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 1.108 2007/11/02 19:52:25 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 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 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 398 (info->flags & V_SAVE)) { 399 if (bp->b_vp == info->vp) { 400 if (bp->b_flags & B_CLUSTEROK) { 401 vfs_bio_awrite(bp); 402 } else { 403 bremfree(bp); 404 bp->b_flags |= B_ASYNC; 405 bwrite(bp); 406 } 407 } else { 408 bremfree(bp); 409 bwrite(bp); 410 } 411 } else if (info->flags & V_SAVE) { 412 /* 413 * Cannot set B_NOCACHE on a clean buffer as this will 414 * destroy the VM backing store which might actually 415 * be dirty (and unsynchronized). 416 */ 417 bremfree(bp); 418 bp->b_flags |= (B_INVAL | B_RELBUF); 419 bp->b_flags &= ~B_ASYNC; 420 brelse(bp); 421 } else { 422 bremfree(bp); 423 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 424 bp->b_flags &= ~B_ASYNC; 425 brelse(bp); 426 } 427 return(0); 428 } 429 430 /* 431 * Truncate a file's buffer and pages to a specified length. This 432 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 433 * sync activity. 434 * 435 * The vnode must be locked. 436 */ 437 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 438 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 439 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 440 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 441 442 int 443 vtruncbuf(struct vnode *vp, off_t length, int blksize) 444 { 445 off_t truncloffset; 446 int count; 447 const char *filename; 448 449 /* 450 * Round up to the *next* block, then destroy the buffers in question. 451 * Since we are only removing some of the buffers we must rely on the 452 * scan count to determine whether a loop is necessary. 453 */ 454 if ((count = (int)(length % blksize)) != 0) 455 truncloffset = length + (blksize - count); 456 else 457 truncloffset = length; 458 459 crit_enter(); 460 do { 461 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 462 vtruncbuf_bp_trunc_cmp, 463 vtruncbuf_bp_trunc, &truncloffset); 464 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 465 vtruncbuf_bp_trunc_cmp, 466 vtruncbuf_bp_trunc, &truncloffset); 467 } while(count); 468 469 /* 470 * For safety, fsync any remaining metadata if the file is not being 471 * truncated to 0. Since the metadata does not represent the entire 472 * dirty list we have to rely on the hit count to ensure that we get 473 * all of it. 474 */ 475 if (length > 0) { 476 do { 477 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 478 vtruncbuf_bp_metasync_cmp, 479 vtruncbuf_bp_metasync, vp); 480 } while (count); 481 } 482 483 /* 484 * Clean out any left over VM backing store. 485 */ 486 crit_exit(); 487 488 vnode_pager_setsize(vp, length); 489 490 crit_enter(); 491 492 /* 493 * It is possible to have in-progress I/O from buffers that were 494 * not part of the truncation. This should not happen if we 495 * are truncating to 0-length. 496 */ 497 filename = TAILQ_FIRST(&vp->v_namecache) ? 498 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 499 500 while ((count = vp->v_track_write.bk_active) > 0) { 501 vp->v_track_write.bk_waitflag = 1; 502 tsleep(&vp->v_track_write, 0, "vbtrunc", 0); 503 if (length == 0) { 504 kprintf("Warning: vtruncbuf(): Had to wait for " 505 "%d buffer I/Os to finish in %s\n", 506 count, filename); 507 } 508 } 509 510 /* 511 * Make sure no buffers were instantiated while we were trying 512 * to clean out the remaining VM pages. This could occur due 513 * to busy dirty VM pages being flushed out to disk. 514 */ 515 do { 516 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 517 vtruncbuf_bp_trunc_cmp, 518 vtruncbuf_bp_trunc, &truncloffset); 519 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 520 vtruncbuf_bp_trunc_cmp, 521 vtruncbuf_bp_trunc, &truncloffset); 522 if (count) { 523 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 524 "left over buffers in %s\n", count, filename); 525 } 526 } while(count); 527 528 crit_exit(); 529 530 return (0); 531 } 532 533 /* 534 * The callback buffer is beyond the new file EOF and must be destroyed. 535 * Note that the compare function must conform to the RB_SCAN's requirements. 536 */ 537 static 538 int 539 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 540 { 541 if (bp->b_loffset >= *(off_t *)data) 542 return(0); 543 return(-1); 544 } 545 546 static 547 int 548 vtruncbuf_bp_trunc(struct buf *bp, void *data) 549 { 550 /* 551 * Do not try to use a buffer we cannot immediately lock, but sleep 552 * anyway to prevent a livelock. The code will loop until all buffers 553 * can be acted upon. 554 */ 555 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 556 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 557 BUF_UNLOCK(bp); 558 } else { 559 bremfree(bp); 560 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 561 bp->b_flags &= ~B_ASYNC; 562 brelse(bp); 563 } 564 return(1); 565 } 566 567 /* 568 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 569 * blocks (with a negative loffset) are scanned. 570 * Note that the compare function must conform to the RB_SCAN's requirements. 