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 1.57 2005/06/06 15:02:28 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/vmmeter.h> 68 #include <sys/vnode.h> 69 70 #include <machine/limits.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_object.h> 74 #include <vm/vm_extern.h> 75 #include <vm/vm_kern.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_page.h> 79 #include <vm/vm_pager.h> 80 #include <vm/vnode_pager.h> 81 #include <vm/vm_zone.h> 82 83 #include <sys/buf2.h> 84 #include <sys/thread2.h> 85 86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 87 88 int numvnodes; 89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 90 int vfs_fastdev = 1; 91 SYSCTL_INT(_vfs, OID_AUTO, fastdev, CTLFLAG_RW, &vfs_fastdev, 0, ""); 92 93 enum vtype iftovt_tab[16] = { 94 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 95 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 96 }; 97 int vttoif_tab[9] = { 98 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 99 S_IFSOCK, S_IFIFO, S_IFMT, 100 }; 101 102 static int reassignbufcalls; 103 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, 104 &reassignbufcalls, 0, ""); 105 static int reassignbufloops; 106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, 107 &reassignbufloops, 0, ""); 108 static int reassignbufsortgood; 109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, 110 &reassignbufsortgood, 0, ""); 111 static int reassignbufsortbad; 112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, 113 &reassignbufsortbad, 0, ""); 114 static int reassignbufmethod = 1; 115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, 116 &reassignbufmethod, 0, ""); 117 118 int nfs_mount_type = -1; 119 static struct lwkt_token spechash_token; 120 struct nfs_public nfs_pub; /* publicly exported FS */ 121 122 int desiredvnodes; 123 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 124 &desiredvnodes, 0, "Maximum number of vnodes"); 125 126 static void vfs_free_addrlist (struct netexport *nep); 127 static int vfs_free_netcred (struct radix_node *rn, void *w); 128 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 129 struct export_args *argp); 130 131 extern int dev_ref_debug; 132 extern struct vnodeopv_entry_desc spec_vnodeop_entries[]; 133 134 /* 135 * Red black tree functions 136 */ 137 static int rb_buf_compare(struct buf *b1, struct buf *b2); 138 RB_GENERATE(buf_rb_tree, buf, b_rbnode, rb_buf_compare); 139 140 static int 141 rb_buf_compare(struct buf *b1, struct buf *b2) 142 { 143 if (b1->b_lblkno < b2->b_lblkno) 144 return(-1); 145 if (b1->b_lblkno > b2->b_lblkno) 146 return(1); 147 return(0); 148 } 149 150 /* 151 * Return 0 if the vnode is already on the free list or cannot be placed 152 * on the free list. Return 1 if the vnode can be placed on the free list. 153 */ 154 static __inline int 155 vshouldfree(struct vnode *vp, int usecount) 156 { 157 if (vp->v_flag & VFREE) 158 return (0); /* already free */ 159 if (vp->v_holdcnt != 0 || vp->v_usecount != usecount) 160 return (0); /* other holderse */ 161 if (vp->v_object && 162 (vp->v_object->ref_count || vp->v_object->resident_page_count)) { 163 return (0); 164 } 165 return (1); 166 } 167 168 /* 169 * Initialize the vnode management data structures. 170 * 171 * Called from vfsinit() 172 */ 173 void 174 vfs_subr_init(void) 175 { 176 /* 177 * Desired vnodes is a result of the physical page count 178 * and the size of kernel's heap. It scales in proportion 179 * to the amount of available physical memory. This can 180 * cause trouble on 64-bit and large memory platforms. 181 */ 182 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */ 183 desiredvnodes = 184 min(maxproc + vmstats.v_page_count /4, 185 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 186 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 187 188 lwkt_token_init(&spechash_token); 189 } 190 191 /* 192 * Knob to control the precision of file timestamps: 193 * 194 * 0 = seconds only; nanoseconds zeroed. 195 * 1 = seconds and nanoseconds, accurate within 1/HZ. 196 * 2 = seconds and nanoseconds, truncated to microseconds. 197 * >=3 = seconds and nanoseconds, maximum precision. 198 */ 199 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 200 201 static int timestamp_precision = TSP_SEC; 202 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 203 ×tamp_precision, 0, ""); 204 205 /* 206 * Get a current timestamp. 207 */ 208 void 209 vfs_timestamp(struct timespec *tsp) 210 { 211 struct timeval tv; 212 213 switch (timestamp_precision) { 214 case TSP_SEC: 215 tsp->tv_sec = time_second; 216 tsp->tv_nsec = 0; 217 break; 218 case TSP_HZ: 219 getnanotime(tsp); 220 break; 221 case TSP_USEC: 222 microtime(&tv); 223 TIMEVAL_TO_TIMESPEC(&tv, tsp); 224 break; 225 case TSP_NSEC: 226 default: 227 nanotime(tsp); 228 break; 229 } 230 } 231 232 /* 233 * Set vnode attributes to VNOVAL 234 */ 235 void 236 vattr_null(struct vattr *vap) 237 { 238 vap->va_type = VNON; 239 vap->va_size = VNOVAL; 240 vap->va_bytes = VNOVAL; 241 vap->va_mode = VNOVAL; 242 vap->va_nlink = VNOVAL; 243 vap->va_uid = VNOVAL; 244 vap->va_gid = VNOVAL; 245 vap->va_fsid = VNOVAL; 246 vap->va_fileid = VNOVAL; 247 vap->va_blocksize = VNOVAL; 248 vap->va_rdev = 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 } 259 260 /* 261 * Update outstanding I/O count and do wakeup if requested. 262 */ 263 void 264 vwakeup(struct buf *bp) 265 { 266 struct vnode *vp; 267 268 if ((vp = bp->b_vp)) { 269 vp->v_numoutput--; 270 if (vp->v_numoutput < 0) 271 panic("vwakeup: neg numoutput"); 272 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 273 vp->v_flag &= ~VBWAIT; 274 wakeup((caddr_t) &vp->v_numoutput); 275 } 276 } 277 } 278 279 /* 280 * Flush out and invalidate all buffers associated with a vnode. 281 * 282 * vp must be locked. 283 */ 284 static int vinvalbuf_bp(struct buf *bp, void *data); 285 286 struct vinvalbuf_bp_info { 287 struct vnode *vp; 288 int slptimeo; 289 int slpflag; 290 int flags; 291 }; 292 293 int 294 vinvalbuf(struct vnode *vp, int flags, struct thread *td, 295 int slpflag, int slptimeo) 296 { 297 struct vinvalbuf_bp_info info; 298 int error; 299 vm_object_t object; 300 301 /* 302 * If we are being asked to save, call fsync to ensure that the inode 303 * is updated. 