1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.50 2004/12/17 00:18:07 dillon Exp $ 41 */ 42 43 /* 44 * External virtual filesystem routines 45 */ 46 #include "opt_ddb.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/domain.h> 54 #include <sys/eventhandler.h> 55 #include <sys/fcntl.h> 56 #include <sys/kernel.h> 57 #include <sys/kthread.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/mount.h> 61 #include <sys/proc.h> 62 #include <sys/reboot.h> 63 #include <sys/socket.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/syslog.h> 67 #include <sys/vmmeter.h> 68 #include <sys/vnode.h> 69 70 #include <machine/limits.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_object.h> 74 #include <vm/vm_extern.h> 75 #include <vm/vm_kern.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_page.h> 79 #include <vm/vm_pager.h> 80 #include <vm/vnode_pager.h> 81 #include <vm/vm_zone.h> 82 83 #include <sys/buf2.h> 84 #include <sys/thread2.h> 85 86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 87 88 int numvnodes; 89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 90 int vfs_fastdev = 1; 91 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, ""); 92 93 enum vtype iftovt_tab[16] = { 94 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 95 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 96 }; 97 int vttoif_tab[9] = { 98 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 99 S_IFSOCK, S_IFIFO, S_IFMT, 100 }; 101 102 static int reassignbufcalls; 103 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, 104 &reassignbufcalls, 0, ""); 105 static int reassignbufloops; 106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, 107 &reassignbufloops, 0, ""); 108 static int reassignbufsortgood; 109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, 110 &reassignbufsortgood, 0, ""); 111 static int reassignbufsortbad; 112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, 113 &reassignbufsortbad, 0, ""); 114 static int reassignbufmethod = 1; 115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, 116 &reassignbufmethod, 0, ""); 117 118 int nfs_mount_type = -1; 119 static struct lwkt_token spechash_token; 120 struct nfs_public nfs_pub; /* publicly exported FS */ 121 122 int desiredvnodes; 123 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 124 &desiredvnodes, 0, "Maximum number of vnodes"); 125 126 static void vfs_free_addrlist (struct netexport *nep); 127 static int vfs_free_netcred (struct radix_node *rn, void *w); 128 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 129 struct export_args *argp); 130 131 extern int dev_ref_debug; 132 extern struct vnodeopv_entry_desc spec_vnodeop_entries[]; 133 134 /* 135 * Return 0 if the vnode is already on the free list or cannot be placed 136 * on the free list. Return 1 if the vnode can be placed on the free list. 137 */ 138 static __inline int 139 vshouldfree(struct vnode *vp, int usecount) 140 { 141 if (vp->v_flag & VFREE) 142 return (0); /* already free */ 143 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount) 144 return (0); /* other holderse */ 145 if (vp->v_object && 146 (vp->v_object->ref_count || vp->v_object->resident_page_count)) { 147 return (0); 148 } 149 return (1); 150 } 151 152 /* 153 * Initialize the vnode management data structures. 154 * 155 * Called from vfsinit() 156 */ 157 void 158 vfs_subr_init(void) 159 { 160 /* 161 * Desired vnodes is a result of the physical page count 162 * and the size of kernel's heap. It scales in proportion 163 * to the amount of available physical memory. This can 164 * cause trouble on 64-bit and large memory platforms. 165 */ 166 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */ 167 desiredvnodes = 168 min(maxproc + vmstats.v_page_count /4, 169 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 170 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 171 172 lwkt_token_init(&spechash_token); 173 } 174 175 /* 176 * Knob to control the precision of file timestamps: 177 * 178 * 0 = seconds only; nanoseconds zeroed. 179 * 1 = seconds and nanoseconds, accurate within 1/HZ. 180 * 2 = seconds and nanoseconds, truncated to microseconds. 181 * >=3 = seconds and nanoseconds, maximum precision. 182 */ 183 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 184 185 static int timestamp_precision = TSP_SEC; 186 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 187 ×tamp_precision, 0, ""); 188 189 /* 190 * Get a current timestamp. 191 */ 192 void 193 vfs_timestamp(struct timespec *tsp) 194 { 195 struct timeval tv; 196 197 switch (timestamp_precision) { 198 case TSP_SEC: 199 tsp->tv_sec = time_second; 200 tsp->tv_nsec = 0; 201 break; 202 case TSP_HZ: 203 getnanotime(tsp); 204 break; 205 case TSP_USEC: 206 microtime(&tv); 207 TIMEVAL_TO_TIMESPEC(&tv, tsp); 208 break; 209 case TSP_NSEC: 210 default: 211 nanotime(tsp); 212 break; 213 } 214 } 215 216 /* 217 * Set vnode attributes to VNOVAL 218 */ 219 void 220 vattr_null(struct vattr *vap) 221 { 222 vap->va_type = VNON; 223 vap->va_size = VNOVAL; 224 vap->va_bytes = VNOVAL; 225 vap->va_mode = VNOVAL; 226 vap->va_nlink = VNOVAL; 227 vap->va_uid = VNOVAL; 228 vap->va_gid = VNOVAL; 229 vap->va_fsid = VNOVAL; 230 vap->va_fileid = VNOVAL; 231 vap->va_blocksize = VNOVAL; 232 vap->va_rdev = VNOVAL; 233 vap->va_atime.tv_sec = VNOVAL; 234 vap->va_atime.tv_nsec = VNOVAL; 235 vap->va_mtime.tv_sec = VNOVAL; 236 vap->va_mtime.tv_nsec = VNOVAL; 237 vap->va_ctime.tv_sec = VNOVAL; 238 vap->va_ctime.tv_nsec = VNOVAL; 239 vap->va_flags = VNOVAL; 240 vap->va_gen = VNOVAL; 241 vap->va_vaflags = 0; 242 } 243 244 /* 245 * Update outstanding I/O count and do wakeup if requested. 