1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 35 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 36 */ 37 38 /* 39 * External virtual filesystem routines 40 */ 41 #include "opt_ddb.h" 42 #include "opt_inet.h" 43 #include "opt_inet6.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/buf.h> 48 #include <sys/conf.h> 49 #include <sys/dirent.h> 50 #include <sys/eventhandler.h> 51 #include <sys/fcntl.h> 52 #include <sys/file.h> 53 #include <sys/kernel.h> 54 #include <sys/kthread.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/mount.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/reboot.h> 61 #include <sys/socket.h> 62 #include <sys/stat.h> 63 #include <sys/sysctl.h> 64 #include <sys/syslog.h> 65 #include <sys/unistd.h> 66 #include <sys/vmmeter.h> 67 #include <sys/vnode.h> 68 69 #include <machine/limits.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_extern.h> 74 #include <vm/vm_kern.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_pager.h> 79 #include <vm/vnode_pager.h> 80 #include <vm/vm_zone.h> 81 82 #include <sys/buf2.h> 83 #include <sys/thread2.h> 84 #include <sys/mplock2.h> 85 #include <vm/vm_page2.h> 86 87 #include <netinet/in.h> 88 89 static MALLOC_DEFINE(M_NETCRED, "Export Host", "Export host address structure"); 90 91 int numvnodes; 92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 93 "Number of vnodes allocated"); 94 int verbose_reclaims; 95 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0, 96 "Output filename of reclaimed vnode(s)"); 97 98 enum vtype iftovt_tab[16] = { 99 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 100 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 101 }; 102 int vttoif_tab[9] = { 103 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 104 S_IFSOCK, S_IFIFO, S_IFMT, 105 }; 106 107 static int reassignbufcalls; 108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 109 0, "Number of times buffers have been reassigned to the proper list"); 110 111 static int check_buf_overlap = 2; /* invasive check */ 112 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap, 113 0, "Enable overlapping buffer checks"); 114 115 int nfs_mount_type = -1; 116 static struct lwkt_token spechash_token; 117 struct nfs_public nfs_pub; /* publicly exported FS */ 118 119 int maxvnodes; 120 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 121 &maxvnodes, 0, "Maximum number of vnodes"); 122 123 static struct radix_node_head *vfs_create_addrlist_af(int af, 124 struct netexport *nep); 125 static void vfs_free_addrlist (struct netexport *nep); 126 static int vfs_free_netcred (struct radix_node *rn, void *w); 127 static void vfs_free_addrlist_af (struct radix_node_head **prnh); 128 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 129 const struct export_args *argp); 130 131 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 132 133 /* 134 * Red black tree functions 135 */ 136 static int rb_buf_compare(struct buf *b1, struct buf *b2); 137 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 138 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 139 140 static int 141 rb_buf_compare(struct buf *b1, struct buf *b2) 142 { 143 if (b1->b_loffset < b2->b_loffset) 144 return(-1); 145 if (b1->b_loffset > b2->b_loffset) 146 return(1); 147 return(0); 148 } 149 150 /* 151 * Initialize the vnode management data structures. 152 * 153 * Called from vfsinit() 154 */ 155 void 156 vfs_subr_init(void) 157 { 158 int factor1; 159 int factor2; 160 161 /* 162 * Desiredvnodes is kern.maxvnodes. We want to scale it 163 * according to available system memory but we may also have 164 * to limit it based on available KVM. 165 * 166 * WARNING! For machines with 64-256M of ram we have to be sure 167 * that the default limit scales down well due to HAMMER 168 * taking up significantly more memory per-vnode vs UFS. 169 * We want around ~5800 on a 128M machine. 170 * 171 * WARNING! Now that KVM is substantially larger (e.g. 8TB+), 172 * also limit maxvnodes based on a 128GB metric. This 173 * gives us something like ~3 millon vnodes. sysctl 174 * can be used to increase it further if desired. 175 * 176 * For disk cachhing purposes, filesystems like HAMMER1 177 * and HAMMER2 will or can be told to cache file data 178 * via the block device instead of excessively in vnodes. 179 */ 180 factor1 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode)); 181 factor2 = 30 * (sizeof(struct vm_object) + sizeof(struct vnode)); 182 maxvnodes = imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 183 KvaSize / factor2); 184 maxvnodes = imax(maxvnodes, maxproc * 8); 185 maxvnodes = imin(maxvnodes, 64LL*1024*1024*1024 / factor2); 186 187 lwkt_token_init(&spechash_token, "spechash"); 188 } 189 190 /* 191 * Knob to control the precision of file timestamps: 192 * 193 * 0 = seconds only; nanoseconds zeroed. 194 * 1 = seconds and nanoseconds, accurate within 1/HZ. 195 * 2 = seconds and nanoseconds, truncated to microseconds. 196 * >=3 = seconds and nanoseconds, maximum precision. 197 */ 198 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 199 200 static int timestamp_precision = TSP_SEC; 201 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 202 ×tamp_precision, 0, "Precision of file timestamps"); 203 204 /* 205 * Get a current timestamp. 206 * 207 * MPSAFE 208 */ 209 void 210 vfs_timestamp(struct timespec *tsp) 211 { 212 struct timeval tv; 213 214 switch (timestamp_precision) { 215 case TSP_SEC: 216 tsp->tv_sec = time_second; 217 tsp->tv_nsec = 0; 218 break; 219 case TSP_HZ: 220 getnanotime(tsp); 221 break; 222 case TSP_USEC: 223 microtime(&tv); 224 TIMEVAL_TO_TIMESPEC(&tv, tsp); 225 break; 226 case TSP_NSEC: 227 default: 228 nanotime(tsp); 229 break; 230 } 231 } 232 233 /* 234 * Set vnode attributes to VNOVAL 235 */ 236 void 237 vattr_null(struct vattr *vap) 238 { 239 vap->va_type = VNON; 240 vap->va_size = VNOVAL; 241 vap->va_bytes = VNOVAL; 242 vap->va_mode = VNOVAL; 243 vap->va_nlink = VNOVAL; 244 vap->va_uid = VNOVAL; 245 vap->va_gid = VNOVAL; 246 vap->va_fsid = VNOVAL; 247 vap->va_fileid = VNOVAL; 248 vap->va_blocksize = VNOVAL; 249 vap->va_rmajor = VNOVAL; 250 vap->va_rminor = VNOVAL; 251 vap->va_atime.tv_sec = VNOVAL; 252 vap->va_atime.tv_nsec = VNOVAL; 253 vap->va_mtime.tv_sec = VNOVAL; 254 vap->va_mtime.tv_nsec = VNOVAL; 255 vap->va_ctime.tv_sec = VNOVAL; 256 vap->va_ctime.tv_nsec = VNOVAL; 257 vap->va_flags = VNOVAL; 258 vap->va_gen = VNOVAL; 259 vap->va_vaflags = 0; 260 /* va_*_uuid fields are only valid if related flags are set */ 261 } 262 263 /* 264 * Flush out and invalidate all buffers associated with a vnode. 265 * 266 * vp must be locked. 267 */ 268 static int vinvalbuf_bp(struct buf *bp, void *data); 269 270 struct vinvalbuf_bp_info { 271 struct vnode *vp; 272 int slptimeo; 273 int lkflags; 274 int flags; 275 int clean; 276 }; 277 278 int 279 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 280 { 281 struct vinvalbuf_bp_info info; 282 vm_object_t object; 283 int error; 284 285 lwkt_gettoken(&vp->v_token); 286 287 /* 288 * If we are being asked to save, call fsync to ensure that the inode 289 * is updated. 290 */ 291 if (flags & V_SAVE) { 292 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 293 if (error) 294 goto done; 295 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 296 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 297 goto done; 298 #if 0 299 /* 300 * Dirty bufs may be left or generated via races 301 * in circumstances where vinvalbuf() is called on 302 * a vnode not undergoing reclamation. Only 303 * panic if we are trying to reclaim the vnode. 304 */ 305 if ((vp->v_flag & VRECLAIMED) && 306 (bio_track_active(&vp->v_track_write) || 307 !RB_EMPTY(&vp->v_rbdirty_tree))) { 308 panic("vinvalbuf: dirty bufs"); 309 } 310 #endif 311 } 312 } 313 info.slptimeo = slptimeo; 314 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 315 if (slpflag & PCATCH) 316 info.lkflags |= LK_PCATCH; 317 info.flags = flags; 318 info.vp = vp; 319 320 /* 321 * Flush the buffer cache until nothing is left, wait for all I/O 322 * to complete. At least one pass is required. We might block 323 * in the pip code so we have to re-check. Order is important. 324 */ 325 do { 326 /* 327 * Flush buffer cache 328 */ 329 if (!RB_EMPTY(&vp->v_rbclean_tree)) { 330 info.clean = 1; 331 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 332 NULL, vinvalbuf_bp, &info); 333 } 334 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 335 info.clean = 0; 336 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 337 NULL, vinvalbuf_bp, &info); 338 } 339 340 /* 341 * Wait for I/O completion. 