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 1.109 2007/11/07 00:46:36 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 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 398 * check. This code will write out the buffer, period. 399 */ 400 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 401 (info->flags & V_SAVE)) { 402 if (bp->b_vp == info->vp) { 403 if (bp->b_flags & B_CLUSTEROK) { 404 vfs_bio_awrite(bp); 405 } else { 406 bremfree(bp); 407 bp->b_flags |= B_ASYNC; 408 bwrite(bp); 409 } 410 } else { 411 bremfree(bp); 412 bwrite(bp); 413 } 414 } else if (info->flags & V_SAVE) { 415 /* 416 * Cannot set B_NOCACHE on a clean buffer as this will 417 * destroy the VM backing store which might actually 418 * be dirty (and unsynchronized). 419 */ 420 bremfree(bp); 421 bp->b_flags |= (B_INVAL | B_RELBUF); 422 bp->b_flags &= ~B_ASYNC; 423 brelse(bp); 424 } else { 425 bremfree(bp); 426 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 427 bp->b_flags &= ~B_ASYNC; 428 brelse(bp); 429 } 430 return(0); 431 } 432 433 /* 434 * Truncate a file's buffer and pages to a specified length. This 435 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 436 * sync activity. 437 * 438 * The vnode must be locked. 439 */ 440 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 441 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 442 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 443 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 444 445 int 446 vtruncbuf(struct vnode *vp, off_t length, int blksize) 447 { 448 off_t truncloffset; 449 int count; 450 const char *filename; 451 452 /* 453 * Round up to the *next* block, then destroy the buffers in question. 454 * Since we are only removing some of the buffers we must rely on the 455 * scan count to determine whether a loop is necessary. 456 */ 457 if ((count = (int)(length % blksize)) != 0) 458 truncloffset = length + (blksize - count); 459 else 460 truncloffset = length; 461 462 crit_enter(); 463 do { 464 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 465 vtruncbuf_bp_trunc_cmp, 466 vtruncbuf_bp_trunc, &truncloffset); 467 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 468 vtruncbuf_bp_trunc_cmp, 469 vtruncbuf_bp_trunc, &truncloffset); 470 } while(count); 471 472 /* 473 * For safety, fsync any remaining metadata if the file is not being 474 * truncated to 0. Since the metadata does not represent the entire 475 * dirty list we have to rely on the hit count to ensure that we get 476 * all of it. 477 */ 478 if (length > 0) { 479 do { 480 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 481 vtruncbuf_bp_metasync_cmp, 482 vtruncbuf_bp_metasync, vp); 483 } while (count); 484 } 485 486 /* 487 * Clean out any left over VM backing store. 488 */ 489 crit_exit(); 490 491 vnode_pager_setsize(vp, length); 492 493 crit_enter(); 494 495 /* 496 * It is possible to have in-progress I/O from buffers that were 497 * not part of the truncation. This should not happen if we 498 * are truncating to 0-length. 499 */ 500 filename = TAILQ_FIRST(&vp->v_namecache) ? 501 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 502 503 while ((count = vp->v_track_write.bk_active) > 0) { 504 vp->v_track_write.bk_waitflag = 1; 505 tsleep(&vp->v_track_write, 0, "vbtrunc", 0); 506 if (length == 0) { 507 kprintf("Warning: vtruncbuf(): Had to wait for " 508 "%d buffer I/Os to finish in %s\n", 509 count, filename); 510 } 511 } 512 513 /* 514 * Make sure no buffers were instantiated while we were trying 515 * to clean out the remaining VM pages. This could occur due 516 * to busy dirty VM pages being flushed out to disk. 517 */ 518 do { 519 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 520 vtruncbuf_bp_trunc_cmp, 521 vtruncbuf_bp_trunc, &truncloffset); 522 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 523 vtruncbuf_bp_trunc_cmp, 524 vtruncbuf_bp_trunc, &truncloffset); 525 if (count) { 526 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 527 "left over buffers in %s\n", count, filename); 528 } 529 } while(count); 530 531 crit_exit(); 532 533 return (0); 534 } 535 536 /* 537 * The callback buffer is beyond the new file EOF and must be destroyed. 538 * Note that the compare function must conform to the RB_SCAN's requirements. 539 */ 540 static 541 int 542 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 543 { 544 if (bp->b_loffset >= *(off_t *)data) 545 return(0); 546 return(-1); 547 } 548 549 static 550 int 551 vtruncbuf_bp_trunc(struct buf *bp, void *data) 552 { 553 /* 554 * Do not try to use a buffer we cannot immediately lock, but sleep 555 * anyway to prevent a livelock. The code will loop until all buffers 556 * can be acted upon. 557 */ 558 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 559 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 560 BUF_UNLOCK(bp); 561 } else { 562 bremfree(bp); 563 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 564 bp->b_flags &= ~B_ASYNC; 565 brelse(bp); 566 } 567 return(1); 568 } 569 570 /* 571 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 572 * blocks (with a negative loffset) are scanned. 573 * Note that the compare function must conform to the RB_SCAN's requirements. 