| 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 8.25 (Berkeley) %G% |
| 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 ((*vfsp->vfc_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 | int newnodes = 0; |
| 289 | int printcnt = 0; |
| 290 | |
| 291 | /* |
| 292 | * Return the next vnode from the free list. |
| 293 | */ |
| 294 | int |
| 295 | getnewvnode(tag, mp, vops, vpp) |
| 296 | enum vtagtype tag; |
| 297 | struct mount *mp; |
| 298 | int (**vops)(); |
| 299 | struct vnode **vpp; |
| 300 | { |
| 301 | struct proc *p = curproc; /* XXX */ |
| 302 | struct vnode *vp; |
| 303 | int s; |
| 304 | int cnt; |
| 305 | |
| 306 | top: |
| 307 | simple_lock(&vnode_free_list_slock); |
| 308 | newnodes++; |
| 309 | if ((vnode_free_list.tqh_first == NULL && |
| 310 | numvnodes < 2 * desiredvnodes) || |
| 311 | numvnodes < desiredvnodes) { |
| 312 | simple_unlock(&vnode_free_list_slock); |
| 313 | vp = (struct vnode *)malloc((u_long)sizeof *vp, |
| 314 | M_VNODE, M_WAITOK); |
| 315 | bzero((char *)vp, sizeof *vp); |
| 316 | vp->v_freelist.tqe_next = (struct vnode *)0xdeadf; |
| 317 | vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; |
| 318 | vp->v_mntvnodes.le_next = (struct vnode *)0xdeadf; |
| 319 | vp->v_mntvnodes.le_prev = (struct vnode **)0xdeadb; |
| 320 | numvnodes++; |
| 321 | vp->v_spare[0] = numvnodes; |
| 322 | } else { |
| 323 | for (vp = vnode_free_list.tqh_first; |
| 324 | vp != NULLVP; vp = vp->v_freelist.tqe_next) { |
| 325 | if (simple_lock_try(&vp->v_interlock)) |
| 326 | break; |
| 327 | } |
| 328 | /* |
| 329 | * Unless this is a bad time of the month, at most |
| 330 | * the first NCPUS items on the free list are |
| 331 | * locked, so this is close enough to being empty. |
| 332 | */ |
| 333 | if (vp == NULLVP) { |
| 334 | simple_unlock(&vnode_free_list_slock); |
| 335 | tablefull("vnode"); |
| 336 | *vpp = 0; |
| 337 | return (ENFILE); |
| 338 | } |
| 339 | if (vp->v_usecount) |
| 340 | panic("free vnode isn't"); |
| 341 | if (vp->v_freelist.tqe_next == (struct vnode *)0xdeadf || |
| 342 | vp->v_freelist.tqe_prev == (struct vnode **)0xdeadb) |
| 343 | panic("getnewvnode: not on queue"); |
| 344 | TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); |
| 345 | vp->v_freelist.tqe_next = (struct vnode *)0xdeadf; |
| 346 | /* see comment on why 0xdeadb is set at end of vgone (below) */ |
| 347 | vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; |
| 348 | simple_unlock(&vnode_free_list_slock); |
| 349 | vp->v_lease = NULL; |
| 350 | if (vp->v_type != VBAD) |
| 351 | vgonel(vp, p); |
| 352 | else |
| 353 | simple_unlock(&vp->v_interlock); |
| 354 | #ifdef DIAGNOSTIC |
| 355 | if (vp->v_data) |
| 356 | panic("cleaned vnode isn't"); |
| 357 | s = splbio(); |
| 358 | if (vp->v_numoutput) |
| 359 | panic("Clean vnode has pending I/O's"); |
| 360 | splx(s); |
| 361 | #endif |
| 362 | vp->v_flag = 0; |
| 363 | vp->v_lastr = 0; |
| 364 | vp->v_ralen = 0; |
| 365 | vp->v_maxra = 0; |
| 366 | vp->v_lastw = 0; |
| 367 | vp->v_lasta = 0; |
| 368 | vp->v_cstart = 0; |
| 369 | vp->v_clen = 0; |
| 370 | vp->v_socket = 0; |
| 371 | } |
| 372 | vp->v_type = VNON; |
| 373 | cache_purge(vp); |
| 374 | vp->v_tag = tag; |
| 375 | vp->v_op = vops; |
| 376 | insmntque(vp, mp); |
| 377 | *vpp = vp; |
| 378 | vp->v_usecount = 1; |
| 379 | vp->v_data = 0; |
| 380 | if (printcnt-- > 0) vprint("getnewvnode got", vp); |
| 381 | return (0); |
| 382 | } |
| 383 | |
| 384 | /* |
| 385 | * Move a vnode from one mount queue to another. |
| 386 | */ |
| 387 | void |
| 388 | insmntque(vp, mp) |
| 389 | struct vnode *vp; |
| 390 | struct mount *mp; |
| 391 | { |
| 392 | |
| 393 | simple_lock(&mntvnode_slock); |
| 394 | /* |
| 395 | * Delete from old mount point vnode list, if on one. |
| 396 | */ |
| 397 | if (vp->v_mount != NULL) { |
| 398 | if (vp->v_mntvnodes.le_next == (struct vnode *)0xdeadf || |
| 399 | vp->v_mntvnodes.le_prev == (struct vnode **)0xdeadb) |
| 400 | panic("insmntque: not on queue"); |
| 401 | LIST_REMOVE(vp, v_mntvnodes); |
| 402 | vp->v_mntvnodes.le_next = (struct vnode *)0xdeadf; |
| 403 | vp->v_mntvnodes.le_prev = (struct vnode **)0xdeadb; |
| 404 | } |
| 405 | /* |
| 406 | * Insert into list of vnodes for the new mount point, if available. |
| 407 | */ |
| 408 | if ((vp->v_mount = mp) != NULL) |
| 409 | LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); |
| 410 | simple_unlock(&mntvnode_slock); |
| 411 | } |
| 412 | |
| 413 | /* |
| 414 | * Update outstanding I/O count and do wakeup if requested. |
| 415 | */ |
| 416 | void |
| 417 | vwakeup(bp) |
| 418 | register struct buf *bp; |
| 419 | { |
| 420 | register struct vnode *vp; |
| 421 | |
| 422 | bp->b_flags &= ~B_WRITEINPROG; |
| 423 | if (vp = bp->b_vp) { |
| 424 | if (--vp->v_numoutput < 0) |
| 425 | panic("vwakeup: neg numoutput"); |
| 426 | if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { |
| 427 | if (vp->v_numoutput < 0) |
| 428 | panic("vwakeup: neg numoutput 2"); |
| 429 | vp->v_flag &= ~VBWAIT; |
| 430 | wakeup((caddr_t)&vp->v_numoutput); |
| 431 | } |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | /* |
| 436 | * Flush out and invalidate all buffers associated with a vnode. |
| 437 | * Called with the underlying object locked. |
| 438 | */ |
| 439 | int |
| 440 | vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) |
| 441 | register struct vnode *vp; |
| 442 | int flags; |
| 443 | struct ucred *cred; |
| 444 | struct proc *p; |
| 445 | int slpflag, slptimeo; |
| 446 | { |
| 447 | register struct buf *bp; |
| 448 | struct buf *nbp, *blist; |
| 449 | int s, error; |
| 450 | |
| 451 | if (flags & V_SAVE) { |
| 452 | if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) |
| 453 | return (error); |
| 454 | if (vp->v_dirtyblkhd.lh_first != NULL) |
| 455 | panic("vinvalbuf: dirty bufs"); |
| 456 | } |
| 457 | for (;;) { |
