vfs_subr.c revision 32724
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 1.125 1998/01/22 17:29:52 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/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 106TAILQ_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; 121struct 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 vfs_object_create(vp, curproc, curproc->p_ucred, 0); 923 simple_lock(&vp->v_interlock); 924 } 925 if (flags & LK_TYPE_MASK) { 926 if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) 927 vrele(vp); 928 return (error); 929 } 930 simple_unlock(&vp->v_interlock); 931 return (0); 932} 933 934void 935vref(struct vnode *vp) 936{ 937 simple_lock(&vp->v_interlock); 938 vp->v_usecount++; 939 simple_unlock(&vp->v_interlock); 940} 941 942/* 943 * Vnode put/release. 944 * If count drops to zero, call inactive routine and return to freelist. 945 */ 946void 947vrele(vp) 948 struct vnode *vp; 949{ 950 struct proc *p = curproc; /* XXX */ 951 952#ifdef DIAGNOSTIC 953 if (vp == NULL) 954 panic("vrele: null vp"); 955#endif 956 simple_lock(&vp->v_interlock); 957 958 if (vp->v_usecount > 1) { 959 960 vp->v_usecount--; 961 simple_unlock(&vp->v_interlock); 962 963 return; 964 } 965 966 if (vp->v_usecount == 1) { 967 968 vp->v_usecount--; 969 970 if (VSHOULDFREE(vp)) 971 vfree(vp); 972 /* 973 * If we are doing a vput, the node is already locked, and we must 974 * call VOP_INACTIVE with the node locked. So, in the case of 975 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 976 */ 977 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 978 VOP_INACTIVE(vp, p); 979 } 980 981 } else { 982#ifdef DIAGNOSTIC 983 vprint("vrele: negative ref count", vp); 984 simple_unlock(&vp->v_interlock); 985#endif 986 panic("vrele: negative ref cnt"); 987 } 988} 989 990void 991vput(vp) 992 struct vnode *vp; 993{ 994 struct proc *p = curproc; /* XXX */ 995 996#ifdef DIAGNOSTIC 997 if (vp == NULL) 998 panic("vput: null vp"); 999#endif 1000 1001 simple_lock(&vp->v_interlock); 1002 1003 if (vp->v_usecount > 1) { 1004 1005 vp->v_usecount--; 1006 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1007 return; 1008 1009 } 1010 1011 if (vp->v_usecount == 1) { 1012 1013 vp->v_usecount--; 1014 if (VSHOULDFREE(vp)) 1015 vfree(vp); 1016 /* 1017 * If we are doing a vput, the node is already locked, and we must 1018 * call VOP_INACTIVE with the node locked. So, in the case of 1019 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1020 */ 1021 simple_unlock(&vp->v_interlock); 1022 VOP_INACTIVE(vp, p); 1023 1024 } else { 1025#ifdef DIAGNOSTIC 1026 vprint("vput: negative ref count", vp); 1027#endif 1028 panic("vput: negative ref cnt"); 1029 } 1030} 1031 1032/* 1033 * Somebody doesn't want the vnode recycled. 1034 */ 1035void 1036vhold(vp) 1037 register struct vnode *vp; 1038{ 1039 1040 simple_lock(&vp->v_interlock); 1041 vp->v_holdcnt++; 1042 if (VSHOULDBUSY(vp)) 1043 vbusy(vp); 1044 simple_unlock(&vp->v_interlock); 1045} 1046 1047/* 1048 * One less who cares about this vnode. 1049 */ 1050void 1051vdrop(vp) 1052 register struct vnode *vp; 1053{ 1054 1055 simple_lock(&vp->v_interlock); 1056 if (vp->v_holdcnt <= 0) 1057 panic("holdrele: holdcnt"); 1058 vp->v_holdcnt--; 1059 if (VSHOULDFREE(vp)) 1060 vfree(vp); 1061 simple_unlock(&vp->v_interlock); 1062} 1063 1064/* 1065 * Remove any vnodes in the vnode table belonging to mount point mp. 1066 * 1067 * If MNT_NOFORCE is specified, there should not be any active ones, 1068 * return error if any are found (nb: this is a user error, not a 1069 * system error). If MNT_FORCE is specified, detach any active vnodes 1070 * that are found. 1071 */ 1072#ifdef DIAGNOSTIC 1073static int busyprt = 0; /* print out busy vnodes */ 1074SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1075#endif 1076 1077int 1078vflush(mp, skipvp, flags) 1079 struct mount *mp; 1080 struct vnode *skipvp; 1081 int flags; 1082{ 1083 struct proc *p = curproc; /* XXX */ 1084 struct vnode *vp, *nvp; 1085 int busy = 0; 1086 1087 simple_lock(&mntvnode_slock); 1088loop: 1089 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 1090 /* 1091 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1092 * Start over if it has (it won't be on the list anymore). 1093 */ 1094 if (vp->v_mount != mp) 1095 goto loop; 1096 nvp = vp->v_mntvnodes.le_next; 1097 /* 1098 * Skip over a selected vnode. 1099 */ 1100 if (vp == skipvp) 1101 continue; 1102 1103 simple_lock(&vp->v_interlock); 1104 /* 1105 * Skip over a vnodes marked VSYSTEM. 