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 8.27 (Berkeley) 05/18/95 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 TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 57 struct mntlist mountlist; /* mounted filesystem list */ 58 static struct simplelock mntid_slock; 59 struct simplelock mntvnode_slock; 60 static struct simplelock spechash_slock; 61 static struct simplelock vnode_free_list_slock; 62 63 /* 64 * Initialize the vnode management data structures. 65 */ 66 void 67 vntblinit() 68 { 69 70 simple_lock_init(&mntvnode_slock); 71 simple_lock_init(&mntid_slock); 72 simple_lock_init(&spechash_slock); 73 TAILQ_INIT(&vnode_free_list); 74 simple_lock_init(&vnode_free_list_slock); 75 CIRCLEQ_INIT(&mountlist); 76 } 77 78 /* 79 * Lock a filesystem. 80 * Used to prevent access to it while mounting and unmounting. 81 */ 82 int 83 vfs_lock(mp) 84 register struct mount *mp; 85 { 86 87 while (mp->mnt_flag & MNT_MLOCK) { 88 mp->mnt_flag |= MNT_MWAIT; 89 tsleep((caddr_t)mp, PVFS, "vfslock", 0); 90 } 91 mp->mnt_flag |= MNT_MLOCK; 92 return (0); 93 } 94 95 /* 96 * Unlock a locked filesystem. 97 * Panic if filesystem is not locked. 98 */ 99 void 100 vfs_unlock(mp) 101 register struct mount *mp; 102 { 103 104 if ((mp->mnt_flag & MNT_MLOCK) == 0) 105 panic("vfs_unlock: not locked"); 106 mp->mnt_flag &= ~MNT_MLOCK; 107 if (mp->mnt_flag & MNT_MWAIT) { 108 mp->mnt_flag &= ~MNT_MWAIT; 109 wakeup((caddr_t)mp); 110 } 111 } 112 113 /* 114 * Mark a mount point as busy. 115 * Used to synchronize access and to delay unmounting. 116 */ 117 int 118 vfs_busy(mp) 119 register struct mount *mp; 120 { 121 122 while (mp->mnt_flag & MNT_MPBUSY) { 123 mp->mnt_flag |= MNT_MPWANT; 124 tsleep((caddr_t)&mp->mnt_flag, PVFS, "vfsbusy", 0); 125 } 126 if (mp->mnt_flag & MNT_UNMOUNT) 127 return (1); 128 mp->mnt_flag |= MNT_MPBUSY; 129 return (0); 130 } 131 132 /* 133 * Free a busy filesystem. 134 * Panic if filesystem is not busy. 135 */ 136 void 137 vfs_unbusy(mp) 138 register struct mount *mp; 139 { 140 141 if ((mp->mnt_flag & MNT_MPBUSY) == 0) 142 panic("vfs_unbusy: not busy"); 143 mp->mnt_flag &= ~MNT_MPBUSY; 144 if (mp->mnt_flag & MNT_MPWANT) { 145 mp->mnt_flag &= ~MNT_MPWANT; 146 wakeup((caddr_t)&mp->mnt_flag); 147 } 148 } 149 150 /* 151 * Lookup a filesystem type, and if found allocate and initialize 152 * a mount structure for it. 153 * 154 * Devname is usually updated by mount(8) after booting. 155 */ 156 int 157 vfs_rootmountalloc(fstypename, devname, mpp) 158 char *fstypename; 159 char *devname; 160 struct mount **mpp; 161 { 162 struct vfsconf *vfsp; 163 struct mount *mp; 164 165 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 166 if (!strcmp(vfsp->vfc_name, fstypename)) 167 break; 168 if (vfsp == NULL) 169 return (ENODEV); 170 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 171 bzero((char *)mp, (u_long)sizeof(struct mount)); 172 LIST_INIT(&mp->mnt_vnodelist); 173 mp->mnt_vfc = vfsp; 174 mp->mnt_op = vfsp->vfc_vfsops; 175 mp->mnt_flag = MNT_RDONLY; 176 mp->mnt_vnodecovered = NULLVP; 177 vfsp->vfc_refcount++; 178 mp->mnt_stat.f_type = vfsp->vfc_typenum; 179 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 180 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 181 mp->mnt_stat.f_mntonname[0] = '/'; 182 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 183 *mpp = mp; 184 return (0); 185 } 186 187 /* 188 * Find an appropriate filesystem to use for the root. If a filesystem 189 * has not been preselected, walk through the list of known filesystems 190 * trying those that have mountroot routines, and try them until one 191 * works or we have tried them all. 192 */ 193 int 194 vfs_mountroot() 195 { 196 struct vfsconf *vfsp; 197 extern int (*mountroot)(void); 198 int error; 199 200 if (mountroot != NULL) 201 return ((*mountroot)()); 202 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 203 if (vfsp->vfc_mountroot == NULL) 204 continue; 205 if ((error = (*vfsp->vfc_mountroot)()) == 0) 206 return (0); 207 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 208 } 209 return (ENODEV); 210 } 211 212 /* 213 * Lookup a mount point by filesystem identifier. 214 */ 215 struct mount * 216 vfs_getvfs(fsid) 217 fsid_t *fsid; 218 { 219 register struct mount *mp; 220 221 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 222 mp = mp->mnt_list.cqe_next) { 223 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 224 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) 225 return (mp); 226 } 227 return ((struct mount *)0); 228 } 229 230 /* 231 * Get a new unique fsid 232 */ 233 void 234 vfs_getnewfsid(mp) 235 struct mount *mp; 236 { 237 static u_short xxxfs_mntid; 238 239 fsid_t tfsid; 240 int mtype; 241 242 simple_lock(&mntid_slock); 243 mtype = mp->mnt_vfc->vfc_typenum; 244 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 245 mp->mnt_stat.f_fsid.val[1] = mtype; 246 if (xxxfs_mntid == 0) 247 ++xxxfs_mntid; 248 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 249 tfsid.val[1] = mtype; 250 if (mountlist.cqh_first != (void *)&mountlist) { 251 while (vfs_getvfs(&tfsid)) { 252 tfsid.val[0]++; 253 xxxfs_mntid++; 254 } 255 } 256 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 257 simple_unlock(&mntid_slock); 258 } 259 260 /* 261 * Set vnode attributes to VNOVAL 262 */ 263 void 264 vattr_null(vap) 265 register struct vattr *vap; 266 { 267 268 vap->va_type = VNON; 269 vap->va_size = vap->va_bytes = VNOVAL; 270 vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = 271 vap->va_fsid = vap->va_fileid = 272 vap->va_blocksize = vap->va_rdev = 273 vap->va_atime.ts_sec = vap->va_atime.ts_nsec = 274 vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec = 275 vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec = 276 vap->va_flags = vap->va_gen = VNOVAL; 277 vap->va_vaflags = 0; 278 } 279 280 /* 281 * Routines having to do with the management of the vnode table. 