vfs_export.c revision 33205
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 1.132 1998/02/09 06:09:35 eivind 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/vm_pager.h> 70#include <vm/vnode_pager.h> 71#include <vm/vm_zone.h> 72#include <sys/sysctl.h> 73 74#include <miscfs/specfs/specdev.h> 75 76static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 77 78static void insmntque __P((struct vnode *vp, struct mount *mp)); 79#ifdef DDB 80static void printlockedvnodes __P((void)); 81#endif 82static void vclean __P((struct vnode *vp, int flags, struct proc *p)); 83static void vfree __P((struct vnode *)); 84static void vgonel __P((struct vnode *vp, struct proc *p)); 85static unsigned long numvnodes; 86SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 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} 105 106static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 107struct tobefreelist vnode_tobefree_list; /* vnode free list */ 108 109static u_long wantfreevnodes = 25; 110SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 111static u_long freevnodes = 0; 112SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 113 114int vfs_ioopt = 0; 115SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 116 117struct mntlist mountlist; /* mounted filesystem list */ 118struct simplelock mountlist_slock; 119static struct simplelock mntid_slock; 120struct simplelock mntvnode_slock; 121static struct simplelock vnode_free_list_slock; 122static struct simplelock spechash_slock; 123struct nfs_public nfs_pub; /* publicly exported FS */ 124static vm_zone_t vnode_zone; 125 126int desiredvnodes; 127SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, &desiredvnodes, 0, ""); 128 129static void vfs_free_addrlist __P((struct netexport *nep)); 130static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 131static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 132 struct export_args *argp)); 133 134/* 135 * Initialize the vnode management data structures. 136 */ 137void 138vntblinit() 139{ 140 141 desiredvnodes = maxproc + cnt.v_page_count / 4; 142 simple_lock_init(&mntvnode_slock); 143 simple_lock_init(&mntid_slock); 144 simple_lock_init(&spechash_slock); 145 TAILQ_INIT(&vnode_free_list); 146 TAILQ_INIT(&vnode_tobefree_list); 147 simple_lock_init(&vnode_free_list_slock); 148 CIRCLEQ_INIT(&mountlist); 149 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 150} 151 152/* 153 * Mark a mount point as busy. Used to synchronize access and to delay 154 * unmounting. Interlock is not released on failure. 155 */ 156int 157vfs_busy(mp, flags, interlkp, p) 158 struct mount *mp; 159 int flags; 160 struct simplelock *interlkp; 161 struct proc *p; 162{ 163 int lkflags; 164 165 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 166 if (flags & LK_NOWAIT) 167 return (ENOENT); 168 mp->mnt_kern_flag |= MNTK_MWAIT; 169 if (interlkp) { 170 simple_unlock(interlkp); 171 } 172 /* 173 * Since all busy locks are shared except the exclusive 174 * lock granted when unmounting, the only place that a 175 * wakeup needs to be done is at the release of the 176 * exclusive lock at the end of dounmount. 177 */ 178 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 179 if (interlkp) { 180 simple_lock(interlkp); 181 } 182 return (ENOENT); 183 } 184 lkflags = LK_SHARED; 185 if (interlkp) 186 lkflags |= LK_INTERLOCK; 187 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 188 panic("vfs_busy: unexpected lock failure"); 189 return (0); 190} 191 192/* 193 * Free a busy filesystem. 194 */ 195void 196vfs_unbusy(mp, p) 197 struct mount *mp; 198 struct proc *p; 199{ 200 201 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 202} 203 204/* 205 * Lookup a filesystem type, and if found allocate and initialize 206 * a mount structure for it. 207 * 208 * Devname is usually updated by mount(8) after booting. 209 */ 210int 211vfs_rootmountalloc(fstypename, devname, mpp) 212 char *fstypename; 213 char *devname; 214 struct mount **mpp; 215{ 216 struct proc *p = curproc; /* XXX */ 217 struct vfsconf *vfsp; 218 struct mount *mp; 219 220 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 221 if (!strcmp(vfsp->vfc_name, fstypename)) 222 break; 223 if (vfsp == NULL) 224 return (ENODEV); 225 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 226 bzero((char *)mp, (u_long)sizeof(struct mount)); 227 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 228 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 229 LIST_INIT(&mp->mnt_vnodelist); 230 mp->mnt_vfc = vfsp; 231 mp->mnt_op = vfsp->vfc_vfsops; 232 mp->mnt_flag = MNT_RDONLY; 233 mp->mnt_vnodecovered = NULLVP; 234 vfsp->vfc_refcount++; 235 mp->mnt_stat.f_type = vfsp->vfc_typenum; 236 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 237 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 238 mp->mnt_stat.f_mntonname[0] = '/'; 239 mp->mnt_stat.f_mntonname[1] = 0; 240 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 241 *mpp = mp; 242 return (0); 243} 244 245/* 246 * Find an appropriate filesystem to use for the root. If a filesystem 247 * has not been preselected, walk through the list of known filesystems 248 * trying those that have mountroot routines, and try them until one 249 * works or we have tried them all. 250 */ 251#ifdef notdef /* XXX JH */ 252int 253lite2_vfs_mountroot() 254{ 255 struct vfsconf *vfsp; 256 extern int (*lite2_mountroot) __P((void)); 257 int error; 258 259 if (lite2_mountroot != NULL) 260 return ((*lite2_mountroot)()); 261 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 262 if (vfsp->vfc_mountroot == NULL) 263 continue; 264 if ((error = (*vfsp->vfc_mountroot)()) == 0) 265 return (0); 266 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 267 } 268 return (ENODEV); 269} 270#endif 271 272/* 273 * Lookup a mount point by filesystem identifier. 274 */ 275struct mount * 276vfs_getvfs(fsid) 277 fsid_t *fsid; 278{ 279 register struct mount *mp; 280 281 simple_lock(&mountlist_slock); 282 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 283 mp = mp->mnt_list.cqe_next) { 284 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 285 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 286 simple_unlock(&mountlist_slock); 287 return (mp); 288 } 289 } 290 simple_unlock(&mountlist_slock); 291 return ((struct mount *) 0); 292} 293 294/* 295 * Get a new unique fsid 296 */ 297void 298vfs_getnewfsid(mp) 299 struct mount *mp; 300{ 301 static u_short xxxfs_mntid; 302 303 fsid_t tfsid; 304 int mtype; 305 306 simple_lock(&mntid_slock); 307 mtype = mp->mnt_vfc->vfc_typenum; 308 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 309 mp->mnt_stat.f_fsid.val[1] = mtype; 310 if (xxxfs_mntid == 0) 311 ++xxxfs_mntid; 312 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 313 tfsid.val[1] = mtype; 314 if (mountlist.cqh_first != (void *)&mountlist) { 315 while (vfs_getvfs(&tfsid)) { 316 tfsid.val[0]++; 317 xxxfs_mntid++; 318 } 319 } 320 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 321 simple_unlock(&mntid_slock); 322} 323 324/* 325 * Set vnode attributes to VNOVAL 326 */ 327void 328vattr_null(vap) 329 register struct vattr *vap; 330{ 331 332 vap->va_type = VNON; 333 vap->va_size = VNOVAL; 334 vap->va_bytes = VNOVAL; 335 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 336 vap->va_fsid = vap->va_fileid = 337 vap->va_blocksize = vap->va_rdev = 338 vap->va_atime.tv_sec = vap->va_atime.tv_nsec = 339 vap->va_mtime.tv_sec = vap->va_mtime.tv_nsec = 340 vap->va_ctime.tv_sec = vap->va_ctime.tv_nsec = 341 vap->va_flags = vap->va_gen = VNOVAL; 342 vap->va_vaflags = 0; 343} 344 345/* 346 * Routines having to do with the management of the vnode table. 