304 */ 305 if (flags & V_SAVE) { 306 crit_enter(); 307 while (vp->v_numoutput) { 308 vp->v_flag |= VBWAIT; 309 error = tsleep((caddr_t)&vp->v_numoutput, 310 slpflag, "vinvlbuf", slptimeo); 311 if (error) { 312 crit_exit(); 313 return (error); 314 } 315 } 316 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 317 crit_exit(); 318 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) 319 return (error); 320 crit_enter(); 321 if (vp->v_numoutput > 0 || 322 !RB_EMPTY(&vp->v_rbdirty_tree)) 323 panic("vinvalbuf: dirty bufs"); 324 } 325 crit_exit(); 326 } 327 crit_enter(); 328 info.slptimeo = slptimeo; 329 info.slpflag = slpflag; 330 info.flags = flags; 331 info.vp = vp; 332 333 /* 334 * Flush the buffer cache until nothing is left. 335 */ 336 while (!RB_EMPTY(&vp->v_rbclean_tree) || 337 !RB_EMPTY(&vp->v_rbdirty_tree)) { 338 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 339 vinvalbuf_bp, &info); 340 if (error == 0) { 341 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 342 vinvalbuf_bp, &info); 343 } 344 } 345 346 /* 347 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 348 * have write I/O in-progress but if there is a VM object then the 349 * VM object can also have read-I/O in-progress. 350 */ 351 do { 352 while (vp->v_numoutput > 0) { 353 vp->v_flag |= VBWAIT; 354 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0); 355 } 356 if (VOP_GETVOBJECT(vp, &object) == 0) { 357 while (object->paging_in_progress) 358 vm_object_pip_sleep(object, "vnvlbx"); 359 } 360 } while (vp->v_numoutput > 0); 361 362 crit_exit(); 363 364 /* 365 * Destroy the copy in the VM cache, too. 366 */ 367 if (VOP_GETVOBJECT(vp, &object) == 0) { 368 vm_object_page_remove(object, 0, 0, 369 (flags & V_SAVE) ? TRUE : FALSE); 370 } 371 372 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 373 panic("vinvalbuf: flush failed"); 374 return (0); 375 } 376 377 static int 378 vinvalbuf_bp(struct buf *bp, void *data) 379 { 380 struct vinvalbuf_bp_info *info = data; 381 int error; 382 383 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 384 error = BUF_TIMELOCK(bp, 385 LK_EXCLUSIVE | LK_SLEEPFAIL, 386 "vinvalbuf", info->slpflag, info->slptimeo); 387 if (error == 0) { 388 BUF_UNLOCK(bp); 389 error = ENOLCK; 390 } 391 if (error == ENOLCK) 392 return(0); 393 return (-error); 394 } 395 /* 396 * XXX Since there are no node locks for NFS, I 397 * believe there is a slight chance that a delayed 398 * write will occur while sleeping just above, so 399 * check for it. Note that vfs_bio_awrite expects 400 * buffers to reside on a queue, while VOP_BWRITE and 401 * brelse do not. 402 */ 403 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 404 (info->flags & V_SAVE)) { 405 if (bp->b_vp == info->vp) { 406 if (bp->b_flags & B_CLUSTEROK) { 407 BUF_UNLOCK(bp); 408 vfs_bio_awrite(bp); 409 } else { 410 bremfree(bp); 411 bp->b_flags |= B_ASYNC; 412 VOP_BWRITE(bp->b_vp, bp); 413 } 414 } else { 415 bremfree(bp); 416 VOP_BWRITE(bp->b_vp, bp); 417 } 418 } else { 419 bremfree(bp); 420 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 421 bp->b_flags &= ~B_ASYNC; 422 brelse(bp); 423 } 424 return(0); 425 } 426 427 /* 428 * Truncate a file's buffer and pages to a specified length. This 429 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 430 * sync activity. 431 * 432 * The vnode must be locked. 433 */ 434 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 435 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 436 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 437 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 438 439 int 440 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize) 441 { 442 daddr_t trunclbn; 443 int count; 444 445 /* 446 * Round up to the *next* lbn, then destroy the buffers in question. 447 * Since we are only removing some of the buffers we must rely on the 448 * scan count to determine whether a loop is necessary. 449 */ 450 trunclbn = (length + blksize - 1) / blksize; 451 452 crit_enter(); 453 do { 454 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 455 vtruncbuf_bp_trunc_cmp, 456 vtruncbuf_bp_trunc, &trunclbn); 457 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 458 vtruncbuf_bp_trunc_cmp, 459 vtruncbuf_bp_trunc, &trunclbn); 460 } while(count); 461 462 /* 463 * For safety, fsync any remaining metadata if the file is not being 464 * truncated to 0. Since the metadata does not represent the entire 465 * dirty list we have to rely on the hit count to ensure that we get 466 * all of it. 467 */ 468 if (length > 0) { 469 do { 470 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 471 vtruncbuf_bp_metasync_cmp, 472 vtruncbuf_bp_metasync, vp); 473 } while (count); 474 } 475 476 /* 477 * Wait for any in-progress I/O to complete before returning (why?) 478 */ 479 while (vp->v_numoutput > 0) { 480 vp->v_flag |= VBWAIT; 481 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0); 482 } 483 484 crit_exit(); 485 486 vnode_pager_setsize(vp, length); 487 488 return (0); 489 } 490 491 /* 492 * The callback buffer is beyond the new file EOF and must be destroyed. 493 * Note that the compare function must conform to the RB_SCAN's requirements. 494 */ 495 static 496 int 497 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 498 { 499 if (bp->b_lblkno >= *(daddr_t *)data) 500 return(0); 501 return(-1); 502 } 503 504 static 505 int 506 vtruncbuf_bp_trunc(struct buf *bp, void *data) 507 { 508 /* 509 * Do not try to use a buffer we cannot immediately lock, but sleep 510 * anyway to prevent a livelock. The code will loop until all buffers 511 * can be acted upon. 512 */ 513 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 514 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 515 BUF_UNLOCK(bp); 516 } else { 517 bremfree(bp); 518 bp->b_flags |= (B_INVAL | B_RELBUF); 519 bp->b_flags &= ~B_ASYNC; 520 brelse(bp); 521 } 522 return(1); 523 } 524 525 /* 526 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 527 * blocks (with a negative lblkno) are scanned. 528 * Note that the compare function must conform to the RB_SCAN's requirements. 