571 */ 572 static int 573 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 574 { 575 if (bp->b_loffset < 0) 576 return(0); 577 return(1); 578 } 579 580 static int 581 vtruncbuf_bp_metasync(struct buf *bp, void *data) 582 { 583 struct vnode *vp = data; 584 585 if (bp->b_flags & B_DELWRI) { 586 /* 587 * Do not try to use a buffer we cannot immediately lock, 588 * but sleep anyway to prevent a livelock. The code will 589 * loop until all buffers can be acted upon. 590 */ 591 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 592 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 593 BUF_UNLOCK(bp); 594 } else { 595 bremfree(bp); 596 if (bp->b_vp == vp) { 597 bp->b_flags |= B_ASYNC; 598 } else { 599 bp->b_flags &= ~B_ASYNC; 600 } 601 bwrite(bp); 602 } 603 return(1); 604 } else { 605 return(0); 606 } 607 } 608 609 /* 610 * vfsync - implements a multipass fsync on a file which understands 611 * dependancies and meta-data. The passed vnode must be locked. The 612 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 613 * 614 * When fsyncing data asynchronously just do one consolidated pass starting 615 * with the most negative block number. This may not get all the data due 616 * to dependancies. 617 * 618 * When fsyncing data synchronously do a data pass, then a metadata pass, 619 * then do additional data+metadata passes to try to get all the data out. 620 */ 621 static int vfsync_wait_output(struct vnode *vp, 622 int (*waitoutput)(struct vnode *, struct thread *)); 623 static int vfsync_data_only_cmp(struct buf *bp, void *data); 624 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 625 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 626 static int vfsync_bp(struct buf *bp, void *data); 627 628 struct vfsync_info { 629 struct vnode *vp; 630 int synchronous; 631 int syncdeps; 632 int lazycount; 633 int lazylimit; 634 int skippedbufs; 635 int (*checkdef)(struct buf *); 636 }; 637 638 int 639 vfsync(struct vnode *vp, int waitfor, int passes, 640 int (*checkdef)(struct buf *), 641 int (*waitoutput)(struct vnode *, struct thread *)) 642 { 643 struct vfsync_info info; 644 int error; 645 646 bzero(&info, sizeof(info)); 647 info.vp = vp; 648 if ((info.checkdef = checkdef) == NULL) 649 info.syncdeps = 1; 650 651 crit_enter_id("vfsync"); 652 653 switch(waitfor) { 654 case MNT_LAZY: 655 /* 656 * Lazy (filesystem syncer typ) Asynchronous plus limit the 657 * number of data (not meta) pages we try to flush to 1MB. 658 * A non-zero return means that lazy limit was reached. 659 */ 660 info.lazylimit = 1024 * 1024; 661 info.syncdeps = 1; 662 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 663 vfsync_lazy_range_cmp, vfsync_bp, &info); 664 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 665 vfsync_meta_only_cmp, vfsync_bp, &info); 666 if (error == 0) 667 vp->v_lazyw = 0; 668 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 669 vn_syncer_add_to_worklist(vp, 1); 670 error = 0; 671 break; 672 case MNT_NOWAIT: 673 /* 674 * Asynchronous. Do a data-only pass and a meta-only pass. 675 */ 676 info.syncdeps = 1; 677 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 678 vfsync_bp, &info); 679 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 680 vfsync_bp, &info); 681 error = 0; 682 break; 683 default: 684 /* 685 * Synchronous. Do a data-only pass, then a meta-data+data 686 * pass, then additional integrated passes to try to get 687 * all the dependancies flushed. 688 */ 689 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 690 vfsync_bp, &info); 691 error = vfsync_wait_output(vp, waitoutput); 692 if (error == 0) { 693 info.skippedbufs = 0; 694 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 695 vfsync_bp, &info); 696 error = vfsync_wait_output(vp, waitoutput); 697 if (info.skippedbufs) 698 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs); 699 } 700 while (error == 0 && passes > 0 && 701 !RB_EMPTY(&vp->v_rbdirty_tree)) { 702 if (--passes == 0) { 703 info.synchronous = 1; 704 info.syncdeps = 1; 705 } 706 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 707 vfsync_bp, &info); 708 if (error < 0) 709 error = -error; 710 info.syncdeps = 1; 711 if (error == 0) 712 error = vfsync_wait_output(vp, waitoutput); 713 } 714 break; 715 } 716 crit_exit_id("vfsync"); 717 return(error); 718 } 719 720 static int 721 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 722 { 723 int error = 0; 724 725 while (vp->v_track_write.bk_active) { 726 vp->v_track_write.bk_waitflag = 1; 727 tsleep(&vp->v_track_write, 0, "fsfsn", 0); 728 } 729 if (waitoutput) 730 error = waitoutput(vp, curthread); 731 return(error); 732 } 733 734 static int 735 vfsync_data_only_cmp(struct buf *bp, void *data) 736 { 737 if (bp->b_loffset < 0) 738 return(-1); 739 return(0); 740 } 741 742 static int 743 vfsync_meta_only_cmp(struct buf *bp, void *data) 744 { 745 if (bp->b_loffset < 0) 746 return(0); 747 return(1); 748 } 749 750 static int 751 vfsync_lazy_range_cmp(struct buf *bp, void *data) 752 { 753 struct vfsync_info *info = data; 754 if (bp->b_loffset < info->vp->v_lazyw) 755 return(-1); 756 return(0); 757 } 758 759 static int 760 vfsync_bp(struct buf *bp, void *data) 761 { 762 struct vfsync_info *info = data; 763 struct vnode *vp = info->vp; 764 int error; 765 766 /* 767 * if syncdeps is not set we do not try to write buffers which have 768 * dependancies. 769 */ 770 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 771 return(0); 772 773 /* 774 * Ignore buffers that we cannot immediately lock. XXX 775 */ 776 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 777 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp); 778 ++info->skippedbufs; 779 return(0); 780 } 781 if ((bp->b_flags & B_DELWRI) == 0) 782 panic("vfsync_bp: buffer not dirty"); 783 if (vp != bp->b_vp) 784 panic("vfsync_bp: buffer vp mismatch"); 785 786 /* 787 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 788 * has been written but an additional handshake with the device 789 * is required before we can dispose of the buffer. We have no idea 790 * how to do this so we have to skip these buffers. 