vfs_export.c revision 32072
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 1.117 1997/12/29 00:22:45 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 111int vfs_ioopt = 0; 112SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 113 114struct mntlist mountlist; /* mounted filesystem list */ 115struct simplelock mountlist_slock; 116static struct simplelock mntid_slock; 117struct simplelock mntvnode_slock; 118struct simplelock vnode_free_list_slock; 119static struct simplelock spechash_slock; 120struct nfs_public nfs_pub; /* publicly exported FS */ 121 122int desiredvnodes; 123SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, ""); 124 125static void vfs_free_addrlist __P((struct netexport *nep)); 126static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 127static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 128 struct export_args *argp)); 129 130/* 131 * Initialize the vnode management data structures. 132 */ 133void 134vntblinit() 135{ 136 137 desiredvnodes = maxproc + cnt.v_page_count / 4; 138 simple_lock_init(&mntvnode_slock); 139 simple_lock_init(&mntid_slock); 140 simple_lock_init(&spechash_slock); 141 TAILQ_INIT(&vnode_free_list); 142 simple_lock_init(&vnode_free_list_slock); 143 CIRCLEQ_INIT(&mountlist); 144} 145 146/* 147 * Mark a mount point as busy. Used to synchronize access and to delay 148 * unmounting. Interlock is not released on failure. 149 */ 150int 151vfs_busy(mp, flags, interlkp, p) 152 struct mount *mp; 153 int flags; 154 struct simplelock *interlkp; 155 struct proc *p; 156{ 157 int lkflags; 158 159 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 160 if (flags & LK_NOWAIT) 161 return (ENOENT); 162 mp->mnt_kern_flag |= MNTK_MWAIT; 163 if (interlkp) { 164 simple_unlock(interlkp); 165 } 166 /* 167 * Since all busy locks are shared except the exclusive 168 * lock granted when unmounting, the only place that a 169 * wakeup needs to be done is at the release of the 170 * exclusive lock at the end of dounmount. 171 */ 172 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 173 if (interlkp) { 174 simple_lock(interlkp); 175 } 176 return (ENOENT); 177 } 178 lkflags = LK_SHARED; 179 if (interlkp) 180 lkflags |= LK_INTERLOCK; 181 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 182 panic("vfs_busy: unexpected lock failure"); 183 return (0); 184} 185 186/* 187 * Free a busy filesystem. 188 */ 189void 190vfs_unbusy(mp, p) 191 struct mount *mp; 192 struct proc *p; 193{ 194 195 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 196} 197 198/* 199 * Lookup a filesystem type, and if found allocate and initialize 200 * a mount structure for it. 201 * 202 * Devname is usually updated by mount(8) after booting. 203 */ 204int 205vfs_rootmountalloc(fstypename, devname, mpp) 206 char *fstypename; 207 char *devname; 208 struct mount **mpp; 209{ 210 struct proc *p = curproc; /* XXX */ 211 struct vfsconf *vfsp; 212 struct mount *mp; 213 214 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 215 if (!strcmp(vfsp->vfc_name, fstypename)) 216 break; 217 if (vfsp == NULL) 218 return (ENODEV); 219 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 220 bzero((char *)mp, (u_long)sizeof(struct mount)); 221 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 222 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 223 LIST_INIT(&mp->mnt_vnodelist); 224 mp->mnt_vfc = vfsp; 225 mp->mnt_op = vfsp->vfc_vfsops; 226 mp->mnt_flag = MNT_RDONLY; 227 mp->mnt_vnodecovered = NULLVP; 228 vfsp->vfc_refcount++; 229 mp->mnt_stat.f_type = vfsp->vfc_typenum; 230 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 231 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 232 mp->mnt_stat.f_mntonname[0] = '/'; 233 mp->mnt_stat.f_mntonname[1] = 0; 234 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 235 *mpp = mp; 236 return (0); 237} 238 239/* 240 * Find an appropriate filesystem to use for the root. If a filesystem 241 * has not been preselected, walk through the list of known filesystems 242 * trying those that have mountroot routines, and try them until one 243 * works or we have tried them all. 244 */ 245#ifdef notdef /* XXX JH */ 246int 247lite2_vfs_mountroot() 248{ 249 struct vfsconf *vfsp; 250 extern int (*lite2_mountroot) __P((void)); 251 int error; 252 253 if (lite2_mountroot != NULL) 254 return ((*lite2_mountroot)()); 255 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 256 if (vfsp->vfc_mountroot == NULL) 257 continue; 258 if ((error = (*vfsp->vfc_mountroot)()) == 0) 259 return (0); 260 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 261 } 262 return (ENODEV); 263} 264#endif 265 266/* 267 * Lookup a mount point by filesystem identifier. 268 */ 269struct mount * 270vfs_getvfs(fsid) 271 fsid_t *fsid; 272{ 273 register struct mount *mp; 274 275 simple_lock(&mountlist_slock); 276 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 277 mp = mp->mnt_list.cqe_next) { 278 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 279 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 280 simple_unlock(&mountlist_slock); 281 return (mp); 282 } 283 } 284 simple_unlock(&mountlist_slock); 285 return ((struct mount *) 0); 286} 287 288/* 289 * Get a new unique fsid 290 */ 291void 292vfs_getnewfsid(mp) 293 struct mount *mp; 294{ 295 static u_short xxxfs_mntid; 296 297 fsid_t tfsid; 298 int mtype; 299 300 simple_lock(&mntid_slock); 301 mtype = mp->mnt_vfc->vfc_typenum; 302 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 303 mp->mnt_stat.f_fsid.val[1] = mtype; 304 if (xxxfs_mntid == 0) 305 ++xxxfs_mntid; 306 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 307 tfsid.val[1] = mtype; 308 if (mountlist.cqh_first != (void *)&mountlist) { 309 while (vfs_getvfs(&tfsid)) { 310 tfsid.val[0]++; 311 xxxfs_mntid++; 312 } 313 } 314 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 315 simple_unlock(&mntid_slock); 316} 317 318/* 319 * Set vnode attributes to VNOVAL 320 */ 321void 322vattr_null(vap) 323 register struct vattr *vap; 324{ 325 326 vap->va_type = VNON; 327 vap->va_size = VNOVAL; 328 vap->va_bytes = VNOVAL; 329 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 330 vap->va_fsid = vap->va_fileid = 331 vap->va_blocksize = vap->va_rdev = 332 vap->va_atime.tv_sec = vap->va_atime.tv_nsec = 333 vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec = 334 vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec = 335 vap->va_flags = vap->va_gen = VNOVAL; 336 vap->va_vaflags = 0; 337} 338 339/* 340 * Routines having to do with the management of the vnode table. 