1106 */ 1107 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1108 simple_unlock(&vp->v_interlock); 1109 continue; 1110 } 1111 /* 1112 * If WRITECLOSE is set, only flush out regular file vnodes 1113 * open for writing. 1114 */ 1115 if ((flags & WRITECLOSE) && 1116 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1117 simple_unlock(&vp->v_interlock); 1118 continue; 1119 } 1120 1121 /* 1122 * With v_usecount == 0, all we need to do is clear out the 1123 * vnode data structures and we are done. 1124 */ 1125 if (vp->v_usecount == 0) { 1126 simple_unlock(&mntvnode_slock); 1127 vgonel(vp, p); 1128 simple_lock(&mntvnode_slock); 1129 continue; 1130 } 1131 1132 /* 1133 * If FORCECLOSE is set, forcibly close the vnode. For block 1134 * or character devices, revert to an anonymous device. For 1135 * all other files, just kill them. 1136 */ 1137 if (flags & FORCECLOSE) { 1138 simple_unlock(&mntvnode_slock); 1139 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1140 vgonel(vp, p); 1141 } else { 1142 vclean(vp, 0, p); 1143 vp->v_op = spec_vnodeop_p; 1144 insmntque(vp, (struct mount *) 0); 1145 } 1146 simple_lock(&mntvnode_slock); 1147 continue; 1148 } 1149#ifdef DIAGNOSTIC 1150 if (busyprt) 1151 vprint("vflush: busy vnode", vp); 1152#endif 1153 simple_unlock(&vp->v_interlock); 1154 busy++; 1155 } 1156 simple_unlock(&mntvnode_slock); 1157 if (busy) 1158 return (EBUSY); 1159 return (0); 1160} 1161 1162/* 1163 * Disassociate the underlying file system from a vnode. 1164 */ 1165static void 1166vclean(vp, flags, p) 1167 struct vnode *vp; 1168 int flags; 1169 struct proc *p; 1170{ 1171 int active; 1172 1173 /* 1174 * Check to see if the vnode is in use. If so we have to reference it 1175 * before we clean it out so that its count cannot fall to zero and 1176 * generate a race against ourselves to recycle it. 1177 */ 1178 if ((active = vp->v_usecount)) 1179 vp->v_usecount++; 1180 1181 /* 1182 * Prevent the vnode from being recycled or brought into use while we 1183 * clean it out. 1184 */ 1185 if (vp->v_flag & VXLOCK) 1186 panic("vclean: deadlock"); 1187 vp->v_flag |= VXLOCK; 1188 /* 1189 * Even if the count is zero, the VOP_INACTIVE routine may still 1190 * have the object locked while it cleans it out. The VOP_LOCK 1191 * ensures that the VOP_INACTIVE routine is done with its work. 1192 * For active vnodes, it ensures that no other activity can 1193 * occur while the underlying object is being cleaned out. 1194 */ 1195 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1196 1197 /* 1198 * Clean out any buffers associated with the vnode. 1199 */ 1200 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1201 if (vp->v_object) { 1202 if (vp->v_object->ref_count == 0) { 1203 /* 1204 * This is a normal way of shutting down the object/vnode 1205 * association. 1206 */ 1207 vm_object_terminate(vp->v_object); 1208 } else { 1209 /* 1210 * Woe to the process that tries to page now :-). 1211 */ 1212 vm_pager_deallocate(vp->v_object); 1213 } 1214 } 1215 1216 /* 1217 * If purging an active vnode, it must be closed and 1218 * deactivated before being reclaimed. Note that the 1219 * VOP_INACTIVE will unlock the vnode. 1220 */ 1221 if (active) { 1222 if (flags & DOCLOSE) 1223 VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); 1224 VOP_INACTIVE(vp, p); 1225 } else { 1226 /* 1227 * Any other processes trying to obtain this lock must first 1228 * wait for VXLOCK to clear, then call the new lock operation. 1229 */ 1230 VOP_UNLOCK(vp, 0, p); 1231 } 1232 /* 1233 * Reclaim the vnode. 1234 */ 1235 if (VOP_RECLAIM(vp, p)) 1236 panic("vclean: cannot reclaim"); 1237 if (active) 1238 vrele(vp); 1239 cache_purge(vp); 1240 if (vp->v_vnlock) { 1241#if 0 /* This is the only place we have LK_DRAINED in the entire kernel ??? */ 1242#ifdef DIAGNOSTIC 1243 if ((vp->v_vnlock->lk_flags & LK_DRAINED) == 0) 1244 vprint("vclean: lock not drained", vp); 1245#endif 1246#endif 1247 FREE(vp->v_vnlock, M_VNODE); 1248 vp->v_vnlock = NULL; 1249 } 1250 1251 /* 1252 * Done with purge, notify sleepers of the grim news. 1253 */ 1254 vp->v_op = dead_vnodeop_p; 1255 vn_pollgone(vp); 1256 vp->v_tag = VT_NON; 1257 vp->v_flag &= ~VXLOCK; 1258 if (vp->v_flag & VXWANT) { 1259 vp->v_flag &= ~VXWANT; 1260 wakeup((caddr_t) vp); 1261 } 1262} 1263 1264/* 1265 * Eliminate all activity associated with the requested vnode 1266 * and with all vnodes aliased to the requested vnode. 