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. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 35 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 36 */ 37 38 /* 39 * External virtual filesystem routines 40 */ 41 #include "opt_ddb.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/buf.h> 48 #include <sys/conf.h> 49 #include <sys/dirent.h> 50 #include <sys/eventhandler.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/mount.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/reboot.h> 61 #include <sys/socket.h> 62 #include <sys/stat.h> 63 #include <sys/sysctl.h> 64 #include <sys/syslog.h> 65 #include <sys/unistd.h> 66 #include <sys/vmmeter.h> 67 #include <sys/vnode.h> 68 69 #include <machine/limits.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_extern.h> 74 #include <vm/vm_kern.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_pager.h> 79 #include <vm/vnode_pager.h> 80 #include <vm/vm_zone.h> 81 82 #include <sys/buf2.h> 83 #include <sys/thread2.h> 84 #include <sys/sysref2.h> 85 #include <sys/mplock2.h> 86 87 #include <netinet/in.h> 88 89 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 90 91 int numvnodes; 92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 93 "Number of vnodes allocated"); 94 int verbose_reclaims; 95 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0, 96 "Output filename of reclaimed vnode(s)"); 97 98 enum vtype iftovt_tab[16] = { 99 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 100 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 101 }; 102 int vttoif_tab[9] = { 103 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 104 S_IFSOCK, S_IFIFO, S_IFMT, 105 }; 106 107 static int reassignbufcalls; 108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 109 0, "Number of times buffers have been reassigned to the proper list"); 110 111 static int check_buf_overlap = 2; /* invasive check */ 112 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap, 113 0, "Enable overlapping buffer checks"); 114 115 int nfs_mount_type = -1; 116 static struct lwkt_token spechash_token; 117 struct nfs_public nfs_pub; /* publicly exported FS */ 118 119 int desiredvnodes; 120 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 121 &desiredvnodes, 0, "Maximum number of vnodes"); 122 123 static struct radix_node_head *vfs_create_addrlist_af(int af, 124 struct netexport *nep); 125 static void vfs_free_addrlist (struct netexport *nep); 126 static int vfs_free_netcred (struct radix_node *rn, void *w); 127 static void vfs_free_addrlist_af (struct radix_node_head **prnh); 128 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 129 const struct export_args *argp); 130 131 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 132 133 /* 134 * Red black tree functions 135 */ 136 static int rb_buf_compare(struct buf *b1, struct buf *b2); 137 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 138 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 139 140 static int 141 rb_buf_compare(struct buf *b1, struct buf *b2) 142 { 143 if (b1->b_loffset < b2->b_loffset) 144 return(-1); 145 if (b1->b_loffset > b2->b_loffset) 146 return(1); 147 return(0); 148 } 149 150 /* 151 * Initialize the vnode management data structures. 152 * 153 * Called from vfsinit() 154 */ 155 void 156 vfs_subr_init(void) 157 { 158 int factor1; 159 int factor2; 160 161 /* 162 * Desiredvnodes is kern.maxvnodes. We want to scale it 163 * according to available system memory but we may also have 164 * to limit it based on available KVM, which is capped on 32 bit 165 * systems, to ~80K vnodes or so. 166 * 167 * WARNING! For machines with 64-256M of ram we have to be sure 168 * that the default limit scales down well due to HAMMER 169 * taking up significantly more memory per-vnode vs UFS. 170 * We want around ~5800 on a 128M machine. 171 */ 172 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode)); 173 factor2 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode)); 174 desiredvnodes = 175 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 176 KvaSize / factor2); 177 desiredvnodes = imax(desiredvnodes, maxproc * 8); 178 179 lwkt_token_init(&spechash_token, "spechash"); 180 } 181 182 /* 183 * Knob to control the precision of file timestamps: 184 * 185 * 0 = seconds only; nanoseconds zeroed. 186 * 1 = seconds and nanoseconds, accurate within 1/HZ. 187 * 2 = seconds and nanoseconds, truncated to microseconds. 188 * >=3 = seconds and nanoseconds, maximum precision. 189 */ 190 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 191 192 static int timestamp_precision = TSP_SEC; 193 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 194 ×tamp_precision, 0, "Precision of file timestamps"); 195 196 /* 197 * Get a current timestamp. 198 * 199 * MPSAFE 200 */ 201 void 202 vfs_timestamp(struct timespec *tsp) 203 { 204 struct timeval tv; 205 206 switch (timestamp_precision) { 207 case TSP_SEC: 208 tsp->tv_sec = time_second; 209 tsp->tv_nsec = 0; 210 break; 211 case TSP_HZ: 212 getnanotime(tsp); 213 break; 214 case TSP_USEC: 215 microtime(&tv); 216 TIMEVAL_TO_TIMESPEC(&tv, tsp); 217 break; 218 case TSP_NSEC: 219 default: 220 nanotime(tsp); 221 break; 222 } 223 } 224 225 /* 226 * Set vnode attributes to VNOVAL 227 */ 228 void 229 vattr_null(struct vattr *vap) 230 { 231 vap->va_type = VNON; 232 vap->va_size = VNOVAL; 233 vap->va_bytes = VNOVAL; 234 vap->va_mode = VNOVAL; 235 vap->va_nlink = VNOVAL; 236 vap->va_uid = VNOVAL; 237 vap->va_gid = VNOVAL; 238 vap->va_fsid = VNOVAL; 239 vap->va_fileid = VNOVAL; 240 vap->va_blocksize = VNOVAL; 241 vap->va_rmajor = VNOVAL; 242 vap->va_rminor = VNOVAL; 243 vap->va_atime.tv_sec = VNOVAL; 244 vap->va_atime.tv_nsec = VNOVAL; 245 vap->va_mtime.tv_sec = VNOVAL; 246 vap->va_mtime.tv_nsec = VNOVAL; 247 vap->va_ctime.tv_sec = VNOVAL; 248 vap->va_ctime.tv_nsec = VNOVAL; 249 vap->va_flags = VNOVAL; 250 vap->va_gen = VNOVAL; 251 vap->va_vaflags = 0; 252 /* va_*_uuid fields are only valid if related flags are set */ 253 } 254 255 /* 256 * Flush out and invalidate all buffers associated with a vnode. 257 * 258 * vp must be locked. 259 */ 260 static int vinvalbuf_bp(struct buf *bp, void *data); 261 262 struct vinvalbuf_bp_info { 263 struct vnode *vp; 264 int slptimeo; 265 int lkflags; 266 int flags; 267 int clean; 268 }; 269 270 int 271 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 272 { 273 struct vinvalbuf_bp_info info; 274 vm_object_t object; 275 int error; 276 277 lwkt_gettoken(&vp->v_token); 278 279 /* 280 * If we are being asked to save, call fsync to ensure that the inode 281 * is updated. 282 */ 283 if (flags & V_SAVE) { 284 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 285 if (error) 286 goto done; 287 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 288 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 289 goto done; 290 #if 0 291 /* 292 * Dirty bufs may be left or generated via races 293 * in circumstances where vinvalbuf() is called on 294 * a vnode not undergoing reclamation. Only 295 * panic if we are trying to reclaim the vnode. 296 */ 297 if ((vp->v_flag & VRECLAIMED) && 298 (bio_track_active(&vp->v_track_write) || 299 !RB_EMPTY(&vp->v_rbdirty_tree))) { 300 panic("vinvalbuf: dirty bufs"); 301 } 302 #endif 303 } 304 } 305 info.slptimeo = slptimeo; 306 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 307 if (slpflag & PCATCH) 308 info.lkflags |= LK_PCATCH; 309 info.flags = flags; 310 info.vp = vp; 311 312 /* 313 * Flush the buffer cache until nothing is left, wait for all I/O 314 * to complete. At least one pass is required. We might block 315 * in the pip code so we have to re-check. Order is important. 316 */ 317 do { 318 /* 319 * Flush buffer cache 320 */ 321 if (!RB_EMPTY(&vp->v_rbclean_tree)) { 322 info.clean = 1; 323 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 324 NULL, vinvalbuf_bp, &info); 325 } 326 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 327 info.clean = 0; 328 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 329 NULL, vinvalbuf_bp, &info); 330 } 331 332 /* 333 * Wait for I/O completion. 