282 */ 283 extern int (**dead_vnodeop_p)(); 284 static void vclean __P((struct vnode *vp, int flag, struct proc *p)); 285 extern void vgonel __P((struct vnode *vp, struct proc *p)); 286 long numvnodes; 287 extern struct vattr va_null; 288 289 /* 290 * Return the next vnode from the free list. 291 */ 292 int 293 getnewvnode(tag, mp, vops, vpp) 294 enum vtagtype tag; 295 struct mount *mp; 296 int (**vops)(); 297 struct vnode **vpp; 298 { 299 struct proc *p = curproc; /* XXX */ 300 struct vnode *vp; 301 int s; 302 int cnt; 303 304 top: 305 simple_lock(&vnode_free_list_slock); 306 if ((vnode_free_list.tqh_first == NULL && 307 numvnodes < 2 * desiredvnodes) || 308 numvnodes < desiredvnodes) { 309 simple_unlock(&vnode_free_list_slock); 310 vp = (struct vnode *)malloc((u_long)sizeof *vp, 311 M_VNODE, M_WAITOK); 312 bzero((char *)vp, sizeof *vp); 313 numvnodes++; 314 } else { 315 for (vp = vnode_free_list.tqh_first; 316 vp != NULLVP; vp = vp->v_freelist.tqe_next) { 317 if (simple_lock_try(&vp->v_interlock)) 318 break; 319 } 320 /* 321 * Unless this is a bad time of the month, at most 322 * the first NCPUS items on the free list are 323 * locked, so this is close enough to being empty. 324 */ 325 if (vp == NULLVP) { 326 simple_unlock(&vnode_free_list_slock); 327 tablefull("vnode"); 328 *vpp = 0; 329 return (ENFILE); 330 } 331 if (vp->v_usecount) 332 panic("free vnode isn't"); 333 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 334 /* see comment on why 0xdeadb is set at end of vgone (below) */ 335 vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; 336 simple_unlock(&vnode_free_list_slock); 337 vp->v_lease = NULL; 338 if (vp->v_type != VBAD) 339 vgonel(vp, p); 340 else 341 simple_unlock(&vp->v_interlock); 342 #ifdef DIAGNOSTIC 343 if (vp->v_data) 344 panic("cleaned vnode isn't"); 345 s = splbio(); 346 if (vp->v_numoutput) 347 panic("Clean vnode has pending I/O's"); 348 splx(s); 349 #endif 350 vp->v_flag = 0; 351 vp->v_lastr = 0; 352 vp->v_ralen = 0; 353 vp->v_maxra = 0; 354 vp->v_lastw = 0; 355 vp->v_lasta = 0; 356 vp->v_cstart = 0; 357 vp->v_clen = 0; 358 vp->v_socket = 0; 359 } 360 vp->v_type = VNON; 361 cache_purge(vp); 362 vp->v_tag = tag; 363 vp->v_op = vops; 364 insmntque(vp, mp); 365 *vpp = vp; 366 vp->v_usecount = 1; 367 vp->v_data = 0; 368 return (0); 369 } 370 371 /* 372 * Move a vnode from one mount queue to another. 373 */ 374 void 375 insmntque(vp, mp) 376 struct vnode *vp; 377 struct mount *mp; 378 { 379 380 simple_lock(&mntvnode_slock); 381 /* 382 * Delete from old mount point vnode list, if on one. 383 */ 384 if (vp->v_mount != NULL) 385 LIST_REMOVE(vp, v_mntvnodes); 386 /* 387 * Insert into list of vnodes for the new mount point, if available. 388 */ 389 if ((vp->v_mount = mp) != NULL) 390 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 391 simple_unlock(&mntvnode_slock); 392 } 393 394 /* 395 * Update outstanding I/O count and do wakeup if requested. 396 */ 397 void 398 vwakeup(bp) 399 register struct buf *bp; 400 { 401 register struct vnode *vp; 402 403 bp->b_flags &= ~B_WRITEINPROG; 404 if (vp = bp->b_vp) { 405 if (--vp->v_numoutput < 0) 406 panic("vwakeup: neg numoutput"); 407 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 408 if (vp->v_numoutput < 0) 409 panic("vwakeup: neg numoutput 2"); 410 vp->v_flag &= ~VBWAIT; 411 wakeup((caddr_t)&vp->v_numoutput); 412 } 413 } 414 } 415 416 /* 417 * Flush out and invalidate all buffers associated with a vnode. 418 * Called with the underlying object locked. 419 */ 420 int 421 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 422 register struct vnode *vp; 423 int flags; 424 struct ucred *cred; 425 struct proc *p; 426 int slpflag, slptimeo; 427 { 428 register struct buf *bp; 429 struct buf *nbp, *blist; 430 int s, error; 431 432 if (flags & V_SAVE) { 433 if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) 434 return (error); 435 if (vp->v_dirtyblkhd.lh_first != NULL) 436 panic("vinvalbuf: dirty bufs"); 437 } 438 for (;;) { 439 if ((blist = vp->v_cleanblkhd.lh_first) && flags & V_SAVEMETA) 440 while (blist && blist->b_lblkno < 0) 441 blist = blist->b_vnbufs.le_next; 442 if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 443 (flags & V_SAVEMETA)) 444 while (blist && blist->b_lblkno < 0) 445 blist = blist->b_vnbufs.le_next; 446 if (!blist) 447 break; 448 449 for (bp = blist; bp; bp = nbp) { 450 nbp = bp->b_vnbufs.le_next; 451 if (flags & V_SAVEMETA && bp->b_lblkno < 0) 452 continue; 453 s = splbio(); 454 if (bp->b_flags & B_BUSY) { 455 bp->b_flags |= B_WANTED; 456 error = tsleep((caddr_t)bp, 457 slpflag | (PRIBIO + 1), "vinvalbuf", 458 slptimeo); 459 splx(s); 460 if (error) 461 return (error); 462 break; 463 } 464 bremfree(bp); 465 bp->b_flags |= B_BUSY; 466 splx(s); 467 /* 468 * XXX Since there are no node locks for NFS, I believe 469 * there is a slight chance that a delayed write will 470 * occur while sleeping just above, so check for it. 471 */ 472 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 473 (void) VOP_BWRITE(bp); 474 break; 475 } 476 bp->b_flags |= B_INVAL; 477 brelse(bp); 478 } 479 } 480 if (!