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 1.63 2005/08/27 20:23:05 joerg Exp $ 41 */ 42 43 /* 44 * External virtual filesystem routines 45 */ 46 #include "opt_ddb.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/domain.h> 54 #include <sys/eventhandler.h> 55 #include <sys/fcntl.h> 56 #include <sys/kernel.h> 57 #include <sys/kthread.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/mount.h> 61 #include <sys/proc.h> 62 #include <sys/reboot.h> 63 #include <sys/socket.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/syslog.h> 67 #include <sys/unistd.h> 68 #include <sys/vmmeter.h> 69 #include <sys/vnode.h> 70 71 #include <machine/limits.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_object.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_kern.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_page.h> 80 #include <vm/vm_pager.h> 81 #include <vm/vnode_pager.h> 82 #include <vm/vm_zone.h> 83 84 #include <sys/buf2.h> 85 #include <sys/thread2.h> 86 87 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 88 89 int numvnodes; 90 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 91 int vfs_fastdev = 1; 92 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, ""); 93 94 enum vtype iftovt_tab[16] = { 95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 97 }; 98 int vttoif_tab[9] = { 99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 100 S_IFSOCK, S_IFIFO, S_IFMT, 101 }; 102 103 static int reassignbufcalls; 104 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, 105 &reassignbufcalls, 0, ""); 106 static int reassignbufloops; 107 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, 108 &reassignbufloops, 0, ""); 109 static int reassignbufsortgood; 110 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, 111 &reassignbufsortgood, 0, ""); 112 static int reassignbufsortbad; 113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, 114 &reassignbufsortbad, 0, ""); 115 static int reassignbufmethod = 1; 116 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, 117 &reassignbufmethod, 0, ""); 118 119 int nfs_mount_type = -1; 120 static struct lwkt_token spechash_token; 121 struct nfs_public nfs_pub; /* publicly exported FS */ 122 123 int desiredvnodes; 124 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 125 &desiredvnodes, 0, "Maximum number of vnodes"); 126 127 static void vfs_free_addrlist (struct netexport *nep); 128 static int vfs_free_netcred (struct radix_node *rn, void *w); 129 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 130 struct export_args *argp); 131 132 extern int dev_ref_debug; 133 extern struct vnodeopv_entry_desc spec_vnodeop_entries[]; 134 135 /* 136 * Red black tree functions 137 */ 138 static int rb_buf_compare(struct buf *b1, struct buf *b2); 139 RB_GENERATE(buf_rb_tree, buf, b_rbnode, rb_buf_compare); 140 141 static int 142 rb_buf_compare(struct buf *b1, struct buf *b2) 143 { 144 if (b1->b_lblkno < b2->b_lblkno) 145 return(-1); 146 if (b1->b_lblkno > b2->b_lblkno) 147 return(1); 148 return(0); 149 } 150 151 /* 152 * Return 0 if the vnode is already on the free list or cannot be placed 153 * on the free list. Return 1 if the vnode can be placed on the free list. 154 */ 155 static __inline int 156 vshouldfree(struct vnode *vp, int usecount) 157 { 158 if (vp->v_flag & VFREE) 159 return (0); /* already free */ 160 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount) 161 return (0); /* other holderse */ 162 if (vp->v_object && 163 (vp->v_object->ref_count || vp->v_object->resident_page_count)) { 164 return (0); 165 } 166 return (1); 167 } 168 169 /* 170 * Initialize the vnode management data structures. 171 * 172 * Called from vfsinit() 173 */ 174 void 175 vfs_subr_init(void) 176 { 177 /* 178 * Desired vnodes is a result of the physical page count 179 * and the size of kernel's heap. It scales in proportion 180 * to the amount of available physical memory. This can 181 * cause trouble on 64-bit and large memory platforms. 182 */ 183 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */ 184 desiredvnodes = 185 min(maxproc + vmstats.v_page_count /4, 186 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 187 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 188 189 lwkt_token_init(&spechash_token); 190 } 191 192 /* 193 * Knob to control the precision of file timestamps: 194 * 195 * 0 = seconds only; nanoseconds zeroed. 196 * 1 = seconds and nanoseconds, accurate within 1/HZ. 197 * 2 = seconds and nanoseconds, truncated to microseconds. 198 * >=3 = seconds and nanoseconds, maximum precision. 199 */ 200 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 201 202 static int timestamp_precision = TSP_SEC; 203 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 204 ×tamp_precision, 0, ""); 205 206 /* 207 * Get a current timestamp. 208 */ 209 void 210 vfs_timestamp(struct timespec *tsp) 211 { 212 struct timeval tv; 213 214 switch (timestamp_precision) { 215 case TSP_SEC: 216 tsp->tv_sec = time_second; 217 tsp->tv_nsec = 0; 218 break; 219 case TSP_HZ: 220 getnanotime(tsp); 221 break; 222 case TSP_USEC: 223 microtime(&tv); 224 TIMEVAL_TO_TIMESPEC(&tv, tsp); 225 break; 226 case TSP_NSEC: 227 default: 228 nanotime(tsp); 229 break; 230 } 231 } 232 233 /* 234 * Set vnode attributes to VNOVAL 235 */ 236 void 237 vattr_null(struct vattr *vap) 238 { 239 vap->va_type = VNON; 240 vap->va_size = VNOVAL; 241 vap->va_bytes = VNOVAL; 242 vap->va_mode = VNOVAL; 243 vap->va_nlink = VNOVAL; 244 vap->va_uid = VNOVAL; 245 vap->va_gid = VNOVAL; 246 vap->va_fsid = VNOVAL; 247 vap->va_fileid = VNOVAL; 248 vap->va_blocksize = VNOVAL; 249 vap->va_rdev = VNOVAL; 250 vap->va_atime.tv_sec = VNOVAL; 251 vap->va_atime.tv_nsec = VNOVAL; 252 vap->va_mtime.tv_sec = VNOVAL; 253 vap->va_mtime.tv_nsec = VNOVAL; 254 vap->va_ctime.tv_sec = VNOVAL; 255 vap->va_ctime.tv_nsec = VNOVAL; 256 vap->va_flags = VNOVAL; 257 vap->va_gen = VNOVAL; 258 vap->va_vaflags = 0; 259 } 260 261 /* 262 * Update outstanding I/O count and do wakeup if requested. 263 */ 264 void 265 vwakeup(struct buf *bp) 266 { 267 struct vnode *vp; 268 269 if ((vp = bp->b_vp)) { 270 vp->v_numoutput--; 271 if (vp->v_numoutput < 0) 272 panic("vwakeup: neg numoutput"); 273 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 274 vp->v_flag &= ~VBWAIT; 275 wakeup((caddr_t) &vp->v_numoutput); 276 } 277 } 278 } 279 280 /* 281 * Flush out and invalidate all buffers associated with a vnode. 282 * 283 * vp must be locked. 