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 maxvnodes; 120 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 121 &maxvnodes, 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 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode)); 173 factor2 = 30 * (sizeof(struct vm_object) + sizeof(struct vnode)); 174 maxvnodes = imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 175 KvaSize / factor2); 176 maxvnodes = imax(maxvnodes, maxproc * 8); 177 178 lwkt_token_init(&spechash_token, "spechash"); 179 } 180 181 /* 182 * Knob to control the precision of file timestamps: 183 * 184 * 0 = seconds only; nanoseconds zeroed. 185 * 1 = seconds and nanoseconds, accurate within 1/HZ. 186 * 2 = seconds and nanoseconds, truncated to microseconds. 187 * >=3 = seconds and nanoseconds, maximum precision. 188 */ 189 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 190 191 static int timestamp_precision = TSP_SEC; 192 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 193 ×tamp_precision, 0, "Precision of file timestamps"); 194 195 /* 196 * Get a current timestamp. 197 * 198 * MPSAFE 199 */ 200 void 201 vfs_timestamp(struct timespec *tsp) 202 { 203 struct timeval tv; 204 205 switch (timestamp_precision) { 206 case TSP_SEC: 207 tsp->tv_sec = time_second; 208 tsp->tv_nsec = 0; 209 break; 210 case TSP_HZ: 211 getnanotime(tsp); 212 break; 213 case TSP_USEC: 214 microtime(&tv); 215 TIMEVAL_TO_TIMESPEC(&tv, tsp); 216 break; 217 case TSP_NSEC: 218 default: 219 nanotime(tsp); 220 break; 221 } 222 } 223 224 /* 225 * Set vnode attributes to VNOVAL 226 */ 227 void 228 vattr_null(struct vattr *vap) 229 { 230 vap->va_type = VNON; 231 vap->va_size = VNOVAL; 232 vap->va_bytes = VNOVAL; 233 vap->va_mode = VNOVAL; 234 vap->va_nlink = VNOVAL; 235 vap->va_uid = VNOVAL; 236 vap->va_gid = VNOVAL; 237 vap->va_fsid = VNOVAL; 238 vap->va_fileid = VNOVAL; 239 vap->va_blocksize = VNOVAL; 240 vap->va_rmajor = VNOVAL; 241 vap->va_rminor = VNOVAL; 242 vap->va_atime.tv_sec = VNOVAL; 243 vap->va_atime.tv_nsec = VNOVAL; 244 vap->va_mtime.tv_sec = VNOVAL; 245 vap->va_mtime.tv_nsec = VNOVAL; 246 vap->va_ctime.tv_sec = VNOVAL; 247 vap->va_ctime.tv_nsec = VNOVAL; 248 vap->va_flags = VNOVAL; 249 vap->va_gen = VNOVAL; 250 vap->va_vaflags = 0; 251 /* va_*_uuid fields are only valid if related flags are set */ 252 } 253 254 /* 255 * Flush out and invalidate all buffers associated with a vnode. 256 * 257 * vp must be locked. 258 */ 259 static int vinvalbuf_bp(struct buf *bp, void *data); 260 261 struct vinvalbuf_bp_info { 262 struct vnode *vp; 263 int slptimeo; 264 int lkflags; 265 int flags; 266 int clean; 267 }; 268 269 int 270 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 271 { 272 struct vinvalbuf_bp_info info; 273 vm_object_t object; 274 int error; 275 276 lwkt_gettoken(&vp->v_token); 277 278 /* 279 * If we are being asked to save, call fsync to ensure that the inode 280 * is updated. 281 */ 282 if (flags & V_SAVE) { 283 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 284 if (error) 285 goto done; 286 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 287 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 288 goto done; 289 #if 0 290 /* 291 * Dirty bufs may be left or generated via races 292 * in circumstances where vinvalbuf() is called on 293 * a vnode not undergoing reclamation. Only 294 * panic if we are trying to reclaim the vnode. 295 */ 296 if ((vp->v_flag & VRECLAIMED) && 297 (bio_track_active(&vp->v_track_write) || 298 !RB_EMPTY(&vp->v_rbdirty_tree))) { 299 panic("vinvalbuf: dirty bufs"); 300 } 301 #endif 302 } 303 } 304 info.slptimeo = slptimeo; 305 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 306 if (slpflag & PCATCH) 307 info.lkflags |= LK_PCATCH; 308 info.flags = flags; 309 info.vp = vp; 310 311 /* 312 * Flush the buffer cache until nothing is left, wait for all I/O 313 * to complete. At least one pass is required. We might block 314 * in the pip code so we have to re-check. Order is important. 315 */ 316 do { 317 /* 318 * Flush buffer cache 319 */ 320 if (!RB_EMPTY(&vp->v_rbclean_tree)) { 321 info.clean = 1; 322 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 323 NULL, vinvalbuf_bp, &info); 324 } 325 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 326 info.clean = 0; 327 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 328 NULL, vinvalbuf_bp, &info); 329 } 330 331 /* 332 * Wait for I/O completion. 333 */ 334 bio_track_wait(&vp->v_track_write, 0, 0); 335 if ((object = vp->v_object) != NULL) 336 refcount_wait(&object->paging_in_progress, "vnvlbx"); 337 } while (bio_track_active(&vp->v_track_write) || 338 !RB_EMPTY(&vp->v_rbclean_tree) || 339 !RB_EMPTY(&vp->v_rbdirty_tree)); 340 341 /* 342 * Destroy the copy in the VM cache, too. 343 */ 344 if ((object = vp->v_object) != NULL) { 345 vm_object_page_remove(object, 0, 0, 346 (flags & V_SAVE) ? TRUE : FALSE); 347 } 348 349 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 350 panic("vinvalbuf: flush failed"); 351 if (!RB_EMPTY(&vp->v_rbhash_tree)) 352 panic("vinvalbuf: flush failed, buffers still present"); 353 error = 0; 354 done: 355 lwkt_reltoken(&vp->v_token); 356 return (error); 357 } 358 359 static int 360 vinvalbuf_bp(struct buf *bp, void *data) 361 { 362 struct vinvalbuf_bp_info *info = data; 363 int error; 364 365 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 366 atomic_add_int(&bp->b_refs, 1); 367 error = BUF_TIMELOCK(bp, info->lkflags, 368 "vinvalbuf", info->slptimeo); 369 atomic_subtract_int(&bp->b_refs, 1); 370 if (error == 0) { 371 BUF_UNLOCK(bp); 372 error = ENOLCK; 373 } 374 if (error == ENOLCK) 375 return(0); 376 return (-error); 377 } 378 KKASSERT(bp->b_vp == info->vp); 379 380 /* 381 * Must check clean/dirty status after successfully locking as 382 * it may race. 383 */ 384 if ((info->clean && (bp->b_flags & B_DELWRI)) || 385 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 386 BUF_UNLOCK(bp); 387 return(0); 388 } 389 390 /* 391 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 392 * check. This code will write out the buffer, period. 393 */ 394 bremfree(bp); 395 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 396 (info->flags & V_SAVE)) { 397 cluster_awrite(bp); 398 } else if (info->flags & V_SAVE) { 399 /* 400 * Cannot set B_NOCACHE on a clean buffer as this will 401 * destroy the VM backing store which might actually 402 * be dirty (and unsynchronized). 403 */ 404 bp->b_flags |= (B_INVAL | B_RELBUF); 405 brelse(bp); 406 } else { 407 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 408 brelse(bp); 409 } 410 return(0); 411 } 412 413 /* 414 * Truncate a file's buffer and pages to a specified length. This 415 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 416 * sync activity. 417 * 418 * The vnode must be locked. 419 */ 420 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 421 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 422 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 423 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 424 425 struct vtruncbuf_info { 426 struct vnode *vp; 427 off_t truncloffset; 428 int clean; 429 }; 430 431 int 432 vtruncbuf(struct vnode *vp, off_t length, int blksize) 433 { 434 struct vtruncbuf_info info; 435 const char *filename; 436 int count; 437 438 /* 439 * Round up to the *next* block, then destroy the buffers in question. 440 * Since we are only removing some of the buffers we must rely on the 441 * scan count to determine whether a loop is necessary. 