(flags & V_SAVEMETA) && 481 (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) 482 panic("vinvalbuf: flush failed"); 483 return (0); 484 } 485 486 /* 487 * Associate a buffer with a vnode. 488 */ 489 void 490 bgetvp(vp, bp) 491 register struct vnode *vp; 492 register struct buf *bp; 493 { 494 495 if (bp->b_vp) 496 panic("bgetvp: not free"); 497 VHOLD(vp); 498 bp->b_vp = vp; 499 if (vp->v_type == VBLK || vp->v_type == VCHR) 500 bp->b_dev = vp->v_rdev; 501 else 502 bp->b_dev = NODEV; 503 /* 504 * Insert onto list for new vnode. 505 */ 506 bufinsvn(bp, &vp->v_cleanblkhd); 507 } 508 509 /* 510 * Disassociate a buffer from a vnode. 511 */ 512 void 513 brelvp(bp) 514 register struct buf *bp; 515 { 516 struct vnode *vp; 517 518 if (bp->b_vp == (struct vnode *) 0) 519 panic("brelvp: NULL"); 520 /* 521 * Delete from old vnode list, if on one. 522 */ 523 if (bp->b_vnbufs.le_next != NOLIST) 524 bufremvn(bp); 525 vp = bp->b_vp; 526 bp->b_vp = (struct vnode *) 0; 527 HOLDRELE(vp); 528 } 529 530 /* 531 * Reassign a buffer from one vnode to another. 532 * Used to assign file specific control information 533 * (indirect blocks) to the vnode to which they belong. 534 */ 535 void 536 reassignbuf(bp, newvp) 537 register struct buf *bp; 538 register struct vnode *newvp; 539 { 540 register struct buflists *listheadp; 541 542 if (newvp == NULL) { 543 printf("reassignbuf: NULL"); 544 return; 545 } 546 /* 547 * Delete from old vnode list, if on one. 548 */ 549 if (bp->b_vnbufs.le_next != NOLIST) 550 bufremvn(bp); 551 /* 552 * If dirty, put on list of dirty buffers; 553 * otherwise insert onto list of clean buffers. 554 */ 555 if (bp->b_flags & B_DELWRI) 556 listheadp = &newvp->v_dirtyblkhd; 557 else 558 listheadp = &newvp->v_cleanblkhd; 559 bufinsvn(bp, listheadp); 560 } 561 562 /* 563 * Create a vnode for a block device. 564 * Used for root filesystem, argdev, and swap areas. 565 * Also used for memory file system special devices. 566 */ 567 int 568 bdevvp(dev, vpp) 569 dev_t dev; 570 struct vnode **vpp; 571 { 572 register struct vnode *vp; 573 struct vnode *nvp; 574 int error; 575 576 if (dev == NODEV) { 577 *vpp = NULLVP; 578 return (ENODEV); 579 } 580 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 581 if (error) { 582 *vpp = NULLVP; 583 return (error); 584 } 585 vp = nvp; 586 vp->v_type = VBLK; 587 if (nvp = checkalias(vp, dev, (struct mount *)0)) { 588 vput(vp); 589 vp = nvp; 590 } 591 *vpp = vp; 592 return (0); 593 } 594 595 /* 596 * Check to see if the new vnode represents a special device 597 * for which we already have a vnode (either because of 598 * bdevvp() or because of a different vnode representing 599 * the same block device). If such an alias exists, deallocate 600 * the existing contents and return the aliased vnode. The 601 * caller is responsible for filling it with its new contents. 602 */ 603 struct vnode * 604 checkalias(nvp, nvp_rdev, mp) 605 register struct vnode *nvp; 606 dev_t nvp_rdev; 607 struct mount *mp; 608 { 609 struct proc *p = curproc; /* XXX */ 610 struct vnode *vp; 611 struct vnode **vpp; 612 613 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 614 return (NULLVP); 615 616 vpp = &speclisth[SPECHASH(nvp_rdev)]; 617 loop: 618 simple_lock(&spechash_slock); 619 for (vp = *vpp; vp; vp = vp->v_specnext) { 620 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 621 continue; 622 /* 623 * Alias, but not in use, so flush it out. 624 */ 625 simple_lock(&vp->v_interlock); 626 if (vp->v_usecount == 0) { 627 simple_unlock(&spechash_slock); 628 vgonel(vp, p); 629 goto loop; 630 } 631 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { 632 simple_unlock(&spechash_slock); 633 goto loop; 634 } 635 break; 636 } 637 if (vp == NULL || vp->v_tag != VT_NON) { 638 MALLOC(nvp->v_specinfo, struct specinfo *, 639 sizeof(struct specinfo), M_VNODE, M_WAITOK); 640 nvp->v_rdev = nvp_rdev; 641 nvp->v_hashchain = vpp; 642 nvp->v_specnext = *vpp; 643 nvp->v_specflags = 0; 644 simple_unlock(&spechash_slock); 645 *vpp = nvp; 646 if (vp != NULLVP) { 647 nvp->v_flag |= VALIASED; 648 vp->v_flag |= VALIASED; 649 vput(vp); 650 } 651 return (NULLVP); 652 } 653 simple_unlock(&spechash_slock); 654 VOP_UNLOCK(vp, 0, p); 655 simple_lock(&vp->v_interlock); 656 vclean(vp, 0, p); 657 vp->v_op = nvp->v_op; 658 vp->v_tag = nvp->v_tag; 659 nvp->v_type = VNON; 660 insmntque(vp, mp); 661 return (vp); 662 } 663 664 /* 665 * Grab a particular vnode from the free list, increment its 666 * reference count and lock it. The vnode lock bit is set the 667 * vnode is being eliminated in vgone. The process is awakened 668 * when the transition is completed, and an error returned to 669 * indicate that the vnode is no longer usable (possibly having 670 * been changed to a new file system type). 