791 */ 792 if (bp->b_flags & B_NEEDCOMMIT) { 793 BUF_UNLOCK(bp); 794 return(0); 795 } 796 797 if (info->synchronous) { 798 /* 799 * Synchronous flushing. An error may be returned. 800 */ 801 bremfree(bp); 802 crit_exit_id("vfsync"); 803 error = bwrite(bp); 804 crit_enter_id("vfsync"); 805 } else { 806 /* 807 * Asynchronous flushing. A negative return value simply 808 * stops the scan and is not considered an error. We use 809 * this to support limited MNT_LAZY flushes. 810 */ 811 vp->v_lazyw = bp->b_loffset; 812 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 813 info->lazycount += vfs_bio_awrite(bp); 814 } else { 815 info->lazycount += bp->b_bufsize; 816 bremfree(bp); 817 crit_exit_id("vfsync"); 818 bawrite(bp); 819 crit_enter_id("vfsync"); 820 } 821 if (info->lazylimit && info->lazycount >= info->lazylimit) 822 error = 1; 823 else 824 error = 0; 825 } 826 return(-error); 827 } 828 829 /* 830 * Associate a buffer with a vnode. 831 */ 832 void 833 bgetvp(struct vnode *vp, struct buf *bp) 834 { 835 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 836 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 837 838 vhold(vp); 839 /* 840 * Insert onto list for new vnode. 841 */ 842 crit_enter(); 843 bp->b_vp = vp; 844 bp->b_flags |= B_HASHED; 845 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) 846 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 847 848 bp->b_flags |= B_VNCLEAN; 849 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 850 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 851 crit_exit(); 852 } 853 854 /* 855 * Disassociate a buffer from a vnode. 856 */ 857 void 858 brelvp(struct buf *bp) 859 { 860 struct vnode *vp; 861 862 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 863 864 /* 865 * Delete from old vnode list, if on one. 866 */ 867 vp = bp->b_vp; 868 crit_enter(); 869 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 870 if (bp->b_flags & B_VNDIRTY) 871 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 872 else 873 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 874 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 875 } 876 if (bp->b_flags & B_HASHED) { 877 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 878 bp->b_flags &= ~B_HASHED; 879 } 880 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 881 vp->v_flag &= ~VONWORKLST; 882 LIST_REMOVE(vp, v_synclist); 883 } 884 crit_exit(); 885 bp->b_vp = NULL; 886 vdrop(vp); 887 } 888 889 /* 890 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 891 * This routine is called when the state of the B_DELWRI bit is changed. 892 */ 893 void 894 reassignbuf(struct buf *bp) 895 { 896 struct vnode *vp = bp->b_vp; 897 int delay; 898 899 KKASSERT(vp != NULL); 900 ++reassignbufcalls; 901 902 /* 903 * B_PAGING flagged buffers cannot be reassigned because their vp 904 * is not fully linked in. 905 */ 906 if (bp->b_flags & B_PAGING) 907 panic("cannot reassign paging buffer"); 908 909 crit_enter(); 910 if (bp->b_flags & B_DELWRI) { 911 /* 912 * Move to the dirty list, add the vnode to the worklist 913 */ 914 if (bp->b_flags & B_VNCLEAN) { 915 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 916 bp->b_flags &= ~B_VNCLEAN; 917 } 918 if ((bp->b_flags & B_VNDIRTY) == 0) { 919 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 920 panic("reassignbuf: dup lblk vp %p bp %p", 921 vp, bp); 922 } 923 bp->b_flags |= B_VNDIRTY; 924 } 925 if ((vp->v_flag & VONWORKLST) == 0) { 926 switch (vp->v_type) { 927 case VDIR: 928 delay = dirdelay; 929 break; 930 case VCHR: 931 case VBLK: 932 if (vp->v_rdev && 933 vp->v_rdev->si_mountpoint != NULL) { 934 delay = metadelay; 935 break; 936 } 937 /* fall through */ 938 default: 939 delay = filedelay; 940 } 941 vn_syncer_add_to_worklist(vp, delay); 942 } 943 } else { 944 /* 945 * Move to the clean list, remove the vnode from the worklist 946 * if no dirty blocks remain. 947 */ 948 if (bp->b_flags & B_VNDIRTY) { 949 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 950 bp->b_flags &= ~B_VNDIRTY; 951 } 952 if ((bp->b_flags & B_VNCLEAN) == 0) { 953 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 954 panic("reassignbuf: dup lblk vp %p bp %p", 955 vp, bp); 956 } 957 bp->b_flags |= B_VNCLEAN; 958 } 959 if ((vp->v_flag & VONWORKLST) && 960 RB_EMPTY(&vp->v_rbdirty_tree)) { 961 vp->v_flag &= ~VONWORKLST; 962 LIST_REMOVE(vp, v_synclist); 963 } 964 } 965 crit_exit(); 966 } 967 968 /* 969 * Create a vnode for a block device. 970 * Used for mounting the root file system. 971 */ 972 int 973 bdevvp(cdev_t dev, struct vnode **vpp) 974 { 975 struct vnode *vp; 976 struct vnode *nvp; 977 int error; 978 979 if (dev == NULL) { 980 *vpp = NULLVP; 981 return (ENXIO); 982 } 983 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0); 984 if (error) { 985 *vpp = NULLVP; 986 return (error); 987 } 988 vp = nvp; 989 vp->v_type = VCHR; 990 vp->v_umajor = dev->si_umajor; 991 vp->v_uminor = dev->si_uminor; 992 vx_unlock(vp); 993 *vpp = vp; 994 return (0); 995 } 996 997 int 998 v_associate_rdev(struct vnode *vp, cdev_t dev) 999 { 1000 lwkt_tokref ilock; 1001 1002 if (dev == NULL) 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 kprintf("Z1"); 1009 vp->v_rdev = reference_dev(dev); 1010 lwkt_gettoken(&ilock, &spechash_token); 1011 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1012 lwkt_reltoken(&ilock); 1013 return(0); 1014 } 1015 1016 void 1017 v_release_rdev(struct vnode *vp) 1018 { 1019 lwkt_tokref ilock; 1020 cdev_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_cdevnext); 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 cdev_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, int x, int y) 1039 { 1040 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1041 panic("addaliasu on non-special vnode"); 1042 nvp->v_umajor = x; 1043 nvp->v_uminor = y; 1044 } 1045 1046 /* 1047 * Disassociate a vnode from its underlying filesystem. 