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 * Caller must ref the vnode but does not have to lock it. 610 */ 611 static int vfsync_wait_output(struct vnode *vp, 612 int (*waitoutput)(struct vnode *, struct thread *)); 613 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 614 static int vfsync_data_only_cmp(struct buf *bp, void *data); 615 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 616 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 617 static int vfsync_bp(struct buf *bp, void *data); 618 619 struct vfsync_info { 620 struct vnode *vp; 621 int synchronous; 622 int syncdeps; 623 int lazycount; 624 int lazylimit; 625 int skippedbufs; 626 int (*checkdef)(struct buf *); 627 int (*cmpfunc)(struct buf *, void *); 628 }; 629 630 int 631 vfsync(struct vnode *vp, int waitfor, int passes, 632 int (*checkdef)(struct buf *), 633 int (*waitoutput)(struct vnode *, struct thread *)) 634 { 635 struct vfsync_info info; 636 int error; 637 638 bzero(&info, sizeof(info)); 639 info.vp = vp; 640 if ((info.checkdef = checkdef) == NULL) 641 info.syncdeps = 1; 642 643 lwkt_gettoken(&vp->v_token); 644 645 switch(waitfor) { 646 case MNT_LAZY | MNT_NOWAIT: 647 case MNT_LAZY: 648 /* 649 * Lazy (filesystem syncer typ) Asynchronous plus limit the 650 * number of data (not meta) pages we try to flush to 1MB. 651 * A non-zero return means that lazy limit was reached. 652 */ 653 info.lazylimit = 1024 * 1024; 654 info.syncdeps = 1; 655 info.cmpfunc = vfsync_lazy_range_cmp; 656 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 657 vfsync_lazy_range_cmp, vfsync_bp, &info); 658 info.cmpfunc = vfsync_meta_only_cmp; 659 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 660 vfsync_meta_only_cmp, vfsync_bp, &info); 661 if (error == 0) 662 vp->v_lazyw = 0; 663 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 664 vn_syncer_add(vp, 1); 665 error = 0; 666 break; 667 case MNT_NOWAIT: 668 /* 669 * Asynchronous. Do a data-only pass and a meta-only pass. 670 */ 671 info.syncdeps = 1; 672 info.cmpfunc = vfsync_data_only_cmp; 673 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 674 vfsync_bp, &info); 675 info.cmpfunc = vfsync_meta_only_cmp; 676 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 677 vfsync_bp, &info); 678 error = 0; 679 break; 680 default: 681 /* 682 * Synchronous. Do a data-only pass, then a meta-data+data 683 * pass, then additional integrated passes to try to get 684 * all the dependancies flushed. 685 */ 686 info.cmpfunc = vfsync_data_only_cmp; 687 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 688 vfsync_bp, &info); 689 error = vfsync_wait_output(vp, waitoutput); 690 if (error == 0) { 691 info.skippedbufs = 0; 692 info.cmpfunc = vfsync_dummy_cmp; 693 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 694 vfsync_bp, &info); 695 error = vfsync_wait_output(vp, waitoutput); 696 if (info.skippedbufs) { 697 kprintf("Warning: vfsync skipped %d dirty " 698 "bufs in pass2!\n", info.skippedbufs); 699 } 700 } 701 while (error == 0 && passes > 0 && 702 !RB_EMPTY(&vp->v_rbdirty_tree) 703 ) { 704 if (--passes == 0) { 705 info.synchronous = 1; 706 info.syncdeps = 1; 707 } 708 info.cmpfunc = vfsync_dummy_cmp; 709 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 710 vfsync_bp, &info); 711 if (error < 0) 712 error = -error; 713 info.syncdeps = 1; 714 if (error == 0) 715 error = vfsync_wait_output(vp, waitoutput); 716 } 717 break; 718 } 719 lwkt_reltoken(&vp->v_token); 720 return(error); 721 } 722 723 static int 724 vfsync_wait_output(struct vnode *vp, 725 int (*waitoutput)(struct vnode *, struct thread *)) 726 { 727 int error; 728 729 error = bio_track_wait(&vp->v_track_write, 0, 0); 730 if (waitoutput) 731 error = waitoutput(vp, curthread); 732 return(error); 733 } 734 735 static int 736 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 737 { 738 return(0); 739 } 740 741 static int 742 vfsync_data_only_cmp(struct buf *bp, void *data) 743 { 744 if (bp->b_loffset < 0) 745 return(-1); 746 return(0); 747 } 748 749 static int 750 vfsync_meta_only_cmp(struct buf *bp, void *data) 751 { 752 if (bp->b_loffset < 0) 753 return(0); 754 return(1); 755 } 756 757 static int 758 vfsync_lazy_range_cmp(struct buf *bp, void *data) 759 { 760 struct vfsync_info *info = data; 761 762 if (bp->b_loffset < info->vp->v_lazyw) 763 return(-1); 764 return(0); 765 } 766 767 static int 768 vfsync_bp(struct buf *bp, void *data) 769 { 770 struct vfsync_info *info = data; 771 struct vnode *vp = info->vp; 772 int error; 773 774 /* 775 * Ignore buffers that we cannot immediately lock. 776 */ 777 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 778 ++info->skippedbufs; 779 return(0); 780 } 781 782 /* 783 * We must revalidate the buffer after locking. 784 */ 785 if ((bp->b_flags & B_DELWRI) == 0 || 786 bp->b_vp != info->vp || 787 info->cmpfunc(bp, data)) { 788 BUF_UNLOCK(bp); 789 return(0); 790 } 791 792 /* 793 * If syncdeps is not set we do not try to write buffers which have 794 * dependancies. 795 */ 796 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 797 BUF_UNLOCK(bp); 798 return(0); 799 } 800 801 /* 802 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 803 * has been written but an additional handshake with the device 804 * is required before we can dispose of the buffer. We have no idea 805 * how to do this so we have to skip these buffers. 806 */ 807 if (bp->b_flags & B_NEEDCOMMIT) { 808 BUF_UNLOCK(bp); 809 return(0); 810 } 811 812 /* 813 * Ask bioops if it is ok to sync. If not the VFS may have 814 * set B_LOCKED so we have to cycle the buffer. 815 */ 816 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 817 bremfree(bp); 818 brelse(bp); 819 return(0); 820 } 821 822 if (info->synchronous) { 823 /* 824 * Synchronous flushing. An error may be returned. 