 /* 1516 * If special device, remove it from special device alias list 1517 * if it is on one. 1518 */ 1519 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1520 simple_lock(&spechash_slock); 1521 if (*vp->v_hashchain == vp) { 1522 *vp->v_hashchain = vp->v_specnext; 1523 } else { 1524 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1525 if (vq->v_specnext != vp) 1526 continue; 1527 vq->v_specnext = vp->v_specnext; 1528 break; 1529 } 1530 if (vq == NULL) 1531 panic("missing bdev"); 1532 } 1533 if (vp->v_flag & VALIASED) { 1534 vx = NULL; 1535 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1536 if (vq->v_rdev != vp->v_rdev || 1537 vq->v_type != vp->v_type) 1538 continue; 1539 if (vx) 1540 break; 1541 vx = vq; 1542 } 1543 if (vx == NULL) 1544 panic("missing alias"); 1545 if (vq == NULL) 1546 vx->v_flag &= ~VALIASED; 1547 vp->v_flag &= ~VALIASED; 1548 } 1549 simple_unlock(&spechash_slock); 1550 FREE(vp->v_specinfo, M_VNODE); 1551 vp->v_specinfo = NULL; 1552 } 1553 1554 /* 1555 * If it is on the freelist and not already at the head, 1556 * move it to the head of the list. The test of the back 1557 * pointer and the reference count of zero is because 1558 * it will be removed from the free list by getnewvnode, 1559 * but will not have its reference count incremented until 1560 * after calling vgone. If the reference count were 1561 * incremented first, vgone would (incorrectly) try to 1562 * close the previous instance of the underlying object. 1563 */ 1564 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1565 simple_lock(&vnode_free_list_slock); 1566 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1567 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1568 simple_unlock(&vnode_free_list_slock); 1569 } 1570 1571 vp->v_type = VBAD; 1572} 1573 1574/* 1575 * Lookup a vnode by device number. 1576 */ 1577int 1578vfinddev(dev, type, vpp) 1579 dev_t dev; 1580 enum vtype type; 1581 struct vnode **vpp; 1582{ 1583 register struct vnode *vp; 1584 int rc = 0; 1585 1586 simple_lock(&spechash_slock); 1587 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1588 if (dev != vp->v_rdev || type != vp->v_type) 1589 continue; 1590 *vpp = vp; 1591 rc = 1; 1592 break; 1593 } 1594 simple_unlock(&spechash_slock); 1595 return (rc); 1596} 1597 1598/* 1599 * Calculate the total number of references to a special device. 1600 */ 1601int 1602vcount(vp) 1603 register struct vnode *vp; 1604{ 1605 struct vnode *vq, *vnext; 1606 int count; 1607 1608loop: 1609 if ((vp->v_flag & VALIASED) == 0) 1610 return (vp->v_usecount); 1611 simple_lock(&spechash_slock); 1612 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1613 vnext = vq->v_specnext; 1614 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1615 continue; 1616 /* 1617 * Alias, but not in use, so flush it out. 1618 */ 1619 if (vq->v_usecount == 0 && vq != vp) { 1620 simple_unlock(&spechash_slock); 1621 vgone(vq); 1622 goto loop; 1623 } 1624 count += vq->v_usecount; 1625 } 1626 simple_unlock(&spechash_slock); 1627 return (count); 1628} 1629 1630/* 1631 * Return true for select/poll. 1632 */ 1633int 1634vop_nopoll(ap) 1635 struct vop_poll_args /* { 1636 struct vnode *a_vp; 1637 int a_events; 1638 struct ucred *a_cred; 1639 struct proc *a_p; 1640 } */ *ap; 1641{ 1642 1643 /* 1644 * Just return what we were asked for. 1645 */ 1646 return (ap->a_events & (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)); 1647} 1648 1649/* 1650 * Print out a description of a vnode. 