1048 * 1049 * The vnode must be VX locked and referenced. In all normal situations 1050 * there are no active references. If vclean_vxlocked() is called while 1051 * there are active references, the vnode is being ripped out and we have 1052 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1053 */ 1054 void 1055 vclean_vxlocked(struct vnode *vp, int flags) 1056 { 1057 int active; 1058 int n; 1059 vm_object_t object; 1060 1061 /* 1062 * If the vnode has already been reclaimed we have nothing to do. 1063 */ 1064 if (vp->v_flag & VRECLAIMED) 1065 return; 1066 vp->v_flag |= VRECLAIMED; 1067 1068 /* 1069 * Scrap the vfs cache 1070 */ 1071 while (cache_inval_vp(vp, 0) != 0) { 1072 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1073 tsleep(vp, 0, "vclninv", 2); 1074 } 1075 1076 /* 1077 * Check to see if the vnode is in use. If so we have to reference it 1078 * before we clean it out so that its count cannot fall to zero and 1079 * generate a race against ourselves to recycle it. 1080 */ 1081 active = sysref_isactive(&vp->v_sysref); 1082 1083 /* 1084 * Clean out any buffers associated with the vnode and destroy its 1085 * object, if it has one. 1086 */ 1087 vinvalbuf(vp, V_SAVE, 0, 0); 1088 1089 /* 1090 * If purging an active vnode (typically during a forced unmount 1091 * or reboot), it must be closed and deactivated before being 1092 * reclaimed. This isn't really all that safe, but what can 1093 * we do? XXX. 1094 * 1095 * Note that neither of these routines unlocks the vnode. 1096 */ 1097 if (active && (flags & DOCLOSE)) { 1098 while ((n = vp->v_opencount) != 0) { 1099 if (vp->v_writecount) 1100 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1101 else 1102 VOP_CLOSE(vp, FNONBLOCK); 1103 if (vp->v_opencount == n) { 1104 kprintf("Warning: unable to force-close" 1105 " vnode %p\n", vp); 1106 break; 1107 } 1108 } 1109 } 1110 1111 /* 1112 * If the vnode has not been deactivated, deactivated it. Deactivation 1113 * can create new buffers and VM pages so we have to call vinvalbuf() 1114 * again to make sure they all get flushed. 1115 * 1116 * This can occur if a file with a link count of 0 needs to be 1117 * truncated. 1118 */ 1119 if ((vp->v_flag & VINACTIVE) == 0) { 1120 vp->v_flag |= VINACTIVE; 1121 VOP_INACTIVE(vp); 1122 vinvalbuf(vp, V_SAVE, 0, 0); 1123 } 1124 1125 /* 1126 * If the vnode has an object, destroy it. 1127 */ 1128 if ((object = vp->v_object) != NULL) { 1129 if (object->ref_count == 0) { 1130 if ((object->flags & OBJ_DEAD) == 0) 1131 vm_object_terminate(object); 1132 } else { 1133 vm_pager_deallocate(object); 1134 } 1135 vp->v_flag &= ~VOBJBUF; 1136 } 1137 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1138 1139 /* 1140 * Reclaim the vnode. 1141 */ 1142 if (VOP_RECLAIM(vp)) 1143 panic("vclean: cannot reclaim"); 1144 1145 /* 1146 * Done with purge, notify sleepers of the grim news. 1147 */ 1148 vp->v_ops = &dead_vnode_vops_p; 1149 vn_pollgone(vp); 1150 vp->v_tag = VT_NON; 1151 1152 /* 1153 * If we are destroying an active vnode, reactivate it now that 1154 * we have reassociated it with deadfs. This prevents the system 1155 * from crashing on the vnode due to it being unexpectedly marked 1156 * as inactive or reclaimed. 1157 */ 1158 if (active && (flags & DOCLOSE)) { 1159 vp->v_flag &= ~(VINACTIVE|VRECLAIMED); 1160 } 1161 } 1162 1163 /* 1164 * Eliminate all activity associated with the requested vnode 1165 * and with all vnodes aliased to the requested vnode. 1166 * 1167 * The vnode must be referenced and vx_lock()'d 1168 * 1169 * revoke { struct vnode *a_vp, int a_flags } 1170 */ 1171 int 1172 vop_stdrevoke(struct vop_revoke_args *ap) 1173 { 1174 struct vnode *vp, *vq; 1175 lwkt_tokref ilock; 1176 cdev_t dev; 1177 1178 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1179 1180 vp = ap->a_vp; 1181 1182 /* 1183 * If the vnode is already dead don't try to revoke it 1184 */ 1185 if (vp->v_flag & VRECLAIMED) 1186 return (0); 1187 1188 /* 1189 * If the vnode has a device association, scrap all vnodes associated 1190 * with the device. Don't let the device disappear on us while we 1191 * are scrapping the vnodes. 1192 * 1193 * The passed vp will probably show up in the list, do not VX lock 1194 * it twice! 1195 */ 1196 if (vp->v_type != VCHR) 1197 return(0); 1198 if ((dev = vp->v_rdev) == NULL) { 1199 if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL) 1200 return(0); 1201 } 1202 reference_dev(dev); 1203 lwkt_gettoken(&ilock, &spechash_token); 1204 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1205 if (vp != vq) 1206 vx_get(vq); 1207 if (vq == SLIST_FIRST(&dev->si_hlist)) 1208 vgone_vxlocked(vq); 1209 if (vp != vq) 1210 vx_put(vq); 1211 } 1212 lwkt_reltoken(&ilock); 1213 release_dev(dev); 1214 return (0); 1215 } 1216 1217 /* 1218 * This is called when the object underlying a vnode is being destroyed, 1219 * such as in a remove(). Try to recycle the vnode immediately if the 1220 * only active reference is our reference. 1221 */ 1222 int 1223 vrecycle(struct vnode *vp) 1224 { 1225 if (vp->v_sysref.refcnt <= 1) { 1226 vgone_vxlocked(vp); 1227 return (1); 1228 } 1229 return (0); 1230 } 1231 1232 /* 1233 * Eliminate all activity associated with a vnode in preparation for reuse. 