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 1.114 2008/05/18 05:54:25 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/unistd.h> 68 #include <sys/vmmeter.h> 69 #include <sys/vnode.h> 70 71 #include <machine/limits.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_object.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_kern.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_page.h> 80 #include <vm/vm_pager.h> 81 #include <vm/vnode_pager.h> 82 #include <vm/vm_zone.h> 83 84 #include <sys/buf2.h> 85 #include <sys/thread2.h> 86 #include <sys/sysref2.h> 87 88 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 89 90 int numvnodes; 91 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 92 int vfs_fastdev = 1; 93 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, ""); 94 95 enum vtype iftovt_tab[16] = { 96 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 97 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 98 }; 99 int vttoif_tab[9] = { 100 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 101 S_IFSOCK, S_IFIFO, S_IFMT, 102 }; 103 104 static int reassignbufcalls; 105 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, 106 &reassignbufcalls, 0, ""); 107 static int reassignbufloops; 108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, 109 &reassignbufloops, 0, ""); 110 static int reassignbufsortgood; 111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, 112 &reassignbufsortgood, 0, ""); 113 static int reassignbufsortbad; 114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, 115 &reassignbufsortbad, 0, ""); 116 static int reassignbufmethod = 1; 117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, 118 &reassignbufmethod, 0, ""); 119 120 int nfs_mount_type = -1; 121 static struct lwkt_token spechash_token; 122 struct nfs_public nfs_pub; /* publicly exported FS */ 123 124 int desiredvnodes; 125 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 126 &desiredvnodes, 0, "Maximum number of vnodes"); 127 128 static void vfs_free_addrlist (struct netexport *nep); 129 static int vfs_free_netcred (struct radix_node *rn, void *w); 130 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 131 const struct export_args *argp); 132 133 extern int dev_ref_debug; 134 135 /* 136 * Red black tree functions 137 */ 138 static int rb_buf_compare(struct buf *b1, struct buf *b2); 139 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 140 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 141 142 static int 143 rb_buf_compare(struct buf *b1, struct buf *b2) 144 { 145 if (b1->b_loffset < b2->b_loffset) 146 return(-1); 147 if (b1->b_loffset > b2->b_loffset) 148 return(1); 149 return(0); 150 } 151 152 /* 153 * Returns non-zero if the vnode is a candidate for lazy msyncing. 154 */ 155 static __inline int 156 vshouldmsync(struct vnode *vp) 157 { 158 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0) 159 return (0); /* other holders */ 160 if (vp->v_object && 161 (vp->v_object->ref_count || vp->v_object->resident_page_count)) { 162 return (0); 163 } 164 return (1); 165 } 166 167 /* 168 * Initialize the vnode management data structures. 169 * 170 * Called from vfsinit() 171 */ 172 void 173 vfs_subr_init(void) 174 { 175 /* 176 * Desired vnodes is a result of the physical page count 177 * and the size of kernel's heap. It scales in proportion 178 * to the amount of available physical memory. This can 179 * cause trouble on 64-bit and large memory platforms. 180 */ 181 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */ 182 desiredvnodes = 183 min(maxproc + vmstats.v_page_count / 4, 184 2 * KvaSize / 185 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 186 187 lwkt_token_init(&spechash_token); 188 } 189 190 /* 191 * Knob to control the precision of file timestamps: 192 * 193 * 0 = seconds only; nanoseconds zeroed. 194 * 1 = seconds and nanoseconds, accurate within 1/HZ. 195 * 2 = seconds and nanoseconds, truncated to microseconds. 196 * >=3 = seconds and nanoseconds, maximum precision. 197 */ 198 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 199 200 static int timestamp_precision = TSP_SEC; 201 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 202 ×tamp_precision, 0, ""); 203 204 /* 205 * Get a current timestamp. 206 */ 207 void 208 vfs_timestamp(struct timespec *tsp) 209 { 210 struct timeval tv; 211 212 switch (timestamp_precision) { 213 case TSP_SEC: 214 tsp->tv_sec = time_second; 215 tsp->tv_nsec = 0; 216 break; 217 case TSP_HZ: 218 getnanotime(tsp); 219 break; 220 case TSP_USEC: 221 microtime(&tv); 222 TIMEVAL_TO_TIMESPEC(&tv, tsp); 223 break; 224 case TSP_NSEC: 225 default: 226 nanotime(tsp); 227 break; 228 } 229 } 230 231 /* 232 * Set vnode attributes to VNOVAL 233 */ 234 void 235 vattr_null(struct vattr *vap) 236 { 237 vap->va_type = VNON; 238 vap->va_size = VNOVAL; 239 vap->va_bytes = VNOVAL; 240 vap->va_mode = VNOVAL; 241 vap->va_nlink = VNOVAL; 242 vap->va_uid = VNOVAL; 243 vap->va_gid = VNOVAL; 244 vap->va_fsid = VNOVAL; 245 vap->va_fileid = VNOVAL; 246 vap->va_blocksize = VNOVAL; 247 vap->va_rmajor = VNOVAL; 248 vap->va_rminor = VNOVAL; 249 vap->va_atime.tv_sec = VNOVAL; 250 vap->va_atime.tv_nsec = VNOVAL; 251 vap->va_mtime.tv_sec = VNOVAL; 252 vap->va_mtime.tv_nsec = VNOVAL; 253 vap->va_ctime.tv_sec = VNOVAL; 254 vap->va_ctime.tv_nsec = VNOVAL; 255 vap->va_flags = VNOVAL; 256 vap->va_gen = VNOVAL; 257 vap->va_vaflags = 0; 258 vap->va_fsmid = VNOVAL; 259 /* va_*_uuid fields are only valid if related flags are set */ 260 } 261 262 /* 263 * Flush out and invalidate all buffers associated with a vnode. 264 * 265 * vp must be locked. 266 */ 267 static int vinvalbuf_bp(struct buf *bp, void *data); 268 269 struct vinvalbuf_bp_info { 270 struct vnode *vp; 271 int slptimeo; 272 int lkflags; 273 int flags; 274 }; 275 276 void 277 vupdatefsmid(struct vnode *vp) 278 { 279 atomic_set_int(&vp->v_flag, VFSMID); 280 } 281 282 int 283 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 284 { 285 struct vinvalbuf_bp_info info; 286 int error; 287 vm_object_t object; 288 289 /* 290 * If we are being asked to save, call fsync to ensure that the inode 291 * is updated. 292 */ 293 if (flags & V_SAVE) { 294 crit_enter(); 295 while (vp->v_track_write.bk_active) { 296 vp->v_track_write.bk_waitflag = 1; 297 error = tsleep(&vp->v_track_write, slpflag, 298 "vinvlbuf", slptimeo); 299 if (error) { 300 crit_exit(); 301 return (error); 302 } 303 } 304 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 305 crit_exit(); 306 if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0) 307 return (error); 308 crit_enter(); 309 if (vp->v_track_write.bk_active > 0 || 310 !RB_EMPTY(&vp->v_rbdirty_tree)) 311 panic("vinvalbuf: dirty bufs"); 312 } 313 crit_exit(); 314 } 315 crit_enter(); 316 info.slptimeo = slptimeo; 317 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 318 if (slpflag & PCATCH) 319 info.lkflags |= LK_PCATCH; 320 info.flags = flags; 321 info.vp = vp; 322 323 /* 324 * Flush the buffer cache until nothing is left. 325 */ 326 while (!RB_EMPTY(&vp->v_rbclean_tree) || 327 !RB_EMPTY(&vp->v_rbdirty_tree)) { 328 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 329 vinvalbuf_bp, &info); 330 if (error == 0) { 331 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 332 vinvalbuf_bp, &info); 333 } 334 } 335 336 /* 337 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 338 * have write I/O in-progress but if there is a VM object then the 339 * VM object can also have read-I/O in-progress. 