vfs_export.c revision 32071
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 1.116 1997/12/19 09:03:28 dyson 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/malloc.h> 53#include <sys/mount.h> 54#include <sys/vnode.h> 55#include <sys/stat.h> 56#include <sys/buf.h> 57#include <sys/poll.h> 58#include <sys/domain.h> 59#include <sys/dirent.h> 60#include <sys/vmmeter.h> 61 62#include <machine/limits.h> 63 64#include <vm/vm.h> 65#include <vm/vm_object.h> 66#include <vm/vm_extern.h> 67#include <vm/pmap.h> 68#include <vm/vm_map.h> 69#include <vm/vnode_pager.h> 70#include <sys/sysctl.h> 71 72#include <miscfs/specfs/specdev.h> 73 74static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 75 76static void insmntque __P((struct vnode *vp, struct mount *mp)); 77#ifdef DDB 78static void printlockedvnodes __P((void)); 79#endif 80static void vbusy __P((struct vnode *)); 81static void vclean __P((struct vnode *vp, int flags, struct proc *p)); 82static void vfree __P((struct vnode *)); 83static void vgonel __P((struct vnode *vp, struct proc *p)); 84static unsigned long numvnodes; 85SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 86static void vputrele __P((struct vnode *vp, int put)); 87 88enum vtype iftovt_tab[16] = { 89 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 90 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 91}; 92int vttoif_tab[9] = { 93 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 94 S_IFSOCK, S_IFIFO, S_IFMT, 95}; 96 97/* 98 * Insq/Remq for the vnode usage lists. 99 */ 100#define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 101#define bufremvn(bp) { \ 102 LIST_REMOVE(bp, b_vnbufs); \ 103 (bp)->b_vnbufs.le_next = NOLIST; \ 104} 105TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 106static u_long wantfreevnodes = 25; 107SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 108static u_long freevnodes = 0; 109SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 110 111struct mntlist mountlist; /* mounted filesystem list */ 112struct simplelock mountlist_slock; 113static struct simplelock mntid_slock; 114struct simplelock mntvnode_slock; 115struct simplelock vnode_free_list_slock; 116static struct simplelock spechash_slock; 117struct nfs_public nfs_pub; /* publicly exported FS */ 118 119int desiredvnodes; 120SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, ""); 121 122static void vfs_free_addrlist __P((struct netexport *nep)); 123static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 124static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 125 struct export_args *argp)); 126 127/* 128 * Initialize the vnode management data structures. 129 */ 130void 131vntblinit() 132{ 133 134 desiredvnodes = maxproc + cnt.v_page_count / 4; 135 simple_lock_init(&mntvnode_slock); 136 simple_lock_init(&mntid_slock); 137 simple_lock_init(&spechash_slock); 138 TAILQ_INIT(&vnode_free_list); 139 simple_lock_init(&vnode_free_list_slock); 140 CIRCLEQ_INIT(&mountlist); 141} 142 143/* 144 * Mark a mount point as busy. Used to synchronize access and to delay 145 * unmounting. Interlock is not released on failure. 146 */ 147int 148vfs_busy(mp, flags, interlkp, p) 149 struct mount *mp; 150 int flags; 151 struct simplelock *interlkp; 152 struct proc *p; 153{ 154 int lkflags; 155 156 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 157 if (flags & LK_NOWAIT) 158 return (ENOENT); 159 mp->mnt_kern_flag |= MNTK_MWAIT; 160 if (interlkp) { 161 simple_unlock(interlkp); 162 } 163 /* 164 * Since all busy locks are shared except the exclusive 165 * lock granted when unmounting, the only place that a 166 * wakeup needs to be done is at the release of the 167 * exclusive lock at the end of dounmount. 168 */ 169 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 170 if (interlkp) { 171 simple_lock(interlkp); 172 } 173 return (ENOENT); 174 } 175 lkflags = LK_SHARED; 176 if (interlkp) 177 lkflags |= LK_INTERLOCK; 178 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 179 panic("vfs_busy: unexpected lock failure"); 180 return (0); 181} 182 183/* 184 * Free a busy filesystem. 185 */ 186void 187vfs_unbusy(mp, p) 188 struct mount *mp; 189 struct proc *p; 190{ 191 192 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 193} 194 195/* 196 * Lookup a filesystem type, and if found allocate and initialize 197 * a mount structure for it. 198 * 199 * Devname is usually updated by mount(8) after booting. 200 */ 201int 202vfs_rootmountalloc(fstypename, devname, mpp) 203 char *fstypename; 204 char *devname; 205 struct mount **mpp; 206{ 207 struct proc *p = curproc; /* XXX */ 208 struct vfsconf *vfsp; 209 struct mount *mp; 210 211 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 212 if (!strcmp(vfsp->vfc_name, fstypename)) 213 break; 214 if (vfsp == NULL) 215 return (ENODEV); 216 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 217 bzero((char *)mp, (u_long)sizeof(struct mount)); 218 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 219 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 220 LIST_INIT(&mp->mnt_vnodelist); 221 mp->mnt_vfc = vfsp; 222 mp->mnt_op = vfsp->vfc_vfsops; 223 mp->mnt_flag = MNT_RDONLY; 224 mp->mnt_vnodecovered = NULLVP; 225 vfsp->vfc_refcount++; 226 mp->mnt_stat.f_type = vfsp->vfc_typenum; 227 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 228 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 229 mp->mnt_stat.f_mntonname[0] = '/'; 230 mp->mnt_stat.f_mntonname[1] = 0; 231 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 232 *mpp = mp; 233 return (0); 234} 235 236/* 237 * Find an appropriate filesystem to use for the root. If a filesystem 238 * has not been preselected, walk through the list of known filesystems 239 * trying those that have mountroot routines, and try them until one 240 * works or we have tried them all. 241 */ 242#ifdef notdef /* XXX JH */ 243int 244lite2_vfs_mountroot() 245{ 246 struct vfsconf *vfsp; 247 extern int (*lite2_mountroot) __P((void)); 248 int error; 249 250 if (lite2_mountroot != NULL) 251 return ((*lite2_mountroot)()); 252 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 253 if (vfsp->vfc_mountroot == NULL) 254 continue; 255 if ((error = (*vfsp->vfc_mountroot)()) == 0) 256 return (0); 257 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 258 } 259 return (ENODEV); 260} 261#endif 262 263/* 264 * Lookup a mount point by filesystem identifier. 