334 */ 335 bio_track_wait(&vp->v_track_write, 0, 0); 336 if ((object = vp->v_object) != NULL) 337 refcount_wait(&object->paging_in_progress, "vnvlbx"); 338 } while (bio_track_active(&vp->v_track_write) || 339 !RB_EMPTY(&vp->v_rbclean_tree) || 340 !RB_EMPTY(&vp->v_rbdirty_tree)); 341 342 /* 343 * Destroy the copy in the VM cache, too. 344 */ 345 if ((object = vp->v_object) != NULL) { 346 vm_object_page_remove(object, 0, 0, 347 (flags & V_SAVE) ? TRUE : FALSE); 348 } 349 350 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 351 panic("vinvalbuf: flush failed"); 352 if (!RB_EMPTY(&vp->v_rbhash_tree)) 353 panic("vinvalbuf: flush failed, buffers still present"); 354 error = 0; 355 done: 356 lwkt_reltoken(&vp->v_token); 357 return (error); 358 } 359 360 static int 361 vinvalbuf_bp(struct buf *bp, void *data) 362 { 363 struct vinvalbuf_bp_info *info = data; 364 int error; 365 366 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 367 atomic_add_int(&bp->b_refs, 1); 368 error = BUF_TIMELOCK(bp, info->lkflags, 369 "vinvalbuf", info->slptimeo); 370 atomic_subtract_int(&bp->b_refs, 1); 371 if (error == 0) { 372 BUF_UNLOCK(bp); 373 error = ENOLCK; 374 } 375 if (error == ENOLCK) 376 return(0); 377 return (-error); 378 } 379 KKASSERT(bp->b_vp == info->vp); 380 381 /* 382 * Must check clean/dirty status after successfully locking as 383 * it may race. 384 */ 385 if ((info->clean && (bp->b_flags & B_DELWRI)) || 386 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 387 BUF_UNLOCK(bp); 388 return(0); 389 } 390 391 /* 392 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 393 * check. This code will write out the buffer, period. 394 */ 395 bremfree(bp); 396 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 397 (info->flags & V_SAVE)) { 398 cluster_awrite(bp); 399 } else if (info->flags & V_SAVE) { 400 /* 401 * Cannot set B_NOCACHE on a clean buffer as this will 402 * destroy the VM backing store which might actually 403 * be dirty (and unsynchronized). 404 */ 405 bp->b_flags |= (B_INVAL | B_RELBUF); 406 brelse(bp); 407 } else { 408 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 409 brelse(bp); 410 } 411 return(0); 412 } 413 414 /* 415 * Truncate a file's buffer and pages to a specified length. This 416 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 417 * sync activity. 418 * 419 * The vnode must be locked. 420 */ 421 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 422 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 423 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 424 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 425 426 struct vtruncbuf_info { 427 struct vnode *vp; 428 off_t truncloffset; 429 int clean; 430 }; 431 432 int 433 vtruncbuf(struct vnode *vp, off_t length, int blksize) 434 { 435 struct vtruncbuf_info info; 436 const char *filename; 437 int count; 438 439 /* 440 * Round up to the *next* block, then destroy the buffers in question. 441 * Since we are only removing some of the buffers we must rely on the 442 * scan count to determine whether a loop is necessary. 443 */ 444 if ((count = (int)(length % blksize)) != 0) 445 info.truncloffset = length + (blksize - count); 446 else 447 info.truncloffset = length; 448 info.vp = vp; 449 450 lwkt_gettoken(&vp->v_token); 451 do { 452 info.clean = 1; 453 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 454 vtruncbuf_bp_trunc_cmp, 455 vtruncbuf_bp_trunc, &info); 456 info.clean = 0; 457 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 458 vtruncbuf_bp_trunc_cmp, 459 vtruncbuf_bp_trunc, &info); 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, &info); 473 } while (count); 474 } 475 476 /* 477 * Clean out any left over VM backing store. 478 * 479 * It is possible to have in-progress I/O from buffers that were 480 * not part of the truncation. This should not happen if we 481 * are truncating to 0-length. 482 */ 483 vnode_pager_setsize(vp, length); 484 bio_track_wait(&vp->v_track_write, 0, 0); 485 486 /* 487 * Debugging only 488 */ 489 spin_lock(&vp->v_spin); 490 filename = TAILQ_FIRST(&vp->v_namecache) ? 491 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 492 spin_unlock(&vp->v_spin); 493 494 /* 495 * Make sure no buffers were instantiated while we were trying 496 * to clean out the remaining VM pages. This could occur due 497 * to busy dirty VM pages being flushed out to disk. 498 */ 499 do { 500 info.clean = 1; 501 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 502 vtruncbuf_bp_trunc_cmp, 503 vtruncbuf_bp_trunc, &info); 504 info.clean = 0; 505 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 506 vtruncbuf_bp_trunc_cmp, 507 vtruncbuf_bp_trunc, &info); 508 if (count) { 509 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 510 "left over buffers in %s\n", count, filename); 511 } 512 } while(count); 513 514 lwkt_reltoken(&vp->v_token); 515 516 return (0); 517 } 518 519 /* 520 * The callback buffer is beyond the new file EOF and must be destroyed. 521 * Note that the compare function must conform to the RB_SCAN's requirements. 522 */ 523 static 524 int 525 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 526 { 527 struct vtruncbuf_info *info = data; 528 529 if (bp->b_loffset >= info->truncloffset) 530 return(0); 531 return(-1); 532 } 533 534 static 535 int 536 vtruncbuf_bp_trunc(struct buf *bp, void *data) 537 { 538 struct vtruncbuf_info *info = data; 539 540 /* 541 * Do not try to use a buffer we cannot immediately lock, but sleep 542 * anyway to prevent a livelock. The code will loop until all buffers 543 * can be acted upon. 544 * 545 * We must always revalidate the buffer after locking it to deal 546 * with MP races. 547 */ 548 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 549 atomic_add_int(&bp->b_refs, 1); 550 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 551 BUF_UNLOCK(bp); 552 atomic_subtract_int(&bp->b_refs, 1); 553 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 554 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 555 bp->b_vp != info->vp || 556 vtruncbuf_bp_trunc_cmp(bp, data)) { 557 BUF_UNLOCK(bp); 558 } else { 559 bremfree(bp); 560 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 561 brelse(bp); 562 } 563 return(1); 564 } 565 566 /* 567 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 568 * blocks (with a negative loffset) are scanned. 569 * Note that the compare function must conform to the RB_SCAN's requirements. 570 */ 571 static int 572 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 573 { 574 if (bp->b_loffset < 0) 575 return(0); 576 return(1); 577 } 578 579 static int 580 vtruncbuf_bp_metasync(struct buf *bp, void *data) 581 { 582 struct vtruncbuf_info *info = data; 583 584 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 585 atomic_add_int(&bp->b_refs, 1); 586 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 587 BUF_UNLOCK(bp); 588 atomic_subtract_int(&bp->b_refs, 1); 589 } else if ((bp->b_flags & B_DELWRI) == 0 || 590 bp->b_vp != info->vp || 591 vtruncbuf_bp_metasync_cmp(bp, data)) { 592 BUF_UNLOCK(bp); 593 } else { 594 bremfree(bp); 595 if (bp->b_vp == info->vp) 596 bawrite(bp); 597 else 598 bwrite(bp); 599 } 600 return(1); 601 } 602 603 /* 604 * vfsync - implements a multipass fsync on a file which understands 605 * dependancies and meta-data. The passed vnode must be locked. The 606 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 607 * 608 * When fsyncing data asynchronously just do one consolidated pass starting 609 * with the most negative block number. This may not get all the data due 610 * to dependancies. 