| 1 | /* |
| 2 | * Copyright (c) 1989 The Regents of the University of California. |
| 3 | * All rights reserved. |
| 4 | * |
| 5 | * %sccs.include.redist.c% |
| 6 | * |
| 7 | * @(#)vfs_subr.c 7.70 (Berkeley) %G% |
| 8 | */ |
| 9 | |
| 10 | /* |
| 11 | * External virtual filesystem routines |
| 12 | */ |
| 13 | |
| 14 | #include <sys/param.h> |
| 15 | #include <sys/proc.h> |
| 16 | #include <sys/mount.h> |
| 17 | #include <sys/time.h> |
| 18 | #include <sys/vnode.h> |
| 19 | #include <sys/stat.h> |
| 20 | #include <sys/specdev.h> |
| 21 | #include <sys/namei.h> |
| 22 | #include <sys/ucred.h> |
| 23 | #include <sys/buf.h> |
| 24 | #include <sys/errno.h> |
| 25 | #include <sys/malloc.h> |
| 26 | |
| 27 | enum vtype iftovt_tab[16] = { |
| 28 | VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, |
| 29 | VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, |
| 30 | }; |
| 31 | int vttoif_tab[9] = { |
| 32 | 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, |
| 33 | S_IFSOCK, S_IFIFO, S_IFMT, |
| 34 | }; |
| 35 | |
| 36 | /* |
| 37 | * Remove a mount point from the list of mounted filesystems. |
| 38 | * Unmount of the root is illegal. |
| 39 | */ |
| 40 | void |
| 41 | vfs_remove(mp) |
| 42 | register struct mount *mp; |
| 43 | { |
| 44 | |
| 45 | if (mp == rootfs) |
| 46 | panic("vfs_remove: unmounting root"); |
| 47 | mp->mnt_prev->mnt_next = mp->mnt_next; |
| 48 | mp->mnt_next->mnt_prev = mp->mnt_prev; |
| 49 | mp->mnt_vnodecovered->v_mountedhere = (struct mount *)0; |
| 50 | vfs_unlock(mp); |
| 51 | } |
| 52 | |
| 53 | /* |
| 54 | * Lock a filesystem. |
| 55 | * Used to prevent access to it while mounting and unmounting. |
| 56 | */ |
| 57 | vfs_lock(mp) |
| 58 | register struct mount *mp; |
| 59 | { |
| 60 | |
| 61 | while(mp->mnt_flag & MNT_MLOCK) { |
| 62 | mp->mnt_flag |= MNT_MWAIT; |
| 63 | sleep((caddr_t)mp, PVFS); |
| 64 | } |
| 65 | mp->mnt_flag |= MNT_MLOCK; |
| 66 | return (0); |
| 67 | } |
| 68 | |
| 69 | /* |
| 70 | * Unlock a locked filesystem. |
| 71 | * Panic if filesystem is not locked. |
| 72 | */ |
| 73 | void |
| 74 | vfs_unlock(mp) |
| 75 | register struct mount *mp; |
| 76 | { |
| 77 | |
| 78 | if ((mp->mnt_flag & MNT_MLOCK) == 0) |
| 79 | panic("vfs_unlock: not locked"); |
| 80 | mp->mnt_flag &= ~MNT_MLOCK; |
| 81 | if (mp->mnt_flag & MNT_MWAIT) { |
| 82 | mp->mnt_flag &= ~MNT_MWAIT; |
| 83 | wakeup((caddr_t)mp); |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | /* |
| 88 | * Mark a mount point as busy. |
| 89 | * Used to synchronize access and to delay unmounting. |
| 90 | */ |
| 91 | vfs_busy(mp) |
| 92 | register struct mount *mp; |
| 93 | { |
| 94 | |
| 95 | while(mp->mnt_flag & MNT_MPBUSY) { |
| 96 | mp->mnt_flag |= MNT_MPWANT; |
| 97 | sleep((caddr_t)&mp->mnt_flag, PVFS); |
| 98 | } |
| 99 | if (mp->mnt_flag & MNT_UNMOUNT) |
| 100 | return (1); |
| 101 | mp->mnt_flag |= MNT_MPBUSY; |
| 102 | return (0); |
| 103 | } |
| 104 | |
| 105 | /* |
| 106 | * Free a busy filesystem. |
| 107 | * Panic if filesystem is not busy. |
| 108 | */ |
| 109 | vfs_unbusy(mp) |
| 110 | register struct mount *mp; |
| 111 | { |
| 112 | |
| 113 | if ((mp->mnt_flag & MNT_MPBUSY) == 0) |
| 114 | panic("vfs_unbusy: not busy"); |
| 115 | mp->mnt_flag &= ~MNT_MPBUSY; |
| 116 | if (mp->mnt_flag & MNT_MPWANT) { |
| 117 | mp->mnt_flag &= ~MNT_MPWANT; |
| 118 | wakeup((caddr_t)&mp->mnt_flag); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | /* |
| 123 | * Lookup a mount point by filesystem identifier. |
| 124 | */ |
| 125 | struct mount * |
| 126 | getvfs(fsid) |
| 127 | fsid_t *fsid; |
| 128 | { |
| 129 | register struct mount *mp; |
| 130 | |
| 131 | mp = rootfs; |
| 132 | do { |
| 133 | if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && |
| 134 | mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { |
| 135 | return (mp); |
| 136 | } |
| 137 | mp = mp->mnt_next; |
| 138 | } while (mp != rootfs); |
| 139 | return ((struct mount *)0); |
| 140 | } |
| 141 | |
| 142 | /* |
| 143 | * Set vnode attributes to VNOVAL |
| 144 | */ |
| 145 | void vattr_null(vap) |
| 146 | register struct vattr *vap; |
| 147 | { |
| 148 | |
| 149 | vap->va_type = VNON; |
| 150 | vap->va_size = vap->va_bytes = VNOVAL; |
| 151 | #ifdef _NOQUAD |
| 152 | vap->va_size_rsv = vap->va_bytes_rsv = VNOVAL; |
| 153 | #endif |
| 154 | vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid = |
| 155 | vap->va_fsid = vap->va_fileid = |
| 156 | vap->va_blocksize = vap->va_rdev = |
| 157 | vap->va_atime.tv_sec = vap->va_atime.tv_usec = |
| 158 | vap->va_mtime.tv_sec = vap->va_mtime.tv_usec = |
| 159 | vap->va_ctime.tv_sec = vap->va_ctime.tv_usec = |
| 160 | vap->va_flags = vap->va_gen = VNOVAL; |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * Routines having to do with the management of the vnode table. |
| 165 | */ |
| 166 | struct vnode *vfreeh, **vfreet; |
| 167 | extern struct vnodeops dead_vnodeops, spec_vnodeops; |
| 168 | extern void vclean(); |
| 169 | long numvnodes; |
| 170 | struct vattr va_null; |
| 171 | |
| 172 | /* |
| 173 | * Initialize the vnode structures and initialize each file system type. |
| 174 | */ |
| 175 | vfsinit() |
| 176 | { |
| 177 | struct vfsops **vfsp; |
| 178 | |
| 179 | /* |
| 180 | * Initialize the vnode name cache |
| 181 | */ |
| 182 | nchinit(); |
| 183 | /* |
| 184 | * Initialize each file system type. |
| 185 | */ |
| 186 | vattr_null(&va_null); |
| 187 | for (vfsp = &vfssw[0]; vfsp <= &vfssw[MOUNT_MAXTYPE]; vfsp++) { |
| 188 | if (*vfsp == NULL) |
| 189 | continue; |
| 190 | (*(*vfsp)->vfs_init)(); |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | /* |
| 195 | * Return the next vnode from the free list. |
| 196 | */ |
| 197 | getnewvnode(tag, mp, vops, vpp) |
| 198 | enum vtagtype tag; |
| 199 | struct mount *mp; |
| 200 | struct vnodeops *vops; |
| 201 | struct vnode **vpp; |
| 202 | { |
| 203 | register struct vnode *vp, *vq; |
| 204 | |
| 205 | if (numvnodes < desiredvnodes) { |
| 206 | vp = (struct vnode *)malloc((u_long)sizeof *vp, |
| 207 | M_VNODE, M_WAITOK); |
| 208 | bzero((char *)vp, sizeof *vp); |
| 209 | numvnodes++; |
| 210 | } else { |
| 211 | if ((vp = vfreeh) == NULL) { |
| 212 | tablefull("vnode"); |
| 213 | *vpp = 0; |
| 214 | return (ENFILE); |
| 215 | } |
| 216 | if (vp->v_usecount) |
| 217 | panic("free vnode isn't"); |
| 218 | if (vq = vp->v_freef) |
| 219 | vq->v_freeb = &vfreeh; |
| 220 | else |
| 221 | vfreet = &vfreeh; |
| 222 | vfreeh = vq; |
| 223 | vp->v_freef = NULL; |
| 224 | vp->v_freeb = NULL; |
| 225 | vp->v_lease = NULL; |
| 226 | if (vp->v_type != VBAD) |
| 227 | vgone(vp); |
| 228 | if (vp->v_data) |
| 229 | panic("cleaned vnode isn't"); |
| 230 | vp->v_flag = 0; |
| 231 | vp->v_lastr = 0; |
| 232 | vp->v_socket = 0; |
| 233 | } |
| 234 | vp->v_type = VNON; |
| 235 | cache_purge(vp); |
| 236 | vp->v_tag = tag; |
| 237 | vp->v_op = vops; |
| 238 | insmntque(vp, mp); |
| 239 | VREF(vp); |
| 240 | *vpp = vp; |
| 241 | return (0); |
| 242 | } |
| 243 | |
| 244 | /* |
| 245 | * Move a vnode from one mount queue to another. |
| 246 | */ |
| 247 | insmntque(vp, mp) |
| 248 | register struct vnode *vp; |
| 249 | register struct mount *mp; |
| 250 | { |
| 251 | register struct vnode *vq; |
| 252 | |
| 253 | /* |
| 254 | * Delete from old mount point vnode list, if on one. |
| 255 | */ |
| 256 | if (vp->v_mountb) { |
| 257 | if (vq = vp->v_mountf) |
| 258 | vq->v_mountb = vp->v_mountb; |
| 259 | *vp->v_mountb = vq; |
| 260 | } |
| 261 | /* |
| 262 | * Insert into list of vnodes for the new mount point, if available. |
| 263 | */ |
| 264 | vp->v_mount = mp; |
| 265 | if (mp == NULL) { |
| 266 | vp->v_mountf = NULL; |
| 267 | vp->v_mountb = NULL; |
| 268 | return; |
| 269 | } |
| 270 | if (vq = mp->mnt_mounth) |
| 271 | vq->v_mountb = &vp->v_mountf; |
| 272 | vp->v_mountf = vq; |
| 273 | vp->v_mountb = &mp->mnt_mounth; |
| 274 | mp->mnt_mounth = vp; |
| 275 | } |
| 276 | |
| 277 | /* |
| 278 | * Make sure all write-behind blocks associated |
| 279 | * with mount point are flushed out (from sync). |
| 280 | */ |
| 281 | mntflushbuf(mountp, flags) |
| 282 | struct mount *mountp; |
| 283 | int flags; |
| 284 | { |
| 285 | register struct vnode *vp; |
| 286 | |
| 287 | if ((mountp->mnt_flag & MNT_MPBUSY) == 0) |
| 288 | panic("mntflushbuf: not busy"); |
| 289 | loop: |
| 290 | for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) { |
| 291 | if (VOP_ISLOCKED(vp)) |
| 292 | continue; |
| 293 | if (vget(vp)) |
| 294 | goto loop; |
| 295 | vflushbuf(vp, flags); |
| 296 | vput(vp); |
| 297 | if (vp->v_mount != mountp) |
| 298 | goto loop; |
| 299 | } |
| 300 | } |
| 301 | |
| 302 | /* |
| 303 | * Flush all dirty buffers associated with a vnode. |
| 304 | */ |
| 305 | vflushbuf(vp, flags) |
| 306 | register struct vnode *vp; |