265 */ 266struct mount * 267vfs_getvfs(fsid) 268 fsid_t *fsid; 269{ 270 register struct mount *mp; 271 272 simple_lock(&mountlist_slock); 273 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 274 mp = mp->mnt_list.cqe_next) { 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 * Get a new unique fsid 287 */ 288void 289vfs_getnewfsid(mp) 290 struct mount *mp; 291{ 292 static u_short xxxfs_mntid; 293 294 fsid_t tfsid; 295 int mtype; 296 297 simple_lock(&mntid_slock); 298 mtype = mp->mnt_vfc->vfc_typenum; 299 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 300 mp->mnt_stat.f_fsid.val[1] = mtype; 301 if (xxxfs_mntid == 0) 302 ++xxxfs_mntid; 303 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 304 tfsid.val[1] = mtype; 305 if (mountlist.cqh_first != (void *)&mountlist) { 306 while (vfs_getvfs(&tfsid)) { 307 tfsid.val[0]++; 308 xxxfs_mntid++; 309 } 310 } 311 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 312 simple_unlock(&mntid_slock); 313} 314 315/* 316 * Set vnode attributes to VNOVAL 317 */ 318void 319vattr_null(vap) 320 register struct vattr *vap; 321{ 322 323 vap->va_type = VNON; 324 vap->va_size = VNOVAL; 325 vap->va_bytes = VNOVAL; 326 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 327 vap->va_fsid = vap->va_fileid = 328 vap->va_blocksize = vap->va_rdev = 329 vap->va_atime.tv_sec = vap->va_atime.tv_nsec = 330 vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec = 331 vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec = 332 vap->va_flags = vap->va_gen = VNOVAL; 333 vap->va_vaflags = 0; 334} 335 336/* 337 * Routines having to do with the management of the vnode table. 338 */ 339extern vop_t **dead_vnodeop_p; 340 341/* 342 * Return the next vnode from the free list. 343 */ 344int 345getnewvnode(tag, mp, vops, vpp) 346 enum vtagtype tag; 347 struct mount *mp; 348 vop_t **vops; 349 struct vnode **vpp; 350{ 351 struct proc *p = curproc; /* XXX */ 352 struct vnode *vp; 353 354 /* 355 * We take the least recently used vnode from the freelist 356 * if we can get it and it has no cached pages, and no 357 * namecache entries are relative to it. 358 * Otherwise we allocate a new vnode 359 */ 360 361 simple_lock(&vnode_free_list_slock); 362 363 if (wantfreevnodes && freevnodes < wantfreevnodes) { 364 vp = NULL; 365 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) { 366 /* 367 * XXX: this is only here to be backwards compatible 368 */ 369 vp = NULL; 370 } else { 371 TAILQ_FOREACH(vp, &vnode_free_list, v_freelist) { 372 if (!simple_lock_try(&vp->v_interlock)) 373 continue; 374 if (vp->v_usecount) 375 panic("free vnode isn't"); 376 377 if (vp->v_object && vp->v_object->resident_page_count) { 378 /* Don't recycle if it's caching some pages */ 379 simple_unlock(&vp->v_interlock); 380 continue; 381 } else if (LIST_FIRST(&vp->v_cache_src)) { 382 /* Don't recycle if active in the namecache */ 383 simple_unlock(&vp->v_interlock); 384 continue; 385 } else { 386 break; 387 } 388 } 389 } 390 391 if (vp) { 392 vp->v_flag |= VDOOMED; 393 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 394 freevnodes--; 395 simple_unlock(&vnode_free_list_slock); 396 cache_purge(vp); 397 vp->v_lease = NULL; 398 if (vp->v_type != VBAD) { 399 vgonel(vp, p); 400 } else { 401 simple_unlock(&vp->v_interlock); 402 } 403 404#ifdef DIAGNOSTIC 405 { 406 int s; 407 408 if (vp->v_data) 409 panic("cleaned vnode isn't"); 410 s = splbio(); 411 if (vp->v_numoutput) 412 panic("Clean vnode has pending I/O's"); 413 splx(s); 414 } 415#endif 416 vp->v_flag = 0; 417 vp->v_lastr = 0; 418 vp->v_lastw = 0; 419 vp->v_lasta = 0; 420 vp->v_cstart = 0; 421 vp->v_clen = 0; 422 vp->v_socket = 0; 423 vp->v_writecount = 0; /* XXX */ 424 } else { 425 simple_unlock(&vnode_free_list_slock); 426 vp = (struct vnode *) malloc((u_long) sizeof *vp, 427 M_VNODE, M_WAITOK); 428 bzero((char *) vp, sizeof *vp); 429 vp->v_dd = vp; 430 cache_purge(vp); 431 LIST_INIT(&vp->v_cache_src); 432 TAILQ_INIT(&vp->v_cache_dst); 433 numvnodes++; 434 } 435 436 vp->v_type = VNON; 437 vp->v_tag = tag; 438 vp->v_op = vops; 439 insmntque(vp, mp); 440 *vpp = vp; 441 vp->v_usecount = 1; 442 vp->v_data = 0; 443 return (0); 444} 445 446/* 447 * Move a vnode from one mount queue to another. 448 */ 449static void 450insmntque(vp, mp) 451 register struct vnode *vp; 452 register struct mount *mp; 453{ 454 455 simple_lock(&mntvnode_slock); 456 /* 457 * Delete from old mount point vnode list, if on one. 458 */ 459 if (vp->v_mount != NULL) 460 LIST_REMOVE(vp, v_mntvnodes); 461 /* 462 * Insert into list of vnodes for the new mount point, if available. 463 */ 464 if ((vp->v_mount = mp) == NULL) { 465 simple_unlock(&mntvnode_slock); 466 return; 467 } 468 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 469 simple_unlock(&mntvnode_slock); 470} 471 472/* 473 * Update outstanding I/O count and do wakeup if requested. 474 */ 475void 476vwakeup(bp) 477 register struct buf *bp; 478{ 479 register struct vnode *vp; 480 481 bp->b_flags &= ~B_WRITEINPROG; 482 if ((vp = bp->b_vp)) { 483 vp->v_numoutput--; 484 if (vp->v_numoutput < 0) 485 panic("vwakeup: neg numoutput"); 486 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 487 vp->v_flag &= ~VBWAIT; 488 wakeup((caddr_t) &vp->v_numoutput); 489 } 490 } 491} 492 493/* 494 * Flush out and invalidate all buffers associated with a vnode. 495 * Called with the underlying object locked. 496 */ 497int 498vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 499 register struct vnode *vp; 500 int flags; 501 struct ucred *cred; 502 struct proc *p; 503 int slpflag, slptimeo; 504{ 505 register struct buf *bp; 506 struct buf *nbp, *blist; 507 int s, error; 508 vm_object_t object; 509 510 if (flags & V_SAVE) { 511 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p))) 512 return (error); 513 if (vp->v_dirtyblkhd.lh_first != NULL) 514 panic("vinvalbuf: dirty bufs"); 515 } 516 517 s = splbio(); 518 for (;;) { 519 if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA)) 520 while (blist && blist->b_lblkno < 0) 521 blist = blist->b_vnbufs.le_next; 522 if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 523 (flags & V_SAVEMETA)) 524 while (blist && blist->b_lblkno < 0) 525 blist = blist->b_vnbufs.le_next; 526 if (!blist) 527 break; 528 529 for (bp = blist; bp; bp = nbp) { 530 nbp = bp->b_vnbufs.le_next; 531 if ((flags & V_SAVEMETA) && bp->b_lblkno < 0) 532 continue; 533 if (bp->b_flags & B_BUSY) { 534 bp->b_flags |= B_WANTED; 535 error = tsleep((caddr_t) bp, 536 slpflag | (PRIBIO + 1), "vinvalbuf", 537 slptimeo); 538 if (error) { 539 splx(s); 540 return (error); 541 } 542 break; 543 } 544 bremfree(bp); 545 bp->b_flags |= B_BUSY; 546 /* 547 * XXX Since there are no node locks for NFS, I 548 * believe there is a slight chance that a delayed 549 * write will occur while sleeping just above, so 550 * check for it. 551 */ 552 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 553 (void) VOP_BWRITE(bp); 554 break; 555 } 556 bp->b_flags |= (B_INVAL|B_NOCACHE|B_RELBUF); 557 brelse(bp); 558 } 559 } 560 561 while (vp->v_numoutput > 0) { 562 vp->v_flag |= VBWAIT; 563 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 564 } 565 566 splx(s); 567 568 /* 569 * Destroy the copy in the VM cache, too. 570 */ 571 object = vp->v_object; 572 if (object != NULL) { 573 vm_object_page_remove(object, 0, object->size, 574 (flags & V_SAVE) ? TRUE : FALSE); 575 } 576 if (!