340 */ 341 do { 342 while (vp->v_track_write.bk_active > 0) { 343 vp->v_track_write.bk_waitflag = 1; 344 tsleep(&vp->v_track_write, 0, "vnvlbv", 0); 345 } 346 if ((object = vp->v_object) != NULL) { 347 while (object->paging_in_progress) 348 vm_object_pip_sleep(object, "vnvlbx"); 349 } 350 } while (vp->v_track_write.bk_active > 0); 351 352 crit_exit(); 353 354 /* 355 * Destroy the copy in the VM cache, too. 356 */ 357 if ((object = vp->v_object) != NULL) { 358 vm_object_page_remove(object, 0, 0, 359 (flags & V_SAVE) ? TRUE : FALSE); 360 } 361 362 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 363 panic("vinvalbuf: flush failed"); 364 if (!RB_EMPTY(&vp->v_rbhash_tree)) 365 panic("vinvalbuf: flush failed, buffers still present"); 366 return (0); 367 } 368 369 static int 370 vinvalbuf_bp(struct buf *bp, void *data) 371 { 372 struct vinvalbuf_bp_info *info = data; 373 int error; 374 375 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 376 error = BUF_TIMELOCK(bp, info->lkflags, 377 "vinvalbuf", info->slptimeo); 378 if (error == 0) { 379 BUF_UNLOCK(bp); 380 error = ENOLCK; 381 } 382 if (error == ENOLCK) 383 return(0); 384 return (-error); 385 } 386 387 KKASSERT(bp->b_vp == info->vp); 388 389 /* 390 * XXX Since there are no node locks for NFS, I 391 * believe there is a slight chance that a delayed 392 * write will occur while sleeping just above, so 393 * check for it. Note that vfs_bio_awrite expects 394 * buffers to reside on a queue, while bwrite() and 395 * brelse() do not. 396 * 397 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 398 * check. This code will write out the buffer, period. 399 */ 400 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 401 (info->flags & V_SAVE)) { 402 if (bp->b_vp == info->vp) { 403 if (bp->b_flags & B_CLUSTEROK) { 404 vfs_bio_awrite(bp); 405 } else { 406 bremfree(bp); 407 bp->b_flags |= B_ASYNC; 408 bwrite(bp); 409 } 410 } else { 411 bremfree(bp); 412 bwrite(bp); 413 } 414 } else if (info->flags & V_SAVE) { 415 /* 416 * Cannot set B_NOCACHE on a clean buffer as this will 417 * destroy the VM backing store which might actually 418 * be dirty (and unsynchronized). 419 */ 420 bremfree(bp); 421 bp->b_flags |= (B_INVAL | B_RELBUF); 422 bp->b_flags &= ~B_ASYNC; 423 brelse(bp); 424 } else { 425 bremfree(bp); 426 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 427 bp->b_flags &= ~B_ASYNC; 428 brelse(bp); 429 } 430 return(0); 431 } 432 433 /* 434 * Truncate a file's buffer and pages to a specified length. This 435 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 436 * sync activity. 437 * 438 * The vnode must be locked. 439 */ 440 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 441 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 442 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 443 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 444 445 int 446 vtruncbuf(struct vnode *vp, off_t length, int blksize) 447 { 448 off_t truncloffset; 449 int count; 450 const char *filename; 451 452 /* 453 * Round up to the *next* block, then destroy the buffers in question. 454 * Since we are only removing some of the buffers we must rely on the 455 * scan count to determine whether a loop is necessary. 456 */ 457 if ((count = (int)(length % blksize)) != 0) 458 truncloffset = length + (blksize - count); 459 else 460 truncloffset = length; 461 462 crit_enter(); 463 do { 464 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 465 vtruncbuf_bp_trunc_cmp, 466 vtruncbuf_bp_trunc, &truncloffset); 467 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 468 vtruncbuf_bp_trunc_cmp, 469 vtruncbuf_bp_trunc, &truncloffset); 470 } while(count); 471 472 /* 473 * For safety, fsync any remaining metadata if the file is not being 474 * truncated to 0. Since the metadata does not represent the entire 475 * dirty list we have to rely on the hit count to ensure that we get 476 * all of it. 477 */ 478 if (length > 0) { 479 do { 480 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 481 vtruncbuf_bp_metasync_cmp, 482 vtruncbuf_bp_metasync, vp); 483 } while (count); 484 } 485 486 /* 487 * Clean out any left over VM backing store. 488 */ 489 crit_exit(); 490 491 vnode_pager_setsize(vp, length); 492 493 crit_enter(); 494 495 /* 496 * It is possible to have in-progress I/O from buffers that were 497 * not part of the truncation. This should not happen if we 498 * are truncating to 0-length. 499 */ 500 filename = TAILQ_FIRST(&vp->v_namecache) ? 501 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 502 503 while ((count = vp->v_track_write.bk_active) > 0) { 504 vp->v_track_write.bk_waitflag = 1; 505 tsleep(&vp->v_track_write, 0, "vbtrunc", 0); 506 if (length == 0) { 507 kprintf("Warning: vtruncbuf(): Had to wait for " 508 "%d buffer I/Os to finish in %s\n", 509 count, filename); 510 } 511 } 512 513 /* 514 * Make sure no buffers were instantiated while we were trying 515 * to clean out the remaining VM pages. This could occur due 516 * to busy dirty VM pages being flushed out to disk. 517 */ 518 do { 519 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 520 vtruncbuf_bp_trunc_cmp, 521 vtruncbuf_bp_trunc, &truncloffset); 522 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 523 vtruncbuf_bp_trunc_cmp, 524 vtruncbuf_bp_trunc, &truncloffset); 525 if (count) { 526 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 527 "left over buffers in %s\n", count, filename); 528 } 529 } while(count); 530 531 crit_exit(); 532 533 return (0); 534 } 535 536 /* 537 * The callback buffer is beyond the new file EOF and must be destroyed. 538 * Note that the compare function must conform to the RB_SCAN's requirements. 539 */ 540 static 541 int 542 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 543 { 544 if (bp->b_loffset >= *(off_t *)data) 545 return(0); 546 return(-1); 547 } 548 549 static 550 int 551 vtruncbuf_bp_trunc(struct buf *bp, void *data) 552 { 553 /* 554 * Do not try to use a buffer we cannot immediately lock, but sleep 555 * anyway to prevent a livelock. The code will loop until all buffers 556 * can be acted upon. 557 */ 558 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 559 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 560 BUF_UNLOCK(bp); 561 } else { 562 bremfree(bp); 563 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 564 bp->b_flags &= ~B_ASYNC; 565 brelse(bp); 566 } 567 return(1); 568 } 569 570 /* 571 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 572 * blocks (with a negative loffset) are scanned. 