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 1.104 2007/05/09 00:53:34 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 } 260 261 /* 262 * Flush out and invalidate all buffers associated with a vnode. 263 * 264 * vp must be locked. 265 */ 266 static int vinvalbuf_bp(struct buf *bp, void *data); 267 268 struct vinvalbuf_bp_info { 269 struct vnode *vp; 270 int slptimeo; 271 int lkflags; 272 int flags; 273 }; 274 275 void 276 vupdatefsmid(struct vnode *vp) 277 { 278 atomic_set_int(&vp->v_flag, VFSMID); 279 } 280 281 int 282 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 283 { 284 struct vinvalbuf_bp_info info; 285 int error; 286 vm_object_t object; 287 288 /* 289 * If we are being asked to save, call fsync to ensure that the inode 290 * is updated. 291 */ 292 if (flags & V_SAVE) { 293 crit_enter(); 294 while (vp->v_track_write.bk_active) { 295 vp->v_track_write.bk_waitflag = 1; 296 error = tsleep(&vp->v_track_write, slpflag, 297 "vinvlbuf", slptimeo); 298 if (error) { 299 crit_exit(); 300 return (error); 301 } 302 } 303 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 304 crit_exit(); 305 if ((error = VOP_FSYNC(vp, MNT_WAIT)) != 0) 306 return (error); 307 crit_enter(); 308 if (vp->v_track_write.bk_active > 0 || 309 !RB_EMPTY(&vp->v_rbdirty_tree)) 310 panic("vinvalbuf: dirty bufs"); 311 } 312 crit_exit(); 313 } 314 crit_enter(); 315 info.slptimeo = slptimeo; 316 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 317 if (slpflag & PCATCH) 318 info.lkflags |= LK_PCATCH; 319 info.flags = flags; 320 info.vp = vp; 321 322 /* 323 * Flush the buffer cache until nothing is left. 324 */ 325 while (!RB_EMPTY(&vp->v_rbclean_tree) || 326 !RB_EMPTY(&vp->v_rbdirty_tree)) { 327 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 328 vinvalbuf_bp, &info); 329 if (error == 0) { 330 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 331 vinvalbuf_bp, &info); 332 } 333 } 334 335 /* 336 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 337 * have write I/O in-progress but if there is a VM object then the 338 * VM object can also have read-I/O in-progress. 339 */ 340 do { 341 while (vp->v_track_write.bk_active > 0) { 342 vp->v_track_write.bk_waitflag = 1; 343 tsleep(&vp->v_track_write, 0, "vnvlbv", 0); 344 } 345 if ((object = vp->v_object) != NULL) { 346 while (object->paging_in_progress) 347 vm_object_pip_sleep(object, "vnvlbx"); 348 } 349 } while (vp->v_track_write.bk_active > 0); 350 351 crit_exit(); 352 353 /* 354 * Destroy the copy in the VM cache, too. 355 */ 356 if ((object = vp->v_object) != NULL) { 357 vm_object_page_remove(object, 0, 0, 358 (flags & V_SAVE) ? TRUE : FALSE); 359 } 360 361 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 362 panic("vinvalbuf: flush failed"); 363 if (!RB_EMPTY(&vp->v_rbhash_tree)) 364 panic("vinvalbuf: flush failed, buffers still present"); 365 return (0); 366 } 367 368 static int 369 vinvalbuf_bp(struct buf *bp, void *data) 370 { 371 struct vinvalbuf_bp_info *info = data; 372 int error; 373 374 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 375 error = BUF_TIMELOCK(bp, info->lkflags, 376 "vinvalbuf", info->slptimeo); 377 if (error == 0) { 378 BUF_UNLOCK(bp); 379 error = ENOLCK; 380 } 381 if (error == ENOLCK) 382 return(0); 383 return (-error); 384 } 385 386 KKASSERT(bp->b_vp == info->vp); 387 388 /* 389 * XXX Since there are no node locks for NFS, I 390 * believe there is a slight chance that a delayed 391 * write will occur while sleeping just above, so 392 * check for it. Note that vfs_bio_awrite expects 393 * buffers to reside on a queue, while bwrite() and 394 * brelse() do not. 395 */ 396 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 397 (info->flags & V_SAVE)) { 398 if (bp->b_vp == info->vp) { 399 if (bp->b_flags & B_CLUSTEROK) { 400 vfs_bio_awrite(bp); 401 } else { 402 bremfree(bp); 403 bp->b_flags |= B_ASYNC; 404 bwrite(bp); 405 } 406 } else { 407 bremfree(bp); 408 bwrite(bp); 409 } 410 } else if (info->flags & V_SAVE) { 411 /* 412 * Cannot set B_NOCACHE on a clean buffer as this will 413 * destroy the VM backing store which might actually 414 * be dirty (and unsynchronized). 415 */ 416 bremfree(bp); 417 bp->b_flags |= (B_INVAL | B_RELBUF); 418 bp->b_flags &= ~B_ASYNC; 419 brelse(bp); 420 } else { 421 bremfree(bp); 422 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 423 bp->b_flags &= ~B_ASYNC; 424 brelse(bp); 425 } 426 return(0); 427 } 428 429 /* 430 * Truncate a file's buffer and pages to a specified length. This 431 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 432 * sync activity. 433 * 434 * The vnode must be locked. 435 */ 436 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 437 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 438 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 439 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 440 441 int 442 vtruncbuf(struct vnode *vp, off_t length, int blksize) 443 { 444 off_t truncloffset; 445 int count; 446 const char *filename; 447 448 /* 449 * Round up to the *next* block, then destroy the buffers in question. 450 * Since we are only removing some of the buffers we must rely on the 451 * scan count to determine whether a loop is necessary. 452 */ 453 if ((count = (int)(length % blksize)) != 0) 454 truncloffset = length + (blksize - count); 455 else 456 truncloffset = length; 457 458 crit_enter(); 459 do { 460 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 461 vtruncbuf_bp_trunc_cmp, 462 vtruncbuf_bp_trunc, &truncloffset); 463 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 464 vtruncbuf_bp_trunc_cmp, 465 vtruncbuf_bp_trunc, &truncloffset); 466 } while(count); 467 468 /* 469 * For safety, fsync any remaining metadata if the file is not being 470 * truncated to 0. Since the metadata does not represent the entire 471 * dirty list we have to rely on the hit count to ensure that we get 472 * all of it. 473 */ 474 if (length > 0) { 475 do { 476 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 477 vtruncbuf_bp_metasync_cmp, 478 vtruncbuf_bp_metasync, vp); 479 } while (count); 480 } 481 482 /* 483 * Clean out any left over VM backing store. 