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/mplock2.h> 85 #include <vm/vm_page2.h> 86 87 #include <netinet/in.h> 88 89 static MALLOC_DEFINE(M_NETCRED, "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 0 735 /* 736 * This case can occur normally because vnode lock might 737 * not be held. 738 */ 739 if (!RB_EMPTY(&vp->v_rbdirty_tree)) 740 kprintf("dirty bufs left after final pass\n"); 741 #endif 742 break; 743 } 744 lwkt_reltoken(&vp->v_token); 745 746 return(error); 747 } 748 749 static int 750 vfsync_wait_output(struct vnode *vp, 751 int (*waitoutput)(struct vnode *, struct thread *)) 752 { 753 int error; 754 755 error = bio_track_wait(&vp->v_track_write, 0, 0); 756 if (waitoutput) 757 error = waitoutput(vp, curthread); 758 return(error); 759 } 760 761 static int 762 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 763 { 764 return(0); 765 } 766 767 static int 768 vfsync_data_only_cmp(struct buf *bp, void *data) 769 { 770 if (bp->b_loffset < 0) 771 return(-1); 772 return(0); 773 } 774 775 static int 776 vfsync_meta_only_cmp(struct buf *bp, void *data) 777 { 778 if (bp->b_loffset < 0) 779 return(0); 780 return(1); 781 } 782 783 static int 784 vfsync_lazy_range_cmp(struct buf *bp, void *data) 785 { 786 struct vfsync_info *info = data; 787 788 if (bp->b_loffset < info->vp->v_lazyw) 789 return(-1); 790 return(0); 791 } 792 793 static int 794 vfsync_bp(struct buf *bp, void *data) 795 { 796 struct vfsync_info *info = data; 797 struct vnode *vp = info->vp; 798 int error; 799 800 if (info->fastpass) { 801 /* 802 * Ignore buffers that we cannot immediately lock. 803 */ 804 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 805 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst1", 1)) { 806 ++info->skippedbufs; 807 return(0); 808 } 809 } 810 } else if (info->synchronous == 0) { 811 /* 812 * Normal pass, give the buffer a little time to become 813 * available to us. 814 */ 815 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst2", hz / 10)) { 816 ++info->skippedbufs; 817 return(0); 818 } 819 } else { 820 /* 821 * Synchronous pass, give the buffer a lot of time before 822 * giving up. 823 */ 824 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst3", hz * 10)) { 825 ++info->skippedbufs; 826 return(0); 827 } 828 } 829 830 /* 831 * We must revalidate the buffer after locking. 832 */ 833 if ((bp->b_flags & B_DELWRI) == 0 || 834 bp->b_vp != info->vp || 835 info->cmpfunc(bp, data)) { 836 BUF_UNLOCK(bp); 837 return(0); 838 } 839 840 /* 841 * If syncdeps is not set we do not try to write buffers which have 842 * dependancies. 843 */ 844 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 845 BUF_UNLOCK(bp); 846 return(0); 847 } 848 849 /* 850 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 851 * has been written but an additional handshake with the device 852 * is required before we can dispose of the buffer. We have no idea 853 * how to do this so we have to skip these buffers. 854 */ 855 if (bp->b_flags & B_NEEDCOMMIT) { 856 BUF_UNLOCK(bp); 857 return(0); 858 } 859 860 /* 861 * Ask bioops if it is ok to sync. If not the VFS may have 862 * set B_LOCKED so we have to cycle the buffer. 863 */ 864 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 865 bremfree(bp); 866 brelse(bp); 867 return(0); 868 } 869 870 if (info->synchronous) { 871 /* 872 * Synchronous flush. An error may be returned and will 873 * stop the scan. 874 */ 875 bremfree(bp); 876 error = bwrite(bp); 877 } else { 878 /* 879 * Asynchronous flush. We use the error return to support 880 * MNT_LAZY flushes. 881 * 882 * In low-memory situations we revert to synchronous 883 * operation. This should theoretically prevent the I/O 884 * path from exhausting memory in a non-recoverable way. 885 */ 886 vp->v_lazyw = bp->b_loffset; 887 bremfree(bp); 888 if (vm_page_count_min(0)) { 889 /* low memory */ 890 info->lazycount += bp->b_bufsize; 891 bwrite(bp); 892 } else { 893 /* normal */ 894 info->lazycount += cluster_awrite(bp); 895 waitrunningbufspace(); 896 /*vm_wait_nominal();*/ 897 } 898 if (info->lazylimit && info->lazycount >= info->lazylimit) 899 error = 1; 900 else 901 error = 0; 902 } 903 return(-error); 904 } 905 906 /* 907 * Associate a buffer with a vnode. 908 * 909 * MPSAFE 910 */ 911 int 912 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 913 { 914 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 915 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 916 917 /* 918 * Insert onto list for new vnode. 919 */ 920 lwkt_gettoken(&vp->v_token); 921 922 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 923 lwkt_reltoken(&vp->v_token); 924 return (EEXIST); 925 } 926 927 /* 928 * Diagnostics (mainly for HAMMER debugging). Check for 929 * overlapping buffers. 930 */ 931 if (check_buf_overlap) { 932 struct buf *bx; 933 bx = buf_rb_hash_RB_PREV(bp); 934 if (bx) { 935 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 936 kprintf("bgetvp: overlapl %016jx/%d %016jx " 937 "bx %p bp %p\n", 938 (intmax_t)bx->b_loffset, 939 bx->b_bufsize, 940 (intmax_t)bp->b_loffset, 941 bx, bp); 942 if (check_buf_overlap > 1) 943 panic("bgetvp - overlapping buffer"); 944 } 945 } 946 bx = buf_rb_hash_RB_NEXT(bp); 947 if (bx) { 948 if (bp->b_loffset + testsize > bx->b_loffset) { 949 kprintf("bgetvp: overlapr %016jx/%d %016jx " 950 "bp %p bx %p\n", 951 (intmax_t)bp->b_loffset, 952 testsize, 953 (intmax_t)bx->b_loffset, 954 bp, bx); 955 if (check_buf_overlap > 1) 956 panic("bgetvp - overlapping buffer"); 957 } 958 } 959 } 960 bp->b_vp = vp; 961 bp->b_flags |= B_HASHED; 962 bp->b_flags |= B_VNCLEAN; 963 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 964 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 965 /*vhold(vp);*/ 966 lwkt_reltoken(&vp->v_token); 967 return(0); 968 } 969 970 /* 971 * Disassociate a buffer from a vnode. 