671 */ 672 int 673 vget(vp, flags, p) 674 struct vnode *vp; 675 int flags; 676 struct proc *p; 677 { 678 int error; 679 680 /* 681 * If the vnode is in the process of being cleaned out for 682 * another use, we wait for the cleaning to finish and then 683 * return failure. Cleaning is determined by checking that 684 * the VXLOCK flag is set. 685 */ 686 if ((flags & LK_INTERLOCK) == 0) 687 simple_lock(&vp->v_interlock); 688 if (vp->v_flag & VXLOCK) { 689 vp->v_flag |= VXWANT; 690 simple_unlock(&vp->v_interlock); 691 tsleep((caddr_t)vp, PINOD, "vget", 0); 692 return (ENOENT); 693 } 694 if (vp->v_usecount == 0) { 695 simple_lock(&vnode_free_list_slock); 696 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 697 simple_unlock(&vnode_free_list_slock); 698 } 699 vp->v_usecount++; 700 if (flags & LK_TYPE_MASK) { 701 if (error = vn_lock(vp, flags | LK_INTERLOCK, p)) 702 vrele(vp); 703 return (error); 704 } 705 simple_unlock(&vp->v_interlock); 706 return (0); 707 } 708 709 /* 710 * Stubs to use when there is no locking to be done on the underlying object. 711 * 712 * Getting a lock just clears the interlock if necessary. 713 */ 714 int 715 vop_nolock(ap) 716 struct vop_lock_args /* { 717 struct vnode *a_vp; 718 int a_flags; 719 struct proc *a_p; 720 } */ *ap; 721 { 722 struct vnode *vp = ap->a_vp; 723 724 /* 725 * Since we are not using the lock manager, we must clear 726 * the interlock here. 727 */ 728 if (ap->a_flags & LK_INTERLOCK) 729 simple_unlock(&vp->v_interlock); 730 return (0); 731 } 732 733 /* 734 * Unlock has nothing to do. 735 */ 736 int 737 vop_nounlock(ap) 738 struct vop_unlock_args /* { 739 struct vnode *a_vp; 740 int a_flags; 741 struct proc *a_p; 742 } */ *ap; 743 { 744 745 return (0); 746 } 747 748 /* 749 * Nothing is ever locked. 750 */ 751 int 752 vop_noislocked(ap) 753 struct vop_islocked_args /* { 754 struct vnode *a_vp; 755 } */ *ap; 756 { 757 758 return (0); 759 } 760 761 /* 762 * Vnode reference. 763 */ 764 void 765 vref(vp) 766 struct vnode *vp; 767 { 768 769 simple_lock(&vp->v_interlock); 770 if (vp->v_usecount <= 0) 771 panic("vref used where vget required"); 772 vp->v_usecount++; 773 simple_unlock(&vp->v_interlock); 774 } 775 776 /* 777 * vput(), just unlock and vrele() 778 */ 779 void 780 vput(vp) 781 struct vnode *vp; 782 { 783 struct proc *p = curproc; /* XXX */ 784 785 #ifdef DIGANOSTIC 786 if (vp == NULL) 787 panic("vput: null vp"); 788 #endif 789 simple_lock(&vp->v_interlock); 790 vp->v_usecount--; 791 if (vp->v_usecount > 0) { 792 simple_unlock(&vp->v_interlock); 793 VOP_UNLOCK(vp, 0, p); 794 return; 795 } 796 #ifdef DIAGNOSTIC 797 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 798 vprint("vput: bad ref count", vp); 799 panic("vput: ref cnt"); 800 } 801 #endif 802 /* 803 * insert at tail of LRU list 804 */ 805 simple_lock(&vnode_free_list_slock); 806 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 807 simple_unlock(&vnode_free_list_slock); 808 simple_unlock(&vp->v_interlock); 809 VOP_INACTIVE(vp, p); 810 } 811 812 /* 813 * Vnode release. 814 * If count drops to zero, call inactive routine and return to freelist. 815 */ 816 void 817 vrele(vp) 818 struct vnode *vp; 819 { 820 struct proc *p = curproc; /* XXX */ 821 822 #ifdef DIAGNOSTIC 823 if (vp == NULL) 824 panic("vrele: null vp"); 825 #endif 826 simple_lock(&vp->v_interlock); 827 vp->v_usecount--; 828 if (vp->v_usecount > 0) { 829 simple_unlock(&vp->v_interlock); 830 return; 831 } 832 #ifdef DIAGNOSTIC 833 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 834 vprint("vrele: bad ref count", vp); 835 panic("vrele: ref cnt"); 836 } 837 #endif 838 /* 839 * insert at tail of LRU list 840 */ 841 simple_lock(&vnode_free_list_slock); 842 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 843 simple_unlock(&vnode_free_list_slock); 844 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) 845 VOP_INACTIVE(vp, p); 846 } 847 848 #ifdef DIAGNOSTIC 849 /* 850 * Page or buffer structure gets a reference. 851 */ 852 void 853 vhold(vp) 854 register struct vnode *vp; 855 { 856 857 simple_lock(&vp->v_interlock); 858 vp->v_holdcnt++; 859 simple_unlock(&vp->v_interlock); 860 } 861 862 /* 863 * Page or buffer structure frees a reference. 864 */ 865 void 866 holdrele(vp) 867 register struct vnode *vp; 868 { 869 870 simple_lock(&vp->v_interlock); 871 if (vp->v_holdcnt <= 0) 872 panic("holdrele: holdcnt"); 873 vp->v_holdcnt--; 874 simple_unlock(&vp->v_interlock); 875 } 876 #endif /* DIAGNOSTIC */ 877 878 /* 879 * Remove any vnodes in the vnode table belonging to mount point mp. 880 * 881 * If MNT_NOFORCE is specified, there should not be any active ones, 882 * return error if any are found (nb: this is a user error, not a 883 * system error). If MNT_FORCE is specified, detach any active vnodes 884 * that are found. 885 */ 886 #ifdef DIAGNOSTIC 887 int busyprt = 0; /* print out busy vnodes */ 888 struct ctldebug debug1 = { "busyprt", &busyprt }; 889 #endif 890 891 int 892 vflush(mp, skipvp, flags) 893 struct mount *mp; 894 struct vnode *skipvp; 895 int flags; 896 { 897 struct proc *p = curproc; /* XXX */ 898 struct vnode *vp, *nvp; 899 int busy = 0; 900 901 #ifdef DIAGNOSTIC 902 if ((mp->mnt_flag & MNT_MPBUSY) == 0) 903 panic("vflush: not busy"); 904 #endif 905 906 simple_lock(&mntvnode_slock); 907 loop: 908 for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { 909 if (vp->v_mount != mp) 910 goto loop; 911 nvp = vp->v_mntvnodes.le_next; 912 /* 913 * Skip over a selected vnode. 