246 */ 247 void 248 vwakeup(struct buf *bp) 249 { 250 struct vnode *vp; 251 252 bp->b_flags &= ~B_WRITEINPROG; 253 if ((vp = bp->b_vp)) { 254 vp->v_numoutput--; 255 if (vp->v_numoutput < 0) 256 panic("vwakeup: neg numoutput"); 257 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 258 vp->v_flag &= ~VBWAIT; 259 wakeup((caddr_t) &vp->v_numoutput); 260 } 261 } 262 } 263 264 /* 265 * Flush out and invalidate all buffers associated with a vnode. 266 * 267 * vp must be locked. 268 */ 269 int 270 vinvalbuf(struct vnode *vp, int flags, struct thread *td, 271 int slpflag, int slptimeo) 272 { 273 struct buf *bp; 274 struct buf *nbp, *blist; 275 int s, error; 276 vm_object_t object; 277 278 if (flags & V_SAVE) { 279 s = splbio(); 280 while (vp->v_numoutput) { 281 vp->v_flag |= VBWAIT; 282 error = tsleep((caddr_t)&vp->v_numoutput, 283 slpflag, "vinvlbuf", slptimeo); 284 if (error) { 285 splx(s); 286 return (error); 287 } 288 } 289 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 290 splx(s); 291 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) 292 return (error); 293 s = splbio(); 294 if (vp->v_numoutput > 0 || 295 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 296 panic("vinvalbuf: dirty bufs"); 297 } 298 splx(s); 299 } 300 s = splbio(); 301 for (;;) { 302 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 303 if (!blist) 304 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 305 if (!blist) 306 break; 307 308 for (bp = blist; bp; bp = nbp) { 309 nbp = TAILQ_NEXT(bp, b_vnbufs); 310 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 311 error = BUF_TIMELOCK(bp, 312 LK_EXCLUSIVE | LK_SLEEPFAIL, 313 "vinvalbuf", slpflag, slptimeo); 314 if (error == ENOLCK) 315 break; 316 splx(s); 317 return (error); 318 } 319 /* 320 * XXX Since there are no node locks for NFS, I 321 * believe there is a slight chance that a delayed 322 * write will occur while sleeping just above, so 323 * check for it. Note that vfs_bio_awrite expects 324 * buffers to reside on a queue, while VOP_BWRITE and 325 * brelse do not. 326 */ 327 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 328 (flags & V_SAVE)) { 329 330 if (bp->b_vp == vp) { 331 if (bp->b_flags & B_CLUSTEROK) { 332 BUF_UNLOCK(bp); 333 vfs_bio_awrite(bp); 334 } else { 335 bremfree(bp); 336 bp->b_flags |= B_ASYNC; 337 VOP_BWRITE(bp->b_vp, bp); 338 } 339 } else { 340 bremfree(bp); 341 (void) VOP_BWRITE(bp->b_vp, bp); 342 } 343 break; 344 } 345 bremfree(bp); 346 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 347 bp->b_flags &= ~B_ASYNC; 348 brelse(bp); 349 } 350 } 351 352 /* 353 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 354 * have write I/O in-progress but if there is a VM object then the 355 * VM object can also have read-I/O in-progress. 356 */ 357 do { 358 while (vp->v_numoutput > 0) { 359 vp->v_flag |= VBWAIT; 360 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0); 361 } 362 if (VOP_GETVOBJECT(vp, &object) == 0) { 363 while (object->paging_in_progress) 364 vm_object_pip_sleep(object, "vnvlbx"); 365 } 366 } while (vp->v_numoutput > 0); 367 368 splx(s); 369 370 /* 371 * Destroy the copy in the VM cache, too. 372 */ 373 if (VOP_GETVOBJECT(vp, &object) == 0) { 374 vm_object_page_remove(object, 0, 0, 375 (flags & V_SAVE) ? TRUE : FALSE); 376 } 377 378 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 379 panic("vinvalbuf: flush failed"); 380 return (0); 381 } 382 383 /* 384 * Truncate a file's buffer and pages to a specified length. This 385 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 386 * sync activity. 387 * 388 * The vnode must be locked. 389 */ 390 int 391 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize) 392 { 393 struct buf *bp; 394 struct buf *nbp; 395 int s, anyfreed; 396 int trunclbn; 397 398 /* 399 * Round up to the *next* lbn. 400 */ 401 trunclbn = (length + blksize - 1) / blksize; 402 403 s = splbio(); 404 restart: 405 anyfreed = 1; 406 for (;anyfreed;) { 407 anyfreed = 0; 408 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 409 nbp = TAILQ_NEXT(bp, b_vnbufs); 410 if (bp->b_lblkno >= trunclbn) { 411 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 412 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 413 goto restart; 414 } else { 415 bremfree(bp); 416 bp->b_flags |= (B_INVAL | B_RELBUF); 417 bp->b_flags &= ~B_ASYNC; 418 brelse(bp); 419 anyfreed = 1; 420 } 421 if (nbp && 422 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 423 (nbp->b_vp != vp) || 424 (nbp->b_flags & B_DELWRI))) { 425 goto restart; 426 } 427 } 428 } 429 430 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 431 nbp = TAILQ_NEXT(bp, b_vnbufs); 432 if (bp->b_lblkno >= trunclbn) { 433 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 434 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 435 goto restart; 436 } else { 437 bremfree(bp); 438 bp->b_flags |= (B_INVAL | B_RELBUF); 439 bp->b_flags &= ~B_ASYNC; 440 brelse(bp); 441 anyfreed = 1; 442 } 443 if (nbp && 444 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 445 (nbp->b_vp != vp) || 446 (nbp->b_flags & B_DELWRI) == 0)) { 447 goto restart; 448 } 449 } 450 } 451 } 452 453 if (length > 0) { 454 restartsync: 455 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 456 nbp = TAILQ_NEXT(bp, b_vnbufs); 457 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 458 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 459 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 460 goto restart; 461 } else { 462 bremfree(bp); 463 if (bp->b_vp == vp) { 464 bp->b_flags |= B_ASYNC; 465 } else { 466 bp->b_flags &= ~B_ASYNC; 467 } 468 VOP_BWRITE(bp->b_vp, bp); 469 } 470 goto restartsync; 471 } 472 473 } 474 } 475 476 while (vp->v_numoutput > 0) { 477 vp->v_flag |= VBWAIT; 478 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0); 479 } 480 481 splx(s); 482 483 vnode_pager_setsize(vp, length); 484 485 return (0); 486 } 487 488 /* 489 * Associate a buffer with a vnode. 