484 */ 485 crit_exit(); 486 487 vnode_pager_setsize(vp, length); 488 489 crit_enter(); 490 491 /* 492 * It is possible to have in-progress I/O from buffers that were 493 * not part of the truncation. This should not happen if we 494 * are truncating to 0-length. 495 */ 496 filename = TAILQ_FIRST(&vp->v_namecache) ? 497 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 498 499 while ((count = vp->v_track_write.bk_active) > 0) { 500 vp->v_track_write.bk_waitflag = 1; 501 tsleep(&vp->v_track_write, 0, "vbtrunc", 0); 502 if (length == 0) { 503 kprintf("Warning: vtruncbuf(): Had to wait for " 504 "%d buffer I/Os to finish in %s\n", 505 count, filename); 506 } 507 } 508 509 /* 510 * Make sure no buffers were instantiated while we were trying 511 * to clean out the remaining VM pages. This could occur due 512 * to busy dirty VM pages being flushed out to disk. 513 */ 514 do { 515 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 516 vtruncbuf_bp_trunc_cmp, 517 vtruncbuf_bp_trunc, &truncloffset); 518 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 519 vtruncbuf_bp_trunc_cmp, 520 vtruncbuf_bp_trunc, &truncloffset); 521 if (count) { 522 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 523 "left over buffers in %s\n", count, filename); 524 } 525 } while(count); 526 527 crit_exit(); 528 529 return (0); 530 } 531 532 /* 533 * The callback buffer is beyond the new file EOF and must be destroyed. 534 * Note that the compare function must conform to the RB_SCAN's requirements. 535 */ 536 static 537 int 538 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 539 { 540 if (bp->b_loffset >= *(off_t *)data) 541 return(0); 542 return(-1); 543 } 544 545 static 546 int 547 vtruncbuf_bp_trunc(struct buf *bp, void *data) 548 { 549 /* 550 * Do not try to use a buffer we cannot immediately lock, but sleep 551 * anyway to prevent a livelock. The code will loop until all buffers 552 * can be acted upon. 553 */ 554 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 555 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 556 BUF_UNLOCK(bp); 557 } else { 558 bremfree(bp); 559 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 560 bp->b_flags &= ~B_ASYNC; 561 brelse(bp); 562 } 563 return(1); 564 } 565 566 /* 567 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 568 * blocks (with a negative loffset) are scanned. 569 * Note that the compare function must conform to the RB_SCAN's requirements. 570 */ 571 static int 572 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 573 { 574 if (bp->b_loffset < 0) 575 return(0); 576 return(1); 577 } 578 579 static int 580 vtruncbuf_bp_metasync(struct buf *bp, void *data) 581 { 582 struct vnode *vp = data; 583 584 if (bp->b_flags & B_DELWRI) { 585 /* 586 * Do not try to use a buffer we cannot immediately lock, 587 * but sleep anyway to prevent a livelock. The code will 588 * loop until all buffers can be acted upon. 589 */ 590 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 591 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 592 BUF_UNLOCK(bp); 593 } else { 594 bremfree(bp); 595 if (bp->b_vp == vp) { 596 bp->b_flags |= B_ASYNC; 597 } else { 598 bp->b_flags &= ~B_ASYNC; 599 } 600 bwrite(bp); 601 } 602 return(1); 603 } else { 604 return(0); 605 } 606 } 607 608 /* 609 * vfsync - implements a multipass fsync on a file which understands 610 * dependancies and meta-data. The passed vnode must be locked. The 611 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 612 * 613 * When fsyncing data asynchronously just do one consolidated pass starting 614 * with the most negative block number. This may not get all the data due 615 * to dependancies. 616 * 617 * When fsyncing data synchronously do a data pass, then a metadata pass, 618 * then do additional data+metadata passes to try to get all the data out. 619 */ 620 static int vfsync_wait_output(struct vnode *vp, 621 int (*waitoutput)(struct vnode *, struct thread *)); 622 static int vfsync_data_only_cmp(struct buf *bp, void *data); 623 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 624 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 625 static int vfsync_bp(struct buf *bp, void *data); 626 627 struct vfsync_info { 628 struct vnode *vp; 629 int synchronous; 630 int syncdeps; 631 int lazycount; 632 int lazylimit; 633 int skippedbufs; 634 int (*checkdef)(struct buf *); 635 }; 636 637 int 638 vfsync(struct vnode *vp, int waitfor, int passes, 639 int (*checkdef)(struct buf *), 640 int (*waitoutput)(struct vnode *, struct thread *)) 641 { 642 struct vfsync_info info; 643 int error; 644 645 bzero(&info, sizeof(info)); 646 info.vp = vp; 647 if ((info.checkdef = checkdef) == NULL) 648 info.syncdeps = 1; 649 650 crit_enter_id("vfsync"); 651 652 switch(waitfor) { 653 case MNT_LAZY: 654 /* 655 * Lazy (filesystem syncer typ) Asynchronous plus limit the 656 * number of data (not meta) pages we try to flush to 1MB. 657 * A non-zero return means that lazy limit was reached. 658 */ 659 info.lazylimit = 1024 * 1024; 660 info.syncdeps = 1; 661 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 662 vfsync_lazy_range_cmp, vfsync_bp, &info); 663 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 664 vfsync_meta_only_cmp, vfsync_bp, &info); 665 if (error == 0) 666 vp->v_lazyw = 0; 667 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 668 vn_syncer_add_to_worklist(vp, 1); 669 error = 0; 670 break; 671 case MNT_NOWAIT: 672 /* 673 * Asynchronous. Do a data-only pass and a meta-only pass. 674 */ 675 info.syncdeps = 1; 676 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 677 vfsync_bp, &info); 678 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 679 vfsync_bp, &info); 680 error = 0; 681 break; 682 default: 683 /* 684 * Synchronous. Do a data-only pass, then a meta-data+data 685 * pass, then additional integrated passes to try to get 686 * all the dependancies flushed. 