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 * %sccs.include.redist.c% 11 * 12 * @(#)vfs_subr.c 8.9 (Berkeley) 01/21/94 13 */ 14 15 /* 16 * External virtual filesystem routines 17 */ 18 19 #include <sys/param.h> 20 #include <sys/systm.h> 21 #include <sys/proc.h> 22 #include <sys/mount.h> 23 #include <sys/time.h> 24 #include <sys/vnode.h> 25 #include <sys/stat.h> 26 #include <sys/namei.h> 27 #include <sys/ucred.h> 28 #include <sys/buf.h> 29 #include <sys/errno.h> 30 #include <sys/malloc.h> 31 #include <sys/domain.h> 32 #include <sys/mbuf.h> 33 34 #include <vm/vm.h> 35 #include <sys/sysctl.h> 36 37 #include <miscfs/specfs/specdev.h> 38 39 enum vtype iftovt_tab[16] = { 40 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 41 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 42 }; 43 int vttoif_tab[9] = { 44 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 45 S_IFSOCK, S_IFIFO, S_IFMT, 46 }; 47 48 /* 49 * Insq/Remq for the vnode usage lists. 50 */ 51 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 52 #define bufremvn(bp) { \ 53 LIST_REMOVE(bp, b_vnbufs); \ 54 (bp)->b_vnbufs.le_next = NOLIST; \ 55 } 56 57 TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 58 struct mntlist mountlist; /* mounted filesystem list */ 59 60 /* 61 * Initialize the vnode management data structures. 62 */ 63 vntblinit() 64 { 65 66 TAILQ_INIT(&vnode_free_list); 67 TAILQ_INIT(&mountlist); 68 } 69 70 /* 71 * Lock a filesystem. 72 * Used to prevent access to it while mounting and unmounting. 73 */ 74 vfs_lock(mp) 75 register struct mount *mp; 76 { 77 78 while(mp->mnt_flag & MNT_MLOCK) { 79 mp->mnt_flag |= MNT_MWAIT; 80 sleep((caddr_t)mp, PVFS); 81 } 82 mp->mnt_flag |= MNT_MLOCK; 83 return (0); 84 } 85 86 /* 87 * Unlock a locked filesystem. 88 * Panic if filesystem is not locked. 89 */ 90 void 91 vfs_unlock(mp) 92 register struct mount *mp; 93 { 94 95 if ((mp->mnt_flag & MNT_MLOCK) == 0) 96 panic("vfs_unlock: not locked"); 97 mp->mnt_flag &= ~MNT_MLOCK; 98 if (mp->mnt_flag & MNT_MWAIT) { 99 mp->mnt_flag &= ~MNT_MWAIT; 100 wakeup((caddr_t)mp); 101 } 102 } 103 104 /* 105 * Mark a mount point as busy. 106 * Used to synchronize access and to delay unmounting. 107 */ 108 vfs_busy(mp) 109 register struct mount *mp; 110 { 111 112 while(mp->mnt_flag & MNT_MPBUSY) { 113 mp->mnt_flag |= MNT_MPWANT; 114 sleep((caddr_t)&mp->mnt_flag, PVFS); 115 } 116 if (mp->mnt_flag & MNT_UNMOUNT) 117 return (1); 118 mp->mnt_flag |= MNT_MPBUSY; 119 return (0); 120 } 121 122 /* 123 * Free a busy filesystem. 124 * Panic if filesystem is not busy. 125 */ 126 vfs_unbusy(mp) 127 register struct mount *mp; 128 { 129 130 if ((mp->mnt_flag & MNT_MPBUSY) == 0) 131 panic("vfs_unbusy: not busy"); 132 mp->mnt_flag &= ~MNT_MPBUSY; 133 if (mp->mnt_flag & MNT_MPWANT) { 134 mp->mnt_flag &= ~MNT_MPWANT; 135 wakeup((caddr_t)&mp->mnt_flag); 136 } 137 } 138 139 /* 140 * Lookup a mount point by filesystem identifier. 141 */ 142 struct mount * 143 getvfs(fsid) 144 fsid_t *fsid; 145 { 146 register struct mount *mp; 147 148 for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) { 149 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 150 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) 151 return (mp); 152 } 153 return ((struct mount *)0); 154 } 155 156 /* 157 * Get a new unique fsid 158 */ 159 void 160 getnewfsid(mp, mtype) 161 struct mount *mp; 162 int mtype; 163 { 164 static u_short xxxfs_mntid; 165 166 fsid_t tfsid; 167 168 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 169 mp->mnt_stat.f_fsid.val[1] = mtype; 170 if (xxxfs_mntid == 0) 171 ++xxxfs_mntid; 172 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 173 tfsid.val[1] = mtype; 174 if (mountlist.tqh_first != NULL) { 175 while (getvfs(&tfsid)) { 176 tfsid.val[0]++; 177 xxxfs_mntid++; 178 } 179 } 180 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 181 } 182 183 /* 184 * Set vnode attributes to VNOVAL 185 */ 186 void vattr_null(vap) 187 register struct vattr *vap; 188 { 189 190 vap->va_type = VNON; 191 vap->va_size = vap->va_bytes = VNOVAL; 192 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 193 vap->va_fsid = vap->va_fileid = 194 vap->va_blocksize = vap->va_rdev = 195 vap->va_atime.ts_sec = vap->va_atime.ts_nsec = 196 vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec = 197 vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec = 198 vap->va_flags = vap->va_gen = VNOVAL; 199 vap->va_vaflags = 0; 200 } 201 202 /* 203 * Routines having to do with the management of the vnode table. 