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 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/buf.h> 46 #include <sys/conf.h> 47 #include <sys/dirent.h> 48 #include <sys/domain.h> 49 #include <sys/eventhandler.h> 50 #include <sys/fcntl.h> 51 #include <sys/file.h> 52 #include <sys/kernel.h> 53 #include <sys/kthread.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/mount.h> 57 #include <sys/priv.h> 58 #include <sys/proc.h> 59 #include <sys/reboot.h> 60 #include <sys/socket.h> 61 #include <sys/stat.h> 62 #include <sys/sysctl.h> 63 #include <sys/syslog.h> 64 #include <sys/unistd.h> 65 #include <sys/vmmeter.h> 66 #include <sys/vnode.h> 67 68 #include <machine/limits.h> 69 70 #include <vm/vm.h> 71 #include <vm/vm_object.h> 72 #include <vm/vm_extern.h> 73 #include <vm/vm_kern.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_map.h> 76 #include <vm/vm_page.h> 77 #include <vm/vm_pager.h> 78 #include <vm/vnode_pager.h> 79 #include <vm/vm_zone.h> 80 81 #include <sys/buf2.h> 82 #include <sys/thread2.h> 83 #include <sys/sysref2.h> 84 #include <sys/mplock2.h> 85 86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 87 88 int numvnodes; 89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 90 "Number of vnodes allocated"); 91 int verbose_reclaims; 92 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0, 93 "Output filename of reclaimed vnode(s)"); 94 95 enum vtype iftovt_tab[16] = { 96 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 97 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 98 }; 99 int vttoif_tab[9] = { 100 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 101 S_IFSOCK, S_IFIFO, S_IFMT, 102 }; 103 104 static int reassignbufcalls; 105 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 106 0, "Number of times buffers have been reassigned to the proper list"); 107 108 static int check_buf_overlap = 2; /* invasive check */ 109 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap, 110 0, "Enable overlapping buffer checks"); 111 112 int nfs_mount_type = -1; 113 static struct lwkt_token spechash_token; 114 struct nfs_public nfs_pub; /* publicly exported FS */ 115 116 int desiredvnodes; 117 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 118 &desiredvnodes, 0, "Maximum number of vnodes"); 119 120 static void vfs_free_addrlist (struct netexport *nep); 121 static int vfs_free_netcred (struct radix_node *rn, void *w); 122 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 123 const struct export_args *argp); 124 125 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 126 127 /* 128 * Red black tree functions 129 */ 130 static int rb_buf_compare(struct buf *b1, struct buf *b2); 131 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 132 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 133 134 static int 135 rb_buf_compare(struct buf *b1, struct buf *b2) 136 { 137 if (b1->b_loffset < b2->b_loffset) 138 return(-1); 139 if (b1->b_loffset > b2->b_loffset) 140 return(1); 141 return(0); 142 } 143 144 /* 145 * Initialize the vnode management data structures. 146 * 147 * Called from vfsinit() 148 */ 149 void 150 vfs_subr_init(void) 151 { 152 int factor1; 153 int factor2; 154 155 /* 156 * Desiredvnodes is kern.maxvnodes. We want to scale it 157 * according to available system memory but we may also have 158 * to limit it based on available KVM, which is capped on 32 bit 159 * systems, to ~80K vnodes or so. 160 * 161 * WARNING! For machines with 64-256M of ram we have to be sure 162 * that the default limit scales down well due to HAMMER 163 * taking up significantly more memory per-vnode vs UFS. 164 * We want around ~5800 on a 128M machine. 165 */ 166 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode)); 167 factor2 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode)); 168 desiredvnodes = 169 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 170 KvaSize / factor2); 171 desiredvnodes = imax(desiredvnodes, maxproc * 8); 172 173 lwkt_token_init(&spechash_token, "spechash"); 174 } 175 176 /* 177 * Knob to control the precision of file timestamps: 178 * 179 * 0 = seconds only; nanoseconds zeroed. 180 * 1 = seconds and nanoseconds, accurate within 1/HZ. 181 * 2 = seconds and nanoseconds, truncated to microseconds. 182 * >=3 = seconds and nanoseconds, maximum precision. 183 */ 184 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 185 186 static int timestamp_precision = TSP_SEC; 187 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 188 ×tamp_precision, 0, "Precision of file timestamps"); 189 190 /* 191 * Get a current timestamp. 192 * 193 * MPSAFE 194 */ 195 void 196 vfs_timestamp(struct timespec *tsp) 197 { 198 struct timeval tv; 199 200 switch (timestamp_precision) { 201 case TSP_SEC: 202 tsp->tv_sec = time_second; 203 tsp->tv_nsec = 0; 204 break; 205 case TSP_HZ: 206 getnanotime(tsp); 207 break; 208 case TSP_USEC: 209 microtime(&tv); 210 TIMEVAL_TO_TIMESPEC(&tv, tsp); 211 break; 212 case TSP_NSEC: 213 default: 214 nanotime(tsp); 215 break; 216 } 217 } 218 219 /* 220 * Set vnode attributes to VNOVAL 221 */ 222 void 223 vattr_null(struct vattr *vap) 224 { 225 vap->va_type = VNON; 226 vap->va_size = VNOVAL; 227 vap->va_bytes = VNOVAL; 228 vap->va_mode = VNOVAL; 229 vap->va_nlink = VNOVAL; 230 vap->va_uid = VNOVAL; 231 vap->va_gid = VNOVAL; 232 vap->va_fsid = VNOVAL; 233 vap->va_fileid = VNOVAL; 234 vap->va_blocksize = VNOVAL; 235 vap->va_rmajor = VNOVAL; 236 vap->va_rminor = VNOVAL; 237 vap->va_atime.tv_sec = VNOVAL; 238 vap->va_atime.tv_nsec = VNOVAL; 239 vap->va_mtime.tv_sec = VNOVAL; 240 vap->va_mtime.tv_nsec = VNOVAL; 241 vap->va_ctime.tv_sec = VNOVAL; 242 vap->va_ctime.tv_nsec = VNOVAL; 243 vap->va_flags = VNOVAL; 244 vap->va_gen = VNOVAL; 245 vap->va_vaflags = 0; 246 /* va_*_uuid fields are only valid if related flags are set */ 247 } 248 249 /* 250 * Flush out and invalidate all buffers associated with a vnode. 251 * 252 * vp must be locked. 253 */ 254 static int vinvalbuf_bp(struct buf *bp, void *data); 255 256 struct vinvalbuf_bp_info { 257 struct vnode *vp; 258 int slptimeo; 259 int lkflags; 260 int flags; 261 int clean; 262 }; 263 264 int 265 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 266 { 267 struct vinvalbuf_bp_info info; 268 vm_object_t object; 269 int error; 270 271 lwkt_gettoken(&vp->v_token); 272 273 /* 274 * If we are being asked to save, call fsync to ensure that the inode 275 * is updated. 276 */ 277 if (flags & V_SAVE) { 278 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 279 if (error) 280 goto done; 281 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 282 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 283 goto done; 284 #if 0 285 /* 286 * Dirty bufs may be left or generated via races 287 * in circumstances where vinvalbuf() is called on 288 * a vnode not undergoing reclamation. Only 289 * panic if we are trying to reclaim the vnode. 290 */ 291 if ((vp->v_flag & VRECLAIMED) && 292 (bio_track_active(&vp->v_track_write) || 293 !RB_EMPTY(&vp->v_rbdirty_tree))) { 294 panic("vinvalbuf: dirty bufs"); 295 } 296 #endif 297 } 298 } 299 info.slptimeo = slptimeo; 300 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 301 if (slpflag & PCATCH) 302 info.lkflags |= LK_PCATCH; 303 info.flags = flags; 304 info.vp = vp; 305 306 /* 307 * Flush the buffer cache until nothing is left, wait for all I/O 308 * to complete. At least one pass is required. We might block 309 * in the pip code so we have to re-check. Order is important. 310 */ 311 do { 312 /* 313 * Flush buffer cache 314 */ 315 if (!RB_EMPTY(&vp->v_rbclean_tree)) { 316 info.clean = 1; 317 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 318 NULL, vinvalbuf_bp, &info); 319 } 320 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 321 info.clean = 0; 322 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 323 NULL, vinvalbuf_bp, &info); 324 } 325 326 /* 327 * Wait for I/O completion. 