| 458 | if ((blist = vp->v_cleanblkhd.lh_first) && flags & V_SAVEMETA) |
| 459 | while (blist && blist->b_lblkno < 0) |
| 460 | blist = blist->b_vnbufs.le_next; |
| 461 | if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && |
| 462 | (flags & V_SAVEMETA)) |
| 463 | while (blist && blist->b_lblkno < 0) |
| 464 | blist = blist->b_vnbufs.le_next; |
| 465 | if (!blist) |
| 466 | break; |
| 467 | |
| 468 | for (bp = blist; bp; bp = nbp) { |
| 469 | nbp = bp->b_vnbufs.le_next; |
| 470 | if (flags & V_SAVEMETA && bp->b_lblkno < 0) |
| 471 | continue; |
| 472 | s = splbio(); |
| 473 | if (bp->b_flags & B_BUSY) { |
| 474 | bp->b_flags |= B_WANTED; |
| 475 | error = tsleep((caddr_t)bp, |
| 476 | slpflag | (PRIBIO + 1), "vinvalbuf", |
| 477 | slptimeo); |
| 478 | splx(s); |
| 479 | if (error) |
| 480 | return (error); |
| 481 | break; |
| 482 | } |
| 483 | bremfree(bp); |
| 484 | bp->b_flags |= B_BUSY; |
| 485 | splx(s); |
| 486 | /* |
| 487 | * XXX Since there are no node locks for NFS, I believe |
| 488 | * there is a slight chance that a delayed write will |
| 489 | * occur while sleeping just above, so check for it. |
| 490 | */ |
| 491 | if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { |
| 492 | (void) VOP_BWRITE(bp); |
| 493 | break; |
| 494 | } |
| 495 | bp->b_flags |= B_INVAL; |
| 496 | brelse(bp); |
| 497 | } |
| 498 | } |
| 499 | if (!(flags & V_SAVEMETA) && |
| 500 | (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first)) |
| 501 | panic("vinvalbuf: flush failed"); |
| 502 | return (0); |
| 503 | } |
| 504 | |
| 505 | /* |
| 506 | * Associate a buffer with a vnode. |
| 507 | */ |
| 508 | void |
| 509 | bgetvp(vp, bp) |
| 510 | register struct vnode *vp; |
| 511 | register struct buf *bp; |
| 512 | { |
| 513 | |
| 514 | if (bp->b_vp) |
| 515 | panic("bgetvp: not free"); |
| 516 | VHOLD(vp); |
| 517 | bp->b_vp = vp; |
| 518 | if (vp->v_type == VBLK || vp->v_type == VCHR) |
| 519 | bp->b_dev = vp->v_rdev; |
| 520 | else |
| 521 | bp->b_dev = NODEV; |
| 522 | /* |
| 523 | * Insert onto list for new vnode. |
| 524 | */ |
| 525 | bufinsvn(bp, &vp->v_cleanblkhd); |
| 526 | } |
| 527 | |
| 528 | /* |
| 529 | * Disassociate a buffer from a vnode. |
| 530 | */ |
| 531 | void |
| 532 | brelvp(bp) |
| 533 | register struct buf *bp; |
| 534 | { |
| 535 | struct vnode *vp; |
| 536 | |
| 537 | if (bp->b_vp == (struct vnode *) 0) |
| 538 | panic("brelvp: NULL"); |
| 539 | /* |
| 540 | * Delete from old vnode list, if on one. |
| 541 | */ |
| 542 | if (bp->b_vnbufs.le_next != NOLIST) |
| 543 | bufremvn(bp); |
| 544 | vp = bp->b_vp; |
| 545 | bp->b_vp = (struct vnode *) 0; |
| 546 | HOLDRELE(vp); |
| 547 | } |
| 548 | |
| 549 | /* |
| 550 | * Reassign a buffer from one vnode to another. |
| 551 | * Used to assign file specific control information |
| 552 | * (indirect blocks) to the vnode to which they belong. |
| 553 | */ |
| 554 | void |
| 555 | reassignbuf(bp, newvp) |
| 556 | register struct buf *bp; |
| 557 | register struct vnode *newvp; |
| 558 | { |
| 559 | register struct buflists *listheadp; |
| 560 | |
| 561 | if (newvp == NULL) { |
| 562 | printf("reassignbuf: NULL"); |
| 563 | return; |
| 564 | } |
| 565 | /* |
| 566 | * Delete from old vnode list, if on one. |
| 567 | */ |
| 568 | if (bp->b_vnbufs.le_next != NOLIST) |
| 569 | bufremvn(bp); |
| 570 | /* |
| 571 | * If dirty, put on list of dirty buffers; |
| 572 | * otherwise insert onto list of clean buffers. |
| 573 | */ |
| 574 | if (bp->b_flags & B_DELWRI) |
| 575 | listheadp = &newvp->v_dirtyblkhd; |
| 576 | else |
| 577 | listheadp = &newvp->v_cleanblkhd; |
| 578 | bufinsvn(bp, listheadp); |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * Create a vnode for a block device. |
| 583 | * Used for root filesystem, argdev, and swap areas. |
| 584 | * Also used for memory file system special devices. |
| 585 | */ |
| 586 | int |
| 587 | bdevvp(dev, vpp) |
| 588 | dev_t dev; |
| 589 | struct vnode **vpp; |
| 590 | { |
| 591 | register struct vnode *vp; |
| 592 | struct vnode *nvp; |
| 593 | int error; |
| 594 | |
| 595 | if (dev == NODEV) { |
| 596 | *vpp = NULLVP; |
| 597 | return (ENODEV); |
| 598 | } |
| 599 | error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); |
| 600 | if (error) { |
| 601 | *vpp = NULLVP; |
| 602 | return (error); |
| 603 | } |
| 604 | vp = nvp; |
| 605 | vp->v_type = VBLK; |
| 606 | if (nvp = checkalias(vp, dev, (struct mount *)0)) { |
| 607 | vput(vp); |
| 608 | vp = nvp; |
| 609 | } |
| 610 | *vpp = vp; |
| 611 | return (0); |
| 612 | } |
| 613 | |
| 614 | /* |
| 615 | * Check to see if the new vnode represents a special device |
| 616 | * for which we already have a vnode (either because of |
| 617 | * bdevvp() or because of a different vnode representing |
| 618 | * the same block device). If such an alias exists, deallocate |
| 619 | * the existing contents and return the aliased vnode. The |
| 620 | * caller is responsible for filling it with its new contents. |
| 621 | */ |
| 622 | struct vnode * |
| 623 | checkalias(nvp, nvp_rdev, mp) |
| 624 | register struct vnode *nvp; |
| 625 | dev_t nvp_rdev; |
| 626 | struct mount *mp; |
| 627 | { |
| 628 | struct proc *p = curproc; /* XXX */ |
| 629 | struct vnode *vp; |
| 630 | struct vnode **vpp; |
| 631 | |
| 632 | if (nvp->v_type != VBLK && nvp->v_type != VCHR) |
| 633 | return (NULLVP); |
| 634 | |
| 635 | vpp = &speclisth[SPECHASH(nvp_rdev)]; |
| 636 | loop: |
| 637 | simple_lock(&spechash_slock); |
| 638 | for (vp = *vpp; vp; vp = vp->v_specnext) { |
| 639 | if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) |
| 640 | continue; |
| 641 | /* |
| 642 | * Alias, but not in use, so flush it out. |
| 643 | */ |
| 644 | simple_lock(&vp->v_interlock); |
| 645 | if (vp->v_usecount == 0) { |
| 646 | simple_unlock(&spechash_slock); |
| 647 | vgonel(vp, p); |