342 */ 343 bio_track_wait(&vp->v_track_write, 0, 0); 344 if ((object = vp->v_object) != NULL) 345 refcount_wait(&object->paging_in_progress, "vnvlbx"); 346 } while (bio_track_active(&vp->v_track_write) || 347 !RB_EMPTY(&vp->v_rbclean_tree) || 348 !RB_EMPTY(&vp->v_rbdirty_tree)); 349 350 /* 351 * Destroy the copy in the VM cache, too. 352 */ 353 if ((object = vp->v_object) != NULL) { 354 vm_object_page_remove(object, 0, 0, 355 (flags & V_SAVE) ? TRUE : FALSE); 356 } 357 358 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 359 panic("vinvalbuf: flush failed"); 360 if (!RB_EMPTY(&vp->v_rbhash_tree)) 361 panic("vinvalbuf: flush failed, buffers still present"); 362 error = 0; 363 done: 364 lwkt_reltoken(&vp->v_token); 365 return (error); 366 } 367 368 static int 369 vinvalbuf_bp(struct buf *bp, void *data) 370 { 371 struct vinvalbuf_bp_info *info = data; 372 int error; 373 374 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 375 atomic_add_int(&bp->b_refs, 1); 376 error = BUF_TIMELOCK(bp, info->lkflags, 377 "vinvalbuf", info->slptimeo); 378 atomic_subtract_int(&bp->b_refs, 1); 379 if (error == 0) { 380 BUF_UNLOCK(bp); 381 error = ENOLCK; 382 } 383 if (error == ENOLCK) 384 return(0); 385 return (-error); 386 } 387 KKASSERT(bp->b_vp == info->vp); 388 389 /* 390 * Must check clean/dirty status after successfully locking as 391 * it may race. 392 */ 393 if ((info->clean && (bp->b_flags & B_DELWRI)) || 394 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 395 BUF_UNLOCK(bp); 396 return(0); 397 } 398 399 /* 400 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 401 * check. This code will write out the buffer, period. 402 */ 403 bremfree(bp); 404 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 405 (info->flags & V_SAVE)) { 406 cluster_awrite(bp); 407 } else if (info->flags & V_SAVE) { 408 /* 409 * Cannot set B_NOCACHE on a clean buffer as this will 410 * destroy the VM backing store which might actually 411 * be dirty (and unsynchronized). 412 */ 413 bp->b_flags |= (B_INVAL | B_RELBUF); 414 brelse(bp); 415 } else { 416 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 417 brelse(bp); 418 } 419 return(0); 420 } 421 422 /* 423 * Truncate a file's buffer and pages to a specified length. This 424 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 425 * sync activity. 426 * 427 * The vnode must be locked. 428 */ 429 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 430 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 431 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 432 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 433 434 struct vtruncbuf_info { 435 struct vnode *vp; 436 off_t truncloffset; 437 int clean; 438 }; 439 440 int 441 vtruncbuf(struct vnode *vp, off_t length, int blksize) 442 { 443 struct vtruncbuf_info info; 444 const char *filename; 445 int count; 446 447 /* 448 * Round up to the *next* block, then destroy the buffers in question. 449 * Since we are only removing some of the buffers we must rely on the 450 * scan count to determine whether a loop is necessary. 451 */ 452 if ((count = (int)(length % blksize)) != 0) 453 info.truncloffset = length + (blksize - count); 454 else 455 info.truncloffset = length; 456 info.vp = vp; 457 458 lwkt_gettoken(&vp->v_token); 459 do { 460 info.clean = 1; 461 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 462 vtruncbuf_bp_trunc_cmp, 463 vtruncbuf_bp_trunc, &info); 464 info.clean = 0; 465 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 466 vtruncbuf_bp_trunc_cmp, 467 vtruncbuf_bp_trunc, &info); 468 } while(count); 469 470 /* 471 * For safety, fsync any remaining metadata if the file is not being 472 * truncated to 0. Since the metadata does not represent the entire 473 * dirty list we have to rely on the hit count to ensure that we get 474 * all of it. 475 */ 476 if (length > 0) { 477 do { 478 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 479 vtruncbuf_bp_metasync_cmp, 480 vtruncbuf_bp_metasync, &info); 481 } while (count); 482 } 483 484 /* 485 * Clean out any left over VM backing store. 486 * 487 * It is possible to have in-progress I/O from buffers that were 488 * not part of the truncation. This should not happen if we 489 * are truncating to 0-length. 490 */ 491 vnode_pager_setsize(vp, length); 492 bio_track_wait(&vp->v_track_write, 0, 0); 493 494 /* 495 * Debugging only 496 */ 497 spin_lock(&vp->v_spin); 498 filename = TAILQ_FIRST(&vp->v_namecache) ? 499 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 500 spin_unlock(&vp->v_spin); 501 502 /* 503 * Make sure no buffers were instantiated while we were trying 504 * to clean out the remaining VM pages. This could occur due 505 * to busy dirty VM pages being flushed out to disk. 506 */ 507 do { 508 info.clean = 1; 509 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 510 vtruncbuf_bp_trunc_cmp, 511 vtruncbuf_bp_trunc, &info); 512 info.clean = 0; 513 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 514 vtruncbuf_bp_trunc_cmp, 515 vtruncbuf_bp_trunc, &info); 516 if (count) { 517 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 518 "left over buffers in %s\n", count, filename); 519 } 520 } while(count); 521 522 lwkt_reltoken(&vp->v_token); 523 524 return (0); 525 } 526 527 /* 528 * The callback buffer is beyond the new file EOF and must be destroyed. 529 * Note that the compare function must conform to the RB_SCAN's requirements. 530 */ 531 static 532 int 533 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 534 { 535 struct vtruncbuf_info *info = data; 536 537 if (bp->b_loffset >= info->truncloffset) 538 return(0); 539 return(-1); 540 } 541 542 static 543 int 544 vtruncbuf_bp_trunc(struct buf *bp, void *data) 545 { 546 struct vtruncbuf_info *info = data; 547 548 /* 549 * Do not try to use a buffer we cannot immediately lock, but sleep 550 * anyway to prevent a livelock. The code will loop until all buffers 551 * can be acted upon. 552 * 553 * We must always revalidate the buffer after locking it to deal 554 * with MP races. 555 */ 556 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 557 atomic_add_int(&bp->b_refs, 1); 558 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 559 BUF_UNLOCK(bp); 560 atomic_subtract_int(&bp->b_refs, 1); 561 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 562 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 563 bp->b_vp != info->vp || 564 vtruncbuf_bp_trunc_cmp(bp, data)) { 565 BUF_UNLOCK(bp); 566 } else { 567 bremfree(bp); 568 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 569 brelse(bp); 570 } 571 return(1); 572 } 573 574 /* 575 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 576 * blocks (with a negative loffset) are scanned. 577 * Note that the compare function must conform to the RB_SCAN's requirements. 578 */ 579 static int 580 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 581 { 582 if (bp->b_loffset < 0) 583 return(0); 584 return(1); 585 } 586 587 static int 588 vtruncbuf_bp_metasync(struct buf *bp, void *data) 589 { 590 struct vtruncbuf_info *info = data; 591 592 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 593 atomic_add_int(&bp->b_refs, 1); 594 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 595 BUF_UNLOCK(bp); 596 atomic_subtract_int(&bp->b_refs, 1); 597 } else if ((bp->b_flags & B_DELWRI) == 0 || 598 bp->b_vp != info->vp || 599 vtruncbuf_bp_metasync_cmp(bp, data)) { 600 BUF_UNLOCK(bp); 601 } else { 602 bremfree(bp); 603 if (bp->b_vp == info->vp) 604 bawrite(bp); 605 else 606 bwrite(bp); 607 } 608 return(1); 609 } 610 611 /* 612 * vfsync - implements a multipass fsync on a file which understands 613 * dependancies and meta-data. The passed vnode must be locked. The 614 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 615 * 616 * When fsyncing data asynchronously just do one consolidated pass starting 617 * with the most negative block number. This may not get all the data due 618 * to dependancies. 