328 */ 329 bio_track_wait(&vp->v_track_write, 0, 0); 330 if ((object = vp->v_object) != NULL) 331 refcount_wait(&object->paging_in_progress, "vnvlbx"); 332 } while (bio_track_active(&vp->v_track_write) || 333 !RB_EMPTY(&vp->v_rbclean_tree) || 334 !RB_EMPTY(&vp->v_rbdirty_tree)); 335 336 /* 337 * Destroy the copy in the VM cache, too. 338 */ 339 if ((object = vp->v_object) != NULL) { 340 vm_object_page_remove(object, 0, 0, 341 (flags & V_SAVE) ? TRUE : FALSE); 342 } 343 344 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 345 panic("vinvalbuf: flush failed"); 346 if (!RB_EMPTY(&vp->v_rbhash_tree)) 347 panic("vinvalbuf: flush failed, buffers still present"); 348 error = 0; 349 done: 350 lwkt_reltoken(&vp->v_token); 351 return (error); 352 } 353 354 static int 355 vinvalbuf_bp(struct buf *bp, void *data) 356 { 357 struct vinvalbuf_bp_info *info = data; 358 int error; 359 360 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 361 atomic_add_int(&bp->b_refs, 1); 362 error = BUF_TIMELOCK(bp, info->lkflags, 363 "vinvalbuf", info->slptimeo); 364 atomic_subtract_int(&bp->b_refs, 1); 365 if (error == 0) { 366 BUF_UNLOCK(bp); 367 error = ENOLCK; 368 } 369 if (error == ENOLCK) 370 return(0); 371 return (-error); 372 } 373 KKASSERT(bp->b_vp == info->vp); 374 375 /* 376 * Must check clean/dirty status after successfully locking as 377 * it may race. 378 */ 379 if ((info->clean && (bp->b_flags & B_DELWRI)) || 380 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 381 BUF_UNLOCK(bp); 382 return(0); 383 } 384 385 /* 386 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 387 * check. This code will write out the buffer, period. 388 */ 389 bremfree(bp); 390 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 391 (info->flags & V_SAVE)) { 392 cluster_awrite(bp); 393 } else if (info->flags & V_SAVE) { 394 /* 395 * Cannot set B_NOCACHE on a clean buffer as this will 396 * destroy the VM backing store which might actually 397 * be dirty (and unsynchronized). 398 */ 399 bp->b_flags |= (B_INVAL | B_RELBUF); 400 brelse(bp); 401 } else { 402 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 403 brelse(bp); 404 } 405 return(0); 406 } 407 408 /* 409 * Truncate a file's buffer and pages to a specified length. This 410 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 411 * sync activity. 412 * 413 * The vnode must be locked. 414 */ 415 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 416 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 417 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 418 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 419 420 struct vtruncbuf_info { 421 struct vnode *vp; 422 off_t truncloffset; 423 int clean; 424 }; 425 426 int 427 vtruncbuf(struct vnode *vp, off_t length, int blksize) 428 { 429 struct vtruncbuf_info info; 430 const char *filename; 431 int count; 432 433 /* 434 * Round up to the *next* block, then destroy the buffers in question. 435 * Since we are only removing some of the buffers we must rely on the 436 * scan count to determine whether a loop is necessary. 437 */ 438 if ((count = (int)(length % blksize)) != 0) 439 info.truncloffset = length + (blksize - count); 440 else 441 info.truncloffset = length; 442 info.vp = vp; 443 444 lwkt_gettoken(&vp->v_token); 445 do { 446 info.clean = 1; 447 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 448 vtruncbuf_bp_trunc_cmp, 449 vtruncbuf_bp_trunc, &info); 450 info.clean = 0; 451 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 452 vtruncbuf_bp_trunc_cmp, 453 vtruncbuf_bp_trunc, &info); 454 } while(count); 455 456 /* 457 * For safety, fsync any remaining metadata if the file is not being 458 * truncated to 0. Since the metadata does not represent the entire 459 * dirty list we have to rely on the hit count to ensure that we get 460 * all of it. 461 */ 462 if (length > 0) { 463 do { 464 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 465 vtruncbuf_bp_metasync_cmp, 466 vtruncbuf_bp_metasync, &info); 467 } while (count); 468 } 469 470 /* 471 * Clean out any left over VM backing store. 472 * 473 * It is possible to have in-progress I/O from buffers that were 474 * not part of the truncation. This should not happen if we 475 * are truncating to 0-length. 476 */ 477 vnode_pager_setsize(vp, length); 478 bio_track_wait(&vp->v_track_write, 0, 0); 479 480 /* 481 * Debugging only 482 */ 483 spin_lock(&vp->v_spin); 484 filename = TAILQ_FIRST(&vp->v_namecache) ? 485 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 486 spin_unlock(&vp->v_spin); 487 488 /* 489 * Make sure no buffers were instantiated while we were trying 490 * to clean out the remaining VM pages. This could occur due 491 * to busy dirty VM pages being flushed out to disk. 492 */ 493 do { 494 info.clean = 1; 495 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 496 vtruncbuf_bp_trunc_cmp, 497 vtruncbuf_bp_trunc, &info); 498 info.clean = 0; 499 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 500 vtruncbuf_bp_trunc_cmp, 501 vtruncbuf_bp_trunc, &info); 502 if (count) { 503 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 504 "left over buffers in %s\n", count, filename); 505 } 506 } while(count); 507 508 lwkt_reltoken(&vp->v_token); 509 510 return (0); 511 } 512 513 /* 514 * The callback buffer is beyond the new file EOF and must be destroyed. 515 * Note that the compare function must conform to the RB_SCAN's requirements. 516 */ 517 static 518 int 519 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 520 { 521 struct vtruncbuf_info *info = data; 522 523 if (bp->b_loffset >= info->truncloffset) 524 return(0); 525 return(-1); 526 } 527 528 static 529 int 530 vtruncbuf_bp_trunc(struct buf *bp, void *data) 531 { 532 struct vtruncbuf_info *info = data; 533 534 /* 535 * Do not try to use a buffer we cannot immediately lock, but sleep 536 * anyway to prevent a livelock. The code will loop until all buffers 537 * can be acted upon. 538 * 539 * We must always revalidate the buffer after locking it to deal 540 * with MP races. 541 */ 542 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 543 atomic_add_int(&bp->b_refs, 1); 544 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 545 BUF_UNLOCK(bp); 546 atomic_subtract_int(&bp->b_refs, 1); 547 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 548 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 549 bp->b_vp != info->vp || 550 vtruncbuf_bp_trunc_cmp(bp, data)) { 551 BUF_UNLOCK(bp); 552 } else { 553 bremfree(bp); 554 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 555 brelse(bp); 556 } 557 return(1); 558 } 559 560 /* 561 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 562 * blocks (with a negative loffset) are scanned. 563 * Note that the compare function must conform to the RB_SCAN's requirements. 564 */ 565 static int 566 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 567 { 568 if (bp->b_loffset < 0) 569 return(0); 570 return(1); 571 } 572 573 static int 574 vtruncbuf_bp_metasync(struct buf *bp, void *data) 575 { 576 struct vtruncbuf_info *info = data; 577 578 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 579 atomic_add_int(&bp->b_refs, 1); 580 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 581 BUF_UNLOCK(bp); 582 atomic_subtract_int(&bp->b_refs, 1); 583 } else if ((bp->b_flags & B_DELWRI) == 0 || 584 bp->b_vp != info->vp || 585 vtruncbuf_bp_metasync_cmp(bp, data)) { 586 BUF_UNLOCK(bp); 587 } else { 588 bremfree(bp); 589 if (bp->b_vp == info->vp) 590 bawrite(bp); 591 else 592 bwrite(bp); 593 } 594 return(1); 595 } 596 597 /* 598 * vfsync - implements a multipass fsync on a file which understands 599 * dependancies and meta-data. The passed vnode must be locked. The 600 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 601 * 602 * When fsyncing data asynchronously just do one consolidated pass starting 603 * with the most negative block number. This may not get all the data due 604 * to dependancies. 