347 */ 348extern vop_t **dead_vnodeop_p; 349 350/* 351 * Return the next vnode from the free list. 352 */ 353int 354getnewvnode(tag, mp, vops, vpp) 355 enum vtagtype tag; 356 struct mount *mp; 357 vop_t **vops; 358 struct vnode **vpp; 359{ 360 int s; 361 struct proc *p = curproc; /* XXX */ 362 struct vnode *vp, *tvp, *nvp; 363 vm_object_t object; 364 TAILQ_HEAD(freelst, vnode) vnode_tmp_list; 365 366 /* 367 * We take the least recently used vnode from the freelist 368 * if we can get it and it has no cached pages, and no 369 * namecache entries are relative to it. 370 * Otherwise we allocate a new vnode 371 */ 372 373 s = splbio(); 374 simple_lock(&vnode_free_list_slock); 375 TAILQ_INIT(&vnode_tmp_list); 376 377 for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) { 378 nvp = TAILQ_NEXT(vp, v_freelist); 379 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 380 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 381 vp->v_flag &= ~VTBFREE; 382 vp->v_flag |= VFREE; 383 if (vp->v_usecount) 384 panic("tobe free vnode isn't"); 385 freevnodes++; 386 } 387 388 if (wantfreevnodes && freevnodes < wantfreevnodes) { 389 vp = NULL; 390 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) { 391 /* 392 * XXX: this is only here to be backwards compatible 393 */ 394 vp = NULL; 395 } else { 396 for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) { 397 398 nvp = TAILQ_NEXT(vp, v_freelist); 399 400 if (!simple_lock_try(&vp->v_interlock)) 401 continue; 402 if (vp->v_usecount) 403 panic("free vnode isn't"); 404 405 object = vp->v_object; 406 if (object && (object->resident_page_count || object->ref_count)) { 407 /* Don't recycle if it's caching some pages */ 408 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 409 TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist); 410 continue; 411 } else if (LIST_FIRST(&vp->v_cache_src)) { 412 /* Don't recycle if active in the namecache */ 413 simple_unlock(&vp->v_interlock); 414 continue; 415 } else { 416 break; 417 } 418 } 419 } 420 421 for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) { 422 nvp = TAILQ_NEXT(tvp, v_freelist); 423 TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist); 424 TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist); 425 simple_unlock(&tvp->v_interlock); 426 } 427 428 if (vp) { 429 vp->v_flag |= VDOOMED; 430 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 431 freevnodes--; 432 simple_unlock(&vnode_free_list_slock); 433 cache_purge(vp); 434 vp->v_lease = NULL; 435 if (vp->v_type != VBAD) { 436 vgonel(vp, p); 437 } else { 438 simple_unlock(&vp->v_interlock); 439 } 440 441#ifdef DIAGNOSTIC 442 { 443 int s; 444 445 if (vp->v_data) 446 panic("cleaned vnode isn't"); 447 s = splbio(); 448 if (vp->v_numoutput) 449 panic("Clean vnode has pending I/O's"); 450 splx(s); 451 } 452#endif 453 vp->v_flag = 0; 454 vp->v_lastr = 0; 455 vp->v_lastw = 0; 456 vp->v_lasta = 0; 457 vp->v_cstart = 0; 458 vp->v_clen = 0; 459 vp->v_socket = 0; 460 vp->v_writecount = 0; /* XXX */ 461 vp->v_maxio = 0; 462 } else { 463 simple_unlock(&vnode_free_list_slock); 464 vp = (struct vnode *) zalloc(vnode_zone); 465 bzero((char *) vp, sizeof *vp); 466 simple_lock_init(&vp->v_interlock); 467 vp->v_dd = vp; 468 cache_purge(vp); 469 LIST_INIT(&vp->v_cache_src); 470 TAILQ_INIT(&vp->v_cache_dst); 471 numvnodes++; 472 } 473 474 vp->v_type = VNON; 475 vp->v_tag = tag; 476 vp->v_op = vops; 477 insmntque(vp, mp); 478 *vpp = vp; 479 vp->v_usecount = 1; 480 vp->v_data = 0; 481 splx(s); 482 return (0); 483} 484 485/* 486 * Move a vnode from one mount queue to another. 487 */ 488static void 489insmntque(vp, mp) 490 register struct vnode *vp; 491 register struct mount *mp; 492{ 493 494 simple_lock(&mntvnode_slock); 495 /* 496 * Delete from old mount point vnode list, if on one. 497 */ 498 if (vp->v_mount != NULL) 499 LIST_REMOVE(vp, v_mntvnodes); 500 /* 501 * Insert into list of vnodes for the new mount point, if available. 502 */ 503 if ((vp->v_mount = mp) == NULL) { 504 simple_unlock(&mntvnode_slock); 505 return; 506 } 507 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 508 simple_unlock(&mntvnode_slock); 509} 510 511/* 512 * Update outstanding I/O count and do wakeup if requested. 513 */ 514void 515vwakeup(bp) 516 register struct buf *bp; 517{ 518 register struct vnode *vp; 519 520 bp->b_flags &= ~B_WRITEINPROG; 521 if ((vp = bp->b_vp)) { 522 vp->v_numoutput--; 523 if (vp->v_numoutput < 0) 524 panic("vwakeup: neg numoutput"); 525 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 526 vp->v_flag &= ~VBWAIT; 527 wakeup((caddr_t) &vp->v_numoutput); 528 } 529 } 530} 531 532/* 533 * Flush out and invalidate all buffers associated with a vnode. 534 * Called with the underlying object locked. 535 */ 536int 537vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 538 register struct vnode *vp; 539 int flags; 540 struct ucred *cred; 541 struct proc *p; 542 int slpflag, slptimeo; 543{ 544 register struct buf *bp; 545 struct buf *nbp, *blist; 546 int s, error; 547 vm_object_t object; 548 549 if (flags & V_SAVE) { 550 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p))) 551 return (error); 552 if (vp->v_dirtyblkhd.lh_first != NULL) 553 panic("vinvalbuf: dirty bufs"); 554 } 555 556 s = splbio(); 557 for (;;) { 558 if ((blist = vp->v_cleanblkhd.lh_first) && (flags & V_SAVEMETA)) 559 while (blist && blist->b_lblkno < 0) 560 blist = blist->b_vnbufs.le_next; 561 if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 562 (flags & V_SAVEMETA)) 563 while (blist && blist->b_lblkno < 0) 564 blist = blist->b_vnbufs.le_next; 565 if (!blist) 566 break; 567 568 for (bp = blist; bp; bp = nbp) { 569 nbp = bp->b_vnbufs.le_next; 570 if ((flags & V_SAVEMETA) && bp->b_lblkno < 0) 571 continue; 572 if (bp->b_flags & B_BUSY) { 573 bp->b_flags |= B_WANTED; 574 error = tsleep((caddr_t) bp, 575 slpflag | (PRIBIO + 1), "vinvalbuf", 576 slptimeo); 577 if (error) { 578 splx(s); 579 return (error); 580 } 581 break; 582 } 583 bremfree(bp); 584 bp->b_flags |= B_BUSY; 585 /* 586 * XXX Since there are no node locks for NFS, I 587 * believe there is a slight chance that a delayed 588 * write will occur while sleeping just above, so 589 * check for it. 590 */ 591 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 592 if (bp->b_vp == vp) { 593 if (bp->b_flags & B_CLUSTEROK) { 594 vfs_bio_awrite(bp); 595 } else { 596 bp->b_flags |= B_ASYNC; 597 VOP_BWRITE(bp); 598 } 599 } else { 600 (void) VOP_BWRITE(bp); 601 } 602 break; 603 } 604 bp->b_flags |= (B_INVAL|B_NOCACHE|B_RELBUF); 605 brelse(bp); 606 } 607 } 608 609 while (vp->v_numoutput > 0) { 610 vp->v_flag |= VBWAIT; 611 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 612 } 613 614 splx(s); 615 616 /* 617 * Destroy the copy in the VM cache, too. 618 */ 619 simple_lock(&vp->v_interlock); 620 object = vp->v_object; 621 if (object != NULL) { 622 if (flags & V_SAVEMETA) 623 vm_object_page_remove(object, 0, object->size, 624 (flags & V_SAVE) ? TRUE : FALSE); 625 else 626 vm_object_page_remove(object, 0, 0, 627 (flags & V_SAVE) ? TRUE : FALSE); 628 } 629 simple_unlock(&vp->v_interlock); 630 631 if (!