| 307 | int flags; |
| 308 | { |
| 309 | register struct buf *bp; |
| 310 | struct buf *nbp; |
| 311 | int s; |
| 312 | |
| 313 | loop: |
| 314 | s = splbio(); |
| 315 | for (bp = vp->v_dirtyblkhd; bp; bp = nbp) { |
| 316 | nbp = bp->b_blockf; |
| 317 | if ((bp->b_flags & B_BUSY)) |
| 318 | continue; |
| 319 | if ((bp->b_flags & B_DELWRI) == 0) |
| 320 | panic("vflushbuf: not dirty"); |
| 321 | bremfree(bp); |
| 322 | bp->b_flags |= B_BUSY; |
| 323 | splx(s); |
| 324 | /* |
| 325 | * Wait for I/O associated with indirect blocks to complete, |
| 326 | * since there is no way to quickly wait for them below. |
| 327 | * NB: This is really specific to ufs, but is done here |
| 328 | * as it is easier and quicker. |
| 329 | */ |
| 330 | if (bp->b_vp == vp || (flags & B_SYNC) == 0) |
| 331 | (void) bawrite(bp); |
| 332 | else |
| 333 | (void) bwrite(bp); |
| 334 | goto loop; |
| 335 | } |
| 336 | splx(s); |
| 337 | if ((flags & B_SYNC) == 0) |
| 338 | return; |
| 339 | s = splbio(); |
| 340 | while (vp->v_numoutput) { |
| 341 | vp->v_flag |= VBWAIT; |
| 342 | sleep((caddr_t)&vp->v_numoutput, PRIBIO + 1); |
| 343 | } |
| 344 | splx(s); |
| 345 | if (vp->v_dirtyblkhd) { |
| 346 | vprint("vflushbuf: dirty", vp); |
| 347 | goto loop; |
| 348 | } |
| 349 | } |
| 350 | |
| 351 | /* |
| 352 | * Update outstanding I/O count and do wakeup if requested. |
| 353 | */ |
| 354 | vwakeup(bp) |
| 355 | register struct buf *bp; |
| 356 | { |
| 357 | register struct vnode *vp; |
| 358 | |
| 359 | bp->b_dirtyoff = bp->b_dirtyend = 0; |
| 360 | if (vp = bp->b_vp) { |
| 361 | vp->v_numoutput--; |
| 362 | if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { |
| 363 | if (vp->v_numoutput < 0) |
| 364 | panic("vwakeup: neg numoutput"); |
| 365 | vp->v_flag &= ~VBWAIT; |
| 366 | wakeup((caddr_t)&vp->v_numoutput); |
| 367 | } |
| 368 | } |
| 369 | } |
| 370 | |
| 371 | /* |
| 372 | * Invalidate in core blocks belonging to closed or umounted filesystem |
| 373 | * |
| 374 | * Go through the list of vnodes associated with the file system; |
| 375 | * for each vnode invalidate any buffers that it holds. Normally |
| 376 | * this routine is preceeded by a bflush call, so that on a quiescent |
| 377 | * filesystem there will be no dirty buffers when we are done. Binval |
| 378 | * returns the count of dirty buffers when it is finished. |
| 379 | */ |
| 380 | mntinvalbuf(mountp) |
| 381 | struct mount *mountp; |
| 382 | { |
| 383 | register struct vnode *vp; |
| 384 | int dirty = 0; |
| 385 | |
| 386 | if ((mountp->mnt_flag & MNT_MPBUSY) == 0) |
| 387 | panic("mntinvalbuf: not busy"); |
| 388 | loop: |
| 389 | for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) { |
| 390 | if (vget(vp)) |
| 391 | goto loop; |
| 392 | dirty += vinvalbuf(vp, 1); |
| 393 | vput(vp); |
| 394 | if (vp->v_mount != mountp) |
| 395 | goto loop; |
| 396 | } |
| 397 | return (dirty); |
| 398 | } |
| 399 | |
| 400 | /* |
| 401 | * Flush out and invalidate all buffers associated with a vnode. |
| 402 | * Called with the underlying object locked. |
| 403 | */ |
| 404 | vinvalbuf(vp, save) |
| 405 | register struct vnode *vp; |
| 406 | int save; |
| 407 | { |
| 408 | register struct buf *bp; |
| 409 | struct buf *nbp, *blist; |
| 410 | int s, dirty = 0; |
| 411 | |
| 412 | for (;;) { |
| 413 | if (blist = vp->v_dirtyblkhd) |
| 414 | /* void */; |
| 415 | else if (blist = vp->v_cleanblkhd) |
| 416 | /* void */; |
| 417 | else |
| 418 | break; |
| 419 | for (bp = blist; bp; bp = nbp) { |
| 420 | nbp = bp->b_blockf; |
| 421 | s = splbio(); |
| 422 | if (bp->b_flags & B_BUSY) { |
| 423 | bp->b_flags |= B_WANTED; |
| 424 | sleep((caddr_t)bp, PRIBIO + 1); |
| 425 | splx(s); |
| 426 | break; |
| 427 | } |
| 428 | bremfree(bp); |
| 429 | bp->b_flags |= B_BUSY; |
| 430 | splx(s); |
| 431 | if (save && (bp->b_flags & B_DELWRI)) { |
| 432 | dirty++; |
| 433 | (void) VOP_BWRITE(bp); |
| 434 | break; |
| 435 | } |
| 436 | if (bp->b_vp != vp) |
| 437 | reassignbuf(bp, bp->b_vp); |
| 438 | else |
| 439 | bp->b_flags |= B_INVAL; |
| 440 | brelse(bp); |
| 441 | } |
| 442 | } |
| 443 | if (vp->v_dirtyblkhd || vp->v_cleanblkhd) |
| 444 | panic("vinvalbuf: flush failed"); |
| 445 | return (dirty); |
| 446 | } |
| 447 | |
| 448 | /* |
| 449 | * Associate a buffer with a vnode. |
| 450 | */ |
| 451 | bgetvp(vp, bp) |
| 452 | register struct vnode *vp; |
| 453 | register struct buf *bp; |