574 */ 575 static int 576 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 577 { 578 if (bp->b_loffset < 0) 579 return(0); 580 return(1); 581 } 582 583 static int 584 vtruncbuf_bp_metasync(struct buf *bp, void *data) 585 { 586 struct vnode *vp = data; 587 588 if (bp->b_flags & B_DELWRI) { 589 /* 590 * Do not try to use a buffer we cannot immediately lock, 591 * but sleep anyway to prevent a livelock. The code will 592 * loop until all buffers can be acted upon. 593 */ 594 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 595 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 596 BUF_UNLOCK(bp); 597 } else { 598 bremfree(bp); 599 if (bp->b_vp == vp) { 600 bp->b_flags |= B_ASYNC; 601 } else { 602 bp->b_flags &= ~B_ASYNC; 603 } 604 bwrite(bp); 605 } 606 return(1); 607 } else { 608 return(0); 609 } 610 } 611 612 /* 613 * vfsync - implements a multipass fsync on a file which understands 614 * dependancies and meta-data. The passed vnode must be locked. The 615 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 616 * 617 * When fsyncing data asynchronously just do one consolidated pass starting 618 * with the most negative block number. This may not get all the data due 619 * to dependancies. 620 * 621 * When fsyncing data synchronously do a data pass, then a metadata pass, 622 * then do additional data+metadata passes to try to get all the data out. 623 */ 624 static int vfsync_wait_output(struct vnode *vp, 625 int (*waitoutput)(struct vnode *, struct thread *)); 626 static int vfsync_data_only_cmp(struct buf *bp, void *data); 627 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 628 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 629 static int vfsync_bp(struct buf *bp, void *data); 630 631 struct vfsync_info { 632 struct vnode *vp; 633 int synchronous; 634 int syncdeps; 635 int lazycount; 636 int lazylimit; 637 int skippedbufs; 638 int (*checkdef)(struct buf *); 639 }; 640 641 int 642 vfsync(struct vnode *vp, int waitfor, int passes, 643 int (*checkdef)(struct buf *), 644 int (*waitoutput)(struct vnode *, struct thread *)) 645 { 646 struct vfsync_info info; 647 int error; 648 649 bzero(&info, sizeof(info)); 650 info.vp = vp; 651 if ((info.checkdef = checkdef) == NULL) 652 info.syncdeps = 1; 653 654 crit_enter_id("vfsync"); 655 656 switch(waitfor) { 657 case MNT_LAZY: 658 /* 659 * Lazy (filesystem syncer typ) Asynchronous plus limit the 660 * number of data (not meta) pages we try to flush to 1MB. 661 * A non-zero return means that lazy limit was reached. 662 */ 663 info.lazylimit = 1024 * 1024; 664 info.syncdeps = 1; 665 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 666 vfsync_lazy_range_cmp, vfsync_bp, &info); 667 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 668 vfsync_meta_only_cmp, vfsync_bp, &info); 669 if (error == 0) 670 vp->v_lazyw = 0; 671 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 672 vn_syncer_add_to_worklist(vp, 1); 673 error = 0; 674 break; 675 case MNT_NOWAIT: 676 /* 677 * Asynchronous. Do a data-only pass and a meta-only pass. 678 */ 679 info.syncdeps = 1; 680 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 681 vfsync_bp, &info); 682 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 683 vfsync_bp, &info); 684 error = 0; 685 break; 686 default: 687 /* 688 * Synchronous. Do a data-only pass, then a meta-data+data 689 * pass, then additional integrated passes to try to get 690 * all the dependancies flushed. 691 */ 692 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 693 vfsync_bp, &info); 694 error = vfsync_wait_output(vp, waitoutput); 695 if (error == 0) { 696 info.skippedbufs = 0; 697 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 698 vfsync_bp, &info); 699 error = vfsync_wait_output(vp, waitoutput); 700 if (info.skippedbufs) 701 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs); 702 } 703 while (error == 0 && passes > 0 && 704 !RB_EMPTY(&vp->v_rbdirty_tree)) { 705 if (--passes == 0) { 706 info.synchronous = 1; 707 info.syncdeps = 1; 708 } 709 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 710 vfsync_bp, &info); 711 if (error < 0) 712 error = -error; 713 info.syncdeps = 1; 714 if (error == 0) 715 error = vfsync_wait_output(vp, waitoutput); 716 } 717 break; 718 } 719 crit_exit_id("vfsync"); 720 return(error); 721 } 722 723 static int 724 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 725 { 726 int error = 0; 727 728 while (vp->v_track_write.bk_active) { 729 vp->v_track_write.bk_waitflag = 1; 730 tsleep(&vp->v_track_write, 0, "fsfsn", 0); 731 } 732 if (waitoutput) 733 error = waitoutput(vp, curthread); 734 return(error); 735 } 736 737 static int 738 vfsync_data_only_cmp(struct buf *bp, void *data) 739 { 740 if (bp->b_loffset < 0) 741 return(-1); 742 return(0); 743 } 744 745 static int 746 vfsync_meta_only_cmp(struct buf *bp, void *data) 747 { 748 if (bp->b_loffset < 0) 749 return(0); 750 return(1); 751 } 752 753 static int 754 vfsync_lazy_range_cmp(struct buf *bp, void *data) 755 { 756 struct vfsync_info *info = data; 757 if (bp->b_loffset < info->vp->v_lazyw) 758 return(-1); 759 return(0); 760 } 761 762 static int 763 vfsync_bp(struct buf *bp, void *data) 764 { 765 struct vfsync_info *info = data; 766 struct vnode *vp = info->vp; 767 int error; 768 769 /* 770 * if syncdeps is not set we do not try to write buffers which have 771 * dependancies. 772 */ 773 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 774 return(0); 775 776 /* 777 * Ignore buffers that we cannot immediately lock. XXX 778 */ 779 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 780 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp); 781 ++info->skippedbufs; 782 return(0); 783 } 784 if ((bp->b_flags & B_DELWRI) == 0) 785 panic("vfsync_bp: buffer not dirty"); 786 if (vp != bp->b_vp) 787 panic("vfsync_bp: buffer vp mismatch"); 788 789 /* 790 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 791 * has been written but an additional handshake with the device 792 * is required before we can dispose of the buffer. We have no idea 793 * how to do this so we have to skip these buffers. 