| 648 | goto loop; |
| 649 | } |
| 650 | if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, p)) { |
| 651 | simple_unlock(&spechash_slock); |
| 652 | goto loop; |
| 653 | } |
| 654 | break; |
| 655 | } |
| 656 | if (vp == NULL || vp->v_tag != VT_NON) { |
| 657 | MALLOC(nvp->v_specinfo, struct specinfo *, |
| 658 | sizeof(struct specinfo), M_VNODE, M_WAITOK); |
| 659 | nvp->v_rdev = nvp_rdev; |
| 660 | nvp->v_hashchain = vpp; |
| 661 | nvp->v_specnext = *vpp; |
| 662 | nvp->v_specflags = 0; |
| 663 | simple_unlock(&spechash_slock); |
| 664 | *vpp = nvp; |
| 665 | if (vp != NULLVP) { |
| 666 | nvp->v_flag |= VALIASED; |
| 667 | vp->v_flag |= VALIASED; |
| 668 | vput(vp); |
| 669 | } |
| 670 | return (NULLVP); |
| 671 | } |
| 672 | simple_unlock(&spechash_slock); |
| 673 | VOP_UNLOCK(vp, 0, p); |
| 674 | simple_lock(&vp->v_interlock); |
| 675 | vclean(vp, 0, p); |
| 676 | vp->v_op = nvp->v_op; |
| 677 | vp->v_tag = nvp->v_tag; |
| 678 | nvp->v_type = VNON; |
| 679 | insmntque(vp, mp); |
| 680 | return (vp); |
| 681 | } |
| 682 | |
| 683 | /* |
| 684 | * Grab a particular vnode from the free list, increment its |
| 685 | * reference count and lock it. The vnode lock bit is set the |
| 686 | * vnode is being eliminated in vgone. The process is awakened |
| 687 | * when the transition is completed, and an error returned to |
| 688 | * indicate that the vnode is no longer usable (possibly having |
| 689 | * been changed to a new file system type). |
| 690 | */ |
| 691 | int |
| 692 | vget(vp, flags, p) |
| 693 | struct vnode *vp; |
| 694 | int flags; |
| 695 | struct proc *p; |
| 696 | { |
| 697 | |
| 698 | /* |
| 699 | * If the vnode is in the process of being cleaned out for |
| 700 | * another use, we wait for the cleaning to finish and then |
| 701 | * return failure. Cleaning is determined by checking that |
| 702 | * the VXLOCK flag is set. |
| 703 | */ |
| 704 | if ((flags & LK_INTERLOCK) == 0) |
| 705 | simple_lock(&vp->v_interlock); |
| 706 | if (vp->v_flag & VXLOCK) { |
| 707 | vp->v_flag |= VXWANT; |
| 708 | simple_unlock(&vp->v_interlock); |
| 709 | tsleep((caddr_t)vp, PINOD, "vget", 0); |
| 710 | return (ENOENT); |
| 711 | } |
| 712 | if (vp->v_usecount == 0) { |
| 713 | if (vp->v_freelist.tqe_next == (struct vnode *)0xdeadf || |
| 714 | vp->v_freelist.tqe_prev == (struct vnode **)0xdeadb) |
| 715 | panic("vget: not on queue"); |
| 716 | TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); |
| 717 | simple_unlock(&vnode_free_list_slock); |
| 718 | } |
| 719 | vp->v_freelist.tqe_next = (struct vnode *)0xdeadf; |
| 720 | vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; |
| 721 | } |
| 722 | vp->v_usecount++; |
| 723 | if (flags & LK_TYPE_MASK) |
| 724 | return (vn_lock(vp, flags | LK_INTERLOCK, p)); |
| 725 | simple_unlock(&vp->v_interlock); |
| 726 | if (printcnt-- > 0) vprint("vget got", vp); |
| 727 | return (0); |
| 728 | } |
| 729 | |
| 730 | int bug_refs = 0; |
| 731 | |
| 732 | /* |
| 733 | * Stubs to use when there is no locking to be done on the underlying object. |
| 734 | * |
| 735 | * Getting a lock just clears the interlock if necessary. |
| 736 | */ |
| 737 | int |
| 738 | vop_nolock(ap) |
| 739 | struct vop_lock_args /* { |
| 740 | struct vnode *a_vp; |
| 741 | int a_flags; |
| 742 | struct proc *a_p; |
| 743 | } */ *ap; |
| 744 | { |
| 745 | struct vnode *vp = ap->a_vp; |
| 746 | |
| 747 | /* |
| 748 | * Since we are not using the lock manager, we must clear |
| 749 | * the interlock here. |
| 750 | */ |
| 751 | if (ap->a_flags & LK_INTERLOCK) |
| 752 | simple_unlock(&vp->v_interlock); |
| 753 | return (0); |
| 754 | } |
| 755 | |
| 756 | /* |
| 757 | * Unlock has nothing to do. |
| 758 | */ |
| 759 | int |
| 760 | vop_nounlock(ap) |
| 761 | struct vop_unlock_args /* { |
| 762 | struct vnode *a_vp; |
| 763 | int a_flags; |
| 764 | struct proc *a_p; |
| 765 | } */ *ap; |
| 766 | { |
| 767 | |
| 768 | return (0); |
| 769 | } |
| 770 | |
| 771 | /* |
| 772 | * Nothing is ever locked. |
| 773 | */ |
| 774 | int |
| 775 | vop_noislocked(ap) |
| 776 | struct vop_islocked_args /* { |
| 777 | struct vnode *a_vp; |
| 778 | } */ *ap; |
| 779 | { |
| 780 | |
| 781 | return (0); |
| 782 | } |
| 783 | |
| 784 | /* |
| 785 | * Vnode reference. |
| 786 | */ |
| 787 | void |
| 788 | vref(vp) |
| 789 | struct vnode *vp; |
| 790 | { |
| 791 | |
| 792 | simple_lock(&vp->v_interlock); |
| 793 | if (vp->v_usecount <= 0) |
| 794 | panic("vref used where vget required"); |
| 795 | if (vp->v_freelist.tqe_next != (struct vnode *)0xdeadf || |
| 796 | vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) |
| 797 | panic("vref: not free"); |
| 798 | vp->v_usecount++; |
| 799 | simple_unlock(&vp->v_interlock); |
| 800 | if (printcnt-- > 0) vprint("vref get", vp); |
| 801 | if (vp->v_type != VBLK && curproc) |
| 802 | curproc->p_spare[0]++; |
| 803 | if (bug_refs) |
| 804 | vprint("vref: "); |
| 805 | } |
| 806 | |
| 807 | /* |
| 808 | * vput(), just unlock and vrele() |
| 809 | */ |
| 810 | void |
| 811 | vput(vp) |
| 812 | struct vnode *vp; |
| 813 | { |
| 814 | struct proc *p = curproc; /* XXX */ |
| 815 | |
| 816 | #ifdef DIGANOSTIC |
| 817 | if (vp == NULL) |
| 818 | panic("vput: null vp"); |
| 819 | #endif |
| 820 | simple_lock(&vp->v_interlock); |
| 821 | vp->v_usecount--; |
| 822 | if (vp->v_usecount > 0) { |
| 823 | simple_unlock(&vp->v_interlock); |
| 824 | VOP_UNLOCK(vp, 0, p); |
| 825 | return; |
| 826 | } |
| 827 | #ifdef DIAGNOSTIC |
| 828 | if (vp->v_usecount < 0 || vp->v_writecount != 0) { |
| 829 | vprint("vput: bad ref count", vp); |
| 830 | panic("vput: ref cnt"); |
| 831 | } |
| 832 | #endif |
| 833 | /* |
| 834 | * insert at tail of LRU list |
| 835 | */ |
| 836 | simple_lock(&vnode_free_list_slock); |
| 837 | TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); |
| 838 | simple_unlock(&vnode_free_list_slock); |