972 * 973 * MPSAFE 974 */ 975 void 976 brelvp(struct buf *bp) 977 { 978 struct vnode *vp; 979 980 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 981 982 /* 983 * Delete from old vnode list, if on one. 984 */ 985 vp = bp->b_vp; 986 lwkt_gettoken(&vp->v_token); 987 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 988 if (bp->b_flags & B_VNDIRTY) 989 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 990 else 991 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 992 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 993 } 994 if (bp->b_flags & B_HASHED) { 995 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 996 bp->b_flags &= ~B_HASHED; 997 } 998 999 /* 1000 * Only remove from synclist when no dirty buffers are left AND 1001 * the VFS has not flagged the vnode's inode as being dirty. 1002 */ 1003 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 1004 RB_EMPTY(&vp->v_rbdirty_tree)) { 1005 vn_syncer_remove(vp, 0); 1006 } 1007 bp->b_vp = NULL; 1008 1009 lwkt_reltoken(&vp->v_token); 1010 1011 /*vdrop(vp);*/ 1012 } 1013 1014 /* 1015 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 1016 * This routine is called when the state of the B_DELWRI bit is changed. 1017 * 1018 * Must be called with vp->v_token held. 1019 * MPSAFE 1020 */ 1021 void 1022 reassignbuf(struct buf *bp) 1023 { 1024 struct vnode *vp = bp->b_vp; 1025 int delay; 1026 1027 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 1028 ++reassignbufcalls; 1029 1030 /* 1031 * B_PAGING flagged buffers cannot be reassigned because their vp 1032 * is not fully linked in. 1033 */ 1034 if (bp->b_flags & B_PAGING) 1035 panic("cannot reassign paging buffer"); 1036 1037 if (bp->b_flags & B_DELWRI) { 1038 /* 1039 * Move to the dirty list, add the vnode to the worklist 1040 */ 1041 if (bp->b_flags & B_VNCLEAN) { 1042 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1043 bp->b_flags &= ~B_VNCLEAN; 1044 } 1045 if ((bp->b_flags & B_VNDIRTY) == 0) { 1046 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 1047 panic("reassignbuf: dup lblk vp %p bp %p", 1048 vp, bp); 1049 } 1050 bp->b_flags |= B_VNDIRTY; 1051 } 1052 if ((vp->v_flag & VONWORKLST) == 0) { 1053 switch (vp->v_type) { 1054 case VDIR: 1055 delay = dirdelay; 1056 break; 1057 case VCHR: 1058 case VBLK: 1059 if (vp->v_rdev && 1060 vp->v_rdev->si_mountpoint != NULL) { 1061 delay = metadelay; 1062 break; 1063 } 1064 /* fall through */ 1065 default: 1066 delay = filedelay; 1067 } 1068 vn_syncer_add(vp, delay); 1069 } 1070 } else { 1071 /* 1072 * Move to the clean list, remove the vnode from the worklist 1073 * if no dirty blocks remain. 1074 */ 1075 if (bp->b_flags & B_VNDIRTY) { 1076 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1077 bp->b_flags &= ~B_VNDIRTY; 1078 } 1079 if ((bp->b_flags & B_VNCLEAN) == 0) { 1080 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1081 panic("reassignbuf: dup lblk vp %p bp %p", 1082 vp, bp); 1083 } 1084 bp->b_flags |= B_VNCLEAN; 1085 } 1086 1087 /* 1088 * Only remove from synclist when no dirty buffers are left 1089 * AND the VFS has not flagged the vnode's inode as being 1090 * dirty. 1091 */ 1092 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1093 VONWORKLST && 1094 RB_EMPTY(&vp->v_rbdirty_tree)) { 1095 vn_syncer_remove(vp, 0); 1096 } 1097 } 1098 } 1099 1100 /* 1101 * Create a vnode for a block device. Used for mounting the root file 1102 * system. 1103 * 1104 * A vref()'d vnode is returned. 1105 */ 1106 extern struct vop_ops *devfs_vnode_dev_vops_p; 1107 int 1108 bdevvp(cdev_t dev, struct vnode **vpp) 1109 { 1110 struct vnode *vp; 1111 struct vnode *nvp; 1112 int error; 1113 1114 if (dev == NULL) { 1115 *vpp = NULLVP; 1116 return (ENXIO); 1117 } 1118 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1119 &nvp, 0, 0); 1120 if (error) { 1121 *vpp = NULLVP; 1122 return (error); 1123 } 1124 vp = nvp; 1125 vp->v_type = VCHR; 1126 #if 0 1127 vp->v_rdev = dev; 1128 #endif 1129 v_associate_rdev(vp, dev); 1130 vp->v_umajor = dev->si_umajor; 1131 vp->v_uminor = dev->si_uminor; 1132 vx_unlock(vp); 1133 *vpp = vp; 1134 return (0); 1135 } 1136 1137 int 1138 v_associate_rdev(struct vnode *vp, cdev_t dev) 1139 { 1140 if (dev == NULL) 1141 return(ENXIO); 1142 if (dev_is_good(dev) == 0) 1143 return(ENXIO); 1144 KKASSERT(vp->v_rdev == NULL); 1145 vp->v_rdev = reference_dev(dev); 1146 lwkt_gettoken(&spechash_token); 1147 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1148 lwkt_reltoken(&spechash_token); 1149 return(0); 1150 } 1151 1152 void 1153 v_release_rdev(struct vnode *vp) 1154 { 1155 cdev_t dev; 1156 1157 if ((dev = vp->v_rdev) != NULL) { 1158 lwkt_gettoken(&spechash_token); 1159 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1160 vp->v_rdev = NULL; 1161 release_dev(dev); 1162 lwkt_reltoken(&spechash_token); 1163 } 1164 } 1165 1166 /* 1167 * Add a vnode to the alias list hung off the cdev_t. We only associate 1168 * the device number with the vnode. The actual device is not associated 1169 * until the vnode is opened (usually in spec_open()), and will be 1170 * disassociated on last close. 1171 */ 1172 void 1173 addaliasu(struct vnode *nvp, int x, int y) 1174 { 1175 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1176 panic("addaliasu on non-special vnode"); 1177 nvp->v_umajor = x; 1178 nvp->v_uminor = y; 1179 } 1180 1181 /* 1182 * Simple call that a filesystem can make to try to get rid of a 1183 * vnode. It will fail if anyone is referencing the vnode (including 1184 * the caller). 1185 * 1186 * The filesystem can check whether its in-memory inode structure still 1187 * references the vp on return. 1188 * 1189 * May only be called if the vnode is in a known state (i.e. being prevented 1190 * from being deallocated by some other condition such as a vfs inode hold). 1191 */ 1192 void 1193 vclean_unlocked(struct vnode *vp) 1194 { 1195 vx_get(vp); 1196 if (VREFCNT(vp) <= 1) 1197 vgone_vxlocked(vp); 1198 vx_put(vp); 1199 } 1200 1201 /* 1202 * Disassociate a vnode from its underlying filesystem. 