442 */ 443 if ((count = (int)(length % blksize)) != 0) 444 info.truncloffset = length + (blksize - count); 445 else 446 info.truncloffset = length; 447 info.vp = vp; 448 449 lwkt_gettoken(&vp->v_token); 450 do { 451 info.clean = 1; 452 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 453 vtruncbuf_bp_trunc_cmp, 454 vtruncbuf_bp_trunc, &info); 455 info.clean = 0; 456 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 457 vtruncbuf_bp_trunc_cmp, 458 vtruncbuf_bp_trunc, &info); 459 } while(count); 460 461 /* 462 * For safety, fsync any remaining metadata if the file is not being 463 * truncated to 0. Since the metadata does not represent the entire 464 * dirty list we have to rely on the hit count to ensure that we get 465 * all of it. 466 */ 467 if (length > 0) { 468 do { 469 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 470 vtruncbuf_bp_metasync_cmp, 471 vtruncbuf_bp_metasync, &info); 472 } while (count); 473 } 474 475 /* 476 * Clean out any left over VM backing store. 477 * 478 * It is possible to have in-progress I/O from buffers that were 479 * not part of the truncation. This should not happen if we 480 * are truncating to 0-length. 481 */ 482 vnode_pager_setsize(vp, length); 483 bio_track_wait(&vp->v_track_write, 0, 0); 484 485 /* 486 * Debugging only 487 */ 488 spin_lock(&vp->v_spin); 489 filename = TAILQ_FIRST(&vp->v_namecache) ? 490 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 491 spin_unlock(&vp->v_spin); 492 493 /* 494 * Make sure no buffers were instantiated while we were trying 495 * to clean out the remaining VM pages. This could occur due 496 * to busy dirty VM pages being flushed out to disk. 497 */ 498 do { 499 info.clean = 1; 500 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 501 vtruncbuf_bp_trunc_cmp, 502 vtruncbuf_bp_trunc, &info); 503 info.clean = 0; 504 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 505 vtruncbuf_bp_trunc_cmp, 506 vtruncbuf_bp_trunc, &info); 507 if (count) { 508 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 509 "left over buffers in %s\n", count, filename); 510 } 511 } while(count); 512 513 lwkt_reltoken(&vp->v_token); 514 515 return (0); 516 } 517 518 /* 519 * The callback buffer is beyond the new file EOF and must be destroyed. 520 * Note that the compare function must conform to the RB_SCAN's requirements. 521 */ 522 static 523 int 524 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 525 { 526 struct vtruncbuf_info *info = data; 527 528 if (bp->b_loffset >= info->truncloffset) 529 return(0); 530 return(-1); 531 } 532 533 static 534 int 535 vtruncbuf_bp_trunc(struct buf *bp, void *data) 536 { 537 struct vtruncbuf_info *info = data; 538 539 /* 540 * Do not try to use a buffer we cannot immediately lock, but sleep 541 * anyway to prevent a livelock. The code will loop until all buffers 542 * can be acted upon. 543 * 544 * We must always revalidate the buffer after locking it to deal 545 * with MP races. 546 */ 547 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 548 atomic_add_int(&bp->b_refs, 1); 549 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 550 BUF_UNLOCK(bp); 551 atomic_subtract_int(&bp->b_refs, 1); 552 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 553 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 554 bp->b_vp != info->vp || 555 vtruncbuf_bp_trunc_cmp(bp, data)) { 556 BUF_UNLOCK(bp); 557 } else { 558 bremfree(bp); 559 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 560 brelse(bp); 561 } 562 return(1); 563 } 564 565 /* 566 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 567 * blocks (with a negative loffset) are scanned. 568 * Note that the compare function must conform to the RB_SCAN's requirements. 569 */ 570 static int 571 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 572 { 573 if (bp->b_loffset < 0) 574 return(0); 575 return(1); 576 } 577 578 static int 579 vtruncbuf_bp_metasync(struct buf *bp, void *data) 580 { 581 struct vtruncbuf_info *info = data; 582 583 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 584 atomic_add_int(&bp->b_refs, 1); 585 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 586 BUF_UNLOCK(bp); 587 atomic_subtract_int(&bp->b_refs, 1); 588 } else if ((bp->b_flags & B_DELWRI) == 0 || 589 bp->b_vp != info->vp || 590 vtruncbuf_bp_metasync_cmp(bp, data)) { 591 BUF_UNLOCK(bp); 592 } else { 593 bremfree(bp); 594 if (bp->b_vp == info->vp) 595 bawrite(bp); 596 else 597 bwrite(bp); 598 } 599 return(1); 600 } 601 602 /* 603 * vfsync - implements a multipass fsync on a file which understands 604 * dependancies and meta-data. The passed vnode must be locked. The 605 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 606 * 607 * When fsyncing data asynchronously just do one consolidated pass starting 608 * with the most negative block number. This may not get all the data due 609 * to dependancies. 610 * 611 * When fsyncing data synchronously do a data pass, then a metadata pass, 612 * then do additional data+metadata passes to try to get all the data out. 613 * 614 * Caller must ref the vnode but does not have to lock it. 615 */ 616 static int vfsync_wait_output(struct vnode *vp, 617 int (*waitoutput)(struct vnode *, struct thread *)); 618 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 619 static int vfsync_data_only_cmp(struct buf *bp, void *data); 620 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 621 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 622 static int vfsync_bp(struct buf *bp, void *data); 623 624 struct vfsync_info { 625 struct vnode *vp; 626 int fastpass; 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 info.fastpass = 1; 694 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 695 vfsync_bp, &info); 696 info.fastpass = 0; 697 error = vfsync_wait_output(vp, waitoutput); 698 if (error == 0) { 699 info.skippedbufs = 0; 700 info.cmpfunc = vfsync_dummy_cmp; 701 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 702 vfsync_bp, &info); 703 error = vfsync_wait_output(vp, waitoutput); 704 if (info.skippedbufs) { 705 kprintf("Warning: vfsync skipped %d dirty " 706 "buf%s in pass2!\n", 707 info.skippedbufs, 708 ((info.skippedbufs > 1) ? "s" : "")); 709 } 710 } 711 while (error == 0 && passes > 0 && 712 !RB_EMPTY(&vp->v_rbdirty_tree) 713 ) { 714 info.skippedbufs = 0; 715 if (--passes == 0) { 716 info.synchronous = 1; 717 info.syncdeps = 1; 718 } 719 info.cmpfunc = vfsync_dummy_cmp; 720 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 721 vfsync_bp, &info); 722 if (error < 0) 723 error = -error; 724 info.syncdeps = 1; 725 if (error == 0) 726 error = vfsync_wait_output(vp, waitoutput); 727 if (info.skippedbufs && passes == 0) { 728 kprintf("Warning: vfsync skipped %d dirty " 729 "buf%s in final pass!\n", 730 info.skippedbufs, 731 ((info.skippedbufs > 1) ? "s" : "")); 732 } 733 } 734 if (!RB_EMPTY(&vp->v_rbdirty_tree)) 735 kprintf("dirty bufs left after final pass\n"); 736 break; 737 } 738 lwkt_reltoken(&vp->v_token); 739 740 return(error); 741 } 742 743 static int 744 vfsync_wait_output(struct vnode *vp, 745 int (*waitoutput)(struct vnode *, struct thread *)) 746 { 747 int error; 748 749 error = bio_track_wait(&vp->v_track_write, 0, 0); 750 if (waitoutput) 751 error = waitoutput(vp, curthread); 752 return(error); 753 } 754 755 static int 756 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 757 { 758 return(0); 759 } 760 761 static int 762 vfsync_data_only_cmp(struct buf *bp, void *data) 763 { 764 if (bp->b_loffset < 0) 765 return(-1); 766 return(0); 767 } 768 769 static int 770 vfsync_meta_only_cmp(struct buf *bp, void *data) 771 { 772 if (bp->b_loffset < 0) 773 return(0); 774 return(1); 775 } 776 777 static int 778 vfsync_lazy_range_cmp(struct buf *bp, void *data) 779 { 780 struct vfsync_info *info = data; 781 782 if (bp->b_loffset < info->vp->v_lazyw) 783 return(-1); 784 return(0); 785 } 786 787 static int 788 vfsync_bp(struct buf *bp, void *data) 789 { 790 struct vfsync_info *info = data; 791 struct vnode *vp = info->vp; 792 int error; 793 794 if (info->fastpass) { 795 /* 796 * Ignore buffers that we cannot immediately lock. 797 */ 798 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 799 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst1", 1)) { 800 ++info->skippedbufs; 801 return(0); 802 } 803 } 804 } else if (info->synchronous == 0) { 805 /* 806 * Normal pass, give the buffer a little time to become 807 * available to us. 808 */ 809 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst2", hz / 10)) { 810 ++info->skippedbufs; 811 return(0); 812 } 813 } else { 814 /* 815 * Synchronous pass, give the buffer a lot of time before 816 * giving up. 817 */ 818 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst3", hz * 10)) { 819 ++info->skippedbufs; 820 return(0); 821 } 822 } 823 824 /* 825 * We must revalidate the buffer after locking. 