825 */ 826 bremfree(bp); 827 error = bwrite(bp); 828 } else { 829 /* 830 * Asynchronous flushing. A negative return value simply 831 * stops the scan and is not considered an error. We use 832 * this to support limited MNT_LAZY flushes. 833 */ 834 vp->v_lazyw = bp->b_loffset; 835 bremfree(bp); 836 info->lazycount += cluster_awrite(bp); 837 waitrunningbufspace(); 838 vm_wait_nominal(); 839 if (info->lazylimit && info->lazycount >= info->lazylimit) 840 error = 1; 841 else 842 error = 0; 843 } 844 return(-error); 845 } 846 847 /* 848 * Associate a buffer with a vnode. 849 * 850 * MPSAFE 851 */ 852 int 853 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 854 { 855 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 856 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 857 858 /* 859 * Insert onto list for new vnode. 860 */ 861 lwkt_gettoken(&vp->v_token); 862 863 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 864 lwkt_reltoken(&vp->v_token); 865 return (EEXIST); 866 } 867 868 /* 869 * Diagnostics (mainly for HAMMER debugging). Check for 870 * overlapping buffers. 871 */ 872 if (check_buf_overlap) { 873 struct buf *bx; 874 bx = buf_rb_hash_RB_PREV(bp); 875 if (bx) { 876 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 877 kprintf("bgetvp: overlapl %016jx/%d %016jx " 878 "bx %p bp %p\n", 879 (intmax_t)bx->b_loffset, 880 bx->b_bufsize, 881 (intmax_t)bp->b_loffset, 882 bx, bp); 883 if (check_buf_overlap > 1) 884 panic("bgetvp - overlapping buffer"); 885 } 886 } 887 bx = buf_rb_hash_RB_NEXT(bp); 888 if (bx) { 889 if (bp->b_loffset + testsize > bx->b_loffset) { 890 kprintf("bgetvp: overlapr %016jx/%d %016jx " 891 "bp %p bx %p\n", 892 (intmax_t)bp->b_loffset, 893 testsize, 894 (intmax_t)bx->b_loffset, 895 bp, bx); 896 if (check_buf_overlap > 1) 897 panic("bgetvp - overlapping buffer"); 898 } 899 } 900 } 901 bp->b_vp = vp; 902 bp->b_flags |= B_HASHED; 903 bp->b_flags |= B_VNCLEAN; 904 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 905 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 906 /*vhold(vp);*/ 907 lwkt_reltoken(&vp->v_token); 908 return(0); 909 } 910 911 /* 912 * Disassociate a buffer from a vnode. 913 * 914 * MPSAFE 915 */ 916 void 917 brelvp(struct buf *bp) 918 { 919 struct vnode *vp; 920 921 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 922 923 /* 924 * Delete from old vnode list, if on one. 925 */ 926 vp = bp->b_vp; 927 lwkt_gettoken(&vp->v_token); 928 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 929 if (bp->b_flags & B_VNDIRTY) 930 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 931 else 932 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 933 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 934 } 935 if (bp->b_flags & B_HASHED) { 936 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 937 bp->b_flags &= ~B_HASHED; 938 } 939 940 /* 941 * Only remove from synclist when no dirty buffers are left AND 942 * the VFS has not flagged the vnode's inode as being dirty. 943 */ 944 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 945 RB_EMPTY(&vp->v_rbdirty_tree)) { 946 vn_syncer_remove(vp); 947 } 948 bp->b_vp = NULL; 949 950 lwkt_reltoken(&vp->v_token); 951 952 /*vdrop(vp);*/ 953 } 954 955 /* 956 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 957 * This routine is called when the state of the B_DELWRI bit is changed. 958 * 959 * Must be called with vp->v_token held. 960 * MPSAFE 961 */ 962 void 963 reassignbuf(struct buf *bp) 964 { 965 struct vnode *vp = bp->b_vp; 966 int delay; 967 968 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 969 ++reassignbufcalls; 970 971 /* 972 * B_PAGING flagged buffers cannot be reassigned because their vp 973 * is not fully linked in. 974 */ 975 if (bp->b_flags & B_PAGING) 976 panic("cannot reassign paging buffer"); 977 978 if (bp->b_flags & B_DELWRI) { 979 /* 980 * Move to the dirty list, add the vnode to the worklist 981 */ 982 if (bp->b_flags & B_VNCLEAN) { 983 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 984 bp->b_flags &= ~B_VNCLEAN; 985 } 986 if ((bp->b_flags & B_VNDIRTY) == 0) { 987 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 988 panic("reassignbuf: dup lblk vp %p bp %p", 989 vp, bp); 990 } 991 bp->b_flags |= B_VNDIRTY; 992 } 993 if ((vp->v_flag & VONWORKLST) == 0) { 994 switch (vp->v_type) { 995 case VDIR: 996 delay = dirdelay; 997 break; 998 case VCHR: 999 case VBLK: 1000 if (vp->v_rdev && 1001 vp->v_rdev->si_mountpoint != NULL) { 1002 delay = metadelay; 1003 break; 1004 } 1005 /* fall through */ 1006 default: 1007 delay = filedelay; 1008 } 1009 vn_syncer_add(vp, delay); 1010 } 1011 } else { 1012 /* 1013 * Move to the clean list, remove the vnode from the worklist 1014 * if no dirty blocks remain. 1015 */ 1016 if (bp->b_flags & B_VNDIRTY) { 1017 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1018 bp->b_flags &= ~B_VNDIRTY; 1019 } 1020 if ((bp->b_flags & B_VNCLEAN) == 0) { 1021 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1022 panic("reassignbuf: dup lblk vp %p bp %p", 1023 vp, bp); 1024 } 1025 bp->b_flags |= B_VNCLEAN; 1026 } 1027 1028 /* 1029 * Only remove from synclist when no dirty buffers are left 1030 * AND the VFS has not flagged the vnode's inode as being 1031 * dirty. 1032 */ 1033 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1034 VONWORKLST && 1035 RB_EMPTY(&vp->v_rbdirty_tree)) { 1036 vn_syncer_remove(vp); 1037 } 1038 } 1039 } 1040 1041 /* 1042 * Create a vnode for a block device. Used for mounting the root file 1043 * system. 1044 * 1045 * A vref()'d vnode is returned. 