341 */ 342extern vop_t **dead_vnodeop_p; 343 344/* 345 * Return the next vnode from the free list. 346 */ 347int 348getnewvnode(tag, mp, vops, vpp) 349 enum vtagtype tag; 350 struct mount *mp; 351 vop_t **vops; 352 struct vnode **vpp; 353{ 354 struct proc *p = curproc; /* XXX */ 355 struct vnode *vp; 356 357 /* 358 * We take the least recently used vnode from the freelist 359 * if we can get it and it has no cached pages, and no 360 * namecache entries are relative to it. 361 * Otherwise we allocate a new vnode 362 */ 363 364 simple_lock(&vnode_free_list_slock); 365 366 if (wantfreevnodes && freevnodes < wantfreevnodes) { 367 vp = NULL; 368 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) { 369 /* 370 * XXX: this is only here to be backwards compatible 371 */ 372 vp = NULL; 373 } else { 374 TAILQ_FOREACH(vp, &vnode_free_list, v_freelist) { 375 if (!simple_lock_try(&vp->v_interlock)) 376 continue; 377 if (vp->v_usecount) 378 panic("free vnode isn't"); 379 380 if (vp->v_object && vp->v_object->resident_page_count) { 381 /* Don't recycle if it's caching some pages */ 382 simple_unlock(&vp->v_interlock); 383 continue; 384 } else if (LIST_FIRST(&vp->v_cache_src)) { 385 /* Don't recycle if active in the namecache */ 386 simple_unlock(&vp->v_interlock); 387 continue; 388 } else { 389 break; 390 } 391 } 392 } 393 394 if (vp) { 395 vp->v_flag |= VDOOMED; 396 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 397 freevnodes--; 398 simple_unlock(&vnode_free_list_slock); 399 cache_purge(vp); 400 vp->v_lease = NULL; 401 if (vp->v_type != VBAD) { 402 vgonel(vp, p); 403 } else { 404 simple_unlock(&vp->v_interlock); 405 } 406 407#ifdef DIAGNOSTIC 408 { 409 int s; 410 411 if (vp->v_data) 412 panic("cleaned vnode isn't"); 413 s = splbio(); 414 if (vp->v_numoutput) 415 panic("Clean vnode has pending I/O's"); 416 splx(s); 417 } 418#endif 419 vp->v_flag = 0; 420 vp->v_lastr = 0; 421 vp->v_lastw = 0; 422 vp->v_lasta = 0; 423 vp->v_cstart = 0; 424 vp->v_clen = 0; 425 vp->v_socket = 0; 426 vp->v_writecount = 0; /* XXX */ 427 } else { 428 simple_unlock(&vnode_free_list_slock); 429 vp = (struct vnode *) malloc((u_long) sizeof *vp, 430 M_VNODE, M_WAITOK); 431 bzero((char *) vp, sizeof *vp); 432 vp->v_dd = vp; 433 cache_purge(vp); 434 LIST_INIT(&vp->v_cache_src); 435 TAILQ_INIT(&vp->v_cache_dst); 436 numvnodes++; 437 } 438 439 vp->v_type = VNON; 440 vp->v_tag = tag; 441 vp->v_op = vops; 442 insmntque(vp, mp); 443 *vpp = vp; 444 vp->v_usecount = 1; 445 vp->v_data = 0; 446 return (0); 447} 448 449/* 450 * Move a vnode from one mount queue to another. 451 */ 452static void 453insmntque(vp, mp) 454 register struct vnode *vp; 455 register struct mount *mp; 456{ 457 458 simple_lock(&mntvnode_slock); 459 /* 460 * Delete from old mount point vnode list, if on one. 461 */ 462 if (vp->v_mount != NULL) 463 LIST_REMOVE(vp, v_mntvnodes); 464 /* 465 * Insert into list of vnodes for the new mount point, if available. 466 */ 467 if ((vp->v_mount = mp) == NULL) { 468 simple_unlock(&mntvnode_slock); 469 return; 470 } 471 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 472 simple_unlock(&mntvnode_slock); 473} 474 475/* 476 * Update outstanding I/O count and do wakeup if requested. 477 */ 478void 479vwakeup(bp) 480 register struct buf *bp; 481{ 482 register struct vnode *vp; 483 484 bp->b_flags &= ~B_WRITEINPROG; 485 if ((vp = bp->b_vp)) { 486 vp->v_numoutput--; 487 if (vp->v_numoutput < 0) 488 panic("vwakeup: neg numoutput"); 489 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 490 vp->v_flag &= ~VBWAIT; 491 wakeup((caddr_t) &vp->v_numoutput); 492 } 493 } 494} 495 496/* 497 * Flush out and invalidate all buffers associated with a vnode. 498 * Called with the underlying object locked. 499 */ 500int 501vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 502 register struct vnode *vp; 503 int flags; 504 struct ucred *cred; 505 struct proc *p; 506 int slpflag, slptimeo; 507{ 508 register struct buf *bp; 509 struct buf *nbp, *blist; 510 int s, error; 511 vm_object_t object; 512 513 if (flags & V_SAVE) { 514 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p))) 515 return (error); 516 if (vp->v_dirtyblkhd.lh_first != NULL) 517 panic("vinvalbuf: dirty bufs"); 518 } 519 520 s = splbio(); 521 for (;;) { 522 if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA)) 523 while (blist && blist->b_lblkno < 0) 524 blist = blist->b_vnbufs.le_next; 525 if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 526 (flags & V_SAVEMETA)) 527 while (blist && blist->b_lblkno < 0) 528 blist = blist->b_vnbufs.le_next; 529 if (!blist) 530 break; 531 532 for (bp = blist; bp; bp = nbp) { 533 nbp = bp->b_vnbufs.le_next; 534 if ((flags & V_SAVEMETA) && bp->b_lblkno < 0) 535 continue; 536 if (bp->b_flags & B_BUSY) { 537 bp->b_flags |= B_WANTED; 538 error = tsleep((caddr_t) bp, 539 slpflag | (PRIBIO + 1), "vinvalbuf", 540 slptimeo); 541 if (error) { 542 splx(s); 543 return (error); 544 } 545 break; 546 } 547 bremfree(bp); 548 bp->b_flags |= B_BUSY; 549 /* 550 * XXX Since there are no node locks for NFS, I 551 * believe there is a slight chance that a delayed 552 * write will occur while sleeping just above, so 553 * check for it. 554 */ 555 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 556 (void) VOP_BWRITE(bp); 557 break; 558 } 559 bp->b_flags |= (B_INVAL|B_NOCACHE|B_RELBUF); 560 brelse(bp); 561 } 562 } 563 564 while (vp->v_numoutput > 0) { 565 vp->v_flag |= VBWAIT; 566 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 567 } 568 569 splx(s); 570 571 /* 572 * Destroy the copy in the VM cache, too. 573 */ 574 object = vp->v_object; 575 if (object != NULL) { 576 vm_object_page_remove(object, 0, object->size, 577 (flags & V_SAVE) ? TRUE : FALSE); 578 } 579 if (!(flags & V_SAVEMETA) && 580 (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) 581 panic("vinvalbuf: flush failed"); 582 return (0); 583} 584 585/* 586 * Associate a buffer with a vnode. 587 */ 588void 589bgetvp(vp, bp) 590 register struct vnode *vp; 591 register struct buf *bp; 592{ 593 int s; 594 595#if defined(DIAGNOSTIC) 596 if (bp->b_vp) 597 panic("bgetvp: not free"); 598#endif 599 vhold(vp); 600 bp->b_vp = vp; 601 if (vp->v_type == VBLK || vp->v_type == VCHR) 602 bp->b_dev = vp->v_rdev; 603 else 604 bp->b_dev = NODEV; 605 /* 606 * Insert onto list for new vnode. 607 */ 608 s = splbio(); 609 bufinsvn(bp, &vp->v_cleanblkhd); 610 splx(s); 611} 612 613/* 614 * Disassociate a buffer from a vnode. 615 */ 616void 617brelvp(bp) 618 register struct buf *bp; 619{ 620 struct vnode *vp; 621 int s; 622 623#if defined(DIAGNOSTIC) 624 if (bp->b_vp == (struct vnode *) 0) 625 panic("brelvp: NULL"); 626#endif 627 628 /* 629 * Delete from old vnode list, if on one. 630 */ 631 s = splbio(); 632 if (bp->b_vnbufs.le_next != NOLIST) 633 bufremvn(bp); 634 splx(s); 635 636 vp = bp->b_vp; 637 bp->b_vp = (struct vnode *) 0; 638 vdrop(vp); 639} 640 641/* 642 * Associate a p-buffer with a vnode. 