826 */ 827 if ((bp->b_flags & B_DELWRI) == 0 || 828 bp->b_vp != info->vp || 829 info->cmpfunc(bp, data)) { 830 BUF_UNLOCK(bp); 831 return(0); 832 } 833 834 /* 835 * If syncdeps is not set we do not try to write buffers which have 836 * dependancies. 837 */ 838 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 839 BUF_UNLOCK(bp); 840 return(0); 841 } 842 843 /* 844 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 845 * has been written but an additional handshake with the device 846 * is required before we can dispose of the buffer. We have no idea 847 * how to do this so we have to skip these buffers. 848 */ 849 if (bp->b_flags & B_NEEDCOMMIT) { 850 BUF_UNLOCK(bp); 851 return(0); 852 } 853 854 /* 855 * Ask bioops if it is ok to sync. If not the VFS may have 856 * set B_LOCKED so we have to cycle the buffer. 857 */ 858 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 859 bremfree(bp); 860 brelse(bp); 861 return(0); 862 } 863 864 if (info->synchronous) { 865 /* 866 * Synchronous flushing. An error may be returned. 867 */ 868 bremfree(bp); 869 error = bwrite(bp); 870 } else { 871 /* 872 * Asynchronous flushing. A negative return value simply 873 * stops the scan and is not considered an error. We use 874 * this to support limited MNT_LAZY flushes. 875 */ 876 vp->v_lazyw = bp->b_loffset; 877 bremfree(bp); 878 info->lazycount += cluster_awrite(bp); 879 waitrunningbufspace(); 880 vm_wait_nominal(); 881 if (info->lazylimit && info->lazycount >= info->lazylimit) 882 error = 1; 883 else 884 error = 0; 885 } 886 return(-error); 887 } 888 889 /* 890 * Associate a buffer with a vnode. 891 * 892 * MPSAFE 893 */ 894 int 895 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 896 { 897 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 898 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 899 900 /* 901 * Insert onto list for new vnode. 902 */ 903 lwkt_gettoken(&vp->v_token); 904 905 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 906 lwkt_reltoken(&vp->v_token); 907 return (EEXIST); 908 } 909 910 /* 911 * Diagnostics (mainly for HAMMER debugging). Check for 912 * overlapping buffers. 913 */ 914 if (check_buf_overlap) { 915 struct buf *bx; 916 bx = buf_rb_hash_RB_PREV(bp); 917 if (bx) { 918 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 919 kprintf("bgetvp: overlapl %016jx/%d %016jx " 920 "bx %p bp %p\n", 921 (intmax_t)bx->b_loffset, 922 bx->b_bufsize, 923 (intmax_t)bp->b_loffset, 924 bx, bp); 925 if (check_buf_overlap > 1) 926 panic("bgetvp - overlapping buffer"); 927 } 928 } 929 bx = buf_rb_hash_RB_NEXT(bp); 930 if (bx) { 931 if (bp->b_loffset + testsize > bx->b_loffset) { 932 kprintf("bgetvp: overlapr %016jx/%d %016jx " 933 "bp %p bx %p\n", 934 (intmax_t)bp->b_loffset, 935 testsize, 936 (intmax_t)bx->b_loffset, 937 bp, bx); 938 if (check_buf_overlap > 1) 939 panic("bgetvp - overlapping buffer"); 940 } 941 } 942 } 943 bp->b_vp = vp; 944 bp->b_flags |= B_HASHED; 945 bp->b_flags |= B_VNCLEAN; 946 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 947 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 948 /*vhold(vp);*/ 949 lwkt_reltoken(&vp->v_token); 950 return(0); 951 } 952 953 /* 954 * Disassociate a buffer from a vnode. 955 * 956 * MPSAFE 957 */ 958 void 959 brelvp(struct buf *bp) 960 { 961 struct vnode *vp; 962 963 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 964 965 /* 966 * Delete from old vnode list, if on one. 967 */ 968 vp = bp->b_vp; 969 lwkt_gettoken(&vp->v_token); 970 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 971 if (bp->b_flags & B_VNDIRTY) 972 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 973 else 974 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 975 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 976 } 977 if (bp->b_flags & B_HASHED) { 978 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 979 bp->b_flags &= ~B_HASHED; 980 } 981 982 /* 983 * Only remove from synclist when no dirty buffers are left AND 984 * the VFS has not flagged the vnode's inode as being dirty. 985 */ 986 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 987 RB_EMPTY(&vp->v_rbdirty_tree)) { 988 vn_syncer_remove(vp, 0); 989 } 990 bp->b_vp = NULL; 991 992 lwkt_reltoken(&vp->v_token); 993 994 /*vdrop(vp);*/ 995 } 996 997 /* 998 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 999 * This routine is called when the state of the B_DELWRI bit is changed. 1000 * 1001 * Must be called with vp->v_token held. 1002 * MPSAFE 1003 */ 1004 void 1005 reassignbuf(struct buf *bp) 1006 { 1007 struct vnode *vp = bp->b_vp; 1008 int delay; 1009 1010 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 1011 ++reassignbufcalls; 1012 1013 /* 1014 * B_PAGING flagged buffers cannot be reassigned because their vp 1015 * is not fully linked in. 1016 */ 1017 if (bp->b_flags & B_PAGING) 1018 panic("cannot reassign paging buffer"); 1019 1020 if (bp->b_flags & B_DELWRI) { 1021 /* 1022 * Move to the dirty list, add the vnode to the worklist 1023 */ 1024 if (bp->b_flags & B_VNCLEAN) { 1025 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1026 bp->b_flags &= ~B_VNCLEAN; 1027 } 1028 if ((bp->b_flags & B_VNDIRTY) == 0) { 1029 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 1030 panic("reassignbuf: dup lblk vp %p bp %p", 1031 vp, bp); 1032 } 1033 bp->b_flags |= B_VNDIRTY; 1034 } 1035 if ((vp->v_flag & VONWORKLST) == 0) { 1036 switch (vp->v_type) { 1037 case VDIR: 1038 delay = dirdelay; 1039 break; 1040 case VCHR: 1041 case VBLK: 1042 if (vp->v_rdev && 1043 vp->v_rdev->si_mountpoint != NULL) { 1044 delay = metadelay; 1045 break; 1046 } 1047 /* fall through */ 1048 default: 1049 delay = filedelay; 1050 } 1051 vn_syncer_add(vp, delay); 1052 } 1053 } else { 1054 /* 1055 * Move to the clean list, remove the vnode from the worklist 1056 * if no dirty blocks remain. 1057 */ 1058 if (bp->b_flags & B_VNDIRTY) { 1059 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1060 bp->b_flags &= ~B_VNDIRTY; 1061 } 1062 if ((bp->b_flags & B_VNCLEAN) == 0) { 1063 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1064 panic("reassignbuf: dup lblk vp %p bp %p", 1065 vp, bp); 1066 } 1067 bp->b_flags |= B_VNCLEAN; 1068 } 1069 1070 /* 1071 * Only remove from synclist when no dirty buffers are left 1072 * AND the VFS has not flagged the vnode's inode as being 1073 * dirty. 1074 */ 1075 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1076 VONWORKLST && 1077 RB_EMPTY(&vp->v_rbdirty_tree)) { 1078 vn_syncer_remove(vp, 0); 1079 } 1080 } 1081 } 1082 1083 /* 1084 * Create a vnode for a block device. Used for mounting the root file 1085 * system. 1086 * 1087 * A vref()'d vnode is returned. 