1234 * 1235 * The vnode must be VX locked and refd and will remain VX locked and refd 1236 * on return. This routine may be called with the vnode in any state, as 1237 * long as it is VX locked. The vnode will be cleaned out and marked 1238 * VRECLAIMED but will not actually be reused until all existing refs and 1239 * holds go away. 1240 * 1241 * NOTE: This routine may be called on a vnode which has not yet been 1242 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1243 * already been reclaimed. 1244 * 1245 * This routine is not responsible for placing us back on the freelist. 1246 * Instead, it happens automatically when the caller releases the VX lock 1247 * (assuming there aren't any other references). 1248 */ 1249 1250 void 1251 vgone_vxlocked(struct vnode *vp) 1252 { 1253 /* 1254 * assert that the VX lock is held. This is an absolute requirement 1255 * now for vgone_vxlocked() to be called. 1256 */ 1257 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1258 1259 /* 1260 * Clean out the filesystem specific data and set the VRECLAIMED 1261 * bit. Also deactivate the vnode if necessary. 1262 */ 1263 vclean_vxlocked(vp, DOCLOSE); 1264 1265 /* 1266 * Delete from old mount point vnode list, if on one. 1267 */ 1268 if (vp->v_mount != NULL) 1269 insmntque(vp, NULL); 1270 1271 /* 1272 * If special device, remove it from special device alias list 1273 * if it is on one. This should normally only occur if a vnode is 1274 * being revoked as the device should otherwise have been released 1275 * naturally. 1276 */ 1277 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1278 v_release_rdev(vp); 1279 } 1280 1281 /* 1282 * Set us to VBAD 1283 */ 1284 vp->v_type = VBAD; 1285 } 1286 1287 /* 1288 * Lookup a vnode by device number. 1289 */ 1290 int 1291 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1292 { 1293 lwkt_tokref ilock; 1294 struct vnode *vp; 1295 1296 lwkt_gettoken(&ilock, &spechash_token); 1297 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1298 if (type == vp->v_type) { 1299 *vpp = vp; 1300 lwkt_reltoken(&ilock); 1301 return (1); 1302 } 1303 } 1304 lwkt_reltoken(&ilock); 1305 return (0); 1306 } 1307 1308 /* 1309 * Calculate the total number of references to a special device. This 1310 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1311 * an overloaded field. Since udev2dev can now return NULL, we have 1312 * to check for a NULL v_rdev. 1313 */ 1314 int 1315 count_dev(cdev_t dev) 1316 { 1317 lwkt_tokref ilock; 1318 struct vnode *vp; 1319 int count = 0; 1320 1321 if (SLIST_FIRST(&dev->si_hlist)) { 1322 lwkt_gettoken(&ilock, &spechash_token); 1323 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1324 if (vp->v_sysref.refcnt > 0) 1325 count += vp->v_sysref.refcnt; 1326 } 1327 lwkt_reltoken(&ilock); 1328 } 1329 return(count); 1330 } 1331 1332 int 1333 count_udev(int x, int y) 1334 { 1335 cdev_t dev; 1336 1337 if ((dev = get_dev(x, y)) == NULL) 1338 return(0); 1339 return(count_dev(dev)); 1340 } 1341 1342 int 1343 vcount(struct vnode *vp) 1344 { 1345 if (vp->v_rdev == NULL) 1346 return(0); 1347 return(count_dev(vp->v_rdev)); 1348 } 1349 1350 /* 1351 * Initialize VMIO for a vnode. This routine MUST be called before a 1352 * VFS can issue buffer cache ops on a vnode. It is typically called 1353 * when a vnode is initialized from its inode. 1354 */ 1355 int 1356 vinitvmio(struct vnode *vp, off_t filesize) 1357 { 1358 vm_object_t object; 1359 int error = 0; 1360 1361 retry: 1362 if ((object = vp->v_object) == NULL) { 1363 object = vnode_pager_alloc(vp, filesize, 0, 0); 1364 /* 1365 * Dereference the reference we just created. This assumes 1366 * that the object is associated with the vp. 1367 */ 1368 object->ref_count--; 1369 vrele(vp); 1370 } else { 1371 if (object->flags & OBJ_DEAD) { 1372 vn_unlock(vp); 1373 vm_object_dead_sleep(object, "vodead"); 1374 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1375 goto retry; 1376 } 1377 } 1378 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1379 vp->v_flag |= VOBJBUF; 1380 return (error); 1381 } 1382 1383 1384 /* 1385 * Print out a description of a vnode. 1386 */ 1387 static char *typename[] = 1388 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1389 1390 void 1391 vprint(char *label, struct vnode *vp) 1392 { 1393 char buf[96]; 1394 1395 if (label != NULL) 1396 kprintf("%s: %p: ", label, (void *)vp); 1397 else 1398 kprintf("%p: ", (void *)vp); 1399 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1400 typename[vp->v_type], 1401 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1402 buf[0] = '\0'; 1403 if (vp->v_flag & VROOT) 1404 strcat(buf, "|VROOT"); 1405 if (vp->v_flag & VTEXT) 1406 strcat(buf, "|VTEXT"); 1407 if (vp->v_flag & VSYSTEM) 1408 strcat(buf, "|VSYSTEM"); 1409 if (vp->v_flag & VFREE) 1410 strcat(buf, "|VFREE"); 1411 if (vp->v_flag & VOBJBUF) 1412 strcat(buf, "|VOBJBUF"); 1413 if (buf[0] != '\0') 1414 kprintf(" flags (%s)", &buf[1]); 1415 if (vp->v_data == NULL) { 1416 kprintf("\n"); 1417 } else { 1418 kprintf("\n\t"); 1419 VOP_PRINT(vp); 1420 } 1421 } 1422 1423 #ifdef DDB 1424 #include <ddb/ddb.h> 1425 1426 static int db_show_locked_vnodes(struct mount *mp, void *data); 1427 1428 /* 1429 * List all of the locked vnodes in the system. 1430 * Called when debugging the kernel. 1431 */ 1432 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1433 { 1434 kprintf("Locked vnodes\n"); 1435 mountlist_scan(db_show_locked_vnodes, NULL, 1436 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1437 } 1438 1439 static int 1440 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1441 { 1442 struct vnode *vp; 1443 1444 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1445 if (vn_islocked(vp)) 1446 vprint((char *)0, vp); 1447 } 1448 return(0); 1449 } 1450 #endif 1451 1452 /* 1453 * Top level filesystem related information gathering. 