529 */ 530 static int 531 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data) 532 { 533 if (bp->b_lblkno < 0) 534 return(0); 535 return(1); 536 } 537 538 static int 539 vtruncbuf_bp_metasync(struct buf *bp, void *data) 540 { 541 struct vnode *vp = data; 542 543 if (bp->b_flags & B_DELWRI) { 544 /* 545 * Do not try to use a buffer we cannot immediately lock, 546 * but sleep anyway to prevent a livelock. The code will 547 * loop until all buffers can be acted upon. 548 */ 549 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 550 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 551 BUF_UNLOCK(bp); 552 } else { 553 bremfree(bp); 554 if (bp->b_vp == vp) { 555 bp->b_flags |= B_ASYNC; 556 } else { 557 bp->b_flags &= ~B_ASYNC; 558 } 559 VOP_BWRITE(bp->b_vp, bp); 560 } 561 return(1); 562 } else { 563 return(0); 564 } 565 } 566 567 /* 568 * vfsync - implements a multipass fsync on a file which understands 569 * dependancies and meta-data. The passed vnode must be locked. The 570 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 571 * 572 * When fsyncing data asynchronously just do one consolidated pass starting 573 * with the most negative block number. This may not get all the data due 574 * to dependancies. 575 * 576 * When fsyncing data synchronously do a data pass, then a metadata pass, 577 * then do additional data+metadata passes to try to get all the data out. 578 */ 579 static int vfsync_wait_output(struct vnode *vp, 580 int (*waitoutput)(struct vnode *, struct thread *)); 581 static int vfsync_data_only_cmp(struct buf *bp, void *data); 582 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 583 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 584 static int vfsync_bp(struct buf *bp, void *data); 585 586 struct vfsync_info { 587 struct vnode *vp; 588 int synchronous; 589 int syncdeps; 590 int lazycount; 591 int lazylimit; 592 daddr_t lbn; 593 int (*checkdef)(struct buf *); 594 }; 595 596 int 597 vfsync(struct vnode *vp, int waitfor, int passes, daddr_t lbn, 598 int (*checkdef)(struct buf *), 599 int (*waitoutput)(struct vnode *, struct thread *)) 600 { 601 struct vfsync_info info; 602 int error; 603 604 bzero(&info, sizeof(info)); 605 info.vp = vp; 606 info.lbn = lbn; 607 if ((info.checkdef = checkdef) == NULL) 608 info.syncdeps = 1; 609 610 crit_enter(); 611 612 switch(waitfor) { 613 case MNT_LAZY: 614 /* 615 * Lazy (filesystem syncer typ) Asynchronous plus limit the 616 * number of data (not meta) pages we try to flush to 1MB. 617 * A non-zero return means that lazy limit was reached. 618 */ 619 info.lazylimit = 1024 * 1024; 620 info.syncdeps = 1; 621 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 622 vfsync_lazy_range_cmp, vfsync_bp, &info); 623 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 624 vfsync_meta_only_cmp, vfsync_bp, &info); 625 if (error == 0) 626 vp->v_lazyw = 0; 627 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 628 vn_syncer_add_to_worklist(vp, 1); 629 error = 0; 630 break; 631 case MNT_NOWAIT: 632 /* 633 * Asynchronous. Do a data-only pass and a meta-only pass. 634 */ 635 info.syncdeps = 1; 636 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 637 vfsync_bp, &info); 638 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 639 vfsync_bp, &info); 640 error = 0; 641 break; 642 default: 643 /* 644 * Synchronous. Do a data-only pass, then a meta-data+data 645 * pass, then additional integrated passes to try to get 646 * all the dependancies flushed. 647 */ 648 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 649 vfsync_bp, &info); 650 error = vfsync_wait_output(vp, waitoutput); 651 if (error == 0) { 652 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 653 vfsync_bp, &info); 654 error = vfsync_wait_output(vp, waitoutput); 655 } 656 while (error == 0 && passes > 0 && 657 !RB_EMPTY(&vp->v_rbdirty_tree)) { 658 if (--passes == 0) { 659 info.synchronous = 1; 660 info.syncdeps = 1; 661 } 662 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 663 vfsync_bp, &info); 664 if (error < 0) 665 error = -error; 666 info.syncdeps = 1; 667 if (error == 0) 668 error = vfsync_wait_output(vp, waitoutput); 669 } 670 break; 671 } 672 crit_exit(); 673 return(error); 674 } 675 676 static int 677 vfsync_wait_output(struct vnode *vp, int (*waitoutput)(struct vnode *, struct thread *)) 678 { 679 int error = 0; 680 681 while (vp->v_numoutput) { 682 vp->v_flag |= VBWAIT; 683 tsleep(&vp->v_numoutput, 0, "fsfsn", 0); 684 } 685 if (waitoutput) 686 error = waitoutput(vp, curthread); 687 return(error); 688 } 689 690 static int 691 vfsync_data_only_cmp(struct buf *bp, void *data) 692 { 693 if (bp->b_lblkno < 0) 694 return(-1); 695 return(0); 696 } 697 698 static int 699 vfsync_meta_only_cmp(struct buf *bp, void *data) 700 { 701 if (bp->b_lblkno < 0) 702 return(0); 703 return(1); 704 } 705 706 static int 707 vfsync_lazy_range_cmp(struct buf *bp, void *data) 708 { 709 struct vfsync_info *info = data; 710 if (bp->b_lblkno < info->vp->v_lazyw) 711 return(-1); 712 return(0); 713 } 714 715 static int 716 vfsync_bp(struct buf *bp, void *data) 717 { 718 struct vfsync_info *info = data; 719 struct vnode *vp = info->vp; 720 int error; 721 722 /* 723 * if syncdeps is not set we do not try to write buffers which have 724 * dependancies. 725 */ 726 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) 727 return(0); 728 729 /* 730 * Ignore buffers that we cannot immediately lock. XXX 731 */ 732 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) 733 return(0); 734 if ((bp->b_flags & B_DELWRI) == 0) 735 panic("vfsync_bp: buffer not dirty"); 736 if (vp != bp->b_vp) 737 panic("vfsync_bp: buffer vp mismatch"); 738 739 /* 740 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 741 * has been written but an additional handshake with the device 742 * is required before we can dispose of the buffer. We have no idea 743 * how to do this so we have to skip these buffers. 744 */ 745 if (bp->b_flags & B_NEEDCOMMIT) { 746 BUF_UNLOCK(bp); 747 return(0); 748 } 749 750 /* 751 * (LEGACY FROM UFS, REMOVE WHEN POSSIBLE) - invalidate any dirty 752 * buffers beyond the file EOF. 753 */ 754 if (info->lbn != (daddr_t)-1 && vp->v_type == VREG && 755 bp->b_lblkno >= info->lbn) { 756 bremfree(bp); 757 bp->b_flags |= B_INVAL | B_NOCACHE; 758 crit_exit(); 759 brelse(bp); 760 crit_enter(); 761 } 762 763 if (info->synchronous) { 764 /* 765 * Synchronous flushing. An error may be returned. 766 */ 767 bremfree(bp); 768 crit_exit(); 769 error = bwrite(bp); 770 crit_enter(); 771 } else { 772 /* 773 * Asynchronous flushing. A negative return value simply 774 * stops the scan and is not considered an error. We use 775 * this to support limited MNT_LAZY flushes. 776 */ 777 vp->v_lazyw = bp->b_lblkno; 778 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 779 BUF_UNLOCK(bp); 780 info->lazycount += vfs_bio_awrite(bp); 781 } else { 782 info->lazycount += bp->b_bufsize; 783 bremfree(bp); 784 crit_exit(); 785 bawrite(bp); 786 crit_enter(); 787 } 788 if (info->lazylimit && info->lazycount >= info->lazylimit) 789 error = 1; 790 else 791 error = 0; 792 } 793 return(-error); 794 } 795 796 /* 797 * Associate a buffer with a vnode. 