573 * Note that the compare function must conform to the RB_SCAN's requirements. 574 */ 575 static int 576 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 577 { 578 if (bp->b_loffset < 0) 579 return(0); 580 return(1); 581 } 582 583 static int 584 vtruncbuf_bp_metasync(struct buf *bp, void *data) 585 { 586 struct vnode *vp = data; 587 588 if (bp->b_flags & B_DELWRI) { 589 /* 590 * Do not try to use a buffer we cannot immediately lock, 591 * but sleep anyway to prevent a livelock. The code will 592 * loop until all buffers can be acted upon. 593 */ 594 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 595 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 596 BUF_UNLOCK(bp); 597 } else { 598 bremfree(bp); 599 if (bp->b_vp == vp) { 600 bp->b_flags |= B_ASYNC; 601 } else { 602 bp->b_flags &= ~B_ASYNC; 603 } 604 bwrite(bp); 605 } 606 return(1); 607 } else { 608 return(0); 609 } 610 } 611 612 /* 613 * vfsync - implements a multipass fsync on a file which understands 614 * dependancies and meta-data. The passed vnode must be locked. The 615 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 616 * 617 * When fsyncing data asynchronously just do one consolidated pass starting 618 * with the most negative block number. This may not get all the data due 619 * to dependancies. 620 * 621 * When fsyncing data synchronously do a data pass, then a metadata pass, 622 * then do additional data+metadata passes to try to get all the data out. 623 */ 624 static int vfsync_wait_output(struct vnode *vp, 625 int (*waitoutput)(struct vnode *, struct thread *)); 626 static int vfsync_data_only_cmp(struct buf *bp, void *data); 627 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 628 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 629 static int vfsync_bp(struct buf *bp, void *data); 630 631 struct vfsync_info { 632 struct vnode *vp; 633 int synchronous; 634 int syncdeps; 635 int lazycount; 636 int lazylimit; 637 int skippedbufs; 638 int (*checkdef)(struct buf *); 639 }; 640 641 int 642 vfsync(struct vnode *vp, int waitfor, int passes, 643 int (*checkdef)(struct buf *), 644 int (*waitoutput)(struct vnode *, struct thread *)) 645 { 646 struct vfsync_info info; 647 int error; 648 649 bzero(&info, sizeof(info)); 650 info.vp = vp; 651 if ((info.checkdef = checkdef) == NULL) 652 info.syncdeps = 1; 653 654 crit_enter_id("vfsync"); 655 656 switch(waitfor) { 657 case MNT_LAZY: 658 /* 659 * Lazy (filesystem syncer typ) Asynchronous plus limit the 660 * number of data (not meta) pages we try to flush to 1MB. 661 * A non-zero return means that lazy limit was reached. 662 */ 663 info.lazylimit = 1024 * 1024; 664 info.syncdeps = 1; 665 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 666 vfsync_lazy_range_cmp, vfsync_bp, &info); 667 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 668 vfsync_meta_only_cmp, vfsync_bp, &info); 669 if (error == 0) 670 vp->v_lazyw = 0; 671 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 672 vn_syncer_add_to_worklist(vp, 1); 673 error = 0; 674 break; 675 case MNT_NOWAIT: 676 /* 677 * Asynchronous. Do a data-only pass and a meta-only pass. 678 */ 679 info.syncdeps = 1; 680 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 681 vfsync_bp, &info); 682 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 683 vfsync_bp, &info); 684 error = 0; 685 break; 686 default: 687 /* 688 * Synchronous. Do a data-only pass, then a meta-data+data 689 * pass, then additional integrated passes to try to get 690 * all the dependancies flushed. 691 */ 692 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 693 vfsync_bp, &info); 694 error = vfsync_wait_output(vp, waitoutput); 695 if (error == 0) { 696 info.skippedbufs = 0; 697 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 698 vfsync_bp, &info); 699 error = vfsync_wait_output(vp, waitoutput); 700 if (info.skippedbufs) 701 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs); 702 } 703 while (error == 0 && passes > 0 && 704 !RB_EMPTY(&vp->v_rbdirty_tree)) { 705 if (--passes == 0) { 706 info.synchronous = 1; 707 info.syncdeps = 1; 708 } 709 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 710 vfsync_bp, &info); 711 if (error < 0) 712 error = -error; 713 info.syncdeps = 1; 714 if (error == 0) 715 error = vfsync_wait_output(vp, waitoutput); 716 } 717 break; 718 } 719 crit_exit_id("vfsync"); 720 return(error); 721 } 722 723 static int 724 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 725 { 726 int error = 0; 727 728 while (vp->v_track_write.bk_active) { 729 vp->v_track_write.bk_waitflag = 1; 730 tsleep(&vp->v_track_write, 0, "fsfsn", 0); 731 } 732 if (waitoutput) 733 error = waitoutput(vp, curthread); 734 return(error); 735 } 736 737 static int 738 vfsync_data_only_cmp(struct buf *bp, void *data) 739 { 740 if (bp->b_loffset < 0) 741 return(-1); 742 return(0); 743 } 744 745 static int 746 vfsync_meta_only_cmp(struct buf *bp, void *data) 747 { 748 if (bp->b_loffset < 0) 749 return(0); 750 return(1); 751 } 752 753 static int 754 vfsync_lazy_range_cmp(struct buf *bp, void *data) 755 { 756 struct vfsync_info *info = data; 757 if (bp->b_loffset < info->vp->v_lazyw) 758 return(-1); 759 return(0); 760 } 761 762 static int 763 vfsync_bp(struct buf *bp, void *data) 764 { 765 struct vfsync_info *info = data; 766 struct vnode *vp = info->vp; 767 int error; 768 769 /* 770 * if syncdeps is not set we do not try to write buffers which have 771 * dependancies. 772 */ 773 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 774 return(0); 775 776 /* 777 * Ignore buffers that we cannot immediately lock. XXX 778 */ 779 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 780 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp); 781 ++info->skippedbufs; 782 return(0); 783 } 784 if ((bp->b_flags & B_DELWRI) == 0) 785 panic("vfsync_bp: buffer not dirty"); 786 if (vp != bp->b_vp) 787 panic("vfsync_bp: buffer vp mismatch"); 788 789 /* 790 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 791 * has been written but an additional handshake with the device 792 * is required before we can dispose of the buffer. We have no idea 793 * how to do this so we have to skip these buffers. 794 */ 795 if (bp->b_flags & B_NEEDCOMMIT) { 796 BUF_UNLOCK(bp); 797 return(0); 798 } 799 800 /* 801 * Ask bioops if it is ok to sync 802 */ 803 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 804 bremfree(bp); 805 brelse(bp); 806 return(0); 807 } 808 809 if (info->synchronous) { 810 /* 811 * Synchronous flushing. An error may be returned. 812 */ 813 bremfree(bp); 814 crit_exit_id("vfsync"); 815 error = bwrite(bp); 816 crit_enter_id("vfsync"); 817 } else { 818 /* 819 * Asynchronous flushing. A negative return value simply 820 * stops the scan and is not considered an error. We use 821 * this to support limited MNT_LAZY flushes. 