611 * 612 * When fsyncing data synchronously do a data pass, then a metadata pass, 613 * then do additional data+metadata passes to try to get all the data out. 614 * 615 * Caller must ref the vnode but does not have to lock it. 616 */ 617 static int vfsync_wait_output(struct vnode *vp, 618 int (*waitoutput)(struct vnode *, struct thread *)); 619 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 620 static int vfsync_data_only_cmp(struct buf *bp, void *data); 621 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 622 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 623 static int vfsync_bp(struct buf *bp, void *data); 624 625 struct vfsync_info { 626 struct vnode *vp; 627 int synchronous; 628 int syncdeps; 629 int lazycount; 630 int lazylimit; 631 int skippedbufs; 632 int (*checkdef)(struct buf *); 633 int (*cmpfunc)(struct buf *, void *); 634 }; 635 636 int 637 vfsync(struct vnode *vp, int waitfor, int passes, 638 int (*checkdef)(struct buf *), 639 int (*waitoutput)(struct vnode *, struct thread *)) 640 { 641 struct vfsync_info info; 642 int error; 643 644 bzero(&info, sizeof(info)); 645 info.vp = vp; 646 if ((info.checkdef = checkdef) == NULL) 647 info.syncdeps = 1; 648 649 lwkt_gettoken(&vp->v_token); 650 651 switch(waitfor) { 652 case MNT_LAZY | MNT_NOWAIT: 653 case MNT_LAZY: 654 /* 655 * Lazy (filesystem syncer typ) Asynchronous plus limit the 656 * number of data (not meta) pages we try to flush to 1MB. 657 * A non-zero return means that lazy limit was reached. 658 */ 659 info.lazylimit = 1024 * 1024; 660 info.syncdeps = 1; 661 info.cmpfunc = vfsync_lazy_range_cmp; 662 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 663 vfsync_lazy_range_cmp, vfsync_bp, &info); 664 info.cmpfunc = vfsync_meta_only_cmp; 665 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 666 vfsync_meta_only_cmp, vfsync_bp, &info); 667 if (error == 0) 668 vp->v_lazyw = 0; 669 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 670 vn_syncer_add(vp, 1); 671 error = 0; 672 break; 673 case MNT_NOWAIT: 674 /* 675 * Asynchronous. Do a data-only pass and a meta-only pass. 676 */ 677 info.syncdeps = 1; 678 info.cmpfunc = vfsync_data_only_cmp; 679 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 680 vfsync_bp, &info); 681 info.cmpfunc = vfsync_meta_only_cmp; 682 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 683 vfsync_bp, &info); 684 error = 0; 685 break; 686 default: 687 /* 688 * Synchronous. Do a data-only pass, then a meta-data+data 689 * pass, then additional integrated passes to try to get 690 * all the dependancies flushed. 691 */ 692 info.cmpfunc = vfsync_data_only_cmp; 693 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 694 vfsync_bp, &info); 695 error = vfsync_wait_output(vp, waitoutput); 696 if (error == 0) { 697 info.skippedbufs = 0; 698 info.cmpfunc = vfsync_dummy_cmp; 699 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 700 vfsync_bp, &info); 701 error = vfsync_wait_output(vp, waitoutput); 702 if (info.skippedbufs) { 703 kprintf("Warning: vfsync skipped %d dirty " 704 "bufs in pass2!\n", info.skippedbufs); 705 } 706 } 707 while (error == 0 && passes > 0 && 708 !RB_EMPTY(&vp->v_rbdirty_tree) 709 ) { 710 if (--passes == 0) { 711 info.synchronous = 1; 712 info.syncdeps = 1; 713 } 714 info.cmpfunc = vfsync_dummy_cmp; 715 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 716 vfsync_bp, &info); 717 if (error < 0) 718 error = -error; 719 info.syncdeps = 1; 720 if (error == 0) 721 error = vfsync_wait_output(vp, waitoutput); 722 } 723 break; 724 } 725 lwkt_reltoken(&vp->v_token); 726 return(error); 727 } 728 729 static int 730 vfsync_wait_output(struct vnode *vp, 731 int (*waitoutput)(struct vnode *, struct thread *)) 732 { 733 int error; 734 735 error = bio_track_wait(&vp->v_track_write, 0, 0); 736 if (waitoutput) 737 error = waitoutput(vp, curthread); 738 return(error); 739 } 740 741 static int 742 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 743 { 744 return(0); 745 } 746 747 static int 748 vfsync_data_only_cmp(struct buf *bp, void *data) 749 { 750 if (bp->b_loffset < 0) 751 return(-1); 752 return(0); 753 } 754 755 static int 756 vfsync_meta_only_cmp(struct buf *bp, void *data) 757 { 758 if (bp->b_loffset < 0) 759 return(0); 760 return(1); 761 } 762 763 static int 764 vfsync_lazy_range_cmp(struct buf *bp, void *data) 765 { 766 struct vfsync_info *info = data; 767 768 if (bp->b_loffset < info->vp->v_lazyw) 769 return(-1); 770 return(0); 771 } 772 773 static int 774 vfsync_bp(struct buf *bp, void *data) 775 { 776 struct vfsync_info *info = data; 777 struct vnode *vp = info->vp; 778 int error; 779 780 /* 781 * Ignore buffers that we cannot immediately lock. 782 */ 783 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 784 ++info->skippedbufs; 785 return(0); 786 } 787 788 /* 789 * We must revalidate the buffer after locking. 790 */ 791 if ((bp->b_flags & B_DELWRI) == 0 || 792 bp->b_vp != info->vp || 793 info->cmpfunc(bp, data)) { 794 BUF_UNLOCK(bp); 795 return(0); 796 } 797 798 /* 799 * If syncdeps is not set we do not try to write buffers which have 800 * dependancies. 801 */ 802 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 803 BUF_UNLOCK(bp); 804 return(0); 805 } 806 807 /* 808 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 809 * has been written but an additional handshake with the device 810 * is required before we can dispose of the buffer. We have no idea 811 * how to do this so we have to skip these buffers. 812 */ 813 if (bp->b_flags & B_NEEDCOMMIT) { 814 BUF_UNLOCK(bp); 815 return(0); 816 } 817 818 /* 819 * Ask bioops if it is ok to sync. If not the VFS may have 820 * set B_LOCKED so we have to cycle the buffer. 821 */ 822 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 823 bremfree(bp); 824 brelse(bp); 825 return(0); 826 } 827 828 if (info->synchronous) { 829 /* 830 * Synchronous flushing. An error may be returned. 831 */ 832 bremfree(bp); 833 error = bwrite(bp); 834 } else { 835 /* 836 * Asynchronous flushing. A negative return value simply 837 * stops the scan and is not considered an error. We use 838 * this to support limited MNT_LAZY flushes. 839 */ 840 vp->v_lazyw = bp->b_loffset; 841 bremfree(bp); 842 info->lazycount += cluster_awrite(bp); 843 waitrunningbufspace(); 844 vm_wait_nominal(); 845 if (info->lazylimit && info->lazycount >= info->lazylimit) 846 error = 1; 847 else 848 error = 0; 849 } 850 return(-error); 851 } 852 853 /* 854 * Associate a buffer with a vnode. 855 * 856 * MPSAFE 857 */ 858 int 859 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 860 { 861 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 862 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 863 864 /* 865 * Insert onto list for new vnode. 866 */ 867 lwkt_gettoken(&vp->v_token); 868 869 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 870 lwkt_reltoken(&vp->v_token); 871 return (EEXIST); 872 } 873 874 /* 875 * Diagnostics (mainly for HAMMER debugging). Check for 876 * overlapping buffers. 877 */ 878 if (check_buf_overlap) { 879 struct buf *bx; 880 bx = buf_rb_hash_RB_PREV(bp); 881 if (bx) { 882 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 883 kprintf("bgetvp: overlapl %016jx/%d %016jx " 884 "bx %p bp %p\n", 885 (intmax_t)bx->b_loffset, 886 bx->b_bufsize, 887 (intmax_t)bp->b_loffset, 888 bx, bp); 889 if (check_buf_overlap > 1) 890 panic("bgetvp - overlapping buffer"); 891 } 892 } 893 bx = buf_rb_hash_RB_NEXT(bp); 894 if (bx) { 895 if (bp->b_loffset + testsize > bx->b_loffset) { 896 kprintf("bgetvp: overlapr %016jx/%d %016jx " 897 "bp %p bx %p\n", 898 (intmax_t)bp->b_loffset, 899 testsize, 900 (intmax_t)bx->b_loffset, 901 bp, bx); 902 if (check_buf_overlap > 1) 903 panic("bgetvp - overlapping buffer"); 904 } 905 } 906 } 907 bp->b_vp = vp; 908 bp->b_flags |= B_HASHED; 909 bp->b_flags |= B_VNCLEAN; 910 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 911 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 912 /*vhold(vp);*/ 913 lwkt_reltoken(&vp->v_token); 914 return(0); 915 } 916 917 /* 918 * Disassociate a buffer from a vnode. 