794 */ 795 if (bp->b_flags & B_NEEDCOMMIT) { 796 BUF_UNLOCK(bp); 797 return(0); 798 } 799 800 if (info->synchronous) { 801 /* 802 * Synchronous flushing. An error may be returned. 803 */ 804 bremfree(bp); 805 crit_exit_id("vfsync"); 806 error = bwrite(bp); 807 crit_enter_id("vfsync"); 808 } else { 809 /* 810 * Asynchronous flushing. A negative return value simply 811 * stops the scan and is not considered an error. We use 812 * this to support limited MNT_LAZY flushes. 813 */ 814 vp->v_lazyw = bp->b_loffset; 815 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 816 info->lazycount += vfs_bio_awrite(bp); 817 } else { 818 info->lazycount += bp->b_bufsize; 819 bremfree(bp); 820 crit_exit_id("vfsync"); 821 bawrite(bp); 822 crit_enter_id("vfsync"); 823 } 824 if (info->lazylimit && info->lazycount >= info->lazylimit) 825 error = 1; 826 else 827 error = 0; 828 } 829 return(-error); 830 } 831 832 /* 833 * Associate a buffer with a vnode. 834 */ 835 void 836 bgetvp(struct vnode *vp, struct buf *bp) 837 { 838 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 839 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 840 841 vhold(vp); 842 /* 843 * Insert onto list for new vnode. 844 */ 845 crit_enter(); 846 bp->b_vp = vp; 847 bp->b_flags |= B_HASHED; 848 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) 849 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 850 851 bp->b_flags |= B_VNCLEAN; 852 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 853 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 854 crit_exit(); 855 } 856 857 /* 858 * Disassociate a buffer from a vnode. 859 */ 860 void 861 brelvp(struct buf *bp) 862 { 863 struct vnode *vp; 864 865 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 866 867 /* 868 * Delete from old vnode list, if on one. 869 */ 870 vp = bp->b_vp; 871 crit_enter(); 872 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 873 if (bp->b_flags & B_VNDIRTY) 874 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 875 else 876 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 877 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 878 } 879 if (bp->b_flags & B_HASHED) { 880 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 881 bp->b_flags &= ~B_HASHED; 882 } 883 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 884 vp->v_flag &= ~VONWORKLST; 885 LIST_REMOVE(vp, v_synclist); 886 } 887 crit_exit(); 888 bp->b_vp = NULL; 889 vdrop(vp); 890 } 891 892 /* 893 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 894 * This routine is called when the state of the B_DELWRI bit is changed. 895 */ 896 void 897 reassignbuf(struct buf *bp) 898 { 899 struct vnode *vp = bp->b_vp; 900 int delay; 901 902 KKASSERT(vp != NULL); 903 ++reassignbufcalls; 904 905 /* 906 * B_PAGING flagged buffers cannot be reassigned because their vp 907 * is not fully linked in. 908 */ 909 if (bp->b_flags & B_PAGING) 910 panic("cannot reassign paging buffer"); 911 912 crit_enter(); 913 if (bp->b_flags & B_DELWRI) { 914 /* 915 * Move to the dirty list, add the vnode to the worklist 916 */ 917 if (bp->b_flags & B_VNCLEAN) { 918 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 919 bp->b_flags &= ~B_VNCLEAN; 920 } 921 if ((bp->b_flags & B_VNDIRTY) == 0) { 922 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 923 panic("reassignbuf: dup lblk vp %p bp %p", 924 vp, bp); 925 } 926 bp->b_flags |= B_VNDIRTY; 927 } 928 if ((vp->v_flag & VONWORKLST) == 0) { 929 switch (vp->v_type) { 930 case VDIR: 931 delay = dirdelay; 932 break; 933 case VCHR: 934 case VBLK: 935 if (vp->v_rdev && 936 vp->v_rdev->si_mountpoint != NULL) { 937 delay = metadelay; 938 break; 939 } 940 /* fall through */ 941 default: 942 delay = filedelay; 943 } 944 vn_syncer_add_to_worklist(vp, delay); 945 } 946 } else { 947 /* 948 * Move to the clean list, remove the vnode from the worklist 949 * if no dirty blocks remain. 950 */ 951 if (bp->b_flags & B_VNDIRTY) { 952 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 953 bp->b_flags &= ~B_VNDIRTY; 954 } 955 if ((bp->b_flags & B_VNCLEAN) == 0) { 956 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 957 panic("reassignbuf: dup lblk vp %p bp %p", 958 vp, bp); 959 } 960 bp->b_flags |= B_VNCLEAN; 961 } 962 if ((vp->v_flag & VONWORKLST) && 963 RB_EMPTY(&vp->v_rbdirty_tree)) { 964 vp->v_flag &= ~VONWORKLST; 965 LIST_REMOVE(vp, v_synclist); 966 } 967 } 968 crit_exit(); 969 } 970 971 /* 972 * Create a vnode for a block device. 973 * Used for mounting the root file system. 974 */ 975 int 976 bdevvp(cdev_t dev, struct vnode **vpp) 977 { 978 struct vnode *vp; 979 struct vnode *nvp; 980 int error; 981 982 if (dev == NULL) { 983 *vpp = NULLVP; 984 return (ENXIO); 985 } 986 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0); 987 if (error) { 988 *vpp = NULLVP; 989 return (error); 990 } 991 vp = nvp; 992 vp->v_type = VCHR; 993 vp->v_umajor = dev->si_umajor; 994 vp->v_uminor = dev->si_uminor; 995 vx_unlock(vp); 996 *vpp = vp; 997 return (0); 998 } 999 1000 int 1001 v_associate_rdev(struct vnode *vp, cdev_t dev) 1002 { 1003 lwkt_tokref ilock; 1004 1005 if (dev == NULL) 1006 return(ENXIO); 1007 if (dev_is_good(dev) == 0) 1008 return(ENXIO); 1009 KKASSERT(vp->v_rdev == NULL); 1010 if (dev_ref_debug) 1011 kprintf("Z1"); 1012 vp->v_rdev = reference_dev(dev); 1013 lwkt_gettoken(&ilock, &spechash_token); 1014 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1015 lwkt_reltoken(&ilock); 1016 return(0); 1017 } 1018 1019 void 1020 v_release_rdev(struct vnode *vp) 1021 { 1022 lwkt_tokref ilock; 1023 cdev_t dev; 1024 1025 if ((dev = vp->v_rdev) != NULL) { 1026 lwkt_gettoken(&ilock, &spechash_token); 1027 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1028 vp->v_rdev = NULL; 1029 release_dev(dev); 1030 lwkt_reltoken(&ilock); 1031 } 1032 } 1033 1034 /* 1035 * Add a vnode to the alias list hung off the cdev_t. We only associate 1036 * the device number with the vnode. The actual device is not associated 1037 * until the vnode is opened (usually in spec_open()), and will be 1038 * disassociated on last close. 