(flags & V_SAVEMETA) && 632 (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) 633 panic("vinvalbuf: flush failed"); 634 return (0); 635} 636 637/* 638 * Associate a buffer with a vnode. 639 */ 640void 641bgetvp(vp, bp) 642 register struct vnode *vp; 643 register struct buf *bp; 644{ 645 int s; 646 647#if defined(DIAGNOSTIC) 648 if (bp->b_vp) 649 panic("bgetvp: not free"); 650#endif 651 vhold(vp); 652 bp->b_vp = vp; 653 if (vp->v_type == VBLK || vp->v_type == VCHR) 654 bp->b_dev = vp->v_rdev; 655 else 656 bp->b_dev = NODEV; 657 /* 658 * Insert onto list for new vnode. 659 */ 660 s = splbio(); 661 bufinsvn(bp, &vp->v_cleanblkhd); 662 splx(s); 663} 664 665/* 666 * Disassociate a buffer from a vnode. 667 */ 668void 669brelvp(bp) 670 register struct buf *bp; 671{ 672 struct vnode *vp; 673 int s; 674 675#if defined(DIAGNOSTIC) 676 if (bp->b_vp == (struct vnode *) 0) 677 panic("brelvp: NULL"); 678#endif 679 680 /* 681 * Delete from old vnode list, if on one. 682 */ 683 s = splbio(); 684 if (bp->b_vnbufs.le_next != NOLIST) 685 bufremvn(bp); 686 splx(s); 687 688 vp = bp->b_vp; 689 bp->b_vp = (struct vnode *) 0; 690 vdrop(vp); 691} 692 693/* 694 * Associate a p-buffer with a vnode. 695 */ 696void 697pbgetvp(vp, bp) 698 register struct vnode *vp; 699 register struct buf *bp; 700{ 701#if defined(DIAGNOSTIC) 702 if (bp->b_vp) 703 panic("pbgetvp: not free"); 704#endif 705 bp->b_vp = vp; 706 if (vp->v_type == VBLK || vp->v_type == VCHR) 707 bp->b_dev = vp->v_rdev; 708 else 709 bp->b_dev = NODEV; 710} 711 712/* 713 * Disassociate a p-buffer from a vnode. 714 */ 715void 716pbrelvp(bp) 717 register struct buf *bp; 718{ 719 720#if defined(DIAGNOSTIC) 721 if (bp->b_vp == (struct vnode *) 0) 722 panic("pbrelvp: NULL"); 723#endif 724 725 bp->b_vp = (struct vnode *) 0; 726} 727 728/* 729 * Reassign a buffer from one vnode to another. 730 * Used to assign file specific control information 731 * (indirect blocks) to the vnode to which they belong. 732 */ 733void 734reassignbuf(bp, newvp) 735 register struct buf *bp; 736 register struct vnode *newvp; 737{ 738 int s; 739 740 if (newvp == NULL) { 741 printf("reassignbuf: NULL"); 742 return; 743 } 744 745 s = splbio(); 746 /* 747 * Delete from old vnode list, if on one. 748 */ 749 if (bp->b_vnbufs.le_next != NOLIST) { 750 bufremvn(bp); 751 vdrop(bp->b_vp); 752 } 753 /* 754 * If dirty, put on list of dirty buffers; otherwise insert onto list 755 * of clean buffers. 756 */ 757 if (bp->b_flags & B_DELWRI) { 758 struct buf *tbp; 759 760 tbp = newvp->v_dirtyblkhd.lh_first; 761 if (!tbp || (tbp->b_lblkno > bp->b_lblkno)) { 762 bufinsvn(bp, &newvp->v_dirtyblkhd); 763 } else { 764 while (tbp->b_vnbufs.le_next && 765 (tbp->b_vnbufs.le_next->b_lblkno < bp->b_lblkno)) { 766 tbp = tbp->b_vnbufs.le_next; 767 } 768 LIST_INSERT_AFTER(tbp, bp, b_vnbufs); 769 } 770 } else { 771 bufinsvn(bp, &newvp->v_cleanblkhd); 772 } 773 bp->b_vp = newvp; 774 vhold(bp->b_vp); 775 splx(s); 776} 777 778#ifndef DEVFS_ROOT 779/* 780 * Create a vnode for a block device. 781 * Used for mounting the root file system. 782 */ 783int 784bdevvp(dev, vpp) 785 dev_t dev; 786 struct vnode **vpp; 787{ 788 register struct vnode *vp; 789 struct vnode *nvp; 790 int error; 791 792 if (dev == NODEV) 793 return (0); 794 error = getnewvnode(VT_NON, (struct mount *) 0, spec_vnodeop_p, &nvp); 795 if (error) { 796 *vpp = 0; 797 return (error); 798 } 799 vp = nvp; 800 vp->v_type = VBLK; 801 if ((nvp = checkalias(vp, dev, (struct mount *) 0))) { 802 vput(vp); 803 vp = nvp; 804 } 805 *vpp = vp; 806 return (0); 807} 808#endif /* !DEVFS_ROOT */ 809 810/* 811 * Check to see if the new vnode represents a special device 812 * for which we already have a vnode (either because of 813 * bdevvp() or because of a different vnode representing 814 * the same block device). If such an alias exists, deallocate 815 * the existing contents and return the aliased vnode. The 816 * caller is responsible for filling it with its new contents. 817 */ 818struct vnode * 819checkalias(nvp, nvp_rdev, mp) 820 register struct vnode *nvp; 821 dev_t nvp_rdev; 822 struct mount *mp; 823{ 824 struct proc *p = curproc; /* XXX */ 825 struct vnode *vp; 826 struct vnode **vpp; 827 828 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 829 return (NULLVP); 830 831 vpp = &speclisth[SPECHASH(nvp_rdev)]; 832loop: 833 simple_lock(&spechash_slock); 834 for (vp = *vpp; vp; vp = vp->v_specnext) { 835 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 836 continue; 837 /* 838 * Alias, but not in use, so flush it out. 839 */ 840 simple_lock(&vp->v_interlock); 841 if (vp->v_usecount == 0) { 842 simple_unlock(&spechash_slock); 843 vgonel(vp, p); 844 goto loop; 845 } 846 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { 847 simple_unlock(&spechash_slock); 848 goto loop; 849 } 850 break; 851 } 852 if (vp == NULL || vp->v_tag != VT_NON) { 853 MALLOC(nvp->v_specinfo, struct specinfo *, 854 sizeof(struct specinfo), M_VNODE, M_WAITOK); 855 nvp->v_rdev = nvp_rdev; 856 nvp->v_hashchain = vpp; 857 nvp->v_specnext = *vpp; 858 nvp->v_specflags = 0; 859 simple_unlock(&spechash_slock); 860 *vpp = nvp; 861 if (vp != NULLVP) { 862 nvp->v_flag |= VALIASED; 863 vp->v_flag |= VALIASED; 864 vput(vp); 865 } 866 return (NULLVP); 867 } 868 simple_unlock(&spechash_slock); 869 VOP_UNLOCK(vp, 0, p); 870 simple_lock(&vp->v_interlock); 871 vclean(vp, 0, p); 872 vp->v_op = nvp->v_op; 873 vp->v_tag = nvp->v_tag; 874 nvp->v_type = VNON; 875 insmntque(vp, mp); 876 return (vp); 877} 878 879/* 880 * Grab a particular vnode from the free list, increment its 881 * reference count and lock it. The vnode lock bit is set the 882 * vnode is being eliminated in vgone. The process is awakened 883 * when the transition is completed, and an error returned to 884 * indicate that the vnode is no longer usable (possibly having 885 * been changed to a new file system type). 886 */ 887int 888vget(vp, flags, p) 889 register struct vnode *vp; 890 int flags; 891 struct proc *p; 892{ 893 int error; 894 895 /* 896 * If the vnode is in the process of being cleaned out for 897 * another use, we wait for the cleaning to finish and then 898 * return failure. Cleaning is determined by checking that 899 * the VXLOCK flag is set. 900 */ 901 if ((flags & LK_INTERLOCK) == 0) { 902 simple_lock(&vp->v_interlock); 903 } 904 if (vp->v_flag & VXLOCK) { 905 vp->v_flag |= VXWANT; 906 simple_unlock(&vp->v_interlock); 907 tsleep((caddr_t)vp, PINOD, "vget", 0); 908 return (ENOENT); 909 } 910 911 vp->v_usecount++; 912 913 if (VSHOULDBUSY(vp)) 914 vbusy(vp); 915 /* 916 * Create the VM object, if needed 917 */ 918 if ((flags & LK_NOOBJ) == 0 && 919 (vp->v_type == VREG) && 920 ((vp->v_object == NULL) || 921 (vp->v_object->flags & OBJ_DEAD))) { 922 /* 923 * XXX 924 * When the object is locked with shared lock, VOP_ISLOCKED() 925 * returns true. 926 */ 927 if (VOP_ISLOCKED(vp)) { 928 simple_unlock(&vp->v_interlock); 929 vfs_object_create(vp, curproc, curproc->p_ucred, 1); 930 } else 931 vfs_object_create(vp, curproc, curproc->p_ucred, 0); 932 simple_lock(&vp->v_interlock); 933 } 934 if (flags & LK_TYPE_MASK) { 935 if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) 936 vrele(vp); 937 return (error); 938 } 939 simple_unlock(&vp->v_interlock); 940 return (0); 941} 942 943void 944vref(struct vnode *vp) 945{ 946 simple_lock(&vp->v_interlock); 947 vp->v_usecount++; 948 simple_unlock(&vp->v_interlock); 949} 950 951/* 952 * Vnode put/release. 