643 */ 644void 645pbgetvp(vp, bp) 646 register struct vnode *vp; 647 register struct buf *bp; 648{ 649#if defined(DIAGNOSTIC) 650 if (bp->b_vp) 651 panic("pbgetvp: not free"); 652#endif 653 bp->b_vp = vp; 654 if (vp->v_type == VBLK || vp->v_type == VCHR) 655 bp->b_dev = vp->v_rdev; 656 else 657 bp->b_dev = NODEV; 658} 659 660/* 661 * Disassociate a p-buffer from a vnode. 662 */ 663void 664pbrelvp(bp) 665 register struct buf *bp; 666{ 667 668#if defined(DIAGNOSTIC) 669 if (bp->b_vp == (struct vnode *) 0) 670 panic("pbrelvp: NULL"); 671#endif 672 673 bp->b_vp = (struct vnode *) 0; 674} 675 676/* 677 * Reassign a buffer from one vnode to another. 678 * Used to assign file specific control information 679 * (indirect blocks) to the vnode to which they belong. 680 */ 681void 682reassignbuf(bp, newvp) 683 register struct buf *bp; 684 register struct vnode *newvp; 685{ 686 int s; 687 688 if (newvp == NULL) { 689 printf("reassignbuf: NULL"); 690 return; 691 } 692 693 s = splbio(); 694 /* 695 * Delete from old vnode list, if on one. 696 */ 697 if (bp->b_vnbufs.le_next != NOLIST) { 698 bufremvn(bp); 699 vdrop(bp->b_vp); 700 } 701 /* 702 * If dirty, put on list of dirty buffers; otherwise insert onto list 703 * of clean buffers. 704 */ 705 if (bp->b_flags & B_DELWRI) { 706 struct buf *tbp; 707 708 tbp = newvp->v_dirtyblkhd.lh_first; 709 if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) { 710 bufinsvn(bp, &newvp->v_dirtyblkhd); 711 } else { 712 while (tbp->b_vnbufs.le_next && 713 (tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) { 714 tbp = tbp->b_vnbufs.le_next; 715 } 716 LIST_INSERT_AFTER(tbp, bp, b_vnbufs); 717 } 718 } else { 719 bufinsvn(bp, &newvp->v_cleanblkhd); 720 } 721 bp->b_vp = newvp; 722 vhold(bp->b_vp); 723 splx(s); 724} 725 726#ifndef DEVFS_ROOT 727/* 728 * Create a vnode for a block device. 729 * Used for mounting the root file system. 730 */ 731int 732bdevvp(dev, vpp) 733 dev_t dev; 734 struct vnode **vpp; 735{ 736 register struct vnode *vp; 737 struct vnode *nvp; 738 int error; 739 740 if (dev == NODEV) 741 return (0); 742 error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp); 743 if (error) { 744 *vpp = 0; 745 return (error); 746 } 747 vp = nvp; 748 vp->v_type = VBLK; 749 if ((nvp = checkalias(vp, dev, (struct mount *) 0))) { 750 vput(vp); 751 vp = nvp; 752 } 753 *vpp = vp; 754 return (0); 755} 756#endif /* !DEVFS_ROOT */ 757 758/* 759 * Check to see if the new vnode represents a special device 760 * for which we already have a vnode (either because of 761 * bdevvp() or because of a different vnode representing 762 * the same block device). If such an alias exists, deallocate 763 * the existing contents and return the aliased vnode. The 764 * caller is responsible for filling it with its new contents. 765 */ 766struct vnode * 767checkalias(nvp, nvp_rdev, mp) 768 register struct vnode *nvp; 769 dev_t nvp_rdev; 770 struct mount *mp; 771{ 772 struct proc *p = curproc; /* XXX */ 773 struct vnode *vp; 774 struct vnode **vpp; 775 776 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 777 return (NULLVP); 778 779 vpp = &speclisth[SPECHASH(nvp_rdev)]; 780loop: 781 simple_lock(&spechash_slock); 782 for (vp = *vpp; vp; vp = vp->v_specnext) { 783 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 784 continue; 785 /* 786 * Alias, but not in use, so flush it out. 787 */ 788 simple_lock(&vp->v_interlock); 789 if (vp->v_usecount == 0) { 790 simple_unlock(&spechash_slock); 791 vgonel(vp, p); 792 goto loop; 793 } 794 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { 795 simple_unlock(&spechash_slock); 796 goto loop; 797 } 798 break; 799 } 800 if (vp == NULL || vp->v_tag != VT_NON) { 801 MALLOC(nvp->v_specinfo, struct specinfo *, 802 sizeof(struct specinfo), M_VNODE, M_WAITOK); 803 nvp->v_rdev = nvp_rdev; 804 nvp->v_hashchain = vpp; 805 nvp->v_specnext = *vpp; 806 nvp->v_specflags = 0; 807 simple_unlock(&spechash_slock); 808 *vpp = nvp; 809 if (vp != NULLVP) { 810 nvp->v_flag |= VALIASED; 811 vp->v_flag |= VALIASED; 812 vput(vp); 813 } 814 return (NULLVP); 815 } 816 simple_unlock(&spechash_slock); 817 VOP_UNLOCK(vp, 0, p); 818 simple_lock(&vp->v_interlock); 819 vclean(vp, 0, p); 820 vp->v_op = nvp->v_op; 821 vp->v_tag = nvp->v_tag; 822 nvp->v_type = VNON; 823 insmntque(vp, mp); 824 return (vp); 825} 826 827/* 828 * Grab a particular vnode from the free list, increment its 829 * reference count and lock it. The vnode lock bit is set the 830 * vnode is being eliminated in vgone. The process is awakened 831 * when the transition is completed, and an error returned to 832 * indicate that the vnode is no longer usable (possibly having 833 * been changed to a new file system type). 834 */ 835int 836vget(vp, flags, p) 837 register struct vnode *vp; 838 int flags; 839 struct proc *p; 840{ 841 int error; 842 843 /* 844 * If the vnode is in the process of being cleaned out for 845 * another use, we wait for the cleaning to finish and then 846 * return failure. Cleaning is determined by checking that 847 * the VXLOCK flag is set. 848 */ 849 if ((flags & LK_INTERLOCK) == 0) { 850 simple_lock(&vp->v_interlock); 851 } 852 if (vp->v_flag & VXLOCK) { 853 vp->v_flag |= VXWANT; 854 simple_unlock(&vp->v_interlock); 855 tsleep((caddr_t)vp, PINOD, "vget", 0); 856 return (ENOENT); 857 } 858 859 vp->v_usecount++; 860 861 if (VSHOULDBUSY(vp)) 862 vbusy(vp); 863 /* 864 * Create the VM object, if needed 865 */ 866 if (((vp->v_type == VREG) || (vp->v_type == VBLK)) && 867 ((vp->v_object == NULL) || 868 (vp->v_object->flags & OBJ_VFS_REF) == 0 || 869 (vp->v_object->flags & OBJ_DEAD))) { 870 simple_unlock(&vp->v_interlock); 871 vfs_object_create(vp, curproc, curproc->p_ucred, 0); 872 simple_lock(&vp->v_interlock); 873 } 874 if (flags & LK_TYPE_MASK) { 875 if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) 876 vrele(vp); 877 return (error); 878 } 879 simple_unlock(&vp->v_interlock); 880 return (0); 881} 882 883/* 884 * Vnode put/release. 885 * If count drops to zero, call inactive routine and return to freelist. 