1039 */ 1040 void 1041 addaliasu(struct vnode *nvp, int x, int y) 1042 { 1043 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1044 panic("addaliasu on non-special vnode"); 1045 nvp->v_umajor = x; 1046 nvp->v_uminor = y; 1047 } 1048 1049 /* 1050 * Disassociate a vnode from its underlying filesystem. 1051 * 1052 * The vnode must be VX locked and referenced. In all normal situations 1053 * there are no active references. If vclean_vxlocked() is called while 1054 * there are active references, the vnode is being ripped out and we have 1055 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1056 */ 1057 void 1058 vclean_vxlocked(struct vnode *vp, int flags) 1059 { 1060 int active; 1061 int n; 1062 vm_object_t object; 1063 1064 /* 1065 * If the vnode has already been reclaimed we have nothing to do. 1066 */ 1067 if (vp->v_flag & VRECLAIMED) 1068 return; 1069 vp->v_flag |= VRECLAIMED; 1070 1071 /* 1072 * Scrap the vfs cache 1073 */ 1074 while (cache_inval_vp(vp, 0) != 0) { 1075 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1076 tsleep(vp, 0, "vclninv", 2); 1077 } 1078 1079 /* 1080 * Check to see if the vnode is in use. If so we have to reference it 1081 * before we clean it out so that its count cannot fall to zero and 1082 * generate a race against ourselves to recycle it. 1083 */ 1084 active = sysref_isactive(&vp->v_sysref); 1085 1086 /* 1087 * Clean out any buffers associated with the vnode and destroy its 1088 * object, if it has one. 1089 */ 1090 vinvalbuf(vp, V_SAVE, 0, 0); 1091 1092 /* 1093 * If purging an active vnode (typically during a forced unmount 1094 * or reboot), it must be closed and deactivated before being 1095 * reclaimed. This isn't really all that safe, but what can 1096 * we do? XXX. 1097 * 1098 * Note that neither of these routines unlocks the vnode. 1099 */ 1100 if (active && (flags & DOCLOSE)) { 1101 while ((n = vp->v_opencount) != 0) { 1102 if (vp->v_writecount) 1103 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1104 else 1105 VOP_CLOSE(vp, FNONBLOCK); 1106 if (vp->v_opencount == n) { 1107 kprintf("Warning: unable to force-close" 1108 " vnode %p\n", vp); 1109 break; 1110 } 1111 } 1112 } 1113 1114 /* 1115 * If the vnode has not been deactivated, deactivated it. Deactivation 1116 * can create new buffers and VM pages so we have to call vinvalbuf() 1117 * again to make sure they all get flushed. 1118 * 1119 * This can occur if a file with a link count of 0 needs to be 1120 * truncated. 1121 */ 1122 if ((vp->v_flag & VINACTIVE) == 0) { 1123 vp->v_flag |= VINACTIVE; 1124 VOP_INACTIVE(vp); 1125 vinvalbuf(vp, V_SAVE, 0, 0); 1126 } 1127 1128 /* 1129 * If the vnode has an object, destroy it. 1130 */ 1131 if ((object = vp->v_object) != NULL) { 1132 if (object->ref_count == 0) { 1133 if ((object->flags & OBJ_DEAD) == 0) 1134 vm_object_terminate(object); 1135 } else { 1136 vm_pager_deallocate(object); 1137 } 1138 vp->v_flag &= ~VOBJBUF; 1139 } 1140 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1141 1142 /* 1143 * Reclaim the vnode. 1144 */ 1145 if (VOP_RECLAIM(vp)) 1146 panic("vclean: cannot reclaim"); 1147 1148 /* 1149 * Done with purge, notify sleepers of the grim news. 1150 */ 1151 vp->v_ops = &dead_vnode_vops_p; 1152 vn_pollgone(vp); 1153 vp->v_tag = VT_NON; 1154 1155 /* 1156 * If we are destroying an active vnode, reactivate it now that 1157 * we have reassociated it with deadfs. This prevents the system 1158 * from crashing on the vnode due to it being unexpectedly marked 1159 * as inactive or reclaimed. 1160 */ 1161 if (active && (flags & DOCLOSE)) { 1162 vp->v_flag &= ~(VINACTIVE|VRECLAIMED); 1163 } 1164 } 1165 1166 /* 1167 * Eliminate all activity associated with the requested vnode 1168 * and with all vnodes aliased to the requested vnode. 1169 * 1170 * The vnode must be referenced and vx_lock()'d 1171 * 1172 * revoke { struct vnode *a_vp, int a_flags } 1173 */ 1174 int 1175 vop_stdrevoke(struct vop_revoke_args *ap) 1176 { 1177 struct vnode *vp, *vq; 1178 lwkt_tokref ilock; 1179 cdev_t dev; 1180 1181 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1182 1183 vp = ap->a_vp; 1184 1185 /* 1186 * If the vnode is already dead don't try to revoke it 1187 */ 1188 if (vp->v_flag & VRECLAIMED) 1189 return (0); 1190 1191 /* 1192 * If the vnode has a device association, scrap all vnodes associated 1193 * with the device. Don't let the device disappear on us while we 1194 * are scrapping the vnodes. 1195 * 1196 * The passed vp will probably show up in the list, do not VX lock 1197 * it twice! 1198 */ 1199 if (vp->v_type != VCHR) 1200 return(0); 1201 if ((dev = vp->v_rdev) == NULL) { 1202 if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL) 1203 return(0); 1204 } 1205 reference_dev(dev); 1206 lwkt_gettoken(&ilock, &spechash_token); 1207 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1208 if (vp != vq) 1209 vx_get(vq); 1210 if (vq == SLIST_FIRST(&dev->si_hlist)) 1211 vgone_vxlocked(vq); 1212 if (vp != vq) 1213 vx_put(vq); 1214 } 1215 lwkt_reltoken(&ilock); 1216 release_dev(dev); 1217 return (0); 1218 } 1219 1220 /* 1221 * This is called when the object underlying a vnode is being destroyed, 1222 * such as in a remove(). Try to recycle the vnode immediately if the 1223 * only active reference is our reference. 