1203 * 1204 * The vnode must be VX locked and referenced. In all normal situations 1205 * there are no active references. If vclean_vxlocked() is called while 1206 * there are active references, the vnode is being ripped out and we have 1207 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1208 */ 1209 void 1210 vclean_vxlocked(struct vnode *vp, int flags) 1211 { 1212 int active; 1213 int n; 1214 vm_object_t object; 1215 struct namecache *ncp; 1216 1217 /* 1218 * If the vnode has already been reclaimed we have nothing to do. 1219 */ 1220 if (vp->v_flag & VRECLAIMED) 1221 return; 1222 1223 /* 1224 * Set flag to interlock operation, flag finalization to ensure 1225 * that the vnode winds up on the inactive list, and set v_act to 0. 1226 */ 1227 vsetflags(vp, VRECLAIMED); 1228 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1229 vp->v_act = 0; 1230 1231 if (verbose_reclaims) { 1232 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1233 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1234 } 1235 1236 /* 1237 * Scrap the vfs cache 1238 */ 1239 while (cache_inval_vp(vp, 0) != 0) { 1240 kprintf("Warning: vnode %p clean/cache_resolution " 1241 "race detected\n", vp); 1242 tsleep(vp, 0, "vclninv", 2); 1243 } 1244 1245 /* 1246 * Check to see if the vnode is in use. If so we have to reference it 1247 * before we clean it out so that its count cannot fall to zero and 1248 * generate a race against ourselves to recycle it. 1249 */ 1250 active = (VREFCNT(vp) > 0); 1251 1252 /* 1253 * Clean out any buffers associated with the vnode and destroy its 1254 * object, if it has one. 1255 */ 1256 vinvalbuf(vp, V_SAVE, 0, 0); 1257 1258 /* 1259 * If purging an active vnode (typically during a forced unmount 1260 * or reboot), it must be closed and deactivated before being 1261 * reclaimed. This isn't really all that safe, but what can 1262 * we do? XXX. 1263 * 1264 * Note that neither of these routines unlocks the vnode. 1265 */ 1266 if (active && (flags & DOCLOSE)) { 1267 while ((n = vp->v_opencount) != 0) { 1268 if (vp->v_writecount) 1269 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1270 else 1271 VOP_CLOSE(vp, FNONBLOCK, NULL); 1272 if (vp->v_opencount == n) { 1273 kprintf("Warning: unable to force-close" 1274 " vnode %p\n", vp); 1275 break; 1276 } 1277 } 1278 } 1279 1280 /* 1281 * If the vnode has not been deactivated, deactivated it. Deactivation 1282 * can create new buffers and VM pages so we have to call vinvalbuf() 1283 * again to make sure they all get flushed. 1284 * 1285 * This can occur if a file with a link count of 0 needs to be 1286 * truncated. 1287 * 1288 * If the vnode is already dead don't try to deactivate it. 1289 */ 1290 if ((vp->v_flag & VINACTIVE) == 0) { 1291 vsetflags(vp, VINACTIVE); 1292 if (vp->v_mount) 1293 VOP_INACTIVE(vp); 1294 vinvalbuf(vp, V_SAVE, 0, 0); 1295 } 1296 1297 /* 1298 * If the vnode has an object, destroy it. 1299 */ 1300 while ((object = vp->v_object) != NULL) { 1301 vm_object_hold(object); 1302 if (object == vp->v_object) 1303 break; 1304 vm_object_drop(object); 1305 } 1306 1307 if (object != NULL) { 1308 if (object->ref_count == 0) { 1309 if ((object->flags & OBJ_DEAD) == 0) 1310 vm_object_terminate(object); 1311 vm_object_drop(object); 1312 vclrflags(vp, VOBJBUF); 1313 } else { 1314 vm_pager_deallocate(object); 1315 vclrflags(vp, VOBJBUF); 1316 vm_object_drop(object); 1317 } 1318 } 1319 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1320 1321 /* 1322 * Reclaim the vnode if not already dead. 1323 */ 1324 if (vp->v_mount && VOP_RECLAIM(vp)) 1325 panic("vclean: cannot reclaim"); 1326 1327 /* 1328 * Done with purge, notify sleepers of the grim news. 1329 */ 1330 vp->v_ops = &dead_vnode_vops_p; 1331 vn_gone(vp); 1332 vp->v_tag = VT_NON; 1333 1334 /* 1335 * If we are destroying an active vnode, reactivate it now that 1336 * we have reassociated it with deadfs. This prevents the system 1337 * from crashing on the vnode due to it being unexpectedly marked 1338 * as inactive or reclaimed. 1339 */ 1340 if (active && (flags & DOCLOSE)) { 1341 vclrflags(vp, VINACTIVE | VRECLAIMED); 1342 } 1343 } 1344 1345 /* 1346 * Eliminate all activity associated with the requested vnode 1347 * and with all vnodes aliased to the requested vnode. 1348 * 1349 * The vnode must be referenced but should not be locked. 1350 */ 1351 int 1352 vrevoke(struct vnode *vp, struct ucred *cred) 1353 { 1354 struct vnode *vq; 1355 struct vnode *vqn; 1356 cdev_t dev; 1357 int error; 1358 1359 /* 1360 * If the vnode has a device association, scrap all vnodes associated 1361 * with the device. Don't let the device disappear on us while we 1362 * are scrapping the vnodes. 1363 * 1364 * The passed vp will probably show up in the list, do not VX lock 1365 * it twice! 1366 * 1367 * Releasing the vnode's rdev here can mess up specfs's call to 1368 * device close, so don't do it. The vnode has been disassociated 1369 * and the device will be closed after the last ref on the related 1370 * fp goes away (if not still open by e.g. the kernel). 1371 */ 1372 if (vp->v_type != VCHR) { 1373 error = fdrevoke(vp, DTYPE_VNODE, cred); 1374 return (error); 1375 } 1376 if ((dev = vp->v_rdev) == NULL) { 1377 return(0); 1378 } 1379 reference_dev(dev); 1380 lwkt_gettoken(&spechash_token); 1381 1382 restart: 1383 vqn = SLIST_FIRST(&dev->si_hlist); 1384 if (vqn) 1385 vhold(vqn); 1386 while ((vq = vqn) != NULL) { 1387 if (VREFCNT(vq) > 0) { 1388 vref(vq); 1389 fdrevoke(vq, DTYPE_VNODE, cred); 1390 /*v_release_rdev(vq);*/ 1391 vrele(vq); 1392 if (vq->v_rdev != dev) { 1393 vdrop(vq); 1394 goto restart; 1395 } 1396 } 1397 vqn = SLIST_NEXT(vq, v_cdevnext); 1398 if (vqn) 1399 vhold(vqn); 1400 vdrop(vq); 1401 } 1402 lwkt_reltoken(&spechash_token); 1403 dev_drevoke(dev); 1404 release_dev(dev); 1405 return (0); 1406 } 1407 1408 /* 1409 * This is called when the object underlying a vnode is being destroyed, 1410 * such as in a remove(). Try to recycle the vnode immediately if the 1411 * only active reference is our reference. 1412 * 1413 * Directory vnodes in the namecache with children cannot be immediately 1414 * recycled because numerous VOP_N*() ops require them to be stable. 