| 454 | { |
| 455 | register struct vnode *vq; |
| 456 | register struct buf *bq; |
| 457 | |
| 458 | if (bp->b_vp) |
| 459 | panic("bgetvp: not free"); |
| 460 | VHOLD(vp); |
| 461 | bp->b_vp = vp; |
| 462 | if (vp->v_type == VBLK || vp->v_type == VCHR) |
| 463 | bp->b_dev = vp->v_rdev; |
| 464 | else |
| 465 | bp->b_dev = NODEV; |
| 466 | /* |
| 467 | * Insert onto list for new vnode. |
| 468 | */ |
| 469 | if (bq = vp->v_cleanblkhd) |
| 470 | bq->b_blockb = &bp->b_blockf; |
| 471 | bp->b_blockf = bq; |
| 472 | bp->b_blockb = &vp->v_cleanblkhd; |
| 473 | vp->v_cleanblkhd = bp; |
| 474 | } |
| 475 | |
| 476 | /* |
| 477 | * Disassociate a buffer from a vnode. |
| 478 | */ |
| 479 | brelvp(bp) |
| 480 | register struct buf *bp; |
| 481 | { |
| 482 | struct buf *bq; |
| 483 | struct vnode *vp; |
| 484 | |
| 485 | if (bp->b_vp == (struct vnode *) 0) |
| 486 | panic("brelvp: NULL"); |
| 487 | /* |
| 488 | * Delete from old vnode list, if on one. |
| 489 | */ |
| 490 | if (bp->b_blockb) { |
| 491 | if (bq = bp->b_blockf) |
| 492 | bq->b_blockb = bp->b_blockb; |
| 493 | *bp->b_blockb = bq; |
| 494 | bp->b_blockf = NULL; |
| 495 | bp->b_blockb = NULL; |
| 496 | } |
| 497 | vp = bp->b_vp; |
| 498 | bp->b_vp = (struct vnode *) 0; |
| 499 | HOLDRELE(vp); |
| 500 | } |
| 501 | |
| 502 | /* |
| 503 | * Reassign a buffer from one vnode to another. |
| 504 | * Used to assign file specific control information |
| 505 | * (indirect blocks) to the vnode to which they belong. |
| 506 | */ |
| 507 | reassignbuf(bp, newvp) |
| 508 | register struct buf *bp; |
| 509 | register struct vnode *newvp; |
| 510 | { |
| 511 | register struct buf *bq, **listheadp; |
| 512 | |
| 513 | if (newvp == NULL) { |
| 514 | printf("reassignbuf: NULL"); |
| 515 | return; |
| 516 | } |
| 517 | /* |
| 518 | * Delete from old vnode list, if on one. |
| 519 | */ |
| 520 | if (bp->b_blockb) { |
| 521 | if (bq = bp->b_blockf) |
| 522 | bq->b_blockb = bp->b_blockb; |
| 523 | *bp->b_blockb = bq; |
| 524 | } |
| 525 | /* |
| 526 | * If dirty, put on list of dirty buffers; |
| 527 | * otherwise insert onto list of clean buffers. |
| 528 | */ |
| 529 | if (bp->b_flags & B_DELWRI) |
| 530 | listheadp = &newvp->v_dirtyblkhd; |
| 531 | else |
| 532 | listheadp = &newvp->v_cleanblkhd; |
| 533 | if (bq = *listheadp) |
| 534 | bq->b_blockb = &bp->b_blockf; |
| 535 | bp->b_blockf = bq; |
| 536 | bp->b_blockb = listheadp; |
| 537 | *listheadp = bp; |
| 538 | } |
| 539 | |
| 540 | /* |
| 541 | * Create a vnode for a block device. |
| 542 | * Used for root filesystem, argdev, and swap areas. |
| 543 | * Also used for memory file system special devices. |
| 544 | */ |
| 545 | bdevvp(dev, vpp) |
| 546 | dev_t dev; |
| 547 | struct vnode **vpp; |
| 548 | { |
| 549 | register struct vnode *vp; |
| 550 | struct vnode *nvp; |
| 551 | int error; |
| 552 | |
| 553 | if (dev == NODEV) |
| 554 | return (0); |
| 555 | error = getnewvnode(VT_NON, (struct mount *)0, &spec_vnodeops, &nvp); |
| 556 | if (error) { |
| 557 | *vpp = 0; |
| 558 | return (error); |
| 559 | } |
| 560 | vp = nvp; |
| 561 | vp->v_type = VBLK; |
| 562 | if (nvp = checkalias(vp, dev, (struct mount *)0)) { |
| 563 | vput(vp); |
| 564 | vp = nvp; |
| 565 | } |
| 566 | *vpp = vp; |
| 567 | return (0); |
| 568 | } |
| 569 | |
| 570 | /* |
| 571 | * Check to see if the new vnode represents a special device |
| 572 | * for which we already have a vnode (either because of |
| 573 | * bdevvp() or because of a different vnode representing |
| 574 | * the same block device). If such an alias exists, deallocate |
| 575 | * the existing contents and return the aliased vnode. The |
| 576 | * caller is responsible for filling it with its new contents. |
| 577 | */ |
| 578 | struct vnode * |
| 579 | checkalias(nvp, nvp_rdev, mp) |
| 580 | register struct vnode *nvp; |
| 581 | dev_t nvp_rdev; |
| 582 | struct mount *mp; |
| 583 | { |
| 584 | register struct vnode *vp; |
| 585 | struct vnode **vpp; |
| 586 | |
| 587 | if (nvp->v_type != VBLK && nvp->v_type != VCHR) |
| 588 | return (NULLVP); |
| 589 | |
| 590 | vpp = &speclisth[SPECHASH(nvp_rdev)]; |
| 591 | loop: |
| 592 | for (vp = *vpp; vp; vp = vp->v_specnext) { |
| 593 | if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) |
| 594 | continue; |
| 595 | /* |
| 596 | * Alias, but not in use, so flush it out. |
| 597 | */ |
| 598 | if (vp->v_usecount == 0) { |