(flags & V_SAVEMETA) && 577 (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) 578 panic("vinvalbuf: flush failed"); 579 return (0); 580} 581 582/* 583 * Associate a buffer with a vnode. 584 */ 585void 586bgetvp(vp, bp) 587 register struct vnode *vp; 588 register struct buf *bp; 589{ 590 int s; 591 592#if defined(DIAGNOSTIC) 593 if (bp->b_vp) 594 panic("bgetvp: not free"); 595#endif 596 vhold(vp); 597 bp->b_vp = vp; 598 if (vp->v_type == VBLK || vp->v_type == VCHR) 599 bp->b_dev = vp->v_rdev; 600 else 601 bp->b_dev = NODEV; 602 /* 603 * Insert onto list for new vnode. 604 */ 605 s = splbio(); 606 bufinsvn(bp, &vp->v_cleanblkhd); 607 splx(s); 608} 609 610/* 611 * Disassociate a buffer from a vnode. 612 */ 613void 614brelvp(bp) 615 register struct buf *bp; 616{ 617 struct vnode *vp; 618 int s; 619 620#if defined(DIAGNOSTIC) 621 if (bp->b_vp == (struct vnode *) 0) 622 panic("brelvp: NULL"); 623#endif 624 625 /* 626 * Delete from old vnode list, if on one. 627 */ 628 s = splbio(); 629 if (bp->b_vnbufs.le_next != NOLIST) 630 bufremvn(bp); 631 splx(s); 632 633 vp = bp->b_vp; 634 bp->b_vp = (struct vnode *) 0; 635 vdrop(vp); 636} 637 638/* 639 * Associate a p-buffer with a vnode. 640 */ 641void 642pbgetvp(vp, bp) 643 register struct vnode *vp; 644 register struct buf *bp; 645{ 646#if defined(DIAGNOSTIC) 647 if (bp->b_vp) 648 panic("pbgetvp: not free"); 649#endif 650 bp->b_vp = vp; 651 if (vp->v_type == VBLK || vp->v_type == VCHR) 652 bp->b_dev = vp->v_rdev; 653 else 654 bp->b_dev = NODEV; 655} 656 657/* 658 * Disassociate a p-buffer from a vnode. 659 */ 660void 661pbrelvp(bp) 662 register struct buf *bp; 663{ 664 665#if defined(DIAGNOSTIC) 666 if (bp->b_vp == (struct vnode *) 0) 667 panic("pbrelvp: NULL"); 668#endif 669 670 bp->b_vp = (struct vnode *) 0; 671} 672 673/* 674 * Reassign a buffer from one vnode to another. 675 * Used to assign file specific control information 676 * (indirect blocks) to the vnode to which they belong. 677 */ 678void 679reassignbuf(bp, newvp) 680 register struct buf *bp; 681 register struct vnode *newvp; 682{ 683 int s; 684 685 if (newvp == NULL) { 686 printf("reassignbuf: NULL"); 687 return; 688 } 689 690 s = splbio(); 691 /* 692 * Delete from old vnode list, if on one. 693 */ 694 if (bp->b_vnbufs.le_next != NOLIST) { 695 bufremvn(bp); 696 vdrop(bp->b_vp); 697 } 698 /* 699 * If dirty, put on list of dirty buffers; otherwise insert onto list 700 * of clean buffers. 701 */ 702 if (bp->b_flags & B_DELWRI) { 703 struct buf *tbp; 704 705 tbp = newvp->v_dirtyblkhd.lh_first; 706 if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) { 707 bufinsvn(bp, &newvp->v_dirtyblkhd); 708 } else { 709 while (tbp->b_vnbufs.le_next && 710 (tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) { 711 tbp = tbp->b_vnbufs.le_next; 712 } 713 LIST_INSERT_AFTER(tbp, bp, b_vnbufs); 714 } 715 } else { 716 bufinsvn(bp, &newvp->v_cleanblkhd); 717 } 718 bp->b_vp = newvp; 719 vhold(bp->b_vp); 720 splx(s); 721} 722 723#ifndef DEVFS_ROOT 724/* 725 * Create a vnode for a block device. 726 * Used for mounting the root file system. 727 */ 728int 729bdevvp(dev, vpp) 730 dev_t dev; 731 struct vnode **vpp; 732{ 733 register struct vnode *vp; 734 struct vnode *nvp; 735 int error; 736 737 if (dev == NODEV) 738 return (0); 739 error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp); 740 if (error) { 741 *vpp = 0; 742 return (error); 743 } 744 vp = nvp; 745 vp->v_type = VBLK; 746 if ((nvp = checkalias(vp, dev, (struct mount *) 0))) { 747 vput(vp); 748 vp = nvp; 749 } 750 *vpp = vp; 751 return (0); 752} 753#endif /* !DEVFS_ROOT */ 754 755/* 756 * Check to see if the new vnode represents a special device 757 * for which we already have a vnode (either because of 758 * bdevvp() or because of a different vnode representing 759 * the same block device). If such an alias exists, deallocate 760 * the existing contents and return the aliased vnode. The 761 * caller is responsible for filling it with its new contents. 762 */ 763struct vnode * 764checkalias(nvp, nvp_rdev, mp) 765 register struct vnode *nvp; 766 dev_t nvp_rdev; 767 struct mount *mp; 768{ 769 struct proc *p = curproc; /* XXX */ 770 struct vnode *vp; 771 struct vnode **vpp; 772 773 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 774 return (NULLVP); 775 776 vpp = &speclisth[SPECHASH(nvp_rdev)]; 777loop: 778 simple_lock(&spechash_slock); 779 for (vp = *vpp; vp; vp = vp->v_specnext) { 780 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 781 continue; 782 /* 783 * Alias, but not in use, so flush it out. 784 */ 785 simple_lock(&vp->v_interlock); 786 if (vp->v_usecount == 0) { 787 simple_unlock(&spechash_slock); 788 vgonel(vp, p); 789 goto loop; 790 } 791 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { 792 simple_unlock(&spechash_slock); 793 goto loop; 794 } 795 break; 796 } 797 if (vp == NULL || vp->v_tag != VT_NON) { 798 MALLOC(nvp->v_specinfo, struct specinfo *, 799 sizeof(struct specinfo), M_VNODE, M_WAITOK); 800 nvp->v_rdev = nvp_rdev; 801 nvp->v_hashchain = vpp; 802 nvp->v_specnext = *vpp; 803 nvp->v_specflags = 0; 804 simple_unlock(&spechash_slock); 805 *vpp = nvp; 806 if (vp != NULLVP) { 807 nvp->v_flag |= VALIASED; 808 vp->v_flag |= VALIASED; 809 vput(vp); 810 } 811 return (NULLVP); 812 } 813 simple_unlock(&spechash_slock); 814 VOP_UNLOCK(vp, 0, p); 815 simple_lock(&vp->v_interlock); 816 vclean(vp, 0, p); 817 vp->v_op = nvp->v_op; 818 vp->v_tag = nvp->v_tag; 819 nvp->v_type = VNON; 820 insmntque(vp, mp); 821 return (vp); 822} 823 824/* 825 * Grab a particular vnode from the free list, increment its 826 * reference count and lock it. The vnode lock bit is set the 827 * vnode is being eliminated in vgone. The process is awakened 828 * when the transition is completed, and an error returned to 829 * indicate that the vnode is no longer usable (possibly having 830 * been changed to a new file system type). 831 */ 832int 833vget(vp, flags, p) 834 register struct vnode *vp; 835 int flags; 836 struct proc *p; 837{ 838 int error; 839 840 /* 841 * If the vnode is in the process of being cleaned out for 842 * another use, we wait for the cleaning to finish and then 843 * return failure. Cleaning is determined by checking that 844 * the VXLOCK flag is set. 845 */ 846 if ((flags & LK_INTERLOCK) == 0) { 847 simple_lock(&vp->v_interlock); 848 } 849 if (vp->v_flag & VXLOCK) { 850 vp->v_flag |= VXWANT; 851 simple_unlock(&vp->v_interlock); 852 tsleep((caddr_t)vp, PINOD, "vget", 0); 853 return (ENOENT); 854 } 855 856 vp->v_usecount++; 857 858 if (VSHOULDBUSY(vp)) 859 vbusy(vp); 860 /* 861 * Create the VM object, if needed 862 */ 863 if (((vp->v_type == VREG) || (vp->v_type == VBLK)) && 864 ((vp->v_object == NULL) || 865 (vp->v_object->flags & OBJ_VFS_REF) == 0 || 866 (vp->v_object->flags & OBJ_DEAD))) { 867 simple_unlock(&vp->v_interlock); 868 vfs_object_create(vp, curproc, curproc->p_ucred, 0); 869 simple_lock(&vp->v_interlock); 870 } 871 if (flags & LK_TYPE_MASK) { 872 if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) 873 vrele(vp); 874 return (error); 875 } 876 simple_unlock(&vp->v_interlock); 877 return (0); 878} 879 880/* 881 * Vnode put/release. 