687 */ 688 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 689 vfsync_bp, &info); 690 error = vfsync_wait_output(vp, waitoutput); 691 if (error == 0) { 692 info.skippedbufs = 0; 693 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 694 vfsync_bp, &info); 695 error = vfsync_wait_output(vp, waitoutput); 696 if (info.skippedbufs) 697 kprintf("Warning: vfsync skipped %d dirty bufs in pass2!\n", info.skippedbufs); 698 } 699 while (error == 0 && passes > 0 && 700 !RB_EMPTY(&vp->v_rbdirty_tree)) { 701 if (--passes == 0) { 702 info.synchronous = 1; 703 info.syncdeps = 1; 704 } 705 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 706 vfsync_bp, &info); 707 if (error < 0) 708 error = -error; 709 info.syncdeps = 1; 710 if (error == 0) 711 error = vfsync_wait_output(vp, waitoutput); 712 } 713 break; 714 } 715 crit_exit_id("vfsync"); 716 return(error); 717 } 718 719 static int 720 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 721 { 722 int error = 0; 723 724 while (vp->v_track_write.bk_active) { 725 vp->v_track_write.bk_waitflag = 1; 726 tsleep(&vp->v_track_write, 0, "fsfsn", 0); 727 } 728 if (waitoutput) 729 error = waitoutput(vp, curthread); 730 return(error); 731 } 732 733 static int 734 vfsync_data_only_cmp(struct buf *bp, void *data) 735 { 736 if (bp->b_loffset < 0) 737 return(-1); 738 return(0); 739 } 740 741 static int 742 vfsync_meta_only_cmp(struct buf *bp, void *data) 743 { 744 if (bp->b_loffset < 0) 745 return(0); 746 return(1); 747 } 748 749 static int 750 vfsync_lazy_range_cmp(struct buf *bp, void *data) 751 { 752 struct vfsync_info *info = data; 753 if (bp->b_loffset < info->vp->v_lazyw) 754 return(-1); 755 return(0); 756 } 757 758 static int 759 vfsync_bp(struct buf *bp, void *data) 760 { 761 struct vfsync_info *info = data; 762 struct vnode *vp = info->vp; 763 int error; 764 765 /* 766 * if syncdeps is not set we do not try to write buffers which have 767 * dependancies. 768 */ 769 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 770 return(0); 771 772 /* 773 * Ignore buffers that we cannot immediately lock. XXX 774 */ 775 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 776 kprintf("Warning: vfsync_bp skipping dirty buffer %p\n", bp); 777 ++info->skippedbufs; 778 return(0); 779 } 780 if ((bp->b_flags & B_DELWRI) == 0) 781 panic("vfsync_bp: buffer not dirty"); 782 if (vp != bp->b_vp) 783 panic("vfsync_bp: buffer vp mismatch"); 784 785 /* 786 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 787 * has been written but an additional handshake with the device 788 * is required before we can dispose of the buffer. We have no idea 789 * how to do this so we have to skip these buffers. 790 */ 791 if (bp->b_flags & B_NEEDCOMMIT) { 792 BUF_UNLOCK(bp); 793 return(0); 794 } 795 796 if (info->synchronous) { 797 /* 798 * Synchronous flushing. An error may be returned. 799 */ 800 bremfree(bp); 801 crit_exit_id("vfsync"); 802 error = bwrite(bp); 803 crit_enter_id("vfsync"); 804 } else { 805 /* 806 * Asynchronous flushing. A negative return value simply 807 * stops the scan and is not considered an error. We use 808 * this to support limited MNT_LAZY flushes. 809 */ 810 vp->v_lazyw = bp->b_loffset; 811 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 812 info->lazycount += vfs_bio_awrite(bp); 813 } else { 814 info->lazycount += bp->b_bufsize; 815 bremfree(bp); 816 crit_exit_id("vfsync"); 817 bawrite(bp); 818 crit_enter_id("vfsync"); 819 } 820 if (info->lazylimit && info->lazycount >= info->lazylimit) 821 error = 1; 822 else 823 error = 0; 824 } 825 return(-error); 826 } 827 828 /* 829 * Associate a buffer with a vnode. 830 */ 831 void 832 bgetvp(struct vnode *vp, struct buf *bp) 833 { 834 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 835 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 836 837 vhold(vp); 838 /* 839 * Insert onto list for new vnode. 840 */ 841 crit_enter(); 842 bp->b_vp = vp; 843 bp->b_flags |= B_HASHED; 844 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) 845 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 846 847 bp->b_flags |= B_VNCLEAN; 848 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 849 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 850 crit_exit(); 851 } 852 853 /* 854 * Disassociate a buffer from a vnode. 855 */ 856 void 857 brelvp(struct buf *bp) 858 { 859 struct vnode *vp; 860 861 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 862 863 /* 864 * Delete from old vnode list, if on one. 865 */ 866 vp = bp->b_vp; 867 crit_enter(); 868 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 869 if (bp->b_flags & B_VNDIRTY) 870 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 871 else 872 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 873 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 874 } 875 if (bp->b_flags & B_HASHED) { 876 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 877 bp->b_flags &= ~B_HASHED; 878 } 879 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 880 vp->v_flag &= ~VONWORKLST; 881 LIST_REMOVE(vp, v_synclist); 882 } 883 crit_exit(); 884 bp->b_vp = NULL; 885 vdrop(vp); 886 } 887 888 /* 889 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 890 * This routine is called when the state of the B_DELWRI bit is changed. 891 */ 892 void 893 reassignbuf(struct buf *bp) 894 { 895 struct vnode *vp = bp->b_vp; 896 int delay; 897 898 KKASSERT(vp != NULL); 899 ++reassignbufcalls; 900 901 /* 902 * B_PAGING flagged buffers cannot be reassigned because their vp 903 * is not fully linked in. 904 */ 905 if (bp->b_flags & B_PAGING) 906 panic("cannot reassign paging buffer"); 907 908 crit_enter(); 909 if (bp->b_flags & B_DELWRI) { 910 /* 911 * Move to the dirty list, add the vnode to the worklist 912 */ 913 if (bp->b_flags & B_VNCLEAN) { 914 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 915 bp->b_flags &= ~B_VNCLEAN; 916 } 917 if ((bp->b_flags & B_VNDIRTY) == 0) { 918 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 919 panic("reassignbuf: dup lblk vp %p bp %p", 920 vp, bp); 921 } 922 bp->b_flags |= B_VNDIRTY; 923 } 924 if ((vp->v_flag & VONWORKLST) == 0) { 925 switch (vp->v_type) { 926 case VDIR: 927 delay = dirdelay; 928 break; 929 case VCHR: 930 case VBLK: 931 if (vp->v_rdev && 932 vp->v_rdev->si_mountpoint != NULL) { 933 delay = metadelay; 934 break; 935 } 936 /* fall through */ 937 default: 938 delay = filedelay; 939 } 940 vn_syncer_add_to_worklist(vp, delay); 941 } 942 } else { 943 /* 944 * Move to the clean list, remove the vnode from the worklist 945 * if no dirty blocks remain. 