1415 * 1416 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1417 * function is a NOP if VRECLAIMED is already set. 1418 */ 1419 int 1420 vrecycle(struct vnode *vp) 1421 { 1422 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1423 if (cache_inval_vp_nonblock(vp)) 1424 return(0); 1425 vgone_vxlocked(vp); 1426 return (1); 1427 } 1428 return (0); 1429 } 1430 1431 /* 1432 * Return the maximum I/O size allowed for strategy calls on VP. 1433 * 1434 * If vp is VCHR or VBLK we dive the device, otherwise we use 1435 * the vp's mount info. 1436 * 1437 * The returned value is clamped at MAXPHYS as most callers cannot use 1438 * buffers larger than that size. 1439 */ 1440 int 1441 vmaxiosize(struct vnode *vp) 1442 { 1443 int maxiosize; 1444 1445 if (vp->v_type == VBLK || vp->v_type == VCHR) 1446 maxiosize = vp->v_rdev->si_iosize_max; 1447 else 1448 maxiosize = vp->v_mount->mnt_iosize_max; 1449 1450 if (maxiosize > MAXPHYS) 1451 maxiosize = MAXPHYS; 1452 return (maxiosize); 1453 } 1454 1455 /* 1456 * Eliminate all activity associated with a vnode in preparation for 1457 * destruction. 1458 * 1459 * The vnode must be VX locked and refd and will remain VX locked and refd 1460 * on return. This routine may be called with the vnode in any state, as 1461 * long as it is VX locked. The vnode will be cleaned out and marked 1462 * VRECLAIMED but will not actually be reused until all existing refs and 1463 * holds go away. 1464 * 1465 * NOTE: This routine may be called on a vnode which has not yet been 1466 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1467 * already been reclaimed. 1468 * 1469 * This routine is not responsible for placing us back on the freelist. 1470 * Instead, it happens automatically when the caller releases the VX lock 1471 * (assuming there aren't any other references). 1472 */ 1473 void 1474 vgone_vxlocked(struct vnode *vp) 1475 { 1476 /* 1477 * assert that the VX lock is held. This is an absolute requirement 1478 * now for vgone_vxlocked() to be called. 1479 */ 1480 KKASSERT(lockinuse(&vp->v_lock)); 1481 1482 /* 1483 * Clean out the filesystem specific data and set the VRECLAIMED 1484 * bit. Also deactivate the vnode if necessary. 1485 * 1486 * The vnode should have automatically been removed from the syncer 1487 * list as syncer/dirty flags cleared during the cleaning. 1488 */ 1489 vclean_vxlocked(vp, DOCLOSE); 1490 1491 /* 1492 * Normally panic if the vnode is still dirty, unless we are doing 1493 * a forced unmount (tmpfs typically). 1494 */ 1495 if (vp->v_flag & VONWORKLST) { 1496 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) { 1497 /* force removal */ 1498 vn_syncer_remove(vp, 1); 1499 } else { 1500 panic("vp %p still dirty in vgone after flush", vp); 1501 } 1502 } 1503 1504 /* 1505 * Delete from old mount point vnode list, if on one. 1506 */ 1507 if (vp->v_mount != NULL) { 1508 KKASSERT(vp->v_data == NULL); 1509 insmntque(vp, NULL); 1510 } 1511 1512 /* 1513 * If special device, remove it from special device alias list 1514 * if it is on one. This should normally only occur if a vnode is 1515 * being revoked as the device should otherwise have been released 1516 * naturally. 1517 */ 1518 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1519 v_release_rdev(vp); 1520 } 1521 1522 /* 1523 * Set us to VBAD 1524 */ 1525 vp->v_type = VBAD; 1526 } 1527 1528 /* 1529 * Lookup a vnode by device number. 1530 * 1531 * Returns non-zero and *vpp set to a vref'd vnode on success. 1532 * Returns zero on failure. 1533 */ 1534 int 1535 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1536 { 1537 struct vnode *vp; 1538 1539 lwkt_gettoken(&spechash_token); 1540 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1541 if (type == vp->v_type) { 1542 *vpp = vp; 1543 vref(vp); 1544 lwkt_reltoken(&spechash_token); 1545 return (1); 1546 } 1547 } 1548 lwkt_reltoken(&spechash_token); 1549 return (0); 1550 } 1551 1552 /* 1553 * Calculate the total number of references to a special device. This 1554 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1555 * an overloaded field. Since udev2dev can now return NULL, we have 1556 * to check for a NULL v_rdev. 1557 */ 1558 int 1559 count_dev(cdev_t dev) 1560 { 1561 struct vnode *vp; 1562 int count = 0; 1563 1564 if (SLIST_FIRST(&dev->si_hlist)) { 1565 lwkt_gettoken(&spechash_token); 1566 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1567 count += vp->v_opencount; 1568 } 1569 lwkt_reltoken(&spechash_token); 1570 } 1571 return(count); 1572 } 1573 1574 int 1575 vcount(struct vnode *vp) 1576 { 1577 if (vp->v_rdev == NULL) 1578 return(0); 1579 return(count_dev(vp->v_rdev)); 1580 } 1581 1582 /* 1583 * Initialize VMIO for a vnode. This routine MUST be called before a 1584 * VFS can issue buffer cache ops on a vnode. It is typically called 1585 * when a vnode is initialized from its inode. 1586 */ 1587 int 1588 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1589 { 1590 vm_object_t object; 1591 int error = 0; 1592 1593 object = vp->v_object; 1594 if (object) { 1595 vm_object_hold(object); 1596 KKASSERT(vp->v_object == object); 1597 } 1598 1599 if (object == NULL) { 1600 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1601 1602 /* 1603 * Dereference the reference we just created. This assumes 1604 * that the object is associated with the vp. Allow it to 1605 * have zero refs. It cannot be destroyed as long as it 1606 * is associated with the vnode. 1607 */ 1608 vm_object_hold(object); 1609 atomic_add_int(&object->ref_count, -1); 1610 vrele(vp); 1611 } else { 1612 KKASSERT((object->flags & OBJ_DEAD) == 0); 1613 } 1614 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1615 vsetflags(vp, VOBJBUF); 1616 vm_object_drop(object); 1617 1618 return (error); 1619 } 1620 1621 1622 /* 1623 * Print out a description of a vnode. 