204 */ 205 extern int (**dead_vnodeop_p)(); 206 extern void vclean(); 207 long numvnodes; 208 extern struct vattr va_null; 209 210 /* 211 * Return the next vnode from the free list. 212 */ 213 getnewvnode(tag, mp, vops, vpp) 214 enum vtagtype tag; 215 struct mount *mp; 216 int (**vops)(); 217 struct vnode **vpp; 218 { 219 register struct vnode *vp; 220 int s; 221 222 if ((vnode_free_list.tqh_first == NULL && 223 numvnodes < 2 * desiredvnodes) || 224 numvnodes < desiredvnodes) { 225 vp = (struct vnode *)malloc((u_long)sizeof *vp, 226 M_VNODE, M_WAITOK); 227 bzero((char *)vp, sizeof *vp); 228 numvnodes++; 229 } else { 230 if ((vp = vnode_free_list.tqh_first) == NULL) { 231 tablefull("vnode"); 232 *vpp = 0; 233 return (ENFILE); 234 } 235 if (vp->v_usecount) 236 panic("free vnode isn't"); 237 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 238 /* see comment on why 0xdeadb is set at end of vgone (below) */ 239 vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; 240 vp->v_lease = NULL; 241 if (vp->v_type != VBAD) 242 vgone(vp); 243 #ifdef DIAGNOSTIC 244 if (vp->v_data) 245 panic("cleaned vnode isn't"); 246 s = splbio(); 247 if (vp->v_numoutput) 248 panic("Clean vnode has pending I/O's"); 249 splx(s); 250 #endif 251 vp->v_flag = 0; 252 vp->v_lastr = 0; 253 vp->v_ralen = 0; 254 vp->v_maxra = 0; 255 vp->v_lastw = 0; 256 vp->v_lasta = 0; 257 vp->v_cstart = 0; 258 vp->v_clen = 0; 259 vp->v_socket = 0; 260 } 261 vp->v_type = VNON; 262 cache_purge(vp); 263 vp->v_tag = tag; 264 vp->v_op = vops; 265 insmntque(vp, mp); 266 *vpp = vp; 267 vp->v_usecount = 1; 268 vp->v_data = 0; 269 return (0); 270 } 271 272 /* 273 * Move a vnode from one mount queue to another. 274 */ 275 insmntque(vp, mp) 276 register struct vnode *vp; 277 register struct mount *mp; 278 { 279 280 /* 281 * Delete from old mount point vnode list, if on one. 282 */ 283 if (vp->v_mount != NULL) 284 LIST_REMOVE(vp, v_mntvnodes); 285 /* 286 * Insert into list of vnodes for the new mount point, if available. 287 */ 288 if ((vp->v_mount = mp) == NULL) 289 return; 290 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 291 } 292 293 /* 294 * Update outstanding I/O count and do wakeup if requested. 295 */ 296 vwakeup(bp) 297 register struct buf *bp; 298 { 299 register struct vnode *vp; 300 301 bp->b_flags &= ~B_WRITEINPROG; 302 if (vp = bp->b_vp) { 303 vp->v_numoutput--; 304 if (vp->v_numoutput < 0) 305 panic("vwakeup: neg numoutput"); 306 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 307 if (vp->v_numoutput < 0) 308 panic("vwakeup: neg numoutput"); 309 vp->v_flag &= ~VBWAIT; 310 wakeup((caddr_t)&vp->v_numoutput); 311 } 312 } 313 } 314 315 /* 316 * Flush out and invalidate all buffers associated with a vnode. 317 * Called with the underlying object locked. 318 */ 319 int 320 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 321 register struct vnode *vp; 322 int flags; 323 struct ucred *cred; 324 struct proc *p; 325 int slpflag, slptimeo; 326 { 327 register struct buf *bp; 328 struct buf *nbp, *blist; 329 int s, error; 330 331 if (flags & V_SAVE) { 332 if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) 333 return (error); 334 if (vp->v_dirtyblkhd.lh_first != NULL) 335 panic("vinvalbuf: dirty bufs"); 336 } 337 for (;;) { 338 if ((blist = vp->v_cleanblkhd.lh_first) && flags & V_SAVEMETA) 339 while (blist && blist->b_lblkno < 0) 