886 */ 887void 888vrele(vp) 889 struct vnode *vp; 890{ 891 struct proc *p = curproc; /* XXX */ 892 893#ifdef DIAGNOSTIC 894 if (vp == NULL) 895 panic("vrele: null vp"); 896#endif 897 simple_lock(&vp->v_interlock); 898 899 if (vp->v_usecount > 1) { 900 901 vp->v_usecount--; 902 simple_unlock(&vp->v_interlock); 903 904 } else if (vp->v_usecount == 1) { 905 906 vp->v_usecount--; 907 908 if (VSHOULDFREE(vp)) 909 vfree(vp); 910 /* 911 * If we are doing a vput, the node is already locked, and we must 912 * call VOP_INACTIVE with the node locked. So, in the case of 913 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 914 */ 915 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 916 VOP_INACTIVE(vp, p); 917 } 918 919 } else { 920#ifdef DIAGNOSTIC 921 vprint("vrele: negative ref count", vp); 922#endif 923 panic("vrele: negative ref cnt"); 924 } 925} 926 927void 928vput(vp) 929 struct vnode *vp; 930{ 931 struct proc *p = curproc; /* XXX */ 932 933#ifdef DIAGNOSTIC 934 if (vp == NULL) 935 panic("vput: null vp"); 936#endif 937 simple_lock(&vp->v_interlock); 938 939 if (vp->v_usecount > 1) { 940 941 vp->v_usecount--; 942 VOP_UNLOCK(vp, LK_INTERLOCK, p); 943 944 } else if (vp->v_usecount == 1) { 945 946 vp->v_usecount--; 947 948 if (VSHOULDFREE(vp)) 949 vfree(vp); 950 /* 951 * If we are doing a vput, the node is already locked, and we must 952 * call VOP_INACTIVE with the node locked. So, in the case of 953 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 954 */ 955 simple_unlock(&vp->v_interlock); 956 VOP_INACTIVE(vp, p); 957 958 } else { 959#ifdef DIAGNOSTIC 960 vprint("vput: negative ref count", vp); 961#endif 962 panic("vput: negative ref cnt"); 963 } 964} 965 966/* 967 * Somebody doesn't want the vnode recycled. 968 */ 969void 970vhold(vp) 971 register struct vnode *vp; 972{ 973 974 simple_lock(&vp->v_interlock); 975 vp->v_holdcnt++; 976 if (VSHOULDBUSY(vp)) 977 vbusy(vp); 978 simple_unlock(&vp->v_interlock); 979} 980 981/* 982 * One less who cares about this vnode. 983 */ 984void 985vdrop(vp) 986 register struct vnode *vp; 987{ 988 989 simple_lock(&vp->v_interlock); 990 if (vp->v_holdcnt <= 0) 991 panic("holdrele: holdcnt"); 992 vp->v_holdcnt--; 993 if (VSHOULDFREE(vp)) 994 vfree(vp); 995 simple_unlock(&vp->v_interlock); 996} 997 998/* 999 * Remove any vnodes in the vnode table belonging to mount point mp. 1000 * 1001 * If MNT_NOFORCE is specified, there should not be any active ones, 1002 * return error if any are found (nb: this is a user error, not a 1003 * system error). If MNT_FORCE is specified, detach any active vnodes 1004 * that are found. 1005 */ 1006#ifdef DIAGNOSTIC 1007static int busyprt = 0; /* print out busy vnodes */ 1008SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1009#endif 1010 1011int 1012vflush(mp, skipvp, flags) 1013 struct mount *mp; 1014 struct vnode *skipvp; 1015 int flags; 1016{ 1017 struct proc *p = curproc; /* XXX */ 1018 struct vnode *vp, *nvp; 1019 int busy = 0; 1020 1021 simple_lock(&mntvnode_slock); 1022loop: 1023 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 1024 /* 1025 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1026 * Start over if it has (it won't be on the list anymore). 1027 */ 1028 if (vp->v_mount != mp) 1029 goto loop; 1030 nvp = vp->v_mntvnodes.le_next; 1031 /* 1032 * Skip over a selected vnode. 1033 */ 1034 if (vp == skipvp) 1035 continue; 1036 1037 simple_lock(&vp->v_interlock); 1038 /* 1039 * Skip over a vnodes marked VSYSTEM. 1040 */ 1041 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1042 simple_unlock(&vp->v_interlock); 1043 continue; 1044 } 1045 /* 1046 * If WRITECLOSE is set, only flush out regular file vnodes 1047 * open for writing. 1048 */ 1049 if ((flags & WRITECLOSE) && 1050 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1051 simple_unlock(&vp->v_interlock); 1052 continue; 1053 } 1054 1055 /* 1056 * With v_usecount == 0, all we need to do is clear out the 1057 * vnode data structures and we are done. 1058 */ 1059 if (vp->v_usecount == 0) { 1060 simple_unlock(&mntvnode_slock); 1061 vgonel(vp, p); 1062 simple_lock(&mntvnode_slock); 1063 continue; 1064 } 1065 1066 /* 1067 * If FORCECLOSE is set, forcibly close the vnode. For block 1068 * or character devices, revert to an anonymous device. For 1069 * all other files, just kill them. 1070 */ 1071 if (flags & FORCECLOSE) { 1072 simple_unlock(&mntvnode_slock); 1073 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1074 vgonel(vp, p); 1075 } else { 1076 vclean(vp, 0, p); 1077 vp->v_op = spec_vnodeop_p; 1078 insmntque(vp, (struct mount *) 0); 1079 } 1080 simple_lock(&mntvnode_slock); 1081 continue; 1082 } 1083#ifdef DIAGNOSTIC 1084 if (busyprt) 1085 vprint("vflush: busy vnode", vp); 1086#endif 1087 simple_unlock(&vp->v_interlock); 1088 busy++; 1089 } 1090 simple_unlock(&mntvnode_slock); 1091 if (busy) 1092 return (EBUSY); 1093 return (0); 1094} 1095 1096/* 1097 * Disassociate the underlying file system from a vnode. 1098 */ 1099static void 1100vclean(vp, flags, p) 1101 struct vnode *vp; 1102 int flags; 1103 struct proc *p; 1104{ 1105 int active, irefed; 1106 vm_object_t object; 1107 1108 /* 1109 * Check to see if the vnode is in use. If so we have to reference it 1110 * before we clean it out so that its count cannot fall to zero and 1111 * generate a race against ourselves to recycle it. 1112 */ 1113 if ((active = vp->v_usecount)) 1114 vp->v_usecount++; 1115 /* 1116 * Prevent the vnode from being recycled or brought into use while we 1117 * clean it out. 1118 */ 1119 if (vp->v_flag & VXLOCK) 1120 panic("vclean: deadlock"); 1121 vp->v_flag |= VXLOCK; 1122 /* 1123 * Even if the count is zero, the VOP_INACTIVE routine may still 1124 * have the object locked while it cleans it out. The VOP_LOCK 1125 * ensures that the VOP_INACTIVE routine is done with its work. 1126 * For active vnodes, it ensures that no other activity can 1127 * occur while the underlying object is being cleaned out. 1128 */ 1129 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1130 1131 object = vp->v_object; 1132 1133 /* 1134 * Clean out any buffers associated with the vnode. 1135 */ 1136 if (flags & DOCLOSE) 1137 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1138 1139 if (vp->v_object && (vp->v_object->flags & OBJ_VFS_REF)) { 1140 vp->v_object->flags &= ~OBJ_VFS_REF; 1141 vm_object_deallocate(object); 1142 } 1143 1144 /* 1145 * If purging an active vnode, it must be closed and 1146 * deactivated before being reclaimed. Note that the 1147 * VOP_INACTIVE will unlock the vnode. 1148 */ 1149 if (active) { 1150 if (flags & DOCLOSE) 1151 VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); 1152 VOP_INACTIVE(vp, p); 1153 } else { 1154 /* 1155 * Any other processes trying to obtain this lock must first 1156 * wait for VXLOCK to clear, then call the new lock operation. 1157 */ 1158 VOP_UNLOCK(vp, 0, p); 1159 } 1160 /* 1161 * Reclaim the vnode. 