605 * 606 * When fsyncing data synchronously do a data pass, then a metadata pass, 607 * then do additional data+metadata passes to try to get all the data out. 608 */ 609 static int vfsync_wait_output(struct vnode *vp, 610 int (*waitoutput)(struct vnode *, struct thread *)); 611 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 612 static int vfsync_data_only_cmp(struct buf *bp, void *data); 613 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 614 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 615 static int vfsync_bp(struct buf *bp, void *data); 616 617 struct vfsync_info { 618 struct vnode *vp; 619 int synchronous; 620 int syncdeps; 621 int lazycount; 622 int lazylimit; 623 int skippedbufs; 624 int (*checkdef)(struct buf *); 625 int (*cmpfunc)(struct buf *, void *); 626 }; 627 628 int 629 vfsync(struct vnode *vp, int waitfor, int passes, 630 int (*checkdef)(struct buf *), 631 int (*waitoutput)(struct vnode *, struct thread *)) 632 { 633 struct vfsync_info info; 634 int error; 635 636 bzero(&info, sizeof(info)); 637 info.vp = vp; 638 if ((info.checkdef = checkdef) == NULL) 639 info.syncdeps = 1; 640 641 lwkt_gettoken(&vp->v_token); 642 643 switch(waitfor) { 644 case MNT_LAZY | MNT_NOWAIT: 645 case MNT_LAZY: 646 /* 647 * Lazy (filesystem syncer typ) Asynchronous plus limit the 648 * number of data (not meta) pages we try to flush to 1MB. 649 * A non-zero return means that lazy limit was reached. 650 */ 651 info.lazylimit = 1024 * 1024; 652 info.syncdeps = 1; 653 info.cmpfunc = vfsync_lazy_range_cmp; 654 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 655 vfsync_lazy_range_cmp, vfsync_bp, &info); 656 info.cmpfunc = vfsync_meta_only_cmp; 657 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 658 vfsync_meta_only_cmp, vfsync_bp, &info); 659 if (error == 0) 660 vp->v_lazyw = 0; 661 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 662 vn_syncer_add(vp, 1); 663 error = 0; 664 break; 665 case MNT_NOWAIT: 666 /* 667 * Asynchronous. Do a data-only pass and a meta-only pass. 668 */ 669 info.syncdeps = 1; 670 info.cmpfunc = vfsync_data_only_cmp; 671 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 672 vfsync_bp, &info); 673 info.cmpfunc = vfsync_meta_only_cmp; 674 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 675 vfsync_bp, &info); 676 error = 0; 677 break; 678 default: 679 /* 680 * Synchronous. Do a data-only pass, then a meta-data+data 681 * pass, then additional integrated passes to try to get 682 * all the dependancies flushed. 683 */ 684 info.cmpfunc = vfsync_data_only_cmp; 685 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 686 vfsync_bp, &info); 687 error = vfsync_wait_output(vp, waitoutput); 688 if (error == 0) { 689 info.skippedbufs = 0; 690 info.cmpfunc = vfsync_dummy_cmp; 691 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 692 vfsync_bp, &info); 693 error = vfsync_wait_output(vp, waitoutput); 694 if (info.skippedbufs) { 695 kprintf("Warning: vfsync skipped %d dirty " 696 "bufs in pass2!\n", info.skippedbufs); 697 } 698 } 699 while (error == 0 && passes > 0 && 700 !RB_EMPTY(&vp->v_rbdirty_tree) 701 ) { 702 if (--passes == 0) { 703 info.synchronous = 1; 704 info.syncdeps = 1; 705 } 706 info.cmpfunc = vfsync_dummy_cmp; 707 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 708 vfsync_bp, &info); 709 if (error < 0) 710 error = -error; 711 info.syncdeps = 1; 712 if (error == 0) 713 error = vfsync_wait_output(vp, waitoutput); 714 } 715 break; 716 } 717 lwkt_reltoken(&vp->v_token); 718 return(error); 719 } 720 721 static int 722 vfsync_wait_output(struct vnode *vp, 723 int (*waitoutput)(struct vnode *, struct thread *)) 724 { 725 int error; 726 727 error = bio_track_wait(&vp->v_track_write, 0, 0); 728 if (waitoutput) 729 error = waitoutput(vp, curthread); 730 return(error); 731 } 732 733 static int 734 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 735 { 736 return(0); 737 } 738 739 static int 740 vfsync_data_only_cmp(struct buf *bp, void *data) 741 { 742 if (bp->b_loffset < 0) 743 return(-1); 744 return(0); 745 } 746 747 static int 748 vfsync_meta_only_cmp(struct buf *bp, void *data) 749 { 750 if (bp->b_loffset < 0) 751 return(0); 752 return(1); 753 } 754 755 static int 756 vfsync_lazy_range_cmp(struct buf *bp, void *data) 757 { 758 struct vfsync_info *info = data; 759 760 if (bp->b_loffset < info->vp->v_lazyw) 761 return(-1); 762 return(0); 763 } 764 765 static int 766 vfsync_bp(struct buf *bp, void *data) 767 { 768 struct vfsync_info *info = data; 769 struct vnode *vp = info->vp; 770 int error; 771 772 /* 773 * Ignore buffers that we cannot immediately lock. 774 */ 775 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 776 ++info->skippedbufs; 777 return(0); 778 } 779 780 /* 781 * We must revalidate the buffer after locking. 782 */ 783 if ((bp->b_flags & B_DELWRI) == 0 || 784 bp->b_vp != info->vp || 785 info->cmpfunc(bp, data)) { 786 BUF_UNLOCK(bp); 787 return(0); 788 } 789 790 /* 791 * If syncdeps is not set we do not try to write buffers which have 792 * dependancies. 793 */ 794 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 795 BUF_UNLOCK(bp); 796 return(0); 797 } 798 799 /* 800 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 801 * has been written but an additional handshake with the device 802 * is required before we can dispose of the buffer. We have no idea 803 * how to do this so we have to skip these buffers. 804 */ 805 if (bp->b_flags & B_NEEDCOMMIT) { 806 BUF_UNLOCK(bp); 807 return(0); 808 } 809 810 /* 811 * Ask bioops if it is ok to sync. If not the VFS may have 812 * set B_LOCKED so we have to cycle the buffer. 813 */ 814 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 815 bremfree(bp); 816 brelse(bp); 817 return(0); 818 } 819 820 if (info->synchronous) { 821 /* 822 * Synchronous flushing. An error may be returned. 823 */ 824 bremfree(bp); 825 error = bwrite(bp); 826 } else { 827 /* 828 * Asynchronous flushing. A negative return value simply 829 * stops the scan and is not considered an error. We use 830 * this to support limited MNT_LAZY flushes. 831 */ 832 vp->v_lazyw = bp->b_loffset; 833 bremfree(bp); 834 info->lazycount += cluster_awrite(bp); 835 waitrunningbufspace(); 836 vm_wait_nominal(); 837 if (info->lazylimit && info->lazycount >= info->lazylimit) 838 error = 1; 839 else 840 error = 0; 841 } 842 return(-error); 843 } 844 845 /* 846 * Associate a buffer with a vnode. 847 * 848 * MPSAFE 849 */ 850 int 851 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 852 { 853 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 854 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 855 856 /* 857 * Insert onto list for new vnode. 858 */ 859 lwkt_gettoken(&vp->v_token); 860 861 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 862 lwkt_reltoken(&vp->v_token); 863 return (EEXIST); 864 } 865 866 /* 867 * Diagnostics (mainly for HAMMER debugging). Check for 868 * overlapping buffers. 869 */ 870 if (check_buf_overlap) { 871 struct buf *bx; 872 bx = buf_rb_hash_RB_PREV(bp); 873 if (bx) { 874 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 875 kprintf("bgetvp: overlapl %016jx/%d %016jx " 876 "bx %p bp %p\n", 877 (intmax_t)bx->b_loffset, 878 bx->b_bufsize, 879 (intmax_t)bp->b_loffset, 880 bx, bp); 881 if (check_buf_overlap > 1) 882 panic("bgetvp - overlapping buffer"); 883 } 884 } 885 bx = buf_rb_hash_RB_NEXT(bp); 886 if (bx) { 887 if (bp->b_loffset + testsize > bx->b_loffset) { 888 kprintf("bgetvp: overlapr %016jx/%d %016jx " 889 "bp %p bx %p\n", 890 (intmax_t)bp->b_loffset, 891 testsize, 892 (intmax_t)bx->b_loffset, 893 bp, bx); 894 if (check_buf_overlap > 1) 895 panic("bgetvp - overlapping buffer"); 896 } 897 } 898 } 899 bp->b_vp = vp; 900 bp->b_flags |= B_HASHED; 901 bp->b_flags |= B_VNCLEAN; 902 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 903 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 904 /*vhold(vp);*/ 905 lwkt_reltoken(&vp->v_token); 906 return(0); 907 } 908 909 /* 910 * Disassociate a buffer from a vnode. 