953 * If count drops to zero, call inactive routine and return to freelist. 954 */ 955void 956vrele(vp) 957 struct vnode *vp; 958{ 959 struct proc *p = curproc; /* XXX */ 960 961#ifdef DIAGNOSTIC 962 if (vp == NULL) 963 panic("vrele: null vp"); 964#endif 965 simple_lock(&vp->v_interlock); 966 967 if (vp->v_usecount > 1) { 968 969 vp->v_usecount--; 970 simple_unlock(&vp->v_interlock); 971 972 return; 973 } 974 975 if (vp->v_usecount == 1) { 976 977 vp->v_usecount--; 978 979 if (VSHOULDFREE(vp)) 980 vfree(vp); 981 /* 982 * If we are doing a vput, the node is already locked, and we must 983 * call VOP_INACTIVE with the node locked. So, in the case of 984 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 985 */ 986 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 987 VOP_INACTIVE(vp, p); 988 } 989 990 } else { 991#ifdef DIAGNOSTIC 992 vprint("vrele: negative ref count", vp); 993 simple_unlock(&vp->v_interlock); 994#endif 995 panic("vrele: negative ref cnt"); 996 } 997} 998 999void 1000vput(vp) 1001 struct vnode *vp; 1002{ 1003 struct proc *p = curproc; /* XXX */ 1004 1005#ifdef DIAGNOSTIC 1006 if (vp == NULL) 1007 panic("vput: null vp"); 1008#endif 1009 1010 simple_lock(&vp->v_interlock); 1011 1012 if (vp->v_usecount > 1) { 1013 1014 vp->v_usecount--; 1015 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1016 return; 1017 1018 } 1019 1020 if (vp->v_usecount == 1) { 1021 1022 vp->v_usecount--; 1023 if (VSHOULDFREE(vp)) 1024 vfree(vp); 1025 /* 1026 * If we are doing a vput, the node is already locked, and we must 1027 * call VOP_INACTIVE with the node locked. So, in the case of 1028 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1029 */ 1030 simple_unlock(&vp->v_interlock); 1031 VOP_INACTIVE(vp, p); 1032 1033 } else { 1034#ifdef DIAGNOSTIC 1035 vprint("vput: negative ref count", vp); 1036#endif 1037 panic("vput: negative ref cnt"); 1038 } 1039} 1040 1041/* 1042 * Somebody doesn't want the vnode recycled. 1043 */ 1044void 1045vhold(vp) 1046 register struct vnode *vp; 1047{ 1048 1049 simple_lock(&vp->v_interlock); 1050 vp->v_holdcnt++; 1051 if (VSHOULDBUSY(vp)) 1052 vbusy(vp); 1053 simple_unlock(&vp->v_interlock); 1054} 1055 1056/* 1057 * One less who cares about this vnode. 1058 */ 1059void 1060vdrop(vp) 1061 register struct vnode *vp; 1062{ 1063 1064 simple_lock(&vp->v_interlock); 1065 if (vp->v_holdcnt <= 0) 1066 panic("holdrele: holdcnt"); 1067 vp->v_holdcnt--; 1068 if (VSHOULDFREE(vp)) 1069 vfree(vp); 1070 simple_unlock(&vp->v_interlock); 1071} 1072 1073/* 1074 * Remove any vnodes in the vnode table belonging to mount point mp. 1075 * 1076 * If MNT_NOFORCE is specified, there should not be any active ones, 1077 * return error if any are found (nb: this is a user error, not a 1078 * system error). If MNT_FORCE is specified, detach any active vnodes 1079 * that are found. 1080 */ 1081#ifdef DIAGNOSTIC 1082static int busyprt = 0; /* print out busy vnodes */ 1083SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1084#endif 1085 1086int 1087vflush(mp, skipvp, flags) 1088 struct mount *mp; 1089 struct vnode *skipvp; 1090 int flags; 1091{ 1092 struct proc *p = curproc; /* XXX */ 1093 struct vnode *vp, *nvp; 1094 int busy = 0; 1095 1096 simple_lock(&mntvnode_slock); 1097loop: 1098 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 1099 /* 1100 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1101 * Start over if it has (it won't be on the list anymore). 1102 */ 1103 if (vp->v_mount != mp) 1104 goto loop; 1105 nvp = vp->v_mntvnodes.le_next; 1106 /* 1107 * Skip over a selected vnode. 1108 */ 1109 if (vp == skipvp) 1110 continue; 1111 1112 simple_lock(&vp->v_interlock); 1113 /* 1114 * Skip over a vnodes marked VSYSTEM. 1115 */ 1116 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1117 simple_unlock(&vp->v_interlock); 1118 continue; 1119 } 1120 /* 1121 * If WRITECLOSE is set, only flush out regular file vnodes 1122 * open for writing. 1123 */ 1124 if ((flags & WRITECLOSE) && 1125 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1126 simple_unlock(&vp->v_interlock); 1127 continue; 1128 } 1129 1130 /* 1131 * With v_usecount == 0, all we need to do is clear out the 1132 * vnode data structures and we are done. 1133 */ 1134 if (vp->v_usecount == 0) { 1135 simple_unlock(&mntvnode_slock); 1136 vgonel(vp, p); 1137 simple_lock(&mntvnode_slock); 1138 continue; 1139 } 1140 1141 /* 1142 * If FORCECLOSE is set, forcibly close the vnode. For block 1143 * or character devices, revert to an anonymous device. For 1144 * all other files, just kill them. 1145 */ 1146 if (flags & FORCECLOSE) { 1147 simple_unlock(&mntvnode_slock); 1148 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1149 vgonel(vp, p); 1150 } else { 1151 vclean(vp, 0, p); 1152 vp->v_op = spec_vnodeop_p; 1153 insmntque(vp, (struct mount *) 0); 1154 } 1155 simple_lock(&mntvnode_slock); 1156 continue; 1157 } 1158#ifdef DIAGNOSTIC 1159 if (busyprt) 1160 vprint("vflush: busy vnode", vp); 1161#endif 1162 simple_unlock(&vp->v_interlock); 1163 busy++; 1164 } 1165 simple_unlock(&mntvnode_slock); 1166 if (busy) 1167 return (EBUSY); 1168 return (0); 1169} 1170 1171/* 1172 * Disassociate the underlying file system from a vnode. 1173 */ 1174static void 1175vclean(vp, flags, p) 1176 struct vnode *vp; 1177 int flags; 1178 struct proc *p; 1179{ 1180 int active; 1181 vm_object_t obj; 1182 1183 /* 1184 * Check to see if the vnode is in use. If so we have to reference it 1185 * before we clean it out so that its count cannot fall to zero and 1186 * generate a race against ourselves to recycle it. 1187 */ 1188 if ((active = vp->v_usecount)) 1189 vp->v_usecount++; 1190 1191 /* 1192 * Prevent the vnode from being recycled or brought into use while we 1193 * clean it out. 1194 */ 1195 if (vp->v_flag & VXLOCK) 1196 panic("vclean: deadlock"); 1197 vp->v_flag |= VXLOCK; 1198 /* 1199 * Even if the count is zero, the VOP_INACTIVE routine may still 1200 * have the object locked while it cleans it out. The VOP_LOCK 1201 * ensures that the VOP_INACTIVE routine is done with its work. 1202 * For active vnodes, it ensures that no other activity can 1203 * occur while the underlying object is being cleaned out. 1204 */ 1205 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1206 1207 /* 1208 * Clean out any buffers associated with the vnode. 1209 */ 1210 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1211 if (obj = vp->v_object) { 1212 if (obj->ref_count == 0) { 1213 /* 1214 * This is a normal way of shutting down the object/vnode 1215 * association. 1216 */ 1217 vm_object_terminate(obj); 1218 } else { 1219 /* 1220 * Woe to the process that tries to page now :-). 