1267 */ 1268int 1269vop_revoke(ap) 1270 struct vop_revoke_args /* { 1271 struct vnode *a_vp; 1272 int a_flags; 1273 } */ *ap; 1274{ 1275 struct vnode *vp, *vq; 1276 struct proc *p = curproc; /* XXX */ 1277 1278#ifdef DIAGNOSTIC 1279 if ((ap->a_flags & REVOKEALL) == 0) 1280 panic("vop_revoke"); 1281#endif 1282 1283 vp = ap->a_vp; 1284 simple_lock(&vp->v_interlock); 1285 1286 if (vp->v_flag & VALIASED) { 1287 /* 1288 * If a vgone (or vclean) is already in progress, 1289 * wait until it is done and return. 1290 */ 1291 if (vp->v_flag & VXLOCK) { 1292 vp->v_flag |= VXWANT; 1293 simple_unlock(&vp->v_interlock); 1294 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1295 return (0); 1296 } 1297 /* 1298 * Ensure that vp will not be vgone'd while we 1299 * are eliminating its aliases. 1300 */ 1301 vp->v_flag |= VXLOCK; 1302 simple_unlock(&vp->v_interlock); 1303 while (vp->v_flag & VALIASED) { 1304 simple_lock(&spechash_slock); 1305 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1306 if (vq->v_rdev != vp->v_rdev || 1307 vq->v_type != vp->v_type || vp == vq) 1308 continue; 1309 simple_unlock(&spechash_slock); 1310 vgone(vq); 1311 break; 1312 } 1313 if (vq == NULLVP) { 1314 simple_unlock(&spechash_slock); 1315 } 1316 } 1317 /* 1318 * Remove the lock so that vgone below will 1319 * really eliminate the vnode after which time 1320 * vgone will awaken any sleepers. 1321 */ 1322 simple_lock(&vp->v_interlock); 1323 vp->v_flag &= ~VXLOCK; 1324 if (vp->v_flag & VXWANT) { 1325 vp->v_flag &= ~VXWANT; 1326 wakeup(vp); 1327 } 1328 } 1329 vgonel(vp, p); 1330 return (0); 1331} 1332 1333/* 1334 * Recycle an unused vnode to the front of the free list. 1335 * Release the passed interlock if the vnode will be recycled. 1336 */ 1337int 1338vrecycle(vp, inter_lkp, p) 1339 struct vnode *vp; 1340 struct simplelock *inter_lkp; 1341 struct proc *p; 1342{ 1343 1344 simple_lock(&vp->v_interlock); 1345 if (vp->v_usecount == 0) { 1346 if (inter_lkp) { 1347 simple_unlock(inter_lkp); 1348 } 1349 vgonel(vp, p); 1350 return (1); 1351 } 1352 simple_unlock(&vp->v_interlock); 1353 return (0); 1354} 1355 1356/* 1357 * Eliminate all activity associated with a vnode 1358 * in preparation for reuse. 1359 */ 1360void 1361vgone(vp) 1362 register struct vnode *vp; 1363{ 1364 struct proc *p = curproc; /* XXX */ 1365 1366 simple_lock(&vp->v_interlock); 1367 vgonel(vp, p); 1368} 1369 1370/* 1371 * vgone, with the vp interlock held. 1372 */ 1373static void 1374vgonel(vp, p) 1375 struct vnode *vp; 1376 struct proc *p; 1377{ 1378 int s; 1379 struct vnode *vq; 1380 struct vnode *vx; 1381 1382 /* 1383 * If a vgone (or vclean) is already in progress, 1384 * wait until it is done and return. 1385 */ 1386 if (vp->v_flag & VXLOCK) { 1387 vp->v_flag |= VXWANT; 1388 simple_unlock(&vp->v_interlock); 1389 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1390 return; 1391 } 1392 1393 /* 1394 * Clean out the filesystem specific data. 1395 */ 1396 vclean(vp, DOCLOSE, p); 1397 simple_lock(&vp->v_interlock); 1398 1399 /* 1400 * Delete from old mount point vnode list, if on one. 1401 */ 1402 if (vp->v_mount != NULL) 1403 insmntque(vp, (struct mount *)0); 1404 /* 1405 * If special device, remove it from special device alias list 1406 * if it is on one. 1407 */ 1408 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1409 simple_lock(&spechash_slock); 1410 if (*vp->v_hashchain == vp) { 1411 *vp->v_hashchain = vp->v_specnext; 1412 } else { 1413 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1414 if (vq->v_specnext != vp) 1415 continue; 1416 vq->v_specnext = vp->v_specnext; 1417 break; 1418 } 1419 if (vq == NULL) 1420 panic("missing bdev"); 1421 } 1422 if (vp->v_flag & VALIASED) { 1423 vx = NULL; 1424 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1425 if (vq->v_rdev != vp->v_rdev || 1426 vq->v_type != vp->v_type) 1427 continue; 1428 if (vx) 1429 break; 1430 vx = vq; 1431 } 1432 if (vx == NULL) 1433 panic("missing alias"); 1434 if (vq == NULL) 1435 vx->v_flag &= ~VALIASED; 1436 vp->v_flag &= ~VALIASED; 1437 } 1438 simple_unlock(&spechash_slock); 1439 FREE(vp->v_specinfo, M_VNODE); 1440 vp->v_specinfo = NULL; 1441 } 1442 1443 /* 1444 * If it is on the freelist and not already at the head, 1445 * move it to the head of the list. The test of the back 1446 * pointer and the reference count of zero is because 1447 * it will be removed from the free list by getnewvnode, 1448 * but will not have its reference count incremented until 1449 * after calling vgone. If the reference count were 1450 * incremented first, vgone would (incorrectly) try to 1451 * close the previous instance of the underlying object. 