1088 */ 1089 extern struct vop_ops *devfs_vnode_dev_vops_p; 1090 int 1091 bdevvp(cdev_t dev, struct vnode **vpp) 1092 { 1093 struct vnode *vp; 1094 struct vnode *nvp; 1095 int error; 1096 1097 if (dev == NULL) { 1098 *vpp = NULLVP; 1099 return (ENXIO); 1100 } 1101 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1102 &nvp, 0, 0); 1103 if (error) { 1104 *vpp = NULLVP; 1105 return (error); 1106 } 1107 vp = nvp; 1108 vp->v_type = VCHR; 1109 #if 0 1110 vp->v_rdev = dev; 1111 #endif 1112 v_associate_rdev(vp, dev); 1113 vp->v_umajor = dev->si_umajor; 1114 vp->v_uminor = dev->si_uminor; 1115 vx_unlock(vp); 1116 *vpp = vp; 1117 return (0); 1118 } 1119 1120 int 1121 v_associate_rdev(struct vnode *vp, cdev_t dev) 1122 { 1123 if (dev == NULL) 1124 return(ENXIO); 1125 if (dev_is_good(dev) == 0) 1126 return(ENXIO); 1127 KKASSERT(vp->v_rdev == NULL); 1128 vp->v_rdev = reference_dev(dev); 1129 lwkt_gettoken(&spechash_token); 1130 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1131 lwkt_reltoken(&spechash_token); 1132 return(0); 1133 } 1134 1135 void 1136 v_release_rdev(struct vnode *vp) 1137 { 1138 cdev_t dev; 1139 1140 if ((dev = vp->v_rdev) != NULL) { 1141 lwkt_gettoken(&spechash_token); 1142 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1143 vp->v_rdev = NULL; 1144 release_dev(dev); 1145 lwkt_reltoken(&spechash_token); 1146 } 1147 } 1148 1149 /* 1150 * Add a vnode to the alias list hung off the cdev_t. We only associate 1151 * the device number with the vnode. The actual device is not associated 1152 * until the vnode is opened (usually in spec_open()), and will be 1153 * disassociated on last close. 1154 */ 1155 void 1156 addaliasu(struct vnode *nvp, int x, int y) 1157 { 1158 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1159 panic("addaliasu on non-special vnode"); 1160 nvp->v_umajor = x; 1161 nvp->v_uminor = y; 1162 } 1163 1164 /* 1165 * Simple call that a filesystem can make to try to get rid of a 1166 * vnode. It will fail if anyone is referencing the vnode (including 1167 * the caller). 1168 * 1169 * The filesystem can check whether its in-memory inode structure still 1170 * references the vp on return. 1171 * 1172 * May only be called if the vnode is in a known state (i.e. being prevented 1173 * from being deallocated by some other condition such as a vfs inode hold). 1174 */ 1175 void 1176 vclean_unlocked(struct vnode *vp) 1177 { 1178 vx_get(vp); 1179 if (VREFCNT(vp) <= 1) 1180 vgone_vxlocked(vp); 1181 vx_put(vp); 1182 } 1183 1184 /* 1185 * Disassociate a vnode from its underlying filesystem. 1186 * 1187 * The vnode must be VX locked and referenced. In all normal situations 1188 * there are no active references. If vclean_vxlocked() is called while 1189 * there are active references, the vnode is being ripped out and we have 1190 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1191 */ 1192 void 1193 vclean_vxlocked(struct vnode *vp, int flags) 1194 { 1195 int active; 1196 int n; 1197 vm_object_t object; 1198 struct namecache *ncp; 1199 1200 /* 1201 * If the vnode has already been reclaimed we have nothing to do. 1202 */ 1203 if (vp->v_flag & VRECLAIMED) 1204 return; 1205 1206 /* 1207 * Set flag to interlock operation, flag finalization to ensure 1208 * that the vnode winds up on the inactive list, and set v_act to 0. 1209 */ 1210 vsetflags(vp, VRECLAIMED); 1211 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1212 vp->v_act = 0; 1213 1214 if (verbose_reclaims) { 1215 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1216 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1217 } 1218 1219 /* 1220 * Scrap the vfs cache 1221 */ 1222 while (cache_inval_vp(vp, 0) != 0) { 1223 kprintf("Warning: vnode %p clean/cache_resolution " 1224 "race detected\n", vp); 1225 tsleep(vp, 0, "vclninv", 2); 1226 } 1227 1228 /* 1229 * Check to see if the vnode is in use. If so we have to reference it 1230 * before we clean it out so that its count cannot fall to zero and 1231 * generate a race against ourselves to recycle it. 1232 */ 1233 active = (VREFCNT(vp) > 0); 1234 1235 /* 1236 * Clean out any buffers associated with the vnode and destroy its 1237 * object, if it has one. 1238 */ 1239 vinvalbuf(vp, V_SAVE, 0, 0); 1240 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1241 1242 /* 1243 * If purging an active vnode (typically during a forced unmount 1244 * or reboot), it must be closed and deactivated before being 1245 * reclaimed. This isn't really all that safe, but what can 1246 * we do? XXX. 1247 * 1248 * Note that neither of these routines unlocks the vnode. 1249 */ 1250 if (active && (flags & DOCLOSE)) { 1251 while ((n = vp->v_opencount) != 0) { 1252 if (vp->v_writecount) 1253 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1254 else 1255 VOP_CLOSE(vp, FNONBLOCK, NULL); 1256 if (vp->v_opencount == n) { 1257 kprintf("Warning: unable to force-close" 1258 " vnode %p\n", vp); 1259 break; 1260 } 1261 } 1262 } 1263 1264 /* 1265 * If the vnode has not been deactivated, deactivated it. Deactivation 1266 * can create new buffers and VM pages so we have to call vinvalbuf() 1267 * again to make sure they all get flushed. 1268 * 1269 * This can occur if a file with a link count of 0 needs to be 1270 * truncated. 1271 * 1272 * If the vnode is already dead don't try to deactivate it. 1273 */ 1274 if ((vp->v_flag & VINACTIVE) == 0) { 1275 vsetflags(vp, VINACTIVE); 1276 if (vp->v_mount) 1277 VOP_INACTIVE(vp); 1278 vinvalbuf(vp, V_SAVE, 0, 0); 1279 } 1280 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1281 1282 /* 1283 * If the vnode has an object, destroy it. 1284 */ 1285 while ((object = vp->v_object) != NULL) { 1286 vm_object_hold(object); 1287 if (object == vp->v_object) 1288 break; 1289 vm_object_drop(object); 1290 } 1291 1292 if (object != NULL) { 1293 if (object->ref_count == 0) { 1294 if ((object->flags & OBJ_DEAD) == 0) 1295 vm_object_terminate(object); 1296 vm_object_drop(object); 1297 vclrflags(vp, VOBJBUF); 1298 } else { 1299 vm_pager_deallocate(object); 1300 vclrflags(vp, VOBJBUF); 1301 vm_object_drop(object); 1302 } 1303 } 1304 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1305 1306 /* 1307 * Reclaim the vnode if not already dead. 1308 */ 1309 if (vp->v_mount && VOP_RECLAIM(vp)) 1310 panic("vclean: cannot reclaim"); 1311 1312 /* 1313 * Done with purge, notify sleepers of the grim news. 1314 */ 1315 vp->v_ops = &dead_vnode_vops_p; 1316 vn_gone(vp); 1317 vp->v_tag = VT_NON; 1318 1319 /* 1320 * If we are destroying an active vnode, reactivate it now that 1321 * we have reassociated it with deadfs. This prevents the system 1322 * from crashing on the vnode due to it being unexpectedly marked 1323 * as inactive or reclaimed. 1324 */ 1325 if (active && (flags & DOCLOSE)) { 1326 vclrflags(vp, VINACTIVE | VRECLAIMED); 1327 } 1328 } 1329 1330 /* 1331 * Eliminate all activity associated with the requested vnode 1332 * and with all vnodes aliased to the requested vnode. 1333 * 1334 * The vnode must be referenced but should not be locked. 1335 */ 1336 int 1337 vrevoke(struct vnode *vp, struct ucred *cred) 1338 { 1339 struct vnode *vq; 1340 struct vnode *vqn; 1341 cdev_t dev; 1342 int error; 1343 1344 /* 1345 * If the vnode has a device association, scrap all vnodes associated 1346 * with the device. Don't let the device disappear on us while we 1347 * are scrapping the vnodes. 1348 * 1349 * The passed vp will probably show up in the list, do not VX lock 1350 * it twice! 1351 * 1352 * Releasing the vnode's rdev here can mess up specfs's call to 1353 * device close, so don't do it. The vnode has been disassociated 1354 * and the device will be closed after the last ref on the related 1355 * fp goes away (if not still open by e.g. the kernel). 