490 */ 491 void 492 bgetvp(struct vnode *vp, struct buf *bp) 493 { 494 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 495 496 vhold(vp); 497 bp->b_vp = vp; 498 bp->b_dev = vn_todev(vp); 499 /* 500 * Insert onto list for new vnode. 501 */ 502 crit_enter(); 503 bp->b_xflags |= BX_VNCLEAN; 504 bp->b_xflags &= ~BX_VNDIRTY; 505 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 506 crit_exit(); 507 } 508 509 /* 510 * Disassociate a buffer from a vnode. 511 */ 512 void 513 brelvp(struct buf *bp) 514 { 515 struct vnode *vp; 516 struct buflists *listheadp; 517 518 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 519 520 /* 521 * Delete from old vnode list, if on one. 522 */ 523 vp = bp->b_vp; 524 crit_enter(); 525 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 526 if (bp->b_xflags & BX_VNDIRTY) 527 listheadp = &vp->v_dirtyblkhd; 528 else 529 listheadp = &vp->v_cleanblkhd; 530 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 531 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 532 } 533 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 534 vp->v_flag &= ~VONWORKLST; 535 LIST_REMOVE(vp, v_synclist); 536 } 537 crit_exit(); 538 bp->b_vp = NULL; 539 vdrop(vp); 540 } 541 542 /* 543 * Associate a p-buffer with a vnode. 544 * 545 * Also sets B_PAGING flag to indicate that vnode is not fully associated 546 * with the buffer. i.e. the bp has not been linked into the vnode or 547 * ref-counted. 548 */ 549 void 550 pbgetvp(struct vnode *vp, struct buf *bp) 551 { 552 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 553 554 bp->b_vp = vp; 555 bp->b_flags |= B_PAGING; 556 bp->b_dev = vn_todev(vp); 557 } 558 559 /* 560 * Disassociate a p-buffer from a vnode. 561 */ 562 void 563 pbrelvp(struct buf *bp) 564 { 565 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 566 567 /* XXX REMOVE ME */ 568 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 569 panic( 570 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 571 bp, 572 (int)bp->b_flags 573 ); 574 } 575 bp->b_vp = (struct vnode *) 0; 576 bp->b_flags &= ~B_PAGING; 577 } 578 579 void 580 pbreassignbuf(struct buf *bp, struct vnode *newvp) 581 { 582 if ((bp->b_flags & B_PAGING) == 0) { 583 panic( 584 "pbreassignbuf() on non phys bp %p", 585 bp 586 ); 587 } 588 bp->b_vp = newvp; 589 } 590 591 /* 592 * Reassign a buffer from one vnode to another. 593 * Used to assign file specific control information 594 * (indirect blocks) to the vnode to which they belong. 595 */ 596 void 597 reassignbuf(struct buf *bp, struct vnode *newvp) 598 { 599 struct buflists *listheadp; 600 int delay; 601 602 if (newvp == NULL) { 603 printf("reassignbuf: NULL"); 604 return; 605 } 606 ++reassignbufcalls; 607 608 /* 609 * B_PAGING flagged buffers cannot be reassigned because their vp 610 * is not fully linked in. 611 */ 612 if (bp->b_flags & B_PAGING) 613 panic("cannot reassign paging buffer"); 614 615 crit_enter(); 616 /* 617 * Delete from old vnode list, if on one. 618 */ 619 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 620 if (bp->b_xflags & BX_VNDIRTY) 621 listheadp = &bp->b_vp->v_dirtyblkhd; 622 else 623 listheadp = &bp->b_vp->v_cleanblkhd; 624 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 625 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 626 if (bp->b_vp != newvp) { 627 vdrop(bp->b_vp); 628 bp->b_vp = NULL; /* for clarification */ 629 } 630 } 631 /* 632 * If dirty, put on list of dirty buffers; otherwise insert onto list 633 * of clean buffers. 634 */ 635 if (bp->b_flags & B_DELWRI) { 636 struct buf *tbp; 637 638 listheadp = &newvp->v_dirtyblkhd; 639 if ((newvp->v_flag & VONWORKLST) == 0) { 640 switch (newvp->v_type) { 641 case VDIR: 642 delay = dirdelay; 643 break; 644 case VCHR: 645 case VBLK: 646 if (newvp->v_rdev && 647 newvp->v_rdev->si_mountpoint != NULL) { 648 delay = metadelay; 649 break; 650 } 651 /* fall through */ 652 default: 653 delay = filedelay; 654 } 655 vn_syncer_add_to_worklist(newvp, delay); 656 } 657 bp->b_xflags |= BX_VNDIRTY; 658 tbp = TAILQ_FIRST(listheadp); 659 if (tbp == NULL || 660 bp->b_lblkno == 0 || 661 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 662 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 663 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 664 ++reassignbufsortgood; 665 } else if (bp->b_lblkno < 0) { 666 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 667 ++reassignbufsortgood; 668 } else if (reassignbufmethod == 1) { 669 /* 670 * New sorting algorithm, only handle sequential case, 671 * otherwise append to end (but before metadata) 672 */ 673 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 674 (tbp->b_xflags & BX_VNDIRTY)) { 675 /* 676 * Found the best place to insert the buffer 677 */ 678 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 679 ++reassignbufsortgood; 680 } else { 681 /* 682 * Missed, append to end, but before meta-data. 683 * We know that the head buffer in the list is 684 * not meta-data due to prior conditionals. 685 * 686 * Indirect effects: NFS second stage write 687 * tends to wind up here, giving maximum 688 * distance between the unstable write and the 689 * commit rpc. 690 */ 691 tbp = TAILQ_LAST(listheadp, buflists); 692 while (tbp && tbp->b_lblkno < 0) 693 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 694 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 695 ++reassignbufsortbad; 696 } 697 } else { 698 /* 699 * Old sorting algorithm, scan queue and insert 700 */ 701 struct buf *ttbp; 702 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 703 (ttbp->b_lblkno < bp->b_lblkno)) { 704 ++reassignbufloops; 705 tbp = ttbp; 706 } 707 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 708 } 709 } else { 710 bp->b_xflags |= BX_VNCLEAN; 711 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 712 if ((newvp->v_flag & VONWORKLST) && 713 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 714 newvp->v_flag &= ~VONWORKLST; 715 LIST_REMOVE(newvp, v_synclist); 716 } 717 } 718 if (bp->b_vp != newvp) { 719 bp->b_vp = newvp; 720 vhold(bp->b_vp); 721 } 722 crit_exit(); 723 } 724 725 /* 726 * Create a vnode for a block device. 