798 */ 799 void 800 bgetvp(struct vnode *vp, struct buf *bp) 801 { 802 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 803 804 vhold(vp); 805 bp->b_vp = vp; 806 bp->b_dev = vn_todev(vp); 807 /* 808 * Insert onto list for new vnode. 809 */ 810 crit_enter(); 811 bp->b_xflags |= BX_VNCLEAN; 812 bp->b_xflags &= ~BX_VNDIRTY; 813 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 814 panic("reassignbuf: dup lblk vp %p bp %p", vp, bp); 815 crit_exit(); 816 } 817 818 /* 819 * Disassociate a buffer from a vnode. 820 */ 821 void 822 brelvp(struct buf *bp) 823 { 824 struct vnode *vp; 825 826 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 827 828 /* 829 * Delete from old vnode list, if on one. 830 */ 831 vp = bp->b_vp; 832 crit_enter(); 833 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 834 if (bp->b_xflags & BX_VNDIRTY) 835 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 836 else 837 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 838 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 839 } 840 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) { 841 vp->v_flag &= ~VONWORKLST; 842 LIST_REMOVE(vp, v_synclist); 843 } 844 crit_exit(); 845 bp->b_vp = NULL; 846 vdrop(vp); 847 } 848 849 /* 850 * Associate a p-buffer with a vnode. 851 * 852 * Also sets B_PAGING flag to indicate that vnode is not fully associated 853 * with the buffer. i.e. the bp has not been linked into the vnode or 854 * ref-counted. 855 */ 856 void 857 pbgetvp(struct vnode *vp, struct buf *bp) 858 { 859 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 860 861 bp->b_vp = vp; 862 bp->b_flags |= B_PAGING; 863 bp->b_dev = vn_todev(vp); 864 } 865 866 /* 867 * Disassociate a p-buffer from a vnode. 868 */ 869 void 870 pbrelvp(struct buf *bp) 871 { 872 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 873 874 bp->b_vp = NULL; 875 bp->b_flags &= ~B_PAGING; 876 } 877 878 void 879 pbreassignbuf(struct buf *bp, struct vnode *newvp) 880 { 881 if ((bp->b_flags & B_PAGING) == 0) { 882 panic( 883 "pbreassignbuf() on non phys bp %p", 884 bp 885 ); 886 } 887 bp->b_vp = newvp; 888 } 889 890 /* 891 * Reassign a buffer from one vnode to another. 892 * Used to assign file specific control information 893 * (indirect blocks) to the vnode to which they belong. 894 */ 895 void 896 reassignbuf(struct buf *bp, struct vnode *newvp) 897 { 898 int delay; 899 900 if (newvp == NULL) { 901 printf("reassignbuf: NULL"); 902 return; 903 } 904 ++reassignbufcalls; 905 906 /* 907 * B_PAGING flagged buffers cannot be reassigned because their vp 908 * is not fully linked in. 909 */ 910 if (bp->b_flags & B_PAGING) 911 panic("cannot reassign paging buffer"); 912 913 crit_enter(); 914 /* 915 * Delete from old vnode list, if on one. 916 */ 917 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 918 if (bp->b_xflags & BX_VNDIRTY) 919 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbdirty_tree, bp); 920 else 921 buf_rb_tree_RB_REMOVE(&bp->b_vp->v_rbclean_tree, bp); 922 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 923 if (bp->b_vp != newvp) { 924 vdrop(bp->b_vp); 925 bp->b_vp = NULL; /* for clarification */ 926 } 927 } 928 /* 929 * If dirty, put on list of dirty buffers; otherwise insert onto list 930 * of clean buffers. 931 */ 932 if (bp->b_flags & B_DELWRI) { 933 if ((newvp->v_flag & VONWORKLST) == 0) { 934 switch (newvp->v_type) { 935 case VDIR: 936 delay = dirdelay; 937 break; 938 case VCHR: 939 case VBLK: 940 if (newvp->v_rdev && 941 newvp->v_rdev->si_mountpoint != NULL) { 942 delay = metadelay; 943 break; 944 } 945 /* fall through */ 946 default: 947 delay = filedelay; 948 } 949 vn_syncer_add_to_worklist(newvp, delay); 950 } 951 bp->b_xflags |= BX_VNDIRTY; 952 if (buf_rb_tree_RB_INSERT(&newvp->v_rbdirty_tree, bp)) 953 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp); 954 } else { 955 bp->b_xflags |= BX_VNCLEAN; 956 if (buf_rb_tree_RB_INSERT(&newvp->v_rbclean_tree, bp)) 957 panic("reassignbuf: dup lblk vp %p bp %p", newvp, bp); 958 if ((newvp->v_flag & VONWORKLST) && 959 RB_EMPTY(&newvp->v_rbdirty_tree)) { 960 newvp->v_flag &= ~VONWORKLST; 961 LIST_REMOVE(newvp, v_synclist); 962 } 963 } 964 if (bp->b_vp != newvp) { 965 bp->b_vp = newvp; 966 vhold(bp->b_vp); 967 } 968 crit_exit(); 969 } 970 971 /* 972 * Create a vnode for a block device. 973 * Used for mounting the root file system. 974 */ 975 int 976 bdevvp(dev_t dev, struct vnode **vpp) 977 { 978 struct vnode *vp; 979 struct vnode *nvp; 980 int error; 981 982 if (dev == NODEV) { 983 *vpp = NULLVP; 984 return (ENXIO); 985 } 986 error = getspecialvnode(VT_NON, NULL, &spec_vnode_vops, &nvp, 0, 0); 987 if (error) { 988 *vpp = NULLVP; 989 return (error); 990 } 991 vp = nvp; 992 vp->v_type = VCHR; 993 vp->v_udev = dev->si_udev; 994 vx_unlock(vp); 995 *vpp = vp; 996 return (0); 997 } 998 999 int 1000 v_associate_rdev(struct vnode *vp, dev_t dev) 1001 { 1002 lwkt_tokref ilock; 1003 1004 if (dev == NULL || dev == NODEV) 1005 return(ENXIO); 1006 if (dev_is_good(dev) == 0) 1007 return(ENXIO); 1008 KKASSERT(vp->v_rdev == NULL); 1009 if (dev_ref_debug) 1010 printf("Z1"); 1011 vp->v_rdev = reference_dev(dev); 1012 lwkt_gettoken(&ilock, &spechash_token); 1013 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_specnext); 1014 lwkt_reltoken(&ilock); 1015 return(0); 1016 } 1017 1018 void 1019 v_release_rdev(struct vnode *vp) 1020 { 1021 lwkt_tokref ilock; 1022 dev_t dev; 1023 1024 if ((dev = vp->v_rdev) != NULL) { 1025 lwkt_gettoken(&ilock, &spechash_token); 1026 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_specnext); 1027 if (dev_ref_debug && vp->v_opencount != 0) { 1028 printf("releasing rdev with non-0 " 1029 "v_opencount(%d) (revoked?)\n", 1030 vp->v_opencount); 1031 } 1032 vp->v_rdev = NULL; 1033 vp->v_opencount = 0; 1034 release_dev(dev); 1035 lwkt_reltoken(&ilock); 1036 } 1037 } 1038 1039 /* 1040 * Add a vnode to the alias list hung off the dev_t. We only associate 1041 * the device number with the vnode. The actual device is not associated 1042 * until the vnode is opened (usually in spec_open()), and will be 1043 * disassociated on last close. 