919 * 920 * MPSAFE 921 */ 922 void 923 brelvp(struct buf *bp) 924 { 925 struct vnode *vp; 926 927 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 928 929 /* 930 * Delete from old vnode list, if on one. 931 */ 932 vp = bp->b_vp; 933 lwkt_gettoken(&vp->v_token); 934 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 935 if (bp->b_flags & B_VNDIRTY) 936 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 937 else 938 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 939 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 940 } 941 if (bp->b_flags & B_HASHED) { 942 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 943 bp->b_flags &= ~B_HASHED; 944 } 945 946 /* 947 * Only remove from synclist when no dirty buffers are left AND 948 * the VFS has not flagged the vnode's inode as being dirty. 949 */ 950 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 951 RB_EMPTY(&vp->v_rbdirty_tree)) { 952 vn_syncer_remove(vp); 953 } 954 bp->b_vp = NULL; 955 956 lwkt_reltoken(&vp->v_token); 957 958 /*vdrop(vp);*/ 959 } 960 961 /* 962 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 963 * This routine is called when the state of the B_DELWRI bit is changed. 964 * 965 * Must be called with vp->v_token held. 966 * MPSAFE 967 */ 968 void 969 reassignbuf(struct buf *bp) 970 { 971 struct vnode *vp = bp->b_vp; 972 int delay; 973 974 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 975 ++reassignbufcalls; 976 977 /* 978 * B_PAGING flagged buffers cannot be reassigned because their vp 979 * is not fully linked in. 980 */ 981 if (bp->b_flags & B_PAGING) 982 panic("cannot reassign paging buffer"); 983 984 if (bp->b_flags & B_DELWRI) { 985 /* 986 * Move to the dirty list, add the vnode to the worklist 987 */ 988 if (bp->b_flags & B_VNCLEAN) { 989 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 990 bp->b_flags &= ~B_VNCLEAN; 991 } 992 if ((bp->b_flags & B_VNDIRTY) == 0) { 993 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 994 panic("reassignbuf: dup lblk vp %p bp %p", 995 vp, bp); 996 } 997 bp->b_flags |= B_VNDIRTY; 998 } 999 if ((vp->v_flag & VONWORKLST) == 0) { 1000 switch (vp->v_type) { 1001 case VDIR: 1002 delay = dirdelay; 1003 break; 1004 case VCHR: 1005 case VBLK: 1006 if (vp->v_rdev && 1007 vp->v_rdev->si_mountpoint != NULL) { 1008 delay = metadelay; 1009 break; 1010 } 1011 /* fall through */ 1012 default: 1013 delay = filedelay; 1014 } 1015 vn_syncer_add(vp, delay); 1016 } 1017 } else { 1018 /* 1019 * Move to the clean list, remove the vnode from the worklist 1020 * if no dirty blocks remain. 1021 */ 1022 if (bp->b_flags & B_VNDIRTY) { 1023 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1024 bp->b_flags &= ~B_VNDIRTY; 1025 } 1026 if ((bp->b_flags & B_VNCLEAN) == 0) { 1027 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1028 panic("reassignbuf: dup lblk vp %p bp %p", 1029 vp, bp); 1030 } 1031 bp->b_flags |= B_VNCLEAN; 1032 } 1033 1034 /* 1035 * Only remove from synclist when no dirty buffers are left 1036 * AND the VFS has not flagged the vnode's inode as being 1037 * dirty. 1038 */ 1039 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1040 VONWORKLST && 1041 RB_EMPTY(&vp->v_rbdirty_tree)) { 1042 vn_syncer_remove(vp); 1043 } 1044 } 1045 } 1046 1047 /* 1048 * Create a vnode for a block device. Used for mounting the root file 1049 * system. 1050 * 1051 * A vref()'d vnode is returned. 1052 */ 1053 extern struct vop_ops *devfs_vnode_dev_vops_p; 1054 int 1055 bdevvp(cdev_t dev, struct vnode **vpp) 1056 { 1057 struct vnode *vp; 1058 struct vnode *nvp; 1059 int error; 1060 1061 if (dev == NULL) { 1062 *vpp = NULLVP; 1063 return (ENXIO); 1064 } 1065 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1066 &nvp, 0, 0); 1067 if (error) { 1068 *vpp = NULLVP; 1069 return (error); 1070 } 1071 vp = nvp; 1072 vp->v_type = VCHR; 1073 #if 0 1074 vp->v_rdev = dev; 1075 #endif 1076 v_associate_rdev(vp, dev); 1077 vp->v_umajor = dev->si_umajor; 1078 vp->v_uminor = dev->si_uminor; 1079 vx_unlock(vp); 1080 *vpp = vp; 1081 return (0); 1082 } 1083 1084 int 1085 v_associate_rdev(struct vnode *vp, cdev_t dev) 1086 { 1087 if (dev == NULL) 1088 return(ENXIO); 1089 if (dev_is_good(dev) == 0) 1090 return(ENXIO); 1091 KKASSERT(vp->v_rdev == NULL); 1092 vp->v_rdev = reference_dev(dev); 1093 lwkt_gettoken(&spechash_token); 1094 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1095 lwkt_reltoken(&spechash_token); 1096 return(0); 1097 } 1098 1099 void 1100 v_release_rdev(struct vnode *vp) 1101 { 1102 cdev_t dev; 1103 1104 if ((dev = vp->v_rdev) != NULL) { 1105 lwkt_gettoken(&spechash_token); 1106 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1107 vp->v_rdev = NULL; 1108 release_dev(dev); 1109 lwkt_reltoken(&spechash_token); 1110 } 1111 } 1112 1113 /* 1114 * Add a vnode to the alias list hung off the cdev_t. We only associate 1115 * the device number with the vnode. The actual device is not associated 1116 * until the vnode is opened (usually in spec_open()), and will be 1117 * disassociated on last close. 1118 */ 1119 void 1120 addaliasu(struct vnode *nvp, int x, int y) 1121 { 1122 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1123 panic("addaliasu on non-special vnode"); 1124 nvp->v_umajor = x; 1125 nvp->v_uminor = y; 1126 } 1127 1128 /* 1129 * Simple call that a filesystem can make to try to get rid of a 1130 * vnode. It will fail if anyone is referencing the vnode (including 1131 * the caller). 1132 * 1133 * The filesystem can check whether its in-memory inode structure still 1134 * references the vp on return. 1135 * 1136 * May only be called if the vnode is in a known state (i.e. being prevented 1137 * from being deallocated by some other condition such as a vfs inode hold). 1138 */ 1139 void 1140 vclean_unlocked(struct vnode *vp) 1141 { 1142 vx_get(vp); 1143 if (VREFCNT(vp) <= 1) 1144 vgone_vxlocked(vp); 1145 vx_put(vp); 1146 } 1147 1148 /* 1149 * Disassociate a vnode from its underlying filesystem. 1150 * 1151 * The vnode must be VX locked and referenced. In all normal situations 1152 * there are no active references. If vclean_vxlocked() is called while 1153 * there are active references, the vnode is being ripped out and we have 1154 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1155 */ 1156 void 1157 vclean_vxlocked(struct vnode *vp, int flags) 1158 { 1159 int active; 1160 int n; 1161 vm_object_t object; 1162 struct namecache *ncp; 1163 1164 /* 1165 * If the vnode has already been reclaimed we have nothing to do. 1166 */ 1167 if (vp->v_flag & VRECLAIMED) 1168 return; 1169 1170 /* 1171 * Set flag to interlock operation, flag finalization to ensure 1172 * that the vnode winds up on the inactive list, and set v_act to 0. 1173 */ 1174 vsetflags(vp, VRECLAIMED); 1175 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1176 vp->v_act = 0; 1177 1178 if (verbose_reclaims) { 1179 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1180 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1181 } 1182 1183 /* 1184 * Scrap the vfs cache 1185 */ 1186 while (cache_inval_vp(vp, 0) != 0) { 1187 kprintf("Warning: vnode %p clean/cache_resolution " 1188 "race detected\n", vp); 1189 tsleep(vp, 0, "vclninv", 2); 1190 } 1191 1192 /* 1193 * Check to see if the vnode is in use. If so we have to reference it 1194 * before we clean it out so that its count cannot fall to zero and 1195 * generate a race against ourselves to recycle it. 1196 */ 1197 active = (VREFCNT(vp) > 0); 1198 1199 /* 1200 * Clean out any buffers associated with the vnode and destroy its 1201 * object, if it has one. 1202 */ 1203 vinvalbuf(vp, V_SAVE, 0, 0); 1204 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1205 1206 /* 1207 * If purging an active vnode (typically during a forced unmount 1208 * or reboot), it must be closed and deactivated before being 1209 * reclaimed. This isn't really all that safe, but what can 1210 * we do? XXX. 1211 * 1212 * Note that neither of these routines unlocks the vnode. 