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 1.89 2006/06/05 21:03:02 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 87 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 88 89 int numvnodes; 90 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 91 int vfs_fastdev = 1; 92 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, ""); 93 94 enum vtype iftovt_tab[16] = { 95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 97 }; 98 int vttoif_tab[9] = { 99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 100 S_IFSOCK, S_IFIFO, S_IFMT, 101 }; 102 103 static int reassignbufcalls; 104 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, 105 &reassignbufcalls, 0, ""); 106 static int reassignbufloops; 107 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, 108 &reassignbufloops, 0, ""); 109 static int reassignbufsortgood; 110 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, 111 &reassignbufsortgood, 0, ""); 112 static int reassignbufsortbad; 113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, 114 &reassignbufsortbad, 0, ""); 115 static int reassignbufmethod = 1; 116 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, 117 &reassignbufmethod, 0, ""); 118 119 int nfs_mount_type = -1; 120 static struct lwkt_token spechash_token; 121 struct nfs_public nfs_pub; /* publicly exported FS */ 122 123 int desiredvnodes; 124 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 125 &desiredvnodes, 0, "Maximum number of vnodes"); 126 127 static void vfs_free_addrlist (struct netexport *nep); 128 static int vfs_free_netcred (struct radix_node *rn, void *w); 129 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 130 struct export_args *argp); 131 132 extern int dev_ref_debug; 133 extern struct vnodeopv_entry_desc spec_vnodeop_entries[]; 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 * Return 0 if the vnode is already on the free list or cannot be placed 154 * on the free list. Return 1 if the vnode can be placed on the free list. 155 */ 156 static __inline int 157 vshouldfree(struct vnode *vp, int usecount) 158 { 159 if (vp->v_flag & VFREE) 160 return (0); /* already free */ 161 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount) 162 return (0); /* other holderse */ 163 if (vp->v_object && 164 (vp->v_object->ref_count || vp->v_object->resident_page_count)) { 165 return (0); 166 } 167 return (1); 168 } 169 170 /* 171 * Initialize the vnode management data structures. 172 * 173 * Called from vfsinit() 174 */ 175 void 176 vfs_subr_init(void) 177 { 178 /* 179 * Desired vnodes is a result of the physical page count 180 * and the size of kernel's heap. It scales in proportion 181 * to the amount of available physical memory. This can 182 * cause trouble on 64-bit and large memory platforms. 183 */ 184 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */ 185 desiredvnodes = 186 min(maxproc + vmstats.v_page_count /4, 187 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 188 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 189 190 lwkt_token_init(&spechash_token); 191 } 192 193 /* 194 * Knob to control the precision of file timestamps: 195 * 196 * 0 = seconds only; nanoseconds zeroed. 197 * 1 = seconds and nanoseconds, accurate within 1/HZ. 198 * 2 = seconds and nanoseconds, truncated to microseconds. 199 * >=3 = seconds and nanoseconds, maximum precision. 200 */ 201 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 202 203 static int timestamp_precision = TSP_SEC; 204 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 205 ×tamp_precision, 0, ""); 206 207 /* 208 * Get a current timestamp. 209 */ 210 void 211 vfs_timestamp(struct timespec *tsp) 212 { 213 struct timeval tv; 214 215 switch (timestamp_precision) { 216 case TSP_SEC: 217 tsp->tv_sec = time_second; 218 tsp->tv_nsec = 0; 219 break; 220 case TSP_HZ: 221 getnanotime(tsp); 222 break; 223 case TSP_USEC: 224 microtime(&tv); 225 TIMEVAL_TO_TIMESPEC(&tv, tsp); 226 break; 227 case TSP_NSEC: 228 default: 229 nanotime(tsp); 230 break; 231 } 232 } 233 234 /* 235 * Set vnode attributes to VNOVAL 236 */ 237 void 238 vattr_null(struct vattr *vap) 239 { 240 vap->va_type = VNON; 241 vap->va_size = VNOVAL; 242 vap->va_bytes = VNOVAL; 243 vap->va_mode = VNOVAL; 244 vap->va_nlink = VNOVAL; 245 vap->va_uid = VNOVAL; 246 vap->va_gid = VNOVAL; 247 vap->va_fsid = VNOVAL; 248 vap->va_fileid = VNOVAL; 249 vap->va_blocksize = VNOVAL; 250 vap->va_rdev = VNOVAL; 251 vap->va_atime.tv_sec = VNOVAL; 252 vap->va_atime.tv_nsec = VNOVAL; 253 vap->va_mtime.tv_sec = VNOVAL; 254 vap->va_mtime.tv_nsec = VNOVAL; 255 vap->va_ctime.tv_sec = VNOVAL; 256 vap->va_ctime.tv_nsec = VNOVAL; 257 vap->va_flags = VNOVAL; 258 vap->va_gen = VNOVAL; 259 vap->va_vaflags = 0; 260 vap->va_fsmid = VNOVAL; 261 } 262 263 /* 264 * Flush out and invalidate all buffers associated with a vnode. 265 * 266 * vp must be locked. 267 */ 268 static int vinvalbuf_bp(struct buf *bp, void *data); 269 270 struct vinvalbuf_bp_info { 271 struct vnode *vp; 272 int slptimeo; 273 int lkflags; 274 int flags; 275 }; 276 277 void 278 vupdatefsmid(struct vnode *vp) 279 { 280 atomic_set_int(&vp->v_flag, VFSMID); 281 } 282 283 int 284 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 285 { 286 struct vinvalbuf_bp_info info; 287 int error; 288 vm_object_t object; 289 290 /* 291 * If we are being asked to save, call fsync to ensure that the inode 292 * is updated. 293 */ 294 if (flags & V_SAVE) { 295 crit_enter(); 296 while (vp->v_track_write.bk_active) { 297 vp->v_track_write.bk_waitflag = 1; 298 error = tsleep(&vp->v_track_write, slpflag, 299 "vinvlbuf", slptimeo); 300 if (error) { 301 crit_exit(); 302 return (error); 303 } 304 } 305 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 306 crit_exit(); 307 if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0) 308 return (error); 309 crit_enter(); 310 if (vp->v_track_write.bk_active > 0 || 311 !RB_EMPTY(&vp->v_rbdirty_tree)) 312 panic("vinvalbuf: dirty bufs"); 313 } 314 crit_exit(); 315 } 316 crit_enter(); 317 info.slptimeo = slptimeo; 318 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 319 if (slpflag & PCATCH) 320 info.lkflags |= LK_PCATCH; 321 info.flags = flags; 322 info.vp = vp; 323 324 /* 325 * Flush the buffer cache until nothing is left. 326 */ 327 while (!RB_EMPTY(&vp->v_rbclean_tree) || 328 !RB_EMPTY(&vp->v_rbdirty_tree)) { 329 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 330 vinvalbuf_bp, &info); 331 if (error == 0) { 332 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 333 vinvalbuf_bp, &info); 334 } 335 } 336 337 /* 338 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 339 * have write I/O in-progress but if there is a VM object then the 340 * VM object can also have read-I/O in-progress. 