340 blist = blist->b_vnbufs.le_next; 341 if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 342 (flags & V_SAVEMETA)) 343 while (blist && blist->b_lblkno < 0) 344 blist = blist->b_vnbufs.le_next; 345 if (!blist) 346 break; 347 348 for (bp = blist; bp; bp = nbp) { 349 nbp = bp->b_vnbufs.le_next; 350 if (flags & V_SAVEMETA && bp->b_lblkno < 0) 351 continue; 352 s = splbio(); 353 if (bp->b_flags & B_BUSY) { 354 bp->b_flags |= B_WANTED; 355 error = tsleep((caddr_t)bp, 356 slpflag | (PRIBIO + 1), "vinvalbuf", 357 slptimeo); 358 splx(s); 359 if (error) 360 return (error); 361 break; 362 } 363 bremfree(bp); 364 bp->b_flags |= B_BUSY; 365 splx(s); 366 /* 367 * XXX Since there are no node locks for NFS, I believe 368 * there is a slight chance that a delayed write will 369 * occur while sleeping just above, so check for it. 370 */ 371 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 372 (void) VOP_BWRITE(bp); 373 break; 374 } 375 bp->b_flags |= B_INVAL; 376 brelse(bp); 377 } 378 } 379 if (!(flags & V_SAVEMETA) && 380 (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) 381 panic("vinvalbuf: flush failed"); 382 return (0); 383 } 384 385 /* 386 * Associate a buffer with a vnode. 387 */ 388 bgetvp(vp, bp) 389 register struct vnode *vp; 390 register struct buf *bp; 391 { 392 393 if (bp->b_vp) 394 panic("bgetvp: not free"); 395 VHOLD(vp); 396 bp->b_vp = vp; 397 if (vp->v_type == VBLK || vp->v_type == VCHR) 398 bp->b_dev = vp->v_rdev; 399 else 400 bp->b_dev = NODEV; 401 /* 402 * Insert onto list for new vnode. 403 */ 404 bufinsvn(bp, &vp->v_cleanblkhd); 405 } 406 407 /* 408 * Disassociate a buffer from a vnode. 409 */ 410 brelvp(bp) 411 register struct buf *bp; 412 { 413 struct vnode *vp; 414 415 if (bp->b_vp == (struct vnode *) 0) 416 panic("brelvp: NULL"); 417 /* 418 * Delete from old vnode list, if on one. 419 */ 420 if (bp->b_vnbufs.le_next != NOLIST) 421 bufremvn(bp); 422 vp = bp->b_vp; 423 bp->b_vp = (struct vnode *) 0; 424 HOLDRELE(vp); 425 } 426 427 /* 428 * Reassign a buffer from one vnode to another. 429 * Used to assign file specific control information 430 * (indirect blocks) to the vnode to which they belong. 431 */ 432 reassignbuf(bp, newvp) 433 register struct buf *bp; 434 register struct vnode *newvp; 435 { 436 register struct buflists *listheadp; 437 438 if (newvp == NULL) { 439 printf("reassignbuf: NULL"); 440 return; 441 } 442 /* 443 * Delete from old vnode list, if on one. 444 */ 445 if (bp->b_vnbufs.le_next != NOLIST) 446 bufremvn(bp); 447 /* 448 * If dirty, put on list of dirty buffers; 449 * otherwise insert onto list of clean buffers. 450 */ 451 if (bp->b_flags & B_DELWRI) 452 listheadp = &newvp->v_dirtyblkhd; 453 else 454 listheadp = &newvp->v_cleanblkhd; 455 bufinsvn(bp, listheadp); 456 } 457 458 /* 459 * Create a vnode for a block device. 460 * Used for root filesystem, argdev, and swap areas. 461 * Also used for memory file system special devices. 462 */ 463 bdevvp(dev, vpp) 464 dev_t dev; 465 struct vnode **vpp; 466 { 467 register struct vnode *vp; 468 struct vnode *nvp; 469 int error; 470 471 if (dev == NODEV) 472 return (0); 473 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 474 if (error) { 475 *vpp = 0; 476 return (error); 477 } 478 vp = nvp; 479 vp->v_type = VBLK; 480 if (nvp = checkalias(vp, dev, (struct mount *)0)) { 481 vput(vp); 482 vp = nvp; 483 } 484 *vpp = vp; 485 return (0); 486 } 487 488 /* 489 * Check to see if the new vnode represents a special device 490 * for which we already have a vnode (either because of 491 * bdevvp() or because of a different vnode representing 492 * the same block device). If such an alias exists, deallocate 493 * the existing contents and return the aliased vnode. The 494 * caller is responsible for filling it with its new contents. 