| 839 | simple_unlock(&vp->v_interlock); |
| 840 | VOP_INACTIVE(vp, p); |
| 841 | } |
| 842 | |
| 843 | /* |
| 844 | * Vnode release. |
| 845 | * If count drops to zero, call inactive routine and return to freelist. |
| 846 | */ |
| 847 | void |
| 848 | vrele(vp) |
| 849 | struct vnode *vp; |
| 850 | { |
| 851 | struct proc *p = curproc; /* XXX */ |
| 852 | |
| 853 | #ifdef DIAGNOSTIC |
| 854 | if (vp == NULL) |
| 855 | panic("vrele: null vp"); |
| 856 | #endif |
| 857 | simple_lock(&vp->v_interlock); |
| 858 | vp->v_usecount--; |
| 859 | if (printcnt-- > 0) vprint("vrele put", vp); |
| 860 | if (vp->v_type != VBLK && curproc) |
| 861 | curproc->p_spare[0]--; |
| 862 | if (bug_refs) |
| 863 | vprint("vref: "); |
| 864 | if (vp->v_usecount > 0) { |
| 865 | simple_unlock(&vp->v_interlock); |
| 866 | return; |
| 867 | } |
| 868 | #ifdef DIAGNOSTIC |
| 869 | if (vp->v_usecount < 0 || vp->v_writecount != 0) { |
| 870 | vprint("vrele: bad ref count", vp); |
| 871 | panic("vrele: ref cnt"); |
| 872 | } |
| 873 | #endif |
| 874 | /* |
| 875 | * insert at tail of LRU list |
| 876 | */ |
| 877 | simple_lock(&vnode_free_list_slock); |
| 878 | if (vp->v_freelist.tqe_next != (struct vnode *)0xdeadf || |
| 879 | vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) |
| 880 | panic("vrele: not free"); |
| 881 | TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); |
| 882 | simple_unlock(&vnode_free_list_slock); |
| 883 | if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) |
| 884 | VOP_INACTIVE(vp, p); |
| 885 | } |
| 886 | |
| 887 | #ifdef DIAGNOSTIC |
| 888 | /* |
| 889 | * Page or buffer structure gets a reference. |
| 890 | */ |
| 891 | void |
| 892 | vhold(vp) |
| 893 | register struct vnode *vp; |
| 894 | { |
| 895 | |
| 896 | simple_lock(&vp->v_interlock); |
| 897 | vp->v_holdcnt++; |
| 898 | simple_unlock(&vp->v_interlock); |
| 899 | } |
| 900 | |
| 901 | /* |
| 902 | * Page or buffer structure frees a reference. |
| 903 | */ |
| 904 | void |
| 905 | holdrele(vp) |
| 906 | register struct vnode *vp; |
| 907 | { |
| 908 | |
| 909 | simple_lock(&vp->v_interlock); |
| 910 | if (vp->v_holdcnt <= 0) |
| 911 | panic("holdrele: holdcnt"); |
| 912 | vp->v_holdcnt--; |
| 913 | simple_unlock(&vp->v_interlock); |
| 914 | } |
| 915 | #endif /* DIAGNOSTIC */ |
| 916 | |
| 917 | /* |
| 918 | * Remove any vnodes in the vnode table belonging to mount point mp. |
| 919 | * |
| 920 | * If MNT_NOFORCE is specified, there should not be any active ones, |
| 921 | * return error if any are found (nb: this is a user error, not a |
| 922 | * system error). If MNT_FORCE is specified, detach any active vnodes |
| 923 | * that are found. |
| 924 | */ |
| 925 | #ifdef DIAGNOSTIC |
| 926 | int busyprt = 0; /* print out busy vnodes */ |
| 927 | struct ctldebug debug1 = { "busyprt", &busyprt }; |
| 928 | #endif |
| 929 | |
| 930 | int |
| 931 | vflush(mp, skipvp, flags) |
| 932 | struct mount *mp; |
| 933 | struct vnode *skipvp; |
| 934 | int flags; |
| 935 | { |
| 936 | struct proc *p = curproc; /* XXX */ |
| 937 | struct vnode *vp, *nvp; |
| 938 | int busy = 0; |
| 939 | |
| 940 | #ifdef DIAGNOSTIC |
| 941 | if ((mp->mnt_flag & MNT_MPBUSY) == 0) |
| 942 | panic("vflush: not busy"); |
| 943 | #endif |
| 944 | |
| 945 | simple_lock(&mntvnode_slock); |
| 946 | loop: |
| 947 | for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) { |
| 948 | if (vp->v_mount != mp) |
| 949 | goto loop; |
| 950 | nvp = vp->v_mntvnodes.le_next; |
| 951 | /* |
| 952 | * Skip over a selected vnode. |
| 953 | */ |
| 954 | if (vp == skipvp) |
| 955 | continue; |
| 956 | |
| 957 | simple_lock(&vp->v_interlock); |
| 958 | /* |
| 959 | * Skip over a vnodes marked VSYSTEM. |
| 960 | */ |
| 961 | if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { |
| 962 | simple_unlock(&vp->v_interlock); |
| 963 | continue; |
| 964 | } |
| 965 | /* |
| 966 | * If WRITECLOSE is set, only flush out regular file |
| 967 | * vnodes open for writing. |
| 968 | */ |
| 969 | if ((flags & WRITECLOSE) && |
| 970 | (vp->v_writecount == 0 || vp->v_type != VREG)) { |
| 971 | simple_unlock(&vp->v_interlock); |
| 972 | continue; |
| 973 | } |
| 974 | /* |
| 975 | * With v_usecount == 0, all we need to do is clear |
| 976 | * out the vnode data structures and we are done. |
| 977 | */ |
| 978 | if (vp->v_usecount == 0) { |
| 979 | simple_unlock(&mntvnode_slock); |
| 980 | vgonel(vp, p); |
| 981 | simple_lock(&mntvnode_slock); |
| 982 | continue; |
| 983 | } |
| 984 | /* |
| 985 | * If FORCECLOSE is set, forcibly close the vnode. |
| 986 | * For block or character devices, revert to an |
| 987 | * anonymous device. For all other files, just kill them. |
| 988 | */ |
| 989 | if (flags & FORCECLOSE) { |
| 990 | simple_unlock(&mntvnode_slock); |
| 991 | if (vp->v_type != VBLK && vp->v_type != VCHR) { |
| 992 | vgonel(vp, p); |
| 993 | } else { |
| 994 | vclean(vp, 0, p); |
| 995 | vp->v_op = spec_vnodeop_p; |
| 996 | insmntque(vp, (struct mount *)0); |
| 997 | } |
| 998 | simple_lock(&mntvnode_slock); |
| 999 | continue; |
| 1000 | } |
| 1001 | #ifdef DIAGNOSTIC |
| 1002 | if (busyprt) |
| 1003 | vprint("vflush: busy vnode", vp); |
| 1004 | #endif |
| 1005 | simple_unlock(&vp->v_interlock); |
| 1006 | busy++; |
| 1007 | } |
| 1008 | simple_unlock(&mntvnode_slock); |
| 1009 | if (busy) |
| 1010 | return (EBUSY); |
| 1011 | return (0); |
| 1012 | } |
| 1013 | |
| 1014 | /* |
| 1015 | * Disassociate the underlying file system from a vnode. |
| 1016 | * The vnode interlock is held on entry. |
| 1017 | */ |
| 1018 | static void |
| 1019 | vclean(vp, flags, p) |
| 1020 | struct vnode *vp; |
| 1021 | int flags; |
| 1022 | struct proc *p; |
| 1023 | { |
| 1024 | int active; |
| 1025 | |
| 1026 | /* |
| 1027 | * Check to see if the vnode is in use. |
| 1028 | * If so we have to reference it before we clean it out |