1651 */ 1652static char *typename[] = 1653{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1654 1655void 1656vprint(label, vp) 1657 char *label; 1658 register struct vnode *vp; 1659{ 1660 char buf[64]; 1661 1662 if (label != NULL) 1663 printf("%s: %x: ", label, vp); 1664 else 1665 printf("%x: ", vp); 1666 printf("type %s, usecount %d, writecount %d, refcount %ld,", 1667 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1668 vp->v_holdcnt); 1669 buf[0] = '\0'; 1670 if (vp->v_flag & VROOT) 1671 strcat(buf, "|VROOT"); 1672 if (vp->v_flag & VTEXT) 1673 strcat(buf, "|VTEXT"); 1674 if (vp->v_flag & VSYSTEM) 1675 strcat(buf, "|VSYSTEM"); 1676 if (vp->v_flag & VXLOCK) 1677 strcat(buf, "|VXLOCK"); 1678 if (vp->v_flag & VXWANT) 1679 strcat(buf, "|VXWANT"); 1680 if (vp->v_flag & VBWAIT) 1681 strcat(buf, "|VBWAIT"); 1682 if (vp->v_flag & VALIASED) 1683 strcat(buf, "|VALIASED"); 1684 if (vp->v_flag & VDOOMED) 1685 strcat(buf, "|VDOOMED"); 1686 if (vp->v_flag & VFREE) 1687 strcat(buf, "|VFREE"); 1688 if (buf[0] != '\0') 1689 printf(" flags (%s)", &buf[1]); 1690 if (vp->v_data == NULL) { 1691 printf("\n"); 1692 } else { 1693 printf("\n\t"); 1694 VOP_PRINT(vp); 1695 } 1696} 1697 1698#ifdef DDB 1699/* 1700 * List all of the locked vnodes in the system. 1701 * Called when debugging the kernel. 1702 */ 1703void 1704printlockedvnodes() 1705{ 1706 struct proc *p = curproc; /* XXX */ 1707 struct mount *mp, *nmp; 1708 struct vnode *vp; 1709 1710 printf("Locked vnodes\n"); 1711 simple_lock(&mountlist_slock); 1712 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1713 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1714 nmp = mp->mnt_list.cqe_next; 1715 continue; 1716 } 1717 for (vp = mp->mnt_vnodelist.lh_first; 1718 vp != NULL; 1719 vp = vp->v_mntvnodes.le_next) { 1720 if (VOP_ISLOCKED(vp)) 1721 vprint((char *)0, vp); 1722 } 1723 simple_lock(&mountlist_slock); 1724 nmp = mp->mnt_list.cqe_next; 1725 vfs_unbusy(mp, p); 1726 } 1727 simple_unlock(&mountlist_slock); 1728} 1729#endif 1730 1731/* 1732 * Top level filesystem related information gathering. 1733 */ 1734static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS); 1735 1736static int 1737vfs_sysctl SYSCTL_HANDLER_ARGS 1738{ 1739 int *name = (int *)arg1 - 1; /* XXX */ 1740 u_int namelen = arg2 + 1; /* XXX */ 1741 struct vfsconf *vfsp; 1742 1743#ifndef NO_COMPAT_PRELITE2 1744 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1745 if (namelen == 1) 1746 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1747#endif 1748 1749#ifdef notyet 1750 /* all sysctl names at this level are at least name and field */ 1751 if (namelen < 2) 1752 return (ENOTDIR); /* overloaded */ 1753 if (name[0] != VFS_GENERIC) { 1754 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1755 if (vfsp->vfc_typenum == name[0]) 1756 break; 1757 if (vfsp == NULL) 1758 return (EOPNOTSUPP); 1759 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1760 oldp, oldlenp, newp, newlen, p)); 1761 } 1762#endif 1763 switch (name[1]) { 1764 case VFS_MAXTYPENUM: 1765 if (namelen != 2) 1766 return (ENOTDIR); 1767 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1768 case VFS_CONF: 1769 if (namelen != 3) 1770 return (ENOTDIR); /* overloaded */ 1771 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1772 if (vfsp->vfc_typenum == name[2]) 1773 break; 1774 if (vfsp == NULL) 1775 return (EOPNOTSUPP); 1776 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1777 } 1778 return (EOPNOTSUPP); 1779} 1780 1781SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1782 "Generic filesystem"); 1783 1784#ifndef NO_COMPAT_PRELITE2 1785 1786static int 1787sysctl_ovfs_conf SYSCTL_HANDLER_ARGS 1788{ 1789 int error; 1790 struct vfsconf *vfsp; 1791 struct ovfsconf ovfs; 1792 1793 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1794 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1795 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1796 ovfs.vfc_index = vfsp->vfc_typenum; 1797 ovfs.vfc_refcount = vfsp->vfc_refcount; 1798 ovfs.vfc_flags = vfsp->vfc_flags; 1799 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1800 if (error) 1801 return error; 1802 } 1803 return 0; 1804} 1805 1806#endif /* !NO_COMPAT_PRELITE2 */ 1807 1808int kinfo_vdebug = 1; 1809int kinfo_vgetfailed; 1810 1811#define KINFO_VNODESLOP 10 1812/* 1813 * Dump vnode list (via sysctl). 1814 * Copyout address of vnode followed by vnode. 1815 */ 1816/* ARGSUSED */ 1817static int 1818sysctl_vnode SYSCTL_HANDLER_ARGS 1819{ 1820 struct proc *p = curproc; /* XXX */ 1821 struct mount *mp, *nmp; 1822 struct vnode *nvp, *vp; 1823 int error; 1824 1825#define VPTRSZ sizeof (struct vnode *) 1826#define VNODESZ sizeof (struct vnode) 1827 1828 req->lock = 0; 1829 if (!req->oldptr) /* Make an estimate */ 1830 return (SYSCTL_OUT(req, 0, 1831 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 1832 1833 simple_lock(&mountlist_slock); 1834 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1835 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1836 nmp = mp->mnt_list.cqe_next; 1837 continue; 1838 } 1839again: 1840 simple_lock(&mntvnode_slock); 1841 for (vp = mp->mnt_vnodelist.lh_first; 1842 vp != NULL; 1843 vp = nvp) { 1844 /* 1845 * Check that the vp is still associated with 1846 * this filesystem. RACE: could have been 1847 * recycled onto the same filesystem. 1848 */ 1849 if (vp->v_mount != mp) { 1850 simple_unlock(&mntvnode_slock); 1851 if (kinfo_vdebug) 1852 printf("kinfo: vp changed\n"); 1853 goto again; 1854 } 1855 nvp = vp->v_mntvnodes.le_next; 1856 simple_unlock(&mntvnode_slock); 1857 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 1858 (error = SYSCTL_OUT(req, vp, VNODESZ))) 1859 return (error); 1860 simple_lock(&mntvnode_slock); 1861 } 1862 simple_unlock(&mntvnode_slock); 1863 simple_lock(&mountlist_slock); 1864 nmp = mp->mnt_list.cqe_next; 1865 vfs_unbusy(mp, p); 1866 } 1867 simple_unlock(&mountlist_slock); 1868 1869 return (0); 1870} 1871 1872/* 1873 * XXX 1874 * Exporting the vnode list on large systems causes them to crash. 1875 * Exporting the vnode list on medium systems causes sysctl to coredump. 1876 */ 1877#if 0 1878SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 1879 0, 0, sysctl_vnode, "S,vnode", ""); 1880#endif 1881 1882/* 1883 * Check to see if a filesystem is mounted on a block device. 1884 */ 1885int 1886vfs_mountedon(vp) 1887 struct vnode *vp; 1888{ 1889 struct vnode *vq; 1890 int error = 0; 1891 1892 if (vp->v_specflags & SI_MOUNTEDON) 1893 return (EBUSY); 1894 if (vp->v_flag & VALIASED) { 1895 simple_lock(&spechash_slock); 1896 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1897 if (vq->v_rdev != vp->v_rdev || 1898 vq->v_type != vp->v_type) 1899 continue; 1900 if (vq->v_specflags & SI_MOUNTEDON) { 1901 error = EBUSY; 1902 break; 1903 } 1904 } 1905 simple_unlock(&spechash_slock); 1906 } 1907 return (error); 1908} 1909 1910/* 1911 * Unmount all filesystems. The list is traversed in reverse order 1912 * of mounting to avoid dependencies. 