341 */ 342 do { 343 while (vp->v_track_write.bk_active > 0) { 344 vp->v_track_write.bk_waitflag = 1; 345 tsleep(&vp->v_track_write, 0, "vnvlbv", 0); 346 } 347 if ((object = vp->v_object) != NULL) { 348 while (object->paging_in_progress) 349 vm_object_pip_sleep(object, "vnvlbx"); 350 } 351 } while (vp->v_track_write.bk_active > 0); 352 353 crit_exit(); 354 355 /* 356 * Destroy the copy in the VM cache, too. 357 */ 358 if ((object = vp->v_object) != NULL) { 359 vm_object_page_remove(object, 0, 0, 360 (flags & V_SAVE) ? TRUE : FALSE); 361 } 362 363 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 364 panic("vinvalbuf: flush failed"); 365 if (!RB_EMPTY(&vp->v_rbhash_tree)) 366 panic("vinvalbuf: flush failed, buffers still present"); 367 return (0); 368 } 369 370 static int 371 vinvalbuf_bp(struct buf *bp, void *data) 372 { 373 struct vinvalbuf_bp_info *info = data; 374 int error; 375 376 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 377 error = BUF_TIMELOCK(bp, info->lkflags, 378 "vinvalbuf", info->slptimeo); 379 if (error == 0) { 380 BUF_UNLOCK(bp); 381 error = ENOLCK; 382 } 383 if (error == ENOLCK) 384 return(0); 385 return (-error); 386 } 387 388 KKASSERT(bp->b_vp == info->vp); 389 390 /* 391 * XXX Since there are no node locks for NFS, I 392 * believe there is a slight chance that a delayed 393 * write will occur while sleeping just above, so 394 * check for it. Note that vfs_bio_awrite expects 395 * buffers to reside on a queue, while bwrite() and 396 * brelse() do not. 397 */ 398 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 399 (info->flags & V_SAVE)) { 400 if (bp->b_vp == info->vp) { 401 if (bp->b_flags & B_CLUSTEROK) { 402 vfs_bio_awrite(bp); 403 } else { 404 bremfree(bp); 405 bp->b_flags |= B_ASYNC; 406 bwrite(bp); 407 } 408 } else { 409 bremfree(bp); 410 bwrite(bp); 411 } 412 } else if (info->flags & V_SAVE) { 413 /* 414 * Cannot set B_NOCACHE on a clean buffer as this will 415 * destroy the VM backing store which might actually 416 * be dirty (and unsynchronized). 417 */ 418 bremfree(bp); 419 bp->b_flags |= (B_INVAL | B_RELBUF); 420 bp->b_flags &= ~B_ASYNC; 421 brelse(bp); 422 } else { 423 bremfree(bp); 424 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 425 bp->b_flags &= ~B_ASYNC; 426 brelse(bp); 427 } 428 return(0); 429 } 430 431 /* 432 * Truncate a file's buffer and pages to a specified length. This 433 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 434 * sync activity. 435 * 436 * The vnode must be locked. 437 */ 438 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 439 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 440 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 441 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 442 443 int 444 vtruncbuf(struct vnode *vp, off_t length, int blksize) 445 { 446 off_t truncloffset; 447 int count; 448 const char *filename; 449 450 /* 451 * Round up to the *next* block, then destroy the buffers in question. 452 * Since we are only removing some of the buffers we must rely on the 453 * scan count to determine whether a loop is necessary. 454 */ 455 if ((count = (int)(length % blksize)) != 0) 456 truncloffset = length + (blksize - count); 457 else 458 truncloffset = length; 459 460 crit_enter(); 461 do { 462 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 463 vtruncbuf_bp_trunc_cmp, 464 vtruncbuf_bp_trunc, &truncloffset); 465 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 466 vtruncbuf_bp_trunc_cmp, 467 vtruncbuf_bp_trunc, &truncloffset); 468 } while(count); 469 470 /* 471 * For safety, fsync any remaining metadata if the file is not being 472 * truncated to 0. Since the metadata does not represent the entire 473 * dirty list we have to rely on the hit count to ensure that we get 474 * all of it. 475 */ 476 if (length > 0) { 477 do { 478 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 479 vtruncbuf_bp_metasync_cmp, 480 vtruncbuf_bp_metasync, vp); 481 } while (count); 482 } 483 484 /* 485 * Clean out any left over VM backing store. 486 */ 487 crit_exit(); 488 489 vnode_pager_setsize(vp, length); 490 491 crit_enter(); 492 493 /* 494 * It is possible to have in-progress I/O from buffers that were 495 * not part of the truncation. This should not happen if we 496 * are truncating to 0-length. 497 */ 498 filename = TAILQ_FIRST(&vp->v_namecache) ? 499 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 500 501 while ((count = vp->v_track_write.bk_active) > 0) { 502 vp->v_track_write.bk_waitflag = 1; 503 tsleep(&vp->v_track_write, 0, "vbtrunc", 0); 504 if (length == 0) { 505 printf("Warning: vtruncbuf(): Had to wait for " 506 "%d buffer I/Os to finish in %s\n", 507 count, filename); 508 } 509 } 510 511 /* 512 * Make sure no buffers were instantiated while we were trying 513 * to clean out the remaining VM pages. This could occur due 514 * to busy dirty VM pages being flushed out to disk. 515 */ 516 do { 517 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 518 vtruncbuf_bp_trunc_cmp, 519 vtruncbuf_bp_trunc, &truncloffset); 520 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 521 vtruncbuf_bp_trunc_cmp, 522 vtruncbuf_bp_trunc, &truncloffset); 523 if (count) { 524 printf("Warning: vtruncbuf(): Had to re-clean %d " 525 "left over buffers in %s\n", count, filename); 526 } 527 } while(count); 528 529 crit_exit(); 530 531 return (0); 532 } 533 534 /* 535 * The callback buffer is beyond the new file EOF and must be destroyed. 536 * Note that the compare function must conform to the RB_SCAN's requirements. 537 */ 538 static 539 int 540 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 541 { 542 if (bp->b_loffset >= *(off_t *)data) 543 return(0); 544 return(-1); 545 } 546 547 static 548 int 549 vtruncbuf_bp_trunc(struct buf *bp, void *data) 550 { 551 /* 552 * Do not try to use a buffer we cannot immediately lock, but sleep 553 * anyway to prevent a livelock. The code will loop until all buffers 554 * can be acted upon. 555 */ 556 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 557 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 558 BUF_UNLOCK(bp); 559 } else { 560 bremfree(bp); 561 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 562 bp->b_flags &= ~B_ASYNC; 563 brelse(bp); 564 } 565 return(1); 566 } 567 568 /* 569 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 570 * blocks (with a negative loffset) are scanned. 571 * Note that the compare function must conform to the RB_SCAN's requirements. 