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 /* 815 * Removed BUF_TIMELOCK(..., 1), even a 1-tick 816 * delay can mess up performance 817 * 818 * Another reason is that during a dirty-buffer 819 * scan a clustered write can start I/O on buffers 820 * ahead of the scan, causing the scan to not 821 * get a lock here. Usually this means the write 822 * is already in progress so, in fact, we *want* 823 * to skip the buffer. 824 */ 825 ++info->skippedbufs; 826 return(0); 827 } 828 } else if (info->synchronous == 0) { 829 /* 830 * Normal pass, give the buffer a little time to become 831 * available to us. 832 */ 833 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst2", hz / 10)) { 834 ++info->skippedbufs; 835 return(0); 836 } 837 } else { 838 /* 839 * Synchronous pass, give the buffer a lot of time before 840 * giving up. 841 */ 842 if (BUF_TIMELOCK(bp, LK_EXCLUSIVE, "bflst3", hz * 10)) { 843 ++info->skippedbufs; 844 return(0); 845 } 846 } 847 848 /* 849 * We must revalidate the buffer after locking. 850 */ 851 if ((bp->b_flags & B_DELWRI) == 0 || 852 bp->b_vp != info->vp || 853 info->cmpfunc(bp, data)) { 854 BUF_UNLOCK(bp); 855 return(0); 856 } 857 858 /* 859 * If syncdeps is not set we do not try to write buffers which have 860 * dependancies. 861 */ 862 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 863 BUF_UNLOCK(bp); 864 return(0); 865 } 866 867 /* 868 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 869 * has been written but an additional handshake with the device 870 * is required before we can dispose of the buffer. We have no idea 871 * how to do this so we have to skip these buffers. 872 */ 873 if (bp->b_flags & B_NEEDCOMMIT) { 874 BUF_UNLOCK(bp); 875 return(0); 876 } 877 878 /* 879 * Ask bioops if it is ok to sync. If not the VFS may have 880 * set B_LOCKED so we have to cycle the buffer. 881 */ 882 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 883 bremfree(bp); 884 brelse(bp); 885 return(0); 886 } 887 888 if (info->synchronous) { 889 /* 890 * Synchronous flush. An error may be returned and will 891 * stop the scan. 892 */ 893 bremfree(bp); 894 error = bwrite(bp); 895 } else { 896 /* 897 * Asynchronous flush. We use the error return to support 898 * MNT_LAZY flushes. 899 * 900 * In low-memory situations we revert to synchronous 901 * operation. This should theoretically prevent the I/O 902 * path from exhausting memory in a non-recoverable way. 903 */ 904 vp->v_lazyw = bp->b_loffset; 905 bremfree(bp); 906 if (vm_page_count_min(0)) { 907 /* low memory */ 908 info->lazycount += bp->b_bufsize; 909 bwrite(bp); 910 } else { 911 /* normal */ 912 info->lazycount += cluster_awrite(bp); 913 waitrunningbufspace(); 914 /*vm_wait_nominal();*/ 915 } 916 if (info->lazylimit && info->lazycount >= info->lazylimit) 917 error = 1; 918 else 919 error = 0; 920 } 921 return(-error); 922 } 923 924 /* 925 * Associate a buffer with a vnode. 926 * 927 * MPSAFE 928 */ 929 int 930 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 931 { 932 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 933 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 934 935 /* 936 * Insert onto list for new vnode. 937 */ 938 lwkt_gettoken(&vp->v_token); 939 940 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 941 lwkt_reltoken(&vp->v_token); 942 return (EEXIST); 943 } 944 945 /* 946 * Diagnostics (mainly for HAMMER debugging). Check for 947 * overlapping buffers. 