1213 */ 1214 if (active && (flags & DOCLOSE)) { 1215 while ((n = vp->v_opencount) != 0) { 1216 if (vp->v_writecount) 1217 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1218 else 1219 VOP_CLOSE(vp, FNONBLOCK, NULL); 1220 if (vp->v_opencount == n) { 1221 kprintf("Warning: unable to force-close" 1222 " vnode %p\n", vp); 1223 break; 1224 } 1225 } 1226 } 1227 1228 /* 1229 * If the vnode has not been deactivated, deactivated it. Deactivation 1230 * can create new buffers and VM pages so we have to call vinvalbuf() 1231 * again to make sure they all get flushed. 1232 * 1233 * This can occur if a file with a link count of 0 needs to be 1234 * truncated. 1235 * 1236 * If the vnode is already dead don't try to deactivate it. 1237 */ 1238 if ((vp->v_flag & VINACTIVE) == 0) { 1239 vsetflags(vp, VINACTIVE); 1240 if (vp->v_mount) 1241 VOP_INACTIVE(vp); 1242 vinvalbuf(vp, V_SAVE, 0, 0); 1243 } 1244 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1245 1246 /* 1247 * If the vnode has an object, destroy it. 1248 */ 1249 while ((object = vp->v_object) != NULL) { 1250 vm_object_hold(object); 1251 if (object == vp->v_object) 1252 break; 1253 vm_object_drop(object); 1254 } 1255 1256 if (object != NULL) { 1257 if (object->ref_count == 0) { 1258 if ((object->flags & OBJ_DEAD) == 0) 1259 vm_object_terminate(object); 1260 vm_object_drop(object); 1261 vclrflags(vp, VOBJBUF); 1262 } else { 1263 vm_pager_deallocate(object); 1264 vclrflags(vp, VOBJBUF); 1265 vm_object_drop(object); 1266 } 1267 } 1268 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1269 1270 /* 1271 * Reclaim the vnode if not already dead. 1272 */ 1273 if (vp->v_mount && VOP_RECLAIM(vp)) 1274 panic("vclean: cannot reclaim"); 1275 1276 /* 1277 * Done with purge, notify sleepers of the grim news. 1278 */ 1279 vp->v_ops = &dead_vnode_vops_p; 1280 vn_gone(vp); 1281 vp->v_tag = VT_NON; 1282 1283 /* 1284 * If we are destroying an active vnode, reactivate it now that 1285 * we have reassociated it with deadfs. This prevents the system 1286 * from crashing on the vnode due to it being unexpectedly marked 1287 * as inactive or reclaimed. 1288 */ 1289 if (active && (flags & DOCLOSE)) { 1290 vclrflags(vp, VINACTIVE | VRECLAIMED); 1291 } 1292 } 1293 1294 /* 1295 * Eliminate all activity associated with the requested vnode 1296 * and with all vnodes aliased to the requested vnode. 1297 * 1298 * The vnode must be referenced but should not be locked. 1299 */ 1300 int 1301 vrevoke(struct vnode *vp, struct ucred *cred) 1302 { 1303 struct vnode *vq; 1304 struct vnode *vqn; 1305 cdev_t dev; 1306 int error; 1307 1308 /* 1309 * If the vnode has a device association, scrap all vnodes associated 1310 * with the device. Don't let the device disappear on us while we 1311 * are scrapping the vnodes. 1312 * 1313 * The passed vp will probably show up in the list, do not VX lock 1314 * it twice! 1315 * 1316 * Releasing the vnode's rdev here can mess up specfs's call to 1317 * device close, so don't do it. The vnode has been disassociated 1318 * and the device will be closed after the last ref on the related 1319 * fp goes away (if not still open by e.g. the kernel). 1320 */ 1321 if (vp->v_type != VCHR) { 1322 error = fdrevoke(vp, DTYPE_VNODE, cred); 1323 return (error); 1324 } 1325 if ((dev = vp->v_rdev) == NULL) { 1326 return(0); 1327 } 1328 reference_dev(dev); 1329 lwkt_gettoken(&spechash_token); 1330 1331 restart: 1332 vqn = SLIST_FIRST(&dev->si_hlist); 1333 if (vqn) 1334 vhold(vqn); 1335 while ((vq = vqn) != NULL) { 1336 if (VREFCNT(vq) > 0) { 1337 vref(vq); 1338 fdrevoke(vq, DTYPE_VNODE, cred); 1339 /*v_release_rdev(vq);*/ 1340 vrele(vq); 1341 if (vq->v_rdev != dev) { 1342 vdrop(vq); 1343 goto restart; 1344 } 1345 } 1346 vqn = SLIST_NEXT(vq, v_cdevnext); 1347 if (vqn) 1348 vhold(vqn); 1349 vdrop(vq); 1350 } 1351 lwkt_reltoken(&spechash_token); 1352 dev_drevoke(dev); 1353 release_dev(dev); 1354 return (0); 1355 } 1356 1357 /* 1358 * This is called when the object underlying a vnode is being destroyed, 1359 * such as in a remove(). Try to recycle the vnode immediately if the 1360 * only active reference is our reference. 1361 * 1362 * Directory vnodes in the namecache with children cannot be immediately 1363 * recycled because numerous VOP_N*() ops require them to be stable. 1364 * 1365 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1366 * function is a NOP if VRECLAIMED is already set. 1367 */ 1368 int 1369 vrecycle(struct vnode *vp) 1370 { 1371 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1372 if (cache_inval_vp_nonblock(vp)) 1373 return(0); 1374 vgone_vxlocked(vp); 1375 return (1); 1376 } 1377 return (0); 1378 } 1379 1380 /* 1381 * Return the maximum I/O size allowed for strategy calls on VP. 1382 * 1383 * If vp is VCHR or VBLK we dive the device, otherwise we use 1384 * the vp's mount info. 1385 * 1386 * The returned value is clamped at MAXPHYS as most callers cannot use 1387 * buffers larger than that size. 1388 */ 1389 int 1390 vmaxiosize(struct vnode *vp) 1391 { 1392 int maxiosize; 1393 1394 if (vp->v_type == VBLK || vp->v_type == VCHR) 1395 maxiosize = vp->v_rdev->si_iosize_max; 1396 else 1397 maxiosize = vp->v_mount->mnt_iosize_max; 1398 1399 if (maxiosize > MAXPHYS) 1400 maxiosize = MAXPHYS; 1401 return (maxiosize); 1402 } 1403 1404 /* 1405 * Eliminate all activity associated with a vnode in preparation for 1406 * destruction. 1407 * 1408 * The vnode must be VX locked and refd and will remain VX locked and refd 1409 * on return. This routine may be called with the vnode in any state, as 1410 * long as it is VX locked. The vnode will be cleaned out and marked 1411 * VRECLAIMED but will not actually be reused until all existing refs and 1412 * holds go away. 1413 * 1414 * NOTE: This routine may be called on a vnode which has not yet been 1415 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1416 * already been reclaimed. 1417 * 1418 * This routine is not responsible for placing us back on the freelist. 1419 * Instead, it happens automatically when the caller releases the VX lock 1420 * (assuming there aren't any other references). 1421 */ 1422 void 1423 vgone_vxlocked(struct vnode *vp) 1424 { 1425 /* 1426 * assert that the VX lock is held. This is an absolute requirement 1427 * now for vgone_vxlocked() to be called. 1428 */ 1429 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1430 1431 /* 1432 * Clean out the filesystem specific data and set the VRECLAIMED 1433 * bit. Also deactivate the vnode if necessary. 1434 * 1435 * The vnode should have automatically been removed from the syncer 1436 * list as syncer/dirty flags cleared during the cleaning. 1437 */ 1438 vclean_vxlocked(vp, DOCLOSE); 1439 KKASSERT((vp->v_flag & VONWORKLST) == 0); 1440 1441 /* 1442 * Delete from old mount point vnode list, if on one. 1443 */ 1444 if (vp->v_mount != NULL) { 1445 KKASSERT(vp->v_data == NULL); 1446 insmntque(vp, NULL); 1447 } 1448 1449 /* 1450 * If special device, remove it from special device alias list 1451 * if it is on one. This should normally only occur if a vnode is 1452 * being revoked as the device should otherwise have been released 1453 * naturally. 1454 */ 1455 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1456 v_release_rdev(vp); 1457 } 1458 1459 /* 1460 * Set us to VBAD 1461 */ 1462 vp->v_type = VBAD; 1463 } 1464 1465 /* 1466 * Lookup a vnode by device number. 1467 * 1468 * Returns non-zero and *vpp set to a vref'd vnode on success. 1469 * Returns zero on failure. 1470 */ 1471 int 1472 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1473 { 1474 struct vnode *vp; 1475 1476 lwkt_gettoken(&spechash_token); 1477 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1478 if (type == vp->v_type) { 1479 *vpp = vp; 1480 vref(vp); 1481 lwkt_reltoken(&spechash_token); 1482 return (1); 1483 } 1484 } 1485 lwkt_reltoken(&spechash_token); 1486 return (0); 1487 } 1488 1489 /* 1490 * Calculate the total number of references to a special device. This 1491 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1492 * an overloaded field. Since udev2dev can now return NULL, we have 1493 * to check for a NULL v_rdev. 1494 */ 1495 int 1496 count_dev(cdev_t dev) 1497 { 1498 struct vnode *vp; 1499 int count = 0; 1500 1501 if (SLIST_FIRST(&dev->si_hlist)) { 1502 lwkt_gettoken(&spechash_token); 1503 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1504 count += vp->v_opencount; 1505 } 1506 lwkt_reltoken(&spechash_token); 1507 } 1508 return(count); 1509 } 1510 1511 int 1512 vcount(struct vnode *vp) 1513 { 1514 if (vp->v_rdev == NULL) 1515 return(0); 1516 return(count_dev(vp->v_rdev)); 1517 } 1518 1519 /* 1520 * Initialize VMIO for a vnode. This routine MUST be called before a 1521 * VFS can issue buffer cache ops on a vnode. It is typically called 1522 * when a vnode is initialized from its inode. 