946 */ 947 if (bp->b_flags & B_VNDIRTY) { 948 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 949 bp->b_flags &= ~B_VNDIRTY; 950 } 951 if ((bp->b_flags & B_VNCLEAN) == 0) { 952 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 953 panic("reassignbuf: dup lblk vp %p bp %p", 954 vp, bp); 955 } 956 bp->b_flags |= B_VNCLEAN; 957 } 958 if ((vp->v_flag & VONWORKLST) && 959 RB_EMPTY(&vp->v_rbdirty_tree)) { 960 vp->v_flag &= ~VONWORKLST; 961 LIST_REMOVE(vp, v_synclist); 962 } 963 } 964 crit_exit(); 965 } 966 967 /* 968 * Create a vnode for a block device. 969 * Used for mounting the root file system. 970 */ 971 int 972 bdevvp(cdev_t dev, struct vnode **vpp) 973 { 974 struct vnode *vp; 975 struct vnode *nvp; 976 int error; 977 978 if (dev == NULL) { 979 *vpp = NULLVP; 980 return (ENXIO); 981 } 982 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops_p, &nvp, 0, 0); 983 if (error) { 984 *vpp = NULLVP; 985 return (error); 986 } 987 vp = nvp; 988 vp->v_type = VCHR; 989 vp->v_umajor = dev->si_umajor; 990 vp->v_uminor = dev->si_uminor; 991 vx_unlock(vp); 992 *vpp = vp; 993 return (0); 994 } 995 996 int 997 v_associate_rdev(struct vnode *vp, cdev_t dev) 998 { 999 lwkt_tokref ilock; 1000 1001 if (dev == NULL) 1002 return(ENXIO); 1003 if (dev_is_good(dev) == 0) 1004 return(ENXIO); 1005 KKASSERT(vp->v_rdev == NULL); 1006 if (dev_ref_debug) 1007 kprintf("Z1"); 1008 vp->v_rdev = reference_dev(dev); 1009 lwkt_gettoken(&ilock, &spechash_token); 1010 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1011 lwkt_reltoken(&ilock); 1012 return(0); 1013 } 1014 1015 void 1016 v_release_rdev(struct vnode *vp) 1017 { 1018 lwkt_tokref ilock; 1019 cdev_t dev; 1020 1021 if ((dev = vp->v_rdev) != NULL) { 1022 lwkt_gettoken(&ilock, &spechash_token); 1023 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1024 vp->v_rdev = NULL; 1025 release_dev(dev); 1026 lwkt_reltoken(&ilock); 1027 } 1028 } 1029 1030 /* 1031 * Add a vnode to the alias list hung off the cdev_t. We only associate 1032 * the device number with the vnode. The actual device is not associated 1033 * until the vnode is opened (usually in spec_open()), and will be 1034 * disassociated on last close. 1035 */ 1036 void 1037 addaliasu(struct vnode *nvp, int x, int y) 1038 { 1039 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1040 panic("addaliasu on non-special vnode"); 1041 nvp->v_umajor = x; 1042 nvp->v_uminor = y; 1043 } 1044 1045 /* 1046 * Disassociate a vnode from its underlying filesystem. 1047 * 1048 * The vnode must be VX locked and referenced. In all normal situations 1049 * there are no active references. If vclean_vxlocked() is called while 1050 * there are active references, the vnode is being ripped out and we have 1051 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1052 */ 1053 void 1054 vclean_vxlocked(struct vnode *vp, int flags) 1055 { 1056 int active; 1057 int n; 1058 vm_object_t object; 1059 1060 /* 1061 * If the vnode has already been reclaimed we have nothing to do. 1062 */ 1063 if (vp->v_flag & VRECLAIMED) 1064 return; 1065 vp->v_flag |= VRECLAIMED; 1066 1067 /* 1068 * Scrap the vfs cache 1069 */ 1070 while (cache_inval_vp(vp, 0) != 0) { 1071 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1072 tsleep(vp, 0, "vclninv", 2); 1073 } 1074 1075 /* 1076 * Check to see if the vnode is in use. If so we have to reference it 1077 * before we clean it out so that its count cannot fall to zero and 1078 * generate a race against ourselves to recycle it. 1079 */ 1080 active = sysref_isactive(&vp->v_sysref); 1081 1082 /* 1083 * Clean out any buffers associated with the vnode and destroy its 1084 * object, if it has one. 1085 */ 1086 vinvalbuf(vp, V_SAVE, 0, 0); 1087 1088 /* 1089 * If purging an active vnode (typically during a forced unmount 1090 * or reboot), it must be closed and deactivated before being 1091 * reclaimed. This isn't really all that safe, but what can 1092 * we do? XXX. 1093 * 1094 * Note that neither of these routines unlocks the vnode. 1095 */ 1096 if (active && (flags & DOCLOSE)) { 1097 while ((n = vp->v_opencount) != 0) { 1098 if (vp->v_writecount) 1099 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1100 else 1101 VOP_CLOSE(vp, FNONBLOCK); 1102 if (vp->v_opencount == n) { 1103 kprintf("Warning: unable to force-close" 1104 " vnode %p\n", vp); 1105 break; 1106 } 1107 } 1108 } 1109 1110 /* 1111 * If the vnode has not be deactivated, deactivated it. Deactivation 1112 * can create new buffers and VM pages so we have to call vinvalbuf() 1113 * again to make sure they all get flushed. 1114 * 1115 * This can occur if a file with a link count of 0 needs to be 1116 * truncated. 1117 */ 1118 if ((vp->v_flag & VINACTIVE) == 0) { 1119 vp->v_flag |= VINACTIVE; 1120 VOP_INACTIVE(vp); 1121 vinvalbuf(vp, V_SAVE, 0, 0); 1122 } 1123 1124 /* 1125 * If the vnode has an object, destroy it. 1126 */ 1127 if ((object = vp->v_object) != NULL) { 1128 if (object->ref_count == 0) { 1129 if ((object->flags & OBJ_DEAD) == 0) 1130 vm_object_terminate(object); 1131 } else { 1132 vm_pager_deallocate(object); 1133 } 1134 vp->v_flag &= ~VOBJBUF; 1135 } 1136 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1137 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 /* 1154 * Eliminate all activity associated with the requested vnode 1155 * and with all vnodes aliased to the requested vnode. 