911 * 912 * MPSAFE 913 */ 914 void 915 brelvp(struct buf *bp) 916 { 917 struct vnode *vp; 918 919 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 920 921 /* 922 * Delete from old vnode list, if on one. 923 */ 924 vp = bp->b_vp; 925 lwkt_gettoken(&vp->v_token); 926 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 927 if (bp->b_flags & B_VNDIRTY) 928 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 929 else 930 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 931 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 932 } 933 if (bp->b_flags & B_HASHED) { 934 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 935 bp->b_flags &= ~B_HASHED; 936 } 937 938 /* 939 * Only remove from synclist when no dirty buffers are left AND 940 * the VFS has not flagged the vnode's inode as being dirty. 941 */ 942 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 943 RB_EMPTY(&vp->v_rbdirty_tree)) { 944 vn_syncer_remove(vp); 945 } 946 bp->b_vp = NULL; 947 948 lwkt_reltoken(&vp->v_token); 949 950 /*vdrop(vp);*/ 951 } 952 953 /* 954 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 955 * This routine is called when the state of the B_DELWRI bit is changed. 956 * 957 * Must be called with vp->v_token held. 958 * MPSAFE 959 */ 960 void 961 reassignbuf(struct buf *bp) 962 { 963 struct vnode *vp = bp->b_vp; 964 int delay; 965 966 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 967 ++reassignbufcalls; 968 969 /* 970 * B_PAGING flagged buffers cannot be reassigned because their vp 971 * is not fully linked in. 972 */ 973 if (bp->b_flags & B_PAGING) 974 panic("cannot reassign paging buffer"); 975 976 if (bp->b_flags & B_DELWRI) { 977 /* 978 * Move to the dirty list, add the vnode to the worklist 979 */ 980 if (bp->b_flags & B_VNCLEAN) { 981 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 982 bp->b_flags &= ~B_VNCLEAN; 983 } 984 if ((bp->b_flags & B_VNDIRTY) == 0) { 985 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 986 panic("reassignbuf: dup lblk vp %p bp %p", 987 vp, bp); 988 } 989 bp->b_flags |= B_VNDIRTY; 990 } 991 if ((vp->v_flag & VONWORKLST) == 0) { 992 switch (vp->v_type) { 993 case VDIR: 994 delay = dirdelay; 995 break; 996 case VCHR: 997 case VBLK: 998 if (vp->v_rdev && 999 vp->v_rdev->si_mountpoint != NULL) { 1000 delay = metadelay; 1001 break; 1002 } 1003 /* fall through */ 1004 default: 1005 delay = filedelay; 1006 } 1007 vn_syncer_add(vp, delay); 1008 } 1009 } else { 1010 /* 1011 * Move to the clean list, remove the vnode from the worklist 1012 * if no dirty blocks remain. 1013 */ 1014 if (bp->b_flags & B_VNDIRTY) { 1015 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1016 bp->b_flags &= ~B_VNDIRTY; 1017 } 1018 if ((bp->b_flags & B_VNCLEAN) == 0) { 1019 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1020 panic("reassignbuf: dup lblk vp %p bp %p", 1021 vp, bp); 1022 } 1023 bp->b_flags |= B_VNCLEAN; 1024 } 1025 1026 /* 1027 * Only remove from synclist when no dirty buffers are left 1028 * AND the VFS has not flagged the vnode's inode as being 1029 * dirty. 1030 */ 1031 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1032 VONWORKLST && 1033 RB_EMPTY(&vp->v_rbdirty_tree)) { 1034 vn_syncer_remove(vp); 1035 } 1036 } 1037 } 1038 1039 /* 1040 * Create a vnode for a block device. Used for mounting the root file 1041 * system. 1042 * 1043 * A vref()'d vnode is returned. 1044 */ 1045 extern struct vop_ops *devfs_vnode_dev_vops_p; 1046 int 1047 bdevvp(cdev_t dev, struct vnode **vpp) 1048 { 1049 struct vnode *vp; 1050 struct vnode *nvp; 1051 int error; 1052 1053 if (dev == NULL) { 1054 *vpp = NULLVP; 1055 return (ENXIO); 1056 } 1057 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1058 &nvp, 0, 0); 1059 if (error) { 1060 *vpp = NULLVP; 1061 return (error); 1062 } 1063 vp = nvp; 1064 vp->v_type = VCHR; 1065 #if 0 1066 vp->v_rdev = dev; 1067 #endif 1068 v_associate_rdev(vp, dev); 1069 vp->v_umajor = dev->si_umajor; 1070 vp->v_uminor = dev->si_uminor; 1071 vx_unlock(vp); 1072 *vpp = vp; 1073 return (0); 1074 } 1075 1076 int 1077 v_associate_rdev(struct vnode *vp, cdev_t dev) 1078 { 1079 if (dev == NULL) 1080 return(ENXIO); 1081 if (dev_is_good(dev) == 0) 1082 return(ENXIO); 1083 KKASSERT(vp->v_rdev == NULL); 1084 vp->v_rdev = reference_dev(dev); 1085 lwkt_gettoken(&spechash_token); 1086 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1087 lwkt_reltoken(&spechash_token); 1088 return(0); 1089 } 1090 1091 void 1092 v_release_rdev(struct vnode *vp) 1093 { 1094 cdev_t dev; 1095 1096 if ((dev = vp->v_rdev) != NULL) { 1097 lwkt_gettoken(&spechash_token); 1098 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1099 vp->v_rdev = NULL; 1100 release_dev(dev); 1101 lwkt_reltoken(&spechash_token); 1102 } 1103 } 1104 1105 /* 1106 * Add a vnode to the alias list hung off the cdev_t. We only associate 1107 * the device number with the vnode. The actual device is not associated 1108 * until the vnode is opened (usually in spec_open()), and will be 1109 * disassociated on last close. 1110 */ 1111 void 1112 addaliasu(struct vnode *nvp, int x, int y) 1113 { 1114 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1115 panic("addaliasu on non-special vnode"); 1116 nvp->v_umajor = x; 1117 nvp->v_uminor = y; 1118 } 1119 1120 /* 1121 * Simple call that a filesystem can make to try to get rid of a 1122 * vnode. It will fail if anyone is referencing the vnode (including 1123 * the caller). 1124 * 1125 * The filesystem can check whether its in-memory inode structure still 1126 * references the vp on return. 1127 * 1128 * May only be called if the vnode is in a known state (i.e. being prevented 1129 * from being deallocated by some other condition such as a vfs inode hold). 1130 */ 1131 void 1132 vclean_unlocked(struct vnode *vp) 1133 { 1134 vx_get(vp); 1135 if (VREFCNT(vp) <= 1) 1136 vgone_vxlocked(vp); 1137 vx_put(vp); 1138 } 1139 1140 /* 1141 * Disassociate a vnode from its underlying filesystem. 1142 * 1143 * The vnode must be VX locked and referenced. In all normal situations 1144 * there are no active references. If vclean_vxlocked() is called while 1145 * there are active references, the vnode is being ripped out and we have 1146 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1147 */ 1148 void 1149 vclean_vxlocked(struct vnode *vp, int flags) 1150 { 1151 int active; 1152 int n; 1153 vm_object_t object; 1154 struct namecache *ncp; 1155 1156 /* 1157 * If the vnode has already been reclaimed we have nothing to do. 1158 */ 1159 if (vp->v_flag & VRECLAIMED) 1160 return; 1161 1162 /* 1163 * Set flag to interlock operation, flag finalization to ensure 1164 * that the vnode winds up on the inactive list, and set v_act to 0. 1165 */ 1166 vsetflags(vp, VRECLAIMED); 1167 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1168 vp->v_act = 0; 1169 1170 if (verbose_reclaims) { 1171 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1172 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1173 } 1174 1175 /* 1176 * Scrap the vfs cache 1177 */ 1178 while (cache_inval_vp(vp, 0) != 0) { 1179 kprintf("Warning: vnode %p clean/cache_resolution " 1180 "race detected\n", vp); 1181 tsleep(vp, 0, "vclninv", 2); 1182 } 1183 1184 /* 1185 * Check to see if the vnode is in use. If so we have to reference it 1186 * before we clean it out so that its count cannot fall to zero and 1187 * generate a race against ourselves to recycle it. 1188 */ 1189 active = (VREFCNT(vp) > 0); 1190 1191 /* 1192 * Clean out any buffers associated with the vnode and destroy its 1193 * object, if it has one. 1194 */ 1195 vinvalbuf(vp, V_SAVE, 0, 0); 1196 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1197 1198 /* 1199 * If purging an active vnode (typically during a forced unmount 1200 * or reboot), it must be closed and deactivated before being 1201 * reclaimed. This isn't really all that safe, but what can 1202 * we do? XXX. 1203 * 1204 * Note that neither of these routines unlocks the vnode. 