572 */ 573 static int 574 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 575 { 576 if (bp->b_loffset < 0) 577 return(0); 578 return(1); 579 } 580 581 static int 582 vtruncbuf_bp_metasync(struct buf *bp, void *data) 583 { 584 struct vnode *vp = data; 585 586 if (bp->b_flags & B_DELWRI) { 587 /* 588 * Do not try to use a buffer we cannot immediately lock, 589 * but sleep anyway to prevent a livelock. The code will 590 * loop until all buffers can be acted upon. 591 */ 592 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 593 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 594 BUF_UNLOCK(bp); 595 } else { 596 bremfree(bp); 597 if (bp->b_vp == vp) { 598 bp->b_flags |= B_ASYNC; 599 } else { 600 bp->b_flags &= ~B_ASYNC; 601 } 602 bwrite(bp); 603 } 604 return(1); 605 } else { 606 return(0); 607 } 608 } 609 610 /* 611 * vfsync - implements a multipass fsync on a file which understands 612 * dependancies and meta-data. The passed vnode must be locked. The 613 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 614 * 615 * When fsyncing data asynchronously just do one consolidated pass starting 616 * with the most negative block number. This may not get all the data due 617 * to dependancies. 618 * 619 * When fsyncing data synchronously do a data pass, then a metadata pass, 620 * then do additional data+metadata passes to try to get all the data out. 621 */ 622 static int vfsync_wait_output(struct vnode *vp, 623 int (*waitoutput)(struct vnode *, struct thread *)); 624 static int vfsync_data_only_cmp(struct buf *bp, void *data); 625 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 626 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 627 static int vfsync_bp(struct buf *bp, void *data); 628 629 struct vfsync_info { 630 struct vnode *vp; 631 int synchronous; 632 int syncdeps; 633 int lazycount; 634 int lazylimit; 635 int skippedbufs; 636 int (*checkdef)(struct buf *); 637 }; 638 639 int 640 vfsync(struct vnode *vp, int waitfor, int passes, 641 int (*checkdef)(struct buf *), 642 int (*waitoutput)(struct vnode *, struct thread *)) 643 { 644 struct vfsync_info info; 645 int error; 646 647 bzero(&info, sizeof(info)); 648 info.vp = vp; 649 if ((info.checkdef = checkdef) == NULL) 650 info.syncdeps = 1; 651 652 crit_enter_id("vfsync"); 653 654 switch(waitfor) { 655 case MNT_LAZY: 656 /* 657 * Lazy (filesystem syncer typ) Asynchronous plus limit the 658 * number of data (not meta) pages we try to flush to 1MB. 659 * A non-zero return means that lazy limit was reached. 660 */ 661 info.lazylimit = 1024 * 1024; 662 info.syncdeps = 1; 663 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 664 vfsync_lazy_range_cmp, vfsync_bp, &info); 665 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 666 vfsync_meta_only_cmp, vfsync_bp, &info); 667 if (error == 0) 668 vp->v_lazyw = 0; 669 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 670 vn_syncer_add_to_worklist(vp, 1); 671 error = 0; 672 break; 673 case MNT_NOWAIT: 674 /* 675 * Asynchronous. Do a data-only pass and a meta-only pass. 676 */ 677 info.syncdeps = 1; 678 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 679 vfsync_bp, &info); 680 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 681 vfsync_bp, &info); 682 error = 0; 683 break; 684 default: 685 /* 686 * Synchronous. Do a data-only pass, then a meta-data+data 687 * pass, then additional integrated passes to try to get 688 * all the dependancies flushed. 689 */ 690 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 691 vfsync_bp, &info); 692 error = vfsync_wait_output(vp, waitoutput); 693 if (error == 0) { 694 info.skippedbufs = 0; 695 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 696 vfsync_bp, &info); 697 error = vfsync_wait_output(vp, waitoutput); 698 if (info.skippedbufs) 699 printf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs); 700 } 701 while (error == 0 && passes > 0 && 702 !RB_EMPTY(&vp->v_rbdirty_tree)) { 703 if (--passes == 0) { 704 info.synchronous = 1; 705 info.syncdeps = 1; 706 } 707 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 708 vfsync_bp, &info); 709 if (error < 0) 710 error = -error; 711 info.syncdeps = 1; 712 if (error == 0) 713 error = vfsync_wait_output(vp, waitoutput); 714 } 715 break; 716 } 717 crit_exit_id("vfsync"); 718 return(error); 719 } 720 721 static int 722 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 723 { 724 int error = 0; 725 726 while (vp->v_track_write.bk_active) { 727 vp->v_track_write.bk_waitflag = 1; 728 tsleep(&vp->v_track_write, 0, "fsfsn", 0); 729 } 730 if (waitoutput) 731 error = waitoutput(vp, curthread); 732 return(error); 733 } 734 735 static int 736 vfsync_data_only_cmp(struct buf *bp, void *data) 737 { 738 if (bp->b_loffset < 0) 739 return(-1); 740 return(0); 741 } 742 743 static int 744 vfsync_meta_only_cmp(struct buf *bp, void *data) 745 { 746 if (bp->b_loffset < 0) 747 return(0); 748 return(1); 749 } 750 751 static int 752 vfsync_lazy_range_cmp(struct buf *bp, void *data) 753 { 754 struct vfsync_info *info = data; 755 if (bp->b_loffset < info->vp->v_lazyw) 756 return(-1); 757 return(0); 758 } 759 760 static int 761 vfsync_bp(struct buf *bp, void *data) 762 { 763 struct vfsync_info *info = data; 764 struct vnode *vp = info->vp; 765 int error; 766 767 /* 768 * if syncdeps is not set we do not try to write buffers which have 769 * dependancies. 770 */ 771 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 772 return(0); 773 774 /* 775 * Ignore buffers that we cannot immediately lock. XXX 776 */ 777 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 778 printf("Warning: vfsync_bp skipping dirty buffer %p\n", bp); 779 ++info->skippedbufs; 780 return(0); 781 } 782 if ((bp->b_flags & B_DELWRI) == 0) 783 panic("vfsync_bp: buffer not dirty"); 784 if (vp != bp->b_vp) 785 panic("vfsync_bp: buffer vp mismatch"); 786 787 /* 788 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 789 * has been written but an additional handshake with the device 790 * is required before we can dispose of the buffer. We have no idea 791 * how to do this so we have to skip these buffers. 792 */ 793 if (bp->b_flags & B_NEEDCOMMIT) { 794 BUF_UNLOCK(bp); 795 return(0); 796 } 797 798 if (info->synchronous) { 799 /* 800 * Synchronous flushing. An error may be returned. 801 */ 802 bremfree(bp); 803 crit_exit_id("vfsync"); 804 error = bwrite(bp); 805 crit_enter_id("vfsync"); 806 } else { 807 /* 808 * Asynchronous flushing. A negative return value simply 809 * stops the scan and is not considered an error. We use 810 * this to support limited MNT_LAZY flushes. 