882 * If count drops to zero, call inactive routine and return to freelist. 883 */ 884void 885vrele(vp) 886 struct vnode *vp; 887{ 888 struct proc *p = curproc; /* XXX */ 889 890#ifdef DIAGNOSTIC 891 if (vp == NULL) 892 panic("vrele: null vp"); 893#endif 894 simple_lock(&vp->v_interlock); 895 896 if (vp->v_usecount > 1) { 897 898 vp->v_usecount--; 899 simple_unlock(&vp->v_interlock); 900 901 } else if (vp->v_usecount == 1) { 902 903 vp->v_usecount--; 904 905 if (VSHOULDFREE(vp)) 906 vfree(vp); 907 /* 908 * If we are doing a vput, the node is already locked, and we must 909 * call VOP_INACTIVE with the node locked. So, in the case of 910 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 911 */ 912 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 913 VOP_INACTIVE(vp, p); 914 } 915 916 } else { 917#ifdef DIAGNOSTIC 918 vprint("vrele: negative ref count", vp); 919#endif 920 panic("vrele: negative ref cnt"); 921 } 922} 923 924void 925vput(vp) 926 struct vnode *vp; 927{ 928 struct proc *p = curproc; /* XXX */ 929 930#ifdef DIAGNOSTIC 931 if (vp == NULL) 932 panic("vput: null vp"); 933#endif 934 simple_lock(&vp->v_interlock); 935 936 if (vp->v_usecount > 1) { 937 938 vp->v_usecount--; 939 VOP_UNLOCK(vp, LK_INTERLOCK, p); 940 941 } else if (vp->v_usecount == 1) { 942 943 vp->v_usecount--; 944 945 if (VSHOULDFREE(vp)) 946 vfree(vp); 947 /* 948 * If we are doing a vput, the node is already locked, and we must 949 * call VOP_INACTIVE with the node locked. So, in the case of 950 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 951 */ 952 simple_unlock(&vp->v_interlock); 953 VOP_INACTIVE(vp, p); 954 955 } else { 956#ifdef DIAGNOSTIC 957 vprint("vput: negative ref count", vp); 958#endif 959 panic("vput: negative ref cnt"); 960 } 961} 962 963/* 964 * Somebody doesn't want the vnode recycled. 965 */ 966void 967vhold(vp) 968 register struct vnode *vp; 969{ 970 971 simple_lock(&vp->v_interlock); 972 vp->v_holdcnt++; 973 if (VSHOULDBUSY(vp)) 974 vbusy(vp); 975 simple_unlock(&vp->v_interlock); 976} 977 978/* 979 * One less who cares about this vnode. 980 */ 981void 982vdrop(vp) 983 register struct vnode *vp; 984{ 985 986 simple_lock(&vp->v_interlock); 987 if (vp->v_holdcnt <= 0) 988 panic("holdrele: holdcnt"); 989 vp->v_holdcnt--; 990 if (VSHOULDFREE(vp)) 991 vfree(vp); 992 simple_unlock(&vp->v_interlock); 993} 994 995/* 996 * Remove any vnodes in the vnode table belonging to mount point mp. 997 * 998 * If MNT_NOFORCE is specified, there should not be any active ones, 999 * return error if any are found (nb: this is a user error, not a 1000 * system error). If MNT_FORCE is specified, detach any active vnodes 1001 * that are found. 1002 */ 1003#ifdef DIAGNOSTIC 1004static int busyprt = 0; /* print out busy vnodes */ 1005SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1006#endif 1007 1008int 1009vflush(mp, skipvp, flags) 1010 struct mount *mp; 1011 struct vnode *skipvp; 1012 int flags; 1013{ 1014 struct proc *p = curproc; /* XXX */ 1015 struct vnode *vp, *nvp; 1016 int busy = 0; 1017 1018 simple_lock(&mntvnode_slock); 1019loop: 1020 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 1021 /* 1022 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1023 * Start over if it has (it won't be on the list anymore). 1024 */ 1025 if (vp->v_mount != mp) 1026 goto loop; 1027 nvp = vp->v_mntvnodes.le_next; 1028 /* 1029 * Skip over a selected vnode. 1030 */ 1031 if (vp == skipvp) 1032 continue; 1033 1034 simple_lock(&vp->v_interlock); 1035 /* 1036 * Skip over a vnodes marked VSYSTEM. 1037 */ 1038 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1039 simple_unlock(&vp->v_interlock); 1040 continue; 1041 } 1042 /* 1043 * If WRITECLOSE is set, only flush out regular file vnodes 1044 * open for writing. 1045 */ 1046 if ((flags & WRITECLOSE) && 1047 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1048 simple_unlock(&vp->v_interlock); 1049 continue; 1050 } 1051 1052 /* 1053 * With v_usecount == 0, all we need to do is clear out the 1054 * vnode data structures and we are done. 1055 */ 1056 if (vp->v_usecount == 0) { 1057 simple_unlock(&mntvnode_slock); 1058 vgonel(vp, p); 1059 simple_lock(&mntvnode_slock); 1060 continue; 1061 } 1062 1063 /* 1064 * If FORCECLOSE is set, forcibly close the vnode. For block 1065 * or character devices, revert to an anonymous device. For 1066 * all other files, just kill them. 1067 */ 1068 if (flags & FORCECLOSE) { 1069 simple_unlock(&mntvnode_slock); 1070 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1071 vgonel(vp, p); 1072 } else { 1073 vclean(vp, 0, p); 1074 vp->v_op = spec_vnodeop_p; 1075 insmntque(vp, (struct mount *) 0); 1076 } 1077 simple_lock(&mntvnode_slock); 1078 continue; 1079 } 1080#ifdef DIAGNOSTIC 1081 if (busyprt) 1082 vprint("vflush: busy vnode", vp); 1083#endif 1084 simple_unlock(&vp->v_interlock); 1085 busy++; 1086 } 1087 simple_unlock(&mntvnode_slock); 1088 if (busy) 1089 return (EBUSY); 1090 return (0); 1091} 1092 1093/* 1094 * Disassociate the underlying file system from a vnode. 1095 */ 1096static void 1097vclean(vp, flags, p) 1098 struct vnode *vp; 1099 int flags; 1100 struct proc *p; 1101{ 1102 int active, irefed; 1103 vm_object_t object; 1104 1105 /* 1106 * Check to see if the vnode is in use. If so we have to reference it 1107 * before we clean it out so that its count cannot fall to zero and 1108 * generate a race against ourselves to recycle it. 1109 */ 1110 if ((active = vp->v_usecount)) 1111 vp->v_usecount++; 1112 /* 1113 * Prevent the vnode from being recycled or brought into use while we 1114 * clean it out. 1115 */ 1116 if (vp->v_flag & VXLOCK) 1117 panic("vclean: deadlock"); 1118 vp->v_flag |= VXLOCK; 1119 /* 1120 * Even if the count is zero, the VOP_INACTIVE routine may still 1121 * have the object locked while it cleans it out. The VOP_LOCK 1122 * ensures that the VOP_INACTIVE routine is done with its work. 1123 * For active vnodes, it ensures that no other activity can 1124 * occur while the underlying object is being cleaned out. 1125 */ 1126 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1127 1128 object = vp->v_object; 1129 1130 /* 1131 * Clean out any buffers associated with the vnode. 1132 */ 1133 if (flags & DOCLOSE) 1134 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1135 1136 if (vp->v_object && (vp->v_object->flags & OBJ_VFS_REF)) { 1137 vp->v_object->flags &= ~OBJ_VFS_REF; 1138 vm_object_deallocate(object); 1139 } 1140 1141 /* 1142 * If purging an active vnode, it must be closed and 1143 * deactivated before being reclaimed. Note that the 1144 * VOP_INACTIVE will unlock the vnode. 