914 */ 915 if (vp == skipvp) 916 continue; 917 918 simple_lock(&vp->v_interlock); 919 /* 920 * Skip over a vnodes marked VSYSTEM. 921 */ 922 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 923 simple_unlock(&vp->v_interlock); 924 continue; 925 } 926 /* 927 * If WRITECLOSE is set, only flush out regular file 928 * vnodes open for writing. 929 */ 930 if ((flags & WRITECLOSE) && 931 (vp->v_writecount == 0 || vp->v_type != VREG)) { 932 simple_unlock(&vp->v_interlock); 933 continue; 934 } 935 /* 936 * With v_usecount == 0, all we need to do is clear 937 * out the vnode data structures and we are done. 938 */ 939 if (vp->v_usecount == 0) { 940 simple_unlock(&mntvnode_slock); 941 vgonel(vp, p); 942 simple_lock(&mntvnode_slock); 943 continue; 944 } 945 /* 946 * If FORCECLOSE is set, forcibly close the vnode. 947 * For block or character devices, revert to an 948 * anonymous device. For all other files, just kill them. 949 */ 950 if (flags & FORCECLOSE) { 951 simple_unlock(&mntvnode_slock); 952 if (vp->v_type != VBLK && vp->v_type != VCHR) { 953 vgonel(vp, p); 954 } else { 955 vclean(vp, 0, p); 956 vp->v_op = spec_vnodeop_p; 957 insmntque(vp, (struct mount *)0); 958 } 959 simple_lock(&mntvnode_slock); 960 continue; 961 } 962 #ifdef DIAGNOSTIC 963 if (busyprt) 964 vprint("vflush: busy vnode", vp); 965 #endif 966 simple_unlock(&vp->v_interlock); 967 busy++; 968 } 969 simple_unlock(&mntvnode_slock); 970 if (busy) 971 return (EBUSY); 972 return (0); 973 } 974 975 /* 976 * Disassociate the underlying file system from a vnode. 977 * The vnode interlock is held on entry. 978 */ 979 static void 980 vclean(vp, flags, p) 981 struct vnode *vp; 982 int flags; 983 struct proc *p; 984 { 985 int active; 986 987 /* 988 * Check to see if the vnode is in use. 989 * If so we have to reference it before we clean it out 990 * so that its count cannot fall to zero and generate a 991 * race against ourselves to recycle it. 992 */ 993 if (active = vp->v_usecount) 994 vp->v_usecount++; 995 /* 996 * Prevent the vnode from being recycled or 997 * brought into use while we clean it out. 998 */ 999 if (vp->v_flag & VXLOCK) 1000 panic("vclean: deadlock"); 1001 vp->v_flag |= VXLOCK; 1002 /* 1003 * Even if the count is zero, the VOP_INACTIVE routine may still 1004 * have the object locked while it cleans it out. The VOP_LOCK 1005 * ensures that the VOP_INACTIVE routine is done with its work. 1006 * For active vnodes, it ensures that no other activity can 1007 * occur while the underlying object is being cleaned out. 1008 */ 1009 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1010 /* 1011 * Clean out any buffers associated with the vnode. 1012 */ 1013 if (flags & DOCLOSE) 1014 vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0); 1015 /* 1016 * If purging an active vnode, it must be closed and 1017 * deactivated before being reclaimed. Note that the 1018 * VOP_INACTIVE will unlock the vnode. 1019 */ 1020 if (active) { 1021 if (flags & DOCLOSE) 1022 VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); 1023 VOP_INACTIVE(vp, p); 1024 } else { 1025 /* 1026 * Any other processes trying to obtain this lock must first 1027 * wait for VXLOCK to clear, then call the new lock operation. 1028 */ 1029 VOP_UNLOCK(vp, 0, p); 1030 } 1031 /* 1032 * Reclaim the vnode. 1033 */ 1034 if (VOP_RECLAIM(vp, p)) 1035 panic("vclean: cannot reclaim"); 1036 if (active) 1037 vrele(vp); 1038 cache_purge(vp); 1039 1040 /* 1041 * Done with purge, notify sleepers of the grim news. 1042 */ 1043 vp->v_op = dead_vnodeop_p; 1044 vp->v_tag = VT_NON; 1045 vp->v_flag &= ~VXLOCK; 1046 if (vp->v_flag & VXWANT) { 1047 vp->v_flag &= ~VXWANT; 1048 wakeup((caddr_t)vp); 1049 } 1050 } 1051 1052 /* 1053 * Eliminate all activity associated with the requested vnode 1054 * and with all vnodes aliased to the requested vnode. 1055 */ 1056 int 1057 vop_revoke(ap) 1058 struct vop_revoke_args /* { 1059 struct vnode *a_vp; 1060 int a_flags; 1061 } */ *ap; 1062 { 1063 struct vnode *vp, *vq; 1064 struct proc *p = curproc; /* XXX */ 1065 1066 #ifdef DIAGNOSTIC 1067 if ((ap->a_flags & REVOKEALL) == 0) 1068 panic("vop_revoke"); 1069 #endif 1070 1071 vp = ap->a_vp; 1072 simple_lock(&vp->v_interlock); 1073 1074 if (vp->v_flag & VALIASED) { 1075 /* 1076 * If a vgone (or vclean) is already in progress, 1077 * wait until it is done and return. 1078 */ 1079 if (vp->v_flag & VXLOCK) { 1080 vp->v_flag |= VXWANT; 1081 simple_unlock(&vp->v_interlock); 1082 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1083 return (0); 1084 } 1085 /* 1086 * Ensure that vp will not be vgone'd while we 1087 * are eliminating its aliases. 1088 */ 1089 vp->v_flag |= VXLOCK; 1090 simple_unlock(&vp->v_interlock); 1091 while (vp->v_flag & VALIASED) { 1092 simple_lock(&spechash_slock); 1093 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1094 if (vq->v_rdev != vp->v_rdev || 1095 vq->v_type != vp->v_type || vp == vq) 1096 continue; 1097 simple_unlock(&spechash_slock); 1098 vgone(vq); 1099 break; 1100 } 1101 if (vq == NULLVP) 1102 simple_unlock(&spechash_slock); 1103 } 1104 /* 1105 * Remove the lock so that vgone below will 1106 * really eliminate the vnode after which time 1107 * vgone will awaken any sleepers. 