1162 */ 1163 if (VOP_RECLAIM(vp, p)) 1164 panic("vclean: cannot reclaim"); 1165 if (active) 1166 vrele(vp); 1167 cache_purge(vp); 1168 if (vp->v_vnlock) { 1169#if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */ 1170#ifdef DIAGNOSTIC 1171 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0) 1172 vprint("vclean: lock not drained", vp); 1173#endif 1174#endif 1175 FREE(vp->v_vnlock, M_VNODE); 1176 vp->v_vnlock = NULL; 1177 } 1178 1179 /* 1180 * Done with purge, notify sleepers of the grim news. 1181 */ 1182 vp->v_op = dead_vnodeop_p; 1183 vn_pollgone(vp); 1184 vp->v_tag = VT_NON; 1185 vp->v_flag &= ~VXLOCK; 1186 if (vp->v_flag & VXWANT) { 1187 vp->v_flag &= ~VXWANT; 1188 wakeup((caddr_t) vp); 1189 } 1190} 1191 1192/* 1193 * Eliminate all activity associated with the requested vnode 1194 * and with all vnodes aliased to the requested vnode. 1195 */ 1196int 1197vop_revoke(ap) 1198 struct vop_revoke_args /* { 1199 struct vnode *a_vp; 1200 int a_flags; 1201 } */ *ap; 1202{ 1203 struct vnode *vp, *vq; 1204 struct proc *p = curproc; /* XXX */ 1205 1206#ifdef DIAGNOSTIC 1207 if ((ap->a_flags & REVOKEALL) == 0) 1208 panic("vop_revoke"); 1209#endif 1210 1211 vp = ap->a_vp; 1212 simple_lock(&vp->v_interlock); 1213 1214 if (vp->v_flag & VALIASED) { 1215 /* 1216 * If a vgone (or vclean) is already in progress, 1217 * wait until it is done and return. 1218 */ 1219 if (vp->v_flag & VXLOCK) { 1220 vp->v_flag |= VXWANT; 1221 simple_unlock(&vp->v_interlock); 1222 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1223 return (0); 1224 } 1225 /* 1226 * Ensure that vp will not be vgone'd while we 1227 * are eliminating its aliases. 1228 */ 1229 vp->v_flag |= VXLOCK; 1230 simple_unlock(&vp->v_interlock); 1231 while (vp->v_flag & VALIASED) { 1232 simple_lock(&spechash_slock); 1233 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1234 if (vq->v_rdev != vp->v_rdev || 1235 vq->v_type != vp->v_type || vp == vq) 1236 continue; 1237 simple_unlock(&spechash_slock); 1238 vgone(vq); 1239 break; 1240 } 1241 if (vq == NULLVP) { 1242 simple_unlock(&spechash_slock); 1243 } 1244 } 1245 /* 1246 * Remove the lock so that vgone below will 1247 * really eliminate the vnode after which time 1248 * vgone will awaken any sleepers. 1249 */ 1250 simple_lock(&vp->v_interlock); 1251 vp->v_flag &= ~VXLOCK; 1252 } 1253 vgonel(vp, p); 1254 return (0); 1255} 1256 1257/* 1258 * Recycle an unused vnode to the front of the free list. 1259 * Release the passed interlock if the vnode will be recycled. 1260 */ 1261int 1262vrecycle(vp, inter_lkp, p) 1263 struct vnode *vp; 1264 struct simplelock *inter_lkp; 1265 struct proc *p; 1266{ 1267 1268 simple_lock(&vp->v_interlock); 1269 if (vp->v_usecount == 0) { 1270 if (inter_lkp) { 1271 simple_unlock(inter_lkp); 1272 } 1273 vgonel(vp, p); 1274 return (1); 1275 } 1276 simple_unlock(&vp->v_interlock); 1277 return (0); 1278} 1279 1280/* 1281 * Eliminate all activity associated with a vnode 1282 * in preparation for reuse. 1283 */ 1284void 1285vgone(vp) 1286 register struct vnode *vp; 1287{ 1288 struct proc *p = curproc; /* XXX */ 1289 1290 simple_lock(&vp->v_interlock); 1291 vgonel(vp, p); 1292} 1293 1294/* 1295 * vgone, with the vp interlock held. 1296 */ 1297static void 1298vgonel(vp, p) 1299 struct vnode *vp; 1300 struct proc *p; 1301{ 1302 struct vnode *vq; 1303 struct vnode *vx; 1304 1305 /* 1306 * If a vgone (or vclean) is already in progress, 1307 * wait until it is done and return. 1308 */ 1309 if (vp->v_flag & VXLOCK) { 1310 vp->v_flag |= VXWANT; 1311 simple_unlock(&vp->v_interlock); 1312 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1313 return; 1314 } 1315 1316 if (vp->v_object) { 1317 vp->v_object->flags |= OBJ_VNODE_GONE; 1318 } 1319 1320 /* 1321 * Clean out the filesystem specific data. 1322 */ 1323 vclean(vp, DOCLOSE, p); 1324 1325 /* 1326 * Delete from old mount point vnode list, if on one. 1327 */ 1328 if (vp->v_mount != NULL) 1329 insmntque(vp, (struct mount *)0); 1330 /* 1331 * If special device, remove it from special device alias list 1332 * if it is on one. 1333 */ 1334 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1335 simple_lock(&spechash_slock); 1336 if (*vp->v_hashchain == vp) { 1337 *vp->v_hashchain = vp->v_specnext; 1338 } else { 1339 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1340 if (vq->v_specnext != vp) 1341 continue; 1342 vq->v_specnext = vp->v_specnext; 1343 break; 1344 } 1345 if (vq == NULL) 1346 panic("missing bdev"); 1347 } 1348 if (vp->v_flag & VALIASED) { 1349 vx = NULL; 1350 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1351 if (vq->v_rdev != vp->v_rdev || 1352 vq->v_type != vp->v_type) 1353 continue; 1354 if (vx) 1355 break; 1356 vx = vq; 1357 } 1358 if (vx == NULL) 1359 panic("missing alias"); 1360 if (vq == NULL) 1361 vx->v_flag &= ~VALIASED; 1362 vp->v_flag &= ~VALIASED; 1363 } 1364 simple_unlock(&spechash_slock); 1365 FREE(vp->v_specinfo, M_VNODE); 1366 vp->v_specinfo = NULL; 1367 } 1368 1369 /* 1370 * If it is on the freelist and not already at the head, 1371 * move it to the head of the list. The test of the back 1372 * pointer and the reference count of zero is because 1373 * it will be removed from the free list by getnewvnode, 1374 * but will not have its reference count incremented until 1375 * after calling vgone. If the reference count were 1376 * incremented first, vgone would (incorrectly) try to 1377 * close the previous instance of the underlying object. 1378 */ 1379 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1380 simple_lock(&vnode_free_list_slock); 1381 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1382 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1383 simple_unlock(&vnode_free_list_slock); 1384 } 1385 1386 vp->v_type = VBAD; 1387} 1388 1389/* 1390 * Lookup a vnode by device number. 1391 */ 1392int 1393vfinddev(dev, type, vpp) 1394 dev_t dev; 1395 enum vtype type; 1396 struct vnode **vpp; 1397{ 1398 register struct vnode *vp; 1399 int rc = 0; 1400 1401 simple_lock(&spechash_slock); 1402 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1403 if (dev != vp->v_rdev || type != vp->v_type) 1404 continue; 1405 *vpp = vp; 1406 rc = 1; 1407 break; 1408 } 1409 simple_unlock(&spechash_slock); 1410 return (rc); 1411} 1412 1413/* 1414 * Calculate the total number of references to a special device. 1415 */ 1416int 1417vcount(vp) 1418 register struct vnode *vp; 1419{ 1420 struct vnode *vq, *vnext; 1421 int count; 1422 1423loop: 1424 if ((vp->v_flag & VALIASED) == 0) 1425 return (vp->v_usecount); 1426 simple_lock(&spechash_slock); 1427 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1428 vnext = vq->v_specnext; 1429 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1430 continue; 1431 /* 1432 * Alias, but not in use, so flush it out. 1433 */ 1434 if (vq->v_usecount == 0 && vq != vp) { 1435 simple_unlock(&spechash_slock); 1436 vgone(vq); 1437 goto loop; 1438 } 1439 count += vq->v_usecount; 1440 } 1441 simple_unlock(&spechash_slock); 1442 return (count); 1443} 1444/* 1445 * Print out a description of a vnode. 