1046 */ 1047 extern struct vop_ops *devfs_vnode_dev_vops_p; 1048 int 1049 bdevvp(cdev_t dev, struct vnode **vpp) 1050 { 1051 struct vnode *vp; 1052 struct vnode *nvp; 1053 int error; 1054 1055 if (dev == NULL) { 1056 *vpp = NULLVP; 1057 return (ENXIO); 1058 } 1059 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1060 &nvp, 0, 0); 1061 if (error) { 1062 *vpp = NULLVP; 1063 return (error); 1064 } 1065 vp = nvp; 1066 vp->v_type = VCHR; 1067 #if 0 1068 vp->v_rdev = dev; 1069 #endif 1070 v_associate_rdev(vp, dev); 1071 vp->v_umajor = dev->si_umajor; 1072 vp->v_uminor = dev->si_uminor; 1073 vx_unlock(vp); 1074 *vpp = vp; 1075 return (0); 1076 } 1077 1078 int 1079 v_associate_rdev(struct vnode *vp, cdev_t dev) 1080 { 1081 if (dev == NULL) 1082 return(ENXIO); 1083 if (dev_is_good(dev) == 0) 1084 return(ENXIO); 1085 KKASSERT(vp->v_rdev == NULL); 1086 vp->v_rdev = reference_dev(dev); 1087 lwkt_gettoken(&spechash_token); 1088 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1089 lwkt_reltoken(&spechash_token); 1090 return(0); 1091 } 1092 1093 void 1094 v_release_rdev(struct vnode *vp) 1095 { 1096 cdev_t dev; 1097 1098 if ((dev = vp->v_rdev) != NULL) { 1099 lwkt_gettoken(&spechash_token); 1100 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1101 vp->v_rdev = NULL; 1102 release_dev(dev); 1103 lwkt_reltoken(&spechash_token); 1104 } 1105 } 1106 1107 /* 1108 * Add a vnode to the alias list hung off the cdev_t. We only associate 1109 * the device number with the vnode. The actual device is not associated 1110 * until the vnode is opened (usually in spec_open()), and will be 1111 * disassociated on last close. 1112 */ 1113 void 1114 addaliasu(struct vnode *nvp, int x, int y) 1115 { 1116 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1117 panic("addaliasu on non-special vnode"); 1118 nvp->v_umajor = x; 1119 nvp->v_uminor = y; 1120 } 1121 1122 /* 1123 * Simple call that a filesystem can make to try to get rid of a 1124 * vnode. It will fail if anyone is referencing the vnode (including 1125 * the caller). 1126 * 1127 * The filesystem can check whether its in-memory inode structure still 1128 * references the vp on return. 1129 * 1130 * May only be called if the vnode is in a known state (i.e. being prevented 1131 * from being deallocated by some other condition such as a vfs inode hold). 1132 */ 1133 void 1134 vclean_unlocked(struct vnode *vp) 1135 { 1136 vx_get(vp); 1137 if (VREFCNT(vp) <= 1) 1138 vgone_vxlocked(vp); 1139 vx_put(vp); 1140 } 1141 1142 /* 1143 * Disassociate a vnode from its underlying filesystem. 1144 * 1145 * The vnode must be VX locked and referenced. In all normal situations 1146 * there are no active references. If vclean_vxlocked() is called while 1147 * there are active references, the vnode is being ripped out and we have 1148 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1149 */ 1150 void 1151 vclean_vxlocked(struct vnode *vp, int flags) 1152 { 1153 int active; 1154 int n; 1155 vm_object_t object; 1156 struct namecache *ncp; 1157 1158 /* 1159 * If the vnode has already been reclaimed we have nothing to do. 1160 */ 1161 if (vp->v_flag & VRECLAIMED) 1162 return; 1163 1164 /* 1165 * Set flag to interlock operation, flag finalization to ensure 1166 * that the vnode winds up on the inactive list, and set v_act to 0. 1167 */ 1168 vsetflags(vp, VRECLAIMED); 1169 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1170 vp->v_act = 0; 1171 1172 if (verbose_reclaims) { 1173 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1174 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1175 } 1176 1177 /* 1178 * Scrap the vfs cache 1179 */ 1180 while (cache_inval_vp(vp, 0) != 0) { 1181 kprintf("Warning: vnode %p clean/cache_resolution " 1182 "race detected\n", vp); 1183 tsleep(vp, 0, "vclninv", 2); 1184 } 1185 1186 /* 1187 * Check to see if the vnode is in use. If so we have to reference it 1188 * before we clean it out so that its count cannot fall to zero and 1189 * generate a race against ourselves to recycle it. 1190 */ 1191 active = (VREFCNT(vp) > 0); 1192 1193 /* 1194 * Clean out any buffers associated with the vnode and destroy its 1195 * object, if it has one. 1196 */ 1197 vinvalbuf(vp, V_SAVE, 0, 0); 1198 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1199 1200 /* 1201 * If purging an active vnode (typically during a forced unmount 1202 * or reboot), it must be closed and deactivated before being 1203 * reclaimed. This isn't really all that safe, but what can 1204 * we do? XXX. 1205 * 1206 * Note that neither of these routines unlocks the vnode. 1207 */ 1208 if (active && (flags & DOCLOSE)) { 1209 while ((n = vp->v_opencount) != 0) { 1210 if (vp->v_writecount) 1211 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1212 else 1213 VOP_CLOSE(vp, FNONBLOCK, NULL); 1214 if (vp->v_opencount == n) { 1215 kprintf("Warning: unable to force-close" 1216 " vnode %p\n", vp); 1217 break; 1218 } 1219 } 1220 } 1221 1222 /* 1223 * If the vnode has not been deactivated, deactivated it. Deactivation 1224 * can create new buffers and VM pages so we have to call vinvalbuf() 1225 * again to make sure they all get flushed. 1226 * 1227 * This can occur if a file with a link count of 0 needs to be 1228 * truncated. 1229 * 1230 * If the vnode is already dead don't try to deactivate it. 1231 */ 1232 if ((vp->v_flag & VINACTIVE) == 0) { 1233 vsetflags(vp, VINACTIVE); 1234 if (vp->v_mount) 1235 VOP_INACTIVE(vp); 1236 vinvalbuf(vp, V_SAVE, 0, 0); 1237 } 1238 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1239 1240 /* 1241 * If the vnode has an object, destroy it. 1242 */ 1243 while ((object = vp->v_object) != NULL) { 1244 vm_object_hold(object); 1245 if (object == vp->v_object) 1246 break; 1247 vm_object_drop(object); 1248 } 1249 1250 if (object != NULL) { 1251 if (object->ref_count == 0) { 1252 if ((object->flags & OBJ_DEAD) == 0) 1253 vm_object_terminate(object); 1254 vm_object_drop(object); 1255 vclrflags(vp, VOBJBUF); 1256 } else { 1257 vm_pager_deallocate(object); 1258 vclrflags(vp, VOBJBUF); 1259 vm_object_drop(object); 1260 } 1261 } 1262 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1263 1264 /* 1265 * Reclaim the vnode if not already dead. 1266 */ 1267 if (vp->v_mount && VOP_RECLAIM(vp)) 1268 panic("vclean: cannot reclaim"); 1269 1270 /* 1271 * Done with purge, notify sleepers of the grim news. 1272 */ 1273 vp->v_ops = &dead_vnode_vops_p; 1274 vn_gone(vp); 1275 vp->v_tag = VT_NON; 1276 1277 /* 1278 * If we are destroying an active vnode, reactivate it now that 1279 * we have reassociated it with deadfs. This prevents the system 1280 * from crashing on the vnode due to it being unexpectedly marked 1281 * as inactive or reclaimed. 