1452 */ 1453 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1454 s = splbio(); 1455 simple_lock(&vnode_free_list_slock); 1456 if (vp->v_flag & VFREE) { 1457 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1458 } else if (vp->v_flag & VTBFREE) { 1459 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 1460 vp->v_flag &= ~VTBFREE; 1461 } 1462 vp->v_flag |= VFREE; 1463 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1464 simple_unlock(&vnode_free_list_slock); 1465 splx(s); 1466 } 1467 1468 vp->v_type = VBAD; 1469 simple_unlock(&vp->v_interlock); 1470} 1471 1472/* 1473 * Lookup a vnode by device number. 1474 */ 1475int 1476vfinddev(dev, type, vpp) 1477 dev_t dev; 1478 enum vtype type; 1479 struct vnode **vpp; 1480{ 1481 register struct vnode *vp; 1482 int rc = 0; 1483 1484 simple_lock(&spechash_slock); 1485 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1486 if (dev != vp->v_rdev || type != vp->v_type) 1487 continue; 1488 *vpp = vp; 1489 rc = 1; 1490 break; 1491 } 1492 simple_unlock(&spechash_slock); 1493 return (rc); 1494} 1495 1496/* 1497 * Calculate the total number of references to a special device. 1498 */ 1499int 1500vcount(vp) 1501 register struct vnode *vp; 1502{ 1503 struct vnode *vq, *vnext; 1504 int count; 1505 1506loop: 1507 if ((vp->v_flag & VALIASED) == 0) 1508 return (vp->v_usecount); 1509 simple_lock(&spechash_slock); 1510 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1511 vnext = vq->v_specnext; 1512 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1513 continue; 1514 /* 1515 * Alias, but not in use, so flush it out. 1516 */ 1517 if (vq->v_usecount == 0 && vq != vp) { 1518 simple_unlock(&spechash_slock); 1519 vgone(vq); 1520 goto loop; 1521 } 1522 count += vq->v_usecount; 1523 } 1524 simple_unlock(&spechash_slock); 1525 return (count); 1526} 1527/* 1528 * Print out a description of a vnode. 1529 */ 1530static char *typename[] = 1531{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1532 1533void 1534vprint(label, vp) 1535 char *label; 1536 register struct vnode *vp; 1537{ 1538 char buf[64]; 1539 1540 if (label != NULL) 1541 printf("%s: %x: ", label, vp); 1542 else 1543 printf("%x: ", vp); 1544 printf("type %s, usecount %d, writecount %d, refcount %ld,", 1545 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1546 vp->v_holdcnt); 1547 buf[0] = '\0'; 1548 if (vp->v_flag & VROOT) 1549 strcat(buf, "|VROOT"); 1550 if (vp->v_flag & VTEXT) 1551 strcat(buf, "|VTEXT"); 1552 if (vp->v_flag & VSYSTEM) 1553 strcat(buf, "|VSYSTEM"); 1554 if (vp->v_flag & VXLOCK) 1555 strcat(buf, "|VXLOCK"); 1556 if (vp->v_flag & VXWANT) 1557 strcat(buf, "|VXWANT"); 1558 if (vp->v_flag & VBWAIT) 1559 strcat(buf, "|VBWAIT"); 1560 if (vp->v_flag & VALIASED) 1561 strcat(buf, "|VALIASED"); 1562 if (vp->v_flag & VDOOMED) 1563 strcat(buf, "|VDOOMED"); 1564 if (vp->v_flag & VFREE) 1565 strcat(buf, "|VFREE"); 1566 if (vp->v_flag & VOBJBUF) 1567 strcat(buf, "|VOBJBUF"); 1568 if (buf[0] != '\0') 1569 printf(" flags (%s)", &buf[1]); 1570 if (vp->v_data == NULL) { 1571 printf("\n"); 1572 } else { 1573 printf("\n\t"); 1574 VOP_PRINT(vp); 1575 } 1576} 1577 1578#ifdef DDB 1579/* 1580 * List all of the locked vnodes in the system. 1581 * Called when debugging the kernel. 1582 */ 1583static void 1584printlockedvnodes() 1585{ 1586 struct proc *p = curproc; /* XXX */ 1587 struct mount *mp, *nmp; 1588 struct vnode *vp; 1589 1590 printf("Locked vnodes\n"); 1591 simple_lock(&mountlist_slock); 1592 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1593 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1594 nmp = mp->mnt_list.cqe_next; 1595 continue; 1596 } 1597 for (vp = mp->mnt_vnodelist.lh_first; 1598 vp != NULL; 1599 vp = vp->v_mntvnodes.le_next) { 1600 if (VOP_ISLOCKED(vp)) 1601 vprint((char *)0, vp); 1602 } 1603 simple_lock(&mountlist_slock); 1604 nmp = mp->mnt_list.cqe_next; 1605 vfs_unbusy(mp, p); 1606 } 1607 simple_unlock(&mountlist_slock); 1608} 1609#endif 1610 1611/* 1612 * Top level filesystem related information gathering. 