1205 */ 1206 if (active && (flags & DOCLOSE)) { 1207 while ((n = vp->v_opencount) != 0) { 1208 if (vp->v_writecount) 1209 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1210 else 1211 VOP_CLOSE(vp, FNONBLOCK, NULL); 1212 if (vp->v_opencount == n) { 1213 kprintf("Warning: unable to force-close" 1214 " vnode %p\n", vp); 1215 break; 1216 } 1217 } 1218 } 1219 1220 /* 1221 * If the vnode has not been deactivated, deactivated it. Deactivation 1222 * can create new buffers and VM pages so we have to call vinvalbuf() 1223 * again to make sure they all get flushed. 1224 * 1225 * This can occur if a file with a link count of 0 needs to be 1226 * truncated. 1227 * 1228 * If the vnode is already dead don't try to deactivate it. 1229 */ 1230 if ((vp->v_flag & VINACTIVE) == 0) { 1231 vsetflags(vp, VINACTIVE); 1232 if (vp->v_mount) 1233 VOP_INACTIVE(vp); 1234 vinvalbuf(vp, V_SAVE, 0, 0); 1235 } 1236 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1237 1238 /* 1239 * If the vnode has an object, destroy it. 1240 */ 1241 while ((object = vp->v_object) != NULL) { 1242 vm_object_hold(object); 1243 if (object == vp->v_object) 1244 break; 1245 vm_object_drop(object); 1246 } 1247 1248 if (object != NULL) { 1249 if (object->ref_count == 0) { 1250 if ((object->flags & OBJ_DEAD) == 0) 1251 vm_object_terminate(object); 1252 vm_object_drop(object); 1253 vclrflags(vp, VOBJBUF); 1254 } else { 1255 vm_pager_deallocate(object); 1256 vclrflags(vp, VOBJBUF); 1257 vm_object_drop(object); 1258 } 1259 } 1260 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1261 1262 /* 1263 * Reclaim the vnode if not already dead. 1264 */ 1265 if (vp->v_mount && VOP_RECLAIM(vp)) 1266 panic("vclean: cannot reclaim"); 1267 1268 /* 1269 * Done with purge, notify sleepers of the grim news. 1270 */ 1271 vp->v_ops = &dead_vnode_vops_p; 1272 vn_gone(vp); 1273 vp->v_tag = VT_NON; 1274 1275 /* 1276 * If we are destroying an active vnode, reactivate it now that 1277 * we have reassociated it with deadfs. This prevents the system 1278 * from crashing on the vnode due to it being unexpectedly marked 1279 * as inactive or reclaimed. 1280 */ 1281 if (active && (flags & DOCLOSE)) { 1282 vclrflags(vp, VINACTIVE | VRECLAIMED); 1283 } 1284 } 1285 1286 /* 1287 * Eliminate all activity associated with the requested vnode 1288 * and with all vnodes aliased to the requested vnode. 1289 * 1290 * The vnode must be referenced but should not be locked. 1291 */ 1292 int 1293 vrevoke(struct vnode *vp, struct ucred *cred) 1294 { 1295 struct vnode *vq; 1296 struct vnode *vqn; 1297 cdev_t dev; 1298 int error; 1299 1300 /* 1301 * If the vnode has a device association, scrap all vnodes associated 1302 * with the device. Don't let the device disappear on us while we 1303 * are scrapping the vnodes. 1304 * 1305 * The passed vp will probably show up in the list, do not VX lock 1306 * it twice! 1307 * 1308 * Releasing the vnode's rdev here can mess up specfs's call to 1309 * device close, so don't do it. The vnode has been disassociated 1310 * and the device will be closed after the last ref on the related 1311 * fp goes away (if not still open by e.g. the kernel). 1312 */ 1313 if (vp->v_type != VCHR) { 1314 error = fdrevoke(vp, DTYPE_VNODE, cred); 1315 return (error); 1316 } 1317 if ((dev = vp->v_rdev) == NULL) { 1318 return(0); 1319 } 1320 reference_dev(dev); 1321 lwkt_gettoken(&spechash_token); 1322 1323 restart: 1324 vqn = SLIST_FIRST(&dev->si_hlist); 1325 if (vqn) 1326 vhold(vqn); 1327 while ((vq = vqn) != NULL) { 1328 if (VREFCNT(vq) > 0) { 1329 vref(vq); 1330 fdrevoke(vq, DTYPE_VNODE, cred); 1331 /*v_release_rdev(vq);*/ 1332 vrele(vq); 1333 if (vq->v_rdev != dev) { 1334 vdrop(vq); 1335 goto restart; 1336 } 1337 } 1338 vqn = SLIST_NEXT(vq, v_cdevnext); 1339 if (vqn) 1340 vhold(vqn); 1341 vdrop(vq); 1342 } 1343 lwkt_reltoken(&spechash_token); 1344 dev_drevoke(dev); 1345 release_dev(dev); 1346 return (0); 1347 } 1348 1349 /* 1350 * This is called when the object underlying a vnode is being destroyed, 1351 * such as in a remove(). Try to recycle the vnode immediately if the 1352 * only active reference is our reference. 1353 * 1354 * Directory vnodes in the namecache with children cannot be immediately 1355 * recycled because numerous VOP_N*() ops require them to be stable. 1356 * 1357 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1358 * function is a NOP if VRECLAIMED is already set. 1359 */ 1360 int 1361 vrecycle(struct vnode *vp) 1362 { 1363 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1364 if (cache_inval_vp_nonblock(vp)) 1365 return(0); 1366 vgone_vxlocked(vp); 1367 return (1); 1368 } 1369 return (0); 1370 } 1371 1372 /* 1373 * Return the maximum I/O size allowed for strategy calls on VP. 1374 * 1375 * If vp is VCHR or VBLK we dive the device, otherwise we use 1376 * the vp's mount info. 1377 * 1378 * The returned value is clamped at MAXPHYS as most callers cannot use 1379 * buffers larger than that size. 1380 */ 1381 int 1382 vmaxiosize(struct vnode *vp) 1383 { 1384 int maxiosize; 1385 1386 if (vp->v_type == VBLK || vp->v_type == VCHR) 1387 maxiosize = vp->v_rdev->si_iosize_max; 1388 else 1389 maxiosize = vp->v_mount->mnt_iosize_max; 1390 1391 if (maxiosize > MAXPHYS) 1392 maxiosize = MAXPHYS; 1393 return (maxiosize); 1394 } 1395 1396 /* 1397 * Eliminate all activity associated with a vnode in preparation for 1398 * destruction. 1399 * 1400 * The vnode must be VX locked and refd and will remain VX locked and refd 1401 * on return. This routine may be called with the vnode in any state, as 1402 * long as it is VX locked. The vnode will be cleaned out and marked 1403 * VRECLAIMED but will not actually be reused until all existing refs and 1404 * holds go away. 1405 * 1406 * NOTE: This routine may be called on a vnode which has not yet been 1407 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1408 * already been reclaimed. 1409 * 1410 * This routine is not responsible for placing us back on the freelist. 1411 * Instead, it happens automatically when the caller releases the VX lock 1412 * (assuming there aren't any other references). 1413 */ 1414 void 1415 vgone_vxlocked(struct vnode *vp) 1416 { 1417 /* 1418 * assert that the VX lock is held. This is an absolute requirement 1419 * now for vgone_vxlocked() to be called. 1420 */ 1421 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1422 1423 /* 1424 * Clean out the filesystem specific data and set the VRECLAIMED 1425 * bit. Also deactivate the vnode if necessary. 1426 * 1427 * The vnode should have automatically been removed from the syncer 1428 * list as syncer/dirty flags cleared during the cleaning. 1429 */ 1430 vclean_vxlocked(vp, DOCLOSE); 1431 KKASSERT((vp->v_flag & VONWORKLST) == 0); 1432 1433 /* 1434 * Delete from old mount point vnode list, if on one. 1435 */ 1436 if (vp->v_mount != NULL) { 1437 KKASSERT(vp->v_data == NULL); 1438 insmntque(vp, NULL); 1439 } 1440 1441 /* 1442 * If special device, remove it from special device alias list 1443 * if it is on one. This should normally only occur if a vnode is 1444 * being revoked as the device should otherwise have been released 1445 * naturally. 1446 */ 1447 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1448 v_release_rdev(vp); 1449 } 1450 1451 /* 1452 * Set us to VBAD 1453 */ 1454 vp->v_type = VBAD; 1455 } 1456 1457 /* 1458 * Lookup a vnode by device number. 1459 * 1460 * Returns non-zero and *vpp set to a vref'd vnode on success. 