1356 */ 1357 if (vp->v_type != VCHR) { 1358 error = fdrevoke(vp, DTYPE_VNODE, cred); 1359 return (error); 1360 } 1361 if ((dev = vp->v_rdev) == NULL) { 1362 return(0); 1363 } 1364 reference_dev(dev); 1365 lwkt_gettoken(&spechash_token); 1366 1367 restart: 1368 vqn = SLIST_FIRST(&dev->si_hlist); 1369 if (vqn) 1370 vhold(vqn); 1371 while ((vq = vqn) != NULL) { 1372 if (VREFCNT(vq) > 0) { 1373 vref(vq); 1374 fdrevoke(vq, DTYPE_VNODE, cred); 1375 /*v_release_rdev(vq);*/ 1376 vrele(vq); 1377 if (vq->v_rdev != dev) { 1378 vdrop(vq); 1379 goto restart; 1380 } 1381 } 1382 vqn = SLIST_NEXT(vq, v_cdevnext); 1383 if (vqn) 1384 vhold(vqn); 1385 vdrop(vq); 1386 } 1387 lwkt_reltoken(&spechash_token); 1388 dev_drevoke(dev); 1389 release_dev(dev); 1390 return (0); 1391 } 1392 1393 /* 1394 * This is called when the object underlying a vnode is being destroyed, 1395 * such as in a remove(). Try to recycle the vnode immediately if the 1396 * only active reference is our reference. 1397 * 1398 * Directory vnodes in the namecache with children cannot be immediately 1399 * recycled because numerous VOP_N*() ops require them to be stable. 1400 * 1401 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1402 * function is a NOP if VRECLAIMED is already set. 1403 */ 1404 int 1405 vrecycle(struct vnode *vp) 1406 { 1407 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1408 if (cache_inval_vp_nonblock(vp)) 1409 return(0); 1410 vgone_vxlocked(vp); 1411 return (1); 1412 } 1413 return (0); 1414 } 1415 1416 /* 1417 * Return the maximum I/O size allowed for strategy calls on VP. 1418 * 1419 * If vp is VCHR or VBLK we dive the device, otherwise we use 1420 * the vp's mount info. 1421 * 1422 * The returned value is clamped at MAXPHYS as most callers cannot use 1423 * buffers larger than that size. 1424 */ 1425 int 1426 vmaxiosize(struct vnode *vp) 1427 { 1428 int maxiosize; 1429 1430 if (vp->v_type == VBLK || vp->v_type == VCHR) 1431 maxiosize = vp->v_rdev->si_iosize_max; 1432 else 1433 maxiosize = vp->v_mount->mnt_iosize_max; 1434 1435 if (maxiosize > MAXPHYS) 1436 maxiosize = MAXPHYS; 1437 return (maxiosize); 1438 } 1439 1440 /* 1441 * Eliminate all activity associated with a vnode in preparation for 1442 * destruction. 1443 * 1444 * The vnode must be VX locked and refd and will remain VX locked and refd 1445 * on return. This routine may be called with the vnode in any state, as 1446 * long as it is VX locked. The vnode will be cleaned out and marked 1447 * VRECLAIMED but will not actually be reused until all existing refs and 1448 * holds go away. 1449 * 1450 * NOTE: This routine may be called on a vnode which has not yet been 1451 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1452 * already been reclaimed. 1453 * 1454 * This routine is not responsible for placing us back on the freelist. 1455 * Instead, it happens automatically when the caller releases the VX lock 1456 * (assuming there aren't any other references). 1457 */ 1458 void 1459 vgone_vxlocked(struct vnode *vp) 1460 { 1461 /* 1462 * assert that the VX lock is held. This is an absolute requirement 1463 * now for vgone_vxlocked() to be called. 1464 */ 1465 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1466 1467 /* 1468 * Clean out the filesystem specific data and set the VRECLAIMED 1469 * bit. Also deactivate the vnode if necessary. 1470 * 1471 * The vnode should have automatically been removed from the syncer 1472 * list as syncer/dirty flags cleared during the cleaning. 1473 */ 1474 vclean_vxlocked(vp, DOCLOSE); 1475 1476 /* 1477 * Normally panic if the vnode is still dirty, unless we are doing 1478 * a forced unmount (tmpfs typically). 1479 */ 1480 if (vp->v_flag & VONWORKLST) { 1481 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) { 1482 /* force removal */ 1483 vn_syncer_remove(vp, 1); 1484 } else { 1485 panic("vp %p still dirty in vgone after flush", vp); 1486 } 1487 } 1488 1489 /* 1490 * Delete from old mount point vnode list, if on one. 1491 */ 1492 if (vp->v_mount != NULL) { 1493 KKASSERT(vp->v_data == NULL); 1494 insmntque(vp, NULL); 1495 } 1496 1497 /* 1498 * If special device, remove it from special device alias list 1499 * if it is on one. This should normally only occur if a vnode is 1500 * being revoked as the device should otherwise have been released 1501 * naturally. 1502 */ 1503 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1504 v_release_rdev(vp); 1505 } 1506 1507 /* 1508 * Set us to VBAD 1509 */ 1510 vp->v_type = VBAD; 1511 } 1512 1513 /* 1514 * Lookup a vnode by device number. 1515 * 1516 * Returns non-zero and *vpp set to a vref'd vnode on success. 1517 * Returns zero on failure. 1518 */ 1519 int 1520 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1521 { 1522 struct vnode *vp; 1523 1524 lwkt_gettoken(&spechash_token); 1525 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1526 if (type == vp->v_type) { 1527 *vpp = vp; 1528 vref(vp); 1529 lwkt_reltoken(&spechash_token); 1530 return (1); 1531 } 1532 } 1533 lwkt_reltoken(&spechash_token); 1534 return (0); 1535 } 1536 1537 /* 1538 * Calculate the total number of references to a special device. This 1539 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1540 * an overloaded field. Since udev2dev can now return NULL, we have 1541 * to check for a NULL v_rdev. 1542 */ 1543 int 1544 count_dev(cdev_t dev) 1545 { 1546 struct vnode *vp; 1547 int count = 0; 1548 1549 if (SLIST_FIRST(&dev->si_hlist)) { 1550 lwkt_gettoken(&spechash_token); 1551 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1552 count += vp->v_opencount; 1553 } 1554 lwkt_reltoken(&spechash_token); 1555 } 1556 return(count); 1557 } 1558 1559 int 1560 vcount(struct vnode *vp) 1561 { 1562 if (vp->v_rdev == NULL) 1563 return(0); 1564 return(count_dev(vp->v_rdev)); 1565 } 1566 1567 /* 1568 * Initialize VMIO for a vnode. This routine MUST be called before a 1569 * VFS can issue buffer cache ops on a vnode. It is typically called 1570 * when a vnode is initialized from its inode. 1571 */ 1572 int 1573 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1574 { 1575 vm_object_t object; 1576 int error = 0; 1577 1578 object = vp->v_object; 1579 if (object) { 1580 vm_object_hold(object); 1581 KKASSERT(vp->v_object == object); 1582 } 1583 1584 if (object == NULL) { 1585 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1586 1587 /* 1588 * Dereference the reference we just created. This assumes 1589 * that the object is associated with the vp. Allow it to 1590 * have zero refs. It cannot be destroyed as long as it 1591 * is associated with the vnode. 1592 */ 1593 vm_object_hold(object); 1594 atomic_add_int(&object->ref_count, -1); 1595 vrele(vp); 1596 } else { 1597 KKASSERT((object->flags & OBJ_DEAD) == 0); 1598 } 1599 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1600 vsetflags(vp, VOBJBUF); 1601 vm_object_drop(object); 1602 1603 return (error); 1604 } 1605 1606 1607 /* 1608 * Print out a description of a vnode. 1609 */ 1610 static char *typename[] = 1611 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1612 1613 void 1614 vprint(char *label, struct vnode *vp) 1615 { 1616 char buf[96]; 1617 1618 if (label != NULL) 1619 kprintf("%s: %p: ", label, (void *)vp); 1620 else 1621 kprintf("%p: ", (void *)vp); 1622 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1623 typename[vp->v_type], 1624 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1625 buf[0] = '\0'; 1626 if (vp->v_flag & VROOT) 1627 strcat(buf, "|VROOT"); 1628 if (vp->v_flag & VPFSROOT) 1629 strcat(buf, "|VPFSROOT"); 1630 if (vp->v_flag & VTEXT) 1631 strcat(buf, "|VTEXT"); 1632 if (vp->v_flag & VSYSTEM) 1633 strcat(buf, "|VSYSTEM"); 1634 if (vp->v_flag & VOBJBUF) 1635 strcat(buf, "|VOBJBUF"); 1636 if (buf[0] != '\0') 1637 kprintf(" flags (%s)", &buf[1]); 1638 if (vp->v_data == NULL) { 1639 kprintf("\n"); 1640 } else { 1641 kprintf("\n\t"); 1642 VOP_PRINT(vp); 1643 } 1644 } 1645 1646 /* 1647 * Do the usual access checking. 