1224 */ 1225 int 1226 vrecycle(struct vnode *vp) 1227 { 1228 if (vp->v_sysref.refcnt <= 1) { 1229 vgone_vxlocked(vp); 1230 return (1); 1231 } 1232 return (0); 1233 } 1234 1235 /* 1236 * Eliminate all activity associated with a vnode in preparation for reuse. 1237 * 1238 * The vnode must be VX locked and refd and will remain VX locked and refd 1239 * on return. This routine may be called with the vnode in any state, as 1240 * long as it is VX locked. The vnode will be cleaned out and marked 1241 * VRECLAIMED but will not actually be reused until all existing refs and 1242 * holds go away. 1243 * 1244 * NOTE: This routine may be called on a vnode which has not yet been 1245 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1246 * already been reclaimed. 1247 * 1248 * This routine is not responsible for placing us back on the freelist. 1249 * Instead, it happens automatically when the caller releases the VX lock 1250 * (assuming there aren't any other references). 1251 */ 1252 1253 void 1254 vgone_vxlocked(struct vnode *vp) 1255 { 1256 /* 1257 * assert that the VX lock is held. This is an absolute requirement 1258 * now for vgone_vxlocked() to be called. 1259 */ 1260 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1261 1262 /* 1263 * Clean out the filesystem specific data and set the VRECLAIMED 1264 * bit. Also deactivate the vnode if necessary. 1265 */ 1266 vclean_vxlocked(vp, DOCLOSE); 1267 1268 /* 1269 * Delete from old mount point vnode list, if on one. 1270 */ 1271 if (vp->v_mount != NULL) 1272 insmntque(vp, NULL); 1273 1274 /* 1275 * If special device, remove it from special device alias list 1276 * if it is on one. This should normally only occur if a vnode is 1277 * being revoked as the device should otherwise have been released 1278 * naturally. 1279 */ 1280 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1281 v_release_rdev(vp); 1282 } 1283 1284 /* 1285 * Set us to VBAD 1286 */ 1287 vp->v_type = VBAD; 1288 } 1289 1290 /* 1291 * Lookup a vnode by device number. 1292 */ 1293 int 1294 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1295 { 1296 lwkt_tokref ilock; 1297 struct vnode *vp; 1298 1299 lwkt_gettoken(&ilock, &spechash_token); 1300 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1301 if (type == vp->v_type) { 1302 *vpp = vp; 1303 lwkt_reltoken(&ilock); 1304 return (1); 1305 } 1306 } 1307 lwkt_reltoken(&ilock); 1308 return (0); 1309 } 1310 1311 /* 1312 * Calculate the total number of references to a special device. This 1313 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1314 * an overloaded field. Since udev2dev can now return NULL, we have 1315 * to check for a NULL v_rdev. 1316 */ 1317 int 1318 count_dev(cdev_t dev) 1319 { 1320 lwkt_tokref ilock; 1321 struct vnode *vp; 1322 int count = 0; 1323 1324 if (SLIST_FIRST(&dev->si_hlist)) { 1325 lwkt_gettoken(&ilock, &spechash_token); 1326 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1327 if (vp->v_sysref.refcnt > 0) 1328 count += vp->v_sysref.refcnt; 1329 } 1330 lwkt_reltoken(&ilock); 1331 } 1332 return(count); 1333 } 1334 1335 int 1336 count_udev(int x, int y) 1337 { 1338 cdev_t dev; 1339 1340 if ((dev = get_dev(x, y)) == NULL) 1341 return(0); 1342 return(count_dev(dev)); 1343 } 1344 1345 int 1346 vcount(struct vnode *vp) 1347 { 1348 if (vp->v_rdev == NULL) 1349 return(0); 1350 return(count_dev(vp->v_rdev)); 1351 } 1352 1353 /* 1354 * Initialize VMIO for a vnode. This routine MUST be called before a 1355 * VFS can issue buffer cache ops on a vnode. It is typically called 1356 * when a vnode is initialized from its inode. 1357 */ 1358 int 1359 vinitvmio(struct vnode *vp, off_t filesize) 1360 { 1361 vm_object_t object; 1362 int error = 0; 1363 1364 retry: 1365 if ((object = vp->v_object) == NULL) { 1366 object = vnode_pager_alloc(vp, filesize, 0, 0); 1367 /* 1368 * Dereference the reference we just created. This assumes 1369 * that the object is associated with the vp. 1370 */ 1371 object->ref_count--; 1372 vrele(vp); 1373 } else { 1374 if (object->flags & OBJ_DEAD) { 1375 vn_unlock(vp); 1376 vm_object_dead_sleep(object, "vodead"); 1377 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1378 goto retry; 1379 } 1380 } 1381 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1382 vp->v_flag |= VOBJBUF; 1383 return (error); 1384 } 1385 1386 1387 /* 1388 * Print out a description of a vnode. 1389 */ 1390 static char *typename[] = 1391 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1392 1393 void 1394 vprint(char *label, struct vnode *vp) 1395 { 1396 char buf[96]; 1397 1398 if (label != NULL) 1399 kprintf("%s: %p: ", label, (void *)vp); 1400 else 1401 kprintf("%p: ", (void *)vp); 1402 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1403 typename[vp->v_type], 1404 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1405 buf[0] = '\0'; 1406 if (vp->v_flag & VROOT) 1407 strcat(buf, "|VROOT"); 1408 if (vp->v_flag & VTEXT) 1409 strcat(buf, "|VTEXT"); 1410 if (vp->v_flag & VSYSTEM) 1411 strcat(buf, "|VSYSTEM"); 1412 if (vp->v_flag & VFREE) 1413 strcat(buf, "|VFREE"); 1414 if (vp->v_flag & VOBJBUF) 1415 strcat(buf, "|VOBJBUF"); 1416 if (buf[0] != '\0') 1417 kprintf(" flags (%s)", &buf[1]); 1418 if (vp->v_data == NULL) { 1419 kprintf("\n"); 1420 } else { 1421 kprintf("\n\t"); 1422 VOP_PRINT(vp); 1423 } 1424 } 1425 1426 #ifdef DDB 1427 #include <ddb/ddb.h> 1428 1429 static int db_show_locked_vnodes(struct mount *mp, void *data); 1430 1431 /* 1432 * List all of the locked vnodes in the system. 1433 * Called when debugging the kernel. 