1624 */ 1625 static char *typename[] = 1626 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1627 1628 void 1629 vprint(char *label, struct vnode *vp) 1630 { 1631 char buf[96]; 1632 1633 if (label != NULL) 1634 kprintf("%s: %p: ", label, (void *)vp); 1635 else 1636 kprintf("%p: ", (void *)vp); 1637 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1638 typename[vp->v_type], 1639 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1640 buf[0] = '\0'; 1641 if (vp->v_flag & VROOT) 1642 strcat(buf, "|VROOT"); 1643 if (vp->v_flag & VPFSROOT) 1644 strcat(buf, "|VPFSROOT"); 1645 if (vp->v_flag & VTEXT) 1646 strcat(buf, "|VTEXT"); 1647 if (vp->v_flag & VSYSTEM) 1648 strcat(buf, "|VSYSTEM"); 1649 if (vp->v_flag & VOBJBUF) 1650 strcat(buf, "|VOBJBUF"); 1651 if (buf[0] != '\0') 1652 kprintf(" flags (%s)", &buf[1]); 1653 if (vp->v_data == NULL) { 1654 kprintf("\n"); 1655 } else { 1656 kprintf("\n\t"); 1657 VOP_PRINT(vp); 1658 } 1659 } 1660 1661 /* 1662 * Do the usual access checking. 1663 * file_mode, uid and gid are from the vnode in question, 1664 * while acc_mode and cred are from the VOP_ACCESS parameter list 1665 */ 1666 int 1667 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1668 mode_t acc_mode, struct ucred *cred) 1669 { 1670 mode_t mask; 1671 int ismember; 1672 1673 /* 1674 * Super-user always gets read/write access, but execute access depends 1675 * on at least one execute bit being set. 1676 */ 1677 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1678 if ((acc_mode & VEXEC) && type != VDIR && 1679 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1680 return (EACCES); 1681 return (0); 1682 } 1683 1684 mask = 0; 1685 1686 /* Otherwise, check the owner. */ 1687 if (cred->cr_uid == uid) { 1688 if (acc_mode & VEXEC) 1689 mask |= S_IXUSR; 1690 if (acc_mode & VREAD) 1691 mask |= S_IRUSR; 1692 if (acc_mode & VWRITE) 1693 mask |= S_IWUSR; 1694 return ((file_mode & mask) == mask ? 0 : EACCES); 1695 } 1696 1697 /* Otherwise, check the groups. */ 1698 ismember = groupmember(gid, cred); 1699 if (cred->cr_svgid == gid || ismember) { 1700 if (acc_mode & VEXEC) 1701 mask |= S_IXGRP; 1702 if (acc_mode & VREAD) 1703 mask |= S_IRGRP; 1704 if (acc_mode & VWRITE) 1705 mask |= S_IWGRP; 1706 return ((file_mode & mask) == mask ? 0 : EACCES); 1707 } 1708 1709 /* Otherwise, check everyone else. */ 1710 if (acc_mode & VEXEC) 1711 mask |= S_IXOTH; 1712 if (acc_mode & VREAD) 1713 mask |= S_IROTH; 1714 if (acc_mode & VWRITE) 1715 mask |= S_IWOTH; 1716 return ((file_mode & mask) == mask ? 0 : EACCES); 1717 } 1718 1719 #ifdef DDB 1720 #include <ddb/ddb.h> 1721 1722 static int db_show_locked_vnodes(struct mount *mp, void *data); 1723 1724 /* 1725 * List all of the locked vnodes in the system. 1726 * Called when debugging the kernel. 1727 */ 1728 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1729 { 1730 kprintf("Locked vnodes\n"); 1731 mountlist_scan(db_show_locked_vnodes, NULL, 1732 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1733 } 1734 1735 static int 1736 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1737 { 1738 struct vnode *vp; 1739 1740 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1741 if (vn_islocked(vp)) 1742 vprint(NULL, vp); 1743 } 1744 return(0); 1745 } 1746 #endif 1747 1748 /* 1749 * Top level filesystem related information gathering. 1750 */ 1751 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1752 1753 static int 1754 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1755 { 1756 int *name = (int *)arg1 - 1; /* XXX */ 1757 u_int namelen = arg2 + 1; /* XXX */ 1758 struct vfsconf *vfsp; 1759 int maxtypenum; 1760 1761 #if 1 || defined(COMPAT_PRELITE2) 1762 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1763 if (namelen == 1) 1764 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1765 #endif 1766 1767 #ifdef notyet 1768 /* all sysctl names at this level are at least name and field */ 1769 if (namelen < 2) 1770 return (ENOTDIR); /* overloaded */ 1771 if (name[0] != VFS_GENERIC) { 1772 vfsp = vfsconf_find_by_typenum(name[0]); 1773 if (vfsp == NULL) 1774 return (EOPNOTSUPP); 1775 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1776 oldp, oldlenp, newp, newlen, p)); 1777 } 1778 #endif 1779 switch (name[1]) { 1780 case VFS_MAXTYPENUM: 1781 if (namelen != 2) 1782 return (ENOTDIR); 1783 maxtypenum = vfsconf_get_maxtypenum(); 1784 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1785 case VFS_CONF: 1786 if (namelen != 3) 1787 return (ENOTDIR); /* overloaded */ 1788 vfsp = vfsconf_find_by_typenum(name[2]); 1789 if (vfsp == NULL) 1790 return (EOPNOTSUPP); 1791 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1792 } 1793 return (EOPNOTSUPP); 1794 } 1795 1796 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1797 "Generic filesystem"); 1798 1799 #if 1 || defined(COMPAT_PRELITE2) 1800 1801 static int 1802 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1803 { 1804 int error; 1805 struct ovfsconf ovfs; 1806 struct sysctl_req *req = (struct sysctl_req*) data; 1807 1808 bzero(&ovfs, sizeof(ovfs)); 1809 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1810 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1811 ovfs.vfc_index = vfsp->vfc_typenum; 1812 ovfs.vfc_refcount = vfsp->vfc_refcount; 1813 ovfs.vfc_flags = vfsp->vfc_flags; 1814 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1815 if (error) 1816 return error; /* abort iteration with error code */ 1817 else 1818 return 0; /* continue iterating with next element */ 1819 } 1820 1821 static int 1822 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1823 { 1824 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1825 } 1826 1827 #endif /* 1 || COMPAT_PRELITE2 */ 1828 1829 /* 1830 * Check to see if a filesystem is mounted on a block device. 1831 */ 1832 int 1833 vfs_mountedon(struct vnode *vp) 1834 { 1835 cdev_t dev; 1836 1837 if ((dev = vp->v_rdev) == NULL) { 1838 /* if (vp->v_type != VBLK) 1839 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1840 } 1841 if (dev != NULL && dev->si_mountpoint) 1842 return (EBUSY); 1843 return (0); 1844 } 1845 1846 /* 1847 * Unmount all filesystems. The list is traversed in reverse order 1848 * of mounting to avoid dependencies. 1849 * 1850 * We want the umountall to be able to break out of its loop if a 1851 * failure occurs, after scanning all possible mounts, so the callback 1852 * returns 0 on error. 