811 */ 812 vp->v_lazyw = bp->b_loffset; 813 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 814 info->lazycount += vfs_bio_awrite(bp); 815 } else { 816 info->lazycount += bp->b_bufsize; 817 bremfree(bp); 818 crit_exit_id("vfsync"); 819 bawrite(bp); 820 crit_enter_id("vfsync"); 821 } 822 if (info->lazylimit && info->lazycount >= info->lazylimit) 823 error = 1; 824 else 825 error = 0; 826 } 827 return(-error); 828 } 829 830 /* 831 * Associate a buffer with a vnode. 832 */ 833 void 834 bgetvp(struct vnode *vp, struct buf *bp) 835 { 836 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 837 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 838 839 vhold(vp); 840 /* 841 * Insert onto list for new vnode. 842 */ 843 crit_enter(); 844 bp->b_vp = vp; 845 bp->b_flags |= B_HASHED; 846 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) 847 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 848 849 bp->b_flags |= B_VNCLEAN; 850 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 851 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 852 crit_exit(); 853 } 854 855 /* 856 * Disassociate a buffer from a vnode. 857 */ 858 void 859 brelvp(struct buf *bp) 860 { 861 struct vnode *vp; 862 863 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 864 865 /* 866 * Delete from old vnode list, if on one. 867 */ 868 vp = bp->b_vp; 869 crit_enter(); 870 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 871 if (bp->b_flags & B_VNDIRTY) 872 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 873 else 874 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 875 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 876 } 877 if (bp->b_flags & B_HASHED) { 878 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 879 bp->b_flags &= ~B_HASHED; 880 } 881 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 882 vp->v_flag &= ~VONWORKLST; 883 LIST_REMOVE(vp, v_synclist); 884 } 885 crit_exit(); 886 bp->b_vp = NULL; 887 vdrop(vp); 888 } 889 890 /* 891 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 892 * This routine is called when the state of the B_DELWRI bit is changed. 893 */ 894 void 895 reassignbuf(struct buf *bp) 896 { 897 struct vnode *vp = bp->b_vp; 898 int delay; 899 900 KKASSERT(vp != NULL); 901 ++reassignbufcalls; 902 903 /* 904 * B_PAGING flagged buffers cannot be reassigned because their vp 905 * is not fully linked in. 906 */ 907 if (bp->b_flags & B_PAGING) 908 panic("cannot reassign paging buffer"); 909 910 crit_enter(); 911 if (bp->b_flags & B_DELWRI) { 912 /* 913 * Move to the dirty list, add the vnode to the worklist 914 */ 915 if (bp->b_flags & B_VNCLEAN) { 916 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 917 bp->b_flags &= ~B_VNCLEAN; 918 } 919 if ((bp->b_flags & B_VNDIRTY) == 0) { 920 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 921 panic("reassignbuf: dup lblk vp %p bp %p", 922 vp, bp); 923 } 924 bp->b_flags |= B_VNDIRTY; 925 } 926 if ((vp->v_flag & VONWORKLST) == 0) { 927 switch (vp->v_type) { 928 case VDIR: 929 delay = dirdelay; 930 break; 931 case VCHR: 932 case VBLK: 933 if (vp->v_rdev && 934 vp->v_rdev->si_mountpoint != NULL) { 935 delay = metadelay; 936 break; 937 } 938 /* fall through */ 939 default: 940 delay = filedelay; 941 } 942 vn_syncer_add_to_worklist(vp, delay); 943 } 944 } else { 945 /* 946 * Move to the clean list, remove the vnode from the worklist 947 * if no dirty blocks remain. 948 */ 949 if (bp->b_flags & B_VNDIRTY) { 950 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 951 bp->b_flags &= ~B_VNDIRTY; 952 } 953 if ((bp->b_flags & B_VNCLEAN) == 0) { 954 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 955 panic("reassignbuf: dup lblk vp %p bp %p", 956 vp, bp); 957 } 958 bp->b_flags |= B_VNCLEAN; 959 } 960 if ((vp->v_flag & VONWORKLST) && 961 RB_EMPTY(&vp->v_rbdirty_tree)) { 962 vp->v_flag &= ~VONWORKLST; 963 LIST_REMOVE(vp, v_synclist); 964 } 965 } 966 crit_exit(); 967 } 968 969 /* 970 * Create a vnode for a block device. 971 * Used for mounting the root file system. 972 */ 973 int 974 bdevvp(dev_t dev, struct vnode **vpp) 975 { 976 struct vnode *vp; 977 struct vnode *nvp; 978 int error; 979 980 if (dev == NODEV) { 981 *vpp = NULLVP; 982 return (ENXIO); 983 } 984 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0); 985 if (error) { 986 *vpp = NULLVP; 987 return (error); 988 } 989 vp = nvp; 990 vp->v_type = VCHR; 991 vp->v_udev = dev->si_udev; 992 vx_unlock(vp); 993 *vpp = vp; 994 return (0); 995 } 996 997 int 998 v_associate_rdev(struct vnode *vp, dev_t dev) 999 { 1000 lwkt_tokref ilock; 1001 1002 if (dev == NULL || dev == NODEV) 1003 return(ENXIO); 1004 if (dev_is_good(dev) == 0) 1005 return(ENXIO); 1006 KKASSERT(vp->v_rdev == NULL); 1007 if (dev_ref_debug) 1008 printf("Z1"); 1009 vp->v_rdev = reference_dev(dev); 1010 lwkt_gettoken(&ilock, &spechash_token); 1011 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext); 1012 lwkt_reltoken(&ilock); 1013 return(0); 1014 } 1015 1016 void 1017 v_release_rdev(struct vnode *vp) 1018 { 1019 lwkt_tokref ilock; 1020 dev_t dev; 1021 1022 if ((dev = vp->v_rdev) != NULL) { 1023 lwkt_gettoken(&ilock, &spechash_token); 1024 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext); 1025 vp->v_rdev = NULL; 1026 release_dev(dev); 1027 lwkt_reltoken(&ilock); 1028 } 1029 } 1030 1031 /* 1032 * Add a vnode to the alias list hung off the dev_t. We only associate 1033 * the device number with the vnode. The actual device is not associated 1034 * until the vnode is opened (usually in spec_open()), and will be 1035 * disassociated on last close. 1036 */ 1037 void 1038 addaliasu(struct vnode *nvp, udev_t nvp_udev) 1039 { 1040 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1041 panic("addaliasu on non-special vnode"); 1042 nvp->v_udev = nvp_udev; 1043 } 1044 1045 /* 1046 * Disassociate a vnode from its underlying filesystem. 1047 * 1048 * The vnode must be VX locked and refd 1049 * 1050 * If there are v_usecount references to the vnode other then ours we have 1051 * to VOP_CLOSE the vnode before we can deactivate and reclaim it. 1052 */ 1053 void 1054 vclean(struct vnode *vp, int flags) 1055 { 1056 int active; 1057 int n; 1058 vm_object_t object; 1059 1060 /* 1061 * If the vnode has already been reclaimed we have nothing to do. 1062 */ 1063 if (vp->v_flag & VRECLAIMED) 1064 return; 1065 vp->v_flag |= VRECLAIMED; 1066 1067 /* 1068 * Scrap the vfs cache 1069 */ 1070 while (cache_inval_vp(vp, 0) != 0) { 1071 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1072 tsleep(vp, 0, "vclninv", 2); 1073 } 1074 1075 /* 1076 * Check to see if the vnode is in use. If so we have to reference it 1077 * before we clean it out so that its count cannot fall to zero and 1078 * generate a race against ourselves to recycle it. 1079 */ 1080 active = (vp->v_usecount > 1); 1081 1082 /* 1083 * Clean out any buffers associated with the vnode and destroy its 1084 * object, if it has one. 1085 */ 1086 vinvalbuf(vp, V_SAVE, 0, 0); 1087 1088 if ((object = vp->v_object) != NULL) { 1089 if (object->ref_count == 0) { 1090 if ((object->flags & OBJ_DEAD) == 0) 1091 vm_object_terminate(object); 1092 } else { 1093 vm_pager_deallocate(object); 1094 } 1095 vp->v_flag &= ~VOBJBUF; 1096 } 1097 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1098 1099 /* 1100 * If purging an active vnode (typically during a forced unmount 1101 * or reboot), it must be closed and deactivated before being 1102 * reclaimed. This isn't really all that safe, but what can 1103 * we do? XXX. 1104 * 1105 * Note that neither of these routines unlocks the vnode. 