1913 */ 1914void 1915vfs_unmountall() 1916{ 1917 struct mount *mp, *nmp; 1918 struct proc *p = initproc; /* XXX XXX should this be proc0? */ 1919 int error; 1920 1921 /* 1922 * Since this only runs when rebooting, it is not interlocked. 1923 */ 1924 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 1925 nmp = mp->mnt_list.cqe_prev; 1926 error = dounmount(mp, MNT_FORCE, p); 1927 if (error) { 1928 printf("unmount of %s failed (", 1929 mp->mnt_stat.f_mntonname); 1930 if (error == EBUSY) 1931 printf("BUSY)\n"); 1932 else 1933 printf("%d)\n", error); 1934 } 1935 } 1936} 1937 1938/* 1939 * Build hash lists of net addresses and hang them off the mount point. 1940 * Called by ufs_mount() to set up the lists of export addresses. 1941 */ 1942static int 1943vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1944 struct export_args *argp) 1945{ 1946 register struct netcred *np; 1947 register struct radix_node_head *rnh; 1948 register int i; 1949 struct radix_node *rn; 1950 struct sockaddr *saddr, *smask = 0; 1951 struct domain *dom; 1952 int error; 1953 1954 if (argp->ex_addrlen == 0) { 1955 if (mp->mnt_flag & MNT_DEFEXPORTED) 1956 return (EPERM); 1957 np = &nep->ne_defexported; 1958 np->netc_exflags = argp->ex_flags; 1959 np->netc_anon = argp->ex_anon; 1960 np->netc_anon.cr_ref = 1; 1961 mp->mnt_flag |= MNT_DEFEXPORTED; 1962 return (0); 1963 } 1964 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1965 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1966 bzero((caddr_t) np, i); 1967 saddr = (struct sockaddr *) (np + 1); 1968 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1969 goto out; 1970 if (saddr->sa_len > argp->ex_addrlen) 1971 saddr->sa_len = argp->ex_addrlen; 1972 if (argp->ex_masklen) { 1973 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 1974 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 1975 if (error) 1976 goto out; 1977 if (smask->sa_len > argp->ex_masklen) 1978 smask->sa_len = argp->ex_masklen; 1979 } 1980 i = saddr->sa_family; 1981 if ((rnh = nep->ne_rtable[i]) == 0) { 1982 /* 1983 * Seems silly to initialize every AF when most are not used, 1984 * do so on demand here 1985 */ 1986 for (dom = domains; dom; dom = dom->dom_next) 1987 if (dom->dom_family == i && dom->dom_rtattach) { 1988 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1989 dom->dom_rtoffset); 1990 break; 1991 } 1992 if ((rnh = nep->ne_rtable[i]) == 0) { 1993 error = ENOBUFS; 1994 goto out; 1995 } 1996 } 1997 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 1998 np->netc_rnodes); 1999 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 2000 error = EPERM; 2001 goto out; 2002 } 2003 np->netc_exflags = argp->ex_flags; 2004 np->netc_anon = argp->ex_anon; 2005 np->netc_anon.cr_ref = 1; 2006 return (0); 2007out: 2008 free(np, M_NETADDR); 2009 return (error); 2010} 2011 2012/* ARGSUSED */ 2013static int 2014vfs_free_netcred(struct radix_node *rn, void *w) 2015{ 2016 register struct radix_node_head *rnh = (struct radix_node_head *) w; 2017 2018 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2019 free((caddr_t) rn, M_NETADDR); 2020 return (0); 2021} 2022 2023/* 2024 * Free the net address hash lists that are hanging off the mount points. 