1108 */ 1109 simple_lock(&vp->v_interlock); 1110 vp->v_flag &= ~VXLOCK; 1111 } 1112 vgonel(vp, p); 1113 return (0); 1114 } 1115 1116 /* 1117 * Recycle an unused vnode to the front of the free list. 1118 * Release the passed interlock if the vnode will be recycled. 1119 */ 1120 int 1121 vrecycle(vp, inter_lkp, p) 1122 struct vnode *vp; 1123 struct simplelock *inter_lkp; 1124 struct proc *p; 1125 { 1126 1127 simple_lock(&vp->v_interlock); 1128 if (vp->v_usecount == 0) { 1129 if (inter_lkp) 1130 simple_unlock(inter_lkp); 1131 vgonel(vp, p); 1132 return (1); 1133 } 1134 simple_unlock(&vp->v_interlock); 1135 return (0); 1136 } 1137 1138 /* 1139 * Eliminate all activity associated with a vnode 1140 * in preparation for reuse. 1141 */ 1142 void 1143 vgone(vp) 1144 struct vnode *vp; 1145 { 1146 struct proc *p = curproc; /* XXX */ 1147 1148 simple_lock(&vp->v_interlock); 1149 vgonel(vp, p); 1150 } 1151 1152 /* 1153 * vgone, with the vp interlock held. 1154 */ 1155 void 1156 vgonel(vp, p) 1157 struct vnode *vp; 1158 struct proc *p; 1159 { 1160 struct vnode *vq; 1161 struct vnode *vx; 1162 1163 /* 1164 * If a vgone (or vclean) is already in progress, 1165 * wait until it is done and return. 1166 */ 1167 if (vp->v_flag & VXLOCK) { 1168 vp->v_flag |= VXWANT; 1169 simple_unlock(&vp->v_interlock); 1170 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1171 return; 1172 } 1173 /* 1174 * Clean out the filesystem specific data. 1175 */ 1176 vclean(vp, DOCLOSE, p); 1177 /* 1178 * Delete from old mount point vnode list, if on one. 1179 */ 1180 if (vp->v_mount != NULL) 1181 insmntque(vp, (struct mount *)0); 1182 /* 1183 * If special device, remove it from special device alias list 1184 * if it is on one. 1185 */ 1186 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1187 simple_lock(&spechash_slock); 1188 if (*vp->v_hashchain == vp) { 1189 *vp->v_hashchain = vp->v_specnext; 1190 } else { 1191 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1192 if (vq->v_specnext != vp) 1193 continue; 1194 vq->v_specnext = vp->v_specnext; 1195 break; 1196 } 1197 if (vq == NULL) 1198 panic("missing bdev"); 1199 } 1200 if (vp->v_flag & VALIASED) { 1201 vx = NULL; 1202 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1203 if (vq->v_rdev != vp->v_rdev || 1204 vq->v_type != vp->v_type) 1205 continue; 1206 if (vx) 1207 break; 1208 vx = vq; 1209 } 1210 if (vx == NULL) 1211 panic("missing alias"); 1212 if (vq == NULL) 1213 vx->v_flag &= ~VALIASED; 1214 vp->v_flag &= ~VALIASED; 1215 } 1216 simple_unlock(&spechash_slock); 1217 FREE(vp->v_specinfo, M_VNODE); 1218 vp->v_specinfo = NULL; 1219 } 1220 /* 1221 * If it is on the freelist and not already at the head, 1222 * move it to the head of the list. The test of the back 1223 * pointer and the reference count of zero is because 1224 * it will be removed from the free list by getnewvnode, 1225 * but will not have its reference count incremented until 1226 * after calling vgone. If the reference count were 1227 * incremented first, vgone would (incorrectly) try to 1228 * close the previous instance of the underlying object. 1229 * So, the back pointer is explicitly set to `0xdeadb' in 1230 * getnewvnode after removing it from the freelist to ensure 1231 * that we do not try to move it here. 1232 */ 1233 if (vp->v_usecount == 0) { 1234 simple_lock(&vnode_free_list_slock); 1235 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1236 vnode_free_list.tqh_first != vp) { 1237 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1238 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1239 } 1240 simple_unlock(&vnode_free_list_slock); 1241 } 1242 vp->v_type = VBAD; 1243 } 1244 1245 /* 1246 * Lookup a vnode by device number. 1247 */ 1248 int 1249 vfinddev(dev, type, vpp) 1250 dev_t dev; 1251 enum vtype type; 1252 struct vnode **vpp; 1253 { 1254 struct vnode *vp; 1255 int rc = 0; 1256 1257 simple_lock(&spechash_slock); 1258 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1259 if (dev != vp->v_rdev || type != vp->v_type) 1260 continue; 1261 *vpp = vp; 1262 rc = 1; 1263 break; 1264 } 1265 simple_unlock(&spechash_slock); 1266 return (rc); 1267 } 1268 1269 /* 1270 * Calculate the total number of references to a special device. 1271 */ 1272 int 1273 vcount(vp) 1274 struct vnode *vp; 1275 { 1276 struct vnode *vq, *vnext; 1277 int count; 1278 1279 loop: 1280 if ((vp->v_flag & VALIASED) == 0) 1281 return (vp->v_usecount); 1282 simple_lock(&spechash_slock); 1283 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1284 vnext = vq->v_specnext; 1285 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1286 continue; 1287 /* 1288 * Alias, but not in use, so flush it out. 1289 */ 1290 if (vq->v_usecount == 0 && vq != vp) { 1291 simple_unlock(&spechash_slock); 1292 vgone(vq); 1293 goto loop; 1294 } 1295 count += vq->v_usecount; 1296 } 1297 simple_unlock(&spechash_slock); 1298 return (count); 1299 } 1300 1301 /* 1302 * Print out a description of a vnode. 