1145 */ 1146 if (active) { 1147 if (flags & DOCLOSE) 1148 VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); 1149 VOP_INACTIVE(vp, p); 1150 } else { 1151 /* 1152 * Any other processes trying to obtain this lock must first 1153 * wait for VXLOCK to clear, then call the new lock operation. 1154 */ 1155 VOP_UNLOCK(vp, 0, p); 1156 } 1157 /* 1158 * Reclaim the vnode. 1159 */ 1160 if (VOP_RECLAIM(vp, p)) 1161 panic("vclean: cannot reclaim"); 1162 if (active) 1163 vrele(vp); 1164 cache_purge(vp); 1165 if (vp->v_vnlock) { 1166#if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */ 1167#ifdef DIAGNOSTIC 1168 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0) 1169 vprint("vclean: lock not drained", vp); 1170#endif 1171#endif 1172 FREE(vp->v_vnlock, M_VNODE); 1173 vp->v_vnlock = NULL; 1174 } 1175 1176 /* 1177 * Done with purge, notify sleepers of the grim news. 1178 */ 1179 vp->v_op = dead_vnodeop_p; 1180 vn_pollgone(vp); 1181 vp->v_tag = VT_NON; 1182 vp->v_flag &= ~VXLOCK; 1183 if (vp->v_flag & VXWANT) { 1184 vp->v_flag &= ~VXWANT; 1185 wakeup((caddr_t) vp); 1186 } 1187} 1188 1189/* 1190 * Eliminate all activity associated with the requested vnode 1191 * and with all vnodes aliased to the requested vnode. 1192 */ 1193int 1194vop_revoke(ap) 1195 struct vop_revoke_args /* { 1196 struct vnode *a_vp; 1197 int a_flags; 1198 } */ *ap; 1199{ 1200 struct vnode *vp, *vq; 1201 struct proc *p = curproc; /* XXX */ 1202 1203#ifdef DIAGNOSTIC 1204 if ((ap->a_flags & REVOKEALL) == 0) 1205 panic("vop_revoke"); 1206#endif 1207 1208 vp = ap->a_vp; 1209 simple_lock(&vp->v_interlock); 1210 1211 if (vp->v_flag & VALIASED) { 1212 /* 1213 * If a vgone (or vclean) is already in progress, 1214 * wait until it is done and return. 1215 */ 1216 if (vp->v_flag & VXLOCK) { 1217 vp->v_flag |= VXWANT; 1218 simple_unlock(&vp->v_interlock); 1219 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1220 return (0); 1221 } 1222 /* 1223 * Ensure that vp will not be vgone'd while we 1224 * are eliminating its aliases. 1225 */ 1226 vp->v_flag |= VXLOCK; 1227 simple_unlock(&vp->v_interlock); 1228 while (vp->v_flag & VALIASED) { 1229 simple_lock(&spechash_slock); 1230 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1231 if (vq->v_rdev != vp->v_rdev || 1232 vq->v_type != vp->v_type || vp == vq) 1233 continue; 1234 simple_unlock(&spechash_slock); 1235 vgone(vq); 1236 break; 1237 } 1238 if (vq == NULLVP) { 1239 simple_unlock(&spechash_slock); 1240 } 1241 } 1242 /* 1243 * Remove the lock so that vgone below will 1244 * really eliminate the vnode after which time 1245 * vgone will awaken any sleepers. 1246 */ 1247 simple_lock(&vp->v_interlock); 1248 vp->v_flag &= ~VXLOCK; 1249 } 1250 vgonel(vp, p); 1251 return (0); 1252} 1253 1254/* 1255 * Recycle an unused vnode to the front of the free list. 1256 * Release the passed interlock if the vnode will be recycled. 1257 */ 1258int 1259vrecycle(vp, inter_lkp, p) 1260 struct vnode *vp; 1261 struct simplelock *inter_lkp; 1262 struct proc *p; 1263{ 1264 1265 simple_lock(&vp->v_interlock); 1266 if (vp->v_usecount == 0) { 1267 if (inter_lkp) { 1268 simple_unlock(inter_lkp); 1269 } 1270 vgonel(vp, p); 1271 return (1); 1272 } 1273 simple_unlock(&vp->v_interlock); 1274 return (0); 1275} 1276 1277/* 1278 * Eliminate all activity associated with a vnode 1279 * in preparation for reuse. 1280 */ 1281void 1282vgone(vp) 1283 register struct vnode *vp; 1284{ 1285 struct proc *p = curproc; /* XXX */ 1286 1287 simple_lock(&vp->v_interlock); 1288 vgonel(vp, p); 1289} 1290 1291/* 1292 * vgone, with the vp interlock held. 1293 */ 1294static void 1295vgonel(vp, p) 1296 struct vnode *vp; 1297 struct proc *p; 1298{ 1299 struct vnode *vq; 1300 struct vnode *vx; 1301 1302 /* 1303 * If a vgone (or vclean) is already in progress, 1304 * wait until it is done and return. 1305 */ 1306 if (vp->v_flag & VXLOCK) { 1307 vp->v_flag |= VXWANT; 1308 simple_unlock(&vp->v_interlock); 1309 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1310 return; 1311 } 1312 1313 if (vp->v_object) { 1314 vp->v_object->flags |= OBJ_VNODE_GONE; 1315 } 1316 1317 /* 1318 * Clean out the filesystem specific data. 1319 */ 1320 vclean(vp, DOCLOSE, p); 1321 1322 /* 1323 * Delete from old mount point vnode list, if on one. 1324 */ 1325 if (vp->v_mount != NULL) 1326 insmntque(vp, (struct mount *)0); 1327 /* 1328 * If special device, remove it from special device alias list 1329 * if it is on one. 1330 */ 1331 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1332 simple_lock(&spechash_slock); 1333 if (*vp->v_hashchain == vp) { 1334 *vp->v_hashchain = vp->v_specnext; 1335 } else { 1336 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1337 if (vq->v_specnext != vp) 1338 continue; 1339 vq->v_specnext = vp->v_specnext; 1340 break; 1341 } 1342 if (vq == NULL) 1343 panic("missing bdev"); 1344 } 1345 if (vp->v_flag & VALIASED) { 1346 vx = NULL; 1347 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1348 if (vq->v_rdev != vp->v_rdev || 1349 vq->v_type != vp->v_type) 1350 continue; 1351 if (vx) 1352 break; 1353 vx = vq; 1354 } 1355 if (vx == NULL) 1356 panic("missing alias"); 1357 if (vq == NULL) 1358 vx->v_flag &= ~VALIASED; 1359 vp->v_flag &= ~VALIASED; 1360 } 1361 simple_unlock(&spechash_slock); 1362 FREE(vp->v_specinfo, M_VNODE); 1363 vp->v_specinfo = NULL; 1364 } 1365 1366 /* 1367 * If it is on the freelist and not already at the head, 1368 * move it to the head of the list. The test of the back 1369 * pointer and the reference count of zero is because 1370 * it will be removed from the free list by getnewvnode, 1371 * but will not have its reference count incremented until 1372 * after calling vgone. If the reference count were 1373 * incremented first, vgone would (incorrectly) try to 1374 * close the previous instance of the underlying object. 1375 */ 1376 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1377 simple_lock(&vnode_free_list_slock); 1378 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1379 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1380 simple_unlock(&vnode_free_list_slock); 1381 } 1382 1383 vp->v_type = VBAD; 1384} 1385 1386/* 1387 * Lookup a vnode by device number. 1388 */ 1389int 1390vfinddev(dev, type, vpp) 1391 dev_t dev; 1392 enum vtype type; 1393 struct vnode **vpp; 1394{ 1395 register struct vnode *vp; 1396 int rc = 0; 1397 1398 simple_lock(&spechash_slock); 1399 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1400 if (dev != vp->v_rdev || type != vp->v_type) 1401 continue; 1402 *vpp = vp; 1403 rc = 1; 1404 break; 1405 } 1406 simple_unlock(&spechash_slock); 1407 return (rc); 1408} 1409 1410/* 1411 * Calculate the total number of references to a special device. 