495 */ 496 struct vnode * 497 checkalias(nvp, nvp_rdev, mp) 498 register struct vnode *nvp; 499 dev_t nvp_rdev; 500 struct mount *mp; 501 { 502 register struct vnode *vp; 503 struct vnode **vpp; 504 505 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 506 return (NULLVP); 507 508 vpp = &speclisth[SPECHASH(nvp_rdev)]; 509 loop: 510 for (vp = *vpp; vp; vp = vp->v_specnext) { 511 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 512 continue; 513 /* 514 * Alias, but not in use, so flush it out. 515 */ 516 if (vp->v_usecount == 0) { 517 vgone(vp); 518 goto loop; 519 } 520 if (vget(vp, 1)) 521 goto loop; 522 break; 523 } 524 if (vp == NULL || vp->v_tag != VT_NON) { 525 MALLOC(nvp->v_specinfo, struct specinfo *, 526 sizeof(struct specinfo), M_VNODE, M_WAITOK); 527 nvp->v_rdev = nvp_rdev; 528 nvp->v_hashchain = vpp; 529 nvp->v_specnext = *vpp; 530 nvp->v_specflags = 0; 531 *vpp = nvp; 532 if (vp != NULL) { 533 nvp->v_flag |= VALIASED; 534 vp->v_flag |= VALIASED; 535 vput(vp); 536 } 537 return (NULLVP); 538 } 539 VOP_UNLOCK(vp); 540 vclean(vp, 0); 541 vp->v_op = nvp->v_op; 542 vp->v_tag = nvp->v_tag; 543 nvp->v_type = VNON; 544 insmntque(vp, mp); 545 return (vp); 546 } 547 548 /* 549 * Grab a particular vnode from the free list, increment its 550 * reference count and lock it. The vnode lock bit is set the 551 * vnode is being eliminated in vgone. The process is awakened 552 * when the transition is completed, and an error returned to 553 * indicate that the vnode is no longer usable (possibly having 554 * been changed to a new file system type). 555 */ 556 vget(vp, lockflag) 557 register struct vnode *vp; 558 int lockflag; 559 { 560 561 if (vp->v_flag & VXLOCK) { 562 vp->v_flag |= VXWANT; 563 sleep((caddr_t)vp, PINOD); 564 return (1); 565 } 566 if (vp->v_usecount == 0) 567 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 568 vp->v_usecount++; 569 if (lockflag) 570 VOP_LOCK(vp); 571 return (0); 572 } 573 574 /* 575 * Vnode reference, just increment the count 576 */ 577 void vref(vp) 578 struct vnode *vp; 579 { 580 581 if (vp->v_usecount <= 0) 582 panic("vref used where vget required"); 583 vp->v_usecount++; 584 } 585 586 /* 587 * vput(), just unlock and vrele() 588 */ 589 void vput(vp) 590 register struct vnode *vp; 591 { 592 593 VOP_UNLOCK(vp); 594 vrele(vp); 595 } 596 597 /* 598 * Vnode release. 599 * If count drops to zero, call inactive routine and return to freelist. 600 */ 601 void vrele(vp) 602 register struct vnode *vp; 603 { 604 605 #ifdef DIAGNOSTIC 606 if (vp == NULL) 607 panic("vrele: null vp"); 608 #endif 609 vp->v_usecount--; 610 if (vp->v_usecount > 0) 611 return; 612 #ifdef DIAGNOSTIC 613 if (vp->v_usecount != 0 || vp->v_writecount != 0) { 614 vprint("vrele: bad ref count", vp); 615 panic("vrele: ref cnt"); 616 } 617 #endif 618 /* 619 * insert at tail of LRU list 620 */ 621 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 622 VOP_INACTIVE(vp); 623 } 624 625 /* 626 * Page or buffer structure gets a reference. 627 */ 628 void vhold(vp) 629 register struct vnode *vp; 630 { 631 632 vp->v_holdcnt++; 633 } 634 635 /* 636 * Page or buffer structure frees a reference. 637 */ 638 void holdrele(vp) 639 register struct vnode *vp; 640 { 641 642 if (vp->v_holdcnt <= 0) 643 panic("holdrele: holdcnt"); 644 vp->v_holdcnt--; 645 } 646 647 /* 648 * Remove any vnodes in the vnode table belonging to mount point mp. 649 * 650 * If MNT_NOFORCE is specified, there should not be any active ones, 651 * return error if any are found (nb: this is a user error, not a 652 * system error). If MNT_FORCE is specified, detach any active vnodes 653 * that are found. 654 */ 655 #ifdef DIAGNOSTIC 656 int busyprt = 0; /* print out busy vnodes */ 657 struct ctldebug debug1 = { "busyprt", &busyprt }; 658 #endif 659 660 vflush(mp, skipvp, flags) 661 struct mount *mp; 662 struct vnode *skipvp; 663 int flags; 664 { 665 register struct vnode *vp, *nvp; 666 int busy = 0; 667 668 if ((mp->mnt_flag & MNT_MPBUSY) == 0) 669 panic("vflush: not busy"); 670 loop: 671 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 672 if (vp->v_mount != mp) 673 goto loop; 674 nvp = vp->v_mntvnodes.le_next; 675 /* 676 * Skip over a selected vnode. 