1044 */ 1045 void 1046 addaliasu(struct vnode *nvp, udev_t nvp_udev) 1047 { 1048 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1049 panic("addaliasu on non-special vnode"); 1050 nvp->v_udev = nvp_udev; 1051 } 1052 1053 /* 1054 * Disassociate a vnode from its underlying filesystem. 1055 * 1056 * The vnode must be VX locked and refd 1057 * 1058 * If there are v_usecount references to the vnode other then ours we have 1059 * to VOP_CLOSE the vnode before we can deactivate and reclaim it. 1060 */ 1061 void 1062 vclean(struct vnode *vp, int flags, struct thread *td) 1063 { 1064 int active; 1065 1066 /* 1067 * If the vnode has already been reclaimed we have nothing to do. 1068 */ 1069 if (vp->v_flag & VRECLAIMED) 1070 return; 1071 vp->v_flag |= VRECLAIMED; 1072 1073 /* 1074 * Scrap the vfs cache 1075 */ 1076 while (cache_inval_vp(vp, 0) != 0) { 1077 printf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1078 tsleep(vp, 0, "vclninv", 2); 1079 } 1080 1081 /* 1082 * Check to see if the vnode is in use. If so we have to reference it 1083 * before we clean it out so that its count cannot fall to zero and 1084 * generate a race against ourselves to recycle it. 1085 */ 1086 active = (vp->v_usecount > 1); 1087 1088 /* 1089 * Clean out any buffers associated with the vnode and destroy its 1090 * object, if it has one. 1091 */ 1092 vinvalbuf(vp, V_SAVE, td, 0, 0); 1093 VOP_DESTROYVOBJECT(vp); 1094 1095 /* 1096 * If purging an active vnode, it must be closed and 1097 * deactivated before being reclaimed. XXX 1098 * 1099 * Note that neither of these routines unlocks the vnode. 1100 */ 1101 if (active) { 1102 if (flags & DOCLOSE) 1103 VOP_CLOSE(vp, FNONBLOCK, td); 1104 } 1105 1106 /* 1107 * If the vnode has not be deactivated, deactivated it. 1108 */ 1109 if ((vp->v_flag & VINACTIVE) == 0) { 1110 vp->v_flag |= VINACTIVE; 1111 VOP_INACTIVE(vp, td); 1112 } 1113 1114 /* 1115 * Reclaim the vnode. 1116 */ 1117 if (VOP_RECLAIM(vp, td)) 1118 panic("vclean: cannot reclaim"); 1119 1120 /* 1121 * Done with purge, notify sleepers of the grim news. 1122 */ 1123 vp->v_ops = &dead_vnode_vops; 1124 vn_pollgone(vp); 1125 vp->v_tag = VT_NON; 1126 } 1127 1128 /* 1129 * Eliminate all activity associated with the requested vnode 1130 * and with all vnodes aliased to the requested vnode. 1131 * 1132 * The vnode must be referenced and vx_lock()'d 1133 * 1134 * revoke { struct vnode *a_vp, int a_flags } 1135 */ 1136 int 1137 vop_stdrevoke(struct vop_revoke_args *ap) 1138 { 1139 struct vnode *vp, *vq; 1140 lwkt_tokref ilock; 1141 dev_t dev; 1142 1143 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1144 1145 vp = ap->a_vp; 1146 1147 /* 1148 * If the vnode is already dead don't try to revoke it 1149 */ 1150 if (vp->v_flag & VRECLAIMED) 1151 return (0); 1152 1153 /* 1154 * If the vnode has a device association, scrap all vnodes associated 1155 * with the device. Don't let the device disappear on us while we 1156 * are scrapping the vnodes. 1157 * 1158 * The passed vp will probably show up in the list, do not VX lock 1159 * it twice! 1160 */ 1161 if (vp->v_type != VCHR && vp->v_type != VBLK) 1162 return(0); 1163 if ((dev = vp->v_rdev) == NULL) { 1164 if ((dev = udev2dev(vp->v_udev, vp->v_type == VBLK)) == NODEV) 1165 return(0); 1166 } 1167 reference_dev(dev); 1168 lwkt_gettoken(&ilock, &spechash_token); 1169 while ((vq = SLIST_FIRST(&dev->si_hlist)) != NULL) { 1170 if (vp == vq || vx_get(vq) == 0) { 1171 if (vq == SLIST_FIRST(&dev->si_hlist)) 1172 vgone(vq); 1173 if (vp != vq) 1174 vx_put(vq); 1175 } 1176 } 1177 lwkt_reltoken(&ilock); 1178 release_dev(dev); 1179 return (0); 1180 } 1181 1182 /* 1183 * Recycle an unused vnode to the front of the free list. 1184 * 1185 * Returns 1 if we were successfully able to recycle the vnode, 1186 * 0 otherwise. 1187 */ 1188 int 1189 vrecycle(struct vnode *vp, struct thread *td) 1190 { 1191 if (vp->v_usecount == 1) { 1192 vgone(vp); 1193 return (1); 1194 } 1195 return (0); 1196 } 1197 1198 /* 1199 * Eliminate all activity associated with a vnode in preparation for reuse. 1200 * 1201 * The vnode must be VX locked and refd and will remain VX locked and refd 1202 * on return. This routine may be called with the vnode in any state, as 1203 * long as it is VX locked. The vnode will be cleaned out and marked 1204 * VRECLAIMED but will not actually be reused until all existing refs and 1205 * holds go away. 1206 * 1207 * NOTE: This routine may be called on a vnode which has not yet been 1208 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1209 * already been reclaimed. 1210 * 1211 * This routine is not responsible for placing us back on the freelist. 1212 * Instead, it happens automatically when the caller releases the VX lock 1213 * (assuming there aren't any other references). 1214 */ 1215 void 1216 vgone(struct vnode *vp) 1217 { 1218 /* 1219 * assert that the VX lock is held. This is an absolute requirement 1220 * now for vgone() to be called. 1221 */ 1222 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1223 1224 /* 1225 * Clean out the filesystem specific data and set the VRECLAIMED 1226 * bit. Also deactivate the vnode if necessary. 1227 */ 1228 vclean(vp, DOCLOSE, curthread); 1229 1230 /* 1231 * Delete from old mount point vnode list, if on one. 1232 */ 1233 if (vp->v_mount != NULL) 1234 insmntque(vp, NULL); 1235 1236 /* 1237 * If special device, remove it from special device alias list 1238 * if it is on one. This should normally only occur if a vnode is 1239 * being revoked as the device should otherwise have been released 1240 * naturally. 1241 */ 1242 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1243 v_release_rdev(vp); 1244 } 1245 1246 /* 1247 * Set us to VBAD 1248 */ 1249 vp->v_type = VBAD; 1250 } 1251 1252 /* 1253 * Lookup a vnode by device number. 1254 */ 1255 int 1256 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp) 1257 { 1258 lwkt_tokref ilock; 1259 struct vnode *vp; 1260 1261 lwkt_gettoken(&ilock, &spechash_token); 1262 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1263 if (type == vp->v_type) { 1264 *vpp = vp; 1265 lwkt_reltoken(&ilock); 1266 return (1); 1267 } 1268 } 1269 lwkt_reltoken(&ilock); 1270 return (0); 1271 } 1272 1273 /* 1274 * Calculate the total number of references to a special device. This 1275 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1276 * an overloaded field. Since udev2dev can now return NODEV, we have 1277 * to check for a NULL v_rdev. 1278 */ 1279 int 1280 count_dev(dev_t dev) 1281 { 1282 lwkt_tokref ilock; 1283 struct vnode *vp; 1284 int count = 0; 1285 1286 if (SLIST_FIRST(&dev->si_hlist)) { 1287 lwkt_gettoken(&ilock, &spechash_token); 1288 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1289 count += vp->v_usecount; 1290 } 1291 lwkt_reltoken(&ilock); 1292 } 1293 return(count); 1294 } 1295 1296 int 1297 count_udev(udev_t udev) 1298 { 1299 dev_t dev; 1300 1301 if ((dev = udev2dev(udev, 0)) == NODEV) 1302 return(0); 1303 return(count_dev(dev)); 1304 } 1305 1306 int 1307 vcount(struct vnode *vp) 1308 { 1309 if (vp->v_rdev == NULL) 1310 return(0); 1311 return(count_dev(vp->v_rdev)); 1312 } 1313 1314 /* 1315 * Print out a description of a vnode. 