1613 */ 1614static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS); 1615 1616static int 1617vfs_sysctl SYSCTL_HANDLER_ARGS 1618{ 1619 int *name = (int *)arg1 - 1; /* XXX */ 1620 u_int namelen = arg2 + 1; /* XXX */ 1621 struct vfsconf *vfsp; 1622 1623#ifndef NO_COMPAT_PRELITE2 1624 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1625 if (namelen == 1) 1626 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1627#endif 1628 1629#ifdef notyet 1630 /* all sysctl names at this level are at least name and field */ 1631 if (namelen < 2) 1632 return (ENOTDIR); /* overloaded */ 1633 if (name[0] != VFS_GENERIC) { 1634 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1635 if (vfsp->vfc_typenum == name[0]) 1636 break; 1637 if (vfsp == NULL) 1638 return (EOPNOTSUPP); 1639 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1640 oldp, oldlenp, newp, newlen, p)); 1641 } 1642#endif 1643 switch (name[1]) { 1644 case VFS_MAXTYPENUM: 1645 if (namelen != 2) 1646 return (ENOTDIR); 1647 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1648 case VFS_CONF: 1649 if (namelen != 3) 1650 return (ENOTDIR); /* overloaded */ 1651 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1652 if (vfsp->vfc_typenum == name[2]) 1653 break; 1654 if (vfsp == NULL) 1655 return (EOPNOTSUPP); 1656 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1657 } 1658 return (EOPNOTSUPP); 1659} 1660 1661SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1662 "Generic filesystem"); 1663 1664#ifndef NO_COMPAT_PRELITE2 1665 1666static int 1667sysctl_ovfs_conf SYSCTL_HANDLER_ARGS 1668{ 1669 int error; 1670 struct vfsconf *vfsp; 1671 struct ovfsconf ovfs; 1672 1673 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1674 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1675 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1676 ovfs.vfc_index = vfsp->vfc_typenum; 1677 ovfs.vfc_refcount = vfsp->vfc_refcount; 1678 ovfs.vfc_flags = vfsp->vfc_flags; 1679 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1680 if (error) 1681 return error; 1682 } 1683 return 0; 1684} 1685 1686#endif /* !NO_COMPAT_PRELITE2 */ 1687 1688static volatile int kinfo_vdebug = 1; 1689 1690#if 0 1691#define KINFO_VNODESLOP 10 1692/* 1693 * Dump vnode list (via sysctl). 1694 * Copyout address of vnode followed by vnode. 1695 */ 1696/* ARGSUSED */ 1697static int 1698sysctl_vnode SYSCTL_HANDLER_ARGS 1699{ 1700 struct proc *p = curproc; /* XXX */ 1701 struct mount *mp, *nmp; 1702 struct vnode *nvp, *vp; 1703 int error; 1704 1705#define VPTRSZ sizeof (struct vnode *) 1706#define VNODESZ sizeof (struct vnode) 1707 1708 req->lock = 0; 1709 if (!req->oldptr) /* Make an estimate */ 1710 return (SYSCTL_OUT(req, 0, 1711 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 1712 1713 simple_lock(&mountlist_slock); 1714 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1715 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 1716 nmp = mp->mnt_list.cqe_next; 1717 continue; 1718 } 1719again: 1720 simple_lock(&mntvnode_slock); 1721 for (vp = mp->mnt_vnodelist.lh_first; 1722 vp != NULL; 1723 vp = nvp) { 1724 /* 1725 * Check that the vp is still associated with 1726 * this filesystem. RACE: could have been 1727 * recycled onto the same filesystem. 1728 */ 1729 if (vp->v_mount != mp) { 1730 simple_unlock(&mntvnode_slock); 1731 if (kinfo_vdebug) 1732 printf("kinfo: vp changed\n"); 1733 goto again; 1734 } 1735 nvp = vp->v_mntvnodes.le_next; 1736 simple_unlock(&mntvnode_slock); 1737 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 1738 (error = SYSCTL_OUT(req, vp, VNODESZ))) 1739 return (error); 1740 simple_lock(&mntvnode_slock); 1741 } 1742 simple_unlock(&mntvnode_slock); 1743 simple_lock(&mountlist_slock); 1744 nmp = mp->mnt_list.cqe_next; 1745 vfs_unbusy(mp, p); 1746 } 1747 simple_unlock(&mountlist_slock); 1748 1749 return (0); 1750} 1751#endif 1752 1753/* 1754 * XXX 1755 * Exporting the vnode list on large systems causes them to crash. 1756 * Exporting the vnode list on medium systems causes sysctl to coredump. 1757 */ 1758#if 0 1759SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 1760 0, 0, sysctl_vnode, "S,vnode", ""); 1761#endif 1762 1763/* 1764 * Check to see if a filesystem is mounted on a block device. 1765 */ 1766int 1767vfs_mountedon(vp) 1768 struct vnode *vp; 1769{ 1770 struct vnode *vq; 1771 int error = 0; 1772 1773 if (vp->v_specflags & SI_MOUNTEDON) 1774 return (EBUSY); 1775 if (vp->v_flag & VALIASED) { 1776 simple_lock(&spechash_slock); 1777 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1778 if (vq->v_rdev != vp->v_rdev || 1779 vq->v_type != vp->v_type) 1780 continue; 1781 if (vq->v_specflags & SI_MOUNTEDON) { 1782 error = EBUSY; 1783 break; 1784 } 1785 } 1786 simple_unlock(&spechash_slock); 1787 } 1788 return (error); 1789} 1790 1791/* 1792 * Unmount all filesystems. The list is traversed in reverse order 1793 * of mounting to avoid dependencies. 