| 599 | vgone(vp); |
| 600 | goto loop; |
| 601 | } |
| 602 | if (vget(vp)) |
| 603 | goto loop; |
| 604 | break; |
| 605 | } |
| 606 | if (vp == NULL || vp->v_tag != VT_NON) { |
| 607 | MALLOC(nvp->v_specinfo, struct specinfo *, |
| 608 | sizeof(struct specinfo), M_VNODE, M_WAITOK); |
| 609 | nvp->v_rdev = nvp_rdev; |
| 610 | nvp->v_hashchain = vpp; |
| 611 | nvp->v_specnext = *vpp; |
| 612 | nvp->v_specflags = 0; |
| 613 | *vpp = nvp; |
| 614 | if (vp != NULL) { |
| 615 | nvp->v_flag |= VALIASED; |
| 616 | vp->v_flag |= VALIASED; |
| 617 | vput(vp); |
| 618 | } |
| 619 | return (NULLVP); |
| 620 | } |
| 621 | VOP_UNLOCK(vp); |
| 622 | vclean(vp, 0); |
| 623 | vp->v_op = nvp->v_op; |
| 624 | vp->v_tag = nvp->v_tag; |
| 625 | nvp->v_type = VNON; |
| 626 | insmntque(vp, mp); |
| 627 | return (vp); |
| 628 | } |
| 629 | |
| 630 | /* |
| 631 | * Grab a particular vnode from the free list, increment its |
| 632 | * reference count and lock it. The vnode lock bit is set the |
| 633 | * vnode is being eliminated in vgone. The process is awakened |
| 634 | * when the transition is completed, and an error returned to |
| 635 | * indicate that the vnode is no longer usable (possibly having |
| 636 | * been changed to a new file system type). |
| 637 | */ |
| 638 | vget(vp) |
| 639 | register struct vnode *vp; |
| 640 | { |
| 641 | register struct vnode *vq; |
| 642 | |
| 643 | if (vp->v_flag & VXLOCK) { |
| 644 | vp->v_flag |= VXWANT; |
| 645 | sleep((caddr_t)vp, PINOD); |
| 646 | return (1); |
| 647 | } |
| 648 | if (vp->v_usecount == 0) { |
| 649 | if (vq = vp->v_freef) |
| 650 | vq->v_freeb = vp->v_freeb; |
| 651 | else |
| 652 | vfreet = vp->v_freeb; |
| 653 | *vp->v_freeb = vq; |
| 654 | vp->v_freef = NULL; |
| 655 | vp->v_freeb = NULL; |
| 656 | } |
| 657 | VREF(vp); |
| 658 | VOP_LOCK(vp); |
| 659 | return (0); |
| 660 | } |
| 661 | |
| 662 | int bug_refs = 0; |
| 663 | |
| 664 | /* |
| 665 | * Vnode reference, just increment the count |
| 666 | */ |
| 667 | void vref(vp) |
| 668 | struct vnode *vp; |
| 669 | { |
| 670 | |
| 671 | vp->v_usecount++; |
| 672 | if (vp->v_type != VBLK && curproc) |
| 673 | curproc->p_spare[0]++; |
| 674 | if (bug_refs) |
| 675 | vprint("vref: "); |
| 676 | } |
| 677 | |
| 678 | /* |
| 679 | * vput(), just unlock and vrele() |
| 680 | */ |
| 681 | void vput(vp) |
| 682 | register struct vnode *vp; |
| 683 | { |
| 684 | |
| 685 | VOP_UNLOCK(vp); |
| 686 | vrele(vp); |
| 687 | } |
| 688 | |
| 689 | /* |
| 690 | * Vnode release. |
| 691 | * If count drops to zero, call inactive routine and return to freelist. |
| 692 | */ |
| 693 | void vrele(vp) |
| 694 | register struct vnode *vp; |
| 695 | { |
| 696 | struct proc *p = curproc; /* XXX */ |
| 697 | |
| 698 | #ifdef DIAGNOSTIC |
| 699 | if (vp == NULL) |
| 700 | panic("vrele: null vp"); |
| 701 | #endif |
| 702 | vp->v_usecount--; |
| 703 | if (vp->v_type != VBLK && curproc) |
| 704 | curproc->p_spare[0]--; |
| 705 | if (bug_refs) |
| 706 | vprint("vref: "); |
| 707 | if (vp->v_usecount > 0) |
| 708 | return; |
| 709 | #ifdef DIAGNOSTIC |
| 710 | if (vp->v_usecount != 0 || vp->v_writecount != 0) { |
| 711 | vprint("vrele: bad ref count", vp); |
| 712 | panic("vrele: ref cnt"); |
| 713 | } |
| 714 | #endif |
| 715 | if (vfreeh == NULLVP) { |
| 716 | /* |
| 717 | * insert into empty list |
| 718 | */ |
| 719 | vfreeh = vp; |
| 720 | vp->v_freeb = &vfreeh; |
| 721 | } else { |
| 722 | /* |
| 723 | * insert at tail of list |
| 724 | */ |
| 725 | *vfreet = vp; |
| 726 | vp->v_freeb = vfreet; |
| 727 | } |
| 728 | vp->v_freef = NULL; |
| 729 | vfreet = &vp->v_freef; |
| 730 | VOP_INACTIVE(vp, p); |
| 731 | } |
| 732 | |
| 733 | /* |
| 734 | * Page or buffer structure gets a reference. |
| 735 | */ |
| 736 | vhold(vp) |
| 737 | register struct vnode *vp; |
| 738 | { |
| 739 | |
| 740 | vp->v_holdcnt++; |
| 741 | } |
| 742 | |
| 743 | /* |
| 744 | * Page or buffer structure frees a reference. |
| 745 | */ |
| 746 | holdrele(vp) |
| 747 | register struct vnode *vp; |
| 748 | { |
| 749 | |
| 750 | if (vp->v_holdcnt <= 0) |
| 751 | panic("holdrele: holdcnt"); |
| 752 | vp->v_holdcnt--; |
| 753 | } |
| 754 | |
| 755 | /* |
| 756 | * Remove any vnodes in the vnode table belonging to mount point mp. |
| 757 | * |
| 758 | * If MNT_NOFORCE is specified, there should not be any active ones, |
| 759 | * return error if any are found (nb: this is a user error, not a |