822 */ 823 vp->v_lazyw = bp->b_loffset; 824 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 825 info->lazycount += vfs_bio_awrite(bp); 826 } else { 827 info->lazycount += bp->b_bufsize; 828 bremfree(bp); 829 crit_exit_id("vfsync"); 830 bawrite(bp); 831 crit_enter_id("vfsync"); 832 } 833 if (info->lazylimit && info->lazycount >= info->lazylimit) 834 error = 1; 835 else 836 error = 0; 837 } 838 return(-error); 839 } 840 841 /* 842 * Associate a buffer with a vnode. 843 */ 844 void 845 bgetvp(struct vnode *vp, struct buf *bp) 846 { 847 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 848 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 849 850 vhold(vp); 851 /* 852 * Insert onto list for new vnode. 853 */ 854 crit_enter(); 855 bp->b_vp = vp; 856 bp->b_flags |= B_HASHED; 857 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) 858 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 859 860 bp->b_flags |= B_VNCLEAN; 861 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 862 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 863 crit_exit(); 864 } 865 866 /* 867 * Disassociate a buffer from a vnode. 868 */ 869 void 870 brelvp(struct buf *bp) 871 { 872 struct vnode *vp; 873 874 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 875 876 /* 877 * Delete from old vnode list, if on one. 878 */ 879 vp = bp->b_vp; 880 crit_enter(); 881 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 882 if (bp->b_flags & B_VNDIRTY) 883 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 884 else 885 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 886 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 887 } 888 if (bp->b_flags & B_HASHED) { 889 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 890 bp->b_flags &= ~B_HASHED; 891 } 892 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 893 vp->v_flag &= ~VONWORKLST; 894 LIST_REMOVE(vp, v_synclist); 895 } 896 crit_exit(); 897 bp->b_vp = NULL; 898 vdrop(vp); 899 } 900 901 /* 902 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 903 * This routine is called when the state of the B_DELWRI bit is changed. 904 */ 905 void 906 reassignbuf(struct buf *bp) 907 { 908 struct vnode *vp = bp->b_vp; 909 int delay; 910 911 KKASSERT(vp != NULL); 912 ++reassignbufcalls; 913 914 /* 915 * B_PAGING flagged buffers cannot be reassigned because their vp 916 * is not fully linked in. 917 */ 918 if (bp->b_flags & B_PAGING) 919 panic("cannot reassign paging buffer"); 920 921 crit_enter(); 922 if (bp->b_flags & B_DELWRI) { 923 /* 924 * Move to the dirty list, add the vnode to the worklist 925 */ 926 if (bp->b_flags & B_VNCLEAN) { 927 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 928 bp->b_flags &= ~B_VNCLEAN; 929 } 930 if ((bp->b_flags & B_VNDIRTY) == 0) { 931 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 932 panic("reassignbuf: dup lblk vp %p bp %p", 933 vp, bp); 934 } 935 bp->b_flags |= B_VNDIRTY; 936 } 937 if ((vp->v_flag & VONWORKLST) == 0) { 938 switch (vp->v_type) { 939 case VDIR: 940 delay = dirdelay; 941 break; 942 case VCHR: 943 case VBLK: 944 if (vp->v_rdev && 945 vp->v_rdev->si_mountpoint != NULL) { 946 delay = metadelay; 947 break; 948 } 949 /* fall through */ 950 default: 951 delay = filedelay; 952 } 953 vn_syncer_add_to_worklist(vp, delay); 954 } 955 } else { 956 /* 957 * Move to the clean list, remove the vnode from the worklist 958 * if no dirty blocks remain. 959 */ 960 if (bp->b_flags & B_VNDIRTY) { 961 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 962 bp->b_flags &= ~B_VNDIRTY; 963 } 964 if ((bp->b_flags & B_VNCLEAN) == 0) { 965 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 966 panic("reassignbuf: dup lblk vp %p bp %p", 967 vp, bp); 968 } 969 bp->b_flags |= B_VNCLEAN; 970 } 971 if ((vp->v_flag & VONWORKLST) && 972 RB_EMPTY(&vp->v_rbdirty_tree)) { 973 vp->v_flag &= ~VONWORKLST; 974 LIST_REMOVE(vp, v_synclist); 975 } 976 } 977 crit_exit(); 978 } 979 980 /* 981 * Create a vnode for a block device. 982 * Used for mounting the root file system. 983 */ 984 int 985 bdevvp(cdev_t dev, struct vnode **vpp) 986 { 987 struct vnode *vp; 988 struct vnode *nvp; 989 int error; 990 991 if (dev == NULL) { 992 *vpp = NULLVP; 993 return (ENXIO); 994 } 995 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0); 996 if (error) { 997 *vpp = NULLVP; 998 return (error); 999 } 1000 vp = nvp; 1001 vp->v_type = VCHR; 1002 vp->v_umajor = dev->si_umajor; 1003 vp->v_uminor = dev->si_uminor; 1004 vx_unlock(vp); 1005 *vpp = vp; 1006 return (0); 1007 } 1008 1009 int 1010 v_associate_rdev(struct vnode *vp, cdev_t dev) 1011 { 1012 lwkt_tokref ilock; 1013 1014 if (dev == NULL) 1015 return(ENXIO); 1016 if (dev_is_good(dev) == 0) 1017 return(ENXIO); 1018 KKASSERT(vp->v_rdev == NULL); 1019 if (dev_ref_debug) 1020 kprintf("Z1"); 1021 vp->v_rdev = reference_dev(dev); 1022 lwkt_gettoken(&ilock, &spechash_token); 1023 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1024 lwkt_reltoken(&ilock); 1025 return(0); 1026 } 1027 1028 void 1029 v_release_rdev(struct vnode *vp) 1030 { 1031 lwkt_tokref ilock; 1032 cdev_t dev; 1033 1034 if ((dev = vp->v_rdev) != NULL) { 1035 lwkt_gettoken(&ilock, &spechash_token); 1036 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1037 vp->v_rdev = NULL; 1038 release_dev(dev); 1039 lwkt_reltoken(&ilock); 1040 } 1041 } 1042 1043 /* 1044 * Add a vnode to the alias list hung off the cdev_t. We only associate 1045 * the device number with the vnode. The actual device is not associated 1046 * until the vnode is opened (usually in spec_open()), and will be 1047 * disassociated on last close. 1048 */ 1049 void 1050 addaliasu(struct vnode *nvp, int x, int y) 1051 { 1052 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1053 panic("addaliasu on non-special vnode"); 1054 nvp->v_umajor = x; 1055 nvp->v_uminor = y; 1056 } 1057 1058 /* 1059 * Disassociate a vnode from its underlying filesystem. 1060 * 1061 * The vnode must be VX locked and referenced. In all normal situations 1062 * there are no active references. If vclean_vxlocked() is called while 1063 * there are active references, the vnode is being ripped out and we have 1064 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1065 */ 1066 void 1067 vclean_vxlocked(struct vnode *vp, int flags) 1068 { 1069 int active; 1070 int n; 1071 vm_object_t object; 1072 1073 /* 1074 * If the vnode has already been reclaimed we have nothing to do. 1075 */ 1076 if (vp->v_flag & VRECLAIMED) 1077 return; 1078 vp->v_flag |= VRECLAIMED; 1079 1080 /* 1081 * Scrap the vfs cache 1082 */ 1083 while (cache_inval_vp(vp, 0) != 0) { 1084 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1085 tsleep(vp, 0, "vclninv", 2); 1086 } 1087 1088 /* 1089 * Check to see if the vnode is in use. If so we have to reference it 1090 * before we clean it out so that its count cannot fall to zero and 1091 * generate a race against ourselves to recycle it. 1092 */ 1093 active = sysref_isactive(&vp->v_sysref); 1094 1095 /* 1096 * Clean out any buffers associated with the vnode and destroy its 1097 * object, if it has one. 1098 */ 1099 vinvalbuf(vp, V_SAVE, 0, 0); 1100 1101 /* 1102 * If purging an active vnode (typically during a forced unmount 1103 * or reboot), it must be closed and deactivated before being 1104 * reclaimed. This isn't really all that safe, but what can 1105 * we do? XXX. 1106 * 1107 * Note that neither of these routines unlocks the vnode. 