1454 */ 1455 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1456 1457 static int 1458 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1459 { 1460 int *name = (int *)arg1 - 1; /* XXX */ 1461 u_int namelen = arg2 + 1; /* XXX */ 1462 struct vfsconf *vfsp; 1463 1464 #if 1 || defined(COMPAT_PRELITE2) 1465 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1466 if (namelen == 1) 1467 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1468 #endif 1469 1470 #ifdef notyet 1471 /* all sysctl names at this level are at least name and field */ 1472 if (namelen < 2) 1473 return (ENOTDIR); /* overloaded */ 1474 if (name[0] != VFS_GENERIC) { 1475 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1476 if (vfsp->vfc_typenum == name[0]) 1477 break; 1478 if (vfsp == NULL) 1479 return (EOPNOTSUPP); 1480 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1481 oldp, oldlenp, newp, newlen, p)); 1482 } 1483 #endif 1484 switch (name[1]) { 1485 case VFS_MAXTYPENUM: 1486 if (namelen != 2) 1487 return (ENOTDIR); 1488 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1489 case VFS_CONF: 1490 if (namelen != 3) 1491 return (ENOTDIR); /* overloaded */ 1492 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1493 if (vfsp->vfc_typenum == name[2]) 1494 break; 1495 if (vfsp == NULL) 1496 return (EOPNOTSUPP); 1497 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1498 } 1499 return (EOPNOTSUPP); 1500 } 1501 1502 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1503 "Generic filesystem"); 1504 1505 #if 1 || defined(COMPAT_PRELITE2) 1506 1507 static int 1508 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1509 { 1510 int error; 1511 struct vfsconf *vfsp; 1512 struct ovfsconf ovfs; 1513 1514 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1515 bzero(&ovfs, sizeof(ovfs)); 1516 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1517 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1518 ovfs.vfc_index = vfsp->vfc_typenum; 1519 ovfs.vfc_refcount = vfsp->vfc_refcount; 1520 ovfs.vfc_flags = vfsp->vfc_flags; 1521 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1522 if (error) 1523 return error; 1524 } 1525 return 0; 1526 } 1527 1528 #endif /* 1 || COMPAT_PRELITE2 */ 1529 1530 /* 1531 * Check to see if a filesystem is mounted on a block device. 1532 */ 1533 int 1534 vfs_mountedon(struct vnode *vp) 1535 { 1536 cdev_t dev; 1537 1538 if ((dev = vp->v_rdev) == NULL) { 1539 if (vp->v_type != VBLK) 1540 dev = get_dev(vp->v_uminor, vp->v_umajor); 1541 } 1542 if (dev != NULL && dev->si_mountpoint) 1543 return (EBUSY); 1544 return (0); 1545 } 1546 1547 /* 1548 * Unmount all filesystems. The list is traversed in reverse order 1549 * of mounting to avoid dependencies. 1550 */ 1551 1552 static int vfs_umountall_callback(struct mount *mp, void *data); 1553 1554 void 1555 vfs_unmountall(void) 1556 { 1557 int count; 1558 1559 do { 1560 count = mountlist_scan(vfs_umountall_callback, 1561 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1562 } while (count); 1563 } 1564 1565 static 1566 int 1567 vfs_umountall_callback(struct mount *mp, void *data) 1568 { 1569 int error; 1570 1571 error = dounmount(mp, MNT_FORCE); 1572 if (error) { 1573 mountlist_remove(mp); 1574 kprintf("unmount of filesystem mounted from %s failed (", 1575 mp->mnt_stat.f_mntfromname); 1576 if (error == EBUSY) 1577 kprintf("BUSY)\n"); 1578 else 1579 kprintf("%d)\n", error); 1580 } 1581 return(1); 1582 } 1583 1584 /* 1585 * Build hash lists of net addresses and hang them off the mount point. 1586 * Called by ufs_mount() to set up the lists of export addresses. 1587 */ 1588 static int 1589 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1590 struct export_args *argp) 1591 { 1592 struct netcred *np; 1593 struct radix_node_head *rnh; 1594 int i; 1595 struct radix_node *rn; 1596 struct sockaddr *saddr, *smask = 0; 1597 struct domain *dom; 1598 int error; 1599 1600 if (argp->ex_addrlen == 0) { 1601 if (mp->mnt_flag & MNT_DEFEXPORTED) 1602 return (EPERM); 1603 np = &nep->ne_defexported; 1604 np->netc_exflags = argp->ex_flags; 1605 np->netc_anon = argp->ex_anon; 1606 np->netc_anon.cr_ref = 1; 1607 mp->mnt_flag |= MNT_DEFEXPORTED; 1608 return (0); 1609 } 1610 1611 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1612 return (EINVAL); 1613 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1614 return (EINVAL); 1615 1616 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1617 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK); 1618 bzero((caddr_t) np, i); 1619 saddr = (struct sockaddr *) (np + 1); 1620 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1621 goto out; 1622 if (saddr->sa_len > argp->ex_addrlen) 1623 saddr->sa_len = argp->ex_addrlen; 1624 if (argp->ex_masklen) { 1625 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1626 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1627 if (error) 1628 goto out; 1629 if (smask->sa_len > argp->ex_masklen) 1630 smask->sa_len = argp->ex_masklen; 1631 } 1632 i = saddr->sa_family; 1633 if ((rnh = nep->ne_rtable[i]) == 0) { 1634 /* 1635 * Seems silly to initialize every AF when most are not used, 1636 * do so on demand here 1637 */ 1638 SLIST_FOREACH(dom, &domains, dom_next) 1639 if (dom->dom_family == i && dom->dom_rtattach) { 1640 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1641 dom->dom_rtoffset); 1642 break; 1643 } 1644 if ((rnh = nep->ne_rtable[i]) == 0) { 1645 error = ENOBUFS; 1646 goto out; 1647 } 1648 } 1649 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1650 np->netc_rnodes); 1651 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1652 error = EPERM; 1653 goto out; 1654 } 1655 np->netc_exflags = argp->ex_flags; 1656 np->netc_anon = argp->ex_anon; 1657 np->netc_anon.cr_ref = 1; 1658 return (0); 1659 out: 1660 kfree(np, M_NETADDR); 1661 return (error); 1662 } 1663 1664 /* ARGSUSED */ 1665 static int 1666 vfs_free_netcred(struct radix_node *rn, void *w) 1667 { 1668 struct radix_node_head *rnh = (struct radix_node_head *) w; 1669 1670 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1671 kfree((caddr_t) rn, M_NETADDR); 1672 return (0); 1673 } 1674 1675 /* 1676 * Free the net address hash lists that are hanging off the mount points. 