| 1029 | * so that its count cannot fall to zero and generate a |
| 1030 | * race against ourselves to recycle it. |
| 1031 | */ |
| 1032 | if (active = vp->v_usecount) |
| 1033 | vp->v_usecount++; |
| 1034 | /* |
| 1035 | * Prevent the vnode from being recycled or |
| 1036 | * brought into use while we clean it out. |
| 1037 | */ |
| 1038 | if (vp->v_flag & VXLOCK) |
| 1039 | panic("vclean: deadlock"); |
| 1040 | vp->v_flag |= VXLOCK; |
| 1041 | /* |
| 1042 | * Even if the count is zero, the VOP_INACTIVE routine may still |
| 1043 | * have the object locked while it cleans it out. The VOP_LOCK |
| 1044 | * ensures that the VOP_INACTIVE routine is done with its work. |
| 1045 | * For active vnodes, it ensures that no other activity can |
| 1046 | * occur while the underlying object is being cleaned out. |
| 1047 | */ |
| 1048 | VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); |
| 1049 | /* |
| 1050 | * Clean out any buffers associated with the vnode. |
| 1051 | */ |
| 1052 | if (flags & DOCLOSE) |
| 1053 | vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0); |
| 1054 | /* |
| 1055 | * If purging an active vnode, it must be closed and |
| 1056 | * deactivated before being reclaimed. Note that the |
| 1057 | * VOP_INACTIVE will unlock the vnode. |
| 1058 | */ |
| 1059 | if (active) { |
| 1060 | if (flags & DOCLOSE) |
| 1061 | VOP_CLOSE(vp, IO_NDELAY, NOCRED, p); |
| 1062 | VOP_INACTIVE(vp, p); |
| 1063 | } else { |
| 1064 | /* |
| 1065 | * Any other processes trying to obtain this lock must first |
| 1066 | * wait for VXLOCK to clear, then call the new lock operation. |
| 1067 | */ |
| 1068 | VOP_UNLOCK(vp, 0, p); |
| 1069 | } |
| 1070 | /* |
| 1071 | * Reclaim the vnode. |
| 1072 | */ |
| 1073 | if (VOP_RECLAIM(vp, p)) |
| 1074 | panic("vclean: cannot reclaim"); |
| 1075 | if (active) |
| 1076 | vrele(vp); |
| 1077 | cache_purge(vp); |
| 1078 | |
| 1079 | /* |
| 1080 | * Done with purge, notify sleepers of the grim news. |
| 1081 | */ |
| 1082 | vp->v_op = dead_vnodeop_p; |
| 1083 | vp->v_tag = VT_NON; |
| 1084 | vp->v_flag &= ~VXLOCK; |
| 1085 | if (vp->v_flag & VXWANT) { |
| 1086 | vp->v_flag &= ~VXWANT; |
| 1087 | wakeup((caddr_t)vp); |
| 1088 | } |
| 1089 | } |
| 1090 | |
| 1091 | /* |
| 1092 | * Eliminate all activity associated with the requested vnode |
| 1093 | * and with all vnodes aliased to the requested vnode. |
| 1094 | */ |
| 1095 | int |
| 1096 | vop_revoke(ap) |
| 1097 | struct vop_revoke_args /* { |
| 1098 | struct vnode *a_vp; |
| 1099 | int a_flags; |
| 1100 | } */ *ap; |
| 1101 | { |
| 1102 | struct vnode *vp, *vq; |
| 1103 | struct proc *p = curproc; /* XXX */ |
| 1104 | |
| 1105 | #ifdef DIAGNOSTIC |
| 1106 | if ((ap->a_flags & REVOKEALL) == 0) |
| 1107 | panic("vop_revoke"); |
| 1108 | #endif |
| 1109 | |
| 1110 | vp = ap->a_vp; |
| 1111 | simple_lock(&vp->v_interlock); |
| 1112 | |
| 1113 | if (vp->v_flag & VALIASED) { |
| 1114 | /* |
| 1115 | * If a vgone (or vclean) is already in progress, |
| 1116 | * wait until it is done and return. |
| 1117 | */ |
| 1118 | if (vp->v_flag & VXLOCK) { |
| 1119 | vp->v_flag |= VXWANT; |
| 1120 | simple_unlock(&vp->v_interlock); |
| 1121 | tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); |
| 1122 | return (0); |
| 1123 | } |
| 1124 | /* |
| 1125 | * Ensure that vp will not be vgone'd while we |
| 1126 | * are eliminating its aliases. |
| 1127 | */ |
| 1128 | vp->v_flag |= VXLOCK; |
| 1129 | simple_unlock(&vp->v_interlock); |
| 1130 | while (vp->v_flag & VALIASED) { |
| 1131 | simple_lock(&spechash_slock); |
| 1132 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 1133 | if (vq->v_rdev != vp->v_rdev || |
| 1134 | vq->v_type != vp->v_type || vp == vq) |
| 1135 | continue; |
| 1136 | simple_unlock(&spechash_slock); |
| 1137 | vgone(vq); |
| 1138 | break; |
| 1139 | } |
| 1140 | if (vq == NULLVP) |
| 1141 | simple_unlock(&spechash_slock); |
| 1142 | } |
| 1143 | /* |
| 1144 | * Remove the lock so that vgone below will |
| 1145 | * really eliminate the vnode after which time |
| 1146 | * vgone will awaken any sleepers. |
| 1147 | */ |
| 1148 | simple_lock(&vp->v_interlock); |
| 1149 | vp->v_flag &= ~VXLOCK; |
| 1150 | } |
| 1151 | vgonel(vp, p); |
| 1152 | return (0); |
| 1153 | } |
| 1154 | |
| 1155 | /* |
| 1156 | * Recycle an unused vnode to the front of the free list. |
| 1157 | * Release the passed interlock if the vnode will be recycled. |
| 1158 | */ |
| 1159 | int |
| 1160 | vrecycle(vp, inter_lkp, p) |
| 1161 | struct vnode *vp; |
| 1162 | struct simplelock *inter_lkp; |
| 1163 | struct proc *p; |
| 1164 | { |
| 1165 | |
| 1166 | simple_lock(&vp->v_interlock); |
| 1167 | if (vp->v_usecount == 0) { |
| 1168 | if (inter_lkp) |
| 1169 | simple_unlock(inter_lkp); |
| 1170 | vgonel(vp, p); |
| 1171 | return (1); |
| 1172 | } |
| 1173 | simple_unlock(&vp->v_interlock); |
| 1174 | return (0); |
| 1175 | } |
| 1176 | |
| 1177 | /* |
| 1178 | * Eliminate all activity associated with a vnode |
| 1179 | * in preparation for reuse. |
| 1180 | */ |
| 1181 | void |
| 1182 | vgone(vp) |
| 1183 | struct vnode *vp; |
| 1184 | { |
| 1185 | struct proc *p = curproc; /* XXX */ |
| 1186 | |
| 1187 | simple_lock(&vp->v_interlock); |
| 1188 | vgonel(vp, p); |
| 1189 | } |
| 1190 | |
| 1191 | /* |
| 1192 | * vgone, with the vp interlock held. |
| 1193 | */ |
| 1194 | void |
| 1195 | vgonel(vp, p) |
| 1196 | struct vnode *vp; |
| 1197 | struct proc *p; |
| 1198 | { |
| 1199 | struct vnode *vq; |
| 1200 | struct vnode *vx; |
| 1201 | |
| 1202 | /* |
| 1203 | * If a vgone (or vclean) is already in progress, |
| 1204 | * wait until it is done and return. |
| 1205 | */ |
| 1206 | if (vp->v_flag & VXLOCK) { |
| 1207 | vp->v_flag |= VXWANT; |
| 1208 | simple_unlock(&vp->v_interlock); |
| 1209 | tsleep((caddr_t)vp, PINOD, "vgone", 0); |
| 1210 | return; |