1316 */ 1317 static char *typename[] = 1318 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1319 1320 void 1321 vprint(char *label, struct vnode *vp) 1322 { 1323 char buf[96]; 1324 1325 if (label != NULL) 1326 printf("%s: %p: ", label, (void *)vp); 1327 else 1328 printf("%p: ", (void *)vp); 1329 printf("type %s, usecount %d, writecount %d, refcount %d,", 1330 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1331 vp->v_holdcnt); 1332 buf[0] = '\0'; 1333 if (vp->v_flag & VROOT) 1334 strcat(buf, "|VROOT"); 1335 if (vp->v_flag & VTEXT) 1336 strcat(buf, "|VTEXT"); 1337 if (vp->v_flag & VSYSTEM) 1338 strcat(buf, "|VSYSTEM"); 1339 if (vp->v_flag & VBWAIT) 1340 strcat(buf, "|VBWAIT"); 1341 if (vp->v_flag & VFREE) 1342 strcat(buf, "|VFREE"); 1343 if (vp->v_flag & VOBJBUF) 1344 strcat(buf, "|VOBJBUF"); 1345 if (buf[0] != '\0') 1346 printf(" flags (%s)", &buf[1]); 1347 if (vp->v_data == NULL) { 1348 printf("\n"); 1349 } else { 1350 printf("\n\t"); 1351 VOP_PRINT(vp); 1352 } 1353 } 1354 1355 #ifdef DDB 1356 #include <ddb/ddb.h> 1357 1358 static int db_show_locked_vnodes(struct mount *mp, void *data); 1359 1360 /* 1361 * List all of the locked vnodes in the system. 1362 * Called when debugging the kernel. 1363 */ 1364 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1365 { 1366 printf("Locked vnodes\n"); 1367 mountlist_scan(db_show_locked_vnodes, NULL, 1368 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1369 } 1370 1371 static int 1372 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1373 { 1374 struct vnode *vp; 1375 1376 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1377 if (VOP_ISLOCKED(vp, NULL)) 1378 vprint((char *)0, vp); 1379 } 1380 return(0); 1381 } 1382 #endif 1383 1384 /* 1385 * Top level filesystem related information gathering. 1386 */ 1387 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1388 1389 static int 1390 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1391 { 1392 int *name = (int *)arg1 - 1; /* XXX */ 1393 u_int namelen = arg2 + 1; /* XXX */ 1394 struct vfsconf *vfsp; 1395 1396 #if 1 || defined(COMPAT_PRELITE2) 1397 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1398 if (namelen == 1) 1399 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1400 #endif 1401 1402 #ifdef notyet 1403 /* all sysctl names at this level are at least name and field */ 1404 if (namelen < 2) 1405 return (ENOTDIR); /* overloaded */ 1406 if (name[0] != VFS_GENERIC) { 1407 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1408 if (vfsp->vfc_typenum == name[0]) 1409 break; 1410 if (vfsp == NULL) 1411 return (EOPNOTSUPP); 1412 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1413 oldp, oldlenp, newp, newlen, p)); 1414 } 1415 #endif 1416 switch (name[1]) { 1417 case VFS_MAXTYPENUM: 1418 if (namelen != 2) 1419 return (ENOTDIR); 1420 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 1421 case VFS_CONF: 1422 if (namelen != 3) 1423 return (ENOTDIR); /* overloaded */ 1424 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 1425 if (vfsp->vfc_typenum == name[2]) 1426 break; 1427 if (vfsp == NULL) 1428 return (EOPNOTSUPP); 1429 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1430 } 1431 return (EOPNOTSUPP); 1432 } 1433 1434 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1435 "Generic filesystem"); 1436 1437 #if 1 || defined(COMPAT_PRELITE2) 1438 1439 static int 1440 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1441 { 1442 int error; 1443 struct vfsconf *vfsp; 1444 struct ovfsconf ovfs; 1445 1446 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 1447 bzero(&ovfs, sizeof(ovfs)); 1448 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1449 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1450 ovfs.vfc_index = vfsp->vfc_typenum; 1451 ovfs.vfc_refcount = vfsp->vfc_refcount; 1452 ovfs.vfc_flags = vfsp->vfc_flags; 1453 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1454 if (error) 1455 return error; 1456 } 1457 return 0; 1458 } 1459 1460 #endif /* 1 || COMPAT_PRELITE2 */ 1461 1462 /* 1463 * Check to see if a filesystem is mounted on a block device. 1464 */ 1465 int 1466 vfs_mountedon(struct vnode *vp) 1467 { 1468 dev_t dev; 1469 1470 if ((dev = vp->v_rdev) == NULL) 1471 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1472 if (dev != NODEV && dev->si_mountpoint) 1473 return (EBUSY); 1474 return (0); 1475 } 1476 1477 /* 1478 * Unmount all filesystems. The list is traversed in reverse order 1479 * of mounting to avoid dependencies. 1480 */ 1481 1482 static int vfs_umountall_callback(struct mount *mp, void *data); 1483 1484 void 1485 vfs_unmountall(void) 1486 { 1487 struct thread *td = curthread; 1488 int count; 1489 1490 if (td->td_proc == NULL) 1491 td = initproc->p_thread; /* XXX XXX use proc0 instead? */ 1492 1493 do { 1494 count = mountlist_scan(vfs_umountall_callback, 1495 &td, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1496 } while (count); 1497 } 1498 1499 static 1500 int 1501 vfs_umountall_callback(struct mount *mp, void *data) 1502 { 1503 struct thread *td = *(struct thread **)data; 1504 int error; 1505 1506 error = dounmount(mp, MNT_FORCE, td); 1507 if (error) { 1508 mountlist_remove(mp); 1509 printf("unmount of filesystem mounted from %s failed (", 1510 mp->mnt_stat.f_mntfromname); 1511 if (error == EBUSY) 1512 printf("BUSY)\n"); 1513 else 1514 printf("%d)\n", error); 1515 } 1516 return(1); 1517 } 1518 1519 /* 1520 * Build hash lists of net addresses and hang them off the mount point. 1521 * Called by ufs_mount() to set up the lists of export addresses. 