284 */ 285 static int vinvalbuf_bp(struct buf *bp, void *data); 286 287 struct vinvalbuf_bp_info { 288 struct vnode *vp; 289 int slptimeo; 290 int slpflag; 291 int flags; 292 }; 293 294 int 295 vinvalbuf(struct vnode *vp, int flags, struct thread *td, 296 int slpflag, int slptimeo) 297 { 298 struct vinvalbuf_bp_info info; 299 int error; 300 vm_object_t object; 301 302 /* 303 * If we are being asked to save, call fsync to ensure that the inode 304 * is updated. 305 */ 306 if (flags & V_SAVE) { 307 crit_enter(); 308 while (vp->v_numoutput) { 309 vp->v_flag |= VBWAIT; 310 error = tsleep((caddr_t)&vp->v_numoutput, 311 slpflag, "vinvlbuf", slptimeo); 312 if (error) { 313 crit_exit(); 314 return (error); 315 } 316 } 317 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 318 crit_exit(); 319 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) 320 return (error); 321 crit_enter(); 322 if (vp->v_numoutput > 0 || 323 !RB_EMPTY(&vp->v_rbdirty_tree)) 324 panic("vinvalbuf: dirty bufs"); 325 } 326 crit_exit(); 327 } 328 crit_enter(); 329 info.slptimeo = slptimeo; 330 info.slpflag = slpflag; 331 info.flags = flags; 332 info.vp = vp; 333 334 /* 335 * Flush the buffer cache until nothing is left. 336 */ 337 while (!RB_EMPTY(&vp->v_rbclean_tree) || 338 !RB_EMPTY(&vp->v_rbdirty_tree)) { 339 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 340 vinvalbuf_bp, &info); 341 if (error == 0) { 342 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 343 vinvalbuf_bp, &info); 344 } 345 } 346 347 /* 348 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 349 * have write I/O in-progress but if there is a VM object then the 350 * VM object can also have read-I/O in-progress. 351 */ 352 do { 353 while (vp->v_numoutput > 0) { 354 vp->v_flag |= VBWAIT; 355 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0); 356 } 357 if (VOP_GETVOBJECT(vp, &object) == 0) { 358 while (object->paging_in_progress) 359 vm_object_pip_sleep(object, "vnvlbx"); 360 } 361 } while (vp->v_numoutput > 0); 362 363 crit_exit(); 364 365 /* 366 * Destroy the copy in the VM cache, too. 367 */ 368 if (VOP_GETVOBJECT(vp, &object) == 0) { 369 vm_object_page_remove(object, 0, 0, 370 (flags & V_SAVE) ? TRUE : FALSE); 371 } 372 373 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 374 panic("vinvalbuf: flush failed"); 375 return (0); 376 } 377 378 static int 379 vinvalbuf_bp(struct buf *bp, void *data) 380 { 381 struct vinvalbuf_bp_info *info = data; 382 int error; 383 384 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 385 error = BUF_TIMELOCK(bp, 386 LK_EXCLUSIVE | LK_SLEEPFAIL, 387 "vinvalbuf", info->slpflag, info->slptimeo); 388 if (error == 0) { 389 BUF_UNLOCK(bp); 390 error = ENOLCK; 391 } 392 if (error == ENOLCK) 393 return(0); 394 return (-error); 395 } 396 /* 397 * XXX Since there are no node locks for NFS, I 398 * believe there is a slight chance that a delayed 399 * write will occur while sleeping just above, so 400 * check for it. Note that vfs_bio_awrite expects 401 * buffers to reside on a queue, while VOP_BWRITE and 402 * brelse do not. 403 */ 404 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 405 (info->flags & V_SAVE)) { 406 if (bp->b_vp == info->vp) { 407 if (bp->b_flags & B_CLUSTEROK) { 408 BUF_UNLOCK(bp); 409 vfs_bio_awrite(bp); 410 } else { 411 bremfree(bp); 412 bp->b_flags |= B_ASYNC; 413 VOP_BWRITE(bp->b_vp, bp); 414 } 415 } else { 416 bremfree(bp); 417 VOP_BWRITE(bp->b_vp, bp); 418 } 419 } else { 420 bremfree(bp); 421 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 422 bp->b_flags &= ~B_ASYNC; 423 brelse(bp); 424 } 425 return(0); 426 } 427 428 /* 429 * Truncate a file's buffer and pages to a specified length. This 430 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 431 * sync activity. 432 * 433 * The vnode must be locked. 434 */ 435 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 436 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 437 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 438 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 439 440 int 441 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize) 442 { 443 daddr_t trunclbn; 444 int count; 445 446 /* 447 * Round up to the *next* lbn, then destroy the buffers in question. 448 * Since we are only removing some of the buffers we must rely on the 449 * scan count to determine whether a loop is necessary. 450 */ 451 trunclbn = (length + blksize - 1) / blksize; 452 453 crit_enter(); 454 do { 455 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 456 vtruncbuf_bp_trunc_cmp, 457 vtruncbuf_bp_trunc, &trunclbn); 458 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 459 vtruncbuf_bp_trunc_cmp, 460 vtruncbuf_bp_trunc, &trunclbn); 461 } while(count); 462 463 /* 464 * For safety, fsync any remaining metadata if the file is not being 465 * truncated to 0. Since the metadata does not represent the entire 466 * dirty list we have to rely on the hit count to ensure that we get 467 * all of it. 468 */ 469 if (length > 0) { 470 do { 471 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 472 vtruncbuf_bp_metasync_cmp, 473 vtruncbuf_bp_metasync, vp); 474 } while (count); 475 } 476 477 /* 478 * Wait for any in-progress I/O to complete before returning (why?) 479 */ 480 while (vp->v_numoutput > 0) { 481 vp->v_flag |= VBWAIT; 482 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0); 483 } 484 485 crit_exit(); 486 487 vnode_pager_setsize(vp, length); 488 489 return (0); 490 } 491 492 /* 493 * The callback buffer is beyond the new file EOF and must be destroyed. 494 * Note that the compare function must conform to the RB_SCAN's requirements. 495 */ 496 static 497 int 498 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 499 { 500 if (bp->b_lblkno >= *(daddr_t *)data) 501 return(0); 502 return(-1); 503 } 504 505 static 506 int 507 vtruncbuf_bp_trunc(struct buf *bp, void *data) 508 { 509 /* 510 * Do not try to use a buffer we cannot immediately lock, but sleep 511 * anyway to prevent a livelock. The code will loop until all buffers 512 * can be acted upon. 513 */ 514 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 515 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 516 BUF_UNLOCK(bp); 517 } else { 518 bremfree(bp); 519 bp->b_flags |= (B_INVAL | B_RELBUF); 520 bp->b_flags &= ~B_ASYNC; 521 brelse(bp); 522 } 523 return(1); 524 } 525 526 /* 527 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 528 * blocks (with a negative lblkno) are scanned. 529 * Note that the compare function must conform to the RB_SCAN's requirements. 