677 */ 678 if (vp == skipvp) 679 continue; 680 /* 681 * Skip over a vnodes marked VSYSTEM. 682 */ 683 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) 684 continue; 685 /* 686 * If WRITECLOSE is set, only flush out regular file 687 * vnodes open for writing. 688 */ 689 if ((flags & WRITECLOSE) && 690 (vp->v_writecount == 0 || vp->v_type != VREG)) 691 continue; 692 /* 693 * With v_usecount == 0, all we need to do is clear 694 * out the vnode data structures and we are done. 695 */ 696 if (vp->v_usecount == 0) { 697 vgone(vp); 698 continue; 699 } 700 /* 701 * If FORCECLOSE is set, forcibly close the vnode. 702 * For block or character devices, revert to an 703 * anonymous device. For all other files, just kill them. 704 */ 705 if (flags & FORCECLOSE) { 706 if (vp->v_type != VBLK && vp->v_type != VCHR) { 707 vgone(vp); 708 } else { 709 vclean(vp, 0); 710 vp->v_op = spec_vnodeop_p; 711 insmntque(vp, (struct mount *)0); 712 } 713 continue; 714 } 715 #ifdef DIAGNOSTIC 716 if (busyprt) 717 vprint("vflush: busy vnode", vp); 718 #endif 719 busy++; 720 } 721 if (busy) 722 return (EBUSY); 723 return (0); 724 } 725 726 /* 727 * Disassociate the underlying file system from a vnode. 728 */ 729 void 730 vclean(vp, flags) 731 register struct vnode *vp; 732 int flags; 733 { 734 int active; 735 736 /* 737 * Check to see if the vnode is in use. 738 * If so we have to reference it before we clean it out 739 * so that its count cannot fall to zero and generate a 740 * race against ourselves to recycle it. 741 */ 742 if (active = vp->v_usecount) 743 VREF(vp); 744 /* 745 * Even if the count is zero, the VOP_INACTIVE routine may still 746 * have the object locked while it cleans it out. The VOP_LOCK 747 * ensures that the VOP_INACTIVE routine is done with its work. 748 * For active vnodes, it ensures that no other activity can 749 * occur while the underlying object is being cleaned out. 750 */ 751 VOP_LOCK(vp); 752 /* 753 * Prevent the vnode from being recycled or 754 * brought into use while we clean it out. 755 */ 756 if (vp->v_flag & VXLOCK) 757 panic("vclean: deadlock"); 758 vp->v_flag |= VXLOCK; 759 /* 760 * Clean out any buffers associated with the vnode. 761 */ 762 if (flags & DOCLOSE) 763 vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0); 764 /* 765 * Any other processes trying to obtain this lock must first 766 * wait for VXLOCK to clear, then call the new lock operation. 767 */ 768 VOP_UNLOCK(vp); 769 /* 770 * If purging an active vnode, it must be closed and 771 * deactivated before being reclaimed. 772 */ 773 if (active) { 774 if (flags & DOCLOSE) 775 VOP_CLOSE(vp, IO_NDELAY, NOCRED, NULL); 776 VOP_INACTIVE(vp); 777 } 778 /* 779 * Reclaim the vnode. 780 */ 781 if (VOP_RECLAIM(vp)) 782 panic("vclean: cannot reclaim"); 783 if (active) 784 vrele(vp); 785 786 /* 787 * Done with purge, notify sleepers of the grim news. 788 */ 789 vp->v_op = dead_vnodeop_p; 790 vp->v_tag = VT_NON; 791 vp->v_flag &= ~VXLOCK; 792 if (vp->v_flag & VXWANT) { 793 vp->v_flag &= ~VXWANT; 794 wakeup((caddr_t)vp); 795 } 796 } 797 798 /* 799 * Eliminate all activity associated with the requested vnode 800 * and with all vnodes aliased to the requested vnode. 801 */ 802 void vgoneall(vp) 803 register struct vnode *vp; 804 { 805 register struct vnode *vq; 806 807 if (vp->v_flag & VALIASED) { 808 /* 809 * If a vgone (or vclean) is already in progress, 810 * wait until it is done and return. 811 */ 812 if (vp->v_flag & VXLOCK) { 813 vp->v_flag |= VXWANT; 814 sleep((caddr_t)vp, PINOD); 815 return; 816 } 817 /* 818 * Ensure that vp will not be vgone'd while we 819 * are eliminating its aliases. 820 */ 821 vp->v_flag |= VXLOCK; 822 while (vp->v_flag & VALIASED) { 823 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 824 if (vq->v_rdev != vp->v_rdev || 825 vq->v_type != vp->v_type || vp == vq) 826 continue; 827 vgone(vq); 828 break; 829 } 830 } 831 /* 832 * Remove the lock so that vgone below will 833 * really eliminate the vnode after which time 834 * vgone will awaken any sleepers. 