1522 */ 1523 static int 1524 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1525 struct export_args *argp) 1526 { 1527 struct netcred *np; 1528 struct radix_node_head *rnh; 1529 int i; 1530 struct radix_node *rn; 1531 struct sockaddr *saddr, *smask = 0; 1532 struct domain *dom; 1533 int error; 1534 1535 if (argp->ex_addrlen == 0) { 1536 if (mp->mnt_flag & MNT_DEFEXPORTED) 1537 return (EPERM); 1538 np = &nep->ne_defexported; 1539 np->netc_exflags = argp->ex_flags; 1540 np->netc_anon = argp->ex_anon; 1541 np->netc_anon.cr_ref = 1; 1542 mp->mnt_flag |= MNT_DEFEXPORTED; 1543 return (0); 1544 } 1545 1546 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1547 return (EINVAL); 1548 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1549 return (EINVAL); 1550 1551 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1552 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 1553 bzero((caddr_t) np, i); 1554 saddr = (struct sockaddr *) (np + 1); 1555 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1556 goto out; 1557 if (saddr->sa_len > argp->ex_addrlen) 1558 saddr->sa_len = argp->ex_addrlen; 1559 if (argp->ex_masklen) { 1560 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1561 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1562 if (error) 1563 goto out; 1564 if (smask->sa_len > argp->ex_masklen) 1565 smask->sa_len = argp->ex_masklen; 1566 } 1567 i = saddr->sa_family; 1568 if ((rnh = nep->ne_rtable[i]) == 0) { 1569 /* 1570 * Seems silly to initialize every AF when most are not used, 1571 * do so on demand here 1572 */ 1573 SLIST_FOREACH(dom, &domains, dom_next) 1574 if (dom->dom_family == i && dom->dom_rtattach) { 1575 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1576 dom->dom_rtoffset); 1577 break; 1578 } 1579 if ((rnh = nep->ne_rtable[i]) == 0) { 1580 error = ENOBUFS; 1581 goto out; 1582 } 1583 } 1584 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1585 np->netc_rnodes); 1586 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1587 error = EPERM; 1588 goto out; 1589 } 1590 np->netc_exflags = argp->ex_flags; 1591 np->netc_anon = argp->ex_anon; 1592 np->netc_anon.cr_ref = 1; 1593 return (0); 1594 out: 1595 free(np, M_NETADDR); 1596 return (error); 1597 } 1598 1599 /* ARGSUSED */ 1600 static int 1601 vfs_free_netcred(struct radix_node *rn, void *w) 1602 { 1603 struct radix_node_head *rnh = (struct radix_node_head *) w; 1604 1605 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1606 free((caddr_t) rn, M_NETADDR); 1607 return (0); 1608 } 1609 1610 /* 1611 * Free the net address hash lists that are hanging off the mount points. 1612 */ 1613 static void 1614 vfs_free_addrlist(struct netexport *nep) 1615 { 1616 int i; 1617 struct radix_node_head *rnh; 1618 1619 for (i = 0; i <= AF_MAX; i++) 1620 if ((rnh = nep->ne_rtable[i])) { 1621 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1622 (caddr_t) rnh); 1623 free((caddr_t) rnh, M_RTABLE); 1624 nep->ne_rtable[i] = 0; 1625 } 1626 } 1627 1628 int 1629 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1630 { 1631 int error; 1632 1633 if (argp->ex_flags & MNT_DELEXPORT) { 1634 if (mp->mnt_flag & MNT_EXPUBLIC) { 1635 vfs_setpublicfs(NULL, NULL, NULL); 1636 mp->mnt_flag &= ~MNT_EXPUBLIC; 1637 } 1638 vfs_free_addrlist(nep); 1639 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1640 } 1641 if (argp->ex_flags & MNT_EXPORTED) { 1642 if (argp->ex_flags & MNT_EXPUBLIC) { 1643 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 1644 return (error); 1645 mp->mnt_flag |= MNT_EXPUBLIC; 1646 } 1647 if ((error = vfs_hang_addrlist(mp, nep, argp))) 1648 return (error); 1649 mp->mnt_flag |= MNT_EXPORTED; 1650 } 1651 return (0); 1652 } 1653 1654 1655 /* 1656 * Set the publicly exported filesystem (WebNFS). Currently, only 1657 * one public filesystem is possible in the spec (RFC 2054 and 2055) 1658 */ 1659 int 1660 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 1661 struct export_args *argp) 1662 { 1663 int error; 1664 struct vnode *rvp; 1665 char *cp; 1666 1667 /* 1668 * mp == NULL -> invalidate the current info, the FS is 1669 * no longer exported. May be called from either vfs_export 1670 * or unmount, so check if it hasn't already been done. 1671 */ 1672 if (mp == NULL) { 1673 if (nfs_pub.np_valid) { 1674 nfs_pub.np_valid = 0; 1675 if (nfs_pub.np_index != NULL) { 1676 FREE(nfs_pub.np_index, M_TEMP); 1677 nfs_pub.np_index = NULL; 1678 } 1679 } 1680 return (0); 1681 } 1682 1683 /* 1684 * Only one allowed at a time. 1685 */ 1686 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 1687 return (EBUSY); 1688 1689 /* 1690 * Get real filehandle for root of exported FS. 1691 */ 1692 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 1693 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 1694 1695 if ((error = VFS_ROOT(mp, &rvp))) 1696 return (error); 1697 1698 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 1699 return (error); 1700 1701 vput(rvp); 1702 1703 /* 1704 * If an indexfile was specified, pull it in. 1705 */ 1706 if (argp->ex_indexfile != NULL) { 1707 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 1708 M_WAITOK); 1709 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 1710 MAXNAMLEN, (size_t *)0); 1711 if (!error) { 1712 /* 1713 * Check for illegal filenames. 1714 */ 1715 for (cp = nfs_pub.np_index; *cp; cp++) { 1716 if (*cp == '/') { 1717 error = EINVAL; 1718 break; 1719 } 1720 } 1721 } 1722 if (error) { 1723 FREE(nfs_pub.np_index, M_TEMP); 1724 return (error); 1725 } 1726 } 1727 1728 nfs_pub.np_mount = mp; 1729 nfs_pub.np_valid = 1; 1730 return (0); 1731 } 1732 1733 struct netcred * 1734 vfs_export_lookup(struct mount *mp, struct netexport *nep, 1735 struct sockaddr *nam) 1736 { 1737 struct netcred *np; 1738 struct radix_node_head *rnh; 1739 struct sockaddr *saddr; 1740 1741 np = NULL; 1742 if (mp->mnt_flag & MNT_EXPORTED) { 1743 /* 1744 * Lookup in the export list first. 1745 */ 1746 if (nam != NULL) { 1747 saddr = nam; 1748 rnh = nep->ne_rtable[saddr->sa_family]; 1749 if (rnh != NULL) { 1750 np = (struct netcred *) 1751 (*rnh->rnh_matchaddr)((char *)saddr, 1752 rnh); 1753 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 1754 np = NULL; 1755 } 1756 } 1757 /* 1758 * If no address match, use the default if it exists. 1759 */ 1760 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1761 np = &nep->ne_defexported; 1762 } 1763 return (np); 1764 } 1765 1766 /* 1767 * perform msync on all vnodes under a mount point. The mount point must 1768 * be locked. This code is also responsible for lazy-freeing unreferenced 1769 * vnodes whos VM objects no longer contain pages. 