1523 */ 1524 int 1525 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1526 { 1527 vm_object_t object; 1528 int error = 0; 1529 1530 object = vp->v_object; 1531 if (object) { 1532 vm_object_hold(object); 1533 KKASSERT(vp->v_object == object); 1534 } 1535 1536 if (object == NULL) { 1537 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1538 1539 /* 1540 * Dereference the reference we just created. This assumes 1541 * that the object is associated with the vp. Allow it to 1542 * have zero refs. It cannot be destroyed as long as it 1543 * is associated with the vnode. 1544 */ 1545 vm_object_hold(object); 1546 atomic_add_int(&object->ref_count, -1); 1547 vrele(vp); 1548 } else { 1549 KKASSERT((object->flags & OBJ_DEAD) == 0); 1550 } 1551 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1552 vsetflags(vp, VOBJBUF); 1553 vm_object_drop(object); 1554 1555 return (error); 1556 } 1557 1558 1559 /* 1560 * Print out a description of a vnode. 1561 */ 1562 static char *typename[] = 1563 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1564 1565 void 1566 vprint(char *label, struct vnode *vp) 1567 { 1568 char buf[96]; 1569 1570 if (label != NULL) 1571 kprintf("%s: %p: ", label, (void *)vp); 1572 else 1573 kprintf("%p: ", (void *)vp); 1574 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1575 typename[vp->v_type], 1576 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1577 buf[0] = '\0'; 1578 if (vp->v_flag & VROOT) 1579 strcat(buf, "|VROOT"); 1580 if (vp->v_flag & VPFSROOT) 1581 strcat(buf, "|VPFSROOT"); 1582 if (vp->v_flag & VTEXT) 1583 strcat(buf, "|VTEXT"); 1584 if (vp->v_flag & VSYSTEM) 1585 strcat(buf, "|VSYSTEM"); 1586 if (vp->v_flag & VOBJBUF) 1587 strcat(buf, "|VOBJBUF"); 1588 if (buf[0] != '\0') 1589 kprintf(" flags (%s)", &buf[1]); 1590 if (vp->v_data == NULL) { 1591 kprintf("\n"); 1592 } else { 1593 kprintf("\n\t"); 1594 VOP_PRINT(vp); 1595 } 1596 } 1597 1598 /* 1599 * Do the usual access checking. 1600 * file_mode, uid and gid are from the vnode in question, 1601 * while acc_mode and cred are from the VOP_ACCESS parameter list 1602 */ 1603 int 1604 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1605 mode_t acc_mode, struct ucred *cred) 1606 { 1607 mode_t mask; 1608 int ismember; 1609 1610 /* 1611 * Super-user always gets read/write access, but execute access depends 1612 * on at least one execute bit being set. 1613 */ 1614 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1615 if ((acc_mode & VEXEC) && type != VDIR && 1616 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1617 return (EACCES); 1618 return (0); 1619 } 1620 1621 mask = 0; 1622 1623 /* Otherwise, check the owner. */ 1624 if (cred->cr_uid == uid) { 1625 if (acc_mode & VEXEC) 1626 mask |= S_IXUSR; 1627 if (acc_mode & VREAD) 1628 mask |= S_IRUSR; 1629 if (acc_mode & VWRITE) 1630 mask |= S_IWUSR; 1631 return ((file_mode & mask) == mask ? 0 : EACCES); 1632 } 1633 1634 /* Otherwise, check the groups. */ 1635 ismember = groupmember(gid, cred); 1636 if (cred->cr_svgid == gid || ismember) { 1637 if (acc_mode & VEXEC) 1638 mask |= S_IXGRP; 1639 if (acc_mode & VREAD) 1640 mask |= S_IRGRP; 1641 if (acc_mode & VWRITE) 1642 mask |= S_IWGRP; 1643 return ((file_mode & mask) == mask ? 0 : EACCES); 1644 } 1645 1646 /* Otherwise, check everyone else. */ 1647 if (acc_mode & VEXEC) 1648 mask |= S_IXOTH; 1649 if (acc_mode & VREAD) 1650 mask |= S_IROTH; 1651 if (acc_mode & VWRITE) 1652 mask |= S_IWOTH; 1653 return ((file_mode & mask) == mask ? 0 : EACCES); 1654 } 1655 1656 #ifdef DDB 1657 #include <ddb/ddb.h> 1658 1659 static int db_show_locked_vnodes(struct mount *mp, void *data); 1660 1661 /* 1662 * List all of the locked vnodes in the system. 1663 * Called when debugging the kernel. 1664 */ 1665 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1666 { 1667 kprintf("Locked vnodes\n"); 1668 mountlist_scan(db_show_locked_vnodes, NULL, 1669 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1670 } 1671 1672 static int 1673 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1674 { 1675 struct vnode *vp; 1676 1677 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1678 if (vn_islocked(vp)) 1679 vprint(NULL, vp); 1680 } 1681 return(0); 1682 } 1683 #endif 1684 1685 /* 1686 * Top level filesystem related information gathering. 1687 */ 1688 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1689 1690 static int 1691 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1692 { 1693 int *name = (int *)arg1 - 1; /* XXX */ 1694 u_int namelen = arg2 + 1; /* XXX */ 1695 struct vfsconf *vfsp; 1696 int maxtypenum; 1697 1698 #if 1 || defined(COMPAT_PRELITE2) 1699 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1700 if (namelen == 1) 1701 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1702 #endif 1703 1704 #ifdef notyet 1705 /* all sysctl names at this level are at least name and field */ 1706 if (namelen < 2) 1707 return (ENOTDIR); /* overloaded */ 1708 if (name[0] != VFS_GENERIC) { 1709 vfsp = vfsconf_find_by_typenum(name[0]); 1710 if (vfsp == NULL) 1711 return (EOPNOTSUPP); 1712 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1713 oldp, oldlenp, newp, newlen, p)); 1714 } 1715 #endif 1716 switch (name[1]) { 1717 case VFS_MAXTYPENUM: 1718 if (namelen != 2) 1719 return (ENOTDIR); 1720 maxtypenum = vfsconf_get_maxtypenum(); 1721 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1722 case VFS_CONF: 1723 if (namelen != 3) 1724 return (ENOTDIR); /* overloaded */ 1725 vfsp = vfsconf_find_by_typenum(name[2]); 1726 if (vfsp == NULL) 1727 return (EOPNOTSUPP); 1728 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1729 } 1730 return (EOPNOTSUPP); 1731 } 1732 1733 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1734 "Generic filesystem"); 1735 1736 #if 1 || defined(COMPAT_PRELITE2) 1737 1738 static int 1739 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1740 { 1741 int error; 1742 struct ovfsconf ovfs; 1743 struct sysctl_req *req = (struct sysctl_req*) data; 1744 1745 bzero(&ovfs, sizeof(ovfs)); 1746 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1747 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1748 ovfs.vfc_index = vfsp->vfc_typenum; 1749 ovfs.vfc_refcount = vfsp->vfc_refcount; 1750 ovfs.vfc_flags = vfsp->vfc_flags; 1751 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1752 if (error) 1753 return error; /* abort iteration with error code */ 1754 else 1755 return 0; /* continue iterating with next element */ 1756 } 1757 1758 static int 1759 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1760 { 1761 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1762 } 1763 1764 #endif /* 1 || COMPAT_PRELITE2 */ 1765 1766 /* 1767 * Check to see if a filesystem is mounted on a block device. 1768 */ 1769 int 1770 vfs_mountedon(struct vnode *vp) 1771 { 1772 cdev_t dev; 1773 1774 if ((dev = vp->v_rdev) == NULL) { 1775 /* if (vp->v_type != VBLK) 1776 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1777 } 1778 if (dev != NULL && dev->si_mountpoint) 1779 return (EBUSY); 1780 return (0); 1781 } 1782 1783 /* 1784 * Unmount all filesystems. The list is traversed in reverse order 1785 * of mounting to avoid dependencies. 