530 */ 531 static int 532 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 533 { 534 if (bp->b_lblkno < 0) 535 return(0); 536 return(1); 537 } 538 539 static int 540 vtruncbuf_bp_metasync(struct buf *bp, void *data) 541 { 542 struct vnode *vp = data; 543 544 if (bp->b_flags & B_DELWRI) { 545 /* 546 * Do not try to use a buffer we cannot immediately lock, 547 * but sleep anyway to prevent a livelock. The code will 548 * loop until all buffers can be acted upon. 549 */ 550 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 551 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 552 BUF_UNLOCK(bp); 553 } else { 554 bremfree(bp); 555 if (bp->b_vp == vp) { 556 bp->b_flags |= B_ASYNC; 557 } else { 558 bp->b_flags &= ~B_ASYNC; 559 } 560 VOP_BWRITE(bp->b_vp, bp); 561 } 562 return(1); 563 } else { 564 return(0); 565 } 566 } 567 568 /* 569 * vfsync - implements a multipass fsync on a file which understands 570 * dependancies and meta-data. The passed vnode must be locked. The 571 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 572 * 573 * When fsyncing data asynchronously just do one consolidated pass starting 574 * with the most negative block number. This may not get all the data due 575 * to dependancies. 576 * 577 * When fsyncing data synchronously do a data pass, then a metadata pass, 578 * then do additional data+metadata passes to try to get all the data out. 579 */ 580 static int vfsync_wait_output(struct vnode *vp, 581 int (*waitoutput)(struct vnode *, struct thread *)); 582 static int vfsync_data_only_cmp(struct buf *bp, void *data); 583 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 584 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 585 static int vfsync_bp(struct buf *bp, void *data); 586 587 struct vfsync_info { 588 struct vnode *vp; 589 int synchronous; 590 int syncdeps; 591 int lazycount; 592 int lazylimit; 593 daddr_t lbn; 594 int (*checkdef)(struct buf *); 595 }; 596 597 int 598 vfsync(struct vnode *vp, int waitfor, int passes, daddr_t lbn, 599 int (*checkdef)(struct buf *), 600 int (*waitoutput)(struct vnode *, struct thread *)) 601 { 602 struct vfsync_info info; 603 int error; 604 605 bzero(&info, sizeof(info)); 606 info.vp = vp; 607 info.lbn = lbn; 608 if ((info.checkdef = checkdef) == NULL) 609 info.syncdeps = 1; 610 611 crit_enter(); 612 613 switch(waitfor) { 614 case MNT_LAZY: 615 /* 616 * Lazy (filesystem syncer typ) Asynchronous plus limit the 617 * number of data (not meta) pages we try to flush to 1MB. 618 * A non-zero return means that lazy limit was reached. 619 */ 620 info.lazylimit = 1024 * 1024; 621 info.syncdeps = 1; 622 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 623 vfsync_lazy_range_cmp, vfsync_bp, &info); 624 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 625 vfsync_meta_only_cmp, vfsync_bp, &info); 626 if (error == 0) 627 vp->v_lazyw = 0; 628 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 629 vn_syncer_add_to_worklist(vp, 1); 630 error = 0; 631 break; 632 case MNT_NOWAIT: 633 /* 634 * Asynchronous. Do a data-only pass and a meta-only pass. 635 */ 636 info.syncdeps = 1; 637 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 638 vfsync_bp, &info); 639 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 640 vfsync_bp, &info); 641 error = 0; 642 break; 643 default: 644 /* 645 * Synchronous. Do a data-only pass, then a meta-data+data 646 * pass, then additional integrated passes to try to get 647 * all the dependancies flushed. 648 */ 649 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 650 vfsync_bp, &info); 651 error = vfsync_wait_output(vp, waitoutput); 652 if (error == 0) { 653 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 654 vfsync_bp, &info); 655 error = vfsync_wait_output(vp, waitoutput); 656 } 657 while (error == 0 && passes > 0 && 658 !RB_EMPTY(&vp->v_rbdirty_tree)) { 659 if (--passes == 0) { 660 info.synchronous = 1; 661 info.syncdeps = 1; 662 } 663 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 664 vfsync_bp, &info); 665 if (error < 0) 666 error = -error; 667 info.syncdeps = 1; 668 if (error == 0) 669 error = vfsync_wait_output(vp, waitoutput); 670 } 671 break; 672 } 673 crit_exit(); 674 return(error); 675 } 676 677 static int 678 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 679 { 680 int error = 0; 681 682 while (vp->v_numoutput) { 683 vp->v_flag |= VBWAIT; 684 tsleep(&vp->v_numoutput, 0, "fsfsn", 0); 685 } 686 if (waitoutput) 687 error = waitoutput(vp, curthread); 688 return(error); 689 } 690 691 static int 692 vfsync_data_only_cmp(struct buf *bp, void *data) 693 { 694 if (bp->b_lblkno < 0) 695 return(-1); 696 return(0); 697 } 698 699 static int 700 vfsync_meta_only_cmp(struct buf *bp, void *data) 701 { 702 if (bp->b_lblkno < 0) 703 return(0); 704 return(1); 705 } 706 707 static int 708 vfsync_lazy_range_cmp(struct buf *bp, void *data) 709 { 710 struct vfsync_info *info = data; 711 if (bp->b_lblkno < info->vp->v_lazyw) 712 return(-1); 713 return(0); 714 } 715 716 static int 717 vfsync_bp(struct buf *bp, void *data) 718 { 719 struct vfsync_info *info = data; 720 struct vnode *vp = info->vp; 721 int error; 722 723 /* 724 * if syncdeps is not set we do not try to write buffers which have 725 * dependancies. 726 */ 727 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 728 return(0); 729 730 /* 731 * Ignore buffers that we cannot immediately lock. XXX 732 */ 733 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 734 return(0); 735 if ((bp->b_flags & B_DELWRI) == 0) 736 panic("vfsync_bp: buffer not dirty"); 737 if (vp != bp->b_vp) 738 panic("vfsync_bp: buffer vp mismatch"); 739 740 /* 741 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 742 * has been written but an additional handshake with the device 743 * is required before we can dispose of the buffer. We have no idea 744 * how to do this so we have to skip these buffers. 745 */ 746 if (bp->b_flags & B_NEEDCOMMIT) { 747 BUF_UNLOCK(bp); 748 return(0); 749 } 750 751 /* 752 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty 753 * buffers beyond the file EOF. 754 */ 755 if (info->lbn != (daddr_t)-1 && vp->v_type == VREG && 756 bp->b_lblkno >= info->lbn) { 757 bremfree(bp); 758 bp->b_flags |= B_INVAL | B_NOCACHE; 759 crit_exit(); 760 brelse(bp); 761 crit_enter(); 762 } 763 764 if (info->synchronous) { 765 /* 766 * Synchronous flushing. An error may be returned. 767 */ 768 bremfree(bp); 769 crit_exit(); 770 error = bwrite(bp); 771 crit_enter(); 772 } else { 773 /* 774 * Asynchronous flushing. A negative return value simply 775 * stops the scan and is not considered an error. We use 776 * this to support limited MNT_LAZY flushes. 