1303 */ 1304 static char *typename[] = 1305 { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; 1306 1307 void 1308 vprint(label, vp) 1309 char *label; 1310 register struct vnode *vp; 1311 { 1312 char buf[64]; 1313 1314 if (label != NULL) 1315 printf("%s: ", label); 1316 printf("type %s, usecount %d, writecount %d, refcount %d,", 1317 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1318 vp->v_holdcnt); 1319 buf[0] = '\0'; 1320 if (vp->v_flag & VROOT) 1321 strcat(buf, "|VROOT"); 1322 if (vp->v_flag & VTEXT) 1323 strcat(buf, "|VTEXT"); 1324 if (vp->v_flag & VSYSTEM) 1325 strcat(buf, "|VSYSTEM"); 1326 if (vp->v_flag & VXLOCK) 1327 strcat(buf, "|VXLOCK"); 1328 if (vp->v_flag & VXWANT) 1329 strcat(buf, "|VXWANT"); 1330 if (vp->v_flag & VBWAIT) 1331 strcat(buf, "|VBWAIT"); 1332 if (vp->v_flag & VALIASED) 1333 strcat(buf, "|VALIASED"); 1334 if (buf[0] != '\0') 1335 printf(" flags (%s)", &buf[1]); 1336 if (vp->v_data == NULL) { 1337 printf("\n"); 1338 } else { 1339 printf("\n\t"); 1340 VOP_PRINT(vp); 1341 } 1342 } 1343 1344 #ifdef DEBUG 1345 /* 1346 * List all of the locked vnodes in the system. 1347 * Called when debugging the kernel. 1348 */ 1349 void 1350 printlockedvnodes() 1351 { 1352 register struct mount *mp; 1353 register struct vnode *vp; 1354 1355 printf("Locked vnodes\n"); 1356 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; 1357 mp = mp->mnt_list.cqe_next) { 1358 for (vp = mp->mnt_vnodelist.lh_first; 1359 vp != NULL; 1360 vp = vp->v_mntvnodes.le_next) { 1361 if (VOP_ISLOCKED(vp)) 1362 vprint((char *)0, vp); 1363 } 1364 } 1365 } 1366 #endif 1367 1368 /* 1369 * Top level filesystem related information gathering. 1370 */ 1371 int 1372 vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) 1373 int *name; 1374 u_int namelen; 1375 void *oldp; 1376 size_t *oldlenp; 1377 void *newp; 1378 size_t newlen; 1379 struct proc *p; 1380 { 1381 struct ctldebug *cdp; 1382 struct vfsconf *vfsp; 1383 1384 /* all sysctl names at this level are at least name and field */ 1385 if (namelen < 2) 1386 return (ENOTDIR); /* overloaded */ 1387 if (name[0] != VFS_GENERIC) { 1388 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1389 if (vfsp->vfc_typenum == name[0]) 1390 break; 1391 if (vfsp == NULL) 1392 return (EOPNOTSUPP); 1393 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1394 oldp, oldlenp, newp, newlen, p)); 1395 } 1396 switch (name[1]) { 1397 case VFS_MAXTYPENUM: 1398 return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf)); 1399 case VFS_CONF: 1400 if (namelen < 3) 1401 return (ENOTDIR); /* overloaded */ 1402 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1403 if (vfsp->vfc_typenum == name[2]) 1404 break; 1405 if (vfsp == NULL) 1406 return (EOPNOTSUPP); 1407 return (sysctl_rdstruct(oldp, oldlenp, newp, vfsp, 1408 sizeof(struct vfsconf))); 1409 } 1410 return (EOPNOTSUPP); 1411 } 1412 1413 int kinfo_vdebug = 1; 1414 int kinfo_vgetfailed; 1415 #define KINFO_VNODESLOP 10 1416 /* 1417 * Dump vnode list (via sysctl). 1418 * Copyout address of vnode followed by vnode. 1419 */ 1420 /* ARGSUSED */ 1421 int 1422 sysctl_vnode(where, sizep) 1423 char *where; 1424 size_t *sizep; 1425 { 1426 register struct mount *mp, *nmp; 1427 struct vnode *nvp, *vp; 1428 register char *bp = where, *savebp; 1429 char *ewhere; 1430 int error; 1431 1432 #define VPTRSZ sizeof (struct vnode *) 1433 #define VNODESZ sizeof (struct vnode) 1434 if (where == NULL) { 1435 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 1436 return (0); 1437 } 1438 ewhere = where + *sizep; 1439 1440 for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { 1441 nmp = mp->mnt_list.cqe_next; 1442 if (vfs_busy(mp)) 1443 continue; 1444 savebp = bp; 1445 again: 1446 simple_lock(&mntvnode_slock); 1447 for (vp = mp->mnt_vnodelist.lh_first; 1448 vp != NULL; 1449 vp = nvp) { 1450 /* 1451 * Check that the vp is still associated with 1452 * this filesystem. RACE: could have been 1453 * recycled onto the same filesystem. 1454 */ 1455 if (vp->v_mount != mp) { 1456 simple_unlock(&mntvnode_slock); 1457 if (kinfo_vdebug) 1458 printf("kinfo: vp changed\n"); 1459 bp = savebp; 1460 goto again; 1461 } 1462 nvp = vp->v_mntvnodes.le_next; 1463 if (bp + VPTRSZ + VNODESZ > ewhere) { 1464 simple_unlock(&mntvnode_slock); 1465 *sizep = bp - where; 1466 return (ENOMEM); 1467 } 1468 simple_unlock(&mntvnode_slock); 1469 if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || 1470 (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ))) 1471 return (error); 1472 bp += VPTRSZ + VNODESZ; 1473 simple_lock(&mntvnode_slock); 1474 } 1475 simple_unlock(&mntvnode_slock); 1476 vfs_unbusy(mp); 1477 } 1478 1479 *sizep = bp - where; 1480 return (0); 1481 } 1482 1483 /* 1484 * Check to see if a filesystem is mounted on a block device. 1485 */ 1486 int 1487 vfs_mountedon(vp) 1488 struct vnode *vp; 1489 { 1490 struct vnode *vq; 1491 int error = 0; 1492 1493 if (vp->v_specflags & SI_MOUNTEDON) 1494 return (EBUSY); 1495 if (vp->v_flag & VALIASED) { 1496 simple_lock(&spechash_slock); 1497 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 1498 if (vq->v_rdev != vp->v_rdev || 1499 vq->v_type != vp->v_type) 1500 continue; 1501 if (vq->v_specflags & SI_MOUNTEDON) { 1502 error = EBUSY; 1503 break; 1504 } 1505 } 1506 simple_unlock(&spechash_slock); 1507 } 1508 return (error); 1509 } 1510 1511 /* 1512 * Unmount all filesystems. The list is traversed in reverse order 1513 * of mounting to avoid dependencies. 