1648 * file_mode, uid and gid are from the vnode in question, 1649 * while acc_mode and cred are from the VOP_ACCESS parameter list 1650 */ 1651 int 1652 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1653 mode_t acc_mode, struct ucred *cred) 1654 { 1655 mode_t mask; 1656 int ismember; 1657 1658 /* 1659 * Super-user always gets read/write access, but execute access depends 1660 * on at least one execute bit being set. 1661 */ 1662 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1663 if ((acc_mode & VEXEC) && type != VDIR && 1664 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1665 return (EACCES); 1666 return (0); 1667 } 1668 1669 mask = 0; 1670 1671 /* Otherwise, check the owner. */ 1672 if (cred->cr_uid == uid) { 1673 if (acc_mode & VEXEC) 1674 mask |= S_IXUSR; 1675 if (acc_mode & VREAD) 1676 mask |= S_IRUSR; 1677 if (acc_mode & VWRITE) 1678 mask |= S_IWUSR; 1679 return ((file_mode & mask) == mask ? 0 : EACCES); 1680 } 1681 1682 /* Otherwise, check the groups. */ 1683 ismember = groupmember(gid, cred); 1684 if (cred->cr_svgid == gid || ismember) { 1685 if (acc_mode & VEXEC) 1686 mask |= S_IXGRP; 1687 if (acc_mode & VREAD) 1688 mask |= S_IRGRP; 1689 if (acc_mode & VWRITE) 1690 mask |= S_IWGRP; 1691 return ((file_mode & mask) == mask ? 0 : EACCES); 1692 } 1693 1694 /* Otherwise, check everyone else. */ 1695 if (acc_mode & VEXEC) 1696 mask |= S_IXOTH; 1697 if (acc_mode & VREAD) 1698 mask |= S_IROTH; 1699 if (acc_mode & VWRITE) 1700 mask |= S_IWOTH; 1701 return ((file_mode & mask) == mask ? 0 : EACCES); 1702 } 1703 1704 #ifdef DDB 1705 #include <ddb/ddb.h> 1706 1707 static int db_show_locked_vnodes(struct mount *mp, void *data); 1708 1709 /* 1710 * List all of the locked vnodes in the system. 1711 * Called when debugging the kernel. 1712 */ 1713 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1714 { 1715 kprintf("Locked vnodes\n"); 1716 mountlist_scan(db_show_locked_vnodes, NULL, 1717 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1718 } 1719 1720 static int 1721 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1722 { 1723 struct vnode *vp; 1724 1725 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1726 if (vn_islocked(vp)) 1727 vprint(NULL, vp); 1728 } 1729 return(0); 1730 } 1731 #endif 1732 1733 /* 1734 * Top level filesystem related information gathering. 1735 */ 1736 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1737 1738 static int 1739 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1740 { 1741 int *name = (int *)arg1 - 1; /* XXX */ 1742 u_int namelen = arg2 + 1; /* XXX */ 1743 struct vfsconf *vfsp; 1744 int maxtypenum; 1745 1746 #if 1 || defined(COMPAT_PRELITE2) 1747 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1748 if (namelen == 1) 1749 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1750 #endif 1751 1752 #ifdef notyet 1753 /* all sysctl names at this level are at least name and field */ 1754 if (namelen < 2) 1755 return (ENOTDIR); /* overloaded */ 1756 if (name[0] != VFS_GENERIC) { 1757 vfsp = vfsconf_find_by_typenum(name[0]); 1758 if (vfsp == NULL) 1759 return (EOPNOTSUPP); 1760 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1761 oldp, oldlenp, newp, newlen, p)); 1762 } 1763 #endif 1764 switch (name[1]) { 1765 case VFS_MAXTYPENUM: 1766 if (namelen != 2) 1767 return (ENOTDIR); 1768 maxtypenum = vfsconf_get_maxtypenum(); 1769 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1770 case VFS_CONF: 1771 if (namelen != 3) 1772 return (ENOTDIR); /* overloaded */ 1773 vfsp = vfsconf_find_by_typenum(name[2]); 1774 if (vfsp == NULL) 1775 return (EOPNOTSUPP); 1776 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1777 } 1778 return (EOPNOTSUPP); 1779 } 1780 1781 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1782 "Generic filesystem"); 1783 1784 #if 1 || defined(COMPAT_PRELITE2) 1785 1786 static int 1787 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1788 { 1789 int error; 1790 struct ovfsconf ovfs; 1791 struct sysctl_req *req = (struct sysctl_req*) data; 1792 1793 bzero(&ovfs, sizeof(ovfs)); 1794 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1795 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1796 ovfs.vfc_index = vfsp->vfc_typenum; 1797 ovfs.vfc_refcount = vfsp->vfc_refcount; 1798 ovfs.vfc_flags = vfsp->vfc_flags; 1799 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1800 if (error) 1801 return error; /* abort iteration with error code */ 1802 else 1803 return 0; /* continue iterating with next element */ 1804 } 1805 1806 static int 1807 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1808 { 1809 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1810 } 1811 1812 #endif /* 1 || COMPAT_PRELITE2 */ 1813 1814 /* 1815 * Check to see if a filesystem is mounted on a block device. 1816 */ 1817 int 1818 vfs_mountedon(struct vnode *vp) 1819 { 1820 cdev_t dev; 1821 1822 if ((dev = vp->v_rdev) == NULL) { 1823 /* if (vp->v_type != VBLK) 1824 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1825 } 1826 if (dev != NULL && dev->si_mountpoint) 1827 return (EBUSY); 1828 return (0); 1829 } 1830 1831 /* 1832 * Unmount all filesystems. The list is traversed in reverse order 1833 * of mounting to avoid dependencies. 1834 * 1835 * We want the umountall to be able to break out of its loop if a 1836 * failure occurs, after scanning all possible mounts, so the callback 1837 * returns 0 on error. 1838 * 1839 * NOTE: Do not call mountlist_remove(mp) on error any more, this will 1840 * confuse mountlist_scan()'s unbusy check. 1841 */ 1842 static int vfs_umountall_callback(struct mount *mp, void *data); 1843 1844 void 1845 vfs_unmountall(void) 1846 { 1847 int count; 1848 1849 do { 1850 count = mountlist_scan(vfs_umountall_callback, 1851 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1852 } while (count); 1853 } 1854 1855 static 1856 int 1857 vfs_umountall_callback(struct mount *mp, void *data) 1858 { 1859 int error; 1860 1861 error = dounmount(mp, MNT_FORCE); 1862 if (error) { 1863 kprintf("unmount of filesystem mounted from %s failed (", 1864 mp->mnt_stat.f_mntfromname); 1865 if (error == EBUSY) 1866 kprintf("BUSY)\n"); 1867 else 1868 kprintf("%d)\n", error); 1869 return 0; 1870 } else { 1871 return 1; 1872 } 1873 } 1874 1875 /* 1876 * Checks the mount flags for parameter mp and put the names comma-separated 1877 * into a string buffer buf with a size limit specified by len. 1878 * 1879 * It returns the number of bytes written into buf, and (*errorp) will be 1880 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1881 * not large enough). The buffer will be 0-terminated if len was not 0. 