727 * Used for mounting the root file system. 728 */ 729 int 730 bdevvp(dev_t dev, struct vnode **vpp) 731 { 732 struct vnode *vp; 733 struct vnode *nvp; 734 int error; 735 736 if (dev == NODEV) { 737 *vpp = NULLVP; 738 return (ENXIO); 739 } 740 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0); 741 if (error) { 742 *vpp = NULLVP; 743 return (error); 744 } 745 vp = nvp; 746 vp->v_type = VCHR; 747 vp->v_udev = dev->si_udev; 748 vx_unlock(vp); 749 *vpp = vp; 750 return (0); 751 } 752 753 int 754 v_associate_rdev(struct vnode *vp, dev_t dev) 755 { 756 lwkt_tokref ilock; 757 758 if (dev == NULL || dev == NODEV) 759 return(ENXIO); 760 if (dev_is_good(dev) == 0) 761 return(ENXIO); 762 KKASSERT(vp->v_rdev == NULL); 763 if (dev_ref_debug) 764 printf("Z1"); 765 vp->v_rdev = reference_dev(dev); 766 lwkt_gettoken(&ilock, &spechash_token); 767 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext); 768 lwkt_reltoken(&ilock); 769 return(0); 770 } 771 772 void 773 v_release_rdev(struct vnode *vp) 774 { 775 lwkt_tokref ilock; 776 dev_t dev; 777 778 if ((dev = vp->v_rdev) != NULL) { 779 lwkt_gettoken(&ilock, &spechash_token); 780 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext); 781 if (dev_ref_debug && vp->v_opencount != 0) { 782 printf("releasing rdev with non-0 " 783 "v_opencount(%d) (revoked?)\n", 784 vp->v_opencount); 785 } 786 vp->v_rdev = NULL; 787 vp->v_opencount = 0; 788 release_dev(dev); 789 lwkt_reltoken(&ilock); 790 } 791 } 792 793 /* 794 * Add a vnode to the alias list hung off the dev_t. We only associate 795 * the device number with the vnode. The actual device is not associated 796 * until the vnode is opened (usually in spec_open()), and will be 797 * disassociated on last close. 798 */ 799 void 800 addaliasu(struct vnode *nvp, udev_t nvp_udev) 801 { 802 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 803 panic("addaliasu on non-special vnode"); 804 nvp->v_udev = nvp_udev; 805 } 806 807 /* 808 * Disassociate a vnode from its underlying filesystem. 809 * 810 * The vnode must be VX locked and refd 811 * 812 * If there are v_usecount references to the vnode other then ours we have 813 * to VOP_CLOSE the vnode before we can deactivate and reclaim it. 814 */ 815 void 816 vclean(struct vnode *vp, int flags, struct thread *td) 817 { 818 int active; 819 820 /* 821 * If the vnode has already been reclaimed we have nothing to do. 822 */ 823 if (vp->v_flag & VRECLAIMED) 824 return; 825 vp->v_flag |= VRECLAIMED; 826 827 /* 828 * Scrap the vfs cache 829 */ 830 cache_inval_vp(vp, 0); 831 832 /* 833 * Check to see if the vnode is in use. If so we have to reference it 834 * before we clean it out so that its count cannot fall to zero and 835 * generate a race against ourselves to recycle it. 836 */ 837 active = (vp->v_usecount > 1); 838 839 /* 840 * Clean out any buffers associated with the vnode and destroy its 841 * object, if it has one. 842 */ 843 vinvalbuf(vp, V_SAVE, td, 0, 0); 844 VOP_DESTROYVOBJECT(vp); 845 846 /* 847 * If purging an active vnode, it must be closed and 848 * deactivated before being reclaimed. XXX 849 * 850 * Note that neither of these routines unlocks the vnode. 851 */ 852 if (active) { 853 if (flags & DOCLOSE) 854 VOP_CLOSE(vp, FNONBLOCK, td); 855 } 856 857 /* 858 * If the vnode has not be deactivated, deactivated it. 859 */ 860 if ((vp->v_flag & VINACTIVE) == 0) { 861 vp->v_flag |= VINACTIVE; 862 VOP_INACTIVE(vp, td); 863 } 864 865 /* 866 * Reclaim the vnode. 867 */ 868 if (VOP_RECLAIM(vp, td)) 869 panic("vclean: cannot reclaim"); 870 871 /* 872 * Done with purge, notify sleepers of the grim news. 873 */ 874 vp->v_ops = &dead_vnode_vops; 875 vn_pollgone(vp); 876 vp->v_tag = VT_NON; 877 } 878 879 /* 880 * Eliminate all activity associated with the requested vnode 881 * and with all vnodes aliased to the requested vnode. 882 * 883 * The vnode must be referenced and vx_lock()'d 884 * 885 * revoke { struct vnode *a_vp, int a_flags } 886 */ 887 int 888 vop_stdrevoke(struct vop_revoke_args *ap) 889 { 890 struct vnode *vp, *vq; 891 lwkt_tokref ilock; 892 dev_t dev; 893 894 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 895 896 vp = ap->a_vp; 897 898 /* 899 * If the vnode is already dead don't try to revoke it 900 */ 901 if (vp->v_flag & VRECLAIMED) 902 return (0); 903 904 /* 905 * If the vnode has a device association, scrap all vnodes associated 906 * with the device. Don't let the device disappear on us while we 907 * are scrapping the vnodes. 908 * 909 * The passed vp will probably show up in the list, do not VX lock 910 * it twice! 911 */ 912 if (vp->v_type != VCHR && vp->v_type != VBLK) 913 return(0); 914 if ((dev = vp->v_rdev) == NULL) { 915 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV) 916 return(0); 917 } 918 reference_dev(dev); 919 lwkt_gettoken(&ilock, &spechash_token); 920 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 921 if (vp == vq || vx_get(vq) == 0) { 922 if (vq == SLIST_FIRST(&dev->si_hlist)) 923 vgone(vq); 924 if (vp != vq) 925 vx_put(vq); 926 } 927 } 928 lwkt_reltoken(&ilock); 929 release_dev(dev); 930 return (0); 931 } 932 933 /* 934 * Recycle an unused vnode to the front of the free list. 935 * 936 * Returns 1 if we were successfully able to recycle the vnode, 937 * 0 otherwise. 938 */ 939 int 940 vrecycle(struct vnode *vp, struct thread *td) 941 { 942 if (vp->v_usecount == 1) { 943 vgone(vp); 944 return (1); 945 } 946 return (0); 947 } 948 949 /* 950 * Eliminate all activity associated with a vnode in preparation for reuse. 