1786 */ 1787 1788 static int vfs_umountall_callback(struct mount *mp, void *data); 1789 1790 void 1791 vfs_unmountall(void) 1792 { 1793 int count; 1794 1795 do { 1796 count = mountlist_scan(vfs_umountall_callback, 1797 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1798 } while (count); 1799 } 1800 1801 static 1802 int 1803 vfs_umountall_callback(struct mount *mp, void *data) 1804 { 1805 int error; 1806 1807 error = dounmount(mp, MNT_FORCE); 1808 if (error) { 1809 mountlist_remove(mp); 1810 kprintf("unmount of filesystem mounted from %s failed (", 1811 mp->mnt_stat.f_mntfromname); 1812 if (error == EBUSY) 1813 kprintf("BUSY)\n"); 1814 else 1815 kprintf("%d)\n", error); 1816 } 1817 return(1); 1818 } 1819 1820 /* 1821 * Checks the mount flags for parameter mp and put the names comma-separated 1822 * into a string buffer buf with a size limit specified by len. 1823 * 1824 * It returns the number of bytes written into buf, and (*errorp) will be 1825 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1826 * not large enough). The buffer will be 0-terminated if len was not 0. 1827 */ 1828 size_t 1829 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1830 char *buf, size_t len, int *errorp) 1831 { 1832 static const struct mountctl_opt optnames[] = { 1833 { MNT_ASYNC, "asynchronous" }, 1834 { MNT_EXPORTED, "NFS exported" }, 1835 { MNT_LOCAL, "local" }, 1836 { MNT_NOATIME, "noatime" }, 1837 { MNT_NODEV, "nodev" }, 1838 { MNT_NOEXEC, "noexec" }, 1839 { MNT_NOSUID, "nosuid" }, 1840 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1841 { MNT_QUOTA, "with-quotas" }, 1842 { MNT_RDONLY, "read-only" }, 1843 { MNT_SYNCHRONOUS, "synchronous" }, 1844 { MNT_UNION, "union" }, 1845 { MNT_NOCLUSTERR, "noclusterr" }, 1846 { MNT_NOCLUSTERW, "noclusterw" }, 1847 { MNT_SUIDDIR, "suiddir" }, 1848 { MNT_SOFTDEP, "soft-updates" }, 1849 { MNT_IGNORE, "ignore" }, 1850 { 0, NULL} 1851 }; 1852 int bwritten; 1853 int bleft; 1854 int optlen; 1855 int actsize; 1856 1857 *errorp = 0; 1858 bwritten = 0; 1859 bleft = len - 1; /* leave room for trailing \0 */ 1860 1861 /* 1862 * Checks the size of the string. If it contains 1863 * any data, then we will append the new flags to 1864 * it. 1865 */ 1866 actsize = strlen(buf); 1867 if (actsize > 0) 1868 buf += actsize; 1869 1870 /* Default flags if no flags passed */ 1871 if (optp == NULL) 1872 optp = optnames; 1873 1874 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1875 *errorp = EINVAL; 1876 return(0); 1877 } 1878 1879 for (; flags && optp->o_opt; ++optp) { 1880 if ((flags & optp->o_opt) == 0) 1881 continue; 1882 optlen = strlen(optp->o_name); 1883 if (bwritten || actsize > 0) { 1884 if (bleft < 2) { 1885 *errorp = ENOSPC; 1886 break; 1887 } 1888 buf[bwritten++] = ','; 1889 buf[bwritten++] = ' '; 1890 bleft -= 2; 1891 } 1892 if (bleft < optlen) { 1893 *errorp = ENOSPC; 1894 break; 1895 } 1896 bcopy(optp->o_name, buf + bwritten, optlen); 1897 bwritten += optlen; 1898 bleft -= optlen; 1899 flags &= ~optp->o_opt; 1900 } 1901 1902 /* 1903 * Space already reserved for trailing \0 1904 */ 1905 buf[bwritten] = 0; 1906 return (bwritten); 1907 } 1908 1909 /* 1910 * Build hash lists of net addresses and hang them off the mount point. 1911 * Called by ufs_mount() to set up the lists of export addresses. 1912 */ 1913 static int 1914 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1915 const struct export_args *argp) 1916 { 1917 struct netcred *np; 1918 struct radix_node_head *rnh; 1919 int i; 1920 struct radix_node *rn; 1921 struct sockaddr *saddr, *smask = NULL; 1922 int error; 1923 1924 if (argp->ex_addrlen == 0) { 1925 if (mp->mnt_flag & MNT_DEFEXPORTED) 1926 return (EPERM); 1927 np = &nep->ne_defexported; 1928 np->netc_exflags = argp->ex_flags; 1929 np->netc_anon = argp->ex_anon; 1930 np->netc_anon.cr_ref = 1; 1931 mp->mnt_flag |= MNT_DEFEXPORTED; 1932 return (0); 1933 } 1934 1935 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1936 return (EINVAL); 1937 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1938 return (EINVAL); 1939 1940 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1941 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1942 saddr = (struct sockaddr *) (np + 1); 1943 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1944 goto out; 1945 if (saddr->sa_len > argp->ex_addrlen) 1946 saddr->sa_len = argp->ex_addrlen; 1947 if (argp->ex_masklen) { 1948 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1949 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1950 if (error) 1951 goto out; 1952 if (smask->sa_len > argp->ex_masklen) 1953 smask->sa_len = argp->ex_masklen; 1954 } 1955 NE_LOCK(nep); 1956 if (nep->ne_maskhead == NULL) { 1957 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) { 1958 error = ENOBUFS; 1959 goto out; 1960 } 1961 } 1962 if((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) { 1963 error = ENOBUFS; 1964 goto out; 1965 } 1966 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1967 np->netc_rnodes); 1968 NE_UNLOCK(nep); 1969 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */ 1970 error = EPERM; 1971 goto out; 1972 } 1973 np->netc_exflags = argp->ex_flags; 1974 np->netc_anon = argp->ex_anon; 1975 np->netc_anon.cr_ref = 1; 1976 return (0); 1977 out: 1978 kfree(np, M_NETADDR); 1979 return (error); 1980 } 1981 1982 /* ARGSUSED */ 1983 static int 1984 vfs_free_netcred(struct radix_node *rn, void *w) 1985 { 1986 struct radix_node_head *rnh = (struct radix_node_head *) w; 1987 1988 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1989 kfree((caddr_t) rn, M_NETADDR); 1990 return (0); 1991 } 1992 1993 static struct radix_node_head * 1994 vfs_create_addrlist_af(int af, struct netexport *nep) 1995 { 1996 struct radix_node_head *rnh = NULL; 1997 #if defined(INET) || defined(INET6) 1998 struct radix_node_head *maskhead = nep->ne_maskhead; 1999 int off; 2000 #endif 2001 2002 NE_ASSERT_LOCKED(nep); 2003 KKASSERT(maskhead != NULL); 2004 switch (af) { 2005 #ifdef INET 2006 case AF_INET: 2007 if ((rnh = nep->ne_inethead) == NULL) { 2008 off = offsetof(struct sockaddr_in, sin_addr) << 3; 2009 if (!rn_inithead((void **)&rnh, maskhead, off)) 2010 return (NULL); 2011 nep->ne_inethead = rnh; 2012 } 2013 break; 2014 #endif 2015 #ifdef INET6 2016 case AF_INET6: 2017 if ((rnh = nep->ne_inet6head) == NULL) { 2018 off = offsetof(struct sockaddr_in6, sin6_addr) << 3; 2019 if (!rn_inithead((void **)&rnh, maskhead, off)) 2020 return (NULL); 2021 nep->ne_inet6head = rnh; 2022 } 2023 break; 2024 #endif 2025 } 2026 return (rnh); 2027 } 2028 2029 static void 2030 vfs_free_addrlist_af(struct radix_node_head **prnh) 2031 { 2032 struct radix_node_head *rnh = *prnh; 2033 2034 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh); 2035 kfree(rnh, M_RTABLE); 2036 *prnh = NULL; 2037 } 2038 2039 /* 2040 * Free the net address hash lists that are hanging off the mount points. 2041 */ 2042 static void 2043 vfs_free_addrlist(struct netexport *nep) 2044 { 2045 NE_LOCK(nep); 2046 if (nep->ne_inethead != NULL) 2047 vfs_free_addrlist_af(&nep->ne_inethead); 2048 if (nep->ne_inet6head != NULL) 2049 vfs_free_addrlist_af(&nep->ne_inet6head); 2050 if (nep->ne_maskhead) 2051 vfs_free_addrlist_af(&nep->ne_maskhead); 2052 NE_UNLOCK(nep); 2053 } 2054 2055 int 2056 vfs_export(struct mount *mp, struct netexport *nep, 2057 const struct export_args *argp) 2058 { 2059 int error; 2060 2061 if (argp->ex_flags & MNT_DELEXPORT) { 2062 if (mp->mnt_flag & MNT_EXPUBLIC) { 2063 vfs_setpublicfs(NULL, NULL, NULL); 2064 mp->mnt_flag &= ~MNT_EXPUBLIC; 2065 } 2066 vfs_free_addrlist(nep); 2067 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2068 } 2069 if (argp->ex_flags & MNT_EXPORTED) { 2070 if (argp->ex_flags & MNT_EXPUBLIC) { 2071 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2072 return (error); 2073 mp->mnt_flag |= MNT_EXPUBLIC; 2074 } 2075 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2076 return (error); 2077 mp->mnt_flag |= MNT_EXPORTED; 2078 } 2079 return (0); 2080 } 2081 2082 2083 /* 2084 * Set the publicly exported filesystem (WebNFS). Currently, only 2085 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2086 */ 2087 int 2088 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2089 const struct export_args *argp) 2090 { 2091 int error; 