1108 */ 1109 if (active && (flags & DOCLOSE)) { 1110 while ((n = vp->v_opencount) != 0) { 1111 if (vp->v_writecount) 1112 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1113 else 1114 VOP_CLOSE(vp, FNONBLOCK); 1115 if (vp->v_opencount == n) { 1116 kprintf("Warning: unable to force-close" 1117 " vnode %p\n", vp); 1118 break; 1119 } 1120 } 1121 } 1122 1123 /* 1124 * If the vnode has not been deactivated, deactivated it. Deactivation 1125 * can create new buffers and VM pages so we have to call vinvalbuf() 1126 * again to make sure they all get flushed. 1127 * 1128 * This can occur if a file with a link count of 0 needs to be 1129 * truncated. 1130 */ 1131 if ((vp->v_flag & VINACTIVE) == 0) { 1132 vp->v_flag |= VINACTIVE; 1133 VOP_INACTIVE(vp); 1134 vinvalbuf(vp, V_SAVE, 0, 0); 1135 } 1136 1137 /* 1138 * If the vnode has an object, destroy it. 1139 */ 1140 if ((object = vp->v_object) != NULL) { 1141 if (object->ref_count == 0) { 1142 if ((object->flags & OBJ_DEAD) == 0) 1143 vm_object_terminate(object); 1144 } else { 1145 vm_pager_deallocate(object); 1146 } 1147 vp->v_flag &= ~VOBJBUF; 1148 } 1149 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1150 1151 /* 1152 * Reclaim the vnode. 1153 */ 1154 if (VOP_RECLAIM(vp)) 1155 panic("vclean: cannot reclaim"); 1156 1157 /* 1158 * Done with purge, notify sleepers of the grim news. 1159 */ 1160 vp->v_ops = &dead_vnode_vops_p; 1161 vn_pollgone(vp); 1162 vp->v_tag = VT_NON; 1163 1164 /* 1165 * If we are destroying an active vnode, reactivate it now that 1166 * we have reassociated it with deadfs. This prevents the system 1167 * from crashing on the vnode due to it being unexpectedly marked 1168 * as inactive or reclaimed. 1169 */ 1170 if (active && (flags & DOCLOSE)) { 1171 vp->v_flag &= ~(VINACTIVE|VRECLAIMED); 1172 } 1173 } 1174 1175 /* 1176 * Eliminate all activity associated with the requested vnode 1177 * and with all vnodes aliased to the requested vnode. 1178 * 1179 * The vnode must be referenced and vx_lock()'d 1180 * 1181 * revoke { struct vnode *a_vp, int a_flags } 1182 */ 1183 int 1184 vop_stdrevoke(struct vop_revoke_args *ap) 1185 { 1186 struct vnode *vp, *vq; 1187 lwkt_tokref ilock; 1188 cdev_t dev; 1189 1190 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1191 1192 vp = ap->a_vp; 1193 1194 /* 1195 * If the vnode is already dead don't try to revoke it 1196 */ 1197 if (vp->v_flag & VRECLAIMED) 1198 return (0); 1199 1200 /* 1201 * If the vnode has a device association, scrap all vnodes associated 1202 * with the device. Don't let the device disappear on us while we 1203 * are scrapping the vnodes. 1204 * 1205 * The passed vp will probably show up in the list, do not VX lock 1206 * it twice! 1207 */ 1208 if (vp->v_type != VCHR) 1209 return(0); 1210 if ((dev = vp->v_rdev) == NULL) { 1211 if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL) 1212 return(0); 1213 } 1214 reference_dev(dev); 1215 lwkt_gettoken(&ilock, &spechash_token); 1216 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1217 if (vp != vq) 1218 vx_get(vq); 1219 if (vq == SLIST_FIRST(&dev->si_hlist)) 1220 vgone_vxlocked(vq); 1221 if (vp != vq) 1222 vx_put(vq); 1223 } 1224 lwkt_reltoken(&ilock); 1225 release_dev(dev); 1226 return (0); 1227 } 1228 1229 /* 1230 * This is called when the object underlying a vnode is being destroyed, 1231 * such as in a remove(). Try to recycle the vnode immediately if the 1232 * only active reference is our reference. 1233 * 1234 * Directory vnodes in the namecache with children cannot be immediately 1235 * recycled because numerous VOP_N*() ops require them to be stable. 1236 */ 1237 int 1238 vrecycle(struct vnode *vp) 1239 { 1240 if (vp->v_sysref.refcnt <= 1) { 1241 if (cache_inval_vp_nonblock(vp)) 1242 return(0); 1243 vgone_vxlocked(vp); 1244 return (1); 1245 } 1246 return (0); 1247 } 1248 1249 /* 1250 * Return the maximum I/O size allowed for strategy calls on VP. 1251 * 1252 * If vp is VCHR or VBLK we dive the device, otherwise we use 1253 * the vp's mount info. 1254 */ 1255 int 1256 vmaxiosize(struct vnode *vp) 1257 { 1258 if (vp->v_type == VBLK || vp->v_type == VCHR) { 1259 return(vp->v_rdev->si_iosize_max); 1260 } else { 1261 return(vp->v_mount->mnt_iosize_max); 1262 } 1263 } 1264 1265 /* 1266 * Eliminate all activity associated with a vnode in preparation for reuse. 1267 * 1268 * The vnode must be VX locked and refd and will remain VX locked and refd 1269 * on return. This routine may be called with the vnode in any state, as 1270 * long as it is VX locked. The vnode will be cleaned out and marked 1271 * VRECLAIMED but will not actually be reused until all existing refs and 1272 * holds go away. 1273 * 1274 * NOTE: This routine may be called on a vnode which has not yet been 1275 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1276 * already been reclaimed. 1277 * 1278 * This routine is not responsible for placing us back on the freelist. 1279 * Instead, it happens automatically when the caller releases the VX lock 1280 * (assuming there aren't any other references). 1281 */ 1282 1283 void 1284 vgone_vxlocked(struct vnode *vp) 1285 { 1286 /* 1287 * assert that the VX lock is held. This is an absolute requirement 1288 * now for vgone_vxlocked() to be called. 1289 */ 1290 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1291 1292 /* 1293 * Clean out the filesystem specific data and set the VRECLAIMED 1294 * bit. Also deactivate the vnode if necessary. 1295 */ 1296 vclean_vxlocked(vp, DOCLOSE); 1297 1298 /* 1299 * Delete from old mount point vnode list, if on one. 1300 */ 1301 if (vp->v_mount != NULL) 1302 insmntque(vp, NULL); 1303 1304 /* 1305 * If special device, remove it from special device alias list 1306 * if it is on one. This should normally only occur if a vnode is 1307 * being revoked as the device should otherwise have been released 1308 * naturally. 1309 */ 1310 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1311 v_release_rdev(vp); 1312 } 1313 1314 /* 1315 * Set us to VBAD 1316 */ 1317 vp->v_type = VBAD; 1318 } 1319 1320 /* 1321 * Lookup a vnode by device number. 