951 * 952 * The vnode must be VX locked and refd and will remain VX locked and refd 953 * on return. This routine may be called with the vnode in any state, as 954 * long as it is VX locked. The vnode will be cleaned out and marked 955 * VRECLAIMED but will not actually be reused until all existing refs and 956 * holds go away. 957 * 958 * NOTE: This routine may be called on a vnode which has not yet been 959 * already been deactivated (VOP_INACTIVE), or on a vnode which has 960 * already been reclaimed. 961 * 962 * This routine is not responsible for placing us back on the freelist. 963 * Instead, it happens automatically when the caller releases the VX lock 964 * (assuming there aren't any other references). 965 */ 966 void 967 vgone(struct vnode *vp) 968 { 969 /* 970 * assert that the VX lock is held. This is an absolute requirement 971 * now for vgone() to be called. 972 */ 973 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 974 975 /* 976 * Clean out the filesystem specific data and set the VRECLAIMED 977 * bit. Also deactivate the vnode if necessary. 978 */ 979 vclean(vp, DOCLOSE, curthread); 980 981 /* 982 * Delete from old mount point vnode list, if on one. 983 */ 984 if (vp->v_mount != NULL) 985 insmntque(vp, NULL); 986 987 /* 988 * If special device, remove it from special device alias list 989 * if it is on one. This should normally only occur if a vnode is 990 * being revoked as the device should otherwise have been released 991 * naturally. 992 */ 993 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 994 v_release_rdev(vp); 995 } 996 997 /* 998 * Set us to VBAD 999 */ 1000 vp->v_type = VBAD; 1001 } 1002 1003 /* 1004 * Lookup a vnode by device number. 1005 */ 1006 int 1007 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp) 1008 { 1009 lwkt_tokref ilock; 1010 struct vnode *vp; 1011 1012 lwkt_gettoken(&ilock, &spechash_token); 1013 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1014 if (type == vp->v_type) { 1015 *vpp = vp; 1016 lwkt_reltoken(&ilock); 1017 return (1); 1018 } 1019 } 1020 lwkt_reltoken(&ilock); 1021 return (0); 1022 } 1023 1024 /* 1025 * Calculate the total number of references to a special device. This 1026 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1027 * an overloaded field. Since udev2dev can now return NODEV, we have 1028 * to check for a NULL v_rdev. 1029 */ 1030 int 1031 count_dev(dev_t dev) 1032 { 1033 lwkt_tokref ilock; 1034 struct vnode *vp; 1035 int count = 0; 1036 1037 if (SLIST_FIRST(&dev->si_hlist)) { 1038 lwkt_gettoken(&ilock, &spechash_token); 1039 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1040 count += vp->v_usecount; 1041 } 1042 lwkt_reltoken(&ilock); 1043 } 1044 return(count); 1045 } 1046 1047 int 1048 count_udev(udev_t udev) 1049 { 1050 dev_t dev; 1051 1052 if ((dev = udev2dev(udev, 0)) == NODEV) 1053 return(0); 1054 return(count_dev(dev)); 1055 } 1056 1057 int 1058 vcount(struct vnode *vp) 1059 { 1060 if (vp->v_rdev == NULL) 1061 return(0); 1062 return(count_dev(vp->v_rdev)); 1063 } 1064 1065 /* 1066 * Print out a description of a vnode. 1067 */ 1068 static char *typename[] = 1069 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1070 1071 void 1072 vprint(char *label, struct vnode *vp) 1073 { 1074 char buf[96]; 1075 1076 if (label != NULL) 1077 printf("%s: %p: ", label, (void *)vp); 1078 else 1079 printf("%p: ", (void *)vp); 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 & VBWAIT) 1091 strcat(buf, "|VBWAIT"); 1092 if (vp->v_flag & VFREE) 1093 strcat(buf, "|VFREE"); 1094 if (vp->v_flag & VOBJBUF) 1095 strcat(buf, "|VOBJBUF"); 1096 if (buf[0] != '\0') 1097 printf(" flags (%s)", &buf[1]); 1098 if (vp->v_data == NULL) { 1099 printf("\n"); 1100 } else { 1101 printf("\n\t"); 1102 VOP_PRINT(vp); 1103 } 1104 } 1105 1106 #ifdef DDB 1107 #include <ddb/ddb.h> 1108 /* 1109 * List all of the locked vnodes in the system. 1110 * Called when debugging the kernel. 1111 */ 1112 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1113 { 1114 struct thread *td = curthread; /* XXX */ 1115 lwkt_tokref ilock; 1116 struct mount *mp, *nmp; 1117 struct vnode *vp; 1118 1119 printf("Locked vnodes\n"); 1120 lwkt_gettoken(&ilock, &mountlist_token); 1121 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 1122 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) { 1123 nmp = TAILQ_NEXT(mp, mnt_list); 1124 continue; 1125 } 1126 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1127 if (VOP_ISLOCKED(vp, NULL)) 1128 vprint((char *)0, vp); 1129 } 1130 lwkt_gettokref(&ilock); 1131 nmp = TAILQ_NEXT(mp, mnt_list); 1132 vfs_unbusy(mp, td); 1133 } 1134 lwkt_reltoken(&ilock); 1135 } 1136 #endif 1137 1138 /* 1139 * Top level filesystem related information gathering. 1140 */ 1141 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1142 1143 static int 1144 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1145 { 1146 int *name = (int *)arg1 - 1; /* XXX */ 1147 u_int namelen = arg2 + 1; /* XXX */ 1148 struct vfsconf *vfsp; 1149 1150 #if 1 || defined(COMPAT_PRELITE2) 1151 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1152 if (namelen == 1) 1153 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1154 #endif 1155 1156 #ifdef notyet 1157 /* all sysctl names at this level are at least name and field */ 1158 if (namelen < 2) 1159 return (ENOTDIR); /* overloaded */ 1160 if (name[0] != VFS_GENERIC) { 1161 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1162 if (vfsp->vfc_typenum == name[0]) 1163 break; 1164 if (vfsp == NULL) 1165 return (EOPNOTSUPP); 1166 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1167 oldp, oldlenp, newp, newlen, p)); 1168 } 1169 #endif 1170 switch (name[1]) { 1171 case VFS_MAXTYPENUM: 1172 if (namelen != 2) 1173 return (ENOTDIR); 1174 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1175 case VFS_CONF: 1176 if (namelen != 