2092 struct vnode *rvp; 2093 char *cp; 2094 2095 /* 2096 * mp == NULL -> invalidate the current info, the FS is 2097 * no longer exported. May be called from either vfs_export 2098 * or unmount, so check if it hasn't already been done. 2099 */ 2100 if (mp == NULL) { 2101 if (nfs_pub.np_valid) { 2102 nfs_pub.np_valid = 0; 2103 if (nfs_pub.np_index != NULL) { 2104 kfree(nfs_pub.np_index, M_TEMP); 2105 nfs_pub.np_index = NULL; 2106 } 2107 } 2108 return (0); 2109 } 2110 2111 /* 2112 * Only one allowed at a time. 2113 */ 2114 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2115 return (EBUSY); 2116 2117 /* 2118 * Get real filehandle for root of exported FS. 2119 */ 2120 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2121 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2122 2123 if ((error = VFS_ROOT(mp, &rvp))) 2124 return (error); 2125 2126 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2127 return (error); 2128 2129 vput(rvp); 2130 2131 /* 2132 * If an indexfile was specified, pull it in. 2133 */ 2134 if (argp->ex_indexfile != NULL) { 2135 int namelen; 2136 2137 error = vn_get_namelen(rvp, &namelen); 2138 if (error) 2139 return (error); 2140 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2141 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2142 namelen, NULL); 2143 if (!error) { 2144 /* 2145 * Check for illegal filenames. 2146 */ 2147 for (cp = nfs_pub.np_index; *cp; cp++) { 2148 if (*cp == '/') { 2149 error = EINVAL; 2150 break; 2151 } 2152 } 2153 } 2154 if (error) { 2155 kfree(nfs_pub.np_index, M_TEMP); 2156 return (error); 2157 } 2158 } 2159 2160 nfs_pub.np_mount = mp; 2161 nfs_pub.np_valid = 1; 2162 return (0); 2163 } 2164 2165 struct netcred * 2166 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2167 struct sockaddr *nam) 2168 { 2169 struct netcred *np; 2170 struct radix_node_head *rnh; 2171 struct sockaddr *saddr; 2172 2173 np = NULL; 2174 if (mp->mnt_flag & MNT_EXPORTED) { 2175 /* 2176 * Lookup in the export list first. 2177 */ 2178 NE_LOCK(nep); 2179 if (nam != NULL) { 2180 saddr = nam; 2181 switch (saddr->sa_family) { 2182 #ifdef INET 2183 case AF_INET: 2184 rnh = nep->ne_inethead; 2185 break; 2186 #endif 2187 #ifdef INET6 2188 case AF_INET6: 2189 rnh = nep->ne_inet6head; 2190 break; 2191 #endif 2192 default: 2193 rnh = NULL; 2194 } 2195 if (rnh != NULL) { 2196 np = (struct netcred *) 2197 (*rnh->rnh_matchaddr)((char *)saddr, 2198 rnh); 2199 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2200 np = NULL; 2201 } 2202 } 2203 NE_UNLOCK(nep); 2204 /* 2205 * If no address match, use the default if it exists. 2206 */ 2207 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2208 np = &nep->ne_defexported; 2209 } 2210 return (np); 2211 } 2212 2213 /* 2214 * perform msync on all vnodes under a mount point. The mount point must 2215 * be locked. This code is also responsible for lazy-freeing unreferenced 2216 * vnodes whos VM objects no longer contain pages. 2217 * 2218 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2219 * 2220 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2221 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2222 * way up in this high level function. 2223 */ 2224 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2225 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2226 2227 void 2228 vfs_msync(struct mount *mp, int flags) 2229 { 2230 int vmsc_flags; 2231 2232 /* 2233 * tmpfs sets this flag to prevent msync(), sync, and the 2234 * filesystem periodic syncer from trying to flush VM pages 2235 * to swap. Only pure memory pressure flushes tmpfs VM pages 2236 * to swap. 2237 */ 2238 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2239 return; 2240 2241 /* 2242 * Ok, scan the vnodes for work. If the filesystem is using the 2243 * syncer thread feature we can use vsyncscan() instead of 2244 * vmntvnodescan(), which is much faster. 2245 */ 2246 vmsc_flags = VMSC_GETVP; 2247 if (flags != MNT_WAIT) 2248 vmsc_flags |= VMSC_NOWAIT; 2249 2250 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2251 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2252 (void *)(intptr_t)flags); 2253 } else { 2254 vmntvnodescan(mp, vmsc_flags, 2255 vfs_msync_scan1, vfs_msync_scan2, 2256 (void *)(intptr_t)flags); 2257 } 2258 } 2259 2260 /* 2261 * scan1 is a fast pre-check. There could be hundreds of thousands of 2262 * vnodes, we cannot afford to do anything heavy weight until we have a 2263 * fairly good indication that there is work to do. 2264 */ 2265 static 2266 int 2267 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2268 { 2269 int flags = (int)(intptr_t)data; 2270 2271 if ((vp->v_flag & VRECLAIMED) == 0) { 2272 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2273 vp->v_object) { 2274 return(0); /* call scan2 */ 2275 } 2276 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2277 (vp->v_flag & VOBJDIRTY) && 2278 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2279 return(0); /* call scan2 */ 2280 } 2281 } 2282 2283 /* 2284 * do not call scan2, continue the loop 2285 */ 2286 return(-1); 2287 } 2288 2289 /* 2290 * This callback is handed a locked vnode. 2291 */ 2292 static 2293 int 2294 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2295 { 2296 vm_object_t obj; 2297 int flags = (int)(intptr_t)data; 2298 2299 if (vp->v_flag & VRECLAIMED) 2300 return(0); 2301 2302 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2303 if ((obj = vp->v_object) != NULL) { 2304 vm_object_page_clean(obj, 0, 0, 2305 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2306 } 2307 } 2308 return(0); 2309 } 2310 2311 /* 2312 * Wake up anyone interested in vp because it is being revoked. 2313 */ 2314 void 2315 vn_gone(struct vnode *vp) 2316 { 2317 lwkt_gettoken(&vp->v_token); 2318 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2319 lwkt_reltoken(&vp->v_token); 2320 } 2321 2322 /* 2323 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2324 * (or v_rdev might be NULL). 2325 */ 2326 cdev_t 2327 vn_todev(struct vnode *vp) 2328 { 2329 if (vp->v_type != VBLK && vp->v_type != VCHR) 2330 return (NULL); 2331 KKASSERT(vp->v_rdev != NULL); 2332 return (vp->v_rdev); 2333 } 2334 2335 /* 2336 * Check if vnode represents a disk device. The vnode does not need to be 2337 * opened. 2338 * 2339 * MPALMOSTSAFE 2340 */ 2341 int 2342 vn_isdisk(struct vnode *vp, int *errp) 2343 { 2344 cdev_t dev; 2345 2346 if (vp->v_type != VCHR) { 2347 if (errp != NULL) 2348 *errp = ENOTBLK; 2349 return (0); 2350 } 2351 2352 dev = vp->v_rdev; 2353 2354 if (dev == NULL) { 2355 if (errp != NULL) 2356 *errp = ENXIO; 2357 return (0); 2358 } 2359 if (dev_is_good(dev) == 0) { 2360 if (errp != NULL) 2361 *errp = ENXIO; 2362 return (0); 2363 } 2364 if ((dev_dflags(dev) & D_DISK) == 0) { 2365 if (errp != NULL) 2366 *errp = ENOTBLK; 2367 return (0); 2368 } 2369 if (errp != NULL) 2370 *errp = 0; 2371 return (1); 2372 } 2373 2374 int 2375 vn_get_namelen(struct vnode *vp, int *namelen) 2376 { 2377 int error; 2378 register_t retval[2]; 2379 2380 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2381 if (error) 2382 return (error); 2383 *namelen = (int)retval[0]; 2384 return (0); 2385 } 2386 2387 int 2388 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2389 uint16_t d_namlen, const char *d_name) 2390 { 2391 struct dirent *dp; 2392 size_t len; 2393 2394 len = _DIRENT_RECLEN(d_namlen); 2395 if (len > uio->uio_resid) 2396 return(1); 2397 2398 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2399 2400 dp->d_ino = d_ino; 2401 dp->d_namlen = d_namlen; 2402 dp->d_type = d_type; 2403 bcopy(d_name, dp->d_name, d_namlen); 2404 2405 *error = uiomove((caddr_t)dp, len, uio); 2406 2407 kfree(dp, M_TEMP); 2408 2409 return(0); 2410 } 2411 2412 void 2413 vn_mark_atime(struct vnode *vp, struct thread *td) 2414 { 2415 struct proc *p = td->td_proc; 2416 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2417 2418 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2419 VOP_MARKATIME(vp, cred); 2420 } 2421 } 2422