1282 */ 1283 if (active && (flags & DOCLOSE)) { 1284 vclrflags(vp, VINACTIVE | VRECLAIMED); 1285 } 1286 } 1287 1288 /* 1289 * Eliminate all activity associated with the requested vnode 1290 * and with all vnodes aliased to the requested vnode. 1291 * 1292 * The vnode must be referenced but should not be locked. 1293 */ 1294 int 1295 vrevoke(struct vnode *vp, struct ucred *cred) 1296 { 1297 struct vnode *vq; 1298 struct vnode *vqn; 1299 cdev_t dev; 1300 int error; 1301 1302 /* 1303 * If the vnode has a device association, scrap all vnodes associated 1304 * with the device. Don't let the device disappear on us while we 1305 * are scrapping the vnodes. 1306 * 1307 * The passed vp will probably show up in the list, do not VX lock 1308 * it twice! 1309 * 1310 * Releasing the vnode's rdev here can mess up specfs's call to 1311 * device close, so don't do it. The vnode has been disassociated 1312 * and the device will be closed after the last ref on the related 1313 * fp goes away (if not still open by e.g. the kernel). 1314 */ 1315 if (vp->v_type != VCHR) { 1316 error = fdrevoke(vp, DTYPE_VNODE, cred); 1317 return (error); 1318 } 1319 if ((dev = vp->v_rdev) == NULL) { 1320 return(0); 1321 } 1322 reference_dev(dev); 1323 lwkt_gettoken(&spechash_token); 1324 1325 restart: 1326 vqn = SLIST_FIRST(&dev->si_hlist); 1327 if (vqn) 1328 vhold(vqn); 1329 while ((vq = vqn) != NULL) { 1330 if (VREFCNT(vq) > 0) { 1331 vref(vq); 1332 fdrevoke(vq, DTYPE_VNODE, cred); 1333 /*v_release_rdev(vq);*/ 1334 vrele(vq); 1335 if (vq->v_rdev != dev) { 1336 vdrop(vq); 1337 goto restart; 1338 } 1339 } 1340 vqn = SLIST_NEXT(vq, v_cdevnext); 1341 if (vqn) 1342 vhold(vqn); 1343 vdrop(vq); 1344 } 1345 lwkt_reltoken(&spechash_token); 1346 dev_drevoke(dev); 1347 release_dev(dev); 1348 return (0); 1349 } 1350 1351 /* 1352 * This is called when the object underlying a vnode is being destroyed, 1353 * such as in a remove(). Try to recycle the vnode immediately if the 1354 * only active reference is our reference. 1355 * 1356 * Directory vnodes in the namecache with children cannot be immediately 1357 * recycled because numerous VOP_N*() ops require them to be stable. 1358 * 1359 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1360 * function is a NOP if VRECLAIMED is already set. 1361 */ 1362 int 1363 vrecycle(struct vnode *vp) 1364 { 1365 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1366 if (cache_inval_vp_nonblock(vp)) 1367 return(0); 1368 vgone_vxlocked(vp); 1369 return (1); 1370 } 1371 return (0); 1372 } 1373 1374 /* 1375 * Return the maximum I/O size allowed for strategy calls on VP. 1376 * 1377 * If vp is VCHR or VBLK we dive the device, otherwise we use 1378 * the vp's mount info. 1379 * 1380 * The returned value is clamped at MAXPHYS as most callers cannot use 1381 * buffers larger than that size. 1382 */ 1383 int 1384 vmaxiosize(struct vnode *vp) 1385 { 1386 int maxiosize; 1387 1388 if (vp->v_type == VBLK || vp->v_type == VCHR) 1389 maxiosize = vp->v_rdev->si_iosize_max; 1390 else 1391 maxiosize = vp->v_mount->mnt_iosize_max; 1392 1393 if (maxiosize > MAXPHYS) 1394 maxiosize = MAXPHYS; 1395 return (maxiosize); 1396 } 1397 1398 /* 1399 * Eliminate all activity associated with a vnode in preparation for 1400 * destruction. 1401 * 1402 * The vnode must be VX locked and refd and will remain VX locked and refd 1403 * on return. This routine may be called with the vnode in any state, as 1404 * long as it is VX locked. The vnode will be cleaned out and marked 1405 * VRECLAIMED but will not actually be reused until all existing refs and 1406 * holds go away. 1407 * 1408 * NOTE: This routine may be called on a vnode which has not yet been 1409 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1410 * already been reclaimed. 1411 * 1412 * This routine is not responsible for placing us back on the freelist. 1413 * Instead, it happens automatically when the caller releases the VX lock 1414 * (assuming there aren't any other references). 1415 */ 1416 void 1417 vgone_vxlocked(struct vnode *vp) 1418 { 1419 /* 1420 * assert that the VX lock is held. This is an absolute requirement 1421 * now for vgone_vxlocked() to be called. 1422 */ 1423 KKASSERT(lockcountnb(&vp->v_lock) == 1); 1424 1425 /* 1426 * Clean out the filesystem specific data and set the VRECLAIMED 1427 * bit. Also deactivate the vnode if necessary. 1428 * 1429 * The vnode should have automatically been removed from the syncer 1430 * list as syncer/dirty flags cleared during the cleaning. 1431 */ 1432 vclean_vxlocked(vp, DOCLOSE); 1433 KKASSERT((vp->v_flag & VONWORKLST) == 0); 1434 1435 /* 1436 * Delete from old mount point vnode list, if on one. 1437 */ 1438 if (vp->v_mount != NULL) { 1439 KKASSERT(vp->v_data == NULL); 1440 insmntque(vp, NULL); 1441 } 1442 1443 /* 1444 * If special device, remove it from special device alias list 1445 * if it is on one. This should normally only occur if a vnode is 1446 * being revoked as the device should otherwise have been released 1447 * naturally. 1448 */ 1449 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1450 v_release_rdev(vp); 1451 } 1452 1453 /* 1454 * Set us to VBAD 1455 */ 1456 vp->v_type = VBAD; 1457 } 1458 1459 /* 1460 * Lookup a vnode by device number. 1461 * 1462 * Returns non-zero and *vpp set to a vref'd vnode on success. 1463 * Returns zero on failure. 