1106 */ 1107 if (active && (flags & DOCLOSE)) { 1108 while ((n = vp->v_opencount) != 0) { 1109 if (vp->v_writecount) 1110 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1111 else 1112 VOP_CLOSE(vp, FNONBLOCK); 1113 if (vp->v_opencount == n) { 1114 printf("Warning: unable to force-close" 1115 " vnode %p\n", vp); 1116 break; 1117 } 1118 } 1119 } 1120 1121 /* 1122 * If the vnode has not be deactivated, deactivated it. 1123 */ 1124 if ((vp->v_flag & VINACTIVE) == 0) { 1125 vp->v_flag |= VINACTIVE; 1126 VOP_INACTIVE(vp); 1127 } 1128 1129 /* 1130 * Reclaim the vnode. 1131 */ 1132 if (VOP_RECLAIM(vp)) 1133 panic("vclean: cannot reclaim"); 1134 1135 /* 1136 * Done with purge, notify sleepers of the grim news. 1137 */ 1138 vp->v_ops = &dead_vnode_vops; 1139 vn_pollgone(vp); 1140 vp->v_tag = VT_NON; 1141 } 1142 1143 /* 1144 * Eliminate all activity associated with the requested vnode 1145 * and with all vnodes aliased to the requested vnode. 1146 * 1147 * The vnode must be referenced and vx_lock()'d 1148 * 1149 * revoke { struct vnode *a_vp, int a_flags } 1150 */ 1151 int 1152 vop_stdrevoke(struct vop_revoke_args *ap) 1153 { 1154 struct vnode *vp, *vq; 1155 lwkt_tokref ilock; 1156 dev_t dev; 1157 1158 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1159 1160 vp = ap->a_vp; 1161 1162 /* 1163 * If the vnode is already dead don't try to revoke it 1164 */ 1165 if (vp->v_flag & VRECLAIMED) 1166 return (0); 1167 1168 /* 1169 * If the vnode has a device association, scrap all vnodes associated 1170 * with the device. Don't let the device disappear on us while we 1171 * are scrapping the vnodes. 1172 * 1173 * The passed vp will probably show up in the list, do not VX lock 1174 * it twice! 1175 */ 1176 if (vp->v_type != VCHR && vp->v_type != VBLK) 1177 return(0); 1178 if ((dev = vp->v_rdev) == NULL) { 1179 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV) 1180 return(0); 1181 } 1182 reference_dev(dev); 1183 lwkt_gettoken(&ilock, &spechash_token); 1184 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1185 if (vp == vq || vx_get(vq) == 0) { 1186 if (vq == SLIST_FIRST(&dev->si_hlist)) 1187 vgone(vq); 1188 if (vp != vq) 1189 vx_put(vq); 1190 } 1191 } 1192 lwkt_reltoken(&ilock); 1193 release_dev(dev); 1194 return (0); 1195 } 1196 1197 /* 1198 * Recycle an unused vnode to the front of the free list. 1199 * 1200 * Returns 1 if we were successfully able to recycle the vnode, 1201 * 0 otherwise. 1202 */ 1203 int 1204 vrecycle(struct vnode *vp) 1205 { 1206 if (vp->v_usecount == 1) { 1207 vgone(vp); 1208 return (1); 1209 } 1210 return (0); 1211 } 1212 1213 /* 1214 * Eliminate all activity associated with a vnode in preparation for reuse. 1215 * 1216 * The vnode must be VX locked and refd and will remain VX locked and refd 1217 * on return. This routine may be called with the vnode in any state, as 1218 * long as it is VX locked. The vnode will be cleaned out and marked 1219 * VRECLAIMED but will not actually be reused until all existing refs and 1220 * holds go away. 1221 * 1222 * NOTE: This routine may be called on a vnode which has not yet been 1223 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1224 * already been reclaimed. 1225 * 1226 * This routine is not responsible for placing us back on the freelist. 1227 * Instead, it happens automatically when the caller releases the VX lock 1228 * (assuming there aren't any other references). 1229 */ 1230 void 1231 vgone(struct vnode *vp) 1232 { 1233 /* 1234 * assert that the VX lock is held. This is an absolute requirement 1235 * now for vgone() to be called. 1236 */ 1237 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1238 1239 /* 1240 * Clean out the filesystem specific data and set the VRECLAIMED 1241 * bit. Also deactivate the vnode if necessary. 1242 */ 1243 vclean(vp, DOCLOSE); 1244 1245 /* 1246 * Delete from old mount point vnode list, if on one. 1247 */ 1248 if (vp->v_mount != NULL) 1249 insmntque(vp, NULL); 1250 1251 /* 1252 * If special device, remove it from special device alias list 1253 * if it is on one. This should normally only occur if a vnode is 1254 * being revoked as the device should otherwise have been released 1255 * naturally. 1256 */ 1257 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1258 v_release_rdev(vp); 1259 } 1260 1261 /* 1262 * Set us to VBAD 1263 */ 1264 vp->v_type = VBAD; 1265 } 1266 1267 /* 1268 * Lookup a vnode by device number. 1269 */ 1270 int 1271 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp) 1272 { 1273 lwkt_tokref ilock; 1274 struct vnode *vp; 1275 1276 lwkt_gettoken(&ilock, &spechash_token); 1277 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1278 if (type == vp->v_type) { 1279 *vpp = vp; 1280 lwkt_reltoken(&ilock); 1281 return (1); 1282 } 1283 } 1284 lwkt_reltoken(&ilock); 1285 return (0); 1286 } 1287 1288 /* 1289 * Calculate the total number of references to a special device. This 1290 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1291 * an overloaded field. Since udev2dev can now return NODEV, we have 1292 * to check for a NULL v_rdev. 1293 */ 1294 int 1295 count_dev(dev_t dev) 1296 { 1297 lwkt_tokref ilock; 1298 struct vnode *vp; 1299 int count = 0; 1300 1301 if (SLIST_FIRST(&dev->si_hlist)) { 1302 lwkt_gettoken(&ilock, &spechash_token); 1303 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1304 count += vp->v_usecount; 1305 } 1306 lwkt_reltoken(&ilock); 1307 } 1308 return(count); 1309 } 1310 1311 int 1312 count_udev(udev_t udev) 1313 { 1314 dev_t dev; 1315 1316 if ((dev = udev2dev(udev, 0)) == NODEV) 1317 return(0); 1318 return(count_dev(dev)); 1319 } 1320 1321 int 1322 vcount(struct vnode *vp) 1323 { 1324 if (vp->v_rdev == NULL) 1325 return(0); 1326 return(count_dev(vp->v_rdev)); 1327 } 1328 1329 /* 1330 * Initialize VMIO for a vnode. This routine MUST be called before a 1331 * VFS can issue buffer cache ops on a vnode. It is typically called 1332 * when a vnode is initialized from its inode. 