1156 * 1157 * The vnode must be referenced and vx_lock()'d 1158 * 1159 * revoke { struct vnode *a_vp, int a_flags } 1160 */ 1161 int 1162 vop_stdrevoke(struct vop_revoke_args *ap) 1163 { 1164 struct vnode *vp, *vq; 1165 lwkt_tokref ilock; 1166 cdev_t dev; 1167 1168 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1169 1170 vp = ap->a_vp; 1171 1172 /* 1173 * If the vnode is already dead don't try to revoke it 1174 */ 1175 if (vp->v_flag & VRECLAIMED) 1176 return (0); 1177 1178 /* 1179 * If the vnode has a device association, scrap all vnodes associated 1180 * with the device. Don't let the device disappear on us while we 1181 * are scrapping the vnodes. 1182 * 1183 * The passed vp will probably show up in the list, do not VX lock 1184 * it twice! 1185 */ 1186 if (vp->v_type != VCHR) 1187 return(0); 1188 if ((dev = vp->v_rdev) == NULL) { 1189 if ((dev = get_dev(vp->v_umajor, vp->v_uminor)) == NULL) 1190 return(0); 1191 } 1192 reference_dev(dev); 1193 lwkt_gettoken(&ilock, &spechash_token); 1194 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1195 if (vp != vq) 1196 vx_get(vq); 1197 if (vq == SLIST_FIRST(&dev->si_hlist)) 1198 vgone_vxlocked(vq); 1199 if (vp != vq) 1200 vx_put(vq); 1201 } 1202 lwkt_reltoken(&ilock); 1203 release_dev(dev); 1204 return (0); 1205 } 1206 1207 /* 1208 * This is called when the object underlying a vnode is being destroyed, 1209 * such as in a remove(). Try to recycle the vnode immediately if the 1210 * only active reference is our reference. 1211 */ 1212 int 1213 vrecycle(struct vnode *vp) 1214 { 1215 if (vp->v_sysref.refcnt == 1) { 1216 vgone_vxlocked(vp); 1217 return (1); 1218 } 1219 return (0); 1220 } 1221 1222 /* 1223 * Eliminate all activity associated with a vnode in preparation for reuse. 1224 * 1225 * The vnode must be VX locked and refd and will remain VX locked and refd 1226 * on return. This routine may be called with the vnode in any state, as 1227 * long as it is VX locked. The vnode will be cleaned out and marked 1228 * VRECLAIMED but will not actually be reused until all existing refs and 1229 * holds go away. 1230 * 1231 * NOTE: This routine may be called on a vnode which has not yet been 1232 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1233 * already been reclaimed. 1234 * 1235 * This routine is not responsible for placing us back on the freelist. 1236 * Instead, it happens automatically when the caller releases the VX lock 1237 * (assuming there aren't any other references). 1238 */ 1239 1240 void 1241 vgone_vxlocked(struct vnode *vp) 1242 { 1243 /* 1244 * assert that the VX lock is held. This is an absolute requirement 1245 * now for vgone_vxlocked() to be called. 1246 */ 1247 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1248 1249 /* 1250 * Clean out the filesystem specific data and set the VRECLAIMED 1251 * bit. Also deactivate the vnode if necessary. 1252 */ 1253 vclean_vxlocked(vp, DOCLOSE); 1254 1255 /* 1256 * Delete from old mount point vnode list, if on one. 1257 */ 1258 if (vp->v_mount != NULL) 1259 insmntque(vp, NULL); 1260 1261 /* 1262 * If special device, remove it from special device alias list 1263 * if it is on one. This should normally only occur if a vnode is 1264 * being revoked as the device should otherwise have been released 1265 * naturally. 1266 */ 1267 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1268 v_release_rdev(vp); 1269 } 1270 1271 /* 1272 * Set us to VBAD 1273 */ 1274 vp->v_type = VBAD; 1275 } 1276 1277 /* 1278 * Lookup a vnode by device number. 1279 */ 1280 int 1281 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1282 { 1283 lwkt_tokref ilock; 1284 struct vnode *vp; 1285 1286 lwkt_gettoken(&ilock, &spechash_token); 1287 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1288 if (type == vp->v_type) { 1289 *vpp = vp; 1290 lwkt_reltoken(&ilock); 1291 return (1); 1292 } 1293 } 1294 lwkt_reltoken(&ilock); 1295 return (0); 1296 } 1297 1298 /* 1299 * Calculate the total number of references to a special device. This 1300 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1301 * an overloaded field. Since udev2dev can now return NULL, we have 1302 * to check for a NULL v_rdev. 1303 */ 1304 int 1305 count_dev(cdev_t dev) 1306 { 1307 lwkt_tokref ilock; 1308 struct vnode *vp; 1309 int count = 0; 1310 1311 if (SLIST_FIRST(&dev->si_hlist)) { 1312 lwkt_gettoken(&ilock, &spechash_token); 1313 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1314 if (vp->v_sysref.refcnt > 0) 1315 count += vp->v_sysref.refcnt; 1316 } 1317 lwkt_reltoken(&ilock); 1318 } 1319 return(count); 1320 } 1321 1322 int 1323 count_udev(int x, int y) 1324 { 1325 cdev_t dev; 1326 1327 if ((dev = get_dev(x, y)) == NULL) 1328 return(0); 1329 return(count_dev(dev)); 1330 } 1331 1332 int 1333 vcount(struct vnode *vp) 1334 { 1335 if (vp->v_rdev == NULL) 1336 return(0); 1337 return(count_dev(vp->v_rdev)); 1338 } 1339 1340 /* 1341 * Initialize VMIO for a vnode. This routine MUST be called before a 1342 * VFS can issue buffer cache ops on a vnode. It is typically called 1343 * when a vnode is initialized from its inode. 1344 */ 1345 int 1346 vinitvmio(struct vnode *vp, off_t filesize) 1347 { 1348 vm_object_t object; 1349 int error = 0; 1350 1351 retry: 1352 if ((object = vp->v_object) == NULL) { 1353 object = vnode_pager_alloc(vp, filesize, 0, 0); 1354 /* 1355 * Dereference the reference we just created. This assumes 1356 * that the object is associated with the vp. 1357 */ 1358 object->ref_count--; 1359 vrele(vp); 1360 } else { 1361 if (object->flags & OBJ_DEAD) { 1362 vn_unlock(vp); 1363 tsleep(object, 0, "vodead", 0); 1364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1365 goto retry; 1366 } 1367 } 1368 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1369 vp->v_flag |= VOBJBUF; 1370 return (error); 1371 } 1372 1373 1374 /* 1375 * Print out a description of a vnode. 