619 * 620 * When fsyncing data synchronously do a data pass, then a metadata pass, 621 * then do additional data+metadata passes to try to get all the data out. 622 * 623 * Caller must ref the vnode but does not have to lock it. 624 */ 625 static int vfsync_wait_output(struct vnode *vp, 626 int (*waitoutput)(struct vnode *, struct thread *)); 627 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 628 static int vfsync_data_only_cmp(struct buf *bp, void *data); 629 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 630 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 631 static int vfsync_bp(struct buf *bp, void *data); 632 633 struct vfsync_info { 634 struct vnode *vp; 635 int fastpass; 636 int synchronous; 637 int syncdeps; 638 int lazycount; 639 int lazylimit; 640 int skippedbufs; 641 int (*checkdef)(struct buf *); 642 int (*cmpfunc)(struct buf *, void *); 643 }; 644 645 int 646 vfsync(struct vnode *vp, int waitfor, int passes, 647 int (*checkdef)(struct buf *), 648 int (*waitoutput)(struct vnode *, struct thread *)) 649 { 650 struct vfsync_info info; 651 int error; 652 653 bzero(&info, sizeof(info)); 654 info.vp = vp; 655 if ((info.checkdef = checkdef) == NULL) 656 info.syncdeps = 1; 657 658 lwkt_gettoken(&vp->v_token); 659 660 switch(waitfor) { 661 case MNT_LAZY | MNT_NOWAIT: 662 case MNT_LAZY: 663 /* 664 * Lazy (filesystem syncer typ) Asynchronous plus limit the 665 * number of data (not meta) pages we try to flush to 1MB. 666 * A non-zero return means that lazy limit was reached. 667 */ 668 info.lazylimit = 1024 * 1024; 669 info.syncdeps = 1; 670 info.cmpfunc = vfsync_lazy_range_cmp; 671 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 672 vfsync_lazy_range_cmp, vfsync_bp, &info); 673 info.cmpfunc = vfsync_meta_only_cmp; 674 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 675 vfsync_meta_only_cmp, vfsync_bp, &info); 676 if (error == 0) 677 vp->v_lazyw = 0; 678 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 679 vn_syncer_add(vp, 1); 680 error = 0; 681 break; 682 case MNT_NOWAIT: 683 /* 684 * Asynchronous. Do a data-only pass and a meta-only pass. 685 */ 686 info.syncdeps = 1; 687 info.cmpfunc = vfsync_data_only_cmp; 688 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 689 vfsync_bp, &info); 690 info.cmpfunc = vfsync_meta_only_cmp; 691 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 692 vfsync_bp, &info); 693 error = 0; 694 break; 695 default: 696 /* 697 * Synchronous. Do a data-only pass, then a meta-data+data 698 * pass, then additional integrated passes to try to get 699 * all the dependancies flushed. 700 */ 701 info.cmpfunc = vfsync_data_only_cmp; 702 info.fastpass = 1; 703 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 704 vfsync_bp, &info); 705 info.fastpass = 0; 706 error = vfsync_wait_output(vp, waitoutput); 707 if (error == 0) { 708 info.skippedbufs = 0; 709 info.cmpfunc = vfsync_dummy_cmp; 710 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 711 vfsync_bp, &info); 712 error = vfsync_wait_output(vp, waitoutput); 713 if (info.skippedbufs) { 714 kprintf("Warning: vfsync skipped %d dirty " 715 "buf%s in pass2!\n", 716 info.skippedbufs, 717 ((info.skippedbufs > 1) ? "s" : "")); 718 } 719 } 720 while (error == 0 && passes > 0 && 721 !RB_EMPTY(&vp->v_rbdirty_tree) 722 ) { 723 info.skippedbufs = 0; 724 if (--passes == 0) { 725 info.synchronous = 1; 726 info.syncdeps = 1; 727 } 728 info.cmpfunc = vfsync_dummy_cmp; 729 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 730 vfsync_bp, &info); 731 if (error < 0) 732 error = -error; 733 info.syncdeps = 1; 734 if (error == 0) 735 error = vfsync_wait_output(vp, waitoutput); 736 if (info.skippedbufs && passes == 0) { 737 kprintf("Warning: vfsync skipped %d dirty " 738 "buf%s in final pass!\n", 739 info.skippedbufs, 740 ((info.skippedbufs > 1) ? "s" : "")); 741 } 742 } 743 #if 0 744 /* 745 * This case can occur normally because vnode lock might 746 * not be held. 747 */ 748 if (!RB_EMPTY(&vp->v_rbdirty_tree)) 749 kprintf("dirty bufs left after final pass\n"); 750 #endif 751 break; 752 } 753 lwkt_reltoken(&vp->v_token); 754 755 return(error); 756 } 757 758 static int 759 vfsync_wait_output(struct vnode *vp, 760 int (*waitoutput)(struct vnode *, struct thread *)) 761 { 762 int error; 763 764 error = bio_track_wait(&vp->v_track_write, 0, 0); 765 if (waitoutput) 766 error = waitoutput(vp, curthread); 767 return(error); 768 } 769 770 static int 771 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 772 { 773 return(0); 774 } 775 776 static int 777 vfsync_data_only_cmp(struct buf *bp, void *data) 778 { 779 if (bp->b_loffset < 0) 780 return(-1); 781 return(0); 782 } 783 784 static int 785 vfsync_meta_only_cmp(struct buf *bp, void *data) 786 { 787 if (bp->b_loffset < 0) 788 return(0); 789 return(1); 790 } 791 792 static int 793 vfsync_lazy_range_cmp(struct buf *bp, void *data) 794 { 795 struct vfsync_info *info = data; 796 797 if (bp->b_loffset < info->vp->v_lazyw) 798 return(-1); 799 return(0); 800 } 801 802 static int 803 vfsync_bp(struct buf *bp, void *data) 804 { 805 struct vfsync_info *info = data; 806 struct vnode *vp = info->vp; 807 int error; 808 809 if (info->fastpass) { 810 /* 811 * Ignore buffers that we cannot immediately lock. 812 */ 813 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 814 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst1", 1)) { 815 ++info->skippedbufs; 816 return(0); 817 } 818 } 819 } else if (info->synchronous == 0) { 820 /* 821 * Normal pass, give the buffer a little time to become 822 * available to us. 823 */ 824 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst2", hz / 10)) { 825 ++info->skippedbufs; 826 return(0); 827 } 828 } else { 829 /* 830 * Synchronous pass, give the buffer a lot of time before 831 * giving up. 832 */ 833 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst3", hz * 10)) { 834 ++info->skippedbufs; 835 return(0); 836 } 837 } 838 839 /* 840 * We must revalidate the buffer after locking. 841 */ 842 if ((bp->b_flags & B_DELWRI) == 0 || 843 bp->b_vp != info->vp || 844 info->cmpfunc(bp, data)) { 845 BUF_UNLOCK(bp); 846 return(0); 847 } 848 849 /* 850 * If syncdeps is not set we do not try to write buffers which have 851 * dependancies. 852 */ 853 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 854 BUF_UNLOCK(bp); 855 return(0); 856 } 857 858 /* 859 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 860 * has been written but an additional handshake with the device 861 * is required before we can dispose of the buffer. We have no idea 862 * how to do this so we have to skip these buffers. 863 */ 864 if (bp->b_flags & B_NEEDCOMMIT) { 865 BUF_UNLOCK(bp); 866 return(0); 867 } 868 869 /* 870 * Ask bioops if it is ok to sync. If not the VFS may have 871 * set B_LOCKED so we have to cycle the buffer. 872 */ 873 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 874 bremfree(bp); 875 brelse(bp); 876 return(0); 877 } 878 879 if (info->synchronous) { 880 /* 881 * Synchronous flush. An error may be returned and will 882 * stop the scan. 883 */ 884 bremfree(bp); 885 error = bwrite(bp); 886 } else { 887 /* 888 * Asynchronous flush. We use the error return to support 889 * MNT_LAZY flushes. 890 * 891 * In low-memory situations we revert to synchronous 892 * operation. This should theoretically prevent the I/O 893 * path from exhausting memory in a non-recoverable way. 894 */ 895 vp->v_lazyw = bp->b_loffset; 896 bremfree(bp); 897 if (vm_page_count_min(0)) { 898 /* low memory */ 899 info->lazycount += bp->b_bufsize; 900 bwrite(bp); 901 } else { 902 /* normal */ 903 info->lazycount += cluster_awrite(bp); 904 waitrunningbufspace(); 905 /*vm_wait_nominal();*/ 906 } 907 if (info->lazylimit && info->lazycount >= info->lazylimit) 908 error = 1; 909 else 910 error = 0; 911 } 912 return(-error); 913 } 914 915 /* 916 * Associate a buffer with a vnode. 917 * 918 * MPSAFE 919 */ 920 int 921 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 922 { 923 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 924 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 925 926 /* 927 * Insert onto list for new vnode. 928 */ 929 lwkt_gettoken(&vp->v_token); 930 931 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 932 lwkt_reltoken(&vp->v_token); 933 return (EEXIST); 934 } 935 936 /* 937 * Diagnostics (mainly for HAMMER debugging). Check for 938 * overlapping buffers. 939 */ 940 if (check_buf_overlap) { 941 struct buf *bx; 942 bx = buf_rb_hash_RB_PREV(bp); 943 if (bx) { 944 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 945 kprintf("bgetvp: overlapl %016jx/%d %016jx " 946 "bx %p bp %p\n", 947 (intmax_t)bx->b_loffset, 948 bx->b_bufsize, 949 (intmax_t)bp->b_loffset, 950 bx, bp); 951 if (check_buf_overlap > 1) 952 panic("bgetvp - overlapping buffer"); 953 } 954 } 955 bx = buf_rb_hash_RB_NEXT(bp); 956 if (bx) { 957 if (bp->b_loffset + testsize > bx->b_loffset) { 958 kprintf("bgetvp: overlapr %016jx/%d %016jx " 959 "bp %p bx %p\n", 960 (intmax_t)bp->b_loffset, 961 testsize, 962 (intmax_t)bx->b_loffset, 963 bp, bx); 964 if (check_buf_overlap > 1) 965 panic("bgetvp - overlapping buffer"); 966 } 967 } 968 } 969 bp->b_vp = vp; 970 bp->b_flags |= B_HASHED; 971 bp->b_flags |= B_VNCLEAN; 972 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 973 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 974 /*vhold(vp);*/ 975 lwkt_reltoken(&vp->v_token); 976 return(0); 977 } 978 979 /* 980 * Disassociate a buffer from a vnode. 