1677 */ 1678 static void 1679 vfs_free_addrlist(struct netexport *nep) 1680 { 1681 int i; 1682 struct radix_node_head *rnh; 1683 1684 for (i = 0; i <= AF_MAX; i++) 1685 if ((rnh = nep->ne_rtable[i])) { 1686 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1687 (caddr_t) rnh); 1688 kfree((caddr_t) rnh, M_RTABLE); 1689 nep->ne_rtable[i] = 0; 1690 } 1691 } 1692 1693 int 1694 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1695 { 1696 int error; 1697 1698 if (argp->ex_flags & MNT_DELEXPORT) { 1699 if (mp->mnt_flag & MNT_EXPUBLIC) { 1700 vfs_setpublicfs(NULL, NULL, NULL); 1701 mp->mnt_flag &= ~MNT_EXPUBLIC; 1702 } 1703 vfs_free_addrlist(nep); 1704 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1705 } 1706 if (argp->ex_flags & MNT_EXPORTED) { 1707 if (argp->ex_flags & MNT_EXPUBLIC) { 1708 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1709 return (error); 1710 mp->mnt_flag |= MNT_EXPUBLIC; 1711 } 1712 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1713 return (error); 1714 mp->mnt_flag |= MNT_EXPORTED; 1715 } 1716 return (0); 1717 } 1718 1719 1720 /* 1721 * Set the publicly exported filesystem (WebNFS). Currently, only 1722 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1723 */ 1724 int 1725 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1726 struct export_args *argp) 1727 { 1728 int error; 1729 struct vnode *rvp; 1730 char *cp; 1731 1732 /* 1733 * mp == NULL -> invalidate the current info, the FS is 1734 * no longer exported. May be called from either vfs_export 1735 * or unmount, so check if it hasn't already been done. 1736 */ 1737 if (mp == NULL) { 1738 if (nfs_pub.np_valid) { 1739 nfs_pub.np_valid = 0; 1740 if (nfs_pub.np_index != NULL) { 1741 FREE(nfs_pub.np_index, M_TEMP); 1742 nfs_pub.np_index = NULL; 1743 } 1744 } 1745 return (0); 1746 } 1747 1748 /* 1749 * Only one allowed at a time. 1750 */ 1751 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1752 return (EBUSY); 1753 1754 /* 1755 * Get real filehandle for root of exported FS. 1756 */ 1757 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1758 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1759 1760 if ((error = VFS_ROOT(mp, &rvp))) 1761 return (error); 1762 1763 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1764 return (error); 1765 1766 vput(rvp); 1767 1768 /* 1769 * If an indexfile was specified, pull it in. 1770 */ 1771 if (argp->ex_indexfile != NULL) { 1772 int namelen; 1773 1774 error = vn_get_namelen(rvp, &namelen); 1775 if (error) 1776 return (error); 1777 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 1778 M_WAITOK); 1779 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1780 namelen, (size_t *)0); 1781 if (!error) { 1782 /* 1783 * Check for illegal filenames. 1784 */ 1785 for (cp = nfs_pub.np_index; *cp; cp++) { 1786 if (*cp == '/') { 1787 error = EINVAL; 1788 break; 1789 } 1790 } 1791 } 1792 if (error) { 1793 FREE(nfs_pub.np_index, M_TEMP); 1794 return (error); 1795 } 1796 } 1797 1798 nfs_pub.np_mount = mp; 1799 nfs_pub.np_valid = 1; 1800 return (0); 1801 } 1802 1803 struct netcred * 1804 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1805 struct sockaddr *nam) 1806 { 1807 struct netcred *np; 1808 struct radix_node_head *rnh; 1809 struct sockaddr *saddr; 1810 1811 np = NULL; 1812 if (mp->mnt_flag & MNT_EXPORTED) { 1813 /* 1814 * Lookup in the export list first. 1815 */ 1816 if (nam != NULL) { 1817 saddr = nam; 1818 rnh = nep->ne_rtable[saddr->sa_family]; 1819 if (rnh != NULL) { 1820 np = (struct netcred *) 1821 (*rnh->rnh_matchaddr)((char *)saddr, 1822 rnh); 1823 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1824 np = NULL; 1825 } 1826 } 1827 /* 1828 * If no address match, use the default if it exists. 1829 */ 1830 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1831 np = &nep->ne_defexported; 1832 } 1833 return (np); 1834 } 1835 1836 /* 1837 * perform msync on all vnodes under a mount point. The mount point must 1838 * be locked. This code is also responsible for lazy-freeing unreferenced 1839 * vnodes whos VM objects no longer contain pages. 1840 * 1841 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1842 * 1843 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 1844 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 1845 * way up in this high level function. 1846 */ 1847 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1848 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1849 1850 void 1851 vfs_msync(struct mount *mp, int flags) 1852 { 1853 int vmsc_flags; 1854 1855 vmsc_flags = VMSC_GETVP; 1856 if (flags != MNT_WAIT) 1857 vmsc_flags |= VMSC_NOWAIT; 1858 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 1859 (void *)flags); 1860 } 1861 1862 /* 1863 * scan1 is a fast pre-check. There could be hundreds of thousands of 1864 * vnodes, we cannot afford to do anything heavy weight until we have a 1865 * fairly good indication that there is work to do. 1866 */ 1867 static 1868 int 1869 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1870 { 1871 int flags = (int)data; 1872 1873 if ((vp->v_flag & VRECLAIMED) == 0) { 1874 if (vshouldmsync(vp)) 1875 return(0); /* call scan2 */ 1876 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1877 (vp->v_flag & VOBJDIRTY) && 1878 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 1879 return(0); /* call scan2 */ 1880 } 1881 } 1882 1883 /* 1884 * do not call scan2, continue the loop 1885 */ 1886 return(-1); 1887 } 1888 1889 /* 1890 * This callback is handed a locked vnode. 