1434 */ 1435 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1436 { 1437 kprintf("Locked vnodes\n"); 1438 mountlist_scan(db_show_locked_vnodes, NULL, 1439 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1440 } 1441 1442 static int 1443 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1444 { 1445 struct vnode *vp; 1446 1447 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1448 if (vn_islocked(vp)) 1449 vprint((char *)0, vp); 1450 } 1451 return(0); 1452 } 1453 #endif 1454 1455 /* 1456 * Top level filesystem related information gathering. 1457 */ 1458 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1459 1460 static int 1461 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1462 { 1463 int *name = (int *)arg1 - 1; /* XXX */ 1464 u_int namelen = arg2 + 1; /* XXX */ 1465 struct vfsconf *vfsp; 1466 1467 #if 1 || defined(COMPAT_PRELITE2) 1468 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1469 if (namelen == 1) 1470 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1471 #endif 1472 1473 #ifdef notyet 1474 /* all sysctl names at this level are at least name and field */ 1475 if (namelen < 2) 1476 return (ENOTDIR); /* overloaded */ 1477 if (name[0] != VFS_GENERIC) { 1478 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1479 if (vfsp->vfc_typenum == name[0]) 1480 break; 1481 if (vfsp == NULL) 1482 return (EOPNOTSUPP); 1483 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1484 oldp, oldlenp, newp, newlen, p)); 1485 } 1486 #endif 1487 switch (name[1]) { 1488 case VFS_MAXTYPENUM: 1489 if (namelen != 2) 1490 return (ENOTDIR); 1491 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1492 case VFS_CONF: 1493 if (namelen != 3) 1494 return (ENOTDIR); /* overloaded */ 1495 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1496 if (vfsp->vfc_typenum == name[2]) 1497 break; 1498 if (vfsp == NULL) 1499 return (EOPNOTSUPP); 1500 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1501 } 1502 return (EOPNOTSUPP); 1503 } 1504 1505 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1506 "Generic filesystem"); 1507 1508 #if 1 || defined(COMPAT_PRELITE2) 1509 1510 static int 1511 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1512 { 1513 int error; 1514 struct vfsconf *vfsp; 1515 struct ovfsconf ovfs; 1516 1517 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1518 bzero(&ovfs, sizeof(ovfs)); 1519 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1520 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1521 ovfs.vfc_index = vfsp->vfc_typenum; 1522 ovfs.vfc_refcount = vfsp->vfc_refcount; 1523 ovfs.vfc_flags = vfsp->vfc_flags; 1524 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1525 if (error) 1526 return error; 1527 } 1528 return 0; 1529 } 1530 1531 #endif /* 1 || COMPAT_PRELITE2 */ 1532 1533 /* 1534 * Check to see if a filesystem is mounted on a block device. 1535 */ 1536 int 1537 vfs_mountedon(struct vnode *vp) 1538 { 1539 cdev_t dev; 1540 1541 if ((dev = vp->v_rdev) == NULL) { 1542 if (vp->v_type != VBLK) 1543 dev = get_dev(vp->v_uminor, vp->v_umajor); 1544 } 1545 if (dev != NULL && dev->si_mountpoint) 1546 return (EBUSY); 1547 return (0); 1548 } 1549 1550 /* 1551 * Unmount all filesystems. The list is traversed in reverse order 1552 * of mounting to avoid dependencies. 1553 */ 1554 1555 static int vfs_umountall_callback(struct mount *mp, void *data); 1556 1557 void 1558 vfs_unmountall(void) 1559 { 1560 int count; 1561 1562 do { 1563 count = mountlist_scan(vfs_umountall_callback, 1564 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1565 } while (count); 1566 } 1567 1568 static 1569 int 1570 vfs_umountall_callback(struct mount *mp, void *data) 1571 { 1572 int error; 1573 1574 error = dounmount(mp, MNT_FORCE); 1575 if (error) { 1576 mountlist_remove(mp); 1577 kprintf("unmount of filesystem mounted from %s failed (", 1578 mp->mnt_stat.f_mntfromname); 1579 if (error == EBUSY) 1580 kprintf("BUSY)\n"); 1581 else 1582 kprintf("%d)\n", error); 1583 } 1584 return(1); 1585 } 1586 1587 /* 1588 * Build hash lists of net addresses and hang them off the mount point. 1589 * Called by ufs_mount() to set up the lists of export addresses. 1590 */ 1591 static int 1592 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1593 struct export_args *argp) 1594 { 1595 struct netcred *np; 1596 struct radix_node_head *rnh; 1597 int i; 1598 struct radix_node *rn; 1599 struct sockaddr *saddr, *smask = 0; 1600 struct domain *dom; 1601 int error; 1602 1603 if (argp->ex_addrlen == 0) { 1604 if (mp->mnt_flag & MNT_DEFEXPORTED) 1605 return (EPERM); 1606 np = &nep->ne_defexported; 1607 np->netc_exflags = argp->ex_flags; 1608 np->netc_anon = argp->ex_anon; 1609 np->netc_anon.cr_ref = 1; 1610 mp->mnt_flag |= MNT_DEFEXPORTED; 1611 return (0); 1612 } 1613 1614 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1615 return (EINVAL); 1616 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1617 return (EINVAL); 1618 1619 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1620 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK); 1621 bzero((caddr_t) np, i); 1622 saddr = (struct sockaddr *) (np + 1); 1623 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1624 goto out; 1625 if (saddr->sa_len > argp->ex_addrlen) 1626 saddr->sa_len = argp->ex_addrlen; 1627 if (argp->ex_masklen) { 1628 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1629 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1630 if (error) 1631 goto out; 1632 if (smask->sa_len > argp->ex_masklen) 1633 smask->sa_len = argp->ex_masklen; 1634 } 1635 i = saddr->sa_family; 1636 if ((rnh = nep->ne_rtable[i]) == 0) { 1637 /* 1638 * Seems silly to initialize every AF when most are not used, 1639 * do so on demand here 1640 */ 1641 SLIST_FOREACH(dom, &domains, dom_next) 1642 if (dom->dom_family == i && dom->dom_rtattach) { 1643 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1644 dom->dom_rtoffset); 1645 break; 1646 } 1647 if ((rnh = nep->ne_rtable[i]) == 0) { 1648 error = ENOBUFS; 1649 goto out; 1650 } 1651 } 1652 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1653 np->netc_rnodes); 1654 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1655 error = EPERM; 1656 goto out; 1657 } 1658 np->netc_exflags = argp->ex_flags; 1659 np->netc_anon = argp->ex_anon; 1660 np->netc_anon.cr_ref = 1; 1661 return (0); 1662 out: 1663 kfree(np, M_NETADDR); 1664 return (error); 1665 } 1666 1667 /* ARGSUSED */ 1668 static int 1669 vfs_free_netcred(struct radix_node *rn, void *w) 1670 { 1671 struct radix_node_head *rnh = (struct radix_node_head *) w; 1672 1673 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1674 kfree((caddr_t) rn, M_NETADDR); 1675 return (0); 1676 } 1677 1678 /* 1679 * Free the net address hash lists that are hanging off the mount points. 