777 */ 778 vp->v_lazyw = bp->b_lblkno; 779 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 780 BUF_UNLOCK(bp); 781 info->lazycount += vfs_bio_awrite(bp); 782 } else { 783 info->lazycount += bp->b_bufsize; 784 bremfree(bp); 785 crit_exit(); 786 bawrite(bp); 787 crit_enter(); 788 } 789 if (info->lazylimit && info->lazycount >= info->lazylimit) 790 error = 1; 791 else 792 error = 0; 793 } 794 return(-error); 795 } 796 797 /* 798 * Associate a buffer with a vnode. 799 */ 800 void 801 bgetvp(struct vnode *vp, struct buf *bp) 802 { 803 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 804 805 vhold(vp); 806 bp->b_vp = vp; 807 bp->b_dev = vn_todev(vp); 808 /* 809 * Insert onto list for new vnode. 810 */ 811 crit_enter(); 812 bp->b_xflags |= BX_VNCLEAN; 813 bp->b_xflags &= ~BX_VNDIRTY; 814 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 815 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 816 crit_exit(); 817 } 818 819 /* 820 * Disassociate a buffer from a vnode. 821 */ 822 void 823 brelvp(struct buf *bp) 824 { 825 struct vnode *vp; 826 827 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 828 829 /* 830 * Delete from old vnode list, if on one. 831 */ 832 vp = bp->b_vp; 833 crit_enter(); 834 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 835 if (bp->b_xflags & BX_VNDIRTY) 836 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 837 else 838 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 839 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 840 } 841 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 842 vp->v_flag &= ~VONWORKLST; 843 LIST_REMOVE(vp, v_synclist); 844 } 845 crit_exit(); 846 bp->b_vp = NULL; 847 vdrop(vp); 848 } 849 850 /* 851 * Associate a p-buffer with a vnode. 852 * 853 * Also sets B_PAGING flag to indicate that vnode is not fully associated 854 * with the buffer. i.e. the bp has not been linked into the vnode or 855 * ref-counted. 856 */ 857 void 858 pbgetvp(struct vnode *vp, struct buf *bp) 859 { 860 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 861 862 bp->b_vp = vp; 863 bp->b_flags |= B_PAGING; 864 bp->b_dev = vn_todev(vp); 865 } 866 867 /* 868 * Disassociate a p-buffer from a vnode. 869 */ 870 void 871 pbrelvp(struct buf *bp) 872 { 873 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 874 875 bp->b_vp = NULL; 876 bp->b_flags &= ~B_PAGING; 877 } 878 879 void 880 pbreassignbuf(struct buf *bp, struct vnode *newvp) 881 { 882 if ((bp->b_flags & B_PAGING) == 0) { 883 panic( 884 "pbreassignbuf() on non phys bp %p", 885 bp 886 ); 887 } 888 bp->b_vp = newvp; 889 } 890 891 /* 892 * Reassign a buffer from one vnode to another. 893 * Used to assign file specific control information 894 * (indirect blocks) to the vnode to which they belong. 895 */ 896 void 897 reassignbuf(struct buf *bp, struct vnode *newvp) 898 { 899 int delay; 900 901 if (newvp == NULL) { 902 printf("reassignbuf: NULL"); 903 return; 904 } 905 ++reassignbufcalls; 906 907 /* 908 * B_PAGING flagged buffers cannot be reassigned because their vp 909 * is not fully linked in. 910 */ 911 if (bp->b_flags & B_PAGING) 912 panic("cannot reassign paging buffer"); 913 914 crit_enter(); 915 /* 916 * Delete from old vnode list, if on one. 917 */ 918 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 919 if (bp->b_xflags & BX_VNDIRTY) 920 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbdirty_tree, bp); 921 else 922 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbclean_tree, bp); 923 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 924 if (bp->b_vp != newvp) { 925 vdrop(bp->b_vp); 926 bp->b_vp = NULL; /* for clarification */ 927 } 928 } 929 /* 930 * If dirty, put on list of dirty buffers; otherwise insert onto list 931 * of clean buffers. 932 */ 933 if (bp->b_flags & B_DELWRI) { 934 if ((newvp->v_flag & VONWORKLST) == 0) { 935 switch (newvp->v_type) { 936 case VDIR: 937 delay = dirdelay; 938 break; 939 case VCHR: 940 case VBLK: 941 if (newvp->v_rdev && 942 newvp->v_rdev->si_mountpoint != NULL) { 943 delay = metadelay; 944 break; 945 } 946 /* fall through */ 947 default: 948 delay = filedelay; 949 } 950 vn_syncer_add_to_worklist(newvp, delay); 951 } 952 bp->b_xflags |= BX_VNDIRTY; 953 if (buf_rb_tree_RB_INSERT(&newvp->v_rbdirty_tree, bp)) 954 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp); 955 } else { 956 bp->b_xflags |= BX_VNCLEAN; 957 if (buf_rb_tree_RB_INSERT(&newvp->v_rbclean_tree, bp)) 958 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp); 959 if ((newvp->v_flag & VONWORKLST) && 960 RB_EMPTY(&newvp->v_rbdirty_tree)) { 961 newvp->v_flag &= ~VONWORKLST; 962 LIST_REMOVE(newvp, v_synclist); 963 } 964 } 965 if (bp->b_vp != newvp) { 966 bp->b_vp = newvp; 967 vhold(bp->b_vp); 968 } 969 crit_exit(); 970 } 971 972 /* 973 * Create a vnode for a block device. 974 * Used for mounting the root file system. 975 */ 976 int 977 bdevvp(dev_t dev, struct vnode **vpp) 978 { 979 struct vnode *vp; 980 struct vnode *nvp; 981 int error; 982 983 if (dev == NODEV) { 984 *vpp = NULLVP; 985 return (ENXIO); 986 } 987 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0); 988 if (error) { 989 *vpp = NULLVP; 990 return (error); 991 } 992 vp = nvp; 993 vp->v_type = VCHR; 994 vp->v_udev = dev->si_udev; 995 vx_unlock(vp); 996 *vpp = vp; 997 return (0); 998 } 999 1000 int 1001 v_associate_rdev(struct vnode *vp, dev_t dev) 1002 { 1003 lwkt_tokref ilock; 1004 1005 if (dev == NULL || dev == NODEV) 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 printf("Z1"); 1012 vp->v_rdev = reference_dev(dev); 1013 lwkt_gettoken(&ilock, &spechash_token); 1014 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext); 1015 lwkt_reltoken(&ilock); 1016 return(0); 1017 } 1018 1019 void 1020 v_release_rdev(struct vnode *vp) 1021 { 1022 lwkt_tokref ilock; 1023 dev_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_specnext); 1028 if (dev_ref_debug && vp->v_opencount != 0) { 1029 printf("releasing rdev with non-0 " 1030 "v_opencount(%d) (revoked?)\n", 1031 vp->v_opencount); 1032 } 1033 vp->v_rdev = NULL; 1034 vp->v_opencount = 0; 1035 release_dev(dev); 1036 lwkt_reltoken(&ilock); 1037 } 1038 } 1039 1040 /* 1041 * Add a vnode to the alias list hung off the dev_t. We only associate 1042 * the device number with the vnode. The actual device is not associated 1043 * until the vnode is opened (usually in spec_open()), and will be 1044 * disassociated on last close. 1045 */ 1046 void 1047 addaliasu(struct vnode *nvp, udev_t nvp_udev) 1048 { 1049 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1050 panic("addaliasu on non-special vnode"); 1051 nvp->v_udev = nvp_udev; 1052 } 1053 1054 /* 1055 * Disassociate a vnode from its underlying filesystem. 