835 */ 836 vp->v_flag &= ~VXLOCK; 837 } 838 vgone(vp); 839 } 840 841 /* 842 * Eliminate all activity associated with a vnode 843 * in preparation for reuse. 844 */ 845 void vgone(vp) 846 register struct vnode *vp; 847 { 848 register struct vnode *vq; 849 struct vnode *vx; 850 851 /* 852 * If a vgone (or vclean) is already in progress, 853 * wait until it is done and return. 854 */ 855 if (vp->v_flag & VXLOCK) { 856 vp->v_flag |= VXWANT; 857 sleep((caddr_t)vp, PINOD); 858 return; 859 } 860 /* 861 * Clean out the filesystem specific data. 862 */ 863 vclean(vp, DOCLOSE); 864 /* 865 * Delete from old mount point vnode list, if on one. 866 */ 867 if (vp->v_mount != NULL) { 868 LIST_REMOVE(vp, v_mntvnodes); 869 vp->v_mount = NULL; 870 } 871 /* 872 * If special device, remove it from special device alias list. 873 */ 874 if (vp->v_type == VBLK || vp->v_type == VCHR) { 875 if (*vp->v_hashchain == vp) { 876 *vp->v_hashchain = vp->v_specnext; 877 } else { 878 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 879 if (vq->v_specnext != vp) 880 continue; 881 vq->v_specnext = vp->v_specnext; 882 break; 883 } 884 if (vq == NULL) 885 panic("missing bdev"); 886 } 887 if (vp->v_flag & VALIASED) { 888 vx = NULL; 889 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 890 if (vq->v_rdev != vp->v_rdev || 891 vq->v_type != vp->v_type) 892 continue; 893 if (vx) 894 break; 895 vx = vq; 896 } 897 if (vx == NULL) 898 panic("missing alias"); 899 if (vq == NULL) 900 vx->v_flag &= ~VALIASED; 901 vp->v_flag &= ~VALIASED; 902 } 903 FREE(vp->v_specinfo, M_VNODE); 904 vp->v_specinfo = NULL; 905 } 906 /* 907 * If it is on the freelist and not already at the head, 908 * move it to the head of the list. The test of the back 909 * pointer and the reference count of zero is because 910 * it will be removed from the free list by getnewvnode, 911 * but will not have its reference count incremented until 912 * after calling vgone. If the reference count were 913 * incremented first, vgone would (incorrectly) try to 914 * close the previous instance of the underlying object. 915 * So, the back pointer is explicitly set to `0xdeadb' in 916 * getnewvnode after removing it from the freelist to ensure 917 * that we do not try to move it here. 918 */ 919 if (vp->v_usecount == 0 && 920 vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb && 921 vnode_free_list.tqh_first != vp) { 922 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 923 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 924 } 925 vp->v_type = VBAD; 926 } 927 928 /* 929 * Lookup a vnode by device number. 930 */ 931 vfinddev(dev, type, vpp) 932 dev_t dev; 933 enum vtype type; 934 struct vnode **vpp; 935 { 936 register struct vnode *vp; 937 938 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 939 if (dev != vp->v_rdev || type != vp->v_type) 940 continue; 941 *vpp = vp; 942 return (1); 943 } 944 return (0); 945 } 946 947 /* 948 * Calculate the total number of references to a special device. 949 */ 950 vcount(vp) 951 register struct vnode *vp; 952 { 953 register struct vnode *vq; 954 int count; 955 956 if ((vp->v_flag & VALIASED) == 0) 957 return (vp->v_usecount); 958 loop: 959 for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 960 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 961 continue; 962 /* 963 * Alias, but not in use, so flush it out. 964 */ 965 if (vq->v_usecount == 0) { 966 vgone(vq); 967 goto loop; 968 } 969 count += vq->v_usecount; 970 } 971 return (count); 972 } 973 974 /* 975 * Print out a description of a vnode. 