948 */ 949 if (check_buf_overlap) { 950 struct buf *bx; 951 bx = buf_rb_hash_RB_PREV(bp); 952 if (bx) { 953 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 954 kprintf("bgetvp: overlapl %016jx/%d %016jx " 955 "bx %p bp %p\n", 956 (intmax_t)bx->b_loffset, 957 bx->b_bufsize, 958 (intmax_t)bp->b_loffset, 959 bx, bp); 960 if (check_buf_overlap > 1) 961 panic("bgetvp - overlapping buffer"); 962 } 963 } 964 bx = buf_rb_hash_RB_NEXT(bp); 965 if (bx) { 966 if (bp->b_loffset + testsize > bx->b_loffset) { 967 kprintf("bgetvp: overlapr %016jx/%d %016jx " 968 "bp %p bx %p\n", 969 (intmax_t)bp->b_loffset, 970 testsize, 971 (intmax_t)bx->b_loffset, 972 bp, bx); 973 if (check_buf_overlap > 1) 974 panic("bgetvp - overlapping buffer"); 975 } 976 } 977 } 978 bp->b_vp = vp; 979 bp->b_flags |= B_HASHED; 980 bp->b_flags |= B_VNCLEAN; 981 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 982 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 983 /*vhold(vp);*/ 984 lwkt_reltoken(&vp->v_token); 985 return(0); 986 } 987 988 /* 989 * Disassociate a buffer from a vnode. 990 * 991 * MPSAFE 992 */ 993 void 994 brelvp(struct buf *bp) 995 { 996 struct vnode *vp; 997 998 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 999 1000 /* 1001 * Delete from old vnode list, if on one. 1002 */ 1003 vp = bp->b_vp; 1004 lwkt_gettoken(&vp->v_token); 1005 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 1006 if (bp->b_flags & B_VNDIRTY) 1007 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1008 else 1009 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1010 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 1011 } 1012 if (bp->b_flags & B_HASHED) { 1013 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 1014 bp->b_flags &= ~B_HASHED; 1015 } 1016 1017 /* 1018 * Only remove from synclist when no dirty buffers are left AND 1019 * the VFS has not flagged the vnode's inode as being dirty. 1020 */ 1021 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST && 1022 RB_EMPTY(&vp->v_rbdirty_tree)) { 1023 vn_syncer_remove(vp, 0); 1024 } 1025 bp->b_vp = NULL; 1026 1027 lwkt_reltoken(&vp->v_token); 1028 1029 /*vdrop(vp);*/ 1030 } 1031 1032 /* 1033 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 1034 * This routine is called when the state of the B_DELWRI bit is changed. 1035 * 1036 * Must be called with vp->v_token held. 1037 * MPSAFE 1038 */ 1039 void 1040 reassignbuf(struct buf *bp) 1041 { 1042 struct vnode *vp = bp->b_vp; 1043 int delay; 1044 1045 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 1046 ++reassignbufcalls; 1047 1048 /* 1049 * B_PAGING flagged buffers cannot be reassigned because their vp 1050 * is not fully linked in. 1051 */ 1052 if (bp->b_flags & B_PAGING) 1053 panic("cannot reassign paging buffer"); 1054 1055 if (bp->b_flags & B_DELWRI) { 1056 /* 1057 * Move to the dirty list, add the vnode to the worklist 1058 */ 1059 if (bp->b_flags & B_VNCLEAN) { 1060 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1061 bp->b_flags &= ~B_VNCLEAN; 1062 } 1063 if ((bp->b_flags & B_VNDIRTY) == 0) { 1064 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 1065 panic("reassignbuf: dup lblk vp %p bp %p", 1066 vp, bp); 1067 } 1068 bp->b_flags |= B_VNDIRTY; 1069 } 1070 if ((vp->v_flag & VONWORKLST) == 0) { 1071 switch (vp->v_type) { 1072 case VDIR: 1073 delay = dirdelay; 1074 break; 1075 case VCHR: 1076 case VBLK: 1077 if (vp->v_rdev && 1078 vp->v_rdev->si_mountpoint != NULL) { 1079 delay = metadelay; 1080 break; 1081 } 1082 /* fall through */ 1083 default: 1084 delay = filedelay; 1085 } 1086 vn_syncer_add(vp, delay); 1087 } 1088 } else { 1089 /* 1090 * Move to the clean list, remove the vnode from the worklist 1091 * if no dirty blocks remain. 