| 1211 | } |
| 1212 | /* |
| 1213 | * Clean out the filesystem specific data. |
| 1214 | */ |
| 1215 | vclean(vp, DOCLOSE, p); |
| 1216 | /* |
| 1217 | * Delete from old mount point vnode list, if on one. |
| 1218 | */ |
| 1219 | if (vp->v_mount != NULL) |
| 1220 | insmntque(vp, (struct mount *)0); |
| 1221 | /* |
| 1222 | * If special device, remove it from special device alias list |
| 1223 | * if it is on one. |
| 1224 | */ |
| 1225 | if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { |
| 1226 | simple_lock(&spechash_slock); |
| 1227 | if (*vp->v_hashchain == vp) { |
| 1228 | *vp->v_hashchain = vp->v_specnext; |
| 1229 | } else { |
| 1230 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 1231 | if (vq->v_specnext != vp) |
| 1232 | continue; |
| 1233 | vq->v_specnext = vp->v_specnext; |
| 1234 | break; |
| 1235 | } |
| 1236 | if (vq == NULL) |
| 1237 | panic("missing bdev"); |
| 1238 | } |
| 1239 | if (vp->v_flag & VALIASED) { |
| 1240 | vx = NULL; |
| 1241 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 1242 | if (vq->v_rdev != vp->v_rdev || |
| 1243 | vq->v_type != vp->v_type) |
| 1244 | continue; |
| 1245 | if (vx) |
| 1246 | break; |
| 1247 | vx = vq; |
| 1248 | } |
| 1249 | if (vx == NULL) |
| 1250 | panic("missing alias"); |
| 1251 | if (vq == NULL) |
| 1252 | vx->v_flag &= ~VALIASED; |
| 1253 | vp->v_flag &= ~VALIASED; |
| 1254 | } |
| 1255 | simple_unlock(&spechash_slock); |
| 1256 | FREE(vp->v_specinfo, M_VNODE); |
| 1257 | vp->v_specinfo = NULL; |
| 1258 | } |
| 1259 | /* |
| 1260 | * If it is on the freelist and not already at the head, |
| 1261 | * move it to the head of the list. The test of the back |
| 1262 | * pointer and the reference count of zero is because |
| 1263 | * it will be removed from the free list by getnewvnode, |
| 1264 | * but will not have its reference count incremented until |
| 1265 | * after calling vgone. If the reference count were |
| 1266 | * incremented first, vgone would (incorrectly) try to |
| 1267 | * close the previous instance of the underlying object. |
| 1268 | * So, the back pointer is explicitly set to `0xdeadb' in |
| 1269 | * getnewvnode after removing it from the freelist to ensure |
| 1270 | * that we do not try to move it here. |
| 1271 | */ |
| 1272 | if (vp->v_usecount == 0) { |
| 1273 | simple_lock(&vnode_free_list_slock); |
| 1274 | if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && |
| 1275 | vnode_free_list.tqh_first != vp) { |
| 1276 | TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); |
| 1277 | TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); |
| 1278 | } |
| 1279 | simple_unlock(&vnode_free_list_slock); |
| 1280 | } |
| 1281 | vp->v_type = VBAD; |
| 1282 | } |
| 1283 | |
| 1284 | /* |
| 1285 | * Lookup a vnode by device number. |
| 1286 | */ |
| 1287 | int |
| 1288 | vfinddev(dev, type, vpp) |
| 1289 | dev_t dev; |
| 1290 | enum vtype type; |
| 1291 | struct vnode **vpp; |
| 1292 | { |
| 1293 | struct vnode *vp; |
| 1294 | int rc = 0; |
| 1295 | |
| 1296 | simple_lock(&spechash_slock); |
| 1297 | for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { |
| 1298 | if (dev != vp->v_rdev || type != vp->v_type) |
| 1299 | continue; |
| 1300 | *vpp = vp; |
| 1301 | rc = 1; |
| 1302 | break; |
| 1303 | } |
| 1304 | simple_unlock(&spechash_slock); |
| 1305 | return (rc); |
| 1306 | } |
| 1307 | |
| 1308 | /* |
| 1309 | * Calculate the total number of references to a special device. |
| 1310 | */ |
| 1311 | int |
| 1312 | vcount(vp) |
| 1313 | struct vnode *vp; |
| 1314 | { |
| 1315 | struct vnode *vq, *vnext; |
| 1316 | int count; |
| 1317 | |
| 1318 | loop: |
| 1319 | if ((vp->v_flag & VALIASED) == 0) |
| 1320 | return (vp->v_usecount); |
| 1321 | simple_lock(&spechash_slock); |
| 1322 | for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { |
| 1323 | vnext = vq->v_specnext; |
| 1324 | if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) |
| 1325 | continue; |
| 1326 | /* |
| 1327 | * Alias, but not in use, so flush it out. |
| 1328 | */ |
| 1329 | if (vq->v_usecount == 0 && vq != vp) { |
| 1330 | simple_unlock(&spechash_slock); |
| 1331 | vgone(vq); |
| 1332 | goto loop; |
| 1333 | } |
| 1334 | count += vq->v_usecount; |
| 1335 | } |
| 1336 | simple_unlock(&spechash_slock); |
| 1337 | return (count); |
| 1338 | } |
| 1339 | |
| 1340 | /* |
| 1341 | * Print out a description of a vnode. |
| 1342 | */ |
| 1343 | static char *typename[] = |
| 1344 | { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; |
| 1345 | |
| 1346 | void |
| 1347 | vprint(label, vp) |
| 1348 | char *label; |
| 1349 | register struct vnode *vp; |
| 1350 | { |
| 1351 | char buf[64]; |
| 1352 | |
| 1353 | if (label != NULL) |
| 1354 | printf("%s: ", label); |
| 1355 | printf("num %d ", vp->v_spare[0]); |
| 1356 | printf("type %s, usecount %d, writecount %d, refcount %d,", |
| 1357 | typename[vp->v_type], vp->v_usecount, vp->v_writecount, |
| 1358 | vp->v_holdcnt); |
| 1359 | buf[0] = '\0'; |
| 1360 | if (vp->v_flag & VROOT) |
| 1361 | strcat(buf, "|VROOT"); |
| 1362 | if (vp->v_flag & VTEXT) |
| 1363 | strcat(buf, "|VTEXT"); |
| 1364 | if (vp->v_flag & VSYSTEM) |
| 1365 | strcat(buf, "|VSYSTEM"); |
| 1366 | if (vp->v_flag & VXLOCK) |
| 1367 | strcat(buf, "|VXLOCK"); |
| 1368 | if (vp->v_flag & VXWANT) |
| 1369 | strcat(buf, "|VXWANT"); |
| 1370 | if (vp->v_flag & VBWAIT) |
| 1371 | strcat(buf, "|VBWAIT"); |
| 1372 | if (vp->v_flag & VALIASED) |
| 1373 | strcat(buf, "|VALIASED"); |
| 1374 | if (buf[0] != '\0') |
| 1375 | printf(" flags (%s)", &buf[1]); |
| 1376 | if (vp->v_data == NULL) { |
| 1377 | printf("\n"); |
| 1378 | } else { |
| 1379 | printf("\n\t"); |
| 1380 | VOP_PRINT(vp); |
| 1381 | } |
| 1382 | } |
| 1383 | |
| 1384 | #ifdef DEBUG |
| 1385 | /* |
| 1386 | * List all of the locked vnodes in the system. |
| 1387 | * Called when debugging the kernel. |