1412 */ 1413int 1414vcount(vp) 1415 register struct vnode *vp; 1416{ 1417 struct vnode *vq, *vnext; 1418 int count; 1419 1420loop: 1421 if ((vp->v_flag & VALIASED) == 0) 1422 return (vp->v_usecount); 1423 simple_lock(&spechash_slock); 1424 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1425 vnext = vq->v_specnext; 1426 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1427 continue; 1428 /* 1429 * Alias, but not in use, so flush it out. 1430 */ 1431 if (vq->v_usecount == 0 && vq != vp) { 1432 simple_unlock(&spechash_slock); 1433 vgone(vq); 1434 goto loop; 1435 } 1436 count += vq->v_usecount; 1437 } 1438 simple_unlock(&spechash_slock); 1439 return (count); 1440} 1441/* 1442 * Print out a description of a vnode. 1443 */ 1444static char *typename[] = 1445{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1446 1447void 1448vprint(label, vp) 1449 char *label; 1450 register struct vnode *vp; 1451{ 1452 char buf[64]; 1453 1454 if (label != NULL) 1455 printf("%s: %x: ", label, vp); 1456 else 1457 printf("%x: ", vp); 1458 printf("type %s, usecount %d, writecount %d, refcount %ld,", 1459 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1460 vp->v_holdcnt); 1461 buf[0] = '\0'; 1462 if (vp->v_flag & VROOT) 1463 strcat(buf, "|VROOT"); 1464 if (vp->v_flag & VTEXT) 1465 strcat(buf, "|VTEXT"); 1466 if (vp->v_flag & VSYSTEM) 1467 strcat(buf, "|VSYSTEM"); 1468 if (vp->v_flag & VXLOCK) 1469 strcat(buf, "|VXLOCK"); 1470 if (vp->v_flag & VXWANT) 1471 strcat(buf, "|VXWANT"); 1472 if (vp->v_flag & VBWAIT) 1473 strcat(buf, "|VBWAIT"); 1474 if (vp->v_flag & VALIASED) 1475 strcat(buf, "|VALIASED"); 1476 if (vp->v_flag & VDOOMED) 1477 strcat(buf, "|VDOOMED"); 1478 if (vp->v_flag & VFREE) 1479 strcat(buf, "|VFREE"); 1480 if (buf[0] != '\0') 1481 printf(" flags (%s)", &buf[1]); 1482 if (vp->v_data == NULL) { 1483 printf("\n"); 1484 } else { 1485 printf("\n\t"); 1486 VOP_PRINT(vp); 1487 } 1488} 1489 1490#ifdef DDB 1491/* 1492 * List all of the locked vnodes in the system. 1493 * Called when debugging the kernel. 1494 */ 1495static void 1496printlockedvnodes() 1497{ 1498 struct proc *p = curproc; /* XXX */ 1499 struct mount *mp, *nmp; 1500 struct vnode *vp; 1501 1502 printf("Locked vnodes\n"); 1503 simple_lock(&mountlist_slock); 1504 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1505 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1506 nmp = mp->mnt_list.cqe_next; 1507 continue; 1508 } 1509 for (vp = mp->mnt_vnodelist.lh_first; 1510 vp != NULL; 1511 vp = vp->v_mntvnodes.le_next) { 1512 if (VOP_ISLOCKED(vp)) 1513 vprint((char *)0, vp); 1514 } 1515 simple_lock(&mountlist_slock); 1516 nmp = mp->mnt_list.cqe_next; 1517 vfs_unbusy(mp, p); 1518 } 1519 simple_unlock(&mountlist_slock); 1520} 1521#endif 1522 1523/* 1524 * Top level filesystem related information gathering. 1525 */ 1526static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS); 1527 1528static int 1529vfs_sysctl SYSCTL_HANDLER_ARGS 1530{ 1531 int *name = (int *)arg1 - 1; /* XXX */ 1532 u_int namelen = arg2 + 1; /* XXX */ 1533 struct vfsconf *vfsp; 1534 1535#ifndef NO_COMPAT_PRELITE2 1536 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1537 if (namelen == 1) 1538 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1539#endif 1540 1541#ifdef notyet 1542 /* all sysctl names at this level are at least name and field */ 1543 if (namelen < 2) 1544 return (ENOTDIR); /* overloaded */ 1545 if (name[0] != VFS_GENERIC) { 1546 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1547 if (vfsp->vfc_typenum == name[0]) 1548 break; 1549 if (vfsp == NULL) 1550 return (EOPNOTSUPP); 1551 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1552 oldp, oldlenp, newp, newlen, p)); 1553 } 1554#endif 1555 switch (name[1]) { 1556 case VFS_MAXTYPENUM: 1557 if (namelen != 2) 1558 return (ENOTDIR); 1559 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1560 case VFS_CONF: 1561 if (namelen != 3) 1562 return (ENOTDIR); /* overloaded */ 1563 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1564 if (vfsp->vfc_typenum == name[2]) 1565 break; 1566 if (vfsp == NULL) 1567 return (EOPNOTSUPP); 1568 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1569 } 1570 return (EOPNOTSUPP); 1571} 1572 1573SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1574 "Generic filesystem"); 1575 1576#ifndef NO_COMPAT_PRELITE2 1577 1578static int 1579sysctl_ovfs_conf SYSCTL_HANDLER_ARGS 1580{ 1581 int error; 1582 struct vfsconf *vfsp; 1583 struct ovfsconf ovfs; 1584 1585 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1586 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1587 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1588 ovfs.vfc_index = vfsp->vfc_typenum; 1589 ovfs.vfc_refcount = vfsp->vfc_refcount; 1590 ovfs.vfc_flags = vfsp->vfc_flags; 1591 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1592 if (error) 1593 return error; 1594 } 1595 return 0; 1596} 1597 1598#endif /* !NO_COMPAT_PRELITE2 */ 1599 1600static volatile int kinfo_vdebug = 1; 1601 1602#if 0 1603#define KINFO_VNODESLOP 10 1604/* 1605 * Dump vnode list (via sysctl). 1606 * Copyout address of vnode followed by vnode. 1607 */ 1608/* ARGSUSED */ 1609static int 1610sysctl_vnode SYSCTL_HANDLER_ARGS 1611{ 1612 struct proc *p = curproc; /* XXX */ 1613 struct mount *mp, *nmp; 1614 struct vnode *nvp, *vp; 1615 int error; 1616 1617#define VPTRSZ sizeof (struct vnode *) 1618#define VNODESZ sizeof (struct vnode) 1619 1620 req->lock = 0; 1621 if (!req->oldptr) /* Make an estimate */ 1622 return (SYSCTL_OUT(req, 0, 1623 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 1624 1625 simple_lock(&mountlist_slock); 1626 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1627 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1628 nmp = mp->mnt_list.cqe_next; 1629 continue; 1630 } 1631again: 1632 simple_lock(&mntvnode_slock); 1633 for (vp = mp->mnt_vnodelist.lh_first; 1634 vp != NULL; 1635 vp = nvp) { 1636 /* 1637 * Check that the vp is still associated with 1638 * this filesystem. RACE: could have been 1639 * recycled onto the same filesystem. 1640 */ 1641 if (vp->v_mount != mp) { 1642 simple_unlock(&mntvnode_slock); 1643 if (kinfo_vdebug) 1644 printf("kinfo: vp changed\n"); 1645 goto again; 1646 } 1647 nvp = vp->v_mntvnodes.le_next; 1648 simple_unlock(&mntvnode_slock); 1649 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 1650 (error = SYSCTL_OUT(req, vp, VNODESZ))) 1651 return (error); 1652 simple_lock(&mntvnode_slock); 1653 } 1654 simple_unlock(&mntvnode_slock); 1655 simple_lock(&mountlist_slock); 1656 nmp = mp->mnt_list.cqe_next; 1657 vfs_unbusy(mp, p); 1658 } 1659 simple_unlock(&mountlist_slock); 1660 1661 return (0); 1662} 1663#endif 1664 1665/* 1666 * XXX 1667 * Exporting the vnode list on large systems causes them to crash. 1668 * Exporting the vnode list on medium systems causes sysctl to coredump. 