1464 */ 1465 int 1466 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1467 { 1468 struct vnode *vp; 1469 1470 lwkt_gettoken(&spechash_token); 1471 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1472 if (type == vp->v_type) { 1473 *vpp = vp; 1474 vref(vp); 1475 lwkt_reltoken(&spechash_token); 1476 return (1); 1477 } 1478 } 1479 lwkt_reltoken(&spechash_token); 1480 return (0); 1481 } 1482 1483 /* 1484 * Calculate the total number of references to a special device. This 1485 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1486 * an overloaded field. Since udev2dev can now return NULL, we have 1487 * to check for a NULL v_rdev. 1488 */ 1489 int 1490 count_dev(cdev_t dev) 1491 { 1492 struct vnode *vp; 1493 int count = 0; 1494 1495 if (SLIST_FIRST(&dev->si_hlist)) { 1496 lwkt_gettoken(&spechash_token); 1497 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1498 count += vp->v_opencount; 1499 } 1500 lwkt_reltoken(&spechash_token); 1501 } 1502 return(count); 1503 } 1504 1505 int 1506 vcount(struct vnode *vp) 1507 { 1508 if (vp->v_rdev == NULL) 1509 return(0); 1510 return(count_dev(vp->v_rdev)); 1511 } 1512 1513 /* 1514 * Initialize VMIO for a vnode. This routine MUST be called before a 1515 * VFS can issue buffer cache ops on a vnode. It is typically called 1516 * when a vnode is initialized from its inode. 1517 */ 1518 int 1519 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1520 { 1521 vm_object_t object; 1522 int error = 0; 1523 1524 object = vp->v_object; 1525 if (object) { 1526 vm_object_hold(object); 1527 KKASSERT(vp->v_object == object); 1528 } 1529 1530 if (object == NULL) { 1531 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1532 1533 /* 1534 * Dereference the reference we just created. This assumes 1535 * that the object is associated with the vp. Allow it to 1536 * have zero refs. It cannot be destroyed as long as it 1537 * is associated with the vnode. 1538 */ 1539 vm_object_hold(object); 1540 atomic_add_int(&object->ref_count, -1); 1541 vrele(vp); 1542 } else { 1543 KKASSERT((object->flags & OBJ_DEAD) == 0); 1544 } 1545 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1546 vsetflags(vp, VOBJBUF); 1547 vm_object_drop(object); 1548 1549 return (error); 1550 } 1551 1552 1553 /* 1554 * Print out a description of a vnode. 1555 */ 1556 static char *typename[] = 1557 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1558 1559 void 1560 vprint(char *label, struct vnode *vp) 1561 { 1562 char buf[96]; 1563 1564 if (label != NULL) 1565 kprintf("%s: %p: ", label, (void *)vp); 1566 else 1567 kprintf("%p: ", (void *)vp); 1568 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1569 typename[vp->v_type], 1570 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1571 buf[0] = '\0'; 1572 if (vp->v_flag & VROOT) 1573 strcat(buf, "|VROOT"); 1574 if (vp->v_flag & VPFSROOT) 1575 strcat(buf, "|VPFSROOT"); 1576 if (vp->v_flag & VTEXT) 1577 strcat(buf, "|VTEXT"); 1578 if (vp->v_flag & VSYSTEM) 1579 strcat(buf, "|VSYSTEM"); 1580 if (vp->v_flag & VOBJBUF) 1581 strcat(buf, "|VOBJBUF"); 1582 if (buf[0] != '\0') 1583 kprintf(" flags (%s)", &buf[1]); 1584 if (vp->v_data == NULL) { 1585 kprintf("\n"); 1586 } else { 1587 kprintf("\n\t"); 1588 VOP_PRINT(vp); 1589 } 1590 } 1591 1592 /* 1593 * Do the usual access checking. 1594 * file_mode, uid and gid are from the vnode in question, 1595 * while acc_mode and cred are from the VOP_ACCESS parameter list 1596 */ 1597 int 1598 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1599 mode_t acc_mode, struct ucred *cred) 1600 { 1601 mode_t mask; 1602 int ismember; 1603 1604 /* 1605 * Super-user always gets read/write access, but execute access depends 1606 * on at least one execute bit being set. 1607 */ 1608 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1609 if ((acc_mode & VEXEC) && type != VDIR && 1610 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1611 return (EACCES); 1612 return (0); 1613 } 1614 1615 mask = 0; 1616 1617 /* Otherwise, check the owner. */ 1618 if (cred->cr_uid == uid) { 1619 if (acc_mode & VEXEC) 1620 mask |= S_IXUSR; 1621 if (acc_mode & VREAD) 1622 mask |= S_IRUSR; 1623 if (acc_mode & VWRITE) 1624 mask |= S_IWUSR; 1625 return ((file_mode & mask) == mask ? 0 : EACCES); 1626 } 1627 1628 /* Otherwise, check the groups. */ 1629 ismember = groupmember(gid, cred); 1630 if (cred->cr_svgid == gid || ismember) { 1631 if (acc_mode & VEXEC) 1632 mask |= S_IXGRP; 1633 if (acc_mode & VREAD) 1634 mask |= S_IRGRP; 1635 if (acc_mode & VWRITE) 1636 mask |= S_IWGRP; 1637 return ((file_mode & mask) == mask ? 0 : EACCES); 1638 } 1639 1640 /* Otherwise, check everyone else. */ 1641 if (acc_mode & VEXEC) 1642 mask |= S_IXOTH; 1643 if (acc_mode & VREAD) 1644 mask |= S_IROTH; 1645 if (acc_mode & VWRITE) 1646 mask |= S_IWOTH; 1647 return ((file_mode & mask) == mask ? 0 : EACCES); 1648 } 1649 1650 #ifdef DDB 1651 #include <ddb/ddb.h> 1652 1653 static int db_show_locked_vnodes(struct mount *mp, void *data); 1654 1655 /* 1656 * List all of the locked vnodes in the system. 1657 * Called when debugging the kernel. 1658 */ 1659 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1660 { 1661 kprintf("Locked vnodes\n"); 1662 mountlist_scan(db_show_locked_vnodes, NULL, 1663 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1664 } 1665 1666 static int 1667 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1668 { 1669 struct vnode *vp; 1670 1671 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1672 if (vn_islocked(vp)) 1673 vprint(NULL, vp); 1674 } 1675 return(0); 1676 } 1677 #endif 1678 1679 /* 1680 * Top level filesystem related information gathering. 