| 1388 | */ |
| 1389 | void |
| 1390 | printlockedvnodes() |
| 1391 | { |
| 1392 | register struct mount *mp; |
| 1393 | register struct vnode *vp; |
| 1394 | |
| 1395 | printf("Locked vnodes\n"); |
| 1396 | for (mp = mountlist.cqh_first; mp != (void *)&mountlist; |
| 1397 | mp = mp->mnt_list.cqe_next) { |
| 1398 | for (vp = mp->mnt_vnodelist.lh_first; |
| 1399 | vp != NULL; |
| 1400 | vp = vp->v_mntvnodes.le_next) { |
| 1401 | if (VOP_ISLOCKED(vp)) |
| 1402 | vprint((char *)0, vp); |
| 1403 | } |
| 1404 | } |
| 1405 | } |
| 1406 | #endif |
| 1407 | |
| 1408 | /* |
| 1409 | * Top level filesystem related information gathering. |
| 1410 | */ |
| 1411 | int |
| 1412 | vfs_sysctl(name, namelen, oldp, oldlenp, newp, newlen, p) |
| 1413 | int *name; |
| 1414 | u_int namelen; |
| 1415 | void *oldp; |
| 1416 | size_t *oldlenp; |
| 1417 | void *newp; |
| 1418 | size_t newlen; |
| 1419 | struct proc *p; |
| 1420 | { |
| 1421 | struct ctldebug *cdp; |
| 1422 | struct vfsconf *vfsp; |
| 1423 | |
| 1424 | /* all sysctl names at this level are at least name and field */ |
| 1425 | if (namelen < 2) |
| 1426 | return (ENOTDIR); /* overloaded */ |
| 1427 | if (name[0] != VFS_GENERIC) { |
| 1428 | for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) |
| 1429 | if (vfsp->vfc_typenum == name[0]) |
| 1430 | break; |
| 1431 | if (vfsp == NULL) |
| 1432 | return (EOPNOTSUPP); |
| 1433 | return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, |
| 1434 | oldp, oldlenp, newp, newlen, p)); |
| 1435 | } |
| 1436 | switch (name[1]) { |
| 1437 | case VFS_MAXTYPENUM: |
| 1438 | return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf)); |
| 1439 | case VFS_CONF: |
| 1440 | if (namelen < 3) |
| 1441 | return (ENOTDIR); /* overloaded */ |
| 1442 | for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) |
| 1443 | if (vfsp->vfc_typenum == name[2]) |
| 1444 | break; |
| 1445 | if (vfsp == NULL) |
| 1446 | return (EOPNOTSUPP); |
| 1447 | return (sysctl_rdstruct(oldp, oldlenp, newp, vfsp, |
| 1448 | sizeof(struct vfsconf))); |
| 1449 | } |
| 1450 | return (EOPNOTSUPP); |
| 1451 | } |
| 1452 | |
| 1453 | int kinfo_vdebug = 1; |
| 1454 | int kinfo_vgetfailed; |
| 1455 | #define KINFO_VNODESLOP 10 |
| 1456 | /* |
| 1457 | * Dump vnode list (via sysctl). |
| 1458 | * Copyout address of vnode followed by vnode. |
| 1459 | */ |
| 1460 | /* ARGSUSED */ |
| 1461 | int |
| 1462 | sysctl_vnode(where, sizep) |
| 1463 | char *where; |
| 1464 | size_t *sizep; |
| 1465 | { |
| 1466 | register struct mount *mp, *nmp; |
| 1467 | struct vnode *nvp, *vp; |
| 1468 | register char *bp = where, *savebp; |
| 1469 | char *ewhere; |
| 1470 | int error; |
| 1471 | |
| 1472 | #define VPTRSZ sizeof (struct vnode *) |
| 1473 | #define VNODESZ sizeof (struct vnode) |
| 1474 | if (where == NULL) { |
| 1475 | *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); |
| 1476 | return (0); |
| 1477 | } |
| 1478 | ewhere = where + *sizep; |
| 1479 | |
| 1480 | for (mp = mountlist.cqh_first; mp != (void *)&mountlist; mp = nmp) { |
| 1481 | nmp = mp->mnt_list.cqe_next; |
| 1482 | if (vfs_busy(mp)) |
| 1483 | continue; |
| 1484 | savebp = bp; |
| 1485 | again: |
| 1486 | simple_lock(&mntvnode_slock); |
| 1487 | for (vp = mp->mnt_vnodelist.lh_first; |
| 1488 | vp != NULL; |
| 1489 | vp = nvp) { |
| 1490 | /* |
| 1491 | * Check that the vp is still associated with |
| 1492 | * this filesystem. RACE: could have been |
| 1493 | * recycled onto the same filesystem. |
| 1494 | */ |
| 1495 | if (vp->v_mount != mp) { |
| 1496 | simple_unlock(&mntvnode_slock); |
| 1497 | if (kinfo_vdebug) |
| 1498 | printf("kinfo: vp changed\n"); |
| 1499 | bp = savebp; |
| 1500 | goto again; |
| 1501 | } |
| 1502 | nvp = vp->v_mntvnodes.le_next; |
| 1503 | if (bp + VPTRSZ + VNODESZ > ewhere) { |
| 1504 | simple_unlock(&mntvnode_slock); |
| 1505 | *sizep = bp - where; |
| 1506 | return (ENOMEM); |
| 1507 | } |
| 1508 | simple_unlock(&mntvnode_slock); |
| 1509 | if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || |
| 1510 | (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ))) |
| 1511 | return (error); |
| 1512 | bp += VPTRSZ + VNODESZ; |
| 1513 | simple_lock(&mntvnode_slock); |
| 1514 | } |
| 1515 | simple_unlock(&mntvnode_slock); |
| 1516 | vfs_unbusy(mp); |
| 1517 | } |
| 1518 | |
| 1519 | *sizep = bp - where; |
| 1520 | return (0); |
| 1521 | } |
| 1522 | |
| 1523 | /* |
| 1524 | * Check to see if a filesystem is mounted on a block device. |
| 1525 | */ |
| 1526 | int |
| 1527 | vfs_mountedon(vp) |
| 1528 | struct vnode *vp; |
| 1529 | { |
| 1530 | struct vnode *vq; |
| 1531 | int error = 0; |
| 1532 | |
| 1533 | if (vp->v_specflags & SI_MOUNTEDON) |
| 1534 | return (EBUSY); |
| 1535 | if (vp->v_flag & VALIASED) { |
| 1536 | simple_lock(&spechash_slock); |
| 1537 | for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { |
| 1538 | if (vq->v_rdev != vp->v_rdev || |
| 1539 | vq->v_type != vp->v_type) |
| 1540 | continue; |
| 1541 | if (vq->v_specflags & SI_MOUNTEDON) { |
| 1542 | error = EBUSY; |
| 1543 | break; |
| 1544 | } |
| 1545 | } |
| 1546 | simple_unlock(&spechash_slock); |
| 1547 | } |
| 1548 | return (error); |
| 1549 | } |
| 1550 | |
| 1551 | /* |
| 1552 | * Unmount all filesystems. The list is traversed in reverse order |
| 1553 | * of mounting to avoid dependencies. |
| 1554 | */ |
| 1555 | void |
| 1556 | vfs_unmountall() |
| 1557 | { |
| 1558 | struct mount *mp, *nmp; |
| 1559 | |
| 1560 | for (mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { |
| 1561 | nmp = mp->mnt_list.cqe_prev; |
| 1562 | (void) dounmount(mp, MNT_FORCE, &proc0); |
| 1563 | } |
| 1564 | } |
| 1565 | |
| 1566 | /* |
| 1567 | * Build hash lists of net addresses and hang them off the mount point. |
| 1568 | * Called by ufs_mount() to set up the lists of export addresses. |
| 1569 | */ |