1514 */ 1515 void 1516 vfs_unmountall() 1517 { 1518 struct mount *mp, *nmp; 1519 1520 for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 1521 nmp = mp->mnt_list.cqe_prev; 1522 (void) dounmount(mp, MNT_FORCE, &proc0); 1523 } 1524 } 1525 1526 /* 1527 * Build hash lists of net addresses and hang them off the mount point. 1528 * Called by ufs_mount() to set up the lists of export addresses. 1529 */ 1530 static int 1531 vfs_hang_addrlist(mp, nep, argp) 1532 struct mount *mp; 1533 struct netexport *nep; 1534 struct export_args *argp; 1535 { 1536 register struct netcred *np; 1537 register struct radix_node_head *rnh; 1538 register int i; 1539 struct radix_node *rn; 1540 struct sockaddr *saddr, *smask = 0; 1541 struct domain *dom; 1542 int error; 1543 1544 if (argp->ex_addrlen == 0) { 1545 if (mp->mnt_flag & MNT_DEFEXPORTED) 1546 return (EPERM); 1547 np = &nep->ne_defexported; 1548 np->netc_exflags = argp->ex_flags; 1549 np->netc_anon = argp->ex_anon; 1550 np->netc_anon.cr_ref = 1; 1551 mp->mnt_flag |= MNT_DEFEXPORTED; 1552 return (0); 1553 } 1554 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1555 np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); 1556 bzero((caddr_t)np, i); 1557 saddr = (struct sockaddr *)(np + 1); 1558 if (error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen)) 1559 goto out; 1560 if (saddr->sa_len > argp->ex_addrlen) 1561 saddr->sa_len = argp->ex_addrlen; 1562 if (argp->ex_masklen) { 1563 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1564 error = copyin(argp->ex_addr, (caddr_t)smask, argp->ex_masklen); 1565 if (error) 1566 goto out; 1567 if (smask->sa_len > argp->ex_masklen) 1568 smask->sa_len = argp->ex_masklen; 1569 } 1570 i = saddr->sa_family; 1571 if ((rnh = nep->ne_rtable[i]) == 0) { 1572 /* 1573 * Seems silly to initialize every AF when most are not 1574 * used, do so on demand here 1575 */ 1576 for (dom = domains; dom; dom = dom->dom_next) 1577 if (dom->dom_family == i && dom->dom_rtattach) { 1578 dom->dom_rtattach((void **)&nep->ne_rtable[i], 1579 dom->dom_rtoffset); 1580 break; 1581 } 1582 if ((rnh = nep->ne_rtable[i]) == 0) { 1583 error = ENOBUFS; 1584 goto out; 1585 } 1586 } 1587 rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh, 1588 np->netc_rnodes); 1589 if (rn == 0) { 1590 /* 1591 * One of the reasons that rnh_addaddr may fail is that 1592 * the entry already exists. To check for this case, we 1593 * look up the entry to see if it is there. If so, we 1594 * do not need to make a new entry but do return success. 1595 */ 1596 free(np, M_NETADDR); 1597 rn = (*rnh->rnh_matchaddr)((caddr_t)saddr, rnh); 1598 if (rn != 0 && (rn->rn_flags & RNF_ROOT) == 0 && 1599 ((struct netcred *)rn)->netc_exflags == argp->ex_flags && 1600 !bcmp((caddr_t)&((struct netcred *)rn)->netc_anon, 1601 (caddr_t)&argp->ex_anon, sizeof(struct ucred))) 1602 return (0); 1603 return (EPERM); 1604 } 1605 np->netc_exflags = argp->ex_flags; 1606 np->netc_anon = argp->ex_anon; 1607 np->netc_anon.cr_ref = 1; 1608 return (0); 1609 out: 1610 free(np, M_NETADDR); 1611 return (error); 1612 } 1613 1614 /* ARGSUSED */ 1615 static int 1616 vfs_free_netcred(rn, w) 1617 struct radix_node *rn; 1618 caddr_t w; 1619 { 1620 register struct radix_node_head *rnh = (struct radix_node_head *)w; 1621 1622 (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh); 1623 free((caddr_t)rn, M_NETADDR); 1624 return (0); 1625 } 1626 1627 /* 1628 * Free the net address hash lists that are hanging off the mount points. 1629 */ 1630 static void 1631 vfs_free_addrlist(nep) 1632 struct netexport *nep; 1633 { 1634 register int i; 1635 register struct radix_node_head *rnh; 1636 1637 for (i = 0; i <= AF_MAX; i++) 1638 if (rnh = nep->ne_rtable[i]) { 1639 (*rnh->rnh_walktree)(rnh, vfs_free_netcred, 1640 (caddr_t)rnh); 1641 free((caddr_t)rnh, M_RTABLE); 1642 nep->ne_rtable[i] = 0; 1643 } 1644 } 1645 1646 int 1647 vfs_export(mp, nep, argp) 1648 struct mount *mp; 1649 struct netexport *nep; 1650 struct export_args *argp; 1651 { 1652 int error; 1653 1654 if (argp->ex_flags & MNT_DELEXPORT) { 1655 vfs_free_addrlist(nep); 1656 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1657 } 1658 if (argp->ex_flags & MNT_EXPORTED) { 1659 if (error = vfs_hang_addrlist(mp, nep, argp)) 1660 return (error); 1661 mp->mnt_flag |= MNT_EXPORTED; 1662 } 1663 return (0); 1664 } 1665 1666 struct netcred * 1667 vfs_export_lookup(mp, nep, nam) 1668 register struct mount *mp; 1669 struct netexport *nep; 1670 struct mbuf *nam; 1671 { 1672 register struct netcred *np; 1673 register struct radix_node_head *rnh; 1674 struct sockaddr *saddr; 1675 1676 np = NULL; 1677 if (mp->mnt_flag & MNT_EXPORTED) { 1678 /* 1679 * Lookup in the export list first. 1680 */ 1681 if (nam != NULL) { 1682 saddr = mtod(nam, struct sockaddr *); 1683 rnh = nep->ne_rtable[saddr->sa_family]; 1684 if (rnh != NULL) { 1685 np = (struct netcred *) 1686 (*rnh->rnh_matchaddr)((caddr_t)saddr, 1687 rnh); 1688 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1689 np = NULL; 1690 } 1691 } 1692 /* 1693 * If no address match, use the default if it exists. 1694 */ 1695 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1696 np = &nep->ne_defexported; 1697 } 1698 return (np); 1699 } 1700