1461 * Returns zero on failure. 1462 */ 1463 int 1464 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1465 { 1466 struct vnode *vp; 1467 1468 lwkt_gettoken(&spechash_token); 1469 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1470 if (type == vp->v_type) { 1471 *vpp = vp; 1472 vref(vp); 1473 lwkt_reltoken(&spechash_token); 1474 return (1); 1475 } 1476 } 1477 lwkt_reltoken(&spechash_token); 1478 return (0); 1479 } 1480 1481 /* 1482 * Calculate the total number of references to a special device. This 1483 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1484 * an overloaded field. Since udev2dev can now return NULL, we have 1485 * to check for a NULL v_rdev. 1486 */ 1487 int 1488 count_dev(cdev_t dev) 1489 { 1490 struct vnode *vp; 1491 int count = 0; 1492 1493 if (SLIST_FIRST(&dev->si_hlist)) { 1494 lwkt_gettoken(&spechash_token); 1495 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1496 count += vp->v_opencount; 1497 } 1498 lwkt_reltoken(&spechash_token); 1499 } 1500 return(count); 1501 } 1502 1503 int 1504 vcount(struct vnode *vp) 1505 { 1506 if (vp->v_rdev == NULL) 1507 return(0); 1508 return(count_dev(vp->v_rdev)); 1509 } 1510 1511 /* 1512 * Initialize VMIO for a vnode. This routine MUST be called before a 1513 * VFS can issue buffer cache ops on a vnode. It is typically called 1514 * when a vnode is initialized from its inode. 1515 */ 1516 int 1517 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1518 { 1519 vm_object_t object; 1520 int error = 0; 1521 1522 object = vp->v_object; 1523 if (object) { 1524 vm_object_hold(object); 1525 KKASSERT(vp->v_object == object); 1526 } 1527 1528 if (object == NULL) { 1529 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1530 1531 /* 1532 * Dereference the reference we just created. This assumes 1533 * that the object is associated with the vp. Allow it to 1534 * have zero refs. It cannot be destroyed as long as it 1535 * is associated with the vnode. 1536 */ 1537 vm_object_hold(object); 1538 atomic_add_int(&object->ref_count, -1); 1539 vrele(vp); 1540 } else { 1541 KKASSERT((object->flags & OBJ_DEAD) == 0); 1542 } 1543 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1544 vsetflags(vp, VOBJBUF); 1545 vm_object_drop(object); 1546 1547 return (error); 1548 } 1549 1550 1551 /* 1552 * Print out a description of a vnode. 1553 */ 1554 static char *typename[] = 1555 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1556 1557 void 1558 vprint(char *label, struct vnode *vp) 1559 { 1560 char buf[96]; 1561 1562 if (label != NULL) 1563 kprintf("%s: %p: ", label, (void *)vp); 1564 else 1565 kprintf("%p: ", (void *)vp); 1566 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1567 typename[vp->v_type], 1568 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1569 buf[0] = '\0'; 1570 if (vp->v_flag & VROOT) 1571 strcat(buf, "|VROOT"); 1572 if (vp->v_flag & VPFSROOT) 1573 strcat(buf, "|VPFSROOT"); 1574 if (vp->v_flag & VTEXT) 1575 strcat(buf, "|VTEXT"); 1576 if (vp->v_flag & VSYSTEM) 1577 strcat(buf, "|VSYSTEM"); 1578 if (vp->v_flag & VOBJBUF) 1579 strcat(buf, "|VOBJBUF"); 1580 if (buf[0] != '\0') 1581 kprintf(" flags (%s)", &buf[1]); 1582 if (vp->v_data == NULL) { 1583 kprintf("\n"); 1584 } else { 1585 kprintf("\n\t"); 1586 VOP_PRINT(vp); 1587 } 1588 } 1589 1590 /* 1591 * Do the usual access checking. 1592 * file_mode, uid and gid are from the vnode in question, 1593 * while acc_mode and cred are from the VOP_ACCESS parameter list 1594 */ 1595 int 1596 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1597 mode_t acc_mode, struct ucred *cred) 1598 { 1599 mode_t mask; 1600 int ismember; 1601 1602 /* 1603 * Super-user always gets read/write access, but execute access depends 1604 * on at least one execute bit being set. 1605 */ 1606 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1607 if ((acc_mode & VEXEC) && type != VDIR && 1608 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1609 return (EACCES); 1610 return (0); 1611 } 1612 1613 mask = 0; 1614 1615 /* Otherwise, check the owner. */ 1616 if (cred->cr_uid == uid) { 1617 if (acc_mode & VEXEC) 1618 mask |= S_IXUSR; 1619 if (acc_mode & VREAD) 1620 mask |= S_IRUSR; 1621 if (acc_mode & VWRITE) 1622 mask |= S_IWUSR; 1623 return ((file_mode & mask) == mask ? 0 : EACCES); 1624 } 1625 1626 /* Otherwise, check the groups. */ 1627 ismember = groupmember(gid, cred); 1628 if (cred->cr_svgid == gid || ismember) { 1629 if (acc_mode & VEXEC) 1630 mask |= S_IXGRP; 1631 if (acc_mode & VREAD) 1632 mask |= S_IRGRP; 1633 if (acc_mode & VWRITE) 1634 mask |= S_IWGRP; 1635 return ((file_mode & mask) == mask ? 0 : EACCES); 1636 } 1637 1638 /* Otherwise, check everyone else. */ 1639 if (acc_mode & VEXEC) 1640 mask |= S_IXOTH; 1641 if (acc_mode & VREAD) 1642 mask |= S_IROTH; 1643 if (acc_mode & VWRITE) 1644 mask |= S_IWOTH; 1645 return ((file_mode & mask) == mask ? 0 : EACCES); 1646 } 1647 1648 #ifdef DDB 1649 #include <ddb/ddb.h> 1650 1651 static int db_show_locked_vnodes(struct mount *mp, void *data); 1652 1653 /* 1654 * List all of the locked vnodes in the system. 1655 * Called when debugging the kernel. 1656 */ 1657 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1658 { 1659 kprintf("Locked vnodes\n"); 1660 mountlist_scan(db_show_locked_vnodes, NULL, 1661 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1662 } 1663 1664 static int 1665 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1666 { 1667 struct vnode *vp; 1668 1669 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1670 if (vn_islocked(vp)) 1671 vprint(NULL, vp); 1672 } 1673 return(0); 1674 } 1675 #endif 1676 1677 /* 1678 * Top level filesystem related information gathering. 1679 */ 1680 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1681 1682 static int 1683 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1684 { 1685 int *name = (int *)arg1 - 1; /* XXX */ 1686 u_int namelen = arg2 + 1; /* XXX */ 1687 struct vfsconf *vfsp; 1688 int maxtypenum; 1689 1690 #if 1 || defined(COMPAT_PRELITE2) 1691 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1692 if (namelen == 1) 1693 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1694 #endif 1695 1696 #ifdef notyet 1697 /* all sysctl names at this level are at least name and field */ 1698 if (namelen < 2) 1699 return (ENOTDIR); /* overloaded */ 1700 if (name[0] != VFS_GENERIC) { 1701 vfsp = vfsconf_find_by_typenum(name[0]); 1702 if (vfsp == NULL) 1703 return (EOPNOTSUPP); 1704 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1705 oldp, oldlenp, newp, newlen, p)); 1706 } 1707 #endif 1708 switch (name[1]) { 1709 case VFS_MAXTYPENUM: 1710 if (namelen != 2) 1711 return (ENOTDIR); 1712 maxtypenum = vfsconf_get_maxtypenum(); 1713 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1714 case VFS_CONF: 1715 if (namelen != 3) 1716 return (ENOTDIR); /* overloaded */ 1717 vfsp = vfsconf_find_by_typenum(name[2]); 1718 if (vfsp == NULL) 1719 return (EOPNOTSUPP); 1720 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1721 } 1722 return (EOPNOTSUPP); 1723 } 1724 1725 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1726 "Generic filesystem"); 1727 1728 #if 1 || defined(COMPAT_PRELITE2) 1729 1730 static int 1731 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1732 { 1733 int error; 1734 struct ovfsconf ovfs; 1735 struct sysctl_req *req = (struct sysctl_req*) data; 1736 1737 bzero(&ovfs, sizeof(ovfs)); 1738 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1739 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1740 ovfs.vfc_index = vfsp->vfc_typenum; 1741 ovfs.vfc_refcount = vfsp->vfc_refcount; 1742 ovfs.vfc_flags = vfsp->vfc_flags; 1743 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1744 if (error) 1745 return error; /* abort iteration with error code */ 1746 else 1747 return 0; /* continue iterating with next element */ 1748 } 1749 1750 static int 1751 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1752 { 1753 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1754 } 1755 1756 #endif /* 1 || COMPAT_PRELITE2 */ 1757 1758 /* 1759 * Check to see if a filesystem is mounted on a block device. 