1853 * 1854 * NOTE: Do not call mountlist_remove(mp) on error any more, this will 1855 * confuse mountlist_scan()'s unbusy check. 1856 */ 1857 static int vfs_umountall_callback(struct mount *mp, void *data); 1858 1859 void 1860 vfs_unmountall(void) 1861 { 1862 int count; 1863 1864 do { 1865 count = mountlist_scan(vfs_umountall_callback, 1866 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1867 } while (count); 1868 } 1869 1870 static 1871 int 1872 vfs_umountall_callback(struct mount *mp, void *data) 1873 { 1874 int error; 1875 1876 error = dounmount(mp, MNT_FORCE); 1877 if (error) { 1878 kprintf("unmount of filesystem mounted from %s failed (", 1879 mp->mnt_stat.f_mntfromname); 1880 if (error == EBUSY) 1881 kprintf("BUSY)\n"); 1882 else 1883 kprintf("%d)\n", error); 1884 return 0; 1885 } else { 1886 return 1; 1887 } 1888 } 1889 1890 /* 1891 * Checks the mount flags for parameter mp and put the names comma-separated 1892 * into a string buffer buf with a size limit specified by len. 1893 * 1894 * It returns the number of bytes written into buf, and (*errorp) will be 1895 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1896 * not large enough). The buffer will be 0-terminated if len was not 0. 1897 */ 1898 size_t 1899 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1900 char *buf, size_t len, int *errorp) 1901 { 1902 static const struct mountctl_opt optnames[] = { 1903 { MNT_RDONLY, "read-only" }, 1904 { MNT_SYNCHRONOUS, "synchronous" }, 1905 { MNT_NOEXEC, "noexec" }, 1906 { MNT_NOSUID, "nosuid" }, 1907 { MNT_NODEV, "nodev" }, 1908 { MNT_AUTOMOUNTED, "automounted" }, 1909 { MNT_ASYNC, "asynchronous" }, 1910 { MNT_SUIDDIR, "suiddir" }, 1911 { MNT_SOFTDEP, "soft-updates" }, 1912 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1913 { MNT_TRIM, "trim" }, 1914 { MNT_NOATIME, "noatime" }, 1915 { MNT_NOCLUSTERR, "noclusterr" }, 1916 { MNT_NOCLUSTERW, "noclusterw" }, 1917 { MNT_EXRDONLY, "NFS read-only" }, 1918 { MNT_EXPORTED, "NFS exported" }, 1919 /* Remaining NFS flags could come here */ 1920 { MNT_LOCAL, "local" }, 1921 { MNT_QUOTA, "with-quotas" }, 1922 /* { MNT_ROOTFS, "rootfs" }, */ 1923 /* { MNT_IGNORE, "ignore" }, */ 1924 { 0, NULL} 1925 }; 1926 int bwritten; 1927 int bleft; 1928 int optlen; 1929 int actsize; 1930 1931 *errorp = 0; 1932 bwritten = 0; 1933 bleft = len - 1; /* leave room for trailing \0 */ 1934 1935 /* 1936 * Checks the size of the string. If it contains 1937 * any data, then we will append the new flags to 1938 * it. 1939 */ 1940 actsize = strlen(buf); 1941 if (actsize > 0) 1942 buf += actsize; 1943 1944 /* Default flags if no flags passed */ 1945 if (optp == NULL) 1946 optp = optnames; 1947 1948 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1949 *errorp = EINVAL; 1950 return(0); 1951 } 1952 1953 for (; flags && optp->o_opt; ++optp) { 1954 if ((flags & optp->o_opt) == 0) 1955 continue; 1956 optlen = strlen(optp->o_name); 1957 if (bwritten || actsize > 0) { 1958 if (bleft < 2) { 1959 *errorp = ENOSPC; 1960 break; 1961 } 1962 buf[bwritten++] = ','; 1963 buf[bwritten++] = ' '; 1964 bleft -= 2; 1965 } 1966 if (bleft < optlen) { 1967 *errorp = ENOSPC; 1968 break; 1969 } 1970 bcopy(optp->o_name, buf + bwritten, optlen); 1971 bwritten += optlen; 1972 bleft -= optlen; 1973 flags &= ~optp->o_opt; 1974 } 1975 1976 /* 1977 * Space already reserved for trailing \0 1978 */ 1979 buf[bwritten] = 0; 1980 return (bwritten); 1981 } 1982 1983 /* 1984 * Build hash lists of net addresses and hang them off the mount point. 1985 * Called by ufs_mount() to set up the lists of export addresses. 1986 */ 1987 static int 1988 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1989 const struct export_args *argp) 1990 { 1991 struct netcred *np; 1992 struct radix_node_head *rnh; 1993 int i; 1994 struct radix_node *rn; 1995 struct sockaddr *saddr, *smask = NULL; 1996 int error; 1997 1998 if (argp->ex_addrlen == 0) { 1999 if (mp->mnt_flag & MNT_DEFEXPORTED) 2000 return (EPERM); 2001 np = &nep->ne_defexported; 2002 np->netc_exflags = argp->ex_flags; 2003 np->netc_anon = argp->ex_anon; 2004 np->netc_anon.cr_ref = 1; 2005 mp->mnt_flag |= MNT_DEFEXPORTED; 2006 return (0); 2007 } 2008 2009 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 2010 return (EINVAL); 2011 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 2012 return (EINVAL); 2013 2014 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2015 np = (struct netcred *)kmalloc(i, M_NETCRED, M_WAITOK | M_ZERO); 2016 saddr = (struct sockaddr *) (np + 1); 2017 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2018 goto out; 2019 if (saddr->sa_len > argp->ex_addrlen) 2020 saddr->sa_len = argp->ex_addrlen; 2021 if (argp->ex_masklen) { 2022 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2023 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2024 if (error) 2025 goto out; 2026 if (smask->sa_len > argp->ex_masklen) 2027 smask->sa_len = argp->ex_masklen; 2028 } 2029 NE_LOCK(nep); 2030 if (nep->ne_maskhead == NULL) { 2031 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) { 2032 error = ENOBUFS; 2033 goto out; 2034 } 2035 } 2036 if ((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) { 2037 error = ENOBUFS; 2038 goto out; 2039 } 2040 rn = (*rnh->rnh_addaddr)((char *)saddr, (char *)smask, rnh, 2041 np->netc_rnodes); 2042 NE_UNLOCK(nep); 2043 if (rn == NULL || np != (struct netcred *)rn) { /* already exists */ 2044 error = EPERM; 2045 goto out; 2046 } 2047 np->netc_exflags = argp->ex_flags; 2048 np->netc_anon = argp->ex_anon; 2049 np->netc_anon.cr_ref = 1; 2050 return (0); 2051 2052 out: 2053 kfree(np, M_NETCRED); 2054 return (error); 2055 } 2056 2057 /* 2058 * Free netcred structures installed in the netexport 2059 */ 2060 static int 2061 vfs_free_netcred(struct radix_node *rn, void *w) 2062 { 2063 struct radix_node_head *rnh = (struct radix_node_head *)w; 2064 2065 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2066 kfree(rn, M_NETCRED); 2067 2068 return (0); 2069 } 2070 2071 /* 2072 * callback to free an element of the mask table installed in the 2073 * netexport. These may be created indirectly and are not netcred 2074 * structures. 