1056 * 1057 * The vnode must be VX locked and refd 1058 * 1059 * If there are v_usecount references to the vnode other then ours we have 1060 * to VOP_CLOSE the vnode before we can deactivate and reclaim it. 1061 */ 1062 void 1063 vclean(struct vnode *vp, int flags, struct thread *td) 1064 { 1065 int active; 1066 1067 /* 1068 * If the vnode has already been reclaimed we have nothing to do. 1069 */ 1070 if (vp->v_flag & VRECLAIMED) 1071 return; 1072 vp->v_flag |= VRECLAIMED; 1073 1074 /* 1075 * Scrap the vfs cache 1076 */ 1077 while (cache_inval_vp(vp, 0) != 0) { 1078 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1079 tsleep(vp, 0, "vclninv", 2); 1080 } 1081 1082 /* 1083 * Check to see if the vnode is in use. If so we have to reference it 1084 * before we clean it out so that its count cannot fall to zero and 1085 * generate a race against ourselves to recycle it. 1086 */ 1087 active = (vp->v_usecount > 1); 1088 1089 /* 1090 * Clean out any buffers associated with the vnode and destroy its 1091 * object, if it has one. 1092 */ 1093 vinvalbuf(vp, V_SAVE, td, 0, 0); 1094 VOP_DESTROYVOBJECT(vp); 1095 1096 /* 1097 * If purging an active vnode, it must be closed and 1098 * deactivated before being reclaimed. XXX 1099 * 1100 * Note that neither of these routines unlocks the vnode. 1101 */ 1102 if (active) { 1103 if (flags & DOCLOSE) 1104 VOP_CLOSE(vp, FNONBLOCK, td); 1105 } 1106 1107 /* 1108 * If the vnode has not be deactivated, deactivated it. 1109 */ 1110 if ((vp->v_flag & VINACTIVE) == 0) { 1111 vp->v_flag |= VINACTIVE; 1112 VOP_INACTIVE(vp, td); 1113 } 1114 1115 /* 1116 * Reclaim the vnode. 1117 */ 1118 if (VOP_RECLAIM(vp, td)) 1119 panic("vclean: cannot reclaim"); 1120 1121 /* 1122 * Done with purge, notify sleepers of the grim news. 1123 */ 1124 vp->v_ops = &dead_vnode_vops; 1125 vn_pollgone(vp); 1126 vp->v_tag = VT_NON; 1127 } 1128 1129 /* 1130 * Eliminate all activity associated with the requested vnode 1131 * and with all vnodes aliased to the requested vnode. 1132 * 1133 * The vnode must be referenced and vx_lock()'d 1134 * 1135 * revoke { struct vnode *a_vp, int a_flags } 1136 */ 1137 int 1138 vop_stdrevoke(struct vop_revoke_args *ap) 1139 { 1140 struct vnode *vp, *vq; 1141 lwkt_tokref ilock; 1142 dev_t dev; 1143 1144 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1145 1146 vp = ap->a_vp; 1147 1148 /* 1149 * If the vnode is already dead don't try to revoke it 1150 */ 1151 if (vp->v_flag & VRECLAIMED) 1152 return (0); 1153 1154 /* 1155 * If the vnode has a device association, scrap all vnodes associated 1156 * with the device. Don't let the device disappear on us while we 1157 * are scrapping the vnodes. 1158 * 1159 * The passed vp will probably show up in the list, do not VX lock 1160 * it twice! 1161 */ 1162 if (vp->v_type != VCHR && vp->v_type != VBLK) 1163 return(0); 1164 if ((dev = vp->v_rdev) == NULL) { 1165 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV) 1166 return(0); 1167 } 1168 reference_dev(dev); 1169 lwkt_gettoken(&ilock, &spechash_token); 1170 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1171 if (vp == vq || vx_get(vq) == 0) { 1172 if (vq == SLIST_FIRST(&dev->si_hlist)) 1173 vgone(vq); 1174 if (vp != vq) 1175 vx_put(vq); 1176 } 1177 } 1178 lwkt_reltoken(&ilock); 1179 release_dev(dev); 1180 return (0); 1181 } 1182 1183 /* 1184 * Recycle an unused vnode to the front of the free list. 1185 * 1186 * Returns 1 if we were successfully able to recycle the vnode, 1187 * 0 otherwise. 1188 */ 1189 int 1190 vrecycle(struct vnode *vp, struct thread *td) 1191 { 1192 if (vp->v_usecount == 1) { 1193 vgone(vp); 1194 return (1); 1195 } 1196 return (0); 1197 } 1198 1199 /* 1200 * Eliminate all activity associated with a vnode in preparation for reuse. 1201 * 1202 * The vnode must be VX locked and refd and will remain VX locked and refd 1203 * on return. This routine may be called with the vnode in any state, as 1204 * long as it is VX locked. The vnode will be cleaned out and marked 1205 * VRECLAIMED but will not actually be reused until all existing refs and 1206 * holds go away. 1207 * 1208 * NOTE: This routine may be called on a vnode which has not yet been 1209 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1210 * already been reclaimed. 1211 * 1212 * This routine is not responsible for placing us back on the freelist. 1213 * Instead, it happens automatically when the caller releases the VX lock 1214 * (assuming there aren't any other references). 1215 */ 1216 void 1217 vgone(struct vnode *vp) 1218 { 1219 /* 1220 * assert that the VX lock is held. This is an absolute requirement 1221 * now for vgone() to be called. 1222 */ 1223 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1224 1225 /* 1226 * Clean out the filesystem specific data and set the VRECLAIMED 1227 * bit. Also deactivate the vnode if necessary. 1228 */ 1229 vclean(vp, DOCLOSE, curthread); 1230 1231 /* 1232 * Delete from old mount point vnode list, if on one. 1233 */ 1234 if (vp->v_mount != NULL) 1235 insmntque(vp, NULL); 1236 1237 /* 1238 * If special device, remove it from special device alias list 1239 * if it is on one. This should normally only occur if a vnode is 1240 * being revoked as the device should otherwise have been released 1241 * naturally. 1242 */ 1243 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1244 v_release_rdev(vp); 1245 } 1246 1247 /* 1248 * Set us to VBAD 1249 */ 1250 vp->v_type = VBAD; 1251 } 1252 1253 /* 1254 * Lookup a vnode by device number. 1255 */ 1256 int 1257 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp) 1258 { 1259 lwkt_tokref ilock; 1260 struct vnode *vp; 1261 1262 lwkt_gettoken(&ilock, &spechash_token); 1263 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1264 if (type == vp->v_type) { 1265 *vpp = vp; 1266 lwkt_reltoken(&ilock); 1267 return (1); 1268 } 1269 } 1270 lwkt_reltoken(&ilock); 1271 return (0); 1272 } 1273 1274 /* 1275 * Calculate the total number of references to a special device. This 1276 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1277 * an overloaded field. Since udev2dev can now return NODEV, we have 1278 * to check for a NULL v_rdev. 1279 */ 1280 int 1281 count_dev(dev_t dev) 1282 { 1283 lwkt_tokref ilock; 1284 struct vnode *vp; 1285 int count = 0; 1286 1287 if (SLIST_FIRST(&dev->si_hlist)) { 1288 lwkt_gettoken(&ilock, &spechash_token); 1289 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1290 count += vp->v_usecount; 1291 } 1292 lwkt_reltoken(&ilock); 1293 } 1294 return(count); 1295 } 1296 1297 int 1298 count_udev(udev_t udev) 1299 { 1300 dev_t dev; 1301 1302 if ((dev = udev2dev(udev, 0)) == NODEV) 1303 return(0); 1304 return(count_dev(dev)); 1305 } 1306 1307 int 1308 vcount(struct vnode *vp) 1309 { 1310 if (vp->v_rdev == NULL) 1311 return(0); 1312 return(count_dev(vp->v_rdev)); 1313 } 1314 1315 /* 1316 * Print out a description of a vnode. 