2025 */ 2026static void 2027vfs_free_addrlist(struct netexport *nep) 2028{ 2029 register int i; 2030 register struct radix_node_head *rnh; 2031 2032 for (i = 0; i <= AF_MAX; i++) 2033 if ((rnh = nep->ne_rtable[i])) { 2034 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2035 (caddr_t) rnh); 2036 free((caddr_t) rnh, M_RTABLE); 2037 nep->ne_rtable[i] = 0; 2038 } 2039} 2040 2041int 2042vfs_export(mp, nep, argp) 2043 struct mount *mp; 2044 struct netexport *nep; 2045 struct export_args *argp; 2046{ 2047 int error; 2048 2049 if (argp->ex_flags & MNT_DELEXPORT) { 2050 if (mp->mnt_flag & MNT_EXPUBLIC) { 2051 vfs_setpublicfs(NULL, NULL, NULL); 2052 mp->mnt_flag &= ~MNT_EXPUBLIC; 2053 } 2054 vfs_free_addrlist(nep); 2055 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2056 } 2057 if (argp->ex_flags & MNT_EXPORTED) { 2058 if (argp->ex_flags & MNT_EXPUBLIC) { 2059 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2060 return (error); 2061 mp->mnt_flag |= MNT_EXPUBLIC; 2062 } 2063 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2064 return (error); 2065 mp->mnt_flag |= MNT_EXPORTED; 2066 } 2067 return (0); 2068} 2069 2070 2071/* 2072 * Set the publicly exported filesystem (WebNFS). Currently, only 2073 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2074 */ 2075int 2076vfs_setpublicfs(mp, nep, argp) 2077 struct mount *mp; 2078 struct netexport *nep; 2079 struct export_args *argp; 2080{ 2081 int error; 2082 struct vnode *rvp; 2083 char *cp; 2084 2085 /* 2086 * mp == NULL -> invalidate the current info, the FS is 2087 * no longer exported. May be called from either vfs_export 2088 * or unmount, so check if it hasn't already been done. 2089 */ 2090 if (mp == NULL) { 2091 if (nfs_pub.np_valid) { 2092 nfs_pub.np_valid = 0; 2093 if (nfs_pub.np_index != NULL) { 2094 FREE(nfs_pub.np_index, M_TEMP); 2095 nfs_pub.np_index = NULL; 2096 } 2097 } 2098 return (0); 2099 } 2100 2101 /* 2102 * Only one allowed at a time. 2103 */ 2104 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2105 return (EBUSY); 2106 2107 /* 2108 * Get real filehandle for root of exported FS. 2109 */ 2110 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2111 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2112 2113 if ((error = VFS_ROOT(mp, &rvp))) 2114 return (error); 2115 2116 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2117 return (error); 2118 2119 vput(rvp); 2120 2121 /* 2122 * If an indexfile was specified, pull it in. 2123 */ 2124 if (argp->ex_indexfile != NULL) { 2125 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2126 M_WAITOK); 2127 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2128 MAXNAMLEN, (size_t *)0); 2129 if (!error) { 2130 /* 2131 * Check for illegal filenames. 2132 */ 2133 for (cp = nfs_pub.np_index; *cp; cp++) { 2134 if (*cp == '/') { 2135 error = EINVAL; 2136 break; 2137 } 2138 } 2139 } 2140 if (error) { 2141 FREE(nfs_pub.np_index, M_TEMP); 2142 return (error); 2143 } 2144 } 2145 2146 nfs_pub.np_mount = mp; 2147 nfs_pub.np_valid = 1; 2148 return (0); 2149} 2150 2151struct netcred * 2152vfs_export_lookup(mp, nep, nam) 2153 register struct mount *mp; 2154 struct netexport *nep; 2155 struct sockaddr *nam; 2156{ 2157 register struct netcred *np; 2158 register struct radix_node_head *rnh; 2159 struct sockaddr *saddr; 2160 2161 np = NULL; 2162 if (mp->mnt_flag & MNT_EXPORTED) { 2163 /* 2164 * Lookup in the export list first. 