1891 */ 1892 static 1893 int 1894 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1895 { 1896 vm_object_t obj; 1897 int flags = (int)data; 1898 1899 if (vp->v_flag & VRECLAIMED) 1900 return(0); 1901 1902 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 1903 if ((obj = vp->v_object) != NULL) { 1904 vm_object_page_clean(obj, 0, 0, 1905 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1906 } 1907 } 1908 return(0); 1909 } 1910 1911 /* 1912 * Record a process's interest in events which might happen to 1913 * a vnode. Because poll uses the historic select-style interface 1914 * internally, this routine serves as both the ``check for any 1915 * pending events'' and the ``record my interest in future events'' 1916 * functions. (These are done together, while the lock is held, 1917 * to avoid race conditions.) 1918 */ 1919 int 1920 vn_pollrecord(struct vnode *vp, int events) 1921 { 1922 lwkt_tokref ilock; 1923 1924 KKASSERT(curthread->td_proc != NULL); 1925 1926 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1927 if (vp->v_pollinfo.vpi_revents & events) { 1928 /* 1929 * This leaves events we are not interested 1930 * in available for the other process which 1931 * which presumably had requested them 1932 * (otherwise they would never have been 1933 * recorded). 1934 */ 1935 events &= vp->v_pollinfo.vpi_revents; 1936 vp->v_pollinfo.vpi_revents &= ~events; 1937 1938 lwkt_reltoken(&ilock); 1939 return events; 1940 } 1941 vp->v_pollinfo.vpi_events |= events; 1942 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo); 1943 lwkt_reltoken(&ilock); 1944 return 0; 1945 } 1946 1947 /* 1948 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1949 * it is possible for us to miss an event due to race conditions, but 1950 * that condition is expected to be rare, so for the moment it is the 1951 * preferred interface. 1952 */ 1953 void 1954 vn_pollevent(struct vnode *vp, int events) 1955 { 1956 lwkt_tokref ilock; 1957 1958 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1959 if (vp->v_pollinfo.vpi_events & events) { 1960 /* 1961 * We clear vpi_events so that we don't 1962 * call selwakeup() twice if two events are 1963 * posted before the polling process(es) is 1964 * awakened. This also ensures that we take at 1965 * most one selwakeup() if the polling process 1966 * is no longer interested. However, it does 1967 * mean that only one event can be noticed at 1968 * a time. (Perhaps we should only clear those 1969 * event bits which we note?) XXX 1970 */ 1971 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1972 vp->v_pollinfo.vpi_revents |= events; 1973 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1974 } 1975 lwkt_reltoken(&ilock); 1976 } 1977 1978 /* 1979 * Wake up anyone polling on vp because it is being revoked. 1980 * This depends on dead_poll() returning POLLHUP for correct 1981 * behavior. 1982 */ 1983 void 1984 vn_pollgone(struct vnode *vp) 1985 { 1986 lwkt_tokref ilock; 1987 1988 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1989 if (vp->v_pollinfo.vpi_events) { 1990 vp->v_pollinfo.vpi_events = 0; 1991 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1992 } 1993 lwkt_reltoken(&ilock); 1994 } 1995 1996 /* 1997 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 1998 * (or v_rdev might be NULL). 1999 */ 2000 cdev_t 2001 vn_todev(struct vnode *vp) 2002 { 2003 if (vp->v_type != VBLK && vp->v_type != VCHR) 2004 return (NULL); 2005 KKASSERT(vp->v_rdev != NULL); 2006 return (vp->v_rdev); 2007 } 2008 2009 /* 2010 * Check if vnode represents a disk device. The vnode does not need to be 2011 * opened. 2012 */ 2013 int 2014 vn_isdisk(struct vnode *vp, int *errp) 2015 { 2016 cdev_t dev; 2017 2018 if (vp->v_type != VCHR) { 2019 if (errp != NULL) 2020 *errp = ENOTBLK; 2021 return (0); 2022 } 2023 2024 if ((dev = vp->v_rdev) == NULL) 2025 dev = get_dev(vp->v_umajor, vp->v_uminor); 2026 2027 if (dev == NULL) { 2028 if (errp != NULL) 2029 *errp = ENXIO; 2030 return (0); 2031 } 2032 if (dev_is_good(dev) == 0) { 2033 if (errp != NULL) 2034 *errp = ENXIO; 2035 return (0); 2036 } 2037 if ((dev_dflags(dev) & D_DISK) == 0) { 2038 if (errp != NULL) 2039 *errp = ENOTBLK; 2040 return (0); 2041 } 2042 if (errp != NULL) 2043 *errp = 0; 2044 return (1); 2045 } 2046 2047 int 2048 vn_get_namelen(struct vnode *vp, int *namelen) 2049 { 2050 int error, retval[2]; 2051 2052 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2053 if (error) 2054 return (error); 2055 *namelen = *retval; 2056 return (0); 2057 } 2058 2059 int 2060 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2061 uint16_t d_namlen, const char *d_name) 2062 { 2063 struct dirent *dp; 2064 size_t len; 2065 2066 len = _DIRENT_RECLEN(d_namlen); 2067 if (len > uio->uio_resid) 2068 return(1); 2069 2070 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2071 2072 dp->d_ino = d_ino; 2073 dp->d_namlen = d_namlen; 2074 dp->d_type = d_type; 2075 bcopy(d_name, dp->d_name, d_namlen); 2076 2077 *error = uiomove((caddr_t)dp, len, uio); 2078 2079 kfree(dp, M_TEMP); 2080 2081 return(0); 2082 } 2083 2084