1221 */ 1222 vm_pager_deallocate(obj); 1223 } 1224 } 1225 1226 /* 1227 * If purging an active vnode, it must be closed and 1228 * deactivated before being reclaimed. Note that the 1229 * VOP_INACTIVE will unlock the vnode. 1230 */ 1231 if (active) { 1232 if (flags & DOCLOSE) 1233 VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); 1234 VOP_INACTIVE(vp, p); 1235 } else { 1236 /* 1237 * Any other processes trying to obtain this lock must first 1238 * wait for VXLOCK to clear, then call the new lock operation. 1239 */ 1240 VOP_UNLOCK(vp, 0, p); 1241 } 1242 /* 1243 * Reclaim the vnode. 1244 */ 1245 if (VOP_RECLAIM(vp, p)) 1246 panic("vclean: cannot reclaim"); 1247 if (active) 1248 vrele(vp); 1249 cache_purge(vp); 1250 if (vp->v_vnlock) { 1251#if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */ 1252#ifdef DIAGNOSTIC 1253 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0) 1254 vprint("vclean: lock not drained", vp); 1255#endif 1256#endif 1257 FREE(vp->v_vnlock, M_VNODE); 1258 vp->v_vnlock = NULL; 1259 } 1260 1261 /* 1262 * Done with purge, notify sleepers of the grim news. 1263 */ 1264 vp->v_op = dead_vnodeop_p; 1265 vn_pollgone(vp); 1266 vp->v_tag = VT_NON; 1267 vp->v_flag &= ~VXLOCK; 1268 if (vp->v_flag & VXWANT) { 1269 vp->v_flag &= ~VXWANT; 1270 wakeup((caddr_t) vp); 1271 } 1272} 1273 1274/* 1275 * Eliminate all activity associated with the requested vnode 1276 * and with all vnodes aliased to the requested vnode. 1277 */ 1278int 1279vop_revoke(ap) 1280 struct vop_revoke_args /* { 1281 struct vnode *a_vp; 1282 int a_flags; 1283 } */ *ap; 1284{ 1285 struct vnode *vp, *vq; 1286 struct proc *p = curproc; /* XXX */ 1287 1288#ifdef DIAGNOSTIC 1289 if ((ap->a_flags & REVOKEALL) == 0) 1290 panic("vop_revoke"); 1291#endif 1292 1293 vp = ap->a_vp; 1294 simple_lock(&vp->v_interlock); 1295 1296 if (vp->v_flag & VALIASED) { 1297 /* 1298 * If a vgone (or vclean) is already in progress, 1299 * wait until it is done and return. 1300 */ 1301 if (vp->v_flag & VXLOCK) { 1302 vp->v_flag |= VXWANT; 1303 simple_unlock(&vp->v_interlock); 1304 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1305 return (0); 1306 } 1307 /* 1308 * Ensure that vp will not be vgone'd while we 1309 * are eliminating its aliases. 1310 */ 1311 vp->v_flag |= VXLOCK; 1312 simple_unlock(&vp->v_interlock); 1313 while (vp->v_flag & VALIASED) { 1314 simple_lock(&spechash_slock); 1315 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1316 if (vq->v_rdev != vp->v_rdev || 1317 vq->v_type != vp->v_type || vp == vq) 1318 continue; 1319 simple_unlock(&spechash_slock); 1320 vgone(vq); 1321 break; 1322 } 1323 if (vq == NULLVP) { 1324 simple_unlock(&spechash_slock); 1325 } 1326 } 1327 /* 1328 * Remove the lock so that vgone below will 1329 * really eliminate the vnode after which time 1330 * vgone will awaken any sleepers. 1331 */ 1332 simple_lock(&vp->v_interlock); 1333 vp->v_flag &= ~VXLOCK; 1334 if (vp->v_flag & VXWANT) { 1335 vp->v_flag &= ~VXWANT; 1336 wakeup(vp); 1337 } 1338 } 1339 vgonel(vp, p); 1340 return (0); 1341} 1342 1343/* 1344 * Recycle an unused vnode to the front of the free list. 1345 * Release the passed interlock if the vnode will be recycled. 1346 */ 1347int 1348vrecycle(vp, inter_lkp, p) 1349 struct vnode *vp; 1350 struct simplelock *inter_lkp; 1351 struct proc *p; 1352{ 1353 1354 simple_lock(&vp->v_interlock); 1355 if (vp->v_usecount == 0) { 1356 if (inter_lkp) { 1357 simple_unlock(inter_lkp); 1358 } 1359 vgonel(vp, p); 1360 return (1); 1361 } 1362 simple_unlock(&vp->v_interlock); 1363 return (0); 1364} 1365 1366/* 1367 * Eliminate all activity associated with a vnode 1368 * in preparation for reuse. 1369 */ 1370void 1371vgone(vp) 1372 register struct vnode *vp; 1373{ 1374 struct proc *p = curproc; /* XXX */ 1375 1376 simple_lock(&vp->v_interlock); 1377 vgonel(vp, p); 1378} 1379 1380/* 1381 * vgone, with the vp interlock held. 1382 */ 1383static void 1384vgonel(vp, p) 1385 struct vnode *vp; 1386 struct proc *p; 1387{ 1388 int s; 1389 struct vnode *vq; 1390 struct vnode *vx; 1391 1392 /* 1393 * If a vgone (or vclean) is already in progress, 1394 * wait until it is done and return. 1395 */ 1396 if (vp->v_flag & VXLOCK) { 1397 vp->v_flag |= VXWANT; 1398 simple_unlock(&vp->v_interlock); 1399 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1400 return; 1401 } 1402 1403 /* 1404 * Clean out the filesystem specific data. 1405 */ 1406 vclean(vp, DOCLOSE, p); 1407 simple_lock(&vp->v_interlock); 1408 1409 /* 1410 * Delete from old mount point vnode list, if on one. 1411 */ 1412 if (vp->v_mount != NULL) 1413 insmntque(vp, (struct mount *)0); 1414 /* 1415 * If special device, remove it from special device alias list 1416 * if it is on one. 1417 */ 1418 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1419 simple_lock(&spechash_slock); 1420 if (*vp->v_hashchain == vp) { 1421 *vp->v_hashchain = vp->v_specnext; 1422 } else { 1423 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1424 if (vq->v_specnext != vp) 1425 continue; 1426 vq->v_specnext = vp->v_specnext; 1427 break; 1428 } 1429 if (vq == NULL) 1430 panic("missing bdev"); 1431 } 1432 if (vp->v_flag & VALIASED) { 1433 vx = NULL; 1434 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1435 if (vq->v_rdev != vp->v_rdev || 1436 vq->v_type != vp->v_type) 1437 continue; 1438 if (vx) 1439 break; 1440 vx = vq; 1441 } 1442 if (vx == NULL) 1443 panic("missing alias"); 1444 if (vq == NULL) 1445 vx->v_flag &= ~VALIASED; 1446 vp->v_flag &= ~VALIASED; 1447 } 1448 simple_unlock(&spechash_slock); 1449 FREE(vp->v_specinfo, M_VNODE); 1450 vp->v_specinfo = NULL; 1451 } 1452 1453 /* 1454 * If it is on the freelist and not already at the head, 1455 * move it to the head of the list. The test of the back 1456 * pointer and the reference count of zero is because 1457 * it will be removed from the free list by getnewvnode, 1458 * but will not have its reference count incremented until 1459 * after calling vgone. If the reference count were 1460 * incremented first, vgone would (incorrectly) try to 1461 * close the previous instance of the underlying object. 1462 */ 1463 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1464 s = splbio(); 1465 simple_lock(&vnode_free_list_slock); 1466 if (vp->v_flag & VFREE) { 1467 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1468 } else if (vp->v_flag & VTBFREE) { 1469 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 1470 vp->v_flag &= ~VTBFREE; 1471 freevnodes++; 1472 } else 1473 freevnodes++; 1474 vp->v_flag |= VFREE; 1475 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1476 simple_unlock(&vnode_free_list_slock); 1477 splx(s); 1478 } 1479 1480 vp->v_type = VBAD; 1481 simple_unlock(&vp->v_interlock); 1482} 1483 1484/* 1485 * Lookup a vnode by device number. 1486 */ 1487int 1488vfinddev(dev, type, vpp) 1489 dev_t dev; 1490 enum vtype type; 1491 struct vnode **vpp; 1492{ 1493 register struct vnode *vp; 1494 int rc = 0; 1495 1496 simple_lock(&spechash_slock); 1497 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1498 if (dev != vp->v_rdev || type != vp->v_type) 1499 continue; 1500 *vpp = vp; 1501 rc = 1; 1502 break; 1503 } 1504 simple_unlock(&spechash_slock); 1505 return (rc); 1506} 1507 1508/* 1509 * Calculate the total number of references to a special device. 