1669 */ 1670#if 0 1671SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 1672 0, 0, sysctl_vnode, "S,vnode", ""); 1673#endif 1674 1675/* 1676 * Check to see if a filesystem is mounted on a block device. 1677 */ 1678int 1679vfs_mountedon(vp) 1680 struct vnode *vp; 1681{ 1682 struct vnode *vq; 1683 int error = 0; 1684 1685 if (vp->v_specflags & SI_MOUNTEDON) 1686 return (EBUSY); 1687 if (vp->v_flag & VALIASED) { 1688 simple_lock(&spechash_slock); 1689 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1690 if (vq->v_rdev != vp->v_rdev || 1691 vq->v_type != vp->v_type) 1692 continue; 1693 if (vq->v_specflags & SI_MOUNTEDON) { 1694 error = EBUSY; 1695 break; 1696 } 1697 } 1698 simple_unlock(&spechash_slock); 1699 } 1700 return (error); 1701} 1702 1703/* 1704 * Unmount all filesystems. The list is traversed in reverse order 1705 * of mounting to avoid dependencies. 1706 */ 1707void 1708vfs_unmountall() 1709{ 1710 struct mount *mp, *nmp; 1711 struct proc *p = initproc; /* XXX XXX should this be proc0? */ 1712 int error; 1713 1714 /* 1715 * Since this only runs when rebooting, it is not interlocked. 1716 */ 1717 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 1718 nmp = mp->mnt_list.cqe_prev; 1719 error = dounmount(mp, MNT_FORCE, p); 1720 if (error) { 1721 printf("unmount of %s failed (", 1722 mp->mnt_stat.f_mntonname); 1723 if (error == EBUSY) 1724 printf("BUSY)\n"); 1725 else 1726 printf("%d)\n", error); 1727 } 1728 } 1729} 1730 1731/* 1732 * Build hash lists of net addresses and hang them off the mount point. 1733 * Called by ufs_mount() to set up the lists of export addresses. 1734 */ 1735static int 1736vfs_hang_addrlist(mp, nep, argp) 1737 struct mount *mp; 1738 struct netexport *nep; 1739 struct export_args *argp; 1740{ 1741 register struct netcred *np; 1742 register struct radix_node_head *rnh; 1743 register int i; 1744 struct radix_node *rn; 1745 struct sockaddr *saddr, *smask = 0; 1746 struct domain *dom; 1747 int error; 1748 1749 if (argp->ex_addrlen == 0) { 1750 if (mp->mnt_flag & MNT_DEFEXPORTED) 1751 return (EPERM); 1752 np = &nep->ne_defexported; 1753 np->netc_exflags = argp->ex_flags; 1754 np->netc_anon = argp->ex_anon; 1755 np->netc_anon.cr_ref = 1; 1756 mp->mnt_flag |= MNT_DEFEXPORTED; 1757 return (0); 1758 } 1759 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1760 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1761 bzero((caddr_t) np, i); 1762 saddr = (struct sockaddr *) (np + 1); 1763 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1764 goto out; 1765 if (saddr->sa_len > argp->ex_addrlen) 1766 saddr->sa_len = argp->ex_addrlen; 1767 if (argp->ex_masklen) { 1768 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 1769 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 1770 if (error) 1771 goto out; 1772 if (smask->sa_len > argp->ex_masklen) 1773 smask->sa_len = argp->ex_masklen; 1774 } 1775 i = saddr->sa_family; 1776 if ((rnh = nep->ne_rtable[i]) == 0) { 1777 /* 1778 * Seems silly to initialize every AF when most are not used, 1779 * do so on demand here 1780 */ 1781 for (dom = domains; dom; dom = dom->dom_next) 1782 if (dom->dom_family == i && dom->dom_rtattach) { 1783 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1784 dom->dom_rtoffset); 1785 break; 1786 } 1787 if ((rnh = nep->ne_rtable[i]) == 0) { 1788 error = ENOBUFS; 1789 goto out; 1790 } 1791 } 1792 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 1793 np->netc_rnodes); 1794 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1795 error = EPERM; 1796 goto out; 1797 } 1798 np->netc_exflags = argp->ex_flags; 1799 np->netc_anon = argp->ex_anon; 1800 np->netc_anon.cr_ref = 1; 1801 return (0); 1802out: 1803 free(np, M_NETADDR); 1804 return (error); 1805} 1806 1807/* ARGSUSED */ 1808static int 1809vfs_free_netcred(rn, w) 1810 struct radix_node *rn; 1811 void *w; 1812{ 1813 register struct radix_node_head *rnh = (struct radix_node_head *) w; 1814 1815 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1816 free((caddr_t) rn, M_NETADDR); 1817 return (0); 1818} 1819 1820/* 1821 * Free the net address hash lists that are hanging off the mount points. 1822 */ 1823static void 1824vfs_free_addrlist(nep) 1825 struct netexport *nep; 1826{ 1827 register int i; 1828 register struct radix_node_head *rnh; 1829 1830 for (i = 0; i <= AF_MAX; i++) 1831 if ((rnh = nep->ne_rtable[i])) { 1832 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1833 (caddr_t) rnh); 1834 free((caddr_t) rnh, M_RTABLE); 1835 nep->ne_rtable[i] = 0; 1836 } 1837} 1838 1839int 1840vfs_export(mp, nep, argp) 1841 struct mount *mp; 1842 struct netexport *nep; 1843 struct export_args *argp; 1844{ 1845 int error; 1846 1847 if (argp->ex_flags & MNT_DELEXPORT) { 1848 if (mp->mnt_flag & MNT_EXPUBLIC) { 1849 vfs_setpublicfs(NULL, NULL, NULL); 1850 mp->mnt_flag &= ~MNT_EXPUBLIC; 1851 } 1852 vfs_free_addrlist(nep); 1853 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1854 } 1855 if (argp->ex_flags & MNT_EXPORTED) { 1856 if (argp->ex_flags & MNT_EXPUBLIC) { 1857 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1858 return (error); 1859 mp->mnt_flag |= MNT_EXPUBLIC; 1860 } 1861 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1862 return (error); 1863 mp->mnt_flag |= MNT_EXPORTED; 1864 } 1865 return (0); 1866} 1867 1868 1869/* 1870 * Set the publicly exported filesystem (WebNFS). Currently, only 1871 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1872 */ 1873int 1874vfs_setpublicfs(mp, nep, argp) 1875 struct mount *mp; 1876 struct netexport *nep; 1877 struct export_args *argp; 1878{ 1879 int error; 1880 struct vnode *rvp; 1881 char *cp; 1882 1883 /* 1884 * mp == NULL -> invalidate the current info, the FS is 1885 * no longer exported. May be called from either vfs_export 1886 * or unmount, so check if it hasn't already been done. 1887 */ 1888 if (mp == NULL) { 1889 if (nfs_pub.np_valid) { 1890 nfs_pub.np_valid = 0; 1891 if (nfs_pub.np_index != NULL) { 1892 FREE(nfs_pub.np_index, M_TEMP); 1893 nfs_pub.np_index = NULL; 1894 } 1895 } 1896 return (0); 1897 } 1898 1899 /* 1900 * Only one allowed at a time. 1901 */ 1902 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1903 return (EBUSY); 1904 1905 /* 1906 * Get real filehandle for root of exported FS. 1907 */ 1908 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1909 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1910 1911 if ((error = VFS_ROOT(mp, &rvp))) 1912 return (error); 1913 1914 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1915 return (error); 1916 1917 vput(rvp); 1918 1919 /* 1920 * If an indexfile was specified, pull it in. 1921 */ 1922 if (argp->ex_indexfile != NULL) { 1923 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 1924 M_WAITOK); 1925 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1926 MAXNAMLEN, (size_t *)0); 1927 if (!error) { 1928 /* 1929 * Check for illegal filenames. 