1760 */ 1761 int 1762 vfs_mountedon(struct vnode *vp) 1763 { 1764 cdev_t dev; 1765 1766 if ((dev = vp->v_rdev) == NULL) { 1767 /* if (vp->v_type != VBLK) 1768 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1769 } 1770 if (dev != NULL && dev->si_mountpoint) 1771 return (EBUSY); 1772 return (0); 1773 } 1774 1775 /* 1776 * Unmount all filesystems. The list is traversed in reverse order 1777 * of mounting to avoid dependencies. 1778 */ 1779 1780 static int vfs_umountall_callback(struct mount *mp, void *data); 1781 1782 void 1783 vfs_unmountall(void) 1784 { 1785 int count; 1786 1787 do { 1788 count = mountlist_scan(vfs_umountall_callback, 1789 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1790 } while (count); 1791 } 1792 1793 static 1794 int 1795 vfs_umountall_callback(struct mount *mp, void *data) 1796 { 1797 int error; 1798 1799 error = dounmount(mp, MNT_FORCE); 1800 if (error) { 1801 mountlist_remove(mp); 1802 kprintf("unmount of filesystem mounted from %s failed (", 1803 mp->mnt_stat.f_mntfromname); 1804 if (error == EBUSY) 1805 kprintf("BUSY)\n"); 1806 else 1807 kprintf("%d)\n", error); 1808 } 1809 return(1); 1810 } 1811 1812 /* 1813 * Checks the mount flags for parameter mp and put the names comma-separated 1814 * into a string buffer buf with a size limit specified by len. 1815 * 1816 * It returns the number of bytes written into buf, and (*errorp) will be 1817 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1818 * not large enough). The buffer will be 0-terminated if len was not 0. 1819 */ 1820 size_t 1821 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1822 char *buf, size_t len, int *errorp) 1823 { 1824 static const struct mountctl_opt optnames[] = { 1825 { MNT_ASYNC, "asynchronous" }, 1826 { MNT_EXPORTED, "NFS exported" }, 1827 { MNT_LOCAL, "local" }, 1828 { MNT_NOATIME, "noatime" }, 1829 { MNT_NODEV, "nodev" }, 1830 { MNT_NOEXEC, "noexec" }, 1831 { MNT_NOSUID, "nosuid" }, 1832 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1833 { MNT_QUOTA, "with-quotas" }, 1834 { MNT_RDONLY, "read-only" }, 1835 { MNT_SYNCHRONOUS, "synchronous" }, 1836 { MNT_UNION, "union" }, 1837 { MNT_NOCLUSTERR, "noclusterr" }, 1838 { MNT_NOCLUSTERW, "noclusterw" }, 1839 { MNT_SUIDDIR, "suiddir" }, 1840 { MNT_SOFTDEP, "soft-updates" }, 1841 { MNT_IGNORE, "ignore" }, 1842 { 0, NULL} 1843 }; 1844 int bwritten; 1845 int bleft; 1846 int optlen; 1847 int actsize; 1848 1849 *errorp = 0; 1850 bwritten = 0; 1851 bleft = len - 1; /* leave room for trailing \0 */ 1852 1853 /* 1854 * Checks the size of the string. If it contains 1855 * any data, then we will append the new flags to 1856 * it. 1857 */ 1858 actsize = strlen(buf); 1859 if (actsize > 0) 1860 buf += actsize; 1861 1862 /* Default flags if no flags passed */ 1863 if (optp == NULL) 1864 optp = optnames; 1865 1866 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1867 *errorp = EINVAL; 1868 return(0); 1869 } 1870 1871 for (; flags && optp->o_opt; ++optp) { 1872 if ((flags & optp->o_opt) == 0) 1873 continue; 1874 optlen = strlen(optp->o_name); 1875 if (bwritten || actsize > 0) { 1876 if (bleft < 2) { 1877 *errorp = ENOSPC; 1878 break; 1879 } 1880 buf[bwritten++] = ','; 1881 buf[bwritten++] = ' '; 1882 bleft -= 2; 1883 } 1884 if (bleft < optlen) { 1885 *errorp = ENOSPC; 1886 break; 1887 } 1888 bcopy(optp->o_name, buf + bwritten, optlen); 1889 bwritten += optlen; 1890 bleft -= optlen; 1891 flags &= ~optp->o_opt; 1892 } 1893 1894 /* 1895 * Space already reserved for trailing \0 1896 */ 1897 buf[bwritten] = 0; 1898 return (bwritten); 1899 } 1900 1901 /* 1902 * Build hash lists of net addresses and hang them off the mount point. 1903 * Called by ufs_mount() to set up the lists of export addresses. 1904 */ 1905 static int 1906 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1907 const struct export_args *argp) 1908 { 1909 struct netcred *np; 1910 struct radix_node_head *rnh; 1911 int i; 1912 struct radix_node *rn; 1913 struct sockaddr *saddr, *smask = NULL; 1914 struct domain *dom; 1915 int error; 1916 1917 if (argp->ex_addrlen == 0) { 1918 if (mp->mnt_flag & MNT_DEFEXPORTED) 1919 return (EPERM); 1920 np = &nep->ne_defexported; 1921 np->netc_exflags = argp->ex_flags; 1922 np->netc_anon = argp->ex_anon; 1923 np->netc_anon.cr_ref = 1; 1924 mp->mnt_flag |= MNT_DEFEXPORTED; 1925 return (0); 1926 } 1927 1928 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1929 return (EINVAL); 1930 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1931 return (EINVAL); 1932 1933 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1934 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1935 saddr = (struct sockaddr *) (np + 1); 1936 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1937 goto out; 1938 if (saddr->sa_len > argp->ex_addrlen) 1939 saddr->sa_len = argp->ex_addrlen; 1940 if (argp->ex_masklen) { 1941 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1942 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1943 if (error) 1944 goto out; 1945 if (smask->sa_len > argp->ex_masklen) 1946 smask->sa_len = argp->ex_masklen; 1947 } 1948 i = saddr->sa_family; 1949 if ((rnh = nep->ne_rtable[i]) == NULL) { 1950 /* 1951 * Seems silly to initialize every AF when most are not used, 1952 * do so on demand here 1953 */ 1954 SLIST_FOREACH(dom, &domains, dom_next) 1955 if (dom->dom_family == i && dom->dom_rtattach) { 1956 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1957 dom->dom_rtoffset); 1958 break; 1959 } 1960 if ((rnh = nep->ne_rtable[i]) == NULL) { 1961 error = ENOBUFS; 1962 goto out; 1963 } 1964 } 1965 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1966 np->netc_rnodes); 1967 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */ 1968 error = EPERM; 1969 goto out; 1970 } 1971 np->netc_exflags = argp->ex_flags; 1972 np->netc_anon = argp->ex_anon; 1973 np->netc_anon.cr_ref = 1; 1974 return (0); 1975 out: 1976 kfree(np, M_NETADDR); 1977 return (error); 1978 } 1979 1980 /* ARGSUSED */ 1981 static int 1982 vfs_free_netcred(struct radix_node *rn, void *w) 1983 { 1984 struct radix_node_head *rnh = (struct radix_node_head *) w; 1985 1986 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1987 kfree((caddr_t) rn, M_NETADDR); 1988 return (0); 1989 } 1990 1991 /* 1992 * Free the net address hash lists that are hanging off the mount points. 1993 */ 1994 static void 1995 vfs_free_addrlist(struct netexport *nep) 1996 { 1997 int i; 1998 struct radix_node_head *rnh; 1999 2000 for (i = 0; i <= AF_MAX; i++) 2001 if ((rnh = nep->ne_rtable[i])) { 2002 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2003 (caddr_t) rnh); 2004 kfree((caddr_t) rnh, M_RTABLE); 2005 nep->ne_rtable[i] = 0; 2006 } 2007 } 2008 2009 int 2010 vfs_export(struct mount *mp, struct netexport *nep, 2011 const struct export_args *argp) 2012 { 2013 int error; 2014 2015 if (argp->ex_flags & MNT_DELEXPORT) { 2016 if (mp->mnt_flag & MNT_EXPUBLIC) { 2017 vfs_setpublicfs(NULL, NULL, NULL); 2018 mp->mnt_flag &= ~MNT_EXPUBLIC; 2019 } 2020 vfs_free_addrlist(nep); 2021 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2022 } 2023 if (argp->ex_flags & MNT_EXPORTED) { 2024 if (argp->ex_flags & MNT_EXPUBLIC) { 2025 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2026 return (error); 2027 mp->mnt_flag |= MNT_EXPUBLIC; 2028 } 2029 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2030 return (error); 2031 mp->mnt_flag |= MNT_EXPORTED; 2032 } 2033 return (0); 2034 } 2035 2036 2037 /* 2038 * Set the publicly exported filesystem (WebNFS). Currently, only 2039 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2040 */ 2041 int 2042 