1446 */ 1447static char *typename[] = 1448{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1449 1450void 1451vprint(label, vp) 1452 char *label; 1453 register struct vnode *vp; 1454{ 1455 char buf[64]; 1456 1457 if (label != NULL) 1458 printf("%s: %x: ", label, vp); 1459 else 1460 printf("%x: ", vp); 1461 printf("type %s, usecount %d, writecount %d, refcount %ld,", 1462 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1463 vp->v_holdcnt); 1464 buf[0] = '\0'; 1465 if (vp->v_flag & VROOT) 1466 strcat(buf, "|VROOT"); 1467 if (vp->v_flag & VTEXT) 1468 strcat(buf, "|VTEXT"); 1469 if (vp->v_flag & VSYSTEM) 1470 strcat(buf, "|VSYSTEM"); 1471 if (vp->v_flag & VXLOCK) 1472 strcat(buf, "|VXLOCK"); 1473 if (vp->v_flag & VXWANT) 1474 strcat(buf, "|VXWANT"); 1475 if (vp->v_flag & VBWAIT) 1476 strcat(buf, "|VBWAIT"); 1477 if (vp->v_flag & VALIASED) 1478 strcat(buf, "|VALIASED"); 1479 if (vp->v_flag & VDOOMED) 1480 strcat(buf, "|VDOOMED"); 1481 if (vp->v_flag & VFREE) 1482 strcat(buf, "|VFREE"); 1483 if (buf[0] != '\0') 1484 printf(" flags (%s)", &buf[1]); 1485 if (vp->v_data == NULL) { 1486 printf("\n"); 1487 } else { 1488 printf("\n\t"); 1489 VOP_PRINT(vp); 1490 } 1491} 1492 1493#ifdef DDB 1494/* 1495 * List all of the locked vnodes in the system. 1496 * Called when debugging the kernel. 1497 */ 1498static void 1499printlockedvnodes() 1500{ 1501 struct proc *p = curproc; /* XXX */ 1502 struct mount *mp, *nmp; 1503 struct vnode *vp; 1504 1505 printf("Locked vnodes\n"); 1506 simple_lock(&mountlist_slock); 1507 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1508 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1509 nmp = mp->mnt_list.cqe_next; 1510 continue; 1511 } 1512 for (vp = mp->mnt_vnodelist.lh_first; 1513 vp != NULL; 1514 vp = vp->v_mntvnodes.le_next) { 1515 if (VOP_ISLOCKED(vp)) 1516 vprint((char *)0, vp); 1517 } 1518 simple_lock(&mountlist_slock); 1519 nmp = mp->mnt_list.cqe_next; 1520 vfs_unbusy(mp, p); 1521 } 1522 simple_unlock(&mountlist_slock); 1523} 1524#endif 1525 1526/* 1527 * Top level filesystem related information gathering. 1528 */ 1529static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS); 1530 1531static int 1532vfs_sysctl SYSCTL_HANDLER_ARGS 1533{ 1534 int *name = (int *)arg1 - 1; /* XXX */ 1535 u_int namelen = arg2 + 1; /* XXX */ 1536 struct vfsconf *vfsp; 1537 1538#ifndef NO_COMPAT_PRELITE2 1539 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1540 if (namelen == 1) 1541 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1542#endif 1543 1544#ifdef notyet 1545 /* all sysctl names at this level are at least name and field */ 1546 if (namelen < 2) 1547 return (ENOTDIR); /* overloaded */ 1548 if (name[0] != VFS_GENERIC) { 1549 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1550 if (vfsp->vfc_typenum == name[0]) 1551 break; 1552 if (vfsp == NULL) 1553 return (EOPNOTSUPP); 1554 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1555 oldp, oldlenp, newp, newlen, p)); 1556 } 1557#endif 1558 switch (name[1]) { 1559 case VFS_MAXTYPENUM: 1560 if (namelen != 2) 1561 return (ENOTDIR); 1562 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1563 case VFS_CONF: 1564 if (namelen != 3) 1565 return (ENOTDIR); /* overloaded */ 1566 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1567 if (vfsp->vfc_typenum == name[2]) 1568 break; 1569 if (vfsp == NULL) 1570 return (EOPNOTSUPP); 1571 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1572 } 1573 return (EOPNOTSUPP); 1574} 1575 1576SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1577 "Generic filesystem"); 1578 1579#ifndef NO_COMPAT_PRELITE2 1580 1581static int 1582sysctl_ovfs_conf SYSCTL_HANDLER_ARGS 1583{ 1584 int error; 1585 struct vfsconf *vfsp; 1586 struct ovfsconf ovfs; 1587 1588 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1589 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1590 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1591 ovfs.vfc_index = vfsp->vfc_typenum; 1592 ovfs.vfc_refcount = vfsp->vfc_refcount; 1593 ovfs.vfc_flags = vfsp->vfc_flags; 1594 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1595 if (error) 1596 return error; 1597 } 1598 return 0; 1599} 1600 1601#endif /* !NO_COMPAT_PRELITE2 */ 1602 1603static volatile int kinfo_vdebug = 1; 1604 1605#if 0 1606#define KINFO_VNODESLOP 10 1607/* 1608 * Dump vnode list (via sysctl). 1609 * Copyout address of vnode followed by vnode. 1610 */ 1611/* ARGSUSED */ 1612static int 1613sysctl_vnode SYSCTL_HANDLER_ARGS 1614{ 1615 struct proc *p = curproc; /* XXX */ 1616 struct mount *mp, *nmp; 1617 struct vnode *nvp, *vp; 1618 int error; 1619 1620#define VPTRSZ sizeof (struct vnode *) 1621#define VNODESZ sizeof (struct vnode) 1622 1623 req->lock = 0; 1624 if (!req->oldptr) /* Make an estimate */ 1625 return (SYSCTL_OUT(req, 0, 1626 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 1627 1628 simple_lock(&mountlist_slock); 1629 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1630 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1631 nmp = mp->mnt_list.cqe_next; 1632 continue; 1633 } 1634again: 1635 simple_lock(&mntvnode_slock); 1636 for (vp = mp->mnt_vnodelist.lh_first; 1637 vp != NULL; 1638 vp = nvp) { 1639 /* 1640 * Check that the vp is still associated with 1641 * this filesystem. RACE: could have been 1642 * recycled onto the same filesystem. 1643 */ 1644 if (vp->v_mount != mp) { 1645 simple_unlock(&mntvnode_slock); 1646 if (kinfo_vdebug) 1647 printf("kinfo: vp changed\n"); 1648 goto again; 1649 } 1650 nvp = vp->v_mntvnodes.le_next; 1651 simple_unlock(&mntvnode_slock); 1652 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 1653 (error = SYSCTL_OUT(req, vp, VNODESZ))) 1654 return (error); 1655 simple_lock(&mntvnode_slock); 1656 } 1657 simple_unlock(&mntvnode_slock); 1658 simple_lock(&mountlist_slock); 1659 nmp = mp->mnt_list.cqe_next; 1660 vfs_unbusy(mp, p); 1661 } 1662 simple_unlock(&mountlist_slock); 1663 1664 return (0); 1665} 1666#endif 1667 1668/* 1669 * XXX 1670 * Exporting the vnode list on large systems causes them to crash. 1671 * Exporting the vnode list on medium systems causes sysctl to coredump. 1672 */ 1673#if 0 1674SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 1675 0, 0, sysctl_vnode, "S,vnode", ""); 1676#endif 1677 1678/* 1679 * Check to see if a filesystem is mounted on a block device. 