1680 */ 1681 static void 1682 vfs_free_addrlist(struct netexport *nep) 1683 { 1684 int i; 1685 struct radix_node_head *rnh; 1686 1687 for (i = 0; i <= AF_MAX; i++) 1688 if ((rnh = nep->ne_rtable[i])) { 1689 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1690 (caddr_t) rnh); 1691 kfree((caddr_t) rnh, M_RTABLE); 1692 nep->ne_rtable[i] = 0; 1693 } 1694 } 1695 1696 int 1697 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1698 { 1699 int error; 1700 1701 if (argp->ex_flags & MNT_DELEXPORT) { 1702 if (mp->mnt_flag & MNT_EXPUBLIC) { 1703 vfs_setpublicfs(NULL, NULL, NULL); 1704 mp->mnt_flag &= ~MNT_EXPUBLIC; 1705 } 1706 vfs_free_addrlist(nep); 1707 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1708 } 1709 if (argp->ex_flags & MNT_EXPORTED) { 1710 if (argp->ex_flags & MNT_EXPUBLIC) { 1711 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1712 return (error); 1713 mp->mnt_flag |= MNT_EXPUBLIC; 1714 } 1715 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1716 return (error); 1717 mp->mnt_flag |= MNT_EXPORTED; 1718 } 1719 return (0); 1720 } 1721 1722 1723 /* 1724 * Set the publicly exported filesystem (WebNFS). Currently, only 1725 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1726 */ 1727 int 1728 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1729 struct export_args *argp) 1730 { 1731 int error; 1732 struct vnode *rvp; 1733 char *cp; 1734 1735 /* 1736 * mp == NULL -> invalidate the current info, the FS is 1737 * no longer exported. May be called from either vfs_export 1738 * or unmount, so check if it hasn't already been done. 1739 */ 1740 if (mp == NULL) { 1741 if (nfs_pub.np_valid) { 1742 nfs_pub.np_valid = 0; 1743 if (nfs_pub.np_index != NULL) { 1744 FREE(nfs_pub.np_index, M_TEMP); 1745 nfs_pub.np_index = NULL; 1746 } 1747 } 1748 return (0); 1749 } 1750 1751 /* 1752 * Only one allowed at a time. 1753 */ 1754 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1755 return (EBUSY); 1756 1757 /* 1758 * Get real filehandle for root of exported FS. 1759 */ 1760 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1761 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1762 1763 if ((error = VFS_ROOT(mp, &rvp))) 1764 return (error); 1765 1766 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1767 return (error); 1768 1769 vput(rvp); 1770 1771 /* 1772 * If an indexfile was specified, pull it in. 1773 */ 1774 if (argp->ex_indexfile != NULL) { 1775 int namelen; 1776 1777 error = vn_get_namelen(rvp, &namelen); 1778 if (error) 1779 return (error); 1780 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 1781 M_WAITOK); 1782 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1783 namelen, (size_t *)0); 1784 if (!error) { 1785 /* 1786 * Check for illegal filenames. 1787 */ 1788 for (cp = nfs_pub.np_index; *cp; cp++) { 1789 if (*cp == '/') { 1790 error = EINVAL; 1791 break; 1792 } 1793 } 1794 } 1795 if (error) { 1796 FREE(nfs_pub.np_index, M_TEMP); 1797 return (error); 1798 } 1799 } 1800 1801 nfs_pub.np_mount = mp; 1802 nfs_pub.np_valid = 1; 1803 return (0); 1804 } 1805 1806 struct netcred * 1807 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1808 struct sockaddr *nam) 1809 { 1810 struct netcred *np; 1811 struct radix_node_head *rnh; 1812 struct sockaddr *saddr; 1813 1814 np = NULL; 1815 if (mp->mnt_flag & MNT_EXPORTED) { 1816 /* 1817 * Lookup in the export list first. 1818 */ 1819 if (nam != NULL) { 1820 saddr = nam; 1821 rnh = nep->ne_rtable[saddr->sa_family]; 1822 if (rnh != NULL) { 1823 np = (struct netcred *) 1824 (*rnh->rnh_matchaddr)((char *)saddr, 1825 rnh); 1826 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1827 np = NULL; 1828 } 1829 } 1830 /* 1831 * If no address match, use the default if it exists. 1832 */ 1833 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1834 np = &nep->ne_defexported; 1835 } 1836 return (np); 1837 } 1838 1839 /* 1840 * perform msync on all vnodes under a mount point. The mount point must 1841 * be locked. This code is also responsible for lazy-freeing unreferenced 1842 * vnodes whos VM objects no longer contain pages. 1843 * 1844 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1845 * 1846 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 1847 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 1848 * way up in this high level function. 