| 760 | * system error). If MNT_FORCE is specified, detach any active vnodes |
| 761 | * that are found. |
| 762 | */ |
| 763 | int busyprt = 0; /* patch to print out busy vnodes */ |
| 764 | |
| 765 | vflush(mp, skipvp, flags) |
| 766 | struct mount *mp; |
| 767 | struct vnode *skipvp; |
| 768 | int flags; |
| 769 | { |
| 770 | register struct vnode *vp, *nvp; |
| 771 | int busy = 0; |
| 772 | |
| 773 | if ((mp->mnt_flag & MNT_MPBUSY) == 0) |
| 774 | panic("vflush: not busy"); |
| 775 | loop: |
| 776 | for (vp = mp->mnt_mounth; vp; vp = nvp) { |
| 777 | if (vp->v_mount != mp) |
| 778 | goto loop; |
| 779 | nvp = vp->v_mountf; |
| 780 | /* |
| 781 | * Skip over a selected vnode. |
| 782 | */ |
| 783 | if (vp == skipvp) |
| 784 | continue; |
| 785 | /* |
| 786 | * Skip over a vnodes marked VSYSTEM. |
| 787 | */ |
| 788 | if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) |
| 789 | continue; |
| 790 | /* |
| 791 | * With v_usecount == 0, all we need to do is clear |
| 792 | * out the vnode data structures and we are done. |
| 793 | */ |
| 794 | if (vp->v_usecount == 0) { |
| 795 | vgone(vp); |
| 796 | continue; |
| 797 | } |
| 798 | /* |
| 799 | * For block or character devices, revert to an |
| 800 | * anonymous device. For all other files, just kill them. |
| 801 | */ |
| 802 | if (flags & FORCECLOSE) { |
| 803 | if (vp->v_type != VBLK && vp->v_type != VCHR) { |
| 804 | vgone(vp); |
| 805 | } else { |
| 806 | vclean(vp, 0); |
| 807 | vp->v_op = &spec_vnodeops; |
| 808 | insmntque(vp, (struct mount *)0); |
| 809 | } |
| 810 | continue; |
| 811 | } |
| 812 | if (busyprt) |
| 813 | vprint("vflush: busy vnode", vp); |
| 814 | busy++; |
| 815 | } |
| 816 | if (busy) |
| 817 | return (EBUSY); |
| 818 | return (0); |
| 819 | } |
| 820 | |
| 821 | /* |
| 822 | * Disassociate the underlying file system from a vnode. |
| 823 | */ |
| 824 | void vclean(vp, flags) |
| 825 | register struct vnode *vp; |
| 826 | int flags; |
| 827 | { |
| 828 | struct vnodeops *origops; |
| 829 | int active; |
| 830 | struct proc *p = curproc; /* XXX */ |
| 831 | |
| 832 | /* |
| 833 | * Check to see if the vnode is in use. |
| 834 | * If so we have to reference it before we clean it out |
| 835 | * so that its count cannot fall to zero and generate a |
| 836 | * race against ourselves to recycle it. |
| 837 | */ |
| 838 | if (active = vp->v_usecount) |
| 839 | VREF(vp); |
| 840 | /* |
| 841 | * Prevent the vnode from being recycled or |
| 842 | * brought into use while we clean it out. |
| 843 | */ |
| 844 | if (vp->v_flag & VXLOCK) |
| 845 | panic("vclean: deadlock"); |
| 846 | vp->v_flag |= VXLOCK; |
| 847 | /* |
| 848 | * Even if the count is zero, the VOP_INACTIVE routine may still |
| 849 | * have the object locked while it cleans it out. The VOP_LOCK |
| 850 | * ensures that the VOP_INACTIVE routine is done with its work. |
| 851 | * For active vnodes, it ensures that no other activity can |
| 852 | * occur while the buffer list is being cleaned out. |
| 853 | */ |
| 854 | VOP_LOCK(vp); |
| 855 | if (flags & DOCLOSE) |
| 856 | vinvalbuf(vp, 1); |
| 857 | /* |
| 858 | * Prevent any further operations on the vnode from |
| 859 | * being passed through to the old file system. |
| 860 | */ |
| 861 | origops = vp->v_op; |
| 862 | vp->v_op = &dead_vnodeops; |
| 863 | vp->v_tag = VT_NON; |
| 864 | /* |
| 865 | * If purging an active vnode, it must be unlocked, closed, |
| 866 | * and deactivated before being reclaimed. |
| 867 | */ |
| 868 | (*(origops->vop_unlock))(vp); |
| 869 | if (active) { |
| 870 | if (flags & DOCLOSE) |
| 871 | (*(origops->vop_close))(vp, IO_NDELAY, NOCRED, p); |
| 872 | (*(origops->vop_inactive))(vp, p); |
| 873 | } |
| 874 | /* |
| 875 | * Reclaim the vnode. |
| 876 | */ |
| 877 | if ((*(origops->vop_reclaim))(vp)) |
| 878 | panic("vclean: cannot reclaim"); |
| 879 | if (active) |
| 880 | vrele(vp); |
| 881 | /* |
| 882 | * Done with purge, notify sleepers in vget of the grim news. |
| 883 | */ |
| 884 | vp->v_flag &= ~VXLOCK; |
| 885 | if (vp->v_flag & VXWANT) { |
| 886 | vp->v_flag &= ~VXWANT; |
| 887 | wakeup((caddr_t)vp); |
| 888 | } |
| 889 | } |
| 890 | |
| 891 | /* |
| 892 | * Eliminate all activity associated with the requested vnode |
| 893 | * and with all vnodes aliased to the requested vnode. |
| 894 | */ |
| 895 | void vgoneall(vp) |
| 896 | register struct vnode *vp; |