1930 */ 1931 for (cp = nfs_pub.np_index; *cp; cp++) { 1932 if (*cp == '/') { 1933 error = EINVAL; 1934 break; 1935 } 1936 } 1937 } 1938 if (error) { 1939 FREE(nfs_pub.np_index, M_TEMP); 1940 return (error); 1941 } 1942 } 1943 1944 nfs_pub.np_mount = mp; 1945 nfs_pub.np_valid = 1; 1946 return (0); 1947} 1948 1949struct netcred * 1950vfs_export_lookup(mp, nep, nam) 1951 register struct mount *mp; 1952 struct netexport *nep; 1953 struct sockaddr *nam; 1954{ 1955 register struct netcred *np; 1956 register struct radix_node_head *rnh; 1957 struct sockaddr *saddr; 1958 1959 np = NULL; 1960 if (mp->mnt_flag & MNT_EXPORTED) { 1961 /* 1962 * Lookup in the export list first. 1963 */ 1964 if (nam != NULL) { 1965 saddr = nam; 1966 rnh = nep->ne_rtable[saddr->sa_family]; 1967 if (rnh != NULL) { 1968 np = (struct netcred *) 1969 (*rnh->rnh_matchaddr)((caddr_t)saddr, 1970 rnh); 1971 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1972 np = NULL; 1973 } 1974 } 1975 /* 1976 * If no address match, use the default if it exists. 1977 */ 1978 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1979 np = &nep->ne_defexported; 1980 } 1981 return (np); 1982} 1983 1984/* 1985 * perform msync on all vnodes under a mount point 1986 * the mount point must be locked. 1987 */ 1988void 1989vfs_msync(struct mount *mp, int flags) { 1990 struct vnode *vp, *nvp; 1991loop: 1992 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 1993 1994 if (vp->v_mount != mp) 1995 goto loop; 1996 nvp = vp->v_mntvnodes.le_next; 1997 if (VOP_ISLOCKED(vp) && (flags != MNT_WAIT)) 1998 continue; 1999 if (vp->v_object && 2000 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2001 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curproc); 2002 vm_object_page_clean(vp->v_object, 0, 0, TRUE); 2003 VOP_UNLOCK(vp, 0, curproc); 2004 } 2005 } 2006} 2007 2008/* 2009 * Create the VM object needed for VMIO and mmap support. This 2010 * is done for all VREG files in the system. Some filesystems might 2011 * afford the additional metadata buffering capability of the 2012 * VMIO code by making the device node be VMIO mode also. 2013 */ 2014int 2015vfs_object_create(vp, p, cred, waslocked) 2016 struct vnode *vp; 2017 struct proc *p; 2018 struct ucred *cred; 2019 int waslocked; 2020{ 2021 struct vattr vat; 2022 vm_object_t object; 2023 int error = 0; 2024 2025 if ((vp->v_type != VREG) && (vp->v_type != VBLK)) 2026 return 0; 2027 2028retry: 2029 if ((object = vp->v_object) == NULL) { 2030 if (vp->v_type == VREG) { 2031 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2032 goto retn; 2033 (void) vnode_pager_alloc(vp, 2034 OFF_TO_IDX(round_page(vat.va_size)), 0, 0); 2035 vp->v_object->flags |= OBJ_VFS_REF; 2036 } else { 2037 /* 2038 * This simply allocates the biggest object possible 2039 * for a VBLK vnode. This should be fixed, but doesn't 2040 * cause any problems (yet). 2041 */ 2042 (void) vnode_pager_alloc(vp, INT_MAX, 0, 0); 2043 vp->v_object->flags |= OBJ_VFS_REF; 2044 } 2045 } else { 2046 if (object->flags & OBJ_DEAD) { 2047 if (waslocked) 2048 VOP_UNLOCK(vp, 0, p); 2049 tsleep(object, PVM, "vodead", 0); 2050 if (waslocked) 2051 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2052 goto retry; 2053 } 2054 if ((object->flags & OBJ_VFS_REF) == 0) { 2055 vm_object_reference(object); 2056 object->flags |= OBJ_VFS_REF; 2057 } 2058 } 2059 if (vp->v_object) 2060 vp->v_flag |= VVMIO; 2061 2062retn: 2063 return error; 2064} 2065 2066static void 2067vfree(vp) 2068 struct vnode *vp; 2069{ 2070 simple_lock(&vnode_free_list_slock); 2071 if (vp->v_flag & VAGE) { 2072 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2073 } else { 2074 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2075 } 2076 freevnodes++; 2077 simple_unlock(&vnode_free_list_slock); 2078 vp->v_flag &= ~VAGE; 2079 vp->v_flag |= VFREE; 2080} 2081 2082static void 2083vbusy(vp) 2084 struct vnode *vp; 2085{ 2086 simple_lock(&vnode_free_list_slock); 2087 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2088 freevnodes--; 2089 simple_unlock(&vnode_free_list_slock); 2090 vp->v_flag &= ~VFREE; 2091} 2092 2093/* 2094 * Record a process's interest in events which might happen to 2095 * a vnode. Because poll uses the historic select-style interface 2096 * internally, this routine serves as both the ``check for any 2097 * pending events'' and the ``record my interest in future events'' 2098 * functions. (These are done together, while the lock is held, 2099 * to avoid race conditions.) 2100 */ 2101int 2102vn_pollrecord(vp, p, events) 2103 struct vnode *vp; 2104 struct proc *p; 2105 short events; 2106{ 2107 simple_lock(&vp->v_pollinfo.vpi_lock); 2108 if (vp->v_pollinfo.vpi_revents & events) { 2109 /* 2110 * This leaves events we are not interested 2111 * in available for the other process which 2112 * which presumably had requested them 2113 * (otherwise they would never have been 2114 * recorded). 2115 */ 2116 events &= vp->v_pollinfo.vpi_revents; 2117 vp->v_pollinfo.vpi_revents &= ~events; 2118 2119 simple_unlock(&vp->v_pollinfo.vpi_lock); 2120 return events; 2121 } 2122 vp->v_pollinfo.vpi_events |= events; 2123 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2124 simple_unlock(&vp->v_pollinfo.vpi_lock); 2125 return 0; 2126} 2127 2128/* 2129 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2130 * it is possible for us to miss an event due to race conditions, but 2131 * that condition is expected to be rare, so for the moment it is the 2132 * preferred interface. 2133 */ 2134void 2135vn_pollevent(vp, events) 2136 struct vnode *vp; 2137 short events; 2138{ 2139 simple_lock(&vp->v_pollinfo.vpi_lock); 2140 if (vp->v_pollinfo.vpi_events & events) { 2141 /* 2142 * We clear vpi_events so that we don't 2143 * call selwakeup() twice if two events are 2144 * posted before the polling process(es) is 2145 * awakened. This also ensures that we take at 2146 * most one selwakeup() if the polling process 2147 * is no longer interested. However, it does 2148 * mean that only one event can be noticed at 2149 * a time. (Perhaps we should only clear those 2150 * event bits which we note?) XXX 2151 */ 2152 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2153 vp->v_pollinfo.vpi_revents |= events; 2154 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2155 } 2156 simple_unlock(&vp->v_pollinfo.vpi_lock); 2157} 2158 2159/* 2160 * Wake up anyone polling on vp because it is being revoked. 2161 * This depends on dead_poll() returning POLLHUP for correct 2162 * behavior. 2163 */ 2164void 2165vn_pollgone(vp) 2166 struct vnode *vp; 2167{ 2168 simple_lock(&vp->v_pollinfo.vpi_lock); 2169 if (vp->v_pollinfo.vpi_events) { 2170 vp->v_pollinfo.vpi_events = 0; 2171 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2172 } 2173 simple_unlock(&vp->v_pollinfo.vpi_lock); 2174} 2175