1849 */ 1850 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1851 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1852 1853 void 1854 vfs_msync(struct mount *mp, int flags) 1855 { 1856 int vmsc_flags; 1857 1858 vmsc_flags = VMSC_GETVP; 1859 if (flags != MNT_WAIT) 1860 vmsc_flags |= VMSC_NOWAIT; 1861 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 1862 (void *)flags); 1863 } 1864 1865 /* 1866 * scan1 is a fast pre-check. There could be hundreds of thousands of 1867 * vnodes, we cannot afford to do anything heavy weight until we have a 1868 * fairly good indication that there is work to do. 1869 */ 1870 static 1871 int 1872 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1873 { 1874 int flags = (int)data; 1875 1876 if ((vp->v_flag & VRECLAIMED) == 0) { 1877 if (vshouldmsync(vp)) 1878 return(0); /* call scan2 */ 1879 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1880 (vp->v_flag & VOBJDIRTY) && 1881 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 1882 return(0); /* call scan2 */ 1883 } 1884 } 1885 1886 /* 1887 * do not call scan2, continue the loop 1888 */ 1889 return(-1); 1890 } 1891 1892 /* 1893 * This callback is handed a locked vnode. 1894 */ 1895 static 1896 int 1897 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1898 { 1899 vm_object_t obj; 1900 int flags = (int)data; 1901 1902 if (vp->v_flag & VRECLAIMED) 1903 return(0); 1904 1905 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 1906 if ((obj = vp->v_object) != NULL) { 1907 vm_object_page_clean(obj, 0, 0, 1908 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1909 } 1910 } 1911 return(0); 1912 } 1913 1914 /* 1915 * Record a process's interest in events which might happen to 1916 * a vnode. Because poll uses the historic select-style interface 1917 * internally, this routine serves as both the ``check for any 1918 * pending events'' and the ``record my interest in future events'' 1919 * functions. (These are done together, while the lock is held, 1920 * to avoid race conditions.) 1921 */ 1922 int 1923 vn_pollrecord(struct vnode *vp, int events) 1924 { 1925 lwkt_tokref ilock; 1926 1927 KKASSERT(curthread->td_proc != NULL); 1928 1929 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1930 if (vp->v_pollinfo.vpi_revents & events) { 1931 /* 1932 * This leaves events we are not interested 1933 * in available for the other process which 1934 * which presumably had requested them 1935 * (otherwise they would never have been 1936 * recorded). 1937 */ 1938 events &= vp->v_pollinfo.vpi_revents; 1939 vp->v_pollinfo.vpi_revents &= ~events; 1940 1941 lwkt_reltoken(&ilock); 1942 return events; 1943 } 1944 vp->v_pollinfo.vpi_events |= events; 1945 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo); 1946 lwkt_reltoken(&ilock); 1947 return 0; 1948 } 1949 1950 /* 1951 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1952 * it is possible for us to miss an event due to race conditions, but 1953 * that condition is expected to be rare, so for the moment it is the 1954 * preferred interface. 1955 */ 1956 void 1957 vn_pollevent(struct vnode *vp, int events) 1958 { 1959 lwkt_tokref ilock; 1960 1961 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1962 if (vp->v_pollinfo.vpi_events & events) { 1963 /* 1964 * We clear vpi_events so that we don't 1965 * call selwakeup() twice if two events are 1966 * posted before the polling process(es) is 1967 * awakened. This also ensures that we take at 1968 * most one selwakeup() if the polling process 1969 * is no longer interested. However, it does 1970 * mean that only one event can be noticed at 1971 * a time. (Perhaps we should only clear those 1972 * event bits which we note?) XXX 1973 */ 1974 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1975 vp->v_pollinfo.vpi_revents |= events; 1976 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1977 } 1978 lwkt_reltoken(&ilock); 1979 } 1980 1981 /* 1982 * Wake up anyone polling on vp because it is being revoked. 1983 * This depends on dead_poll() returning POLLHUP for correct 1984 * behavior. 1985 */ 1986 void 1987 vn_pollgone(struct vnode *vp) 1988 { 1989 lwkt_tokref ilock; 1990 1991 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1992 if (vp->v_pollinfo.vpi_events) { 1993 vp->v_pollinfo.vpi_events = 0; 1994 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1995 } 1996 lwkt_reltoken(&ilock); 1997 } 1998 1999 /* 2000 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2001 * (or v_rdev might be NULL). 2002 */ 2003 cdev_t 2004 vn_todev(struct vnode *vp) 2005 { 2006 if (vp->v_type != VBLK && vp->v_type != VCHR) 2007 return (NULL); 2008 KKASSERT(vp->v_rdev != NULL); 2009 return (vp->v_rdev); 2010 } 2011 2012 /* 2013 * Check if vnode represents a disk device. The vnode does not need to be 2014 * opened. 2015 */ 2016 int 2017 vn_isdisk(struct vnode *vp, int *errp) 2018 { 2019 cdev_t dev; 2020 2021 if (vp->v_type != VCHR) { 2022 if (errp != NULL) 2023 *errp = ENOTBLK; 2024 return (0); 2025 } 2026 2027 if ((dev = vp->v_rdev) == NULL) 2028 dev = get_dev(vp->v_umajor, vp->v_uminor); 2029 2030 if (dev == NULL) { 2031 if (errp != NULL) 2032 *errp = ENXIO; 2033 return (0); 2034 } 2035 if (dev_is_good(dev) == 0) { 2036 if (errp != NULL) 2037 *errp = ENXIO; 2038 return (0); 2039 } 2040 if ((dev_dflags(dev) & D_DISK) == 0) { 2041 if (errp != NULL) 2042 *errp = ENOTBLK; 2043 return (0); 2044 } 2045 if (errp != NULL) 2046 *errp = 0; 2047 return (1); 2048 } 2049 2050 int 2051 vn_get_namelen(struct vnode *vp, int *namelen) 2052 { 2053 int error, retval[2]; 2054 2055 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2056 if (error) 2057 return (error); 2058 *namelen = *retval; 2059 return (0); 2060 } 2061 2062 int 2063 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2064 uint16_t d_namlen, const char *d_name) 2065 { 2066 struct dirent *dp; 2067 size_t len; 2068 2069 len = _DIRENT_RECLEN(d_namlen); 2070 if (len > uio->uio_resid) 2071 return(1); 2072 2073 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2074 2075 dp->d_ino = d_ino; 2076 dp->d_namlen = d_namlen; 2077 dp->d_type = d_type; 2078 bcopy(d_name, dp->d_name, d_namlen); 2079 2080 *error = uiomove((caddr_t)dp, len, uio); 2081 2082 kfree(dp, M_TEMP); 2083 2084 return(0); 2085 } 2086 2087