1770 * 1771 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 1772 */ 1773 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 1774 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 1775 1776 void 1777 vfs_msync(struct mount *mp, int flags) 1778 { 1779 vmntvnodescan(mp, VMSC_REFVP, vfs_msync_scan1, vfs_msync_scan2, 1780 (void *)flags); 1781 } 1782 1783 /* 1784 * scan1 is a fast pre-check. There could be hundreds of thousands of 1785 * vnodes, we cannot afford to do anything heavy weight until we have a 1786 * fairly good indication that there is work to do. 1787 */ 1788 static 1789 int 1790 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 1791 { 1792 int flags = (int)data; 1793 1794 if ((vp->v_flag & VRECLAIMED) == 0) { 1795 if (vshouldfree(vp, 0)) 1796 return(0); /* call scan2 */ 1797 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1798 (vp->v_flag & VOBJDIRTY) && 1799 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1800 return(0); /* call scan2 */ 1801 } 1802 } 1803 1804 /* 1805 * do not call scan2, continue the loop 1806 */ 1807 return(-1); 1808 } 1809 1810 static 1811 int 1812 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 1813 { 1814 vm_object_t obj; 1815 int flags = (int)data; 1816 1817 if (vp->v_flag & VRECLAIMED) 1818 return(0); 1819 1820 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 1821 (vp->v_flag & VOBJDIRTY) && 1822 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 1823 if (VOP_GETVOBJECT(vp, &obj) == 0) { 1824 vm_object_page_clean(obj, 0, 0, 1825 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 1826 } 1827 } 1828 return(0); 1829 } 1830 1831 /* 1832 * Create the VM object needed for VMIO and mmap support. This 1833 * is done for all VREG files in the system. Some filesystems might 1834 * afford the additional metadata buffering capability of the 1835 * VMIO code by making the device node be VMIO mode also. 1836 * 1837 * vp must be locked when vfs_object_create is called. 1838 */ 1839 int 1840 vfs_object_create(struct vnode *vp, struct thread *td) 1841 { 1842 return (VOP_CREATEVOBJECT(vp, td)); 1843 } 1844 1845 /* 1846 * Record a process's interest in events which might happen to 1847 * a vnode. Because poll uses the historic select-style interface 1848 * internally, this routine serves as both the ``check for any 1849 * pending events'' and the ``record my interest in future events'' 1850 * functions. (These are done together, while the lock is held, 1851 * to avoid race conditions.) 1852 */ 1853 int 1854 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 1855 { 1856 lwkt_tokref ilock; 1857 1858 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1859 if (vp->v_pollinfo.vpi_revents & events) { 1860 /* 1861 * This leaves events we are not interested 1862 * in available for the other process which 1863 * which presumably had requested them 1864 * (otherwise they would never have been 1865 * recorded). 1866 */ 1867 events &= vp->v_pollinfo.vpi_revents; 1868 vp->v_pollinfo.vpi_revents &= ~events; 1869 1870 lwkt_reltoken(&ilock); 1871 return events; 1872 } 1873 vp->v_pollinfo.vpi_events |= events; 1874 selrecord(td, &vp->v_pollinfo.vpi_selinfo); 1875 lwkt_reltoken(&ilock); 1876 return 0; 1877 } 1878 1879 /* 1880 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 1881 * it is possible for us to miss an event due to race conditions, but 1882 * that condition is expected to be rare, so for the moment it is the 1883 * preferred interface. 1884 */ 1885 void 1886 vn_pollevent(struct vnode *vp, int events) 1887 { 1888 lwkt_tokref ilock; 1889 1890 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1891 if (vp->v_pollinfo.vpi_events & events) { 1892 /* 1893 * We clear vpi_events so that we don't 1894 * call selwakeup() twice if two events are 1895 * posted before the polling process(es) is 1896 * awakened. This also ensures that we take at 1897 * most one selwakeup() if the polling process 1898 * is no longer interested. However, it does 1899 * mean that only one event can be noticed at 1900 * a time. (Perhaps we should only clear those 1901 * event bits which we note?) XXX 1902 */ 1903 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 1904 vp->v_pollinfo.vpi_revents |= events; 1905 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1906 } 1907 lwkt_reltoken(&ilock); 1908 } 1909 1910 /* 1911 * Wake up anyone polling on vp because it is being revoked. 1912 * This depends on dead_poll() returning POLLHUP for correct 1913 * behavior. 1914 */ 1915 void 1916 vn_pollgone(struct vnode *vp) 1917 { 1918 lwkt_tokref ilock; 1919 1920 lwkt_gettoken(&ilock, &vp->v_pollinfo.vpi_token); 1921 if (vp->v_pollinfo.vpi_events) { 1922 vp->v_pollinfo.vpi_events = 0; 1923 selwakeup(&vp->v_pollinfo.vpi_selinfo); 1924 } 1925 lwkt_reltoken(&ilock); 1926 } 1927 1928 /* 1929 * extract the dev_t from a VBLK or VCHR. The vnode must have been opened 1930 * (or v_rdev might be NULL). 1931 */ 1932 dev_t 1933 vn_todev(struct vnode *vp) 1934 { 1935 if (vp->v_type != VBLK && vp->v_type != VCHR) 1936 return (NODEV); 1937 KKASSERT(vp->v_rdev != NULL); 1938 return (vp->v_rdev); 1939 } 1940 1941 /* 1942 * Check if vnode represents a disk device. The vnode does not need to be 1943 * opened. 1944 */ 1945 int 1946 vn_isdisk(struct vnode *vp, int *errp) 1947 { 1948 dev_t dev; 1949 1950 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1951 if (errp != NULL) 1952 *errp = ENOTBLK; 1953 return (0); 1954 } 1955 1956 if ((dev = vp->v_rdev) == NULL) 1957 dev = udev2dev(vp->v_udev, (vp->v_type == VBLK)); 1958 if (dev == NULL || dev == NODEV) { 1959 if (errp != NULL) 1960 *errp = ENXIO; 1961 return (0); 1962 } 1963 if (dev_is_good(dev) == 0) { 1964 if (errp != NULL) 1965 *errp = ENXIO; 1966 return (0); 1967 } 1968 if ((dev_dflags(dev) & D_DISK) == 0) { 1969 if (errp != NULL) 1970 *errp = ENOTBLK; 1971 return (0); 1972 } 1973 if (errp != NULL) 1974 *errp = 0; 1975 return (1); 1976 } 1977 1978 #ifdef DEBUG_VFS_LOCKS 1979 1980 void 1981 assert_vop_locked(struct vnode *vp, const char *str) 1982 { 1983 if (vp && IS_LOCKING_VFS(vp) && !VOP_ISLOCKED(vp, NULL)) { 1984 panic("%s: %p is not locked shared but should be", str, vp); 1985 } 1986 } 1987 1988 void 1989 assert_vop_unlocked(struct vnode *vp, const char *str) 1990 { 1991 if (vp && IS_LOCKING_VFS(vp)) { 1992 if (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) { 1993 panic("%s: %p is locked but should not be", str, vp); 1994 } 1995 } 1996 } 1997 1998 #endif 1999