1322 */ 1323 int 1324 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1325 { 1326 lwkt_tokref ilock; 1327 struct vnode *vp; 1328 1329 lwkt_gettoken(&ilock, &spechash_token); 1330 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1331 if (type == vp->v_type) { 1332 *vpp = vp; 1333 lwkt_reltoken(&ilock); 1334 return (1); 1335 } 1336 } 1337 lwkt_reltoken(&ilock); 1338 return (0); 1339 } 1340 1341 /* 1342 * Calculate the total number of references to a special device. This 1343 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1344 * an overloaded field. Since udev2dev can now return NULL, we have 1345 * to check for a NULL v_rdev. 1346 */ 1347 int 1348 count_dev(cdev_t dev) 1349 { 1350 lwkt_tokref ilock; 1351 struct vnode *vp; 1352 int count = 0; 1353 1354 if (SLIST_FIRST(&dev->si_hlist)) { 1355 lwkt_gettoken(&ilock, &spechash_token); 1356 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1357 if (vp->v_sysref.refcnt > 0) 1358 count += vp->v_sysref.refcnt; 1359 } 1360 lwkt_reltoken(&ilock); 1361 } 1362 return(count); 1363 } 1364 1365 int 1366 count_udev(int x, int y) 1367 { 1368 cdev_t dev; 1369 1370 if ((dev = get_dev(x, y)) == NULL) 1371 return(0); 1372 return(count_dev(dev)); 1373 } 1374 1375 int 1376 vcount(struct vnode *vp) 1377 { 1378 if (vp->v_rdev == NULL) 1379 return(0); 1380 return(count_dev(vp->v_rdev)); 1381 } 1382 1383 /* 1384 * Initialize VMIO for a vnode. This routine MUST be called before a 1385 * VFS can issue buffer cache ops on a vnode. It is typically called 1386 * when a vnode is initialized from its inode. 1387 */ 1388 int 1389 vinitvmio(struct vnode *vp, off_t filesize) 1390 { 1391 vm_object_t object; 1392 int error = 0; 1393 1394 retry: 1395 if ((object = vp->v_object) == NULL) { 1396 object = vnode_pager_alloc(vp, filesize, 0, 0); 1397 /* 1398 * Dereference the reference we just created. This assumes 1399 * that the object is associated with the vp. 1400 */ 1401 object->ref_count--; 1402 vrele(vp); 1403 } else { 1404 if (object->flags & OBJ_DEAD) { 1405 vn_unlock(vp); 1406 vm_object_dead_sleep(object, "vodead"); 1407 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1408 goto retry; 1409 } 1410 } 1411 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1412 vp->v_flag |= VOBJBUF; 1413 return (error); 1414 } 1415 1416 1417 /* 1418 * Print out a description of a vnode. 1419 */ 1420 static char *typename[] = 1421 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1422 1423 void 1424 vprint(char *label, struct vnode *vp) 1425 { 1426 char buf[96]; 1427 1428 if (label != NULL) 1429 kprintf("%s: %p: ", label, (void *)vp); 1430 else 1431 kprintf("%p: ", (void *)vp); 1432 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1433 typename[vp->v_type], 1434 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1435 buf[0] = '\0'; 1436 if (vp->v_flag & VROOT) 1437 strcat(buf, "|VROOT"); 1438 if (vp->v_flag & VTEXT) 1439 strcat(buf, "|VTEXT"); 1440 if (vp->v_flag & VSYSTEM) 1441 strcat(buf, "|VSYSTEM"); 1442 if (vp->v_flag & VFREE) 1443 strcat(buf, "|VFREE"); 1444 if (vp->v_flag & VOBJBUF) 1445 strcat(buf, "|VOBJBUF"); 1446 if (buf[0] != '\0') 1447 kprintf(" flags (%s)", &buf[1]); 1448 if (vp->v_data == NULL) { 1449 kprintf("\n"); 1450 } else { 1451 kprintf("\n\t"); 1452 VOP_PRINT(vp); 1453 } 1454 } 1455 1456 #ifdef DDB 1457 #include <ddb/ddb.h> 1458 1459 static int db_show_locked_vnodes(struct mount *mp, void *data); 1460 1461 /* 1462 * List all of the locked vnodes in the system. 1463 * Called when debugging the kernel. 1464 */ 1465 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1466 { 1467 kprintf("Locked vnodes\n"); 1468 mountlist_scan(db_show_locked_vnodes, NULL, 1469 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1470 } 1471 1472 static int 1473 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1474 { 1475 struct vnode *vp; 1476 1477 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1478 if (vn_islocked(vp)) 1479 vprint((char *)0, vp); 1480 } 1481 return(0); 1482 } 1483 #endif 1484 1485 /* 1486 * Top level filesystem related information gathering. 1487 */ 1488 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1489 1490 static int 1491 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1492 { 1493 int *name = (int *)arg1 - 1; /* XXX */ 1494 u_int namelen = arg2 + 1; /* XXX */ 1495 struct vfsconf *vfsp; 1496 1497 #if 1 || defined(COMPAT_PRELITE2) 1498 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1499 if (namelen == 1) 1500 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1501 #endif 1502 1503 #ifdef notyet 1504 /* all sysctl names at this level are at least name and field */ 1505 if (namelen < 2) 1506 return (ENOTDIR); /* overloaded */ 1507 if (name[0] != VFS_GENERIC) { 1508 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1509 if (vfsp->vfc_typenum == name[0]) 1510 break; 1511 if (vfsp == NULL) 1512 return (EOPNOTSUPP); 1513 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1514 oldp, oldlenp, newp, newlen, p)); 1515 } 1516 #endif 1517 switch (name[1]) { 1518 case VFS_MAXTYPENUM: 1519 if (namelen != 2) 1520 return (ENOTDIR); 1521 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1522 case VFS_CONF: 1523 if (namelen != 3) 1524 return (ENOTDIR); /* overloaded */ 1525 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1526 if (vfsp->vfc_typenum == name[2]) 1527 break; 1528 if (vfsp == NULL) 1529 return (EOPNOTSUPP); 1530 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1531 } 1532 return (EOPNOTSUPP); 1533 } 1534 1535 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1536 "Generic filesystem"); 1537 1538 #if 1 || defined(COMPAT_PRELITE2) 1539 1540 static int 1541 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1542 { 1543 int error; 1544 struct vfsconf *vfsp; 1545 struct ovfsconf ovfs; 1546 1547 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1548 bzero(&ovfs, sizeof(ovfs)); 1549 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1550 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1551 ovfs.vfc_index = vfsp->vfc_typenum; 1552 ovfs.vfc_refcount = vfsp->vfc_refcount; 1553 ovfs.vfc_flags = vfsp->vfc_flags; 1554 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1555 if (error) 1556 return error; 1557 } 1558 return 0; 1559 } 1560 1561 #endif /* 1 || COMPAT_PRELITE2 */ 1562 1563 /* 1564 * Check to see if a filesystem is mounted on a block device. 1565 */ 1566 int 1567 vfs_mountedon(struct vnode *vp) 1568 { 1569 cdev_t dev; 1570 1571 if ((dev = vp->v_rdev) == NULL) { 1572 if (vp->v_type != VBLK) 1573 dev = get_dev(vp->v_uminor, vp->v_umajor); 1574 } 1575 if (dev != NULL && dev->si_mountpoint) 1576 return (EBUSY); 1577 return (0); 1578 } 1579 1580 /* 1581 * Unmount all filesystems. The list is traversed in reverse order 1582 * of mounting to avoid dependencies. 