1333 */ 1334 int 1335 vinitvmio(struct vnode *vp, off_t filesize) 1336 { 1337 vm_object_t object; 1338 int error = 0; 1339 1340 retry: 1341 if ((object = vp->v_object) == NULL) { 1342 object = vnode_pager_alloc(vp, filesize, 0, 0); 1343 /* 1344 * Dereference the reference we just created. This assumes 1345 * that the object is associated with the vp. 1346 */ 1347 object->ref_count--; 1348 vp->v_usecount--; 1349 } else { 1350 if (object->flags & OBJ_DEAD) { 1351 VOP_UNLOCK(vp, 0); 1352 tsleep(object, 0, "vodead", 0); 1353 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1354 goto retry; 1355 } 1356 } 1357 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1358 vp->v_flag |= VOBJBUF; 1359 return (error); 1360 } 1361 1362 1363 /* 1364 * Print out a description of a vnode. 1365 */ 1366 static char *typename[] = 1367 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1368 1369 void 1370 vprint(char *label, struct vnode *vp) 1371 { 1372 char buf[96]; 1373 1374 if (label != NULL) 1375 printf("%s: %p: ", label, (void *)vp); 1376 else 1377 printf("%p: ", (void *)vp); 1378 printf("type %s, usecount %d, writecount %d, refcount %d,", 1379 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1380 vp->v_holdcnt); 1381 buf[0] = '\0'; 1382 if (vp->v_flag & VROOT) 1383 strcat(buf, "|VROOT"); 1384 if (vp->v_flag & VTEXT) 1385 strcat(buf, "|VTEXT"); 1386 if (vp->v_flag & VSYSTEM) 1387 strcat(buf, "|VSYSTEM"); 1388 if (vp->v_flag & VFREE) 1389 strcat(buf, "|VFREE"); 1390 if (vp->v_flag & VOBJBUF) 1391 strcat(buf, "|VOBJBUF"); 1392 if (buf[0] != '\0') 1393 printf(" flags (%s)", &buf[1]); 1394 if (vp->v_data == NULL) { 1395 printf("\n"); 1396 } else { 1397 printf("\n\t"); 1398 VOP_PRINT(vp); 1399 } 1400 } 1401 1402 #ifdef DDB 1403 #include <ddb/ddb.h> 1404 1405 static int db_show_locked_vnodes(struct mount *mp, void *data); 1406 1407 /* 1408 * List all of the locked vnodes in the system. 1409 * Called when debugging the kernel. 1410 */ 1411 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1412 { 1413 printf("Locked vnodes\n"); 1414 mountlist_scan(db_show_locked_vnodes, NULL, 1415 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1416 } 1417 1418 static int 1419 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1420 { 1421 struct vnode *vp; 1422 1423 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1424 if (VOP_ISLOCKED(vp, NULL)) 1425 vprint((char *)0, vp); 1426 } 1427 return(0); 1428 } 1429 #endif 1430 1431 /* 1432 * Top level filesystem related information gathering. 1433 */ 1434 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1435 1436 static int 1437 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1438 { 1439 int *name = (int *)arg1 - 1; /* XXX */ 1440 u_int namelen = arg2 + 1; /* XXX */ 1441 struct vfsconf *vfsp; 1442 1443 #if 1 || defined(COMPAT_PRELITE2) 1444 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1445 if (namelen == 1) 1446 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1447 #endif 1448 1449 #ifdef notyet 1450 /* all sysctl names at this level are at least name and field */ 1451 if (namelen < 2) 1452 return (ENOTDIR); /* overloaded */ 1453 if (name[0] != VFS_GENERIC) { 1454 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1455 if (vfsp->vfc_typenum == name[0]) 1456 break; 1457 if (vfsp == NULL) 1458 return (EOPNOTSUPP); 1459 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1460 oldp, oldlenp, newp, newlen, p)); 1461 } 1462 #endif 1463 switch (name[1]) { 1464 case VFS_MAXTYPENUM: 1465 if (namelen != 2) 1466 return (ENOTDIR); 1467 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1468 case VFS_CONF: 1469 if (namelen != 3) 1470 return (ENOTDIR); /* overloaded */ 1471 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1472 if (vfsp->vfc_typenum == name[2]) 1473 break; 1474 if (vfsp == NULL) 1475 return (EOPNOTSUPP); 1476 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1477 } 1478 return (EOPNOTSUPP); 1479 } 1480 1481 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1482 "Generic filesystem"); 1483 1484 #if 1 || defined(COMPAT_PRELITE2) 1485 1486 static int 1487 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1488 { 1489 int error; 1490 struct vfsconf *vfsp; 1491 struct ovfsconf ovfs; 1492 1493 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1494 bzero(&ovfs, sizeof(ovfs)); 1495 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1496 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1497 ovfs.vfc_index = vfsp->vfc_typenum; 1498 ovfs.vfc_refcount = vfsp->vfc_refcount; 1499 ovfs.vfc_flags = vfsp->vfc_flags; 1500 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1501 if (error) 1502 return error; 1503 } 1504 return 0; 1505 } 1506 1507 #endif /* 1 || COMPAT_PRELITE2 */ 1508 1509 /* 1510 * Check to see if a filesystem is mounted on a block device. 1511 */ 1512 int 1513 vfs_mountedon(struct vnode *vp) 1514 { 1515 dev_t dev; 1516 1517 if ((dev = vp->v_rdev) == NULL) 1518 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1519 if (dev != NODEV && dev->si_mountpoint) 1520 return (EBUSY); 1521 return (0); 1522 } 1523 1524 /* 1525 * Unmount all filesystems. The list is traversed in reverse order 1526 * of mounting to avoid dependencies. 1527 */ 1528 1529 static int vfs_umountall_callback(struct mount *mp, void *data); 1530 1531 void 1532 vfs_unmountall(void) 1533 { 1534 struct thread *td = curthread; 1535 int count; 1536 1537 if (td->td_proc == NULL) 1538 td = initproc->p_thread; /* XXX XXX use proc0 instead? */ 1539 1540 do { 1541 count = mountlist_scan(vfs_umountall_callback, 1542 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1543 } while (count); 1544 } 1545 1546 static 1547 int 1548 vfs_umountall_callback(struct mount *mp, void *data) 1549 { 1550 int error; 1551 1552 error = dounmount(mp, MNT_FORCE); 1553 if (error) { 1554 mountlist_remove(mp); 1555 printf("unmount of filesystem mounted from %s failed (", 1556 mp->mnt_stat.f_mntfromname); 1557 if (error == EBUSY) 1558 printf("BUSY)\n"); 1559 else 1560 printf("%d)\n", error); 1561 } 1562 return(1); 1563 } 1564 1565 /* 1566 * Build hash lists of net addresses and hang them off the mount point. 1567 * Called by ufs_mount() to set up the lists of export addresses. 