1680 */ 1681int 1682vfs_mountedon(vp) 1683 struct vnode *vp; 1684{ 1685 struct vnode *vq; 1686 int error = 0; 1687 1688 if (vp->v_specflags & SI_MOUNTEDON) 1689 return (EBUSY); 1690 if (vp->v_flag & VALIASED) { 1691 simple_lock(&spechash_slock); 1692 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1693 if (vq->v_rdev != vp->v_rdev || 1694 vq->v_type != vp->v_type) 1695 continue; 1696 if (vq->v_specflags & SI_MOUNTEDON) { 1697 error = EBUSY; 1698 break; 1699 } 1700 } 1701 simple_unlock(&spechash_slock); 1702 } 1703 return (error); 1704} 1705 1706/* 1707 * Unmount all filesystems. The list is traversed in reverse order 1708 * of mounting to avoid dependencies. 1709 */ 1710void 1711vfs_unmountall() 1712{ 1713 struct mount *mp, *nmp; 1714 struct proc *p = initproc; /* XXX XXX should this be proc0? */ 1715 int error; 1716 1717 /* 1718 * Since this only runs when rebooting, it is not interlocked. 1719 */ 1720 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 1721 nmp = mp->mnt_list.cqe_prev; 1722 error = dounmount(mp, MNT_FORCE, p); 1723 if (error) { 1724 printf("unmount of %s failed (", 1725 mp->mnt_stat.f_mntonname); 1726 if (error == EBUSY) 1727 printf("BUSY)\n"); 1728 else 1729 printf("%d)\n", error); 1730 } 1731 } 1732} 1733 1734/* 1735 * Build hash lists of net addresses and hang them off the mount point. 1736 * Called by ufs_mount() to set up the lists of export addresses. 1737 */ 1738static int 1739vfs_hang_addrlist(mp, nep, argp) 1740 struct mount *mp; 1741 struct netexport *nep; 1742 struct export_args *argp; 1743{ 1744 register struct netcred *np; 1745 register struct radix_node_head *rnh; 1746 register int i; 1747 struct radix_node *rn; 1748 struct sockaddr *saddr, *smask = 0; 1749 struct domain *dom; 1750 int error; 1751 1752 if (argp->ex_addrlen == 0) { 1753 if (mp->mnt_flag & MNT_DEFEXPORTED) 1754 return (EPERM); 1755 np = &nep->ne_defexported; 1756 np->netc_exflags = argp->ex_flags; 1757 np->netc_anon = argp->ex_anon; 1758 np->netc_anon.cr_ref = 1; 1759 mp->mnt_flag |= MNT_DEFEXPORTED; 1760 return (0); 1761 } 1762 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1763 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1764 bzero((caddr_t) np, i); 1765 saddr = (struct sockaddr *) (np + 1); 1766 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1767 goto out; 1768 if (saddr->sa_len > argp->ex_addrlen) 1769 saddr->sa_len = argp->ex_addrlen; 1770 if (argp->ex_masklen) { 1771 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 1772 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 1773 if (error) 1774 goto out; 1775 if (smask->sa_len > argp->ex_masklen) 1776 smask->sa_len = argp->ex_masklen; 1777 } 1778 i = saddr->sa_family; 1779 if ((rnh = nep->ne_rtable[i]) == 0) { 1780 /* 1781 * Seems silly to initialize every AF when most are not used, 1782 * do so on demand here 1783 */ 1784 for (dom = domains; dom; dom = dom->dom_next) 1785 if (dom->dom_family == i && dom->dom_rtattach) { 1786 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1787 dom->dom_rtoffset); 1788 break; 1789 } 1790 if ((rnh = nep->ne_rtable[i]) == 0) { 1791 error = ENOBUFS; 1792 goto out; 1793 } 1794 } 1795 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 1796 np->netc_rnodes); 1797 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1798 error = EPERM; 1799 goto out; 1800 } 1801 np->netc_exflags = argp->ex_flags; 1802 np->netc_anon = argp->ex_anon; 1803 np->netc_anon.cr_ref = 1; 1804 return (0); 1805out: 1806 free(np, M_NETADDR); 1807 return (error); 1808} 1809 1810/* ARGSUSED */ 1811static int 1812vfs_free_netcred(rn, w) 1813 struct radix_node *rn; 1814 void *w; 1815{ 1816 register struct radix_node_head *rnh = (struct radix_node_head *) w; 1817 1818 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1819 free((caddr_t) rn, M_NETADDR); 1820 return (0); 1821} 1822 1823/* 1824 * Free the net address hash lists that are hanging off the mount points. 1825 */ 1826static void 1827vfs_free_addrlist(nep) 1828 struct netexport *nep; 1829{ 1830 register int i; 1831 register struct radix_node_head *rnh; 1832 1833 for (i = 0; i <= AF_MAX; i++) 1834 if ((rnh = nep->ne_rtable[i])) { 1835 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1836 (caddr_t) rnh); 1837 free((caddr_t) rnh, M_RTABLE); 1838 nep->ne_rtable[i] = 0; 1839 } 1840} 1841 1842int 1843vfs_export(mp, nep, argp) 1844 struct mount *mp; 1845 struct netexport *nep; 1846 struct export_args *argp; 1847{ 1848 int error; 1849 1850 if (argp->ex_flags & MNT_DELEXPORT) { 1851 if (mp->mnt_flag & MNT_EXPUBLIC) { 1852 vfs_setpublicfs(NULL, NULL, NULL); 1853 mp->mnt_flag &= ~MNT_EXPUBLIC; 1854 } 1855 vfs_free_addrlist(nep); 1856 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1857 } 1858 if (argp->ex_flags & MNT_EXPORTED) { 1859 if (argp->ex_flags & MNT_EXPUBLIC) { 1860 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1861 return (error); 1862 mp->mnt_flag |= MNT_EXPUBLIC; 1863 } 1864 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1865 return (error); 1866 mp->mnt_flag |= MNT_EXPORTED; 1867 } 1868 return (0); 1869} 1870 1871 1872/* 1873 * Set the publicly exported filesystem (WebNFS). Currently, only 1874 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1875 */ 1876int 1877vfs_setpublicfs(mp, nep, argp) 1878 struct mount *mp; 1879 struct netexport *nep; 1880 struct export_args *argp; 1881{ 1882 int error; 1883 struct vnode *rvp; 1884 char *cp; 1885 1886 /* 1887 * mp == NULL -> invalidate the current info, the FS is 1888 * no longer exported. May be called from either vfs_export 1889 * or unmount, so check if it hasn't already been done. 1890 */ 1891 if (mp == NULL) { 1892 if (nfs_pub.np_valid) { 1893 nfs_pub.np_valid = 0; 1894 if (nfs_pub.np_index != NULL) { 1895 FREE(nfs_pub.np_index, M_TEMP); 1896 nfs_pub.np_index = NULL; 1897 } 1898 } 1899 return (0); 1900 } 1901 1902 /* 1903 * Only one allowed at a time. 1904 */ 1905 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1906 return (EBUSY); 1907 1908 /* 1909 * Get real filehandle for root of exported FS. 1910 */ 1911 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1912 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1913 1914 if ((error = VFS_ROOT(mp, &rvp))) 1915 return (error); 1916 1917 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1918 return (error); 1919 1920 vput(rvp); 1921 1922 /* 1923 * If an indexfile was specified, pull it in. 1924 */ 1925 if (argp->ex_indexfile != NULL) { 1926 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 1927 M_WAITOK); 1928 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1929 MAXNAMLEN, (size_t *)0); 1930 if (!error) { 1931 /* 1932 * Check for illegal filenames. 