1794 */ 1795void 1796vfs_unmountall() 1797{ 1798 struct mount *mp, *nmp; 1799 struct proc *p = initproc; /* XXX XXX should this be proc0? */ 1800 int error; 1801 1802 /* 1803 * Since this only runs when rebooting, it is not interlocked. 1804 */ 1805 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 1806 nmp = mp->mnt_list.cqe_prev; 1807 error = dounmount(mp, MNT_FORCE, p); 1808 if (error) { 1809 printf("unmount of %s failed (", 1810 mp->mnt_stat.f_mntonname); 1811 if (error == EBUSY) 1812 printf("BUSY)\n"); 1813 else 1814 printf("%d)\n", error); 1815 } 1816 } 1817} 1818 1819/* 1820 * Build hash lists of net addresses and hang them off the mount point. 1821 * Called by ufs_mount() to set up the lists of export addresses. 1822 */ 1823static int 1824vfs_hang_addrlist(mp, nep, argp) 1825 struct mount *mp; 1826 struct netexport *nep; 1827 struct export_args *argp; 1828{ 1829 register struct netcred *np; 1830 register struct radix_node_head *rnh; 1831 register int i; 1832 struct radix_node *rn; 1833 struct sockaddr *saddr, *smask = 0; 1834 struct domain *dom; 1835 int error; 1836 1837 if (argp->ex_addrlen == 0) { 1838 if (mp->mnt_flag & MNT_DEFEXPORTED) 1839 return (EPERM); 1840 np = &nep->ne_defexported; 1841 np->netc_exflags = argp->ex_flags; 1842 np->netc_anon = argp->ex_anon; 1843 np->netc_anon.cr_ref = 1; 1844 mp->mnt_flag |= MNT_DEFEXPORTED; 1845 return (0); 1846 } 1847 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1848 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1849 bzero((caddr_t) np, i); 1850 saddr = (struct sockaddr *) (np + 1); 1851 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1852 goto out; 1853 if (saddr->sa_len > argp->ex_addrlen) 1854 saddr->sa_len = argp->ex_addrlen; 1855 if (argp->ex_masklen) { 1856 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 1857 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 1858 if (error) 1859 goto out; 1860 if (smask->sa_len > argp->ex_masklen) 1861 smask->sa_len = argp->ex_masklen; 1862 } 1863 i = saddr->sa_family; 1864 if ((rnh = nep->ne_rtable[i]) == 0) { 1865 /* 1866 * Seems silly to initialize every AF when most are not used, 1867 * do so on demand here 1868 */ 1869 for (dom = domains; dom; dom = dom->dom_next) 1870 if (dom->dom_family == i && dom->dom_rtattach) { 1871 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1872 dom->dom_rtoffset); 1873 break; 1874 } 1875 if ((rnh = nep->ne_rtable[i]) == 0) { 1876 error = ENOBUFS; 1877 goto out; 1878 } 1879 } 1880 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 1881 np->netc_rnodes); 1882 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1883 error = EPERM; 1884 goto out; 1885 } 1886 np->netc_exflags = argp->ex_flags; 1887 np->netc_anon = argp->ex_anon; 1888 np->netc_anon.cr_ref = 1; 1889 return (0); 1890out: 1891 free(np, M_NETADDR); 1892 return (error); 1893} 1894 1895/* ARGSUSED */ 1896static int 1897vfs_free_netcred(rn, w) 1898 struct radix_node *rn; 1899 void *w; 1900{ 1901 register struct radix_node_head *rnh = (struct radix_node_head *) w; 1902 1903 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1904 free((caddr_t) rn, M_NETADDR); 1905 return (0); 1906} 1907 1908/* 1909 * Free the net address hash lists that are hanging off the mount points. 1910 */ 1911static void 1912vfs_free_addrlist(nep) 1913 struct netexport *nep; 1914{ 1915 register int i; 1916 register struct radix_node_head *rnh; 1917 1918 for (i = 0; i <= AF_MAX; i++) 1919 if ((rnh = nep->ne_rtable[i])) { 1920 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1921 (caddr_t) rnh); 1922 free((caddr_t) rnh, M_RTABLE); 1923 nep->ne_rtable[i] = 0; 1924 } 1925} 1926 1927int 1928vfs_export(mp, nep, argp) 1929 struct mount *mp; 1930 struct netexport *nep; 1931 struct export_args *argp; 1932{ 1933 int error; 1934 1935 if (argp->ex_flags & MNT_DELEXPORT) { 1936 if (mp->mnt_flag & MNT_EXPUBLIC) { 1937 vfs_setpublicfs(NULL, NULL, NULL); 1938 mp->mnt_flag &= ~MNT_EXPUBLIC; 1939 } 1940 vfs_free_addrlist(nep); 1941 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1942 } 1943 if (argp->ex_flags & MNT_EXPORTED) { 1944 if (argp->ex_flags & MNT_EXPUBLIC) { 1945 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1946 return (error); 1947 mp->mnt_flag |= MNT_EXPUBLIC; 1948 } 1949 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1950 return (error); 1951 mp->mnt_flag |= MNT_EXPORTED; 1952 } 1953 return (0); 1954} 1955 1956 1957/* 1958 * Set the publicly exported filesystem (WebNFS). Currently, only 1959 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1960 */ 1961int 1962vfs_setpublicfs(mp, nep, argp) 1963 struct