3) 1177 return (ENOTDIR); /* overloaded */ 1178 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1179 if (vfsp->vfc_typenum == name[2]) 1180 break; 1181 if (vfsp == NULL) 1182 return (EOPNOTSUPP); 1183 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1184 } 1185 return (EOPNOTSUPP); 1186 } 1187 1188 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1189 "Generic filesystem"); 1190 1191 #if 1 || defined(COMPAT_PRELITE2) 1192 1193 static int 1194 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1195 { 1196 int error; 1197 struct vfsconf *vfsp; 1198 struct ovfsconf ovfs; 1199 1200 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1201 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1202 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1203 ovfs.vfc_index = vfsp->vfc_typenum; 1204 ovfs.vfc_refcount = vfsp->vfc_refcount; 1205 ovfs.vfc_flags = vfsp->vfc_flags; 1206 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1207 if (error) 1208 return error; 1209 } 1210 return 0; 1211 } 1212 1213 #endif /* 1 || COMPAT_PRELITE2 */ 1214 1215 #if 0 1216 #define KINFO_VNODESLOP 10 1217 /* 1218 * Dump vnode list (via sysctl). 1219 * Copyout address of vnode followed by vnode. 1220 */ 1221 /* ARGSUSED */ 1222 static int 1223 sysctl_vnode(SYSCTL_HANDLER_ARGS) 1224 { 1225 struct proc *p = curproc; /* XXX */ 1226 struct mount *mp, *nmp; 1227 struct vnode *nvp, *vp; 1228 lwkt_tokref ilock; 1229 lwkt_tokref jlock; 1230 int error; 1231 1232 #define VPTRSZ sizeof (struct vnode *) 1233 #define VNODESZ sizeof (struct vnode) 1234 1235 req->lock = 0; 1236 if (!req->oldptr) /* Make an estimate */ 1237 return (SYSCTL_OUT(req, 0, 1238 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 1239 1240 lwkt_gettoken(&ilock, &mountlist_token); 1241 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 1242 if (vfs_busy(mp, LK_NOWAIT, &ilock, p)) { 1243 nmp = TAILQ_NEXT(mp, mnt_list); 1244 continue; 1245 } 1246 lwkt_gettoken(&jlock, &mntvnode_token); 1247 again: 1248 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 1249 vp != NULL; 1250 vp = nvp) { 1251 /* 1252 * Check that the vp is still associated with 1253 * this filesystem. RACE: could have been 1254 * recycled onto the same filesystem. 1255 */ 1256 if (vp->v_mount != mp) 1257 goto again; 1258 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1259 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 1260 (error = SYSCTL_OUT(req, vp, VNODESZ))) { 1261 lwkt_reltoken(&jlock); 1262 return (error); 1263 } 1264 } 1265 lwkt_reltoken(&jlock); 1266 lwkt_gettokref(&ilock); 1267 nmp = TAILQ_NEXT(mp, mnt_list); /* ZZZ */ 1268 vfs_unbusy(mp, p); 1269 } 1270 lwkt_reltoken(&ilock); 1271 1272 return (0); 1273 } 1274 #endif 1275 1276 /* 1277 * XXX 1278 * Exporting the vnode list on large systems causes them to crash. 1279 * Exporting the vnode list on medium systems causes sysctl to coredump. 1280 */ 1281 #if 0 1282 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 1283 0, 0, sysctl_vnode, "S,vnode", ""); 1284 #endif 1285 1286 /* 1287 * Check to see if a filesystem is mounted on a block device. 1288 */ 1289 int 1290 vfs_mountedon(struct vnode *vp) 1291 { 1292 dev_t dev; 1293 1294 if ((dev = vp->v_rdev) == NULL) 1295 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1296 if (dev != NODEV && dev->si_mountpoint) 1297 return (EBUSY); 1298 return (0); 1299 } 1300 1301 /* 1302 * Unmount all filesystems. The list is traversed in reverse order 1303 * of mounting to avoid dependencies. 1304 */ 1305 void 1306 vfs_unmountall(void) 1307 { 1308 struct mount *mp; 1309 struct thread *td = curthread; 1310 int error; 1311 1312 if (td->td_proc == NULL) 1313 td = initproc->p_thread; /* XXX XXX use proc0 instead? */ 1314 1315 /* 1316 * Since this only runs when rebooting, it is not interlocked. 1317 */ 1318 while(!TAILQ_EMPTY(&mountlist)) { 1319 mp = TAILQ_LAST(&mountlist, mntlist); 1320 error = dounmount(mp, MNT_FORCE, td); 1321 if (error) { 1322 TAILQ_REMOVE(&mountlist, mp, mnt_list); 1323 printf("unmount of %s failed (", 1324 mp->mnt_stat.f_mntonname); 1325 if (error == EBUSY) 1326 printf("BUSY)\n"); 1327 else 1328 printf("%d)\n", error); 1329 } else { 1330 /* The unmount has removed mp from the mountlist */ 1331 } 1332 } 1333 } 1334 1335 /* 1336 * Build hash lists of net addresses and hang them off the mount point. 1337 * Called by ufs_mount() to set up the lists of export addresses. 1338 */ 1339 static int 1340 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1341 struct export_args *argp) 1342 { 1343 struct netcred *np; 1344 struct radix_node_head *rnh; 1345 int i; 1346 struct radix_node *rn; 1347 struct sockaddr *saddr, *smask = 0; 1348 struct domain *dom; 1349 int error; 1350 1351 if (argp->ex_addrlen == 0) { 1352 if (mp->mnt_flag & MNT_DEFEXPORTED) 1353 return (EPERM); 1354 np = &nep->ne_defexported; 1355 np->netc_exflags = argp->ex_flags; 1356 np->netc_anon = argp->ex_anon; 1357 np->netc_anon.cr_ref = 1; 1358 mp->mnt_flag |= MNT_DEFEXPORTED; 1359 return (0); 1360 } 1361 1362 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1363 return (EINVAL); 1364 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1365 return (EINVAL); 1366 1367 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1368 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1369 bzero((caddr_t) np, i); 1370 saddr = (struct sockaddr *) (np + 1); 1371 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1372 goto out; 1373 if (saddr->sa_len > argp->ex_addrlen) 1374 saddr->sa_len = argp->ex_addrlen; 1375 if (argp->ex_masklen) { 1376 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1377 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1378 if (error) 1379 goto out; 1380 if (smask->sa_len > argp->ex_masklen) 1381 smask->sa_len = argp->ex_masklen; 1382 } 1383 i = saddr->sa_family; 1384 