981 * 982 * MPSAFE 983 */ 984 void 985 brelvp(struct buf *bp) 986 { 987 struct vnode *vp; 988 989 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 990 991 /* 992 * Delete from old vnode list, if on one. 993 */ 994 vp = bp->b_vp; 995 lwkt_gettoken(&vp->v_token); 996 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 997 if (bp->b_flags & B_VNDIRTY) 998 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 999 else 1000 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1001 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 1002 } 1003 if (bp->b_flags & B_HASHED) { 1004 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 1005 bp->b_flags &= ~B_HASHED; 1006 } 1007 1008 /* 1009 * Only remove from synclist when no dirty buffers are left AND 1010 * the VFS has not flagged the vnode's inode as being dirty. 1011 */ 1012 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 1013 RB_EMPTY(&vp->v_rbdirty_tree)) { 1014 vn_syncer_remove(vp, 0); 1015 } 1016 bp->b_vp = NULL; 1017 1018 lwkt_reltoken(&vp->v_token); 1019 1020 /*vdrop(vp);*/ 1021 } 1022 1023 /* 1024 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 1025 * This routine is called when the state of the B_DELWRI bit is changed. 1026 * 1027 * Must be called with vp->v_token held. 1028 * MPSAFE 1029 */ 1030 void 1031 reassignbuf(struct buf *bp) 1032 { 1033 struct vnode *vp = bp->b_vp; 1034 int delay; 1035 1036 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 1037 ++reassignbufcalls; 1038 1039 /* 1040 * B_PAGING flagged buffers cannot be reassigned because their vp 1041 * is not fully linked in. 1042 */ 1043 if (bp->b_flags & B_PAGING) 1044 panic("cannot reassign paging buffer"); 1045 1046 if (bp->b_flags & B_DELWRI) { 1047 /* 1048 * Move to the dirty list, add the vnode to the worklist 1049 */ 1050 if (bp->b_flags & B_VNCLEAN) { 1051 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1052 bp->b_flags &= ~B_VNCLEAN; 1053 } 1054 if ((bp->b_flags & B_VNDIRTY) == 0) { 1055 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 1056 panic("reassignbuf: dup lblk vp %p bp %p", 1057 vp, bp); 1058 } 1059 bp->b_flags |= B_VNDIRTY; 1060 } 1061 if ((vp->v_flag & VONWORKLST) == 0) { 1062 switch (vp->v_type) { 1063 case VDIR: 1064 delay = dirdelay; 1065 break; 1066 case VCHR: 1067 case VBLK: 1068 if (vp->v_rdev && 1069 vp->v_rdev->si_mountpoint != NULL) { 1070 delay = metadelay; 1071 break; 1072 } 1073 /* fall through */ 1074 default: 1075 delay = filedelay; 1076 } 1077 vn_syncer_add(vp, delay); 1078 } 1079 } else { 1080 /* 1081 * Move to the clean list, remove the vnode from the worklist 1082 * if no dirty blocks remain. 1083 */ 1084 if (bp->b_flags & B_VNDIRTY) { 1085 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1086 bp->b_flags &= ~B_VNDIRTY; 1087 } 1088 if ((bp->b_flags & B_VNCLEAN) == 0) { 1089 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1090 panic("reassignbuf: dup lblk vp %p bp %p", 1091 vp, bp); 1092 } 1093 bp->b_flags |= B_VNCLEAN; 1094 } 1095 1096 /* 1097 * Only remove from synclist when no dirty buffers are left 1098 * AND the VFS has not flagged the vnode's inode as being 1099 * dirty. 1100 */ 1101 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1102 VONWORKLST && 1103 RB_EMPTY(&vp->v_rbdirty_tree)) { 1104 vn_syncer_remove(vp, 0); 1105 } 1106 } 1107 } 1108 1109 /* 1110 * Create a vnode for a block device. Used for mounting the root file 1111 * system. 1112 * 1113 * A vref()'d vnode is returned. 1114 */ 1115 extern struct vop_ops *devfs_vnode_dev_vops_p; 1116 int 1117 bdevvp(cdev_t dev, struct vnode **vpp) 1118 { 1119 struct vnode *vp; 1120 struct vnode *nvp; 1121 int error; 1122 1123 if (dev == NULL) { 1124 *vpp = NULLVP; 1125 return (ENXIO); 1126 } 1127 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1128 &nvp, 0, 0); 1129 if (error) { 1130 *vpp = NULLVP; 1131 return (error); 1132 } 1133 vp = nvp; 1134 vp->v_type = VCHR; 1135 #if 0 1136 vp->v_rdev = dev; 1137 #endif 1138 v_associate_rdev(vp, dev); 1139 vp->v_umajor = dev->si_umajor; 1140 vp->v_uminor = dev->si_uminor; 1141 vx_unlock(vp); 1142 *vpp = vp; 1143 return (0); 1144 } 1145 1146 int 1147 v_associate_rdev(struct vnode *vp, cdev_t dev) 1148 { 1149 if (dev == NULL) 1150 return(ENXIO); 1151 if (dev_is_good(dev) == 0) 1152 return(ENXIO); 1153 KKASSERT(vp->v_rdev == NULL); 1154 vp->v_rdev = reference_dev(dev); 1155 lwkt_gettoken(&spechash_token); 1156 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1157 lwkt_reltoken(&spechash_token); 1158 return(0); 1159 } 1160 1161 void 1162 v_release_rdev(struct vnode *vp) 1163 { 1164 cdev_t dev; 1165 1166 if ((dev = vp->v_rdev) != NULL) { 1167 lwkt_gettoken(&spechash_token); 1168 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1169 vp->v_rdev = NULL; 1170 release_dev(dev); 1171 lwkt_reltoken(&spechash_token); 1172 } 1173 } 1174 1175 /* 1176 * Add a vnode to the alias list hung off the cdev_t. We only associate 1177 * the device number with the vnode. The actual device is not associated 1178 * until the vnode is opened (usually in spec_open()), and will be 1179 * disassociated on last close. 1180 */ 1181 void 1182 addaliasu(struct vnode *nvp, int x, int y) 1183 { 1184 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1185 panic("addaliasu on non-special vnode"); 1186 nvp->v_umajor = x; 1187 nvp->v_uminor = y; 1188 } 1189 1190 /* 1191 * Simple call that a filesystem can make to try to get rid of a 1192 * vnode. It will fail if anyone is referencing the vnode (including 1193 * the caller). 1194 * 1195 * The filesystem can check whether its in-memory inode structure still 1196 * references the vp on return. 1197 * 1198 * May only be called if the vnode is in a known state (i.e. being prevented 1199 * from being deallocated by some other condition such as a vfs inode hold). 1200 */ 1201 void 1202 vclean_unlocked(struct vnode *vp) 1203 { 1204 vx_get(vp); 1205 if (VREFCNT(vp) <= 1) 1206 vgone_vxlocked(vp); 1207 vx_put(vp); 1208 } 1209 1210 /* 1211 * Disassociate a vnode from its underlying filesystem. 1212 * 1213 * The vnode must be VX locked and referenced. In all normal situations 1214 * there are no active references. If vclean_vxlocked() is called while 1215 * there are active references, the vnode is being ripped out and we have 1216 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1217 */ 1218 void 1219 vclean_vxlocked(struct vnode *vp, int flags) 1220 { 1221 int active; 1222 int n; 1223 vm_object_t object; 1224 struct namecache *ncp; 1225 1226 /* 1227 * If the vnode has already been reclaimed we have nothing to do. 1228 */ 1229 if (vp->v_flag & VRECLAIMED) 1230 return; 1231 1232 /* 1233 * Set flag to interlock operation, flag finalization to ensure 1234 * that the vnode winds up on the inactive list, and set v_act to 0. 1235 */ 1236 vsetflags(vp, VRECLAIMED); 1237 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1238 vp->v_act = 0; 1239 1240 if (verbose_reclaims) { 1241 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1242 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1243 } 1244 1245 /* 1246 * Scrap the vfs cache 1247 */ 1248 while (cache_inval_vp(vp, 0) != 0) { 1249 kprintf("Warning: vnode %p clean/cache_resolution " 1250 "race detected\n", vp); 1251 tsleep(vp, 0, "vclninv", 2); 1252 } 1253 1254 /* 1255 * Check to see if the vnode is in use. If so we have to reference it 1256 * before we clean it out so that its count cannot fall to zero and 1257 * generate a race against ourselves to recycle it. 1258 */ 1259 active = (VREFCNT(vp) > 0); 1260 1261 /* 1262 * Clean out any buffers associated with the vnode and destroy its 1263 * object, if it has one. 1264 */ 1265 vinvalbuf(vp, V_SAVE, 0, 0); 1266 1267 /* 1268 * If purging an active vnode (typically during a forced unmount 1269 * or reboot), it must be closed and deactivated before being 1270 * reclaimed. This isn't really all that safe, but what can 1271 * we do? XXX. 1272 * 1273 * Note that neither of these routines unlocks the vnode. 1274 */ 1275 if (active && (flags & DOCLOSE)) { 1276 while ((n = vp->v_opencount) != 0) { 1277 if (vp->v_writecount) 1278 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1279 else 1280 VOP_CLOSE(vp, FNONBLOCK, NULL); 1281 if (vp->v_opencount == n) { 1282 kprintf("Warning: unable to force-close" 1283 " vnode %p\n", vp); 1284 break; 1285 } 1286 } 1287 } 1288 1289 /* 1290 * If the vnode has not been deactivated, deactivated it. Deactivation 1291 * can create new buffers and VM pages so we have to call vinvalbuf() 1292 * again to make sure they all get flushed. 