1681 */ 1682 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1683 1684 static int 1685 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1686 { 1687 int *name = (int *)arg1 - 1; /* XXX */ 1688 u_int namelen = arg2 + 1; /* XXX */ 1689 struct vfsconf *vfsp; 1690 int maxtypenum; 1691 1692 #if 1 || defined(COMPAT_PRELITE2) 1693 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1694 if (namelen == 1) 1695 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1696 #endif 1697 1698 #ifdef notyet 1699 /* all sysctl names at this level are at least name and field */ 1700 if (namelen < 2) 1701 return (ENOTDIR); /* overloaded */ 1702 if (name[0] != VFS_GENERIC) { 1703 vfsp = vfsconf_find_by_typenum(name[0]); 1704 if (vfsp == NULL) 1705 return (EOPNOTSUPP); 1706 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1707 oldp, oldlenp, newp, newlen, p)); 1708 } 1709 #endif 1710 switch (name[1]) { 1711 case VFS_MAXTYPENUM: 1712 if (namelen != 2) 1713 return (ENOTDIR); 1714 maxtypenum = vfsconf_get_maxtypenum(); 1715 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1716 case VFS_CONF: 1717 if (namelen != 3) 1718 return (ENOTDIR); /* overloaded */ 1719 vfsp = vfsconf_find_by_typenum(name[2]); 1720 if (vfsp == NULL) 1721 return (EOPNOTSUPP); 1722 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1723 } 1724 return (EOPNOTSUPP); 1725 } 1726 1727 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1728 "Generic filesystem"); 1729 1730 #if 1 || defined(COMPAT_PRELITE2) 1731 1732 static int 1733 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1734 { 1735 int error; 1736 struct ovfsconf ovfs; 1737 struct sysctl_req *req = (struct sysctl_req*) data; 1738 1739 bzero(&ovfs, sizeof(ovfs)); 1740 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1741 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1742 ovfs.vfc_index = vfsp->vfc_typenum; 1743 ovfs.vfc_refcount = vfsp->vfc_refcount; 1744 ovfs.vfc_flags = vfsp->vfc_flags; 1745 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1746 if (error) 1747 return error; /* abort iteration with error code */ 1748 else 1749 return 0; /* continue iterating with next element */ 1750 } 1751 1752 static int 1753 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1754 { 1755 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1756 } 1757 1758 #endif /* 1 || COMPAT_PRELITE2 */ 1759 1760 /* 1761 * Check to see if a filesystem is mounted on a block device. 1762 */ 1763 int 1764 vfs_mountedon(struct vnode *vp) 1765 { 1766 cdev_t dev; 1767 1768 if ((dev = vp->v_rdev) == NULL) { 1769 /* if (vp->v_type != VBLK) 1770 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1771 } 1772 if (dev != NULL && dev->si_mountpoint) 1773 return (EBUSY); 1774 return (0); 1775 } 1776 1777 /* 1778 * Unmount all filesystems. The list is traversed in reverse order 1779 * of mounting to avoid dependencies. 1780 */ 1781 1782 static int vfs_umountall_callback(struct mount *mp, void *data); 1783 1784 void 1785 vfs_unmountall(void) 1786 { 1787 int count; 1788 1789 do { 1790 count = mountlist_scan(vfs_umountall_callback, 1791 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1792 } while (count); 1793 } 1794 1795 static 1796 int 1797 vfs_umountall_callback(struct mount *mp, void *data) 1798 { 1799 int error; 1800 1801 error = dounmount(mp, MNT_FORCE); 1802 if (error) { 1803 mountlist_remove(mp); 1804 kprintf("unmount of filesystem mounted from %s failed (", 1805 mp->mnt_stat.f_mntfromname); 1806 if (error == EBUSY) 1807 kprintf("BUSY)\n"); 1808 else 1809 kprintf("%d)\n", error); 1810 } 1811 return(1); 1812 } 1813 1814 /* 1815 * Checks the mount flags for parameter mp and put the names comma-separated 1816 * into a string buffer buf with a size limit specified by len. 1817 * 1818 * It returns the number of bytes written into buf, and (*errorp) will be 1819 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1820 * not large enough). The buffer will be 0-terminated if len was not 0. 1821 */ 1822 size_t 1823 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1824 char *buf, size_t len, int *errorp) 1825 { 1826 static const struct mountctl_opt optnames[] = { 1827 { MNT_ASYNC, "asynchronous" }, 1828 { MNT_EXPORTED, "NFS exported" }, 1829 { MNT_LOCAL, "local" }, 1830 { MNT_NOATIME, "noatime" }, 1831 { MNT_NODEV, "nodev" }, 1832 { MNT_NOEXEC, "noexec" }, 1833 { MNT_NOSUID, "nosuid" }, 1834 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1835 { MNT_QUOTA, "with-quotas" }, 1836 { MNT_RDONLY, "read-only" }, 1837 { MNT_SYNCHRONOUS, "synchronous" }, 1838 { MNT_UNION, "union" }, 1839 { MNT_NOCLUSTERR, "noclusterr" }, 1840 { MNT_NOCLUSTERW, "noclusterw" }, 1841 { MNT_SUIDDIR, "suiddir" }, 1842 { MNT_SOFTDEP, "soft-updates" }, 1843 { MNT_IGNORE, "ignore" }, 1844 { 0, NULL} 1845 }; 1846 int bwritten; 1847 int bleft; 1848 int optlen; 1849 int actsize; 1850 1851 *errorp = 0; 1852 bwritten = 0; 1853 bleft = len - 1; /* leave room for trailing \0 */ 1854 1855 /* 1856 * Checks the size of the string. If it contains 1857 * any data, then we will append the new flags to 1858 * it. 1859 */ 1860 actsize = strlen(buf); 1861 if (actsize > 0) 1862 buf += actsize; 1863 1864 /* Default flags if no flags passed */ 1865 if (optp == NULL) 1866 optp = optnames; 1867 1868 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1869 *errorp = EINVAL; 1870 return(0); 1871 } 1872 1873 for (; flags && optp->o_opt; ++optp) { 1874 if ((flags & optp->o_opt) == 0) 1875 continue; 1876 optlen = strlen(optp->o_name); 1877 if (bwritten || actsize > 0) { 1878 if (bleft < 2) { 1879 *errorp = ENOSPC; 1880 break; 1881 } 1882 buf[bwritten++] = ','; 1883 buf[bwritten++] = ' '; 1884 bleft -= 2; 1885 } 1886 if (bleft < optlen) { 1887 *errorp = ENOSPC; 1888 break; 1889 } 1890 bcopy(optp->o_name, buf + bwritten, optlen); 1891 bwritten += optlen; 1892 bleft -= optlen; 1893 flags &= ~optp->o_opt; 1894 } 1895 1896 /* 1897 * Space already reserved for trailing \0 1898 */ 1899 buf[bwritten] = 0; 1900 return (bwritten); 1901 } 1902 1903 /* 1904 * Build hash lists of net addresses and hang them off the mount point. 1905 * Called by ufs_mount() to set up the lists of export addresses. 