1882 */ 1883 size_t 1884 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1885 char *buf, size_t len, int *errorp) 1886 { 1887 static const struct mountctl_opt optnames[] = { 1888 { MNT_RDONLY, "read-only" }, 1889 { MNT_SYNCHRONOUS, "synchronous" }, 1890 { MNT_NOEXEC, "noexec" }, 1891 { MNT_NOSUID, "nosuid" }, 1892 { MNT_NODEV, "nodev" }, 1893 { MNT_AUTOMOUNTED, "automounted" }, 1894 { MNT_ASYNC, "asynchronous" }, 1895 { MNT_SUIDDIR, "suiddir" }, 1896 { MNT_SOFTDEP, "soft-updates" }, 1897 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1898 { MNT_TRIM, "trim" }, 1899 { MNT_NOATIME, "noatime" }, 1900 { MNT_NOCLUSTERR, "noclusterr" }, 1901 { MNT_NOCLUSTERW, "noclusterw" }, 1902 { MNT_EXRDONLY, "NFS read-only" }, 1903 { MNT_EXPORTED, "NFS exported" }, 1904 /* Remaining NFS flags could come here */ 1905 { MNT_LOCAL, "local" }, 1906 { MNT_QUOTA, "with-quotas" }, 1907 /* { MNT_ROOTFS, "rootfs" }, */ 1908 /* { MNT_IGNORE, "ignore" }, */ 1909 { 0, NULL} 1910 }; 1911 int bwritten; 1912 int bleft; 1913 int optlen; 1914 int actsize; 1915 1916 *errorp = 0; 1917 bwritten = 0; 1918 bleft = len - 1; /* leave room for trailing \0 */ 1919 1920 /* 1921 * Checks the size of the string. If it contains 1922 * any data, then we will append the new flags to 1923 * it. 1924 */ 1925 actsize = strlen(buf); 1926 if (actsize > 0) 1927 buf += actsize; 1928 1929 /* Default flags if no flags passed */ 1930 if (optp == NULL) 1931 optp = optnames; 1932 1933 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1934 *errorp = EINVAL; 1935 return(0); 1936 } 1937 1938 for (; flags && optp->o_opt; ++optp) { 1939 if ((flags & optp->o_opt) == 0) 1940 continue; 1941 optlen = strlen(optp->o_name); 1942 if (bwritten || actsize > 0) { 1943 if (bleft < 2) { 1944 *errorp = ENOSPC; 1945 break; 1946 } 1947 buf[bwritten++] = ','; 1948 buf[bwritten++] = ' '; 1949 bleft -= 2; 1950 } 1951 if (bleft < optlen) { 1952 *errorp = ENOSPC; 1953 break; 1954 } 1955 bcopy(optp->o_name, buf + bwritten, optlen); 1956 bwritten += optlen; 1957 bleft -= optlen; 1958 flags &= ~optp->o_opt; 1959 } 1960 1961 /* 1962 * Space already reserved for trailing \0 1963 */ 1964 buf[bwritten] = 0; 1965 return (bwritten); 1966 } 1967 1968 /* 1969 * Build hash lists of net addresses and hang them off the mount point. 1970 * Called by ufs_mount() to set up the lists of export addresses. 1971 */ 1972 static int 1973 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1974 const struct export_args *argp) 1975 { 1976 struct netcred *np; 1977 struct radix_node_head *rnh; 1978 int i; 1979 struct radix_node *rn; 1980 struct sockaddr *saddr, *smask = NULL; 1981 int error; 1982 1983 if (argp->ex_addrlen == 0) { 1984 if (mp->mnt_flag & MNT_DEFEXPORTED) 1985 return (EPERM); 1986 np = &nep->ne_defexported; 1987 np->netc_exflags = argp->ex_flags; 1988 np->netc_anon = argp->ex_anon; 1989 np->netc_anon.cr_ref = 1; 1990 mp->mnt_flag |= MNT_DEFEXPORTED; 1991 return (0); 1992 } 1993 1994 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1995 return (EINVAL); 1996 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1997 return (EINVAL); 1998 1999 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2000 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 2001 saddr = (struct sockaddr *) (np + 1); 2002 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2003 goto out; 2004 if (saddr->sa_len > argp->ex_addrlen) 2005 saddr->sa_len = argp->ex_addrlen; 2006 if (argp->ex_masklen) { 2007 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2008 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2009 if (error) 2010 goto out; 2011 if (smask->sa_len > argp->ex_masklen) 2012 smask->sa_len = argp->ex_masklen; 2013 } 2014 NE_LOCK(nep); 2015 if (nep->ne_maskhead == NULL) { 2016 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) { 2017 error = ENOBUFS; 2018 goto out; 2019 } 2020 } 2021 if((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) { 2022 error = ENOBUFS; 2023 goto out; 2024 } 2025 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 2026 np->netc_rnodes); 2027 NE_UNLOCK(nep); 2028 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */ 2029 error = EPERM; 2030 goto out; 2031 } 2032 np->netc_exflags = argp->ex_flags; 2033 np->netc_anon = argp->ex_anon; 2034 np->netc_anon.cr_ref = 1; 2035 return (0); 2036 out: 2037 kfree(np, M_NETADDR); 2038 return (error); 2039 } 2040 2041 /* ARGSUSED */ 2042 static int 2043 vfs_free_netcred(struct radix_node *rn, void *w) 2044 { 2045 struct radix_node_head *rnh = (struct radix_node_head *) w; 2046 2047 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2048 kfree((caddr_t) rn, M_NETADDR); 2049 return (0); 2050 } 2051 2052 static struct radix_node_head * 2053 vfs_create_addrlist_af(int af, struct netexport *nep) 2054 { 2055 struct radix_node_head *rnh = NULL; 2056 #if defined(INET) || defined(INET6) 2057 struct radix_node_head *maskhead = nep->ne_maskhead; 2058 int off; 2059 #endif 2060 2061 NE_ASSERT_LOCKED(nep); 2062 KKASSERT(maskhead != NULL); 2063 switch (af) { 2064 #ifdef INET 2065 case AF_INET: 2066 if ((rnh = nep->ne_inethead) == NULL) { 2067 off = offsetof(struct sockaddr_in, sin_addr) << 3; 2068 if (!rn_inithead((void **)&rnh, maskhead, off)) 2069 return (NULL); 2070 nep->ne_inethead = rnh; 2071 } 2072 break; 2073 #endif 2074 #ifdef INET6 2075 case AF_INET6: 2076 if ((rnh = nep->ne_inet6head) == NULL) { 2077 off = offsetof(struct sockaddr_in6, sin6_addr) << 3; 2078 if (!rn_inithead((void **)&rnh, maskhead, off)) 2079 return (NULL); 2080 nep->ne_inet6head = rnh; 2081 } 2082 break; 2083 #endif 2084 } 2085 return (rnh); 2086 } 2087 2088 static void 2089 vfs_free_addrlist_af(struct radix_node_head **prnh) 2090 { 2091 struct radix_node_head *rnh = *prnh; 2092 2093 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh); 2094 kfree(rnh, M_RTABLE); 2095 *prnh = NULL; 2096 } 2097 2098 /* 2099 * Free the net address hash lists that are hanging off the mount points. 2100 */ 2101 static void 2102 vfs_free_addrlist(struct netexport *nep) 2103 { 2104 NE_LOCK(nep); 2105 if (nep->ne_inethead != NULL) 2106 vfs_free_addrlist_af(&nep->ne_inethead); 2107 if (nep->ne_inet6head != NULL) 2108 vfs_free_addrlist_af(&nep->ne_inet6head); 2109 if (nep->ne_maskhead) 2110 vfs_free_addrlist_af(&nep->ne_maskhead); 2111 NE_UNLOCK(nep); 2112 } 2113 2114 int 2115 vfs_export(struct mount *mp, struct netexport *nep, 2116 const struct export_args *argp) 2117 { 2118 int error; 2119 2120 if (argp->ex_flags & MNT_DELEXPORT) { 2121 if (mp->mnt_flag & MNT_EXPUBLIC) { 2122 vfs_setpublicfs(NULL, NULL, NULL); 2123 mp->mnt_flag &= ~MNT_EXPUBLIC; 2124 } 2125 vfs_free_addrlist(nep); 2126 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2127 } 2128 if (argp->ex_flags & MNT_EXPORTED) { 2129 if (argp->ex_flags & MNT_EXPUBLIC) { 2130 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2131 return (error); 2132 mp->mnt_flag |= MNT_EXPUBLIC; 2133 } 2134 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2135 return (error); 2136 mp->mnt_flag |= MNT_EXPORTED; 2137 } 2138 return (0); 2139 } 2140 2141 2142 /* 2143 * Set the publicly exported filesystem (WebNFS). Currently, only 2144 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2145 */ 2146 int 2147 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2148 const struct export_args *argp) 2149 { 2150 int error; 2151 struct vnode *rvp; 2152 char *cp; 2153 2154 /* 2155 * mp == NULL -> invalidate the current info, the FS is 2156 * no longer exported. May be called from either vfs_export 2157 * or unmount, so check if it hasn't already been done. 2158 */ 2159 if (mp == NULL) { 2160 if (nfs_pub.np_valid) { 2161 nfs_pub.np_valid = 0; 2162 if (nfs_pub.np_index != NULL) { 2163 kfree(nfs_pub.np_index, M_TEMP); 2164 nfs_pub.np_index = NULL; 2165 } 2166 } 2167 return (0); 2168 } 2169 2170 /* 2171 * Only one allowed at a time. 2172 */ 2173 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2174 return (EBUSY); 2175 2176 /* 2177 * Get real filehandle for root of exported FS. 2178 */ 2179 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2180 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2181 2182 if ((error = VFS_ROOT(mp, &rvp))) 2183 return (error); 2184 2185 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2186 return (error); 2187 2188 vput(rvp); 2189 2190 /* 2191 * If an indexfile was specified, pull it in. 2192 */ 2193 if (argp->ex_indexfile != NULL) { 2194 int namelen; 2195 2196 error = vn_get_namelen(rvp, &namelen); 2197 if (error) 2198 return (error); 2199 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2200 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2201 namelen, NULL); 2202 if (!error) { 2203 /* 2204 * Check for illegal filenames. 