mount *mp; 1964 struct netexport *nep; 1965 struct export_args *argp; 1966{ 1967 int error; 1968 struct vnode *rvp; 1969 char *cp; 1970 1971 /* 1972 * mp == NULL -> invalidate the current info, the FS is 1973 * no longer exported. May be called from either vfs_export 1974 * or unmount, so check if it hasn't already been done. 1975 */ 1976 if (mp == NULL) { 1977 if (nfs_pub.np_valid) { 1978 nfs_pub.np_valid = 0; 1979 if (nfs_pub.np_index != NULL) { 1980 FREE(nfs_pub.np_index, M_TEMP); 1981 nfs_pub.np_index = NULL; 1982 } 1983 } 1984 return (0); 1985 } 1986 1987 /* 1988 * Only one allowed at a time. 1989 */ 1990 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1991 return (EBUSY); 1992 1993 /* 1994 * Get real filehandle for root of exported FS. 1995 */ 1996 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1997 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1998 1999 if ((error = VFS_ROOT(mp, &rvp))) 2000 return (error); 2001 2002 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2003 return (error); 2004 2005 vput(rvp); 2006 2007 /* 2008 * If an indexfile was specified, pull it in. 2009 */ 2010 if (argp->ex_indexfile != NULL) { 2011 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2012 M_WAITOK); 2013 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2014 MAXNAMLEN, (size_t *)0); 2015 if (!error) { 2016 /* 2017 * Check for illegal filenames. 2018 */ 2019 for (cp = nfs_pub.np_index; *cp; cp++) { 2020 if (*cp == '/') { 2021 error = EINVAL; 2022 break; 2023 } 2024 } 2025 } 2026 if (error) { 2027 FREE(nfs_pub.np_index, M_TEMP); 2028 return (error); 2029 } 2030 } 2031 2032 nfs_pub.np_mount = mp; 2033 nfs_pub.np_valid = 1; 2034 return (0); 2035} 2036 2037struct netcred * 2038vfs_export_lookup(mp, nep, nam) 2039 register struct mount *mp; 2040 struct netexport *nep; 2041 struct sockaddr *nam; 2042{ 2043 register struct netcred *np; 2044 register struct radix_node_head *rnh; 2045 struct sockaddr *saddr; 2046 2047 np = NULL; 2048 if (mp->mnt_flag & MNT_EXPORTED) { 2049 /* 2050 * Lookup in the export list first. 2051 */ 2052 if (nam != NULL) { 2053 saddr = nam; 2054 rnh = nep->ne_rtable[saddr->sa_family]; 2055 if (rnh != NULL) { 2056 np = (struct netcred *) 2057 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2058 rnh); 2059 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2060 np = NULL; 2061 } 2062 } 2063 /* 2064 * If no address match, use the default if it exists. 2065 */ 2066 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2067 np = &nep->ne_defexported; 2068 } 2069 return (np); 2070} 2071 2072/* 2073 * perform msync on all vnodes under a mount point 2074 * the mount point must be locked. 2075 */ 2076void 2077vfs_msync(struct mount *mp, int flags) { 2078 struct vnode *vp, *nvp; 2079 int anyio, tries; 2080 2081 tries = 5; 2082loop: 2083 anyio = 0; 2084 for (vp = mp->mnt_vnodelist.lh_first; vp != NULL; vp = nvp) { 2085 2086 nvp = vp->v_mntvnodes.le_next; 2087 2088 if (vp->v_mount != mp) { 2089 goto loop; 2090 } 2091 2092 if ((vp->v_flag & VXLOCK) || 2093 (VOP_ISLOCKED(vp) && (flags != MNT_WAIT))) { 2094 continue; 2095 } 2096 2097 simple_lock(&vp->v_interlock); 2098 if (vp->v_object && 2099 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2100 if (!vget(vp, 2101 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) { 2102 if (vp->v_object) { 2103 vm_object_page_clean(vp->v_object, 0, 0, TRUE); 2104 anyio = 1; 2105 } 2106 vput(vp); 2107 } 2108 } else { 2109 simple_unlock(&vp->v_interlock); 2110 } 2111 } 2112 if (anyio && (--tries > 0)) 2113 goto loop; 2114} 2115 2116/* 2117 * Create the VM object needed for VMIO and mmap support. This 2118 * is done for all VREG files in the system. Some filesystems might 2119 * afford the additional metadata buffering capability of the 2120 * VMIO code by making the device node be VMIO mode also. 2121 * 2122 * If !waslocked, must be called with interlock. 2123 */ 2124int 2125vfs_object_create(vp, p, cred, waslocked) 2126 struct vnode *vp; 2127 struct proc *p; 2128 struct ucred *cred; 2129 int waslocked; 2130{ 2131 struct vattr vat; 2132 vm_object_t object; 2133 int error = 0; 2134 2135 if ((vp->v_type != VREG) && (vp->v_type != VBLK)) { 2136 return 0; 2137 } 2138 2139 if (!waslocked) 2140 vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_RETRY, p); 2141 2142retry: 2143 if ((object = vp->v_object) == NULL) { 2144 if (vp->v_type == VREG) { 2145 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2146 goto retn; 2147 object = vnode_pager_alloc(vp, 2148 OFF_TO_IDX(round_page(vat.va_size)), 0, 0); 2149 } else if (major(vp->v_rdev) < nblkdev) { 2150 /* 2151 * This simply allocates the biggest object possible 2152 * for a VBLK vnode. This should be fixed, but doesn't 2153 * cause any problems (yet). 