1583 */ 1584 1585 static int vfs_umountall_callback(struct mount *mp, void *data); 1586 1587 void 1588 vfs_unmountall(void) 1589 { 1590 int count; 1591 1592 do { 1593 count = mountlist_scan(vfs_umountall_callback, 1594 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1595 } while (count); 1596 } 1597 1598 static 1599 int 1600 vfs_umountall_callback(struct mount *mp, void *data) 1601 { 1602 int error; 1603 1604 error = dounmount(mp, MNT_FORCE); 1605 if (error) { 1606 mountlist_remove(mp); 1607 kprintf("unmount of filesystem mounted from %s failed (", 1608 mp->mnt_stat.f_mntfromname); 1609 if (error == EBUSY) 1610 kprintf("BUSY)\n"); 1611 else 1612 kprintf("%d)\n", error); 1613 } 1614 return(1); 1615 } 1616 1617 /* 1618 * Build hash lists of net addresses and hang them off the mount point. 1619 * Called by ufs_mount() to set up the lists of export addresses. 1620 */ 1621 static int 1622 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1623 const struct export_args *argp) 1624 { 1625 struct netcred *np; 1626 struct radix_node_head *rnh; 1627 int i; 1628 struct radix_node *rn; 1629 struct sockaddr *saddr, *smask = 0; 1630 struct domain *dom; 1631 int error; 1632 1633 if (argp->ex_addrlen == 0) { 1634 if (mp->mnt_flag & MNT_DEFEXPORTED) 1635 return (EPERM); 1636 np = &nep->ne_defexported; 1637 np->netc_exflags = argp->ex_flags; 1638 np->netc_anon = argp->ex_anon; 1639 np->netc_anon.cr_ref = 1; 1640 mp->mnt_flag |= MNT_DEFEXPORTED; 1641 return (0); 1642 } 1643 1644 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1645 return (EINVAL); 1646 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1647 return (EINVAL); 1648 1649 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1650 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1651 saddr = (struct sockaddr *) (np + 1); 1652 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1653 goto out; 1654 if (saddr->sa_len > argp->ex_addrlen) 1655 saddr->sa_len = argp->ex_addrlen; 1656 if (argp->ex_masklen) { 1657 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1658 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1659 if (error) 1660 goto out; 1661 if (smask->sa_len > argp->ex_masklen) 1662 smask->sa_len = argp->ex_masklen; 1663 } 1664 i = saddr->sa_family; 1665 if ((rnh = nep->ne_rtable[i]) == 0) { 1666 /* 1667 * Seems silly to initialize every AF when most are not used, 1668 * do so on demand here 1669 */ 1670 SLIST_FOREACH(dom, &domains, dom_next) 1671 if (dom->dom_family == i && dom->dom_rtattach) { 1672 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1673 dom->dom_rtoffset); 1674 break; 1675 } 1676 if ((rnh = nep->ne_rtable[i]) == 0) { 1677 error = ENOBUFS; 1678 goto out; 1679 } 1680 } 1681 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1682 np->netc_rnodes); 1683 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1684 error = EPERM; 1685 goto out; 1686 } 1687 np->netc_exflags = argp->ex_flags; 1688 np->netc_anon = argp->ex_anon; 1689 np->netc_anon.cr_ref = 1; 1690 return (0); 1691 out: 1692 kfree(np, M_NETADDR); 1693 return (error); 1694 } 1695 1696 /* ARGSUSED */ 1697 static int 1698 vfs_free_netcred(struct radix_node *rn, void *w) 1699 { 1700 struct radix_node_head *rnh = (struct radix_node_head *) w; 1701 1702 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1703 kfree((caddr_t) rn, M_NETADDR); 1704 return (0); 1705 } 1706 1707 /* 1708 * Free the net address hash lists that are hanging off the mount points. 1709 */ 1710 static void 1711 vfs_free_addrlist(struct netexport *nep) 1712 { 1713 int i; 1714 struct radix_node_head *rnh; 1715 1716 for (i = 0; i <= AF_MAX; i++) 1717 if ((rnh = nep->ne_rtable[i])) { 1718 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1719 (caddr_t) rnh); 1720 kfree((caddr_t) rnh, M_RTABLE); 1721 nep->ne_rtable[i] = 0; 1722 } 1723 } 1724 1725 int 1726 vfs_export(struct mount *mp, struct netexport *nep, 1727 const struct export_args *argp) 1728 { 1729 int error; 1730 1731 if (argp->ex_flags & MNT_DELEXPORT) { 1732 if (mp->mnt_flag & MNT_EXPUBLIC) { 1733 vfs_setpublicfs(NULL, NULL, NULL); 1734 mp->mnt_flag &= ~MNT_EXPUBLIC; 1735 } 1736 vfs_free_addrlist(nep); 1737 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1738 } 1739 if (argp->ex_flags & MNT_EXPORTED) { 1740 if (argp->ex_flags & MNT_EXPUBLIC) { 1741 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1742 return (error); 1743 mp->mnt_flag |= MNT_EXPUBLIC; 1744 } 1745 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1746 return (error); 1747 mp->mnt_flag |= MNT_EXPORTED; 1748 } 1749 return (0); 1750 } 1751 1752 1753 /* 1754 * Set the publicly exported filesystem (WebNFS). Currently, only 1755 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1756 */ 1757 int 1758 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1759 const struct export_args *argp) 1760 { 1761 int error; 1762 struct vnode *rvp; 1763 char *cp; 1764 1765 /* 1766 * mp == NULL -> invalidate the current info, the FS is 1767 * no longer exported. May be called from either vfs_export 1768 * or unmount, so check if it hasn't already been done. 1769 */ 1770 if (mp == NULL) { 1771 if (nfs_pub.np_valid) { 1772 nfs_pub.np_valid = 0; 1773 if (nfs_pub.np_index != NULL) { 1774 FREE(nfs_pub.np_index, M_TEMP); 1775 nfs_pub.np_index = NULL; 1776 } 1777 } 1778 return (0); 1779 } 1780 1781 /* 1782 * Only one allowed at a time. 1783 */ 1784 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1785 return (EBUSY); 1786 1787 /* 1788 * Get real filehandle for root of exported FS. 1789 */ 1790 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1791 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1792 1793 if ((error = VFS_ROOT(mp, &rvp))) 1794 return (error); 1795 1796 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1797 return (error); 1798 1799 vput(rvp); 1800 1801 /* 1802 * If an indexfile was specified, pull it in. 1803 */ 1804 if (argp->ex_indexfile != NULL) { 1805 int namelen; 1806 1807 error = vn_get_namelen(rvp, &namelen); 1808 if (error) 1809 return (error); 1810 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 1811 M_WAITOK); 1812 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1813 namelen, (size_t *)0); 1814 if (!error) { 1815 /* 1816 * Check for illegal filenames. 1817 */ 1818 for (cp = nfs_pub.np_index; *cp; cp++) { 1819 if (*cp == '/') { 1820 error = EINVAL; 1821 break; 1822 } 1823 } 1824 } 1825 if (error) { 1826 FREE(nfs_pub.np_index, M_TEMP); 1827 return (error); 1828 } 1829 } 1830 1831 nfs_pub.np_mount = mp; 1832 nfs_pub.np_valid = 1; 1833 return (0); 1834 } 1835 1836 struct netcred * 1837 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1838 struct sockaddr *nam) 1839 { 1840 struct netcred *np; 1841 struct radix_node_head *rnh; 1842 struct sockaddr *saddr; 1843 1844 np = NULL; 1845 if (mp->mnt_flag & MNT_EXPORTED) { 1846 /* 1847 * Lookup in the export list first. 1848 */ 1849 if (nam != NULL) { 1850 saddr = nam; 1851 rnh = nep->ne_rtable[saddr->sa_family]; 1852 if (rnh != NULL) { 1853 np = (struct netcred *) 1854 (*rnh->rnh_matchaddr)((char *)saddr, 1855 rnh); 1856 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1857 np = NULL; 1858 } 1859 } 1860 /* 1861 * If no address match, use the default if it exists. 