| 897 | { |
| 898 | register struct vnode *vq; |
| 899 | |
| 900 | if (vp->v_flag & VALIASED) { |
| 901 | /* |
| 902 | * If a vgone (or vclean) is already in progress, |
| 903 | * wait until it is done and return. |
| 904 | */ |
| 905 | if (vp->v_flag & VXLOCK) { |
| 906 | vp->v_flag |= VXWANT; |
| 907 | sleep((caddr_t)vp, PINOD); |
| 908 | return; |
| 909 | } |
| 910 | /* |
| 911 | * Ensure that vp will not be vgone'd while we |
| 912 | * are eliminating its aliases. |
| 913 | */ |
| 914 | vp->v_flag |= VXLOCK; |
| 915 | while (vp->v_flag & VALIASED) { |
| 916 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 917 | if (vq->v_rdev != vp->v_rdev || |
| 918 | vq->v_type != vp->v_type || vp == vq) |
| 919 | continue; |
| 920 | vgone(vq); |
| 921 | break; |
| 922 | } |
| 923 | } |
| 924 | /* |
| 925 | * Remove the lock so that vgone below will |
| 926 | * really eliminate the vnode after which time |
| 927 | * vgone will awaken any sleepers. |
| 928 | */ |
| 929 | vp->v_flag &= ~VXLOCK; |
| 930 | } |
| 931 | vgone(vp); |
| 932 | } |
| 933 | |
| 934 | /* |
| 935 | * Eliminate all activity associated with a vnode |
| 936 | * in preparation for reuse. |
| 937 | */ |
| 938 | void vgone(vp) |
| 939 | register struct vnode *vp; |
| 940 | { |
| 941 | register struct vnode *vq; |
| 942 | struct vnode *vx; |
| 943 | |
| 944 | /* |
| 945 | * If a vgone (or vclean) is already in progress, |
| 946 | * wait until it is done and return. |
| 947 | */ |
| 948 | if (vp->v_flag & VXLOCK) { |
| 949 | vp->v_flag |= VXWANT; |
| 950 | sleep((caddr_t)vp, PINOD); |
| 951 | return; |
| 952 | } |
| 953 | /* |
| 954 | * Clean out the filesystem specific data. |
| 955 | */ |
| 956 | vclean(vp, DOCLOSE); |
| 957 | /* |
| 958 | * Delete from old mount point vnode list, if on one. |
| 959 | */ |
| 960 | if (vp->v_mountb) { |
| 961 | if (vq = vp->v_mountf) |
| 962 | vq->v_mountb = vp->v_mountb; |
| 963 | *vp->v_mountb = vq; |
| 964 | vp->v_mountf = NULL; |
| 965 | vp->v_mountb = NULL; |
| 966 | vp->v_mount = NULL; |
| 967 | } |
| 968 | /* |
| 969 | * If special device, remove it from special device alias list. |
| 970 | */ |
| 971 | if (vp->v_type == VBLK || vp->v_type == VCHR) { |
| 972 | if (*vp->v_hashchain == vp) { |
| 973 | *vp->v_hashchain = vp->v_specnext; |
| 974 | } else { |
| 975 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 976 | if (vq->v_specnext != vp) |
| 977 | continue; |
| 978 | vq->v_specnext = vp->v_specnext; |
| 979 | break; |
| 980 | } |
| 981 | if (vq == NULL) |
| 982 | panic("missing bdev"); |
| 983 | } |
| 984 | if (vp->v_flag & VALIASED) { |
| 985 | vx = NULL; |
| 986 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 987 | if (vq->v_rdev != vp->v_rdev || |
| 988 | vq->v_type != vp->v_type) |
| 989 | continue; |
| 990 | if (vx) |
| 991 | break; |
| 992 | vx = vq; |
| 993 | } |
| 994 | if (vx == NULL) |
| 995 | panic("missing alias"); |
| 996 | if (vq == NULL) |
| 997 | vx->v_flag &= ~VALIASED; |
| 998 | vp->v_flag &= ~VALIASED; |
| 999 | } |
| 1000 | FREE(vp->v_specinfo, M_VNODE); |
| 1001 | vp->v_specinfo = NULL; |
| 1002 | } |
| 1003 | /* |
| 1004 | * If it is on the freelist, move it to the head of the list. |
| 1005 | */ |
| 1006 | if (vp->v_freeb) { |
| 1007 | if (vq = vp->v_freef) |
| 1008 | vq->v_freeb = vp->v_freeb; |
| 1009 | else |
| 1010 | vfreet = vp->v_freeb; |
| 1011 | *vp->v_freeb = vq; |
| 1012 | vp->v_freef = vfreeh; |
| 1013 | vp->v_freeb = &vfreeh; |
| 1014 | vfreeh->v_freeb = &vp->v_freef; |
| 1015 | vfreeh = vp; |
| 1016 | } |
| 1017 | vp->v_type = VBAD; |
| 1018 | } |
| 1019 | |
| 1020 | /* |
| 1021 | * Lookup a vnode by device number. |
| 1022 | */ |
| 1023 | vfinddev(dev, type, vpp) |
| 1024 | dev_t dev; |
| 1025 | enum vtype type; |
| 1026 | struct vnode **vpp; |
| 1027 | { |
| 1028 | register struct vnode *vp; |
| 1029 | |
| 1030 | for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { |
| 1031 | if (dev != vp->v_rdev || type != vp->v_type) |
| 1032 | continue; |
| 1033 | *vpp = vp; |
| 1034 | return (0); |
| 1035 | } |
| 1036 | return (1); |
| 1037 | } |
| 1038 | |
| 1039 | /* |
| 1040 | * Calculate the total number of references to a special device. |
| 1041 | */ |
| 1042 | vcount(vp) |
| 1043 | register struct vnode *vp; |
| 1044 | { |
| 1045 | register struct vnode *vq; |
| 1046 | int count; |
| 1047 | |
| 1048 | if ((vp->v_flag & VALIASED) == 0) |