1092 */ 1093 if (bp->b_flags & B_VNDIRTY) { 1094 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1095 bp->b_flags &= ~B_VNDIRTY; 1096 } 1097 if ((bp->b_flags & B_VNCLEAN) == 0) { 1098 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1099 panic("reassignbuf: dup lblk vp %p bp %p", 1100 vp, bp); 1101 } 1102 bp->b_flags |= B_VNCLEAN; 1103 } 1104 1105 /* 1106 * Only remove from synclist when no dirty buffers are left 1107 * AND the VFS has not flagged the vnode's inode as being 1108 * dirty. 1109 */ 1110 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == 1111 VONWORKLST && 1112 RB_EMPTY(&vp->v_rbdirty_tree)) { 1113 vn_syncer_remove(vp, 0); 1114 } 1115 } 1116 } 1117 1118 /* 1119 * Create a vnode for a block device. Used for mounting the root file 1120 * system. 1121 * 1122 * A vref()'d vnode is returned. 1123 */ 1124 extern struct vop_ops *devfs_vnode_dev_vops_p; 1125 int 1126 bdevvp(cdev_t dev, struct vnode **vpp) 1127 { 1128 struct vnode *vp; 1129 struct vnode *nvp; 1130 int error; 1131 1132 if (dev == NULL) { 1133 *vpp = NULLVP; 1134 return (ENXIO); 1135 } 1136 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1137 &nvp, 0, 0); 1138 if (error) { 1139 *vpp = NULLVP; 1140 return (error); 1141 } 1142 vp = nvp; 1143 vp->v_type = VCHR; 1144 #if 0 1145 vp->v_rdev = dev; 1146 #endif 1147 v_associate_rdev(vp, dev); 1148 vp->v_umajor = dev->si_umajor; 1149 vp->v_uminor = dev->si_uminor; 1150 vx_unlock(vp); 1151 *vpp = vp; 1152 return (0); 1153 } 1154 1155 int 1156 v_associate_rdev(struct vnode *vp, cdev_t dev) 1157 { 1158 if (dev == NULL) 1159 return(ENXIO); 1160 if (dev_is_good(dev) == 0) 1161 return(ENXIO); 1162 KKASSERT(vp->v_rdev == NULL); 1163 vp->v_rdev = reference_dev(dev); 1164 lwkt_gettoken(&spechash_token); 1165 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1166 lwkt_reltoken(&spechash_token); 1167 return(0); 1168 } 1169 1170 void 1171 v_release_rdev(struct vnode *vp) 1172 { 1173 cdev_t dev; 1174 1175 if ((dev = vp->v_rdev) != NULL) { 1176 lwkt_gettoken(&spechash_token); 1177 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1178 vp->v_rdev = NULL; 1179 release_dev(dev); 1180 lwkt_reltoken(&spechash_token); 1181 } 1182 } 1183 1184 /* 1185 * Add a vnode to the alias list hung off the cdev_t. We only associate 1186 * the device number with the vnode. The actual device is not associated 1187 * until the vnode is opened (usually in spec_open()), and will be 1188 * disassociated on last close. 1189 */ 1190 void 1191 addaliasu(struct vnode *nvp, int x, int y) 1192 { 1193 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1194 panic("addaliasu on non-special vnode"); 1195 nvp->v_umajor = x; 1196 nvp->v_uminor = y; 1197 } 1198 1199 /* 1200 * Simple call that a filesystem can make to try to get rid of a 1201 * vnode. It will fail if anyone is referencing the vnode (including 1202 * the caller). 1203 * 1204 * The filesystem can check whether its in-memory inode structure still 1205 * references the vp on return. 1206 * 1207 * May only be called if the vnode is in a known state (i.e. being prevented 1208 * from being deallocated by some other condition such as a vfs inode hold). 1209 */ 1210 void 1211 vclean_unlocked(struct vnode *vp) 1212 { 1213 vx_get(vp); 1214 if (VREFCNT(vp) <= 1) 1215 vgone_vxlocked(vp); 1216 vx_put(vp); 1217 } 1218 1219 /* 1220 * Disassociate a vnode from its underlying filesystem. 