1862 */ 1863 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1864 np = &nep->ne_defexported; 1865 } 1866 return (np); 1867 } 1868 1869 /* 1870 * perform msync on all vnodes under a mount point. The mount point must 1871 * be locked. This code is also responsible for lazy-freeing unreferenced 1872 * vnodes whos VM objects no longer contain pages. 1873 * 1874 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1875 * 1876 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 1877 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 1878 * way up in this high level function. 1879 */ 1880 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1881 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1882 1883 void 1884 vfs_msync(struct mount *mp, int flags) 1885 { 1886 int vmsc_flags; 1887 1888 vmsc_flags = VMSC_GETVP; 1889 if (flags != MNT_WAIT) 1890 vmsc_flags |= VMSC_NOWAIT; 1891 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 1892 (void *)flags); 1893 } 1894 1895 /* 1896 * scan1 is a fast pre-check. There could be hundreds of thousands of 1897 * vnodes, we cannot afford to do anything heavy weight until we have a 1898 * fairly good indication that there is work to do. 1899 */ 1900 static 1901 int 1902 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1903 { 1904 int flags = (int)data; 1905 1906 if ((vp->v_flag & VRECLAIMED) == 0) { 1907 if (vshouldmsync(vp)) 1908 return(0); /* call scan2 */ 1909 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1910 (vp->v_flag & VOBJDIRTY) && 1911 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 1912 return(0); /* call scan2 */ 1913 } 1914 } 1915 1916 /* 1917 * do not call scan2, continue the loop 1918 */ 1919 return(-1); 1920 } 1921 1922 /* 1923 * This callback is handed a locked vnode. 1924 */ 1925 static 1926 int 1927 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1928 { 1929 vm_object_t obj; 1930 int flags = (int)data; 1931 1932 if (vp->v_flag & VRECLAIMED) 1933 return(0); 1934 1935 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 1936 if ((obj = vp->v_object) != NULL) { 1937 vm_object_page_clean(obj, 0, 0, 1938 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1939 } 1940 } 1941 return(0); 1942 } 1943 1944 /* 1945 * Record a process's interest in events which might happen to 1946 * a vnode. Because poll uses the historic select-style interface 1947 * internally, this routine serves as both the ``check for any 1948 * pending events'' and the ``record my interest in future events'' 1949 * functions. (These are done together, while the lock is held, 1950 * to avoid race conditions.) 1951 */ 1952 int 1953 vn_pollrecord(struct vnode *vp, int events) 1954 { 1955 lwkt_tokref ilock; 1956 1957 KKASSERT(curthread->td_proc != NULL); 1958 1959 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1960 if (vp->v_pollinfo.vpi_revents & events) { 1961 /* 1962 * This leaves events we are not interested 1963 * in available for the other process which 1964 * which presumably had requested them 1965 * (otherwise they would never have been 1966 * recorded). 1967 */ 1968 events &= vp->v_pollinfo.vpi_revents; 1969 vp->v_pollinfo.vpi_revents &= ~events; 1970 1971 lwkt_reltoken(&ilock); 1972 return events; 1973 } 1974 vp->v_pollinfo.vpi_events |= events; 1975 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo); 1976 lwkt_reltoken(&ilock); 1977 return 0; 1978 } 1979 1980 /* 1981 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1982 * it is possible for us to miss an event due to race conditions, but 1983 * that condition is expected to be rare, so for the moment it is the 1984 * preferred interface. 1985 */ 1986 void 1987 vn_pollevent(struct vnode *vp, int events) 1988 { 1989 lwkt_tokref ilock; 1990 1991 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1992 if (vp->v_pollinfo.vpi_events & events) { 1993 /* 1994 * We clear vpi_events so that we don't 1995 * call selwakeup() twice if two events are 1996 * posted before the polling process(es) is 1997 * awakened. This also ensures that we take at 1998 * most one selwakeup() if the polling process 1999 * is no longer interested. However, it does 2000 * mean that only one event can be noticed at 2001 * a time. (Perhaps we should only clear those 2002 * event bits which we note?) XXX 2003 */ 2004 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2005 vp->v_pollinfo.vpi_revents |= events; 2006 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2007 } 2008 lwkt_reltoken(&ilock); 2009 } 2010 2011 /* 2012 * Wake up anyone polling on vp because it is being revoked. 2013 * This depends on dead_poll() returning POLLHUP for correct 2014 * behavior. 2015 */ 2016 void 2017 vn_pollgone(struct vnode *vp) 2018 { 2019 lwkt_tokref ilock; 2020 2021 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 2022 if (vp->v_pollinfo.vpi_events) { 2023 vp->v_pollinfo.vpi_events = 0; 2024 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2025 } 2026 lwkt_reltoken(&ilock); 2027 } 2028 2029 /* 2030 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2031 * (or v_rdev might be NULL). 2032 */ 2033 cdev_t 2034 vn_todev(struct vnode *vp) 2035 { 2036 if (vp->v_type != VBLK && vp->v_type != VCHR) 2037 return (NULL); 2038 KKASSERT(vp->v_rdev != NULL); 2039 return (vp->v_rdev); 2040 } 2041 2042 /* 2043 * Check if vnode represents a disk device. The vnode does not need to be 2044 * opened. 2045 */ 2046 int 2047 vn_isdisk(struct vnode *vp, int *errp) 2048 { 2049 cdev_t dev; 2050 2051 if (vp->v_type != VCHR) { 2052 if (errp != NULL) 2053 *errp = ENOTBLK; 2054 return (0); 2055 } 2056 2057 if ((dev = vp->v_rdev) == NULL) 2058 dev = get_dev(vp->v_umajor, vp->v_uminor); 2059 2060 if (dev == NULL) { 2061 if (errp != NULL) 2062 *errp = ENXIO; 2063 return (0); 2064 } 2065 if (dev_is_good(dev) == 0) { 2066 if (errp != NULL) 2067 *errp = ENXIO; 2068 return (0); 2069 } 2070 if ((dev_dflags(dev) & D_DISK) == 0) { 2071 if (errp != NULL) 2072 *errp = ENOTBLK; 2073 return (0); 2074 } 2075 if (errp != NULL) 2076 *errp = 0; 2077 return (1); 2078 } 2079 2080 int 2081 vn_get_namelen(struct vnode *vp, int *namelen) 2082 { 2083 int error, retval[2]; 2084 2085 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2086 if (error) 2087 return (error); 2088 *namelen = *retval; 2089 return (0); 2090 } 2091 2092 int 2093 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2094 uint16_t d_namlen, const char *d_name) 2095 { 2096 struct dirent *dp; 2097 size_t len; 2098 2099 len = _DIRENT_RECLEN(d_namlen); 2100 if (len > uio->uio_resid) 2101 return(1); 2102 2103 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2104 2105 dp->d_ino = d_ino; 2106 dp->d_namlen = d_namlen; 2107 dp->d_type = d_type; 2108 bcopy(d_name, dp->d_name, d_namlen); 2109 2110 *error = uiomove((caddr_t)dp, len, uio); 2111 2112 kfree(dp, M_TEMP); 2113 2114 return(0); 2115 } 2116 2117