976 */ 977 static char *typename[] = 978 { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; 979 980 vprint(label, vp) 981 char *label; 982 register struct vnode *vp; 983 { 984 char buf[64]; 985 986 if (label != NULL) 987 printf("%s: ", label); 988 printf("type %s, usecount %d, writecount %d, refcount %d,", 989 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 990 vp->v_holdcnt); 991 buf[0] = '\0'; 992 if (vp->v_flag & VROOT) 993 strcat(buf, "|VROOT"); 994 if (vp->v_flag & VTEXT) 995 strcat(buf, "|VTEXT"); 996 if (vp->v_flag & VSYSTEM) 997 strcat(buf, "|VSYSTEM"); 998 if (vp->v_flag & VXLOCK) 999 strcat(buf, "|VXLOCK"); 1000 if (vp->v_flag & VXWANT) 1001 strcat(buf, "|VXWANT"); 1002 if (vp->v_flag & VBWAIT) 1003 strcat(buf, "|VBWAIT"); 1004 if (vp->v_flag & VALIASED) 1005 strcat(buf, "|VALIASED"); 1006 if (buf[0] != '\0') 1007 printf(" flags (%s)", &buf[1]); 1008 if (vp->v_data == NULL) { 1009 printf("\n"); 1010 } else { 1011 printf("\n\t"); 1012 VOP_PRINT(vp); 1013 } 1014 } 1015 1016 #ifdef DEBUG 1017 /* 1018 * List all of the locked vnodes in the system. 1019 * Called when debugging the kernel. 1020 */ 1021 printlockedvnodes() 1022 { 1023 register struct mount *mp; 1024 register struct vnode *vp; 1025 1026 printf("Locked vnodes\n"); 1027 for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) { 1028 for (vp = mp->mnt_vnodelist.lh_first; 1029 vp != NULL; 1030 vp = vp->v_mntvnodes.le_next) 1031 if (VOP_ISLOCKED(vp)) 1032 vprint((char *)0, vp); 1033 } 1034 } 1035 #endif 1036 1037 int kinfo_vdebug = 1; 1038 int kinfo_vgetfailed; 1039 #define KINFO_VNODESLOP 10 1040 /* 1041 * Dump vnode list (via sysctl). 1042 * Copyout address of vnode followed by vnode. 1043 */ 1044 /* ARGSUSED */ 1045 sysctl_vnode(where, sizep) 1046 char *where; 1047 size_t *sizep; 1048 { 1049 register struct mount *mp, *nmp; 1050 struct vnode *vp; 1051 register char *bp = where, *savebp; 1052 char *ewhere; 1053 int error; 1054 1055 #define VPTRSZ sizeof (struct vnode *) 1056 #define VNODESZ sizeof (struct vnode) 1057 if (where == NULL) { 1058 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 1059 return (0); 1060 } 1061 ewhere = where + *sizep; 1062 1063 for (mp = mountlist.tqh_first; mp != NULL; mp = nmp) { 1064 nmp = mp->mnt_list.tqe_next; 1065 if (vfs_busy(mp)) 1066 continue; 1067 savebp = bp; 1068 again: 1069 for (vp = mp->mnt_vnodelist.lh_first; 1070 vp != NULL; 1071 vp = vp->v_mntvnodes.le_next) { 1072 /* 1073 * Check that the vp is still associated with 1074 * this filesystem. RACE: could have been 1075 * recycled onto the same filesystem. 1076 */ 1077 if (vp->v_mount != mp) { 1078 if (kinfo_vdebug) 1079 printf("kinfo: vp changed\n"); 1080 bp = savebp; 1081 goto again; 1082 } 1083 if (bp + VPTRSZ + VNODESZ > ewhere) { 1084 *sizep = bp - where; 1085 return (ENOMEM); 1086 } 1087 if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || 1088 (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ))) 1089 return (error); 1090 bp += VPTRSZ + VNODESZ; 1091 } 1092 vfs_unbusy(mp); 1093 } 1094 1095 *sizep = bp - where; 1096 return (0); 1097 } 1098 1099 /* 1100 * Check to see if a filesystem is mounted on a block device. 1101 */ 1102 int 1103 vfs_mountedon(vp) 1104 register struct vnode *vp; 1105 { 1106 register struct vnode *vq; 1107 1108 if (vp->v_specflags & SI_MOUNTEDON) 1109 return (EBUSY); 1110 if (vp->v_flag & VALIASED) { 1111 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1112 if (vq->v_rdev != vp->v_rdev || 1113 vq->v_type != vp->v_type) 1114 continue; 1115 if (vq->v_specflags & SI_MOUNTEDON) 1116 return (EBUSY); 1117 } 1118 } 1119 return (0); 1120 } 1121 1122 /* 1123 * Build hash lists of net addresses and hang them off the mount point. 1124 * Called by ufs_mount() to set up the lists of export addresses. 