1293 * 1294 * This can occur if a file with a link count of 0 needs to be 1295 * truncated. 1296 * 1297 * If the vnode is already dead don't try to deactivate it. 1298 */ 1299 if ((vp->v_flag & VINACTIVE) == 0) { 1300 vsetflags(vp, VINACTIVE); 1301 if (vp->v_mount) 1302 VOP_INACTIVE(vp); 1303 vinvalbuf(vp, V_SAVE, 0, 0); 1304 } 1305 1306 /* 1307 * If the vnode has an object, destroy it. 1308 */ 1309 while ((object = vp->v_object) != NULL) { 1310 vm_object_hold(object); 1311 if (object == vp->v_object) 1312 break; 1313 vm_object_drop(object); 1314 } 1315 1316 if (object != NULL) { 1317 if (object->ref_count == 0) { 1318 if ((object->flags & OBJ_DEAD) == 0) 1319 vm_object_terminate(object); 1320 vm_object_drop(object); 1321 vclrflags(vp, VOBJBUF); 1322 } else { 1323 vm_pager_deallocate(object); 1324 vclrflags(vp, VOBJBUF); 1325 vm_object_drop(object); 1326 } 1327 } 1328 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1329 1330 if (vp->v_flag & VOBJDIRTY) 1331 vclrobjdirty(vp); 1332 1333 /* 1334 * Reclaim the vnode if not already dead. 1335 */ 1336 if (vp->v_mount && VOP_RECLAIM(vp)) 1337 panic("vclean: cannot reclaim"); 1338 1339 /* 1340 * Done with purge, notify sleepers of the grim news. 1341 */ 1342 vp->v_ops = &dead_vnode_vops_p; 1343 vn_gone(vp); 1344 vp->v_tag = VT_NON; 1345 1346 /* 1347 * If we are destroying an active vnode, reactivate it now that 1348 * we have reassociated it with deadfs. This prevents the system 1349 * from crashing on the vnode due to it being unexpectedly marked 1350 * as inactive or reclaimed. 1351 */ 1352 if (active && (flags & DOCLOSE)) { 1353 vclrflags(vp, VINACTIVE | VRECLAIMED); 1354 } 1355 } 1356 1357 /* 1358 * Eliminate all activity associated with the requested vnode 1359 * and with all vnodes aliased to the requested vnode. 1360 * 1361 * The vnode must be referenced but should not be locked. 1362 */ 1363 int 1364 vrevoke(struct vnode *vp, struct ucred *cred) 1365 { 1366 struct vnode *vq; 1367 struct vnode *vqn; 1368 cdev_t dev; 1369 int error; 1370 1371 /* 1372 * If the vnode has a device association, scrap all vnodes associated 1373 * with the device. Don't let the device disappear on us while we 1374 * are scrapping the vnodes. 1375 * 1376 * The passed vp will probably show up in the list, do not VX lock 1377 * it twice! 1378 * 1379 * Releasing the vnode's rdev here can mess up specfs's call to 1380 * device close, so don't do it. The vnode has been disassociated 1381 * and the device will be closed after the last ref on the related 1382 * fp goes away (if not still open by e.g. the kernel). 1383 */ 1384 if (vp->v_type != VCHR) { 1385 error = fdrevoke(vp, DTYPE_VNODE, cred); 1386 return (error); 1387 } 1388 if ((dev = vp->v_rdev) == NULL) { 1389 return(0); 1390 } 1391 reference_dev(dev); 1392 lwkt_gettoken(&spechash_token); 1393 1394 restart: 1395 vqn = SLIST_FIRST(&dev->si_hlist); 1396 if (vqn) 1397 vhold(vqn); 1398 while ((vq = vqn) != NULL) { 1399 if (VREFCNT(vq) > 0) { 1400 vref(vq); 1401 fdrevoke(vq, DTYPE_VNODE, cred); 1402 /*v_release_rdev(vq);*/ 1403 vrele(vq); 1404 if (vq->v_rdev != dev) { 1405 vdrop(vq); 1406 goto restart; 1407 } 1408 } 1409 vqn = SLIST_NEXT(vq, v_cdevnext); 1410 if (vqn) 1411 vhold(vqn); 1412 vdrop(vq); 1413 } 1414 lwkt_reltoken(&spechash_token); 1415 dev_drevoke(dev); 1416 release_dev(dev); 1417 return (0); 1418 } 1419 1420 /* 1421 * This is called when the object underlying a vnode is being destroyed, 1422 * such as in a remove(). Try to recycle the vnode immediately if the 1423 * only active reference is our reference. 1424 * 1425 * Directory vnodes in the namecache with children cannot be immediately 1426 * recycled because numerous VOP_N*() ops require them to be stable. 1427 * 1428 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1429 * function is a NOP if VRECLAIMED is already set. 1430 */ 1431 int 1432 vrecycle(struct vnode *vp) 1433 { 1434 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1435 if (cache_inval_vp_nonblock(vp)) 1436 return(0); 1437 vgone_vxlocked(vp); 1438 return (1); 1439 } 1440 return (0); 1441 } 1442 1443 /* 1444 * Return the maximum I/O size allowed for strategy calls on VP. 1445 * 1446 * If vp is VCHR or VBLK we dive the device, otherwise we use 1447 * the vp's mount info. 1448 * 1449 * The returned value is clamped at MAXPHYS as most callers cannot use 1450 * buffers larger than that size. 1451 */ 1452 int 1453 vmaxiosize(struct vnode *vp) 1454 { 1455 int maxiosize; 1456 1457 if (vp->v_type == VBLK || vp->v_type == VCHR) 1458 maxiosize = vp->v_rdev->si_iosize_max; 1459 else 1460 maxiosize = vp->v_mount->mnt_iosize_max; 1461 1462 if (maxiosize > MAXPHYS) 1463 maxiosize = MAXPHYS; 1464 return (maxiosize); 1465 } 1466 1467 /* 1468 * Eliminate all activity associated with a vnode in preparation for 1469 * destruction. 1470 * 1471 * The vnode must be VX locked and refd and will remain VX locked and refd 1472 * on return. This routine may be called with the vnode in any state, as 1473 * long as it is VX locked. The vnode will be cleaned out and marked 1474 * VRECLAIMED but will not actually be reused until all existing refs and 1475 * holds go away. 1476 * 1477 * NOTE: This routine may be called on a vnode which has not yet been 1478 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1479 * already been reclaimed. 1480 * 1481 * This routine is not responsible for placing us back on the freelist. 1482 * Instead, it happens automatically when the caller releases the VX lock 1483 * (assuming there aren't any other references). 1484 */ 1485 void 1486 vgone_vxlocked(struct vnode *vp) 1487 { 1488 /* 1489 * assert that the VX lock is held. This is an absolute requirement 1490 * now for vgone_vxlocked() to be called. 1491 */ 1492 KKASSERT(lockinuse(&vp->v_lock)); 1493 1494 /* 1495 * Clean out the filesystem specific data and set the VRECLAIMED 1496 * bit. Also deactivate the vnode if necessary. 1497 * 1498 * The vnode should have automatically been removed from the syncer 1499 * list as syncer/dirty flags cleared during the cleaning. 1500 */ 1501 vclean_vxlocked(vp, DOCLOSE); 1502 1503 /* 1504 * Normally panic if the vnode is still dirty, unless we are doing 1505 * a forced unmount (tmpfs typically). 1506 */ 1507 if (vp->v_flag & VONWORKLST) { 1508 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) { 1509 /* force removal */ 1510 vn_syncer_remove(vp, 1); 1511 } else { 1512 panic("vp %p still dirty in vgone after flush", vp); 1513 } 1514 } 1515 1516 /* 1517 * Delete from old mount point vnode list, if on one. 1518 */ 1519 if (vp->v_mount != NULL) { 1520 KKASSERT(vp->v_data == NULL); 1521 insmntque(vp, NULL); 1522 } 1523 1524 /* 1525 * If special device, remove it from special device alias list 1526 * if it is on one. This should normally only occur if a vnode is 1527 * being revoked as the device should otherwise have been released 1528 * naturally. 1529 */ 1530 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1531 v_release_rdev(vp); 1532 } 1533 1534 /* 1535 * Set us to VBAD 1536 */ 1537 vp->v_type = VBAD; 1538 } 1539 1540 /* 1541 * Lookup a vnode by device number. 1542 * 1543 * Returns non-zero and *vpp set to a vref'd vnode on success. 1544 * Returns zero on failure. 1545 */ 1546 int 1547 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1548 { 1549 struct vnode *vp; 1550 1551 lwkt_gettoken(&spechash_token); 1552 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1553 if (type == vp->v_type) { 1554 *vpp = vp; 1555 vref(vp); 1556 lwkt_reltoken(&spechash_token); 1557 return (1); 1558 } 1559 } 1560 lwkt_reltoken(&spechash_token); 1561 return (0); 1562 } 1563 1564 /* 1565 * Calculate the total number of references to a special device. This 1566 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1567 * an overloaded field. Since udev2dev can now return NULL, we have 1568 * to check for a NULL v_rdev. 1569 */ 1570 int 1571 count_dev(cdev_t dev) 1572 { 1573 struct vnode *vp; 1574 int count = 0; 1575 1576 if (SLIST_FIRST(&dev->si_hlist)) { 1577 lwkt_gettoken(&spechash_token); 1578 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1579 count += vp->v_opencount; 1580 } 1581 lwkt_reltoken(&spechash_token); 1582 } 1583 return(count); 1584 } 1585 1586 int 1587 vcount(struct vnode *vp) 1588 { 1589 if (vp->v_rdev == NULL) 1590 return(0); 1591 return(count_dev(vp->v_rdev)); 1592 } 1593 1594 /* 1595 * Initialize VMIO for a vnode. This routine MUST be called before a 1596 * VFS can issue buffer cache ops on a vnode. It is typically called 1597 * when a vnode is initialized from its inode. 