1376 */ 1377 static char *typename[] = 1378 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1379 1380 void 1381 vprint(char *label, struct vnode *vp) 1382 { 1383 char buf[96]; 1384 1385 if (label != NULL) 1386 kprintf("%s: %p: ", label, (void *)vp); 1387 else 1388 kprintf("%p: ", (void *)vp); 1389 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1390 typename[vp->v_type], 1391 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1392 buf[0] = '\0'; 1393 if (vp->v_flag & VROOT) 1394 strcat(buf, "|VROOT"); 1395 if (vp->v_flag & VTEXT) 1396 strcat(buf, "|VTEXT"); 1397 if (vp->v_flag & VSYSTEM) 1398 strcat(buf, "|VSYSTEM"); 1399 if (vp->v_flag & VFREE) 1400 strcat(buf, "|VFREE"); 1401 if (vp->v_flag & VOBJBUF) 1402 strcat(buf, "|VOBJBUF"); 1403 if (buf[0] != '\0') 1404 kprintf(" flags (%s)", &buf[1]); 1405 if (vp->v_data == NULL) { 1406 kprintf("\n"); 1407 } else { 1408 kprintf("\n\t"); 1409 VOP_PRINT(vp); 1410 } 1411 } 1412 1413 #ifdef DDB 1414 #include <ddb/ddb.h> 1415 1416 static int db_show_locked_vnodes(struct mount *mp, void *data); 1417 1418 /* 1419 * List all of the locked vnodes in the system. 1420 * Called when debugging the kernel. 1421 */ 1422 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1423 { 1424 kprintf("Locked vnodes\n"); 1425 mountlist_scan(db_show_locked_vnodes, NULL, 1426 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1427 } 1428 1429 static int 1430 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1431 { 1432 struct vnode *vp; 1433 1434 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1435 if (vn_islocked(vp)) 1436 vprint((char *)0, vp); 1437 } 1438 return(0); 1439 } 1440 #endif 1441 1442 /* 1443 * Top level filesystem related information gathering. 1444 */ 1445 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1446 1447 static int 1448 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1449 { 1450 int *name = (int *)arg1 - 1; /* XXX */ 1451 u_int namelen = arg2 + 1; /* XXX */ 1452 struct vfsconf *vfsp; 1453 1454 #if 1 || defined(COMPAT_PRELITE2) 1455 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1456 if (namelen == 1) 1457 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1458 #endif 1459 1460 #ifdef notyet 1461 /* all sysctl names at this level are at least name and field */ 1462 if (namelen < 2) 1463 return (ENOTDIR); /* overloaded */ 1464 if (name[0] != VFS_GENERIC) { 1465 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1466 if (vfsp->vfc_typenum == name[0]) 1467 break; 1468 if (vfsp == NULL) 1469 return (EOPNOTSUPP); 1470 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1471 oldp, oldlenp, newp, newlen, p)); 1472 } 1473 #endif 1474 switch (name[1]) { 1475 case VFS_MAXTYPENUM: 1476 if (namelen != 2) 1477 return (ENOTDIR); 1478 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1479 case VFS_CONF: 1480 if (namelen != 3) 1481 return (ENOTDIR); /* overloaded */ 1482 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1483 if (vfsp->vfc_typenum == name[2]) 1484 break; 1485 if (vfsp == NULL) 1486 return (EOPNOTSUPP); 1487 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1488 } 1489 return (EOPNOTSUPP); 1490 } 1491 1492 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1493 "Generic filesystem"); 1494 1495 #if 1 || defined(COMPAT_PRELITE2) 1496 1497 static int 1498 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1499 { 1500 int error; 1501 struct vfsconf *vfsp; 1502 struct ovfsconf ovfs; 1503 1504 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1505 bzero(&ovfs, sizeof(ovfs)); 1506 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1507 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1508 ovfs.vfc_index = vfsp->vfc_typenum; 1509 ovfs.vfc_refcount = vfsp->vfc_refcount; 1510 ovfs.vfc_flags = vfsp->vfc_flags; 1511 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1512 if (error) 1513 return error; 1514 } 1515 return 0; 1516 } 1517 1518 #endif /* 1 || COMPAT_PRELITE2 */ 1519 1520 /* 1521 * Check to see if a filesystem is mounted on a block device. 1522 */ 1523 int 1524 vfs_mountedon(struct vnode *vp) 1525 { 1526 cdev_t dev; 1527 1528 if ((dev = vp->v_rdev) == NULL) { 1529 if (vp->v_type != VBLK) 1530 dev = get_dev(vp->v_uminor, vp->v_umajor); 1531 } 1532 if (dev != NULL && dev->si_mountpoint) 1533 return (EBUSY); 1534 return (0); 1535 } 1536 1537 /* 1538 * Unmount all filesystems. The list is traversed in reverse order 1539 * of mounting to avoid dependencies. 1540 */ 1541 1542 static int vfs_umountall_callback(struct mount *mp, void *data); 1543 1544 void 1545 vfs_unmountall(void) 1546 { 1547 int count; 1548 1549 do { 1550 count = mountlist_scan(vfs_umountall_callback, 1551 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1552 } while (count); 1553 } 1554 1555 static 1556 int 1557 vfs_umountall_callback(struct mount *mp, void *data) 1558 { 1559 int error; 1560 1561 error = dounmount(mp, MNT_FORCE); 1562 if (error) { 1563 mountlist_remove(mp); 1564 kprintf("unmount of filesystem mounted from %s failed (", 1565 mp->mnt_stat.f_mntfromname); 1566 if (error == EBUSY) 1567 kprintf("BUSY)\n"); 1568 else 1569 kprintf("%d)\n", error); 1570 } 1571 return(1); 1572 } 1573 1574 /* 1575 * Build hash lists of net addresses and hang them off the mount point. 1576 * Called by ufs_mount() to set up the lists of export addresses. 1577 */ 1578 static int 1579 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1580 struct export_args *argp) 1581 { 1582 struct netcred *np; 1583 struct radix_node_head *rnh; 1584 int i; 1585 struct radix_node *rn; 1586 struct sockaddr *saddr, *smask = 0; 1587 struct domain *dom; 1588 int error; 1589 1590 if (argp->ex_addrlen == 0) { 1591 if (mp->mnt_flag & MNT_DEFEXPORTED) 1592 return (EPERM); 1593 np = &nep->ne_defexported; 1594 np->netc_exflags = argp->ex_flags; 1595 np->netc_anon = argp->ex_anon; 1596 np->netc_anon.cr_ref = 1; 1597 mp->mnt_flag |= MNT_DEFEXPORTED; 1598 return (0); 1599 } 1600 1601 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1602 return (EINVAL); 1603 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1604 return (EINVAL); 1605 1606 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1607 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK); 1608 bzero((caddr_t) np, i); 1609 saddr = (struct sockaddr *) (np + 1); 1610 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1611 goto out; 1612 if (saddr->sa_len > argp->ex_addrlen) 1613 saddr->sa_len = argp->ex_addrlen; 1614 if (argp->ex_masklen) { 1615 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1616 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1617 if (error) 1618 goto out; 1619 if (smask->sa_len > argp->ex_masklen) 1620 smask->sa_len = argp->ex_masklen; 1621 } 1622 i = saddr->sa_family; 1623 if ((rnh = nep->ne_rtable[i]) == 0) { 1624 /* 1625 * Seems silly to initialize every AF when most are not used, 1626 * do so on demand here 1627 */ 1628 SLIST_FOREACH(dom, &domains, dom_next) 1629 if (dom->dom_family == i && dom->dom_rtattach) { 1630 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1631 dom->dom_rtoffset); 1632 break; 1633 } 1634 if ((rnh = nep->ne_rtable[i]) == 0) { 1635 error = ENOBUFS; 1636 goto out; 1637 } 1638 } 1639 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1640 np->netc_rnodes); 1641 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1642 error = EPERM; 1643 goto out; 1644 } 1645 np->netc_exflags = argp->ex_flags; 1646 np->netc_anon = argp->ex_anon; 1647 np->netc_anon.cr_ref = 1; 1648 return (0); 1649 out: 1650 kfree(np, M_NETADDR); 1651 return (error); 1652 } 1653 1654 /* ARGSUSED */ 1655 static int 1656 vfs_free_netcred(struct radix_node *rn, void *w) 1657 { 1658 struct radix_node_head *rnh = (struct radix_node_head *) w; 1659 1660 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1661 kfree((caddr_t) rn, M_NETADDR); 1662 return (0); 1663 } 1664 1665 /* 1666 * Free the net address hash lists that are hanging off the mount points. 1667 */ 1668 static void 1669 vfs_free_addrlist(struct netexport *nep) 1670 { 1671 int i; 1672 struct radix_node_head *rnh; 1673 1674 for (i = 0; i <= AF_MAX; i++) 1675 if ((rnh = nep->ne_rtable[i])) { 1676 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1677 (caddr_t) rnh); 1678 kfree((caddr_t) rnh, M_RTABLE); 1679 nep->ne_rtable[i] = 0; 1680 } 1681 } 1682 1683 int 1684 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1685 { 1686 int error; 1687 1688 if (argp->ex_flags & MNT_DELEXPORT) { 1689 if (mp->mnt_flag & MNT_EXPUBLIC) { 1690 vfs_setpublicfs(NULL, NULL, NULL); 1691 mp->mnt_flag &= ~MNT_EXPUBLIC; 1692 } 1693 vfs_free_addrlist(nep); 1694 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1695 } 1696 if (argp->ex_flags & MNT_EXPORTED) { 1697 if (argp->ex_flags & MNT_EXPUBLIC) { 1698 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1699 return (error); 1700 mp->mnt_flag |= MNT_EXPUBLIC; 1701 } 1702 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1703 return (error); 1704 mp->mnt_flag |= MNT_EXPORTED; 1705 } 1706 return (0); 1707 } 1708 1709 1710 /* 1711 * Set the publicly exported filesystem (WebNFS). Currently, only 1712 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1713 */ 1714 int 1715 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1716 struct export_args *argp) 1717 { 1718 int error; 1719 struct vnode *rvp; 1720 char *cp; 1721 1722 /* 1723 * mp == NULL -> invalidate the current info, the FS is 1724 * no longer exported. May be called from either vfs_export 1725 * or unmount, so check if it hasn't already been done. 1726 */ 1727 if (mp == NULL) { 1728 if (nfs_pub.np_valid) { 1729 nfs_pub.np_valid = 0; 1730 if (nfs_pub.np_index != NULL) { 1731 FREE(nfs_pub.np_index, M_TEMP); 1732 nfs_pub.np_index = NULL; 1733 } 1734 } 1735 return (0); 1736 } 1737 1738 /* 1739 * Only one allowed at a time. 1740 */ 1741 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1742 return (EBUSY); 1743 1744 /* 1745 * Get real filehandle for root of exported FS. 1746 */ 1747 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1748 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1749 1750 if ((error = VFS_ROOT(mp, &rvp))) 1751 return (error); 1752 1753 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1754 return (error); 1755 1756 vput(rvp); 1757 1758 /* 1759 * If an indexfile was specified, pull it in. 1760 */ 1761 if (argp->ex_indexfile != NULL) { 1762 int namelen; 1763 1764 error = vn_get_namelen(rvp, &namelen); 1765 if (error) 1766 return (error); 1767 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 1768 M_WAITOK); 1769 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1770 namelen, (size_t *)0); 1771 if (!error) { 1772 /* 1773 * Check for illegal filenames. 1774 */ 1775 for (cp = nfs_pub.np_index; *cp; cp++) { 1776 if (*cp == '/') { 1777 error = EINVAL; 1778 break; 1779 } 1780 } 1781 } 1782 if (error) { 1783 FREE(nfs_pub.np_index, M_TEMP); 1784 return (error); 1785 } 1786 } 1787 1788 nfs_pub.np_mount = mp; 1789 nfs_pub.np_valid = 1; 1790 return (0); 1791 } 1792 1793 struct netcred * 1794 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1795 struct sockaddr *nam) 1796 { 1797 struct netcred *np; 1798 struct radix_node_head *rnh; 1799 struct sockaddr *saddr; 1800 1801 np = NULL; 1802 if (mp->mnt_flag & MNT_EXPORTED) { 1803 /* 1804 * Lookup in the export list first. 1805 */ 1806 if (nam != NULL) { 1807 saddr = nam; 1808 rnh = nep->ne_rtable[saddr->sa_family]; 1809 if (rnh != NULL) { 1810 np = (struct netcred *) 1811 (*rnh->rnh_matchaddr)((char *)saddr, 1812 rnh); 1813 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1814 np = NULL; 1815 } 1816 } 1817 /* 1818 * If no address match, use the default if it exists. 1819 */ 1820 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1821 np = &nep->ne_defexported; 1822 } 1823 return (np); 1824 } 1825 1826 /* 1827 * perform msync on all vnodes under a mount point. The mount point must 1828 * be locked. This code is also responsible for lazy-freeing unreferenced 1829 * vnodes whos VM objects no longer contain pages. 1830 * 1831 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1832 * 1833 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 1834 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 1835 * way up in this high level function. 