1221 * 1222 * The vnode must be VX locked and referenced. In all normal situations 1223 * there are no active references. If vclean_vxlocked() is called while 1224 * there are active references, the vnode is being ripped out and we have 1225 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1226 */ 1227 void 1228 vclean_vxlocked(struct vnode *vp, int flags) 1229 { 1230 int active; 1231 int n; 1232 vm_object_t object; 1233 struct namecache *ncp; 1234 1235 /* 1236 * If the vnode has already been reclaimed we have nothing to do. 1237 */ 1238 if (vp->v_flag & VRECLAIMED) 1239 return; 1240 1241 /* 1242 * Set flag to interlock operation, flag finalization to ensure 1243 * that the vnode winds up on the inactive list, and set v_act to 0. 1244 */ 1245 vsetflags(vp, VRECLAIMED); 1246 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1247 vp->v_act = 0; 1248 1249 if (verbose_reclaims) { 1250 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1251 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1252 } 1253 1254 /* 1255 * Scrap the vfs cache 1256 */ 1257 while (cache_inval_vp(vp, 0) != 0) { 1258 kprintf("Warning: vnode %p clean/cache_resolution " 1259 "race detected\n", vp); 1260 tsleep(vp, 0, "vclninv", 2); 1261 } 1262 1263 /* 1264 * Check to see if the vnode is in use. If so we have to reference it 1265 * before we clean it out so that its count cannot fall to zero and 1266 * generate a race against ourselves to recycle it. 1267 */ 1268 active = (VREFCNT(vp) > 0); 1269 1270 /* 1271 * Clean out any buffers associated with the vnode and destroy its 1272 * object, if it has one. 1273 */ 1274 vinvalbuf(vp, V_SAVE, 0, 0); 1275 1276 /* 1277 * If purging an active vnode (typically during a forced unmount 1278 * or reboot), it must be closed and deactivated before being 1279 * reclaimed. This isn't really all that safe, but what can 1280 * we do? XXX. 1281 * 1282 * Note that neither of these routines unlocks the vnode. 1283 */ 1284 if (active && (flags & DOCLOSE)) { 1285 while ((n = vp->v_opencount) != 0) { 1286 if (vp->v_writecount) 1287 VOP_CLOSE(vp, FWRITE|FNONBLOCK, NULL); 1288 else 1289 VOP_CLOSE(vp, FNONBLOCK, NULL); 1290 if (vp->v_opencount == n) { 1291 kprintf("Warning: unable to force-close" 1292 " vnode %p\n", vp); 1293 break; 1294 } 1295 } 1296 } 1297 1298 /* 1299 * If the vnode has not been deactivated, deactivated it. Deactivation 1300 * can create new buffers and VM pages so we have to call vinvalbuf() 1301 * again to make sure they all get flushed. 1302 * 1303 * This can occur if a file with a link count of 0 needs to be 1304 * truncated. 1305 * 1306 * If the vnode is already dead don't try to deactivate it. 1307 */ 1308 if ((vp->v_flag & VINACTIVE) == 0) { 1309 vsetflags(vp, VINACTIVE); 1310 if (vp->v_mount) 1311 VOP_INACTIVE(vp); 1312 vinvalbuf(vp, V_SAVE, 0, 0); 1313 } 1314 1315 /* 1316 * If the vnode has an object, destroy it. 1317 */ 1318 while ((object = vp->v_object) != NULL) { 1319 vm_object_hold(object); 1320 if (object == vp->v_object) 1321 break; 1322 vm_object_drop(object); 1323 } 1324 1325 if (object != NULL) { 1326 if (object->ref_count == 0) { 1327 if ((object->flags & OBJ_DEAD) == 0) 1328 vm_object_terminate(object); 1329 vm_object_drop(object); 1330 vclrflags(vp, VOBJBUF); 1331 } else { 1332 vm_pager_deallocate(object); 1333 vclrflags(vp, VOBJBUF); 1334 vm_object_drop(object); 1335 } 1336 } 1337 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1338 1339 if (vp->v_flag & VOBJDIRTY) 1340 vclrobjdirty(vp); 1341 1342 /* 1343 * Reclaim the vnode if not already dead. 1344 */ 1345 if (vp->v_mount && VOP_RECLAIM(vp)) 1346 panic("vclean: cannot reclaim"); 1347 1348 /* 1349 * Done with purge, notify sleepers of the grim news. 1350 */ 1351 vp->v_ops = &dead_vnode_vops_p; 1352 vn_gone(vp); 1353 vp->v_tag = VT_NON; 1354 1355 /* 1356 * If we are destroying an active vnode, reactivate it now that 1357 * we have reassociated it with deadfs. This prevents the system 1358 * from crashing on the vnode due to it being unexpectedly marked 1359 * as inactive or reclaimed. 