1598 */ 1599 int 1600 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1601 { 1602 vm_object_t object; 1603 int error = 0; 1604 1605 object = vp->v_object; 1606 if (object) { 1607 vm_object_hold(object); 1608 KKASSERT(vp->v_object == object); 1609 } 1610 1611 if (object == NULL) { 1612 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1613 1614 /* 1615 * Dereference the reference we just created. This assumes 1616 * that the object is associated with the vp. Allow it to 1617 * have zero refs. It cannot be destroyed as long as it 1618 * is associated with the vnode. 1619 */ 1620 vm_object_hold(object); 1621 atomic_add_int(&object->ref_count, -1); 1622 vrele(vp); 1623 } else { 1624 KKASSERT((object->flags & OBJ_DEAD) == 0); 1625 } 1626 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1627 vsetflags(vp, VOBJBUF); 1628 vm_object_drop(object); 1629 1630 return (error); 1631 } 1632 1633 1634 /* 1635 * Print out a description of a vnode. 1636 */ 1637 static char *typename[] = 1638 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1639 1640 void 1641 vprint(char *label, struct vnode *vp) 1642 { 1643 char buf[96]; 1644 1645 if (label != NULL) 1646 kprintf("%s: %p: ", label, (void *)vp); 1647 else 1648 kprintf("%p: ", (void *)vp); 1649 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1650 typename[vp->v_type], 1651 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1652 buf[0] = '\0'; 1653 if (vp->v_flag & VROOT) 1654 strcat(buf, "|VROOT"); 1655 if (vp->v_flag & VPFSROOT) 1656 strcat(buf, "|VPFSROOT"); 1657 if (vp->v_flag & VTEXT) 1658 strcat(buf, "|VTEXT"); 1659 if (vp->v_flag & VSYSTEM) 1660 strcat(buf, "|VSYSTEM"); 1661 if (vp->v_flag & VOBJBUF) 1662 strcat(buf, "|VOBJBUF"); 1663 if (buf[0] != '\0') 1664 kprintf(" flags (%s)", &buf[1]); 1665 if (vp->v_data == NULL) { 1666 kprintf("\n"); 1667 } else { 1668 kprintf("\n\t"); 1669 VOP_PRINT(vp); 1670 } 1671 } 1672 1673 /* 1674 * Do the usual access checking. 1675 * file_mode, uid and gid are from the vnode in question, 1676 * while acc_mode and cred are from the VOP_ACCESS parameter list 1677 */ 1678 int 1679 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1680 mode_t acc_mode, struct ucred *cred) 1681 { 1682 mode_t mask; 1683 int ismember; 1684 1685 /* 1686 * Super-user always gets read/write access, but execute access depends 1687 * on at least one execute bit being set. 1688 */ 1689 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1690 if ((acc_mode & VEXEC) && type != VDIR && 1691 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1692 return (EACCES); 1693 return (0); 1694 } 1695 1696 mask = 0; 1697 1698 /* Otherwise, check the owner. */ 1699 if (cred->cr_uid == uid) { 1700 if (acc_mode & VEXEC) 1701 mask |= S_IXUSR; 1702 if (acc_mode & VREAD) 1703 mask |= S_IRUSR; 1704 if (acc_mode & VWRITE) 1705 mask |= S_IWUSR; 1706 return ((file_mode & mask) == mask ? 0 : EACCES); 1707 } 1708 1709 /* Otherwise, check the groups. */ 1710 ismember = groupmember(gid, cred); 1711 if (cred->cr_svgid == gid || ismember) { 1712 if (acc_mode & VEXEC) 1713 mask |= S_IXGRP; 1714 if (acc_mode & VREAD) 1715 mask |= S_IRGRP; 1716 if (acc_mode & VWRITE) 1717 mask |= S_IWGRP; 1718 return ((file_mode & mask) == mask ? 0 : EACCES); 1719 } 1720 1721 /* Otherwise, check everyone else. */ 1722 if (acc_mode & VEXEC) 1723 mask |= S_IXOTH; 1724 if (acc_mode & VREAD) 1725 mask |= S_IROTH; 1726 if (acc_mode & VWRITE) 1727 mask |= S_IWOTH; 1728 return ((file_mode & mask) == mask ? 0 : EACCES); 1729 } 1730 1731 #ifdef DDB 1732 #include <ddb/ddb.h> 1733 1734 static int db_show_locked_vnodes(struct mount *mp, void *data); 1735 1736 /* 1737 * List all of the locked vnodes in the system. 1738 * Called when debugging the kernel. 1739 */ 1740 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1741 { 1742 kprintf("Locked vnodes\n"); 1743 mountlist_scan(db_show_locked_vnodes, NULL, 1744 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1745 } 1746 1747 static int 1748 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1749 { 1750 struct vnode *vp; 1751 1752 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1753 if (vn_islocked(vp)) 1754 vprint(NULL, vp); 1755 } 1756 return(0); 1757 } 1758 #endif 1759 1760 /* 1761 * Top level filesystem related information gathering. 1762 */ 1763 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1764 1765 static int 1766 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1767 { 1768 int *name = (int *)arg1 - 1; /* XXX */ 1769 u_int namelen = arg2 + 1; /* XXX */ 1770 struct vfsconf *vfsp; 1771 int maxtypenum; 1772 1773 #if 1 || defined(COMPAT_PRELITE2) 1774 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1775 if (namelen == 1) 1776 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1777 #endif 1778 1779 #ifdef notyet 1780 /* all sysctl names at this level are at least name and field */ 1781 if (namelen < 2) 1782 return (ENOTDIR); /* overloaded */ 1783 if (name[0] != VFS_GENERIC) { 1784 vfsp = vfsconf_find_by_typenum(name[0]); 1785 if (vfsp == NULL) 1786 return (EOPNOTSUPP); 1787 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1788 oldp, oldlenp, newp, newlen, p)); 1789 } 1790 #endif 1791 switch (name[1]) { 1792 case VFS_MAXTYPENUM: 1793 if (namelen != 2) 1794 return (ENOTDIR); 1795 maxtypenum = vfsconf_get_maxtypenum(); 1796 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1797 case VFS_CONF: 1798 if (namelen != 3) 1799 return (ENOTDIR); /* overloaded */ 1800 vfsp = vfsconf_find_by_typenum(name[2]); 1801 if (vfsp == NULL) 1802 return (EOPNOTSUPP); 1803 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1804 } 1805 return (EOPNOTSUPP); 1806 } 1807 1808 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1809 "Generic filesystem"); 1810 1811 #if 1 || defined(COMPAT_PRELITE2) 1812 1813 static int 1814 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1815 { 1816 int error; 1817 struct ovfsconf ovfs; 1818 struct sysctl_req *req = (struct sysctl_req*) data; 1819 1820 bzero(&ovfs, sizeof(ovfs)); 1821 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1822 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1823 ovfs.vfc_index = vfsp->vfc_typenum; 1824 ovfs.vfc_refcount = vfsp->vfc_refcount; 1825 ovfs.vfc_flags = vfsp->vfc_flags; 1826 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1827 if (error) 1828 return error; /* abort iteration with error code */ 1829 else 1830 return 0; /* continue iterating with next element */ 1831 } 1832 1833 static int 1834 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1835 { 1836 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1837 } 1838 1839 #endif /* 1 || COMPAT_PRELITE2 */ 1840 1841 /* 1842 * Check to see if a filesystem is mounted on a block device. 1843 */ 1844 int 1845 vfs_mountedon(struct vnode *vp) 1846 { 1847 cdev_t dev; 1848 1849 if ((dev = vp->v_rdev) == NULL) { 1850 /* if (vp->v_type != VBLK) 1851 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1852 } 1853 if (dev != NULL && dev->si_mountpoint) 1854 return (EBUSY); 1855 return (0); 1856 } 1857 1858 /* 1859 * Unmount all filesystems. The list is traversed in reverse order 1860 * of mounting to avoid dependencies. 1861 * 1862 * We want the umountall to be able to break out of its loop if a 1863 * failure occurs, after scanning all possible mounts, so the callback 1864 * returns 0 on error. 1865 * 1866 * NOTE: Do not call mountlist_remove(mp) on error any more, this will 1867 * confuse mountlist_scan()'s unbusy check. 1868 */ 1869 static int vfs_umountall_callback(struct mount *mp, void *data); 1870 1871 void 1872 vfs_unmountall(int halting) 1873 { 1874 int count; 1875 1876 do { 1877 count = mountlist_scan(vfs_umountall_callback, &halting, 1878 MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1879 } while (count); 1880 } 1881 1882 static 1883 int 1884 vfs_umountall_callback(struct mount *mp, void *data) 1885 { 1886 int error; 1887 int halting = *(int *)data; 1888 1889 /* 1890 * NOTE: When halting, dounmount will disconnect but leave 1891 * certain mount points intact. e.g. devfs. 1892 */ 1893 error = dounmount(mp, MNT_FORCE, halting); 1894 if (error) { 1895 kprintf("unmount of filesystem mounted from %s failed (", 1896 mp->mnt_stat.f_mntfromname); 1897 if (error == EBUSY) 1898 kprintf("BUSY)\n"); 1899 else 1900 kprintf("%d)\n", error); 1901 return 0; 1902 } else { 1903 return 1; 1904 } 1905 } 1906 1907 /* 1908 * Checks the mount flags for parameter mp and put the names comma-separated 1909 * into a string buffer buf with a size limit specified by len. 1910 * 1911 * It returns the number of bytes written into buf, and (*errorp) will be 1912 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1913 * not large enough). The buffer will be 0-terminated if len was not 0. 