if ((rnh = nep->ne_rtable[i]) == 0) { 1385 /* 1386 * Seems silly to initialize every AF when most are not used, 1387 * do so on demand here 1388 */ 1389 for (dom = domains; dom; dom = dom->dom_next) 1390 if (dom->dom_family == i && dom->dom_rtattach) { 1391 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1392 dom->dom_rtoffset); 1393 break; 1394 } 1395 if ((rnh = nep->ne_rtable[i]) == 0) { 1396 error = ENOBUFS; 1397 goto out; 1398 } 1399 } 1400 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1401 np->netc_rnodes); 1402 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1403 error = EPERM; 1404 goto out; 1405 } 1406 np->netc_exflags = argp->ex_flags; 1407 np->netc_anon = argp->ex_anon; 1408 np->netc_anon.cr_ref = 1; 1409 return (0); 1410 out: 1411 free(np, M_NETADDR); 1412 return (error); 1413 } 1414 1415 /* ARGSUSED */ 1416 static int 1417 vfs_free_netcred(struct radix_node *rn, void *w) 1418 { 1419 struct radix_node_head *rnh = (struct radix_node_head *) w; 1420 1421 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1422 free((caddr_t) rn, M_NETADDR); 1423 return (0); 1424 } 1425 1426 /* 1427 * Free the net address hash lists that are hanging off the mount points. 1428 */ 1429 static void 1430 vfs_free_addrlist(struct netexport *nep) 1431 { 1432 int i; 1433 struct radix_node_head *rnh; 1434 1435 for (i = 0; i <= AF_MAX; i++) 1436 if ((rnh = nep->ne_rtable[i])) { 1437 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1438 (caddr_t) rnh); 1439 free((caddr_t) rnh, M_RTABLE); 1440 nep->ne_rtable[i] = 0; 1441 } 1442 } 1443 1444 int 1445 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1446 { 1447 int error; 1448 1449 if (argp->ex_flags & MNT_DELEXPORT) { 1450 if (mp->mnt_flag & MNT_EXPUBLIC) { 1451 vfs_setpublicfs(NULL, NULL, NULL); 1452 mp->mnt_flag &= ~MNT_EXPUBLIC; 1453 } 1454 vfs_free_addrlist(nep); 1455 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1456 } 1457 if (argp->ex_flags & MNT_EXPORTED) { 1458 if (argp->ex_flags & MNT_EXPUBLIC) { 1459 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1460 return (error); 1461 mp->mnt_flag |= MNT_EXPUBLIC; 1462 } 1463 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1464 return (error); 1465 mp->mnt_flag |= MNT_EXPORTED; 1466 } 1467 return (0); 1468 } 1469 1470 1471 /* 1472 * Set the publicly exported filesystem (WebNFS). Currently, only 1473 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1474 */ 1475 int 1476 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1477 struct export_args *argp) 1478 { 1479 int error; 1480 struct vnode *rvp; 1481 char *cp; 1482 1483 /* 1484 * mp == NULL -> invalidate the current info, the FS is 1485 * no longer exported. May be called from either vfs_export 1486 * or unmount, so check if it hasn't already been done. 1487 */ 1488 if (mp == NULL) { 1489 if (nfs_pub.np_valid) { 1490 nfs_pub.np_valid = 0; 1491 if (nfs_pub.np_index != NULL) { 1492 FREE(nfs_pub.np_index, M_TEMP); 1493 nfs_pub.np_index = NULL; 1494 } 1495 } 1496 return (0); 1497 } 1498 1499 /* 1500 * Only one allowed at a time. 1501 */ 1502 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1503 return (EBUSY); 1504 1505 /* 1506 * Get real filehandle for root of exported FS. 1507 */ 1508 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1509 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1510 1511 if ((error = VFS_ROOT(mp, &rvp))) 1512 return (error); 1513 1514 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1515 return (error); 1516 1517 vput(rvp); 1518 1519 /* 1520 * If an indexfile was specified, pull it in. 1521 */ 1522 if (argp->ex_indexfile != NULL) { 1523 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 1524 M_WAITOK); 1525 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1526 MAXNAMLEN, (size_t *)0); 1527 if (!error) { 1528 /* 1529 * Check for illegal filenames. 1530 */ 1531 for (cp = nfs_pub.np_index; *cp; cp++) { 1532 if (*cp == '/') { 1533 error = EINVAL; 1534 break; 1535 } 1536 } 1537 } 1538 if (error) { 1539 FREE(nfs_pub.np_index, M_TEMP); 1540 return (error); 1541 } 1542 } 1543 1544 nfs_pub.np_mount = mp; 1545 nfs_pub.np_valid = 1; 1546 return (0); 1547 } 1548 1549 struct netcred * 1550 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1551 struct sockaddr *nam) 1552 { 1553 struct netcred *np; 1554 struct radix_node_head *rnh; 1555 struct sockaddr *saddr; 1556 1557 np = NULL; 1558 if (mp->mnt_flag & MNT_EXPORTED) { 1559 /* 1560 * Lookup in the export list first. 1561 */ 1562 if (nam != NULL) { 1563 saddr = nam; 1564 rnh = nep->ne_rtable[saddr->sa_family]; 1565 if (rnh != NULL) { 1566 np = (struct netcred *) 1567 (*rnh->rnh_matchaddr)((char *)saddr, 1568 rnh); 1569 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1570 np = NULL; 1571 } 1572 } 1573 /* 1574 * If no address match, use the default if it exists. 1575 */ 1576 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1577 np = &nep->ne_defexported; 1578 } 1579 return (np); 1580 } 1581 1582 /* 1583 * perform msync on all vnodes under a mount point. The mount point must 1584 * be locked. This code is also responsible for lazy-freeing unreferenced 1585 * vnodes whos VM objects no longer contain pages. 