| 1049 | return (vp->v_usecount); |
| 1050 | loop: |
| 1051 | for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 1052 | if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) |
| 1053 | continue; |
| 1054 | /* |
| 1055 | * Alias, but not in use, so flush it out. |
| 1056 | */ |
| 1057 | if (vq->v_usecount == 0) { |
| 1058 | vgone(vq); |
| 1059 | goto loop; |
| 1060 | } |
| 1061 | count += vq->v_usecount; |
| 1062 | } |
| 1063 | return (count); |
| 1064 | } |
| 1065 | |
| 1066 | /* |
| 1067 | * Print out a description of a vnode. |
| 1068 | */ |
| 1069 | static char *typename[] = |
| 1070 | { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; |
| 1071 | |
| 1072 | vprint(label, vp) |
| 1073 | char *label; |
| 1074 | register struct vnode *vp; |
| 1075 | { |
| 1076 | char buf[64]; |
| 1077 | |
| 1078 | if (label != NULL) |
| 1079 | printf("%s: ", label); |
| 1080 | printf("type %s, usecount %d, writecount %d, refcount %d,", |
| 1081 | typename[vp->v_type], vp->v_usecount, vp->v_writecount, |
| 1082 | vp->v_holdcnt); |
| 1083 | buf[0] = '\0'; |
| 1084 | if (vp->v_flag & VROOT) |
| 1085 | strcat(buf, "|VROOT"); |
| 1086 | if (vp->v_flag & VTEXT) |
| 1087 | strcat(buf, "|VTEXT"); |
| 1088 | if (vp->v_flag & VSYSTEM) |
| 1089 | strcat(buf, "|VSYSTEM"); |
| 1090 | if (vp->v_flag & VXLOCK) |
| 1091 | strcat(buf, "|VXLOCK"); |
| 1092 | if (vp->v_flag & VXWANT) |
| 1093 | strcat(buf, "|VXWANT"); |
| 1094 | if (vp->v_flag & VBWAIT) |
| 1095 | strcat(buf, "|VBWAIT"); |
| 1096 | if (vp->v_flag & VALIASED) |
| 1097 | strcat(buf, "|VALIASED"); |
| 1098 | if (buf[0] != '\0') |
| 1099 | printf(" flags (%s)", &buf[1]); |
| 1100 | printf("\n\t"); |
| 1101 | VOP_PRINT(vp); |
| 1102 | } |
| 1103 | |
| 1104 | #ifdef DEBUG |
| 1105 | /* |
| 1106 | * List all of the locked vnodes in the system. |
| 1107 | * Called when debugging the kernel. |
| 1108 | */ |
| 1109 | printlockedvnodes() |
| 1110 | { |
| 1111 | register struct mount *mp; |
| 1112 | register struct vnode *vp; |
| 1113 | |
| 1114 | printf("Locked vnodes\n"); |
| 1115 | mp = rootfs; |
| 1116 | do { |
| 1117 | for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) |
| 1118 | if (VOP_ISLOCKED(vp)) |
| 1119 | vprint((char *)0, vp); |
| 1120 | mp = mp->mnt_next; |
| 1121 | } while (mp != rootfs); |
| 1122 | } |
| 1123 | #endif |
| 1124 | |
| 1125 | int kinfo_vdebug = 1; |
| 1126 | int kinfo_vgetfailed; |
| 1127 | #define KINFO_VNODESLOP 10 |
| 1128 | /* |
| 1129 | * Dump vnode list (via kinfo). |
| 1130 | * Copyout address of vnode followed by vnode. |
| 1131 | */ |
| 1132 | /* ARGSUSED */ |
| 1133 | kinfo_vnode(op, where, acopysize, arg, aneeded) |
| 1134 | int op; |
| 1135 | char *where; |
| 1136 | int *acopysize, arg, *aneeded; |
| 1137 | { |
| 1138 | register struct mount *mp = rootfs; |
| 1139 | struct mount *omp; |
| 1140 | struct vnode *vp; |
| 1141 | register char *bp = where, *savebp; |
| 1142 | char *ewhere = where + *acopysize; |
| 1143 | int error; |
| 1144 | |
| 1145 | #define VPTRSZ sizeof (struct vnode *) |
| 1146 | #define VNODESZ sizeof (struct vnode) |
| 1147 | if (where == NULL) { |
| 1148 | *aneeded = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); |
| 1149 | return (0); |
| 1150 | } |
| 1151 | |
| 1152 | do { |
| 1153 | if (vfs_busy(mp)) { |
| 1154 | mp = mp->mnt_next; |
| 1155 | continue; |
| 1156 | } |
| 1157 | savebp = bp; |
| 1158 | again: |
| 1159 | for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) { |
| 1160 | /* |
| 1161 | * Check that the vp is still associated with |
| 1162 | * this filesystem. RACE: could have been |
| 1163 | * recycled onto the same filesystem. |
| 1164 | */ |
| 1165 | if (vp->v_mount != mp) { |
| 1166 | if (kinfo_vdebug) |
| 1167 | printf("kinfo: vp changed\n"); |
| 1168 | bp = savebp; |
| 1169 | goto again; |
| 1170 | } |
| 1171 | if ((bp + VPTRSZ + VNODESZ <= ewhere) && |
| 1172 | ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || |
| 1173 | (error = copyout((caddr_t)vp, bp + VPTRSZ, |
| 1174 | VNODESZ)))) |
| 1175 | return (error); |
| 1176 | bp += VPTRSZ + VNODESZ; |
| 1177 | } |
| 1178 | omp = mp; |
| 1179 | mp = mp->mnt_next; |
| 1180 | vfs_unbusy(omp); |
| 1181 | } while (mp != rootfs); |
| 1182 | |
| 1183 | *aneeded = bp - where; |
| 1184 | if (bp > ewhere) |
| 1185 | *acopysize = ewhere - where; |
| 1186 | else |
| 1187 | *acopysize = bp - where; |
| 1188 | return (0); |
| 1189 | } |