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2043 const struct export_args *argp) 2044 { 2045 int error; 2046 struct vnode *rvp; 2047 char *cp; 2048 2049 /* 2050 * mp == NULL -> invalidate the current info, the FS is 2051 * no longer exported. May be called from either vfs_export 2052 * or unmount, so check if it hasn't already been done. 2053 */ 2054 if (mp == NULL) { 2055 if (nfs_pub.np_valid) { 2056 nfs_pub.np_valid = 0; 2057 if (nfs_pub.np_index != NULL) { 2058 kfree(nfs_pub.np_index, M_TEMP); 2059 nfs_pub.np_index = NULL; 2060 } 2061 } 2062 return (0); 2063 } 2064 2065 /* 2066 * Only one allowed at a time. 2067 */ 2068 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2069 return (EBUSY); 2070 2071 /* 2072 * Get real filehandle for root of exported FS. 2073 */ 2074 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2075 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2076 2077 if ((error = VFS_ROOT(mp, &rvp))) 2078 return (error); 2079 2080 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2081 return (error); 2082 2083 vput(rvp); 2084 2085 /* 2086 * If an indexfile was specified, pull it in. 2087 */ 2088 if (argp->ex_indexfile != NULL) { 2089 int namelen; 2090 2091 error = vn_get_namelen(rvp, &namelen); 2092 if (error) 2093 return (error); 2094 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2095 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2096 namelen, NULL); 2097 if (!error) { 2098 /* 2099 * Check for illegal filenames. 2100 */ 2101 for (cp = nfs_pub.np_index; *cp; cp++) { 2102 if (*cp == '/') { 2103 error = EINVAL; 2104 break; 2105 } 2106 } 2107 } 2108 if (error) { 2109 kfree(nfs_pub.np_index, M_TEMP); 2110 return (error); 2111 } 2112 } 2113 2114 nfs_pub.np_mount = mp; 2115 nfs_pub.np_valid = 1; 2116 return (0); 2117 } 2118 2119 struct netcred * 2120 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2121 struct sockaddr *nam) 2122 { 2123 struct netcred *np; 2124 struct radix_node_head *rnh; 2125 struct sockaddr *saddr; 2126 2127 np = NULL; 2128 if (mp->mnt_flag & MNT_EXPORTED) { 2129 /* 2130 * Lookup in the export list first. 2131 */ 2132 if (nam != NULL) { 2133 saddr = nam; 2134 rnh = nep->ne_rtable[saddr->sa_family]; 2135 if (rnh != NULL) { 2136 np = (struct netcred *) 2137 (*rnh->rnh_matchaddr)((char *)saddr, 2138 rnh); 2139 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2140 np = NULL; 2141 } 2142 } 2143 /* 2144 * If no address match, use the default if it exists. 2145 */ 2146 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2147 np = &nep->ne_defexported; 2148 } 2149 return (np); 2150 } 2151 2152 /* 2153 * perform msync on all vnodes under a mount point. The mount point must 2154 * be locked. This code is also responsible for lazy-freeing unreferenced 2155 * vnodes whos VM objects no longer contain pages. 2156 * 2157 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2158 * 2159 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2160 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2161 * way up in this high level function. 2162 */ 2163 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2164 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2165 2166 void 2167 vfs_msync(struct mount *mp, int flags) 2168 { 2169 int vmsc_flags; 2170 2171 /* 2172 * tmpfs sets this flag to prevent msync(), sync, and the 2173 * filesystem periodic syncer from trying to flush VM pages 2174 * to swap. Only pure memory pressure flushes tmpfs VM pages 2175 * to swap. 2176 */ 2177 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2178 return; 2179 2180 /* 2181 * Ok, scan the vnodes for work. If the filesystem is using the 2182 * syncer thread feature we can use vsyncscan() instead of 2183 * vmntvnodescan(), which is much faster. 2184 */ 2185 vmsc_flags = VMSC_GETVP; 2186 if (flags != MNT_WAIT) 2187 vmsc_flags |= VMSC_NOWAIT; 2188 2189 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2190 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2191 (void *)(intptr_t)flags); 2192 } else { 2193 vmntvnodescan(mp, vmsc_flags, 2194 vfs_msync_scan1, vfs_msync_scan2, 2195 (void *)(intptr_t)flags); 2196 } 2197 } 2198 2199 /* 2200 * scan1 is a fast pre-check. There could be hundreds of thousands of 2201 * vnodes, we cannot afford to do anything heavy weight until we have a 2202 * fairly good indication that there is work to do. 2203 */ 2204 static 2205 int 2206 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2207 { 2208 int flags = (int)(intptr_t)data; 2209 2210 if ((vp->v_flag & VRECLAIMED) == 0) { 2211 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2212 vp->v_object) { 2213 return(0); /* call scan2 */ 2214 } 2215 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2216 (vp->v_flag & VOBJDIRTY) && 2217 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2218 return(0); /* call scan2 */ 2219 } 2220 } 2221 2222 /* 2223 * do not call scan2, continue the loop 2224 */ 2225 return(-1); 2226 } 2227 2228 /* 2229 * This callback is handed a locked vnode. 2230 */ 2231 static 2232 int 2233 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2234 { 2235 vm_object_t obj; 2236 int flags = (int)(intptr_t)data; 2237 2238 if (vp->v_flag & VRECLAIMED) 2239 return(0); 2240 2241 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2242 if ((obj = vp->v_object) != NULL) { 2243 vm_object_page_clean(obj, 0, 0, 2244 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2245 } 2246 } 2247 return(0); 2248 } 2249 2250 /* 2251 * Wake up anyone interested in vp because it is being revoked. 2252 */ 2253 void 2254 vn_gone(struct vnode *vp) 2255 { 2256 lwkt_gettoken(&vp->v_token); 2257 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2258 lwkt_reltoken(&vp->v_token); 2259 } 2260 2261 /* 2262 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2263 * (or v_rdev might be NULL). 2264 */ 2265 cdev_t 2266 vn_todev(struct vnode *vp) 2267 { 2268 if (vp->v_type != VBLK && vp->v_type != VCHR) 2269 return (NULL); 2270 KKASSERT(vp->v_rdev != NULL); 2271 return (vp->v_rdev); 2272 } 2273 2274 /* 2275 * Check if vnode represents a disk device. The vnode does not need to be 2276 * opened. 2277 * 2278 * MPALMOSTSAFE 2279 */ 2280 int 2281 vn_isdisk(struct vnode *vp, int *errp) 2282 { 2283 cdev_t dev; 2284 2285 if (vp->v_type != VCHR) { 2286 if (errp != NULL) 2287 *errp = ENOTBLK; 2288 return (0); 2289 } 2290 2291 dev = vp->v_rdev; 2292 2293 if (dev == NULL) { 2294 if (errp != NULL) 2295 *errp = ENXIO; 2296 return (0); 2297 } 2298 if (dev_is_good(dev) == 0) { 2299 if (errp != NULL) 2300 *errp = ENXIO; 2301 return (0); 2302 } 2303 if ((dev_dflags(dev) & D_DISK) == 0) { 2304 if (errp != NULL) 2305 *errp = ENOTBLK; 2306 return (0); 2307 } 2308 if (errp != NULL) 2309 *errp = 0; 2310 return (1); 2311 } 2312 2313 int 2314 vn_get_namelen(struct vnode *vp, int *namelen) 2315 { 2316 int error; 2317 register_t retval[2]; 2318 2319 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2320 if (error) 2321 return (error); 2322 *namelen = (int)retval[0]; 2323 return (0); 2324 } 2325 2326 int 2327 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2328 uint16_t d_namlen, const char *d_name) 2329 { 2330 struct dirent *dp; 2331 size_t len; 2332 2333 len = _DIRENT_RECLEN(d_namlen); 2334 if (len > uio->uio_resid) 2335 return(1); 2336 2337 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2338 2339 dp->d_ino = d_ino; 2340 dp->d_namlen = d_namlen; 2341 dp->d_type = d_type; 2342 bcopy(d_name, dp->d_name, d_namlen); 2343 2344 *error = uiomove((caddr_t)dp, len, uio); 2345 2346 kfree(dp, M_TEMP); 2347 2348 return(0); 2349 } 2350 2351 void 2352 vn_mark_atime(struct vnode *vp, struct thread *td) 2353 { 2354 struct proc *p = td->td_proc; 2355 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2356 2357 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2358 VOP_MARKATIME(vp, cred); 2359 } 2360 } 2361