1568 */ 1569 static int 1570 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1571 struct export_args *argp) 1572 { 1573 struct netcred *np; 1574 struct radix_node_head *rnh; 1575 int i; 1576 struct radix_node *rn; 1577 struct sockaddr *saddr, *smask = 0; 1578 struct domain *dom; 1579 int error; 1580 1581 if (argp->ex_addrlen == 0) { 1582 if (mp->mnt_flag & MNT_DEFEXPORTED) 1583 return (EPERM); 1584 np = &nep->ne_defexported; 1585 np->netc_exflags = argp->ex_flags; 1586 np->netc_anon = argp->ex_anon; 1587 np->netc_anon.cr_ref = 1; 1588 mp->mnt_flag |= MNT_DEFEXPORTED; 1589 return (0); 1590 } 1591 1592 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1593 return (EINVAL); 1594 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1595 return (EINVAL); 1596 1597 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1598 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1599 bzero((caddr_t) np, i); 1600 saddr = (struct sockaddr *) (np + 1); 1601 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1602 goto out; 1603 if (saddr->sa_len > argp->ex_addrlen) 1604 saddr->sa_len = argp->ex_addrlen; 1605 if (argp->ex_masklen) { 1606 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1607 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1608 if (error) 1609 goto out; 1610 if (smask->sa_len > argp->ex_masklen) 1611 smask->sa_len = argp->ex_masklen; 1612 } 1613 i = saddr->sa_family; 1614 if ((rnh = nep->ne_rtable[i]) == 0) { 1615 /* 1616 * Seems silly to initialize every AF when most are not used, 1617 * do so on demand here 1618 */ 1619 SLIST_FOREACH(dom, &domains, dom_next) 1620 if (dom->dom_family == i && dom->dom_rtattach) { 1621 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1622 dom->dom_rtoffset); 1623 break; 1624 } 1625 if ((rnh = nep->ne_rtable[i]) == 0) { 1626 error = ENOBUFS; 1627 goto out; 1628 } 1629 } 1630 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1631 np->netc_rnodes); 1632 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1633 error = EPERM; 1634 goto out; 1635 } 1636 np->netc_exflags = argp->ex_flags; 1637 np->netc_anon = argp->ex_anon; 1638 np->netc_anon.cr_ref = 1; 1639 return (0); 1640 out: 1641 free(np, M_NETADDR); 1642 return (error); 1643 } 1644 1645 /* ARGSUSED */ 1646 static int 1647 vfs_free_netcred(struct radix_node *rn, void *w) 1648 { 1649 struct radix_node_head *rnh = (struct radix_node_head *) w; 1650 1651 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1652 free((caddr_t) rn, M_NETADDR); 1653 return (0); 1654 } 1655 1656 /* 1657 * Free the net address hash lists that are hanging off the mount points. 1658 */ 1659 static void 1660 vfs_free_addrlist(struct netexport *nep) 1661 { 1662 int i; 1663 struct radix_node_head *rnh; 1664 1665 for (i = 0; i <= AF_MAX; i++) 1666 if ((rnh = nep->ne_rtable[i])) { 1667 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1668 (caddr_t) rnh); 1669 free((caddr_t) rnh, M_RTABLE); 1670 nep->ne_rtable[i] = 0; 1671 } 1672 } 1673 1674 int 1675 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1676 { 1677 int error; 1678 1679 if (argp->ex_flags & MNT_DELEXPORT) { 1680 if (mp->mnt_flag & MNT_EXPUBLIC) { 1681 vfs_setpublicfs(NULL, NULL, NULL); 1682 mp->mnt_flag &= ~MNT_EXPUBLIC; 1683 } 1684 vfs_free_addrlist(nep); 1685 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1686 } 1687 if (argp->ex_flags & MNT_EXPORTED) { 1688 if (argp->ex_flags & MNT_EXPUBLIC) { 1689 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1690 return (error); 1691 mp->mnt_flag |= MNT_EXPUBLIC; 1692 } 1693 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1694 return (error); 1695 mp->mnt_flag |= MNT_EXPORTED; 1696 } 1697 return (0); 1698 } 1699 1700 1701 /* 1702 * Set the publicly exported filesystem (WebNFS). Currently, only 1703 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1704 */ 1705 int 1706 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1707 struct export_args *argp) 1708 { 1709 int error; 1710 struct vnode *rvp; 1711 char *cp; 1712 1713 /* 1714 * mp == NULL -> invalidate the current info, the FS is 1715 * no longer exported. May be called from either vfs_export 1716 * or unmount, so check if it hasn't already been done. 1717 */ 1718 if (mp == NULL) { 1719 if (nfs_pub.np_valid) { 1720 nfs_pub.np_valid = 0; 1721 if (nfs_pub.np_index != NULL) { 1722 FREE(nfs_pub.np_index, M_TEMP); 1723 nfs_pub.np_index = NULL; 1724 } 1725 } 1726 return (0); 1727 } 1728 1729 /* 1730 * Only one allowed at a time. 1731 */ 1732 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1733 return (EBUSY); 1734 1735 /* 1736 * Get real filehandle for root of exported FS. 1737 */ 1738 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1739 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1740 1741 if ((error = VFS_ROOT(mp, &rvp))) 1742 return (error); 1743 1744 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1745 return (error); 1746 1747 vput(rvp); 1748 1749 /* 1750 * If an indexfile was specified, pull it in. 1751 */ 1752 if (argp->ex_indexfile != NULL) { 1753 int namelen; 1754 1755 error = vn_get_namelen(rvp, &namelen); 1756 if (error) 1757 return (error); 1758 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 1759 M_WAITOK); 1760 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1761 namelen, (size_t *)0); 1762 if (!error) { 1763 /* 1764 * Check for illegal filenames. 1765 */ 1766 for (cp = nfs_pub.np_index; *cp; cp++) { 1767 if (*cp == '/') { 1768 error = EINVAL; 1769 break; 1770 } 1771 } 1772 } 1773 if (error) { 1774 FREE(nfs_pub.np_index, M_TEMP); 1775 return (error); 1776 } 1777 } 1778 1779 nfs_pub.np_mount = mp; 1780 nfs_pub.np_valid = 1; 1781 return (0); 1782 } 1783 1784 struct netcred * 1785 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1786 struct sockaddr *nam) 1787 { 1788 struct netcred *np; 1789 struct radix_node_head *rnh; 1790 struct sockaddr *saddr; 1791 1792 np = NULL; 1793 if (mp->mnt_flag & MNT_EXPORTED) { 1794 /* 1795 * Lookup in the export list first. 1796 */ 1797 if (nam != NULL) { 1798 saddr = nam; 1799 rnh = nep->ne_rtable[saddr->sa_family]; 1800 if (rnh != NULL) { 1801 np = (struct netcred *) 1802 (*rnh->rnh_matchaddr)((char *)saddr, 1803 rnh); 1804 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1805 np = NULL; 1806 } 1807 } 1808 /* 1809 * If no address match, use the default if it exists. 1810 */ 1811 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1812 np = &nep->ne_defexported; 1813 } 1814 return (np); 1815 } 1816 1817 /* 1818 * perform msync on all vnodes under a mount point. The mount point must 1819 * be locked. This code is also responsible for lazy-freeing unreferenced 1820 * vnodes whos VM objects no longer contain pages. 1821 * 1822 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1823 * 1824 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 1825 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 1826 * way up in this high level function. 