1317 */ 1318 static char *typename[] = 1319 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1320 1321 void 1322 vprint(char *label, struct vnode *vp) 1323 { 1324 char buf[96]; 1325 1326 if (label != NULL) 1327 printf("%s: %p: ", label, (void *)vp); 1328 else 1329 printf("%p: ", (void *)vp); 1330 printf("type %s, usecount %d, writecount %d, refcount %d,", 1331 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1332 vp->v_holdcnt); 1333 buf[0] = '\0'; 1334 if (vp->v_flag & VROOT) 1335 strcat(buf, "|VROOT"); 1336 if (vp->v_flag & VTEXT) 1337 strcat(buf, "|VTEXT"); 1338 if (vp->v_flag & VSYSTEM) 1339 strcat(buf, "|VSYSTEM"); 1340 if (vp->v_flag & VBWAIT) 1341 strcat(buf, "|VBWAIT"); 1342 if (vp->v_flag & VFREE) 1343 strcat(buf, "|VFREE"); 1344 if (vp->v_flag & VOBJBUF) 1345 strcat(buf, "|VOBJBUF"); 1346 if (buf[0] != '\0') 1347 printf(" flags (%s)", &buf[1]); 1348 if (vp->v_data == NULL) { 1349 printf("\n"); 1350 } else { 1351 printf("\n\t"); 1352 VOP_PRINT(vp); 1353 } 1354 } 1355 1356 #ifdef DDB 1357 #include <ddb/ddb.h> 1358 1359 static int db_show_locked_vnodes(struct mount *mp, void *data); 1360 1361 /* 1362 * List all of the locked vnodes in the system. 1363 * Called when debugging the kernel. 1364 */ 1365 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1366 { 1367 printf("Locked vnodes\n"); 1368 mountlist_scan(db_show_locked_vnodes, NULL, 1369 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1370 } 1371 1372 static int 1373 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1374 { 1375 struct vnode *vp; 1376 1377 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1378 if (VOP_ISLOCKED(vp, NULL)) 1379 vprint((char *)0, vp); 1380 } 1381 return(0); 1382 } 1383 #endif 1384 1385 /* 1386 * Top level filesystem related information gathering. 1387 */ 1388 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1389 1390 static int 1391 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1392 { 1393 int *name = (int *)arg1 - 1; /* XXX */ 1394 u_int namelen = arg2 + 1; /* XXX */ 1395 struct vfsconf *vfsp; 1396 1397 #if 1 || defined(COMPAT_PRELITE2) 1398 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1399 if (namelen == 1) 1400 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1401 #endif 1402 1403 #ifdef notyet 1404 /* all sysctl names at this level are at least name and field */ 1405 if (namelen < 2) 1406 return (ENOTDIR); /* overloaded */ 1407 if (name[0] != VFS_GENERIC) { 1408 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1409 if (vfsp->vfc_typenum == name[0]) 1410 break; 1411 if (vfsp == NULL) 1412 return (EOPNOTSUPP); 1413 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1414 oldp, oldlenp, newp, newlen, p)); 1415 } 1416 #endif 1417 switch (name[1]) { 1418 case VFS_MAXTYPENUM: 1419 if (namelen != 2) 1420 return (ENOTDIR); 1421 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1422 case VFS_CONF: 1423 if (namelen != 3) 1424 return (ENOTDIR); /* overloaded */ 1425 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1426 if (vfsp->vfc_typenum == name[2]) 1427 break; 1428 if (vfsp == NULL) 1429 return (EOPNOTSUPP); 1430 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1431 } 1432 return (EOPNOTSUPP); 1433 } 1434 1435 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1436 "Generic filesystem"); 1437 1438 #if 1 || defined(COMPAT_PRELITE2) 1439 1440 static int 1441 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1442 { 1443 int error; 1444 struct vfsconf *vfsp; 1445 struct ovfsconf ovfs; 1446 1447 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1448 bzero(&ovfs, sizeof(ovfs)); 1449 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1450 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1451 ovfs.vfc_index = vfsp->vfc_typenum; 1452 ovfs.vfc_refcount = vfsp->vfc_refcount; 1453 ovfs.vfc_flags = vfsp->vfc_flags; 1454 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1455 if (error) 1456 return error; 1457 } 1458 return 0; 1459 } 1460 1461 #endif /* 1 || COMPAT_PRELITE2 */ 1462 1463 /* 1464 * Check to see if a filesystem is mounted on a block device. 1465 */ 1466 int 1467 vfs_mountedon(struct vnode *vp) 1468 { 1469 dev_t dev; 1470 1471 if ((dev = vp->v_rdev) == NULL) 1472 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1473 if (dev != NODEV && dev->si_mountpoint) 1474 return (EBUSY); 1475 return (0); 1476 } 1477 1478 /* 1479 * Unmount all filesystems. The list is traversed in reverse order 1480 * of mounting to avoid dependencies. 1481 */ 1482 1483 static int vfs_umountall_callback(struct mount *mp, void *data); 1484 1485 void 1486 vfs_unmountall(void) 1487 { 1488 struct thread *td = curthread; 1489 int count; 1490 1491 if (td->td_proc == NULL) 1492 td = initproc->p_thread; /* XXX XXX use proc0 instead? */ 1493 1494 do { 1495 count = mountlist_scan(vfs_umountall_callback, 1496 &td, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1497 } while (count); 1498 } 1499 1500 static 1501 int 1502 vfs_umountall_callback(struct mount *mp, void *data) 1503 { 1504 struct thread *td = *(struct thread **)data; 1505 int error; 1506 1507 error = dounmount(mp, MNT_FORCE, td); 1508 if (error) { 1509 mountlist_remove(mp); 1510 printf("unmount of filesystem mounted from %s failed (", 1511 mp->mnt_stat.f_mntfromname); 1512 if (error == EBUSY) 1513 printf("BUSY)\n"); 1514 else 1515 printf("%d)\n", error); 1516 } 1517 return(1); 1518 } 1519 1520 /* 1521 * Build hash lists of net addresses and hang them off the mount point. 1522 * Called by ufs_mount() to set up the lists of export addresses. 