1836 */ 1837 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1838 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1839 1840 void 1841 vfs_msync(struct mount *mp, int flags) 1842 { 1843 int vmsc_flags; 1844 1845 vmsc_flags = VMSC_GETVP; 1846 if (flags != MNT_WAIT) 1847 vmsc_flags |= VMSC_NOWAIT; 1848 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 1849 (void *)flags); 1850 } 1851 1852 /* 1853 * scan1 is a fast pre-check. There could be hundreds of thousands of 1854 * vnodes, we cannot afford to do anything heavy weight until we have a 1855 * fairly good indication that there is work to do. 1856 */ 1857 static 1858 int 1859 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1860 { 1861 int flags = (int)data; 1862 1863 if ((vp->v_flag & VRECLAIMED) == 0) { 1864 if (vshouldmsync(vp)) 1865 return(0); /* call scan2 */ 1866 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1867 (vp->v_flag & VOBJDIRTY) && 1868 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 1869 return(0); /* call scan2 */ 1870 } 1871 } 1872 1873 /* 1874 * do not call scan2, continue the loop 1875 */ 1876 return(-1); 1877 } 1878 1879 /* 1880 * This callback is handed a locked vnode. 1881 */ 1882 static 1883 int 1884 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1885 { 1886 vm_object_t obj; 1887 int flags = (int)data; 1888 1889 if (vp->v_flag & VRECLAIMED) 1890 return(0); 1891 1892 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 1893 if ((obj = vp->v_object) != NULL) { 1894 vm_object_page_clean(obj, 0, 0, 1895 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1896 } 1897 } 1898 return(0); 1899 } 1900 1901 /* 1902 * Record a process's interest in events which might happen to 1903 * a vnode. Because poll uses the historic select-style interface 1904 * internally, this routine serves as both the ``check for any 1905 * pending events'' and the ``record my interest in future events'' 1906 * functions. (These are done together, while the lock is held, 1907 * to avoid race conditions.) 1908 */ 1909 int 1910 vn_pollrecord(struct vnode *vp, int events) 1911 { 1912 lwkt_tokref ilock; 1913 1914 KKASSERT(curthread->td_proc != NULL); 1915 1916 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1917 if (vp->v_pollinfo.vpi_revents & events) { 1918 /* 1919 * This leaves events we are not interested 1920 * in available for the other process which 1921 * which presumably had requested them 1922 * (otherwise they would never have been 1923 * recorded). 1924 */ 1925 events &= vp->v_pollinfo.vpi_revents; 1926 vp->v_pollinfo.vpi_revents &= ~events; 1927 1928 lwkt_reltoken(&ilock); 1929 return events; 1930 } 1931 vp->v_pollinfo.vpi_events |= events; 1932 selrecord(curthread, &vp->v_pollinfo.vpi_selinfo); 1933 lwkt_reltoken(&ilock); 1934 return 0; 1935 } 1936 1937 /* 1938 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1939 * it is possible for us to miss an event due to race conditions, but 1940 * that condition is expected to be rare, so for the moment it is the 1941 * preferred interface. 1942 */ 1943 void 1944 vn_pollevent(struct vnode *vp, int events) 1945 { 1946 lwkt_tokref ilock; 1947 1948 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1949 if (vp->v_pollinfo.vpi_events & events) { 1950 /* 1951 * We clear vpi_events so that we don't 1952 * call selwakeup() twice if two events are 1953 * posted before the polling process(es) is 1954 * awakened. This also ensures that we take at 1955 * most one selwakeup() if the polling process 1956 * is no longer interested. However, it does 1957 * mean that only one event can be noticed at 1958 * a time. (Perhaps we should only clear those 1959 * event bits which we note?) XXX 1960 */ 1961 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1962 vp->v_pollinfo.vpi_revents |= events; 1963 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1964 } 1965 lwkt_reltoken(&ilock); 1966 } 1967 1968 /* 1969 * Wake up anyone polling on vp because it is being revoked. 1970 * This depends on dead_poll() returning POLLHUP for correct 1971 * behavior. 1972 */ 1973 void 1974 vn_pollgone(struct vnode *vp) 1975 { 1976 lwkt_tokref ilock; 1977 1978 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1979 if (vp->v_pollinfo.vpi_events) { 1980 vp->v_pollinfo.vpi_events = 0; 1981 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1982 } 1983 lwkt_reltoken(&ilock); 1984 } 1985 1986 /* 1987 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 1988 * (or v_rdev might be NULL). 1989 */ 1990 cdev_t 1991 vn_todev(struct vnode *vp) 1992 { 1993 if (vp->v_type != VBLK && vp->v_type != VCHR) 1994 return (NULL); 1995 KKASSERT(vp->v_rdev != NULL); 1996 return (vp->v_rdev); 1997 } 1998 1999 /* 2000 * Check if vnode represents a disk device. The vnode does not need to be 2001 * opened. 2002 */ 2003 int 2004 vn_isdisk(struct vnode *vp, int *errp) 2005 { 2006 cdev_t dev; 2007 2008 if (vp->v_type != VCHR) { 2009 if (errp != NULL) 2010 *errp = ENOTBLK; 2011 return (0); 2012 } 2013 2014 if ((dev = vp->v_rdev) == NULL) 2015 dev = get_dev(vp->v_umajor, vp->v_uminor); 2016 2017 if (dev == NULL) { 2018 if (errp != NULL) 2019 *errp = ENXIO; 2020 return (0); 2021 } 2022 if (dev_is_good(dev) == 0) { 2023 if (errp != NULL) 2024 *errp = ENXIO; 2025 return (0); 2026 } 2027 if ((dev_dflags(dev) & D_DISK) == 0) { 2028 if (errp != NULL) 2029 *errp = ENOTBLK; 2030 return (0); 2031 } 2032 if (errp != NULL) 2033 *errp = 0; 2034 return (1); 2035 } 2036 2037 int 2038 vn_get_namelen(struct vnode *vp, int *namelen) 2039 { 2040 int error, retval[2]; 2041 2042 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2043 if (error) 2044 return (error); 2045 *namelen = *retval; 2046 return (0); 2047 } 2048 2049 int 2050 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2051 uint16_t d_namlen, const char *d_name) 2052 { 2053 struct dirent *dp; 2054 size_t len; 2055 2056 len = _DIRENT_RECLEN(d_namlen); 2057 if (len > uio->uio_resid) 2058 return(1); 2059 2060 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2061 2062 dp->d_ino = d_ino; 2063 dp->d_namlen = d_namlen; 2064 dp->d_type = d_type; 2065 bcopy(d_name, dp->d_name, d_namlen); 2066 2067 *error = uiomove((caddr_t)dp, len, uio); 2068 2069 kfree(dp, M_TEMP); 2070 2071 return(0); 2072 } 2073 2074