1586 * 1587 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1588 */ 1589 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1590 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1591 1592 void 1593 vfs_msync(struct mount *mp, int flags) 1594 { 1595 vmntvnodescan(mp, VMSC_REFVP, vfs_msync_scan1, vfs_msync_scan2, 1596 (void *)flags); 1597 } 1598 1599 /* 1600 * scan1 is a fast pre-check. There could be hundreds of thousands of 1601 * vnodes, we cannot afford to do anything heavy weight until we have a 1602 * fairly good indication that there is work to do. 1603 */ 1604 static 1605 int 1606 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1607 { 1608 int flags = (int)data; 1609 1610 if ((vp->v_flag & VRECLAIMED) == 0) { 1611 if (vshouldfree(vp, 0)) 1612 return(0); /* call scan2 */ 1613 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1614 (vp->v_flag & VOBJDIRTY) && 1615 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1616 return(0); /* call scan2 */ 1617 } 1618 } 1619 1620 /* 1621 * do not call scan2, continue the loop 1622 */ 1623 return(-1); 1624 } 1625 1626 static 1627 int 1628 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1629 { 1630 vm_object_t obj; 1631 int flags = (int)data; 1632 1633 if (vp->v_flag & VRECLAIMED) 1634 return(0); 1635 1636 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1637 (vp->v_flag & VOBJDIRTY) && 1638 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1639 if (VOP_GETVOBJECT(vp, &obj) == 0) { 1640 vm_object_page_clean(obj, 0, 0, 1641 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1642 } 1643 } 1644 return(0); 1645 } 1646 1647 /* 1648 * Create the VM object needed for VMIO and mmap support. This 1649 * is done for all VREG files in the system. Some filesystems might 1650 * afford the additional metadata buffering capability of the 1651 * VMIO code by making the device node be VMIO mode also. 1652 * 1653 * vp must be locked when vfs_object_create is called. 1654 */ 1655 int 1656 vfs_object_create(struct vnode *vp, struct thread *td) 1657 { 1658 return (VOP_CREATEVOBJECT(vp, td)); 1659 } 1660 1661 /* 1662 * Record a process's interest in events which might happen to 1663 * a vnode. Because poll uses the historic select-style interface 1664 * internally, this routine serves as both the ``check for any 1665 * pending events'' and the ``record my interest in future events'' 1666 * functions. (These are done together, while the lock is held, 1667 * to avoid race conditions.) 1668 */ 1669 int 1670 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 1671 { 1672 lwkt_tokref ilock; 1673 1674 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1675 if (vp->v_pollinfo.vpi_revents & events) { 1676 /* 1677 * This leaves events we are not interested 1678 * in available for the other process which 1679 * which presumably had requested them 1680 * (otherwise they would never have been 1681 * recorded). 1682 */ 1683 events &= vp->v_pollinfo.vpi_revents; 1684 vp->v_pollinfo.vpi_revents &= ~events; 1685 1686 lwkt_reltoken(&ilock); 1687 return events; 1688 } 1689 vp->v_pollinfo.vpi_events |= events; 1690 selrecord(td, &vp->v_pollinfo.vpi_selinfo); 1691 lwkt_reltoken(&ilock); 1692 return 0; 1693 } 1694 1695 /* 1696 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1697 * it is possible for us to miss an event due to race conditions, but 1698 * that condition is expected to be rare, so for the moment it is the 1699 * preferred interface. 1700 */ 1701 void 1702 vn_pollevent(struct vnode *vp, int events) 1703 { 1704 lwkt_tokref ilock; 1705 1706 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1707 if (vp->v_pollinfo.vpi_events & events) { 1708 /* 1709 * We clear vpi_events so that we don't 1710 * call selwakeup() twice if two events are 1711 * posted before the polling process(es) is 1712 * awakened. This also ensures that we take at 1713 * most one selwakeup() if the polling process 1714 * is no longer interested. However, it does 1715 * mean that only one event can be noticed at 1716 * a time. (Perhaps we should only clear those 1717 * event bits which we note?) XXX 1718 */ 1719 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1720 vp->v_pollinfo.vpi_revents |= events; 1721 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1722 } 1723 lwkt_reltoken(&ilock); 1724 } 1725 1726 /* 1727 * Wake up anyone polling on vp because it is being revoked. 1728 * This depends on dead_poll() returning POLLHUP for correct 1729 * behavior. 1730 */ 1731 void 1732 vn_pollgone(struct vnode *vp) 1733 { 1734 lwkt_tokref ilock; 1735 1736 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1737 if (vp->v_pollinfo.vpi_events) { 1738 vp->v_pollinfo.vpi_events = 0; 1739 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1740 } 1741 lwkt_reltoken(&ilock); 1742 } 1743 1744 /* 1745 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened 1746 * (or v_rdev might be NULL). 1747 */ 1748 dev_t 1749 vn_todev(struct vnode *vp) 1750 { 1751 if (vp->v_type != VBLK && vp->v_type != VCHR) 1752 return (NODEV); 1753 KKASSERT(vp->v_rdev != NULL); 1754 return (vp->v_rdev); 1755 } 1756 1757 /* 1758 * Check if vnode represents a disk device. The vnode does not need to be 1759 * opened. 1760 */ 1761 int 1762 vn_isdisk(struct vnode *vp, int *errp) 1763 { 1764 dev_t dev; 1765 1766 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1767 if (errp != NULL) 1768 *errp = ENOTBLK; 1769 return (0); 1770 } 1771 1772 if ((dev = vp->v_rdev) == NULL) 1773 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1774 if (dev == NULL || dev == NODEV) { 1775 if (errp != NULL) 1776 *errp = ENXIO; 1777 return (0); 1778 } 1779 if (dev_is_good(dev) == 0) { 1780 if (errp != NULL) 1781 *errp = ENXIO; 1782 return (0); 1783 } 1784 if ((dev_dflags(dev) & D_DISK) == 0) { 1785 if (errp != NULL) 1786 *errp = ENOTBLK; 1787 return (0); 1788 } 1789 if (errp != NULL) 1790 *errp = 0; 1791 return (1); 1792 } 1793 1794 #ifdef DEBUG_VFS_LOCKS 1795 1796 void 1797 assert_vop_locked(struct vnode *vp, const char *str) 1798 { 1799 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) { 1800 panic("%s: %p is not locked shared but should be", str, vp); 1801 } 1802 } 1803 1804 void 1805 assert_vop_unlocked(struct vnode *vp, const char *str) 1806 { 1807 if (vp && IS_LOCKING_VFS(vp)) { 1808 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) { 1809 panic("%s: %p is locked but should not be", str, vp); 1810 } 1811 } 1812 } 1813 1814 #endif 1815