1523 */ 1524 static int 1525 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1526 struct export_args *argp) 1527 { 1528 struct netcred *np; 1529 struct radix_node_head *rnh; 1530 int i; 1531 struct radix_node *rn; 1532 struct sockaddr *saddr, *smask = 0; 1533 struct domain *dom; 1534 int error; 1535 1536 if (argp->ex_addrlen == 0) { 1537 if (mp->mnt_flag & MNT_DEFEXPORTED) 1538 return (EPERM); 1539 np = &nep->ne_defexported; 1540 np->netc_exflags = argp->ex_flags; 1541 np->netc_anon = argp->ex_anon; 1542 np->netc_anon.cr_ref = 1; 1543 mp->mnt_flag |= MNT_DEFEXPORTED; 1544 return (0); 1545 } 1546 1547 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1548 return (EINVAL); 1549 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1550 return (EINVAL); 1551 1552 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1553 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1554 bzero((caddr_t) np, i); 1555 saddr = (struct sockaddr *) (np + 1); 1556 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1557 goto out; 1558 if (saddr->sa_len > argp->ex_addrlen) 1559 saddr->sa_len = argp->ex_addrlen; 1560 if (argp->ex_masklen) { 1561 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1562 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1563 if (error) 1564 goto out; 1565 if (smask->sa_len > argp->ex_masklen) 1566 smask->sa_len = argp->ex_masklen; 1567 } 1568 i = saddr->sa_family; 1569 if ((rnh = nep->ne_rtable[i]) == 0) { 1570 /* 1571 * Seems silly to initialize every AF when most are not used, 1572 * do so on demand here 1573 */ 1574 SLIST_FOREACH(dom, &domains, dom_next) 1575 if (dom->dom_family == i && dom->dom_rtattach) { 1576 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1577 dom->dom_rtoffset); 1578 break; 1579 } 1580 if ((rnh = nep->ne_rtable[i]) == 0) { 1581 error = ENOBUFS; 1582 goto out; 1583 } 1584 } 1585 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1586 np->netc_rnodes); 1587 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1588 error = EPERM; 1589 goto out; 1590 } 1591 np->netc_exflags = argp->ex_flags; 1592 np->netc_anon = argp->ex_anon; 1593 np->netc_anon.cr_ref = 1; 1594 return (0); 1595 out: 1596 free(np, M_NETADDR); 1597 return (error); 1598 } 1599 1600 /* ARGSUSED */ 1601 static int 1602 vfs_free_netcred(struct radix_node *rn, void *w) 1603 { 1604 struct radix_node_head *rnh = (struct radix_node_head *) w; 1605 1606 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1607 free((caddr_t) rn, M_NETADDR); 1608 return (0); 1609 } 1610 1611 /* 1612 * Free the net address hash lists that are hanging off the mount points. 1613 */ 1614 static void 1615 vfs_free_addrlist(struct netexport *nep) 1616 { 1617 int i; 1618 struct radix_node_head *rnh; 1619 1620 for (i = 0; i <= AF_MAX; i++) 1621 if ((rnh = nep->ne_rtable[i])) { 1622 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1623 (caddr_t) rnh); 1624 free((caddr_t) rnh, M_RTABLE); 1625 nep->ne_rtable[i] = 0; 1626 } 1627 } 1628 1629 int 1630 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1631 { 1632 int error; 1633 1634 if (argp->ex_flags & MNT_DELEXPORT) { 1635 if (mp->mnt_flag & MNT_EXPUBLIC) { 1636 vfs_setpublicfs(NULL, NULL, NULL); 1637 mp->mnt_flag &= ~MNT_EXPUBLIC; 1638 } 1639 vfs_free_addrlist(nep); 1640 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1641 } 1642 if (argp->ex_flags & MNT_EXPORTED) { 1643 if (argp->ex_flags & MNT_EXPUBLIC) { 1644 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1645 return (error); 1646 mp->mnt_flag |= MNT_EXPUBLIC; 1647 } 1648 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1649 return (error); 1650 mp->mnt_flag |= MNT_EXPORTED; 1651 } 1652 return (0); 1653 } 1654 1655 1656 /* 1657 * Set the publicly exported filesystem (WebNFS). Currently, only 1658 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1659 */ 1660 int 1661 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1662 struct export_args *argp) 1663 { 1664 int error; 1665 struct vnode *rvp; 1666 char *cp; 1667 1668 /* 1669 * mp == NULL -> invalidate the current info, the FS is 1670 * no longer exported. May be called from either vfs_export 1671 * or unmount, so check if it hasn't already been done. 1672 */ 1673 if (mp == NULL) { 1674 if (nfs_pub.np_valid) { 1675 nfs_pub.np_valid = 0; 1676 if (nfs_pub.np_index != NULL) { 1677 FREE(nfs_pub.np_index, M_TEMP); 1678 nfs_pub.np_index = NULL; 1679 } 1680 } 1681 return (0); 1682 } 1683 1684 /* 1685 * Only one allowed at a time. 1686 */ 1687 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1688 return (EBUSY); 1689 1690 /* 1691 * Get real filehandle for root of exported FS. 1692 */ 1693 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1694 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1695 1696 if ((error = VFS_ROOT(mp, &rvp))) 1697 return (error); 1698 1699 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1700 return (error); 1701 1702 vput(rvp); 1703 1704 /* 1705 * If an indexfile was specified, pull it in. 1706 */ 1707 if (argp->ex_indexfile != NULL) { 1708 int namelen; 1709 1710 error = vn_get_namelen(rvp, &namelen); 1711 if (error) 1712 return (error); 1713 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 1714 M_WAITOK); 1715 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1716 namelen, (size_t *)0); 1717 if (!error) { 1718 /* 1719 * Check for illegal filenames. 1720 */ 1721 for (cp = nfs_pub.np_index; *cp; cp++) { 1722 if (*cp == '/') { 1723 error = EINVAL; 1724 break; 1725 } 1726 } 1727 } 1728 if (error) { 1729 FREE(nfs_pub.np_index, M_TEMP); 1730 return (error); 1731 } 1732 } 1733 1734 nfs_pub.np_mount = mp; 1735 nfs_pub.np_valid = 1; 1736 return (0); 1737 } 1738 1739 struct netcred * 1740 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1741 struct sockaddr *nam) 1742 { 1743 struct netcred *np; 1744 struct radix_node_head *rnh; 1745 struct sockaddr *saddr; 1746 1747 np = NULL; 1748 if (mp->mnt_flag & MNT_EXPORTED) { 1749 /* 1750 * Lookup in the export list first. 1751 */ 1752 if (nam != NULL) { 1753 saddr = nam; 1754 rnh = nep->ne_rtable[saddr->sa_family]; 1755 if (rnh != NULL) { 1756 np = (struct netcred *) 1757 (*rnh->rnh_matchaddr)((char *)saddr, 1758 rnh); 1759 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1760 np = NULL; 1761 } 1762 } 1763 /* 1764 * If no address match, use the default if it exists. 1765 */ 1766 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1767 np = &nep->ne_defexported; 1768 } 1769 return (np); 1770 } 1771 1772 /* 1773 * perform msync on all vnodes under a mount point. The mount point must 1774 * be locked. This code is also responsible for lazy-freeing unreferenced 1775 * vnodes whos VM objects no longer contain pages. 