1914 */ 1915 size_t 1916 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1917 char *buf, size_t len, int *errorp) 1918 { 1919 static const struct mountctl_opt optnames[] = { 1920 { MNT_RDONLY, "read-only" }, 1921 { MNT_SYNCHRONOUS, "synchronous" }, 1922 { MNT_NOEXEC, "noexec" }, 1923 { MNT_NOSUID, "nosuid" }, 1924 { MNT_NODEV, "nodev" }, 1925 { MNT_AUTOMOUNTED, "automounted" }, 1926 { MNT_ASYNC, "asynchronous" }, 1927 { MNT_SUIDDIR, "suiddir" }, 1928 { MNT_SOFTDEP, "soft-updates" }, 1929 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1930 { MNT_TRIM, "trim" }, 1931 { MNT_NOATIME, "noatime" }, 1932 { MNT_NOCLUSTERR, "noclusterr" }, 1933 { MNT_NOCLUSTERW, "noclusterw" }, 1934 { MNT_EXRDONLY, "NFS read-only" }, 1935 { MNT_EXPORTED, "NFS exported" }, 1936 /* Remaining NFS flags could come here */ 1937 { MNT_LOCAL, "local" }, 1938 { MNT_QUOTA, "with-quotas" }, 1939 /* { MNT_ROOTFS, "rootfs" }, */ 1940 /* { MNT_IGNORE, "ignore" }, */ 1941 { 0, NULL} 1942 }; 1943 int bwritten; 1944 int bleft; 1945 int optlen; 1946 int actsize; 1947 1948 *errorp = 0; 1949 bwritten = 0; 1950 bleft = len - 1; /* leave room for trailing \0 */ 1951 1952 /* 1953 * Checks the size of the string. If it contains 1954 * any data, then we will append the new flags to 1955 * it. 1956 */ 1957 actsize = strlen(buf); 1958 if (actsize > 0) 1959 buf += actsize; 1960 1961 /* Default flags if no flags passed */ 1962 if (optp == NULL) 1963 optp = optnames; 1964 1965 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1966 *errorp = EINVAL; 1967 return(0); 1968 } 1969 1970 for (; flags && optp->o_opt; ++optp) { 1971 if ((flags & optp->o_opt) == 0) 1972 continue; 1973 optlen = strlen(optp->o_name); 1974 if (bwritten || actsize > 0) { 1975 if (bleft < 2) { 1976 *errorp = ENOSPC; 1977 break; 1978 } 1979 buf[bwritten++] = ','; 1980 buf[bwritten++] = ' '; 1981 bleft -= 2; 1982 } 1983 if (bleft < optlen) { 1984 *errorp = ENOSPC; 1985 break; 1986 } 1987 bcopy(optp->o_name, buf + bwritten, optlen); 1988 bwritten += optlen; 1989 bleft -= optlen; 1990 flags &= ~optp->o_opt; 1991 } 1992 1993 /* 1994 * Space already reserved for trailing \0 1995 */ 1996 buf[bwritten] = 0; 1997 return (bwritten); 1998 } 1999 2000 /* 2001 * Build hash lists of net addresses and hang them off the mount point. 2002 * Called by ufs_mount() to set up the lists of export addresses. 2003 */ 2004 static int 2005 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 2006 const struct export_args *argp) 2007 { 2008 struct netcred *np; 2009 struct radix_node_head *rnh; 2010 int i; 2011 struct radix_node *rn; 2012 struct sockaddr *saddr, *smask = NULL; 2013 int error; 2014 2015 if (argp->ex_addrlen == 0) { 2016 if (mp->mnt_flag & MNT_DEFEXPORTED) 2017 return (EPERM); 2018 np = &nep->ne_defexported; 2019 np->netc_exflags = argp->ex_flags; 2020 np->netc_anon = argp->ex_anon; 2021 np->netc_anon.cr_ref = 1; 2022 mp->mnt_flag |= MNT_DEFEXPORTED; 2023 return (0); 2024 } 2025 2026 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 2027 return (EINVAL); 2028 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 2029 return (EINVAL); 2030 2031 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2032 np = (struct netcred *)kmalloc(i, M_NETCRED, M_WAITOK | M_ZERO); 2033 saddr = (struct sockaddr *) (np + 1); 2034 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2035 goto out; 2036 if (saddr->sa_len > argp->ex_addrlen) 2037 saddr->sa_len = argp->ex_addrlen; 2038 if (argp->ex_masklen) { 2039 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2040 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2041 if (error) 2042 goto out; 2043 if (smask->sa_len > argp->ex_masklen) 2044 smask->sa_len = argp->ex_masklen; 2045 } 2046 NE_LOCK(nep); 2047 if (nep->ne_maskhead == NULL) { 2048 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) { 2049 error = ENOBUFS; 2050 goto out; 2051 } 2052 } 2053 if ((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) { 2054 error = ENOBUFS; 2055 goto out; 2056 } 2057 rn = (*rnh->rnh_addaddr)((char *)saddr, (char *)smask, rnh, 2058 np->netc_rnodes); 2059 NE_UNLOCK(nep); 2060 if (rn == NULL || np != (struct netcred *)rn) { /* already exists */ 2061 error = EPERM; 2062 goto out; 2063 } 2064 np->netc_exflags = argp->ex_flags; 2065 np->netc_anon = argp->ex_anon; 2066 np->netc_anon.cr_ref = 1; 2067 return (0); 2068 2069 out: 2070 kfree(np, M_NETCRED); 2071 return (error); 2072 } 2073 2074 /* 2075 * Free netcred structures installed in the netexport 2076 */ 2077 static int 2078 vfs_free_netcred(struct radix_node *rn, void *w) 2079 { 2080 struct radix_node_head *rnh = (struct radix_node_head *)w; 2081 2082 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2083 kfree(rn, M_NETCRED); 2084 2085 return (0); 2086 } 2087 2088 /* 2089 * callback to free an element of the mask table installed in the 2090 * netexport. These may be created indirectly and are not netcred 2091 * structures. 2092 */ 2093 static int 2094 vfs_free_netcred_mask(struct radix_node *rn, void *w) 2095 { 2096 struct radix_node_head *rnh = (struct radix_node_head *)w; 2097 2098 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2099 kfree(rn, M_RTABLE); 2100 2101 return (0); 2102 } 2103 2104 static struct radix_node_head * 2105 vfs_create_addrlist_af(int af, struct netexport *nep) 2106 { 2107 struct radix_node_head *rnh = NULL; 2108 #if defined(INET) || defined(INET6) 2109 struct radix_node_head *maskhead = nep->ne_maskhead; 2110 int off; 2111 #endif 2112 2113 NE_ASSERT_LOCKED(nep); 2114 KKASSERT(maskhead != NULL); 2115 switch (af) { 2116 #ifdef INET 2117 case AF_INET: 2118 if ((rnh = nep->ne_inethead) == NULL) { 2119 off = offsetof(struct sockaddr_in, sin_addr) << 3; 2120 if (!rn_inithead((void **)&rnh, maskhead, off)) 2121 return (NULL); 2122 nep->ne_inethead = rnh; 2123 } 2124 break; 2125 #endif 2126 #ifdef INET6 2127 case AF_INET6: 2128 if ((rnh = nep->ne_inet6head) == NULL) { 2129 off = offsetof(struct sockaddr_in6, sin6_addr) << 3; 2130 if (!rn_inithead((void **)&rnh, maskhead, off)) 2131 return (NULL); 2132 nep->ne_inet6head = rnh; 2133 } 2134 break; 2135 #endif 2136 } 2137 return (rnh); 2138 } 2139 2140 /* 2141 * helper function for freeing netcred elements 2142 */ 2143 static void 2144 vfs_free_addrlist_af(struct radix_node_head **prnh) 2145 { 2146 struct radix_node_head *rnh = *prnh; 2147 2148 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh); 2149 kfree(rnh, M_RTABLE); 2150 *prnh = NULL; 2151 } 2152 2153 /* 2154 * helper function for freeing mask elements 2155 */ 2156 static void 2157 vfs_free_addrlist_masks(struct radix_node_head **prnh) 2158 { 2159 struct radix_node_head *rnh = *prnh; 2160 2161 (*rnh->rnh_walktree) (rnh, vfs_free_netcred_mask, rnh); 2162 kfree(rnh, M_RTABLE); 2163 *prnh = NULL; 2164 } 2165 2166 /* 2167 * Free the net address hash lists that are hanging off the mount points. 2168 */ 2169 static void 2170 vfs_free_addrlist(struct netexport *nep) 2171 { 2172 NE_LOCK(nep); 2173 if (nep->ne_inethead != NULL) 2174 vfs_free_addrlist_af(&nep->ne_inethead); 2175 if (nep->ne_inet6head != NULL) 2176 vfs_free_addrlist_af(&nep->ne_inet6head); 2177 if (nep->ne_maskhead) 2178 vfs_free_addrlist_masks(&nep->ne_maskhead); 2179 NE_UNLOCK(nep); 2180 } 2181 2182 int 2183 vfs_export(struct mount *mp, struct netexport *nep, 2184 const struct export_args *argp) 2185 { 2186 int error; 2187 2188 if (argp->ex_flags & MNT_DELEXPORT) { 2189 if (mp->mnt_flag & MNT_EXPUBLIC) { 2190 vfs_setpublicfs(NULL, NULL, NULL); 2191 mp->mnt_flag &= ~MNT_EXPUBLIC; 2192 } 2193 vfs_free_addrlist(nep); 2194 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2195 } 2196 if (argp->ex_flags & MNT_EXPORTED) { 2197 if (argp->ex_flags & MNT_EXPUBLIC) { 2198 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2199 return (error); 2200 mp->mnt_flag |= MNT_EXPUBLIC; 2201 } 2202 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2203 return (error); 2204 mp->mnt_flag |= MNT_EXPORTED; 2205 } 2206 return (0); 2207 } 2208 2209 2210 /* 2211 * Set the publicly exported filesystem (WebNFS). Currently, only 2212 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2213 */ 2214 int 2215 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2216 const struct export_args *argp) 2217 { 2218 int error; 2219 struct vnode *rvp; 2220 char *cp; 2221 2222 /* 2223 * mp == NULL -> invalidate the current info, the FS is 2224 * no longer exported. May be called from either vfs_export 2225 * or unmount, so check if it hasn't already been done. 2226 */ 2227 if (mp == NULL) { 2228 if (nfs_pub.np_valid) { 2229 nfs_pub.np_valid = 0; 2230 if (nfs_pub.np_index != NULL) { 2231 kfree(nfs_pub.np_index, M_TEMP); 2232 nfs_pub.np_index = NULL; 2233 } 2234 } 2235 return (0); 2236 } 2237 2238 /* 2239 * Only one allowed at a time. 2240 */ 2241 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2242 return (EBUSY); 2243 2244 /* 2245 * Get real filehandle for root of exported FS. 2246 */ 2247 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2248 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2249 2250 if ((error = VFS_ROOT(mp, &rvp))) 2251 return (error); 2252 2253 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2254 return (error); 2255 2256 vput(rvp); 2257 2258 /* 2259 * If an indexfile was specified, pull it in. 2260 */ 2261 if (argp->ex_indexfile != NULL) { 2262 int namelen; 2263 2264 error = vn_get_namelen(rvp, &namelen); 2265 if (error) 2266 return (error); 2267 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2268 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2269 namelen, NULL); 2270 if (!error) { 2271 /* 2272 * Check for illegal filenames. 