2075 */ 2076 static int 2077 vfs_free_netcred_mask(struct radix_node *rn, void *w) 2078 { 2079 struct radix_node_head *rnh = (struct radix_node_head *)w; 2080 2081 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2082 kfree(rn, M_RTABLE); 2083 2084 return (0); 2085 } 2086 2087 static struct radix_node_head * 2088 vfs_create_addrlist_af(int af, struct netexport *nep) 2089 { 2090 struct radix_node_head *rnh = NULL; 2091 #if defined(INET) || defined(INET6) 2092 struct radix_node_head *maskhead = nep->ne_maskhead; 2093 int off; 2094 #endif 2095 2096 NE_ASSERT_LOCKED(nep); 2097 KKASSERT(maskhead != NULL); 2098 switch (af) { 2099 #ifdef INET 2100 case AF_INET: 2101 if ((rnh = nep->ne_inethead) == NULL) { 2102 off = offsetof(struct sockaddr_in, sin_addr) << 3; 2103 if (!rn_inithead((void **)&rnh, maskhead, off)) 2104 return (NULL); 2105 nep->ne_inethead = rnh; 2106 } 2107 break; 2108 #endif 2109 #ifdef INET6 2110 case AF_INET6: 2111 if ((rnh = nep->ne_inet6head) == NULL) { 2112 off = offsetof(struct sockaddr_in6, sin6_addr) << 3; 2113 if (!rn_inithead((void **)&rnh, maskhead, off)) 2114 return (NULL); 2115 nep->ne_inet6head = rnh; 2116 } 2117 break; 2118 #endif 2119 } 2120 return (rnh); 2121 } 2122 2123 /* 2124 * helper function for freeing netcred elements 2125 */ 2126 static void 2127 vfs_free_addrlist_af(struct radix_node_head **prnh) 2128 { 2129 struct radix_node_head *rnh = *prnh; 2130 2131 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh); 2132 kfree(rnh, M_RTABLE); 2133 *prnh = NULL; 2134 } 2135 2136 /* 2137 * helper function for freeing mask elements 2138 */ 2139 static void 2140 vfs_free_addrlist_masks(struct radix_node_head **prnh) 2141 { 2142 struct radix_node_head *rnh = *prnh; 2143 2144 (*rnh->rnh_walktree) (rnh, vfs_free_netcred_mask, rnh); 2145 kfree(rnh, M_RTABLE); 2146 *prnh = NULL; 2147 } 2148 2149 /* 2150 * Free the net address hash lists that are hanging off the mount points. 2151 */ 2152 static void 2153 vfs_free_addrlist(struct netexport *nep) 2154 { 2155 NE_LOCK(nep); 2156 if (nep->ne_inethead != NULL) 2157 vfs_free_addrlist_af(&nep->ne_inethead); 2158 if (nep->ne_inet6head != NULL) 2159 vfs_free_addrlist_af(&nep->ne_inet6head); 2160 if (nep->ne_maskhead) 2161 vfs_free_addrlist_masks(&nep->ne_maskhead); 2162 NE_UNLOCK(nep); 2163 } 2164 2165 int 2166 vfs_export(struct mount *mp, struct netexport *nep, 2167 const struct export_args *argp) 2168 { 2169 int error; 2170 2171 if (argp->ex_flags & MNT_DELEXPORT) { 2172 if (mp->mnt_flag & MNT_EXPUBLIC) { 2173 vfs_setpublicfs(NULL, NULL, NULL); 2174 mp->mnt_flag &= ~MNT_EXPUBLIC; 2175 } 2176 vfs_free_addrlist(nep); 2177 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2178 } 2179 if (argp->ex_flags & MNT_EXPORTED) { 2180 if (argp->ex_flags & MNT_EXPUBLIC) { 2181 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2182 return (error); 2183 mp->mnt_flag |= MNT_EXPUBLIC; 2184 } 2185 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2186 return (error); 2187 mp->mnt_flag |= MNT_EXPORTED; 2188 } 2189 return (0); 2190 } 2191 2192 2193 /* 2194 * Set the publicly exported filesystem (WebNFS). Currently, only 2195 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2196 */ 2197 int 2198 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2199 const struct export_args *argp) 2200 { 2201 int error; 2202 struct vnode *rvp; 2203 char *cp; 2204 2205 /* 2206 * mp == NULL -> invalidate the current info, the FS is 2207 * no longer exported. May be called from either vfs_export 2208 * or unmount, so check if it hasn't already been done. 2209 */ 2210 if (mp == NULL) { 2211 if (nfs_pub.np_valid) { 2212 nfs_pub.np_valid = 0; 2213 if (nfs_pub.np_index != NULL) { 2214 kfree(nfs_pub.np_index, M_TEMP); 2215 nfs_pub.np_index = NULL; 2216 } 2217 } 2218 return (0); 2219 } 2220 2221 /* 2222 * Only one allowed at a time. 2223 */ 2224 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2225 return (EBUSY); 2226 2227 /* 2228 * Get real filehandle for root of exported FS. 2229 */ 2230 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2231 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2232 2233 if ((error = VFS_ROOT(mp, &rvp))) 2234 return (error); 2235 2236 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2237 return (error); 2238 2239 vput(rvp); 2240 2241 /* 2242 * If an indexfile was specified, pull it in. 2243 */ 2244 if (argp->ex_indexfile != NULL) { 2245 int namelen; 2246 2247 error = vn_get_namelen(rvp, &namelen); 2248 if (error) 2249 return (error); 2250 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2251 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2252 namelen, NULL); 2253 if (!error) { 2254 /* 2255 * Check for illegal filenames. 2256 */ 2257 for (cp = nfs_pub.np_index; *cp; cp++) { 2258 if (*cp == '/') { 2259 error = EINVAL; 2260 break; 2261 } 2262 } 2263 } 2264 if (error) { 2265 kfree(nfs_pub.np_index, M_TEMP); 2266 return (error); 2267 } 2268 } 2269 2270 nfs_pub.np_mount = mp; 2271 nfs_pub.np_valid = 1; 2272 return (0); 2273 } 2274 2275 struct netcred * 2276 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2277 struct sockaddr *nam) 2278 { 2279 struct netcred *np; 2280 struct radix_node_head *rnh; 2281 struct sockaddr *saddr; 2282 2283 np = NULL; 2284 if (mp->mnt_flag & MNT_EXPORTED) { 2285 /* 2286 * Lookup in the export list first. 2287 */ 2288 NE_LOCK(nep); 2289 if (nam != NULL) { 2290 saddr = nam; 2291 switch (saddr->sa_family) { 2292 #ifdef INET 2293 case AF_INET: 2294 rnh = nep->ne_inethead; 2295 break; 2296 #endif 2297 #ifdef INET6 2298 case AF_INET6: 2299 rnh = nep->ne_inet6head; 2300 break; 2301 #endif 2302 default: 2303 rnh = NULL; 2304 } 2305 if (rnh != NULL) { 2306 np = (struct netcred *) 2307 (*rnh->rnh_matchaddr)((char *)saddr, 2308 rnh); 2309 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2310 np = NULL; 2311 } 2312 } 2313 NE_UNLOCK(nep); 2314 /* 2315 * If no address match, use the default if it exists. 