1776 * 1777 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1778 */ 1779 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1780 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1781 1782 void 1783 vfs_msync(struct mount *mp, int flags) 1784 { 1785 vmntvnodescan(mp, VMSC_REFVP, vfs_msync_scan1, vfs_msync_scan2, 1786 (void *)flags); 1787 } 1788 1789 /* 1790 * scan1 is a fast pre-check. There could be hundreds of thousands of 1791 * vnodes, we cannot afford to do anything heavy weight until we have a 1792 * fairly good indication that there is work to do. 1793 */ 1794 static 1795 int 1796 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1797 { 1798 int flags = (int)data; 1799 1800 if ((vp->v_flag & VRECLAIMED) == 0) { 1801 if (vshouldfree(vp, 0)) 1802 return(0); /* call scan2 */ 1803 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1804 (vp->v_flag & VOBJDIRTY) && 1805 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1806 return(0); /* call scan2 */ 1807 } 1808 } 1809 1810 /* 1811 * do not call scan2, continue the loop 1812 */ 1813 return(-1); 1814 } 1815 1816 static 1817 int 1818 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1819 { 1820 vm_object_t obj; 1821 int flags = (int)data; 1822 1823 if (vp->v_flag & VRECLAIMED) 1824 return(0); 1825 1826 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1827 (vp->v_flag & VOBJDIRTY) && 1828 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1829 if (VOP_GETVOBJECT(vp, &obj) == 0) { 1830 vm_object_page_clean(obj, 0, 0, 1831 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1832 } 1833 } 1834 return(0); 1835 } 1836 1837 /* 1838 * Create the VM object needed for VMIO and mmap support. This 1839 * is done for all VREG files in the system. Some filesystems might 1840 * afford the additional metadata buffering capability of the 1841 * VMIO code by making the device node be VMIO mode also. 1842 * 1843 * vp must be locked when vfs_object_create is called. 1844 */ 1845 int 1846 vfs_object_create(struct vnode *vp, struct thread *td) 1847 { 1848 return (VOP_CREATEVOBJECT(vp, td)); 1849 } 1850 1851 /* 1852 * Record a process's interest in events which might happen to 1853 * a vnode. Because poll uses the historic select-style interface 1854 * internally, this routine serves as both the ``check for any 1855 * pending events'' and the ``record my interest in future events'' 1856 * functions. (These are done together, while the lock is held, 1857 * to avoid race conditions.) 1858 */ 1859 int 1860 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 1861 { 1862 lwkt_tokref ilock; 1863 1864 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1865 if (vp->v_pollinfo.vpi_revents & events) { 1866 /* 1867 * This leaves events we are not interested 1868 * in available for the other process which 1869 * which presumably had requested them 1870 * (otherwise they would never have been 1871 * recorded). 1872 */ 1873 events &= vp->v_pollinfo.vpi_revents; 1874 vp->v_pollinfo.vpi_revents &= ~events; 1875 1876 lwkt_reltoken(&ilock); 1877 return events; 1878 } 1879 vp->v_pollinfo.vpi_events |= events; 1880 selrecord(td, &vp->v_pollinfo.vpi_selinfo); 1881 lwkt_reltoken(&ilock); 1882 return 0; 1883 } 1884 1885 /* 1886 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1887 * it is possible for us to miss an event due to race conditions, but 1888 * that condition is expected to be rare, so for the moment it is the 1889 * preferred interface. 1890 */ 1891 void 1892 vn_pollevent(struct vnode *vp, int events) 1893 { 1894 lwkt_tokref ilock; 1895 1896 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1897 if (vp->v_pollinfo.vpi_events & events) { 1898 /* 1899 * We clear vpi_events so that we don't 1900 * call selwakeup() twice if two events are 1901 * posted before the polling process(es) is 1902 * awakened. This also ensures that we take at 1903 * most one selwakeup() if the polling process 1904 * is no longer interested. However, it does 1905 * mean that only one event can be noticed at 1906 * a time. (Perhaps we should only clear those 1907 * event bits which we note?) XXX 1908 */ 1909 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1910 vp->v_pollinfo.vpi_revents |= events; 1911 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1912 } 1913 lwkt_reltoken(&ilock); 1914 } 1915 1916 /* 1917 * Wake up anyone polling on vp because it is being revoked. 1918 * This depends on dead_poll() returning POLLHUP for correct 1919 * behavior. 1920 */ 1921 void 1922 vn_pollgone(struct vnode *vp) 1923 { 1924 lwkt_tokref ilock; 1925 1926 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1927 if (vp->v_pollinfo.vpi_events) { 1928 vp->v_pollinfo.vpi_events = 0; 1929 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1930 } 1931 lwkt_reltoken(&ilock); 1932 } 1933 1934 /* 1935 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened 1936 * (or v_rdev might be NULL). 1937 */ 1938 dev_t 1939 vn_todev(struct vnode *vp) 1940 { 1941 if (vp->v_type != VBLK && vp->v_type != VCHR) 1942 return (NODEV); 1943 KKASSERT(vp->v_rdev != NULL); 1944 return (vp->v_rdev); 1945 } 1946 1947 /* 1948 * Check if vnode represents a disk device. The vnode does not need to be 1949 * opened. 1950 */ 1951 int 1952 vn_isdisk(struct vnode *vp, int *errp) 1953 { 1954 dev_t dev; 1955 1956 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1957 if (errp != NULL) 1958 *errp = ENOTBLK; 1959 return (0); 1960 } 1961 1962 if ((dev = vp->v_rdev) == NULL) 1963 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1964 if (dev == NULL || dev == NODEV) { 1965 if (errp != NULL) 1966 *errp = ENXIO; 1967 return (0); 1968 } 1969 if (dev_is_good(dev) == 0) { 1970 if (errp != NULL) 1971 *errp = ENXIO; 1972 return (0); 1973 } 1974 if ((dev_dflags(dev) & D_DISK) == 0) { 1975 if (errp != NULL) 1976 *errp = ENOTBLK; 1977 return (0); 1978 } 1979 if (errp != NULL) 1980 *errp = 0; 1981 return (1); 1982 } 1983 1984 #ifdef DEBUG_VFS_LOCKS 1985 1986 void 1987 assert_vop_locked(struct vnode *vp, const char *str) 1988 { 1989 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) { 1990 panic("%s: %p is not locked shared but should be", str, vp); 1991 } 1992 } 1993 1994 void 1995 assert_vop_unlocked(struct vnode *vp, const char *str) 1996 { 1997 if (vp && IS_LOCKING_VFS(vp)) { 1998 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) { 1999 panic("%s: %p is locked but should not be", str, vp); 2000 } 2001 } 2002 } 2003 2004 #endif 2005 2006 int 2007 vn_get_namelen(struct vnode *vp, int *namelen) 2008 { 2009 int error, retval[2]; 2010 2011 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2012 if (error) 2013 return (error); 2014 *namelen = *retval; 2015 return (0); 2016 } 2017 2018 int 2019 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2020 uint16_t d_namlen, const char *d_name) 2021 { 2022 struct dirent *dp; 2023 size_t len; 2024 2025 len = _DIRENT_RECLEN(d_namlen); 2026 if (len > uio->uio_resid) 2027 return(1); 2028 2029 dp = malloc(len, M_TEMP, M_WAITOK | M_ZERO); 2030 2031 dp->d_ino = d_ino; 2032 dp->d_namlen = d_namlen; 2033 dp->d_type = d_type; 2034 bcopy(d_name, dp->d_name, d_namlen); 2035 2036 *error = uiomove((caddr_t)dp, len, uio); 2037 2038 free(dp, M_TEMP); 2039 2040 return(0); 2041 } 2042