2273 */ 2274 for (cp = nfs_pub.np_index; *cp; cp++) { 2275 if (*cp == '/') { 2276 error = EINVAL; 2277 break; 2278 } 2279 } 2280 } 2281 if (error) { 2282 kfree(nfs_pub.np_index, M_TEMP); 2283 return (error); 2284 } 2285 } 2286 2287 nfs_pub.np_mount = mp; 2288 nfs_pub.np_valid = 1; 2289 return (0); 2290 } 2291 2292 struct netcred * 2293 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2294 struct sockaddr *nam) 2295 { 2296 struct netcred *np; 2297 struct radix_node_head *rnh; 2298 struct sockaddr *saddr; 2299 2300 np = NULL; 2301 if (mp->mnt_flag & MNT_EXPORTED) { 2302 /* 2303 * Lookup in the export list first. 2304 */ 2305 NE_LOCK(nep); 2306 if (nam != NULL) { 2307 saddr = nam; 2308 switch (saddr->sa_family) { 2309 #ifdef INET 2310 case AF_INET: 2311 rnh = nep->ne_inethead; 2312 break; 2313 #endif 2314 #ifdef INET6 2315 case AF_INET6: 2316 rnh = nep->ne_inet6head; 2317 break; 2318 #endif 2319 default: 2320 rnh = NULL; 2321 } 2322 if (rnh != NULL) { 2323 np = (struct netcred *) 2324 (*rnh->rnh_matchaddr)((char *)saddr, 2325 rnh); 2326 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2327 np = NULL; 2328 } 2329 } 2330 NE_UNLOCK(nep); 2331 /* 2332 * If no address match, use the default if it exists. 2333 */ 2334 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2335 np = &nep->ne_defexported; 2336 } 2337 return (np); 2338 } 2339 2340 /* 2341 * perform msync on all vnodes under a mount point. The mount point must 2342 * be locked. This code is also responsible for lazy-freeing unreferenced 2343 * vnodes whos VM objects no longer contain pages. 2344 * 2345 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2346 * 2347 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2348 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2349 * way up in this high level function. 2350 */ 2351 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2352 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2353 2354 void 2355 vfs_msync(struct mount *mp, int flags) 2356 { 2357 int vmsc_flags; 2358 2359 /* 2360 * tmpfs sets this flag to prevent msync(), sync, and the 2361 * filesystem periodic syncer from trying to flush VM pages 2362 * to swap. Only pure memory pressure flushes tmpfs VM pages 2363 * to swap. 2364 */ 2365 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2366 return; 2367 2368 /* 2369 * Ok, scan the vnodes for work. If the filesystem is using the 2370 * syncer thread feature we can use vsyncscan() instead of 2371 * vmntvnodescan(), which is much faster. 2372 */ 2373 vmsc_flags = VMSC_GETVP; 2374 if (flags != MNT_WAIT) 2375 vmsc_flags |= VMSC_NOWAIT; 2376 2377 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2378 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2379 (void *)(intptr_t)flags); 2380 } else { 2381 vmntvnodescan(mp, vmsc_flags, 2382 vfs_msync_scan1, vfs_msync_scan2, 2383 (void *)(intptr_t)flags); 2384 } 2385 } 2386 2387 /* 2388 * scan1 is a fast pre-check. There could be hundreds of thousands of 2389 * vnodes, we cannot afford to do anything heavy weight until we have a 2390 * fairly good indication that there is work to do. 2391 */ 2392 static 2393 int 2394 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2395 { 2396 int flags = (int)(intptr_t)data; 2397 2398 if ((vp->v_flag & VRECLAIMED) == 0) { 2399 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2400 vp->v_object) { 2401 return(0); /* call scan2 */ 2402 } 2403 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2404 (vp->v_flag & VOBJDIRTY) && 2405 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2406 return(0); /* call scan2 */ 2407 } 2408 } 2409 2410 /* 2411 * do not call scan2, continue the loop 2412 */ 2413 return(-1); 2414 } 2415 2416 /* 2417 * This callback is handed a locked vnode. 2418 */ 2419 static 2420 int 2421 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2422 { 2423 vm_object_t obj; 2424 int flags = (int)(intptr_t)data; 2425 2426 if (vp->v_flag & VRECLAIMED) 2427 return(0); 2428 2429 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2430 if ((obj = vp->v_object) != NULL) { 2431 vm_object_page_clean(obj, 0, 0, 2432 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2433 } 2434 } 2435 return(0); 2436 } 2437 2438 /* 2439 * Wake up anyone interested in vp because it is being revoked. 2440 */ 2441 void 2442 vn_gone(struct vnode *vp) 2443 { 2444 lwkt_gettoken(&vp->v_token); 2445 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2446 lwkt_reltoken(&vp->v_token); 2447 } 2448 2449 /* 2450 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2451 * (or v_rdev might be NULL). 2452 */ 2453 cdev_t 2454 vn_todev(struct vnode *vp) 2455 { 2456 if (vp->v_type != VBLK && vp->v_type != VCHR) 2457 return (NULL); 2458 KKASSERT(vp->v_rdev != NULL); 2459 return (vp->v_rdev); 2460 } 2461 2462 /* 2463 * Check if vnode represents a disk device. The vnode does not need to be 2464 * opened. 2465 * 2466 * MPALMOSTSAFE 2467 */ 2468 int 2469 vn_isdisk(struct vnode *vp, int *errp) 2470 { 2471 cdev_t dev; 2472 2473 if (vp->v_type != VCHR) { 2474 if (errp != NULL) 2475 *errp = ENOTBLK; 2476 return (0); 2477 } 2478 2479 dev = vp->v_rdev; 2480 2481 if (dev == NULL) { 2482 if (errp != NULL) 2483 *errp = ENXIO; 2484 return (0); 2485 } 2486 if (dev_is_good(dev) == 0) { 2487 if (errp != NULL) 2488 *errp = ENXIO; 2489 return (0); 2490 } 2491 if ((dev_dflags(dev) & D_DISK) == 0) { 2492 if (errp != NULL) 2493 *errp = ENOTBLK; 2494 return (0); 2495 } 2496 if (errp != NULL) 2497 *errp = 0; 2498 return (1); 2499 } 2500 2501 int 2502 vn_get_namelen(struct vnode *vp, int *namelen) 2503 { 2504 int error; 2505 register_t retval[2]; 2506 2507 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2508 if (error) 2509 return (error); 2510 *namelen = (int)retval[0]; 2511 return (0); 2512 } 2513 2514 int 2515 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2516 uint16_t d_namlen, const char *d_name) 2517 { 2518 struct dirent *dp; 2519 size_t len; 2520 2521 len = _DIRENT_RECLEN(d_namlen); 2522 if (len > uio->uio_resid) 2523 return(1); 2524 2525 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2526 2527 dp->d_ino = d_ino; 2528 dp->d_namlen = d_namlen; 2529 dp->d_type = d_type; 2530 bcopy(d_name, dp->d_name, d_namlen); 2531 2532 *error = uiomove((caddr_t)dp, len, uio); 2533 2534 kfree(dp, M_TEMP); 2535 2536 return(0); 2537 } 2538 2539 void 2540 vn_mark_atime(struct vnode *vp, struct thread *td) 2541 { 2542 struct proc *p = td->td_proc; 2543 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2544 2545 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2546 VOP_MARKATIME(vp, cred); 2547 } 2548 } 2549 2550 /* 2551 * Calculate the number of entries in an inode-related chained hash table. 2552 * With today's memory sizes, maxvnodes can wind up being a very large 2553 * number. There is no reason to waste memory, so tolerate some stacking. 2554 */ 2555 int 2556 vfs_inodehashsize(void) 2557 { 2558 int hsize; 2559 2560 hsize = 32; 2561 while (hsize < maxvnodes) 2562 hsize <<= 1; 2563 while (hsize > maxvnodes * 2) 2564 hsize >>= 1; /* nominal 2x stacking */ 2565 2566 if (maxvnodes > 1024 * 1024) 2567 hsize >>= 1; /* nominal 8x stacking */ 2568 2569 if (maxvnodes > 128 * 1024) 2570 hsize >>= 1; /* nominal 4x stacking */ 2571 2572 if (hsize < 16) 2573 hsize = 16; 2574 2575 return hsize; 2576 } 2577