2205 */ 2206 for (cp = nfs_pub.np_index; *cp; cp++) { 2207 if (*cp == '/') { 2208 error = EINVAL; 2209 break; 2210 } 2211 } 2212 } 2213 if (error) { 2214 kfree(nfs_pub.np_index, M_TEMP); 2215 return (error); 2216 } 2217 } 2218 2219 nfs_pub.np_mount = mp; 2220 nfs_pub.np_valid = 1; 2221 return (0); 2222 } 2223 2224 struct netcred * 2225 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2226 struct sockaddr *nam) 2227 { 2228 struct netcred *np; 2229 struct radix_node_head *rnh; 2230 struct sockaddr *saddr; 2231 2232 np = NULL; 2233 if (mp->mnt_flag & MNT_EXPORTED) { 2234 /* 2235 * Lookup in the export list first. 2236 */ 2237 NE_LOCK(nep); 2238 if (nam != NULL) { 2239 saddr = nam; 2240 switch (saddr->sa_family) { 2241 #ifdef INET 2242 case AF_INET: 2243 rnh = nep->ne_inethead; 2244 break; 2245 #endif 2246 #ifdef INET6 2247 case AF_INET6: 2248 rnh = nep->ne_inet6head; 2249 break; 2250 #endif 2251 default: 2252 rnh = NULL; 2253 } 2254 if (rnh != NULL) { 2255 np = (struct netcred *) 2256 (*rnh->rnh_matchaddr)((char *)saddr, 2257 rnh); 2258 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2259 np = NULL; 2260 } 2261 } 2262 NE_UNLOCK(nep); 2263 /* 2264 * If no address match, use the default if it exists. 2265 */ 2266 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2267 np = &nep->ne_defexported; 2268 } 2269 return (np); 2270 } 2271 2272 /* 2273 * perform msync on all vnodes under a mount point. The mount point must 2274 * be locked. This code is also responsible for lazy-freeing unreferenced 2275 * vnodes whos VM objects no longer contain pages. 2276 * 2277 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2278 * 2279 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2280 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2281 * way up in this high level function. 2282 */ 2283 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2284 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2285 2286 void 2287 vfs_msync(struct mount *mp, int flags) 2288 { 2289 int vmsc_flags; 2290 2291 /* 2292 * tmpfs sets this flag to prevent msync(), sync, and the 2293 * filesystem periodic syncer from trying to flush VM pages 2294 * to swap. Only pure memory pressure flushes tmpfs VM pages 2295 * to swap. 2296 */ 2297 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2298 return; 2299 2300 /* 2301 * Ok, scan the vnodes for work. If the filesystem is using the 2302 * syncer thread feature we can use vsyncscan() instead of 2303 * vmntvnodescan(), which is much faster. 2304 */ 2305 vmsc_flags = VMSC_GETVP; 2306 if (flags != MNT_WAIT) 2307 vmsc_flags |= VMSC_NOWAIT; 2308 2309 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2310 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2311 (void *)(intptr_t)flags); 2312 } else { 2313 vmntvnodescan(mp, vmsc_flags, 2314 vfs_msync_scan1, vfs_msync_scan2, 2315 (void *)(intptr_t)flags); 2316 } 2317 } 2318 2319 /* 2320 * scan1 is a fast pre-check. There could be hundreds of thousands of 2321 * vnodes, we cannot afford to do anything heavy weight until we have a 2322 * fairly good indication that there is work to do. 2323 */ 2324 static 2325 int 2326 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2327 { 2328 int flags = (int)(intptr_t)data; 2329 2330 if ((vp->v_flag & VRECLAIMED) == 0) { 2331 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2332 vp->v_object) { 2333 return(0); /* call scan2 */ 2334 } 2335 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2336 (vp->v_flag & VOBJDIRTY) && 2337 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2338 return(0); /* call scan2 */ 2339 } 2340 } 2341 2342 /* 2343 * do not call scan2, continue the loop 2344 */ 2345 return(-1); 2346 } 2347 2348 /* 2349 * This callback is handed a locked vnode. 2350 */ 2351 static 2352 int 2353 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2354 { 2355 vm_object_t obj; 2356 int flags = (int)(intptr_t)data; 2357 2358 if (vp->v_flag & VRECLAIMED) 2359 return(0); 2360 2361 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2362 if ((obj = vp->v_object) != NULL) { 2363 vm_object_page_clean(obj, 0, 0, 2364 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2365 } 2366 } 2367 return(0); 2368 } 2369 2370 /* 2371 * Wake up anyone interested in vp because it is being revoked. 2372 */ 2373 void 2374 vn_gone(struct vnode *vp) 2375 { 2376 lwkt_gettoken(&vp->v_token); 2377 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2378 lwkt_reltoken(&vp->v_token); 2379 } 2380 2381 /* 2382 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2383 * (or v_rdev might be NULL). 2384 */ 2385 cdev_t 2386 vn_todev(struct vnode *vp) 2387 { 2388 if (vp->v_type != VBLK && vp->v_type != VCHR) 2389 return (NULL); 2390 KKASSERT(vp->v_rdev != NULL); 2391 return (vp->v_rdev); 2392 } 2393 2394 /* 2395 * Check if vnode represents a disk device. The vnode does not need to be 2396 * opened. 2397 * 2398 * MPALMOSTSAFE 2399 */ 2400 int 2401 vn_isdisk(struct vnode *vp, int *errp) 2402 { 2403 cdev_t dev; 2404 2405 if (vp->v_type != VCHR) { 2406 if (errp != NULL) 2407 *errp = ENOTBLK; 2408 return (0); 2409 } 2410 2411 dev = vp->v_rdev; 2412 2413 if (dev == NULL) { 2414 if (errp != NULL) 2415 *errp = ENXIO; 2416 return (0); 2417 } 2418 if (dev_is_good(dev) == 0) { 2419 if (errp != NULL) 2420 *errp = ENXIO; 2421 return (0); 2422 } 2423 if ((dev_dflags(dev) & D_DISK) == 0) { 2424 if (errp != NULL) 2425 *errp = ENOTBLK; 2426 return (0); 2427 } 2428 if (errp != NULL) 2429 *errp = 0; 2430 return (1); 2431 } 2432 2433 int 2434 vn_get_namelen(struct vnode *vp, int *namelen) 2435 { 2436 int error; 2437 register_t retval[2]; 2438 2439 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2440 if (error) 2441 return (error); 2442 *namelen = (int)retval[0]; 2443 return (0); 2444 } 2445 2446 int 2447 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2448 uint16_t d_namlen, const char *d_name) 2449 { 2450 struct dirent *dp; 2451 size_t len; 2452 2453 len = _DIRENT_RECLEN(d_namlen); 2454 if (len > uio->uio_resid) 2455 return(1); 2456 2457 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2458 2459 dp->d_ino = d_ino; 2460 dp->d_namlen = d_namlen; 2461 dp->d_type = d_type; 2462 bcopy(d_name, dp->d_name, d_namlen); 2463 2464 *error = uiomove((caddr_t)dp, len, uio); 2465 2466 kfree(dp, M_TEMP); 2467 2468 return(0); 2469 } 2470 2471 void 2472 vn_mark_atime(struct vnode *vp, struct thread *td) 2473 { 2474 struct proc *p = td->td_proc; 2475 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2476 2477 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2478 VOP_MARKATIME(vp, cred); 2479 } 2480 } 2481 2482 /* 2483 * Calculate the number of entries in an inode-related chained hash table. 2484 * With today's memory sizes, maxvnodes can wind up being a very large 2485 * number. There is no reason to waste memory, so tolerate some stacking. 2486 */ 2487 int 2488 vfs_inodehashsize(void) 2489 { 2490 int hsize; 2491 2492 hsize = 32; 2493 while (hsize < maxvnodes) 2494 hsize <<= 1; 2495 while (hsize > maxvnodes * 2) 2496 hsize >>= 1; /* nominal 2x stacking */ 2497 2498 if (maxvnodes > 1024 * 1024) 2499 hsize >>= 1; /* nominal 8x stacking */ 2500 2501 if (maxvnodes > 128 * 1024) 2502 hsize >>= 1; /* nominal 4x stacking */ 2503 2504 if (hsize < 16) 2505 hsize = 16; 2506 2507 return hsize; 2508 } 2509