2154 */ 2155 object = vnode_pager_alloc(vp, INT_MAX, 0, 0); 2156 } 2157 object->ref_count--; 2158 vp->v_usecount--; 2159 } else { 2160 if (object->flags & OBJ_DEAD) { 2161 VOP_UNLOCK(vp, 0, p); 2162 tsleep(object, PVM, "vodead", 0); 2163 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2164 goto retry; 2165 } 2166 } 2167 2168 if (vp->v_object) { 2169 vp->v_flag |= VOBJBUF; 2170 } 2171 2172retn: 2173 if (!waslocked) { 2174 simple_lock(&vp->v_interlock); 2175 VOP_UNLOCK(vp, LK_INTERLOCK, p); 2176 } 2177 2178 return error; 2179} 2180 2181static void 2182vfree(vp) 2183 struct vnode *vp; 2184{ 2185 int s; 2186 2187 s = splbio(); 2188 simple_lock(&vnode_free_list_slock); 2189 if (vp->v_flag & VTBFREE) { 2190 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 2191 vp->v_flag &= ~VTBFREE; 2192 } 2193 if (vp->v_flag & VAGE) { 2194 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2195 } else { 2196 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2197 } 2198 freevnodes++; 2199 simple_unlock(&vnode_free_list_slock); 2200 vp->v_flag &= ~VAGE; 2201 vp->v_flag |= VFREE; 2202 splx(s); 2203} 2204 2205void 2206vbusy(vp) 2207 struct vnode *vp; 2208{ 2209 int s; 2210 2211 s = splbio(); 2212 simple_lock(&vnode_free_list_slock); 2213 if (vp->v_flag & VTBFREE) { 2214 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 2215 vp->v_flag &= ~VTBFREE; 2216 } else { 2217 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2218 freevnodes--; 2219 } 2220 simple_unlock(&vnode_free_list_slock); 2221 vp->v_flag &= ~VFREE; 2222 splx(s); 2223} 2224 2225/* 2226 * Record a process's interest in events which might happen to 2227 * a vnode. Because poll uses the historic select-style interface 2228 * internally, this routine serves as both the ``check for any 2229 * pending events'' and the ``record my interest in future events'' 2230 * functions. (These are done together, while the lock is held, 2231 * to avoid race conditions.) 2232 */ 2233int 2234vn_pollrecord(vp, p, events) 2235 struct vnode *vp; 2236 struct proc *p; 2237 short events; 2238{ 2239 simple_lock(&vp->v_pollinfo.vpi_lock); 2240 if (vp->v_pollinfo.vpi_revents & events) { 2241 /* 2242 * This leaves events we are not interested 2243 * in available for the other process which 2244 * which presumably had requested them 2245 * (otherwise they would never have been 2246 * recorded). 2247 */ 2248 events &= vp->v_pollinfo.vpi_revents; 2249 vp->v_pollinfo.vpi_revents &= ~events; 2250 2251 simple_unlock(&vp->v_pollinfo.vpi_lock); 2252 return events; 2253 } 2254 vp->v_pollinfo.vpi_events |= events; 2255 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2256 simple_unlock(&vp->v_pollinfo.vpi_lock); 2257 return 0; 2258} 2259 2260/* 2261 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2262 * it is possible for us to miss an event due to race conditions, but 2263 * that condition is expected to be rare, so for the moment it is the 2264 * preferred interface. 2265 */ 2266void 2267vn_pollevent(vp, events) 2268 struct vnode *vp; 2269 short events; 2270{ 2271 simple_lock(&vp->v_pollinfo.vpi_lock); 2272 if (vp->v_pollinfo.vpi_events & events) { 2273 /* 2274 * We clear vpi_events so that we don't 2275 * call selwakeup() twice if two events are 2276 * posted before the polling process(es) is 2277 * awakened. This also ensures that we take at 2278 * most one selwakeup() if the polling process 2279 * is no longer interested. However, it does 2280 * mean that only one event can be noticed at 2281 * a time. (Perhaps we should only clear those 2282 * event bits which we note?) XXX 2283 */ 2284 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2285 vp->v_pollinfo.vpi_revents |= events; 2286 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2287 } 2288 simple_unlock(&vp->v_pollinfo.vpi_lock); 2289} 2290 2291/* 2292 * Wake up anyone polling on vp because it is being revoked. 2293 * This depends on dead_poll() returning POLLHUP for correct 2294 * behavior. 2295 */ 2296void 2297vn_pollgone(vp) 2298 struct vnode *vp; 2299{ 2300 simple_lock(&vp->v_pollinfo.vpi_lock); 2301 if (vp->v_pollinfo.vpi_events) { 2302 vp->v_pollinfo.vpi_events = 0; 2303 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2304 } 2305 simple_unlock(&vp->v_pollinfo.vpi_lock); 2306} 2307