1827 */ 1828 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1829 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1830 1831 void 1832 vfs_msync(struct mount *mp, int flags) 1833 { 1834 int vmsc_flags; 1835 1836 vmsc_flags = VMSC_GETVP; 1837 if (flags != MNT_WAIT) 1838 vmsc_flags |= VMSC_NOWAIT; 1839 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 1840 (void *)flags); 1841 } 1842 1843 /* 1844 * scan1 is a fast pre-check. There could be hundreds of thousands of 1845 * vnodes, we cannot afford to do anything heavy weight until we have a 1846 * fairly good indication that there is work to do. 1847 */ 1848 static 1849 int 1850 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1851 { 1852 int flags = (int)data; 1853 1854 if ((vp->v_flag & VRECLAIMED) == 0) { 1855 if (vshouldfree(vp, 0)) 1856 return(0); /* call scan2 */ 1857 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1858 (vp->v_flag & VOBJDIRTY) && 1859 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1860 return(0); /* call scan2 */ 1861 } 1862 } 1863 1864 /* 1865 * do not call scan2, continue the loop 1866 */ 1867 return(-1); 1868 } 1869 1870 /* 1871 * This callback is handed a locked vnode. 1872 */ 1873 static 1874 int 1875 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1876 { 1877 vm_object_t obj; 1878 int flags = (int)data; 1879 1880 if (vp->v_flag & VRECLAIMED) 1881 return(0); 1882 1883 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 1884 if ((obj = vp->v_object) != NULL) { 1885 vm_object_page_clean(obj, 0, 0, 1886 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1887 } 1888 } 1889 return(0); 1890 } 1891 1892 /* 1893 * Record a process's interest in events which might happen to 1894 * a vnode. Because poll uses the historic select-style interface 1895 * internally, this routine serves as both the ``check for any 1896 * pending events'' and the ``record my interest in future events'' 1897 * functions. (These are done together, while the lock is held, 1898 * to avoid race conditions.) 1899 */ 1900 int 1901 vn_pollrecord(struct vnode *vp, int events) 1902 { 1903 lwkt_tokref ilock; 1904 1905 KKASSERT(curthread->td_proc != NULL); 1906 1907 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1908 if (vp->v_pollinfo.vpi_revents & events) { 1909 /* 1910 * This leaves events we are not interested 1911 * in available for the other process which 1912 * which presumably had requested them 1913 * (otherwise they would never have been 1914 * recorded). 1915 */ 1916 events &= vp->v_pollinfo.vpi_revents; 1917 vp->v_pollinfo.vpi_revents &= ~events; 1918 1919 lwkt_reltoken(&ilock); 1920 return events; 1921 } 1922 vp->v_pollinfo.vpi_events |= events; 1923 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo); 1924 lwkt_reltoken(&ilock); 1925 return 0; 1926 } 1927 1928 /* 1929 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1930 * it is possible for us to miss an event due to race conditions, but 1931 * that condition is expected to be rare, so for the moment it is the 1932 * preferred interface. 1933 */ 1934 void 1935 vn_pollevent(struct vnode *vp, int events) 1936 { 1937 lwkt_tokref ilock; 1938 1939 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1940 if (vp->v_pollinfo.vpi_events & events) { 1941 /* 1942 * We clear vpi_events so that we don't 1943 * call selwakeup() twice if two events are 1944 * posted before the polling process(es) is 1945 * awakened. This also ensures that we take at 1946 * most one selwakeup() if the polling process 1947 * is no longer interested. However, it does 1948 * mean that only one event can be noticed at 1949 * a time. (Perhaps we should only clear those 1950 * event bits which we note?) XXX 1951 */ 1952 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1953 vp->v_pollinfo.vpi_revents |= events; 1954 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1955 } 1956 lwkt_reltoken(&ilock); 1957 } 1958 1959 /* 1960 * Wake up anyone polling on vp because it is being revoked. 1961 * This depends on dead_poll() returning POLLHUP for correct 1962 * behavior. 1963 */ 1964 void 1965 vn_pollgone(struct vnode *vp) 1966 { 1967 lwkt_tokref ilock; 1968 1969 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1970 if (vp->v_pollinfo.vpi_events) { 1971 vp->v_pollinfo.vpi_events = 0; 1972 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1973 } 1974 lwkt_reltoken(&ilock); 1975 } 1976 1977 /* 1978 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened 1979 * (or v_rdev might be NULL). 1980 */ 1981 dev_t 1982 vn_todev(struct vnode *vp) 1983 { 1984 if (vp->v_type != VBLK && vp->v_type != VCHR) 1985 return (NODEV); 1986 KKASSERT(vp->v_rdev != NULL); 1987 return (vp->v_rdev); 1988 } 1989 1990 /* 1991 * Check if vnode represents a disk device. The vnode does not need to be 1992 * opened. 1993 */ 1994 int 1995 vn_isdisk(struct vnode *vp, int *errp) 1996 { 1997 dev_t dev; 1998 1999 if (vp->v_type != VBLK && vp->v_type != VCHR) { 2000 if (errp != NULL) 2001 *errp = ENOTBLK; 2002 return (0); 2003 } 2004 2005 if ((dev = vp->v_rdev) == NULL) 2006 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 2007 if (dev == NULL || dev == NODEV) { 2008 if (errp != NULL) 2009 *errp = ENXIO; 2010 return (0); 2011 } 2012 if (dev_is_good(dev) == 0) { 2013 if (errp != NULL) 2014 *errp = ENXIO; 2015 return (0); 2016 } 2017 if ((dev_dflags(dev) & D_DISK) == 0) { 2018 if (errp != NULL) 2019 *errp = ENOTBLK; 2020 return (0); 2021 } 2022 if (errp != NULL) 2023 *errp = 0; 2024 return (1); 2025 } 2026 2027 #ifdef DEBUG_VFS_LOCKS 2028 2029 void 2030 assert_vop_locked(struct vnode *vp, const char *str) 2031 { 2032 if (vp && !VOP_ISLOCKED(vp, NULL)) { 2033 panic("%s: %p is not locked shared but should be", str, vp); 2034 } 2035 } 2036 2037 void 2038 assert_vop_unlocked(struct vnode *vp, const char *str) 2039 { 2040 if (vp) { 2041 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) { 2042 panic("%s: %p is locked but should not be", str, vp); 2043 } 2044 } 2045 } 2046 2047 #endif 2048 2049 int 2050 vn_get_namelen(struct vnode *vp, int *namelen) 2051 { 2052 int error, retval[2]; 2053 2054 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2055 if (error) 2056 return (error); 2057 *namelen = *retval; 2058 return (0); 2059 } 2060 2061 int 2062 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2063 uint16_t d_namlen, const char *d_name) 2064 { 2065 struct dirent *dp; 2066 size_t len; 2067 2068 len = _DIRENT_RECLEN(d_namlen); 2069 if (len > uio->uio_resid) 2070 return(1); 2071 2072 dp = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 2073 2074 dp->d_ino = d_ino; 2075 dp->d_namlen = d_namlen; 2076 dp->d_type = d_type; 2077 bcopy(d_name, dp->d_name, d_namlen); 2078 2079 *error = uiomove((caddr_t)dp, len, uio); 2080 2081 free(dp, M_TEMP); 2082 2083 return(0); 2084 } 2085 2086