1125 */ 1126 static int 1127 vfs_hang_addrlist(mp, nep, argp) 1128 struct mount *mp; 1129 struct netexport *nep; 1130 struct export_args *argp; 1131 { 1132 register struct netcred *np; 1133 register struct radix_node_head *rnh; 1134 register int i; 1135 struct radix_node *rn; 1136 struct sockaddr *saddr, *smask = 0; 1137 struct domain *dom; 1138 int error; 1139 1140 if (argp->ex_addrlen == 0) { 1141 if (mp->mnt_flag & MNT_DEFEXPORTED) 1142 return (EPERM); 1143 np = &nep->ne_defexported; 1144 np->netc_exflags = argp->ex_flags; 1145 np->netc_anon = argp->ex_anon; 1146 np->netc_anon.cr_ref = 1; 1147 mp->mnt_flag |= MNT_DEFEXPORTED; 1148 return (0); 1149 } 1150 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1151 np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); 1152 bzero((caddr_t)np, i); 1153 saddr = (struct sockaddr *)(np + 1); 1154 if (error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen)) 1155 goto out; 1156 if (saddr->sa_len > argp->ex_addrlen) 1157 saddr->sa_len = argp->ex_addrlen; 1158 if (argp->ex_masklen) { 1159 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1160 error = copyin(argp->ex_addr, (caddr_t)smask, argp->ex_masklen); 1161 if (error) 1162 goto out; 1163 if (smask->sa_len > argp->ex_masklen) 1164 smask->sa_len = argp->ex_masklen; 1165 } 1166 i = saddr->sa_family; 1167 if ((rnh = nep->ne_rtable[i]) == 0) { 1168 /* 1169 * Seems silly to initialize every AF when most are not 1170 * used, do so on demand here 1171 */ 1172 for (dom = domains; dom; dom = dom->dom_next) 1173 if (dom->dom_family == i && dom->dom_rtattach) { 1174 dom->dom_rtattach((void **)&nep->ne_rtable[i], 1175 dom->dom_rtoffset); 1176 break; 1177 } 1178 if ((rnh = nep->ne_rtable[i]) == 0) { 1179 error = ENOBUFS; 1180 goto out; 1181 } 1182 } 1183 rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh, 1184 np->netc_rnodes); 1185 if (rn == 0 || np != (struct netcred *)rn) { /* already exists */ 1186 error = EPERM; 1187 goto out; 1188 } 1189 np->netc_exflags = argp->ex_flags; 1190 np->netc_anon = argp->ex_anon; 1191 np->netc_anon.cr_ref = 1; 1192 return (0); 1193 out: 1194 free(np, M_NETADDR); 1195 return (error); 1196 } 1197 1198 /* ARGSUSED */ 1199 static int 1200 vfs_free_netcred(rn, w) 1201 struct radix_node *rn; 1202 caddr_t w; 1203 { 1204 register struct radix_node_head *rnh = (struct radix_node_head *)w; 1205 1206 (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh); 1207 free((caddr_t)rn, M_NETADDR); 1208 return (0); 1209 } 1210 1211 /* 1212 * Free the net address hash lists that are hanging off the mount points. 1213 */ 1214 static void 1215 vfs_free_addrlist(nep) 1216 struct netexport *nep; 1217 { 1218 register int i; 1219 register struct radix_node_head *rnh; 1220 1221 for (i = 0; i <= AF_MAX; i++) 1222 if (rnh = nep->ne_rtable[i]) { 1223 (*rnh->rnh_walktree)(rnh, vfs_free_netcred, 1224 (caddr_t)rnh); 1225 free((caddr_t)rnh, M_RTABLE); 1226 nep->ne_rtable[i] = 0; 1227 } 1228 } 1229 1230 int 1231 vfs_export(mp, nep, argp) 1232 struct mount *mp; 1233 struct netexport *nep; 1234 struct export_args *argp; 1235 { 1236 int error; 1237 1238 if (argp->ex_flags & MNT_DELEXPORT) { 1239 vfs_free_addrlist(nep); 1240 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1241 } 1242 if (argp->ex_flags & MNT_EXPORTED) { 1243 if (error = vfs_hang_addrlist(mp, nep, argp)) 1244 return (error); 1245 mp->mnt_flag |= MNT_EXPORTED; 1246 } 1247 return (0); 1248 } 1249 1250 struct netcred * 1251 vfs_export_lookup(mp, nep, nam) 1252 register struct mount *mp; 1253 struct netexport *nep; 1254 struct mbuf *nam; 1255 { 1256 register struct netcred *np; 1257 register struct radix_node_head *rnh; 1258 struct sockaddr *saddr; 1259 1260 np = NULL; 1261 if (mp->mnt_flag & MNT_EXPORTED) { 1262 /* 1263 * Lookup in the export list first. 1264 */ 1265 if (nam != NULL) { 1266 saddr = mtod(nam, struct sockaddr *); 1267 rnh = nep->ne_rtable[saddr->sa_family]; 1268 if (rnh != NULL) { 1269 np = (struct netcred *) 1270 (*rnh->rnh_matchaddr)((caddr_t)saddr, 1271 rnh); 1272 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1273 np = NULL; 1274 } 1275 } 1276 /* 1277 * If no address match, use the default if it exists. 1278 */ 1279 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1280 np = &nep->ne_defexported; 1281 } 1282 return (np); 1283 } 1284