2316 */ 2317 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2318 np = &nep->ne_defexported; 2319 } 2320 return (np); 2321 } 2322 2323 /* 2324 * perform msync on all vnodes under a mount point. The mount point must 2325 * be locked. This code is also responsible for lazy-freeing unreferenced 2326 * vnodes whos VM objects no longer contain pages. 2327 * 2328 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2329 * 2330 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2331 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2332 * way up in this high level function. 2333 */ 2334 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2335 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2336 2337 void 2338 vfs_msync(struct mount *mp, int flags) 2339 { 2340 int vmsc_flags; 2341 2342 /* 2343 * tmpfs sets this flag to prevent msync(), sync, and the 2344 * filesystem periodic syncer from trying to flush VM pages 2345 * to swap. Only pure memory pressure flushes tmpfs VM pages 2346 * to swap. 2347 */ 2348 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2349 return; 2350 2351 /* 2352 * Ok, scan the vnodes for work. If the filesystem is using the 2353 * syncer thread feature we can use vsyncscan() instead of 2354 * vmntvnodescan(), which is much faster. 2355 */ 2356 vmsc_flags = VMSC_GETVP; 2357 if (flags != MNT_WAIT) 2358 vmsc_flags |= VMSC_NOWAIT; 2359 2360 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2361 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2362 (void *)(intptr_t)flags); 2363 } else { 2364 vmntvnodescan(mp, vmsc_flags, 2365 vfs_msync_scan1, vfs_msync_scan2, 2366 (void *)(intptr_t)flags); 2367 } 2368 } 2369 2370 /* 2371 * scan1 is a fast pre-check. There could be hundreds of thousands of 2372 * vnodes, we cannot afford to do anything heavy weight until we have a 2373 * fairly good indication that there is work to do. 2374 */ 2375 static 2376 int 2377 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2378 { 2379 int flags = (int)(intptr_t)data; 2380 2381 if ((vp->v_flag & VRECLAIMED) == 0) { 2382 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2383 vp->v_object) { 2384 return(0); /* call scan2 */ 2385 } 2386 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2387 (vp->v_flag & VOBJDIRTY) && 2388 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2389 return(0); /* call scan2 */ 2390 } 2391 } 2392 2393 /* 2394 * do not call scan2, continue the loop 2395 */ 2396 return(-1); 2397 } 2398 2399 /* 2400 * This callback is handed a locked vnode. 2401 */ 2402 static 2403 int 2404 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2405 { 2406 vm_object_t obj; 2407 int flags = (int)(intptr_t)data; 2408 2409 if (vp->v_flag & VRECLAIMED) 2410 return(0); 2411 2412 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2413 if ((obj = vp->v_object) != NULL) { 2414 vm_object_page_clean(obj, 0, 0, 2415 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2416 } 2417 } 2418 return(0); 2419 } 2420 2421 /* 2422 * Wake up anyone interested in vp because it is being revoked. 2423 */ 2424 void 2425 vn_gone(struct vnode *vp) 2426 { 2427 lwkt_gettoken(&vp->v_token); 2428 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2429 lwkt_reltoken(&vp->v_token); 2430 } 2431 2432 /* 2433 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2434 * (or v_rdev might be NULL). 2435 */ 2436 cdev_t 2437 vn_todev(struct vnode *vp) 2438 { 2439 if (vp->v_type != VBLK && vp->v_type != VCHR) 2440 return (NULL); 2441 KKASSERT(vp->v_rdev != NULL); 2442 return (vp->v_rdev); 2443 } 2444 2445 /* 2446 * Check if vnode represents a disk device. The vnode does not need to be 2447 * opened. 2448 * 2449 * MPALMOSTSAFE 2450 */ 2451 int 2452 vn_isdisk(struct vnode *vp, int *errp) 2453 { 2454 cdev_t dev; 2455 2456 if (vp->v_type != VCHR) { 2457 if (errp != NULL) 2458 *errp = ENOTBLK; 2459 return (0); 2460 } 2461 2462 dev = vp->v_rdev; 2463 2464 if (dev == NULL) { 2465 if (errp != NULL) 2466 *errp = ENXIO; 2467 return (0); 2468 } 2469 if (dev_is_good(dev) == 0) { 2470 if (errp != NULL) 2471 *errp = ENXIO; 2472 return (0); 2473 } 2474 if ((dev_dflags(dev) & D_DISK) == 0) { 2475 if (errp != NULL) 2476 *errp = ENOTBLK; 2477 return (0); 2478 } 2479 if (errp != NULL) 2480 *errp = 0; 2481 return (1); 2482 } 2483 2484 int 2485 vn_get_namelen(struct vnode *vp, int *namelen) 2486 { 2487 int error; 2488 register_t retval[2]; 2489 2490 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2491 if (error) 2492 return (error); 2493 *namelen = (int)retval[0]; 2494 return (0); 2495 } 2496 2497 int 2498 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2499 uint16_t d_namlen, const char *d_name) 2500 { 2501 struct dirent *dp; 2502 size_t len; 2503 2504 len = _DIRENT_RECLEN(d_namlen); 2505 if (len > uio->uio_resid) 2506 return(1); 2507 2508 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2509 2510 dp->d_ino = d_ino; 2511 dp->d_namlen = d_namlen; 2512 dp->d_type = d_type; 2513 bcopy(d_name, dp->d_name, d_namlen); 2514 2515 *error = uiomove((caddr_t)dp, len, uio); 2516 2517 kfree(dp, M_TEMP); 2518 2519 return(0); 2520 } 2521 2522 void 2523 vn_mark_atime(struct vnode *vp, struct thread *td) 2524 { 2525 struct proc *p = td->td_proc; 2526 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2527 2528 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2529 VOP_MARKATIME(vp, cred); 2530 } 2531 } 2532 2533 /* 2534 * Calculate the number of entries in an inode-related chained hash table. 2535 * With today's memory sizes, maxvnodes can wind up being a very large 2536 * number. There is no reason to waste memory, so tolerate some stacking. 2537 */ 2538 int 2539 vfs_inodehashsize(void) 2540 { 2541 int hsize; 2542 2543 hsize = 32; 2544 while (hsize < maxvnodes) 2545 hsize <<= 1; 2546 while (hsize > maxvnodes * 2) 2547 hsize >>= 1; /* nominal 2x stacking */ 2548 2549 if (maxvnodes > 1024 * 1024) 2550 hsize >>= 1; /* nominal 8x stacking */ 2551 2552 if (maxvnodes > 128 * 1024) 2553 hsize >>= 1; /* nominal 4x stacking */ 2554 2555 if (hsize < 16) 2556 hsize = 16; 2557 2558 return hsize; 2559 } 2560