1906 */ 1907 static int 1908 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1909 const struct export_args *argp) 1910 { 1911 struct netcred *np; 1912 struct radix_node_head *rnh; 1913 int i; 1914 struct radix_node *rn; 1915 struct sockaddr *saddr, *smask = NULL; 1916 struct domain *dom; 1917 int error; 1918 1919 if (argp->ex_addrlen == 0) { 1920 if (mp->mnt_flag & MNT_DEFEXPORTED) 1921 return (EPERM); 1922 np = &nep->ne_defexported; 1923 np->netc_exflags = argp->ex_flags; 1924 np->netc_anon = argp->ex_anon; 1925 np->netc_anon.cr_ref = 1; 1926 mp->mnt_flag |= MNT_DEFEXPORTED; 1927 return (0); 1928 } 1929 1930 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1931 return (EINVAL); 1932 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1933 return (EINVAL); 1934 1935 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1936 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1937 saddr = (struct sockaddr *) (np + 1); 1938 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1939 goto out; 1940 if (saddr->sa_len > argp->ex_addrlen) 1941 saddr->sa_len = argp->ex_addrlen; 1942 if (argp->ex_masklen) { 1943 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1944 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1945 if (error) 1946 goto out; 1947 if (smask->sa_len > argp->ex_masklen) 1948 smask->sa_len = argp->ex_masklen; 1949 } 1950 i = saddr->sa_family; 1951 if ((rnh = nep->ne_rtable[i]) == NULL) { 1952 /* 1953 * Seems silly to initialize every AF when most are not used, 1954 * do so on demand here 1955 */ 1956 SLIST_FOREACH(dom, &domains, dom_next) 1957 if (dom->dom_family == i && dom->dom_rtattach) { 1958 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1959 dom->dom_rtoffset); 1960 break; 1961 } 1962 if ((rnh = nep->ne_rtable[i]) == NULL) { 1963 error = ENOBUFS; 1964 goto out; 1965 } 1966 } 1967 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1968 np->netc_rnodes); 1969 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */ 1970 error = EPERM; 1971 goto out; 1972 } 1973 np->netc_exflags = argp->ex_flags; 1974 np->netc_anon = argp->ex_anon; 1975 np->netc_anon.cr_ref = 1; 1976 return (0); 1977 out: 1978 kfree(np, M_NETADDR); 1979 return (error); 1980 } 1981 1982 /* ARGSUSED */ 1983 static int 1984 vfs_free_netcred(struct radix_node *rn, void *w) 1985 { 1986 struct radix_node_head *rnh = (struct radix_node_head *) w; 1987 1988 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1989 kfree((caddr_t) rn, M_NETADDR); 1990 return (0); 1991 } 1992 1993 /* 1994 * Free the net address hash lists that are hanging off the mount points. 1995 */ 1996 static void 1997 vfs_free_addrlist(struct netexport *nep) 1998 { 1999 int i; 2000 struct radix_node_head *rnh; 2001 2002 for (i = 0; i <= AF_MAX; i++) 2003 if ((rnh = nep->ne_rtable[i])) { 2004 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2005 (caddr_t) rnh); 2006 kfree((caddr_t) rnh, M_RTABLE); 2007 nep->ne_rtable[i] = 0; 2008 } 2009 } 2010 2011 int 2012 vfs_export(struct mount *mp, struct netexport *nep, 2013 const struct export_args *argp) 2014 { 2015 int error; 2016 2017 if (argp->ex_flags & MNT_DELEXPORT) { 2018 if (mp->mnt_flag & MNT_EXPUBLIC) { 2019 vfs_setpublicfs(NULL, NULL, NULL); 2020 mp->mnt_flag &= ~MNT_EXPUBLIC; 2021 } 2022 vfs_free_addrlist(nep); 2023 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2024 } 2025 if (argp->ex_flags & MNT_EXPORTED) { 2026 if (argp->ex_flags & MNT_EXPUBLIC) { 2027 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2028 return (error); 2029 mp->mnt_flag |= MNT_EXPUBLIC; 2030 } 2031 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2032 return (error); 2033 mp->mnt_flag |= MNT_EXPORTED; 2034 } 2035 return (0); 2036 } 2037 2038 2039 /* 2040 * Set the publicly exported filesystem (WebNFS). Currently, only 2041 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2042 */ 2043 int 2044 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2045 const struct export_args *argp) 2046 { 2047 int error; 2048 struct vnode *rvp; 2049 char *cp; 2050 2051 /* 2052 * mp == NULL -> invalidate the current info, the FS is 2053 * no longer exported. May be called from either vfs_export 2054 * or unmount, so check if it hasn't already been done. 2055 */ 2056 if (mp == NULL) { 2057 if (nfs_pub.np_valid) { 2058 nfs_pub.np_valid = 0; 2059 if (nfs_pub.np_index != NULL) { 2060 kfree(nfs_pub.np_index, M_TEMP); 2061 nfs_pub.np_index = NULL; 2062 } 2063 } 2064 return (0); 2065 } 2066 2067 /* 2068 * Only one allowed at a time. 2069 */ 2070 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2071 return (EBUSY); 2072 2073 /* 2074 * Get real filehandle for root of exported FS. 2075 */ 2076 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2077 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2078 2079 if ((error = VFS_ROOT(mp, &rvp))) 2080 return (error); 2081 2082 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2083 return (error); 2084 2085 vput(rvp); 2086 2087 /* 2088 * If an indexfile was specified, pull it in. 2089 */ 2090 if (argp->ex_indexfile != NULL) { 2091 int namelen; 2092 2093 error = vn_get_namelen(rvp, &namelen); 2094 if (error) 2095 return (error); 2096 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2097 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2098 namelen, NULL); 2099 if (!error) { 2100 /* 2101 * Check for illegal filenames. 2102 */ 2103 for (cp = nfs_pub.np_index; *cp; cp++) { 2104 if (*cp == '/') { 2105 error = EINVAL; 2106 break; 2107 } 2108 } 2109 } 2110 if (error) { 2111 kfree(nfs_pub.np_index, M_TEMP); 2112 return (error); 2113 } 2114 } 2115 2116 nfs_pub.np_mount = mp; 2117 nfs_pub.np_valid = 1; 2118 return (0); 2119 } 2120 2121 struct netcred * 2122 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2123 struct sockaddr *nam) 2124 { 2125 struct netcred *np; 2126 struct radix_node_head *rnh; 2127 struct sockaddr *saddr; 2128 2129 np = NULL; 2130 if (mp->mnt_flag & MNT_EXPORTED) { 2131 /* 2132 * Lookup in the export list first. 