1933 */ 1934 for (cp = nfs_pub.np_index; *cp; cp++) { 1935 if (*cp == '/') { 1936 error = EINVAL; 1937 break; 1938 } 1939 } 1940 } 1941 if (error) { 1942 FREE(nfs_pub.np_index, M_TEMP); 1943 return (error); 1944 } 1945 } 1946 1947 nfs_pub.np_mount = mp; 1948 nfs_pub.np_valid = 1; 1949 return (0); 1950} 1951 1952struct netcred * 1953vfs_export_lookup(mp, nep, nam) 1954 register struct mount *mp; 1955 struct netexport *nep; 1956 struct sockaddr *nam; 1957{ 1958 register struct netcred *np; 1959 register struct radix_node_head *rnh; 1960 struct sockaddr *saddr; 1961 1962 np = NULL; 1963 if (mp->mnt_flag & MNT_EXPORTED) { 1964 /* 1965 * Lookup in the export list first. 1966 */ 1967 if (nam != NULL) { 1968 saddr = nam; 1969 rnh = nep->ne_rtable[saddr->sa_family]; 1970 if (rnh != NULL) { 1971 np = (struct netcred *) 1972 (*rnh->rnh_matchaddr)((caddr_t)saddr, 1973 rnh); 1974 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1975 np = NULL; 1976 } 1977 } 1978 /* 1979 * If no address match, use the default if it exists. 1980 */ 1981 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1982 np = &nep->ne_defexported; 1983 } 1984 return (np); 1985} 1986 1987/* 1988 * perform msync on all vnodes under a mount point 1989 * the mount point must be locked. 1990 */ 1991void 1992vfs_msync(struct mount *mp, int flags) { 1993 struct vnode *vp, *nvp; 1994loop: 1995 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 1996 1997 if (vp->v_mount != mp) 1998 goto loop; 1999 nvp = vp->v_mntvnodes.le_next; 2000 if (VOP_ISLOCKED(vp) && (flags != MNT_WAIT)) 2001 continue; 2002 if (vp->v_object && 2003 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2004 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curproc); 2005 vm_object_page_clean(vp->v_object, 0, 0, TRUE); 2006 VOP_UNLOCK(vp, 0, curproc); 2007 } 2008 } 2009} 2010 2011/* 2012 * Create the VM object needed for VMIO and mmap support. This 2013 * is done for all VREG files in the system. Some filesystems might 2014 * afford the additional metadata buffering capability of the 2015 * VMIO code by making the device node be VMIO mode also. 2016 */ 2017int 2018vfs_object_create(vp, p, cred, waslocked) 2019 struct vnode *vp; 2020 struct proc *p; 2021 struct ucred *cred; 2022 int waslocked; 2023{ 2024 struct vattr vat; 2025 vm_object_t object; 2026 int error = 0; 2027 2028 if ((vp->v_type != VREG) && (vp->v_type != VBLK)) 2029 return 0; 2030 2031retry: 2032 if ((object = vp->v_object) == NULL) { 2033 if (vp->v_type == VREG) { 2034 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2035 goto retn; 2036 (void) vnode_pager_alloc(vp, 2037 OFF_TO_IDX(round_page(vat.va_size)), 0, 0); 2038 vp->v_object->flags |= OBJ_VFS_REF; 2039 } else { 2040 /* 2041 * This simply allocates the biggest object possible 2042 * for a VBLK vnode. This should be fixed, but doesn't 2043 * cause any problems (yet). 2044 */ 2045 (void) vnode_pager_alloc(vp, INT_MAX, 0, 0); 2046 vp->v_object->flags |= OBJ_VFS_REF; 2047 } 2048 } else { 2049 if (object->flags & OBJ_DEAD) { 2050 if (waslocked) 2051 VOP_UNLOCK(vp, 0, p); 2052 tsleep(object, PVM, "vodead", 0); 2053 if (waslocked) 2054 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2055 goto retry; 2056 } 2057 if ((object->flags & OBJ_VFS_REF) == 0) { 2058 vm_object_reference(object); 2059 object->flags |= OBJ_VFS_REF; 2060 } 2061 } 2062 if (vp->v_object) 2063 vp->v_flag |= VVMIO; 2064 2065retn: 2066 return error; 2067} 2068 2069static void 2070vfree(vp) 2071 struct vnode *vp; 2072{ 2073 simple_lock(&vnode_free_list_slock); 2074 if (vp->v_flag & VAGE) { 2075 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2076 } else { 2077 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2078 } 2079 freevnodes++; 2080 simple_unlock(&vnode_free_list_slock); 2081 vp->v_flag &= ~VAGE; 2082 vp->v_flag |= VFREE; 2083} 2084 2085static void 2086vbusy(vp) 2087 struct vnode *vp; 2088{ 2089 simple_lock(&vnode_free_list_slock); 2090 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2091 freevnodes--; 2092 simple_unlock(&vnode_free_list_slock); 2093 vp->v_flag &= ~VFREE; 2094} 2095 2096/* 2097 * Record a process's interest in events which might happen to 2098 * a vnode. Because poll uses the historic select-style interface 2099 * internally, this routine serves as both the ``check for any 2100 * pending events'' and the ``record my interest in future events'' 2101 * functions. (These are done together, while the lock is held, 2102 * to avoid race conditions.) 2103 */ 2104int 2105vn_pollrecord(vp, p, events) 2106 struct vnode *vp; 2107 struct proc *p; 2108 short events; 2109{ 2110 simple_lock(&vp->v_pollinfo.vpi_lock); 2111 if (vp->v_pollinfo.vpi_revents & events) { 2112 /* 2113 * This leaves events we are not interested 2114 * in available for the other process which 2115 * which presumably had requested them 2116 * (otherwise they would never have been 2117 * recorded). 2118 */ 2119 events &= vp->v_pollinfo.vpi_revents; 2120 vp->v_pollinfo.vpi_revents &= ~events; 2121 2122 simple_unlock(&vp->v_pollinfo.vpi_lock); 2123 return events; 2124 } 2125 vp->v_pollinfo.vpi_events |= events; 2126 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2127 simple_unlock(&vp->v_pollinfo.vpi_lock); 2128 return 0; 2129} 2130 2131/* 2132 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2133 * it is possible for us to miss an event due to race conditions, but 2134 * that condition is expected to be rare, so for the moment it is the 2135 * preferred interface. 2136 */ 2137void 2138vn_pollevent(vp, events) 2139 struct vnode *vp; 2140 short events; 2141{ 2142 simple_lock(&vp->v_pollinfo.vpi_lock); 2143 if (vp->v_pollinfo.vpi_events & events) { 2144 /* 2145 * We clear vpi_events so that we don't 2146 * call selwakeup() twice if two events are 2147 * posted before the polling process(es) is 2148 * awakened. This also ensures that we take at 2149 * most one selwakeup() if the polling process 2150 * is no longer interested. However, it does 2151 * mean that only one event can be noticed at 2152 * a time. (Perhaps we should only clear those 2153 * event bits which we note?) XXX 2154 */ 2155 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2156 vp->v_pollinfo.vpi_revents |= events; 2157 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2158 } 2159 simple_unlock(&vp->v_pollinfo.vpi_lock); 2160} 2161 2162/* 2163 * Wake up anyone polling on vp because it is being revoked. 2164 * This depends on dead_poll() returning POLLHUP for correct 2165 * behavior. 2166 */ 2167void 2168vn_pollgone(vp) 2169 struct vnode *vp; 2170{ 2171 simple_lock(&vp->v_pollinfo.vpi_lock); 2172 if (vp->v_pollinfo.vpi_events) { 2173 vp->v_pollinfo.vpi_events = 0; 2174 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2175 } 2176 simple_unlock(&vp->v_pollinfo.vpi_lock); 2177} 2178