1510 */ 1511int 1512vcount(vp) 1513 register struct vnode *vp; 1514{ 1515 struct vnode *vq, *vnext; 1516 int count; 1517 1518loop: 1519 if ((vp->v_flag & VALIASED) == 0) 1520 return (vp->v_usecount); 1521 simple_lock(&spechash_slock); 1522 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1523 vnext = vq->v_specnext; 1524 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1525 continue; 1526 /* 1527 * Alias, but not in use, so flush it out. 1528 */ 1529 if (vq->v_usecount == 0 && vq != vp) { 1530 simple_unlock(&spechash_slock); 1531 vgone(vq); 1532 goto loop; 1533 } 1534 count += vq->v_usecount; 1535 } 1536 simple_unlock(&spechash_slock); 1537 return (count); 1538} 1539/* 1540 * Print out a description of a vnode. 1541 */ 1542static char *typename[] = 1543{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1544 1545void 1546vprint(label, vp) 1547 char *label; 1548 register struct vnode *vp; 1549{ 1550 char buf[64]; 1551 1552 if (label != NULL) 1553 printf("%s: %x: ", label, vp); 1554 else 1555 printf("%x: ", vp); 1556 printf("type %s, usecount %d, writecount %d, refcount %ld,", 1557 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1558 vp->v_holdcnt); 1559 buf[0] = '\0'; 1560 if (vp->v_flag & VROOT) 1561 strcat(buf, "|VROOT"); 1562 if (vp->v_flag & VTEXT) 1563 strcat(buf, "|VTEXT"); 1564 if (vp->v_flag & VSYSTEM) 1565 strcat(buf, "|VSYSTEM"); 1566 if (vp->v_flag & VXLOCK) 1567 strcat(buf, "|VXLOCK"); 1568 if (vp->v_flag & VXWANT) 1569 strcat(buf, "|VXWANT"); 1570 if (vp->v_flag & VBWAIT) 1571 strcat(buf, "|VBWAIT"); 1572 if (vp->v_flag & VALIASED) 1573 strcat(buf, "|VALIASED"); 1574 if (vp->v_flag & VDOOMED) 1575 strcat(buf, "|VDOOMED"); 1576 if (vp->v_flag & VFREE) 1577 strcat(buf, "|VFREE"); 1578 if (vp->v_flag & VOBJBUF) 1579 strcat(buf, "|VOBJBUF"); 1580 if (buf[0] != '\0') 1581 printf(" flags (%s)", &buf[1]); 1582 if (vp->v_data == NULL) { 1583 printf("\n"); 1584 } else { 1585 printf("\n\t"); 1586 VOP_PRINT(vp); 1587 } 1588} 1589 1590#ifdef DDB 1591/* 1592 * List all of the locked vnodes in the system. 1593 * Called when debugging the kernel. 1594 */ 1595static void 1596printlockedvnodes() 1597{ 1598 struct proc *p = curproc; /* XXX */ 1599 struct mount *mp, *nmp; 1600 struct vnode *vp; 1601 1602 printf("Locked vnodes\n"); 1603 simple_lock(&mountlist_slock); 1604 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1605 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1606 nmp = mp->mnt_list.cqe_next; 1607 continue; 1608 } 1609 for (vp = mp->mnt_vnodelist.lh_first; 1610 vp != NULL; 1611 vp = vp->v_mntvnodes.le_next) { 1612 if (VOP_ISLOCKED(vp)) 1613 vprint((char *)0, vp); 1614 } 1615 simple_lock(&mountlist_slock); 1616 nmp = mp->mnt_list.cqe_next; 1617 vfs_unbusy(mp, p); 1618 } 1619 simple_unlock(&mountlist_slock); 1620} 1621#endif 1622 1623/* 1624 * Top level filesystem related information gathering. 1625 */ 1626static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS); 1627 1628static int 1629vfs_sysctl SYSCTL_HANDLER_ARGS 1630{ 1631 int *name = (int *)arg1 - 1; /* XXX */ 1632 u_int namelen = arg2 + 1; /* XXX */ 1633 struct vfsconf *vfsp; 1634 1635#ifndef NO_COMPAT_PRELITE2 1636 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1637 if (namelen == 1) 1638 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1639#endif 1640 1641#ifdef notyet 1642 /* all sysctl names at this level are at least name and field */ 1643 if (namelen < 2) 1644 return (ENOTDIR); /* overloaded */ 1645 if (name[0] != VFS_GENERIC) { 1646 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1647 if (vfsp->vfc_typenum == name[0]) 1648 break; 1649 if (vfsp == NULL) 1650 return (EOPNOTSUPP); 1651 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1652 oldp, oldlenp, newp, newlen, p)); 1653 } 1654#endif 1655 switch (name[1]) { 1656 case VFS_MAXTYPENUM: 1657 if (namelen != 2) 1658 return (ENOTDIR); 1659 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1660 case VFS_CONF: 1661 if (namelen != 3) 1662 return (ENOTDIR); /* overloaded */ 1663 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1664 if (vfsp->vfc_typenum == name[2]) 1665 break; 1666 if (vfsp == NULL) 1667 return (EOPNOTSUPP); 1668 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1669 } 1670 return (EOPNOTSUPP); 1671} 1672 1673SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1674 "Generic filesystem"); 1675 1676#ifndef NO_COMPAT_PRELITE2 1677 1678static int 1679sysctl_ovfs_conf SYSCTL_HANDLER_ARGS 1680{ 1681 int error; 1682 struct vfsconf *vfsp; 1683 struct ovfsconf ovfs; 1684 1685 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1686 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1687 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1688 ovfs.vfc_index = vfsp->vfc_typenum; 1689 ovfs.vfc_refcount = vfsp->vfc_refcount; 1690 ovfs.vfc_flags = vfsp->vfc_flags; 1691 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1692 if (error) 1693 return error; 1694 } 1695 return 0; 1696} 1697 1698#endif /* !NO_COMPAT_PRELITE2 */ 1699 1700static volatile int kinfo_vdebug = 1; 1701 1702#if 0 1703#define KINFO_VNODESLOP 10 1704/* 1705 * Dump vnode list (via sysctl). 1706 * Copyout address of vnode followed by vnode. 1707 */ 1708/* ARGSUSED */ 1709static int 1710sysctl_vnode SYSCTL_HANDLER_ARGS 1711{ 1712 struct proc *p = curproc; /* XXX */ 1713 struct mount *mp, *nmp; 1714 struct vnode *nvp, *vp; 1715 int error; 1716 1717#define VPTRSZ sizeof (struct vnode *) 1718#define VNODESZ sizeof (struct vnode) 1719 1720 req->lock = 0; 1721 if (!req->oldptr) /* Make an estimate */ 1722 return (SYSCTL_OUT(req, 0, 1723 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 1724 1725 simple_lock(&mountlist_slock); 1726 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1727 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1728 nmp = mp->mnt_list.cqe_next; 1729 continue; 1730 } 1731again: 1732 simple_lock(&mntvnode_slock); 1733 for (vp = mp->mnt_vnodelist.lh_first; 1734 vp != NULL; 1735 vp = nvp) { 1736 /* 1737 * Check that the vp is still associated with 1738 * this filesystem. RACE: could have been 1739 * recycled onto the same filesystem. 1740 */ 1741 if (vp->v_mount != mp) { 1742 simple_unlock(&mntvnode_slock); 1743 if (kinfo_vdebug) 1744 printf("kinfo: vp changed\n"); 1745 goto again; 1746 } 1747 nvp = vp->v_mntvnodes.le_next; 1748 simple_unlock(&mntvnode_slock); 1749 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 1750 (error = SYSCTL_OUT(req, vp, VNODESZ))) 1751 return (error); 1752 simple_lock(&mntvnode_slock); 1753 } 1754 simple_unlock(&mntvnode_slock); 1755 simple_lock(&mountlist_slock); 1756 nmp = mp->mnt_list.cqe_next; 1757 vfs_unbusy(mp, p); 1758 } 1759 simple_unlock(&mountlist_slock); 1760 1761 return (0); 1762} 1763#endif 1764 1765/* 1766 * XXX 1767 * Exporting the vnode list on large systems causes them to crash. 1768 * Exporting the vnode list on medium systems causes sysctl to coredump. 1769 */ 1770#if 0 1771SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 1772 0, 0, sysctl_vnode, "S,vnode", ""); 1773#endif 1774 1775/* 1776 * Check to see if a filesystem is mounted on a block device. 