2133 */ 2134 if (nam != NULL) { 2135 saddr = nam; 2136 rnh = nep->ne_rtable[saddr->sa_family]; 2137 if (rnh != NULL) { 2138 np = (struct netcred *) 2139 (*rnh->rnh_matchaddr)((char *)saddr, 2140 rnh); 2141 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2142 np = NULL; 2143 } 2144 } 2145 /* 2146 * If no address match, use the default if it exists. 2147 */ 2148 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2149 np = &nep->ne_defexported; 2150 } 2151 return (np); 2152 } 2153 2154 /* 2155 * perform msync on all vnodes under a mount point. The mount point must 2156 * be locked. This code is also responsible for lazy-freeing unreferenced 2157 * vnodes whos VM objects no longer contain pages. 2158 * 2159 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2160 * 2161 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2162 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2163 * way up in this high level function. 2164 */ 2165 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2166 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2167 2168 void 2169 vfs_msync(struct mount *mp, int flags) 2170 { 2171 int vmsc_flags; 2172 2173 /* 2174 * tmpfs sets this flag to prevent msync(), sync, and the 2175 * filesystem periodic syncer from trying to flush VM pages 2176 * to swap. Only pure memory pressure flushes tmpfs VM pages 2177 * to swap. 2178 */ 2179 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2180 return; 2181 2182 /* 2183 * Ok, scan the vnodes for work. If the filesystem is using the 2184 * syncer thread feature we can use vsyncscan() instead of 2185 * vmntvnodescan(), which is much faster. 2186 */ 2187 vmsc_flags = VMSC_GETVP; 2188 if (flags != MNT_WAIT) 2189 vmsc_flags |= VMSC_NOWAIT; 2190 2191 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2192 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2193 (void *)(intptr_t)flags); 2194 } else { 2195 vmntvnodescan(mp, vmsc_flags, 2196 vfs_msync_scan1, vfs_msync_scan2, 2197 (void *)(intptr_t)flags); 2198 } 2199 } 2200 2201 /* 2202 * scan1 is a fast pre-check. There could be hundreds of thousands of 2203 * vnodes, we cannot afford to do anything heavy weight until we have a 2204 * fairly good indication that there is work to do. 2205 */ 2206 static 2207 int 2208 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2209 { 2210 int flags = (int)(intptr_t)data; 2211 2212 if ((vp->v_flag & VRECLAIMED) == 0) { 2213 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2214 vp->v_object) { 2215 return(0); /* call scan2 */ 2216 } 2217 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2218 (vp->v_flag & VOBJDIRTY) && 2219 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2220 return(0); /* call scan2 */ 2221 } 2222 } 2223 2224 /* 2225 * do not call scan2, continue the loop 2226 */ 2227 return(-1); 2228 } 2229 2230 /* 2231 * This callback is handed a locked vnode. 2232 */ 2233 static 2234 int 2235 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2236 { 2237 vm_object_t obj; 2238 int flags = (int)(intptr_t)data; 2239 2240 if (vp->v_flag & VRECLAIMED) 2241 return(0); 2242 2243 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2244 if ((obj = vp->v_object) != NULL) { 2245 vm_object_page_clean(obj, 0, 0, 2246 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2247 } 2248 } 2249 return(0); 2250 } 2251 2252 /* 2253 * Wake up anyone interested in vp because it is being revoked. 2254 */ 2255 void 2256 vn_gone(struct vnode *vp) 2257 { 2258 lwkt_gettoken(&vp->v_token); 2259 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2260 lwkt_reltoken(&vp->v_token); 2261 } 2262 2263 /* 2264 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2265 * (or v_rdev might be NULL). 2266 */ 2267 cdev_t 2268 vn_todev(struct vnode *vp) 2269 { 2270 if (vp->v_type != VBLK && vp->v_type != VCHR) 2271 return (NULL); 2272 KKASSERT(vp->v_rdev != NULL); 2273 return (vp->v_rdev); 2274 } 2275 2276 /* 2277 * Check if vnode represents a disk device. The vnode does not need to be 2278 * opened. 2279 * 2280 * MPALMOSTSAFE 2281 */ 2282 int 2283 vn_isdisk(struct vnode *vp, int *errp) 2284 { 2285 cdev_t dev; 2286 2287 if (vp->v_type != VCHR) { 2288 if (errp != NULL) 2289 *errp = ENOTBLK; 2290 return (0); 2291 } 2292 2293 dev = vp->v_rdev; 2294 2295 if (dev == NULL) { 2296 if (errp != NULL) 2297 *errp = ENXIO; 2298 return (0); 2299 } 2300 if (dev_is_good(dev) == 0) { 2301 if (errp != NULL) 2302 *errp = ENXIO; 2303 return (0); 2304 } 2305 if ((dev_dflags(dev) & D_DISK) == 0) { 2306 if (errp != NULL) 2307 *errp = ENOTBLK; 2308 return (0); 2309 } 2310 if (errp != NULL) 2311 *errp = 0; 2312 return (1); 2313 } 2314 2315 int 2316 vn_get_namelen(struct vnode *vp, int *namelen) 2317 { 2318 int error; 2319 register_t retval[2]; 2320 2321 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2322 if (error) 2323 return (error); 2324 *namelen = (int)retval[0]; 2325 return (0); 2326 } 2327 2328 int 2329 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2330 uint16_t d_namlen, const char *d_name) 2331 { 2332 struct dirent *dp; 2333 size_t len; 2334 2335 len = _DIRENT_RECLEN(d_namlen); 2336 if (len > uio->uio_resid) 2337 return(1); 2338 2339 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2340 2341 dp->d_ino = d_ino; 2342 dp->d_namlen = d_namlen; 2343 dp->d_type = d_type; 2344 bcopy(d_name, dp->d_name, d_namlen); 2345 2346 *error = uiomove((caddr_t)dp, len, uio); 2347 2348 kfree(dp, M_TEMP); 2349 2350 return(0); 2351 } 2352 2353 void 2354 vn_mark_atime(struct vnode *vp, struct thread *td) 2355 { 2356 struct proc *p = td->td_proc; 2357 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2358 2359 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2360 VOP_MARKATIME(vp, cred); 2361 } 2362 } 2363