| 1570 | static int |
| 1571 | vfs_hang_addrlist(mp, nep, argp) |
| 1572 | struct mount *mp; |
| 1573 | struct netexport *nep; |
| 1574 | struct export_args *argp; |
| 1575 | { |
| 1576 | register struct netcred *np; |
| 1577 | register struct radix_node_head *rnh; |
| 1578 | register int i; |
| 1579 | struct radix_node *rn; |
| 1580 | struct sockaddr *saddr, *smask = 0; |
| 1581 | struct domain *dom; |
| 1582 | int error; |
| 1583 | |
| 1584 | if (argp->ex_addrlen == 0) { |
| 1585 | if (mp->mnt_flag & MNT_DEFEXPORTED) |
| 1586 | return (EPERM); |
| 1587 | np = &nep->ne_defexported; |
| 1588 | np->netc_exflags = argp->ex_flags; |
| 1589 | np->netc_anon = argp->ex_anon; |
| 1590 | np->netc_anon.cr_ref = 1; |
| 1591 | mp->mnt_flag |= MNT_DEFEXPORTED; |
| 1592 | return (0); |
| 1593 | } |
| 1594 | i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; |
| 1595 | np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); |
| 1596 | bzero((caddr_t)np, i); |
| 1597 | saddr = (struct sockaddr *)(np + 1); |
| 1598 | if (error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen)) |
| 1599 | goto out; |
| 1600 | if (saddr->sa_len > argp->ex_addrlen) |
| 1601 | saddr->sa_len = argp->ex_addrlen; |
| 1602 | if (argp->ex_masklen) { |
| 1603 | smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); |
| 1604 | error = copyin(argp->ex_addr, (caddr_t)smask, argp->ex_masklen); |
| 1605 | if (error) |
| 1606 | goto out; |
| 1607 | if (smask->sa_len > argp->ex_masklen) |
| 1608 | smask->sa_len = argp->ex_masklen; |
| 1609 | } |
| 1610 | i = saddr->sa_family; |
| 1611 | if ((rnh = nep->ne_rtable[i]) == 0) { |
| 1612 | /* |
| 1613 | * Seems silly to initialize every AF when most are not |
| 1614 | * used, do so on demand here |
| 1615 | */ |
| 1616 | for (dom = domains; dom; dom = dom->dom_next) |
| 1617 | if (dom->dom_family == i && dom->dom_rtattach) { |
| 1618 | dom->dom_rtattach((void **)&nep->ne_rtable[i], |
| 1619 | dom->dom_rtoffset); |
| 1620 | break; |
| 1621 | } |
| 1622 | if ((rnh = nep->ne_rtable[i]) == 0) { |
| 1623 | error = ENOBUFS; |
| 1624 | goto out; |
| 1625 | } |
| 1626 | } |
| 1627 | rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh, |
| 1628 | np->netc_rnodes); |
| 1629 | if (rn == 0) { |
| 1630 | /* |
| 1631 | * One of the reasons that rnh_addaddr may fail is that |
| 1632 | * the entry already exists. To check for this case, we |
| 1633 | * look up the entry to see if it is there. If so, we |
| 1634 | * do not need to make a new entry but do return success. |
| 1635 | */ |
| 1636 | free(np, M_NETADDR); |
| 1637 | rn = (*rnh->rnh_matchaddr)((caddr_t)saddr, rnh); |
| 1638 | if (rn != 0 && (rn->rn_flags & RNF_ROOT) == 0 && |
| 1639 | ((struct netcred *)rn)->netc_exflags == argp->ex_flags && |
| 1640 | !bcmp((caddr_t)&((struct netcred *)rn)->netc_anon, |
| 1641 | (caddr_t)&argp->ex_anon, sizeof(struct ucred))) |
| 1642 | return (0); |
| 1643 | return (EPERM); |
| 1644 | } |
| 1645 | np->netc_exflags = argp->ex_flags; |
| 1646 | np->netc_anon = argp->ex_anon; |
| 1647 | np->netc_anon.cr_ref = 1; |
| 1648 | return (0); |
| 1649 | out: |
| 1650 | free(np, M_NETADDR); |
| 1651 | return (error); |
| 1652 | } |
| 1653 | |
| 1654 | /* ARGSUSED */ |
| 1655 | static int |
| 1656 | vfs_free_netcred(rn, w) |
| 1657 | struct radix_node *rn; |
| 1658 | caddr_t w; |
| 1659 | { |
| 1660 | register struct radix_node_head *rnh = (struct radix_node_head *)w; |
| 1661 | |
| 1662 | (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh); |
| 1663 | free((caddr_t)rn, M_NETADDR); |
| 1664 | return (0); |
| 1665 | } |
| 1666 | |
| 1667 | /* |
| 1668 | * Free the net address hash lists that are hanging off the mount points. |
| 1669 | */ |
| 1670 | static void |
| 1671 | vfs_free_addrlist(nep) |
| 1672 | struct netexport *nep; |
| 1673 | { |
| 1674 | register int i; |
| 1675 | register struct radix_node_head *rnh; |
| 1676 | |
| 1677 | for (i = 0; i <= AF_MAX; i++) |
| 1678 | if (rnh = nep->ne_rtable[i]) { |
| 1679 | (*rnh->rnh_walktree)(rnh, vfs_free_netcred, |
| 1680 | (caddr_t)rnh); |
| 1681 | free((caddr_t)rnh, M_RTABLE); |
| 1682 | nep->ne_rtable[i] = 0; |
| 1683 | } |
| 1684 | } |
| 1685 | |
| 1686 | int |
| 1687 | vfs_export(mp, nep, argp) |
| 1688 | struct mount *mp; |
| 1689 | struct netexport *nep; |
| 1690 | struct export_args *argp; |
| 1691 | { |
| 1692 | int error; |
| 1693 | |
| 1694 | if (argp->ex_flags & MNT_DELEXPORT) { |
| 1695 | vfs_free_addrlist(nep); |
| 1696 | mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); |
| 1697 | } |
| 1698 | if (argp->ex_flags & MNT_EXPORTED) { |
| 1699 | if (error = vfs_hang_addrlist(mp, nep, argp)) |
| 1700 | return (error); |
| 1701 | mp->mnt_flag |= MNT_EXPORTED; |
| 1702 | } |
| 1703 | return (0); |
| 1704 | } |
| 1705 | |
| 1706 | struct netcred * |
| 1707 | vfs_export_lookup(mp, nep, nam) |
| 1708 | register struct mount *mp; |
| 1709 | struct netexport *nep; |
| 1710 | struct mbuf *nam; |
| 1711 | { |
| 1712 | register struct netcred *np; |
| 1713 | register struct radix_node_head *rnh; |
| 1714 | struct sockaddr *saddr; |
| 1715 | |
| 1716 | np = NULL; |
| 1717 | if (mp->mnt_flag & MNT_EXPORTED) { |
| 1718 | /* |
| 1719 | * Lookup in the export list first. |
| 1720 | */ |
| 1721 | if (nam != NULL) { |
| 1722 | saddr = mtod(nam, struct sockaddr *); |
| 1723 | rnh = nep->ne_rtable[saddr->sa_family]; |
| 1724 | if (rnh != NULL) { |
| 1725 | np = (struct netcred *) |
| 1726 | (*rnh->rnh_matchaddr)((caddr_t)saddr, |
| 1727 | rnh); |
| 1728 | if (np && np->netc_rnodes->rn_flags & RNF_ROOT) |
| 1729 | np = NULL; |
| 1730 | } |
| 1731 | } |
| 1732 | /* |
| 1733 | * If no address match, use the default if it exists. |
| 1734 | */ |
| 1735 | if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) |
| 1736 | np = &nep->ne_defexported; |
| 1737 | } |
| 1738 | return (np); |
| 1739 | } |