1777 */ 1778int 1779vfs_mountedon(vp) 1780 struct vnode *vp; 1781{ 1782 struct vnode *vq; 1783 int error = 0; 1784 1785 if (vp->v_specflags & SI_MOUNTEDON) 1786 return (EBUSY); 1787 if (vp->v_flag & VALIASED) { 1788 simple_lock(&spechash_slock); 1789 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1790 if (vq->v_rdev != vp->v_rdev || 1791 vq->v_type != vp->v_type) 1792 continue; 1793 if (vq->v_specflags & SI_MOUNTEDON) { 1794 error = EBUSY; 1795 break; 1796 } 1797 } 1798 simple_unlock(&spechash_slock); 1799 } 1800 return (error); 1801} 1802 1803/* 1804 * Unmount all filesystems. The list is traversed in reverse order 1805 * of mounting to avoid dependencies. 1806 */ 1807void 1808vfs_unmountall() 1809{ 1810 struct mount *mp, *nmp; 1811 struct proc *p = initproc; /* XXX XXX should this be proc0? */ 1812 int error; 1813 1814 /* 1815 * Since this only runs when rebooting, it is not interlocked. 1816 */ 1817 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 1818 nmp = mp->mnt_list.cqe_prev; 1819 error = dounmount(mp, MNT_FORCE, p); 1820 if (error) { 1821 printf("unmount of %s failed (", 1822 mp->mnt_stat.f_mntonname); 1823 if (error == EBUSY) 1824 printf("BUSY)\n"); 1825 else 1826 printf("%d)\n", error); 1827 } 1828 } 1829} 1830 1831/* 1832 * Build hash lists of net addresses and hang them off the mount point. 1833 * Called by ufs_mount() to set up the lists of export addresses. 1834 */ 1835static int 1836vfs_hang_addrlist(mp, nep, argp) 1837 struct mount *mp; 1838 struct netexport *nep; 1839 struct export_args *argp; 1840{ 1841 register struct netcred *np; 1842 register struct radix_node_head *rnh; 1843 register int i; 1844 struct radix_node *rn; 1845 struct sockaddr *saddr, *smask = 0; 1846 struct domain *dom; 1847 int error; 1848 1849 if (argp->ex_addrlen == 0) { 1850 if (mp->mnt_flag & MNT_DEFEXPORTED) 1851 return (EPERM); 1852 np = &nep->ne_defexported; 1853 np->netc_exflags = argp->ex_flags; 1854 np->netc_anon = argp->ex_anon; 1855 np->netc_anon.cr_ref = 1; 1856 mp->mnt_flag |= MNT_DEFEXPORTED; 1857 return (0); 1858 } 1859 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1860 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1861 bzero((caddr_t) np, i); 1862 saddr = (struct sockaddr *) (np + 1); 1863 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1864 goto out; 1865 if (saddr->sa_len > argp->ex_addrlen) 1866 saddr->sa_len = argp->ex_addrlen; 1867 if (argp->ex_masklen) { 1868 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 1869 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 1870 if (error) 1871 goto out; 1872 if (smask->sa_len > argp->ex_masklen) 1873 smask->sa_len = argp->ex_masklen; 1874 } 1875 i = saddr->sa_family; 1876 if ((rnh = nep->ne_rtable[i]) == 0) { 1877 /* 1878 * Seems silly to initialize every AF when most are not used, 1879 * do so on demand here 1880 */ 1881 for (dom = domains; dom; dom = dom->dom_next) 1882 if (dom->dom_family == i && dom->dom_rtattach) { 1883 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1884 dom->dom_rtoffset); 1885 break; 1886 } 1887 if ((rnh = nep->ne_rtable[i]) == 0) { 1888 error = ENOBUFS; 1889 goto out; 1890 } 1891 } 1892 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 1893 np->netc_rnodes); 1894 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1895 error = EPERM; 1896 goto out; 1897 } 1898 np->netc_exflags = argp->ex_flags; 1899 np->netc_anon = argp->ex_anon; 1900 np->netc_anon.cr_ref = 1; 1901 return (0); 1902out: 1903 free(np, M_NETADDR); 1904 return (error); 1905} 1906 1907/* ARGSUSED */ 1908static int 1909vfs_free_netcred(rn, w) 1910 struct radix_node *rn; 1911 void *w; 1912{ 1913 register struct radix_node_head *rnh = (struct radix_node_head *) w; 1914 1915 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1916 free((caddr_t) rn, M_NETADDR); 1917 return (0); 1918} 1919 1920/* 1921 * Free the net address hash lists that are hanging off the mount points. 1922 */ 1923static void 1924vfs_free_addrlist(nep) 1925 struct netexport *nep; 1926{ 1927 register int i; 1928 register struct radix_node_head *rnh; 1929 1930 for (i = 0; i <= AF_MAX; i++) 1931 if ((rnh = nep->ne_rtable[i])) { 1932 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1933 (caddr_t) rnh); 1934 free((caddr_t) rnh, M_RTABLE); 1935 nep->ne_rtable[i] = 0; 1936 } 1937} 1938 1939int 1940vfs_export(mp, nep, argp) 1941 struct mount *mp; 1942 struct netexport *nep; 1943 struct export_args *argp; 1944{ 1945 int error; 1946 1947 if (argp->ex_flags & MNT_DELEXPORT) { 1948 if (mp->mnt_flag & MNT_EXPUBLIC) { 1949 vfs_setpublicfs(NULL, NULL, NULL); 1950 mp->mnt_flag &= ~MNT_EXPUBLIC; 1951 } 1952 vfs_free_addrlist(nep); 1953 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1954 } 1955 if (argp->ex_flags & MNT_EXPORTED) { 1956 if (argp->ex_flags & MNT_EXPUBLIC) { 1957 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1958 return (error); 1959 mp->mnt_flag |= MNT_EXPUBLIC; 1960 } 1961 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1962 return (error); 1963 mp->mnt_flag |= MNT_EXPORTED; 1964 } 1965 return (0); 1966} 1967 1968 1969/* 1970 * Set the publicly exported filesystem (WebNFS). Currently, only 1971 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1972 */ 1973int 1974vfs_setpublicfs(mp, nep, argp) 1975 struct mount *mp; 1976 struct netexport *nep; 1977 struct export_args *argp; 1978{ 1979 int error; 1980 struct vnode *rvp; 1981 char *cp; 1982 1983 /* 1984 * mp == NULL -> invalidate the current info, the FS is 1985 * no longer exported. May be called from either vfs_export 1986 * or unmount, so check if it hasn't already been done. 1987 */ 1988 if (mp == NULL) { 1989 if (nfs_pub.np_valid) { 1990 nfs_pub.np_valid = 0; 1991 if (nfs_pub.np_index != NULL) { 1992 FREE(nfs_pub.np_index, M_TEMP); 1993 nfs_pub.np_index = NULL; 1994 } 1995 } 1996 return (0); 1997 } 1998 1999 /* 2000 * Only one allowed at a time. 2001 */ 2002 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2003 return (EBUSY); 2004 2005 /* 2006 * Get real filehandle for root of exported FS. 2007 */ 2008 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2009 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2010 2011 if ((error = VFS_ROOT(mp, &rvp))) 2012 return (error); 2013 2014 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2015 return (error); 2016 2017 vput(rvp); 2018 2019 /* 2020 * If an indexfile was specified, pull it in. 2021 */ 2022 if (argp->ex_indexfile != NULL) { 2023 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2024 M_WAITOK); 2025 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2026 MAXNAMLEN, (size_t *)0); 2027 if (!error) { 2028 /* 2029 * Check for illegal filenames. 2030 */ 2031 for (cp = nfs_pub.np_index; *cp; cp++) { 2032 if (*cp == '/') { 2033 error = EINVAL; 2034 break; 2035 } 2036 } 2037 } 2038 if (error) { 2039 FREE(nfs_pub.np_index, M_TEMP); 2040 return (error); 2041 } 2042 } 2043 2044 nfs_pub.np_mount = mp; 2045 nfs_pub.np_valid = 1; 2046 return (0); 2047} 2048 2049struct netcred * 2050vfs_export_lookup(mp, nep, nam) 2051 register struct mount *mp; 2052 struct netexport *nep; 2053 struct sockaddr *nam; 2054{ 2055 register struct netcred *np; 2056 register struct radix_node_head *rnh; 2057 struct sockaddr *saddr; 2058 2059 np = NULL; 2060 if (mp->mnt_flag & MNT_EXPORTED) { 2061 /* 2062 * Lookup in the export list first. 