2165 */ 2166 if (nam != NULL) { 2167 saddr = nam; 2168 rnh = nep->ne_rtable[saddr->sa_family]; 2169 if (rnh != NULL) { 2170 np = (struct netcred *) 2171 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2172 rnh); 2173 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2174 np = NULL; 2175 } 2176 } 2177 /* 2178 * If no address match, use the default if it exists. 2179 */ 2180 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2181 np = &nep->ne_defexported; 2182 } 2183 return (np); 2184} 2185 2186/* 2187 * perform msync on all vnodes under a mount point 2188 * the mount point must be locked. 2189 */ 2190void 2191vfs_msync(struct mount *mp, int flags) { 2192 struct vnode *vp, *nvp; 2193loop: 2194 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 2195 2196 if (vp->v_mount != mp) 2197 goto loop; 2198 nvp = vp->v_mntvnodes.le_next; 2199 if (VOP_ISLOCKED(vp) && (flags != MNT_WAIT)) 2200 continue; 2201 if (vp->v_object && 2202 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2203 vm_object_page_clean(vp->v_object, 0, 0, TRUE, TRUE); 2204 } 2205 } 2206} 2207 2208/* 2209 * Create the VM object needed for VMIO and mmap support. This 2210 * is done for all VREG files in the system. Some filesystems might 2211 * afford the additional metadata buffering capability of the 2212 * VMIO code by making the device node be VMIO mode also. 2213 */ 2214int 2215vfs_object_create(vp, p, cred, waslocked) 2216 struct vnode *vp; 2217 struct proc *p; 2218 struct ucred *cred; 2219 int waslocked; 2220{ 2221 struct vattr vat; 2222 vm_object_t object; 2223 int error = 0; 2224 2225retry: 2226 if ((object = vp->v_object) == NULL) { 2227 if (vp->v_type == VREG) { 2228 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2229 goto retn; 2230 (void) vnode_pager_alloc(vp, 2231 OFF_TO_IDX(round_page(vat.va_size)), 0, 0); 2232 } else { 2233 /* 2234 * This simply allocates the biggest object possible 2235 * for a VBLK vnode. This should be fixed, but doesn't 2236 * cause any problems (yet). 2237 */ 2238 (void) vnode_pager_alloc(vp, INT_MAX, 0, 0); 2239 } 2240 vp->v_object->flags |= OBJ_VFS_REF; 2241 } else { 2242 if (object->flags & OBJ_DEAD) { 2243 if (waslocked) 2244 VOP_UNLOCK(vp, 0, p); 2245 tsleep(object, PVM, "vodead", 0); 2246 if (waslocked) 2247 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2248 goto retry; 2249 } 2250 if ((object->flags & OBJ_VFS_REF) == 0) { 2251 object->flags |= OBJ_VFS_REF; 2252 vm_object_reference(object); 2253 } 2254 } 2255 if (vp->v_object) 2256 vp->v_flag |= VVMIO; 2257 2258retn: 2259 return error; 2260} 2261 2262void 2263vfree(vp) 2264 struct vnode *vp; 2265{ 2266 simple_lock(&vnode_free_list_slock); 2267 if (vp->v_flag & VAGE) { 2268 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2269 } else { 2270 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2271 } 2272 freevnodes++; 2273 simple_unlock(&vnode_free_list_slock); 2274 vp->v_flag &= ~VAGE; 2275 vp->v_flag |= VFREE; 2276} 2277 2278void 2279vbusy(vp) 2280 struct vnode *vp; 2281{ 2282 simple_lock(&vnode_free_list_slock); 2283 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2284 freevnodes--; 2285 simple_unlock(&vnode_free_list_slock); 2286 vp->v_flag &= ~VFREE; 2287} 2288