2063 */ 2064 if (nam != NULL) { 2065 saddr = nam; 2066 rnh = nep->ne_rtable[saddr->sa_family]; 2067 if (rnh != NULL) { 2068 np = (struct netcred *) 2069 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2070 rnh); 2071 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2072 np = NULL; 2073 } 2074 } 2075 /* 2076 * If no address match, use the default if it exists. 2077 */ 2078 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2079 np = &nep->ne_defexported; 2080 } 2081 return (np); 2082} 2083 2084/* 2085 * perform msync on all vnodes under a mount point 2086 * the mount point must be locked. 2087 */ 2088void 2089vfs_msync(struct mount *mp, int flags) { 2090 struct vnode *vp, *nvp; 2091 int anyio, tries; 2092 2093 tries = 5; 2094loop: 2095 anyio = 0; 2096 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 2097 2098 nvp = vp->v_mntvnodes.le_next; 2099 2100 if (vp->v_mount != mp) { 2101 goto loop; 2102 } 2103 2104 if ((vp->v_flag & VXLOCK) || 2105 (VOP_ISLOCKED(vp) && (flags != MNT_WAIT))) { 2106 continue; 2107 } 2108 2109 simple_lock(&vp->v_interlock); 2110 if (vp->v_object && 2111 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2112 if (!vget(vp, 2113 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) { 2114 if (vp->v_object) { 2115 vm_object_page_clean(vp->v_object, 0, 0, TRUE); 2116 anyio = 1; 2117 } 2118 vput(vp); 2119 } 2120 } else { 2121 simple_unlock(&vp->v_interlock); 2122 } 2123 } 2124 if (anyio && (--tries > 0)) 2125 goto loop; 2126} 2127 2128/* 2129 * Create the VM object needed for VMIO and mmap support. This 2130 * is done for all VREG files in the system. Some filesystems might 2131 * afford the additional metadata buffering capability of the 2132 * VMIO code by making the device node be VMIO mode also. 2133 * 2134 * If !waslocked, must be called with interlock. 2135 */ 2136int 2137vfs_object_create(vp, p, cred, waslocked) 2138 struct vnode *vp; 2139 struct proc *p; 2140 struct ucred *cred; 2141 int waslocked; 2142{ 2143 struct vattr vat; 2144 vm_object_t object; 2145 int error = 0; 2146 2147 if ((vp->v_type != VREG) && (vp->v_type != VBLK)) { 2148 if (!waslocked) 2149 simple_unlock(&vp->v_interlock); 2150 return 0; 2151 } 2152 2153 if (!waslocked) 2154 vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY, p); 2155 2156retry: 2157 if ((object = vp->v_object) == NULL) { 2158 if (vp->v_type == VREG) { 2159 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2160 goto retn; 2161 object = vnode_pager_alloc(vp, 2162 OFF_TO_IDX(round_page(vat.va_size)), 0, 0); 2163 } else if (major(vp->v_rdev) < nblkdev) { 2164 /* 2165 * This simply allocates the biggest object possible 2166 * for a VBLK vnode. This should be fixed, but doesn't 2167 * cause any problems (yet). 2168 */ 2169 object = vnode_pager_alloc(vp, INT_MAX, 0, 0); 2170 } 2171 object->ref_count--; 2172 vp->v_usecount--; 2173 } else { 2174 if (object->flags & OBJ_DEAD) { 2175 VOP_UNLOCK(vp, 0, p); 2176 tsleep(object, PVM, "vodead", 0); 2177 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2178 goto retry; 2179 } 2180 } 2181 2182 if (vp->v_object) { 2183 vp->v_flag |= VOBJBUF; 2184 } 2185 2186retn: 2187 if (!waslocked) { 2188 simple_lock(&vp->v_interlock); 2189 VOP_UNLOCK(vp, LK_INTERLOCK, p); 2190 } 2191 2192 return error; 2193} 2194 2195static void 2196vfree(vp) 2197 struct vnode *vp; 2198{ 2199 int s; 2200 2201 s = splbio(); 2202 simple_lock(&vnode_free_list_slock); 2203 if (vp->v_flag & VTBFREE) { 2204 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 2205 vp->v_flag &= ~VTBFREE; 2206 } 2207 if (vp->v_flag & VAGE) { 2208 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2209 } else { 2210 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2211 } 2212 freevnodes++; 2213 simple_unlock(&vnode_free_list_slock); 2214 vp->v_flag &= ~VAGE; 2215 vp->v_flag |= VFREE; 2216 splx(s); 2217} 2218 2219void 2220vbusy(vp) 2221 struct vnode *vp; 2222{ 2223 int s; 2224 2225 s = splbio(); 2226 simple_lock(&vnode_free_list_slock); 2227 if (vp->v_flag & VTBFREE) { 2228 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 2229 vp->v_flag &= ~VTBFREE; 2230 } else { 2231 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2232 freevnodes--; 2233 } 2234 simple_unlock(&vnode_free_list_slock); 2235 vp->v_flag &= ~VFREE; 2236 splx(s); 2237} 2238 2239/* 2240 * Record a process's interest in events which might happen to 2241 * a vnode. Because poll uses the historic select-style interface 2242 * internally, this routine serves as both the ``check for any 2243 * pending events'' and the ``record my interest in future events'' 2244 * functions. (These are done together, while the lock is held, 2245 * to avoid race conditions.) 2246 */ 2247int 2248vn_pollrecord(vp, p, events) 2249 struct vnode *vp; 2250 struct proc *p; 2251 short events; 2252{ 2253 simple_lock(&vp->v_pollinfo.vpi_lock); 2254 if (vp->v_pollinfo.vpi_revents & events) { 2255 /* 2256 * This leaves events we are not interested 2257 * in available for the other process which 2258 * which presumably had requested them 2259 * (otherwise they would never have been 2260 * recorded). 2261 */ 2262 events &= vp->v_pollinfo.vpi_revents; 2263 vp->v_pollinfo.vpi_revents &= ~events; 2264 2265 simple_unlock(&vp->v_pollinfo.vpi_lock); 2266 return events; 2267 } 2268 vp->v_pollinfo.vpi_events |= events; 2269 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2270 simple_unlock(&vp->v_pollinfo.vpi_lock); 2271 return 0; 2272} 2273 2274/* 2275 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2276 * it is possible for us to miss an event due to race conditions, but 2277 * that condition is expected to be rare, so for the moment it is the 2278 * preferred interface. 2279 */ 2280void 2281vn_pollevent(vp, events) 2282 struct vnode *vp; 2283 short events; 2284{ 2285 simple_lock(&vp->v_pollinfo.vpi_lock); 2286 if (vp->v_pollinfo.vpi_events & events) { 2287 /* 2288 * We clear vpi_events so that we don't 2289 * call selwakeup() twice if two events are 2290 * posted before the polling process(es) is 2291 * awakened. This also ensures that we take at 2292 * most one selwakeup() if the polling process 2293 * is no longer interested. However, it does 2294 * mean that only one event can be noticed at 2295 * a time. (Perhaps we should only clear those 2296 * event bits which we note?) XXX 2297 */ 2298 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2299 vp->v_pollinfo.vpi_revents |= events; 2300 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2301 } 2302 simple_unlock(&vp->v_pollinfo.vpi_lock); 2303} 2304 2305/* 2306 * Wake up anyone polling on vp because it is being revoked. 2307 * This depends on dead_poll() returning POLLHUP for correct 2308 * behavior. 2309 */ 2310void 2311vn_pollgone(vp) 2312 struct vnode *vp; 2313{ 2314 simple_lock(&vp->v_pollinfo.vpi_lock); 2315 if (vp->v_pollinfo.vpi_events) { 2316 vp->v_pollinfo.vpi_events = 0; 2317 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2318 } 2319 simple_unlock(&vp->v_pollinfo.vpi_lock); 2320} 2321