1360 */ 1361 if (active && (flags & DOCLOSE)) { 1362 vclrflags(vp, VINACTIVE | VRECLAIMED); 1363 } 1364 } 1365 1366 /* 1367 * Eliminate all activity associated with the requested vnode 1368 * and with all vnodes aliased to the requested vnode. 1369 * 1370 * The vnode must be referenced but should not be locked. 1371 */ 1372 int 1373 vrevoke(struct vnode *vp, struct ucred *cred) 1374 { 1375 struct vnode *vq; 1376 struct vnode *vqn; 1377 cdev_t dev; 1378 int error; 1379 1380 /* 1381 * If the vnode has a device association, scrap all vnodes associated 1382 * with the device. Don't let the device disappear on us while we 1383 * are scrapping the vnodes. 1384 * 1385 * The passed vp will probably show up in the list, do not VX lock 1386 * it twice! 1387 * 1388 * Releasing the vnode's rdev here can mess up specfs's call to 1389 * device close, so don't do it. The vnode has been disassociated 1390 * and the device will be closed after the last ref on the related 1391 * fp goes away (if not still open by e.g. the kernel). 1392 */ 1393 if (vp->v_type != VCHR) { 1394 error = fdrevoke(vp, DTYPE_VNODE, cred); 1395 return (error); 1396 } 1397 if ((dev = vp->v_rdev) == NULL) { 1398 return(0); 1399 } 1400 reference_dev(dev); 1401 lwkt_gettoken(&spechash_token); 1402 1403 restart: 1404 vqn = SLIST_FIRST(&dev->si_hlist); 1405 if (vqn) 1406 vhold(vqn); 1407 while ((vq = vqn) != NULL) { 1408 if (VREFCNT(vq) > 0) { 1409 vref(vq); 1410 fdrevoke(vq, DTYPE_VNODE, cred); 1411 /*v_release_rdev(vq);*/ 1412 vrele(vq); 1413 if (vq->v_rdev != dev) { 1414 vdrop(vq); 1415 goto restart; 1416 } 1417 } 1418 vqn = SLIST_NEXT(vq, v_cdevnext); 1419 if (vqn) 1420 vhold(vqn); 1421 vdrop(vq); 1422 } 1423 lwkt_reltoken(&spechash_token); 1424 dev_drevoke(dev); 1425 release_dev(dev); 1426 return (0); 1427 } 1428 1429 /* 1430 * This is called when the object underlying a vnode is being destroyed, 1431 * such as in a remove(). Try to recycle the vnode immediately if the 1432 * only active reference is our reference. 1433 * 1434 * Directory vnodes in the namecache with children cannot be immediately 1435 * recycled because numerous VOP_N*() ops require them to be stable. 1436 * 1437 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1438 * function is a NOP if VRECLAIMED is already set. 1439 */ 1440 int 1441 vrecycle(struct vnode *vp) 1442 { 1443 if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1444 if (cache_inval_vp_nonblock(vp)) 1445 return(0); 1446 vgone_vxlocked(vp); 1447 return (1); 1448 } 1449 return (0); 1450 } 1451 1452 /* 1453 * Return the maximum I/O size allowed for strategy calls on VP. 1454 * 1455 * If vp is VCHR or VBLK we dive the device, otherwise we use 1456 * the vp's mount info. 1457 * 1458 * The returned value is clamped at MAXPHYS as most callers cannot use 1459 * buffers larger than that size. 1460 */ 1461 int 1462 vmaxiosize(struct vnode *vp) 1463 { 1464 int maxiosize; 1465 1466 if (vp->v_type == VBLK || vp->v_type == VCHR) 1467 maxiosize = vp->v_rdev->si_iosize_max; 1468 else 1469 maxiosize = vp->v_mount->mnt_iosize_max; 1470 1471 if (maxiosize > MAXPHYS) 1472 maxiosize = MAXPHYS; 1473 return (maxiosize); 1474 } 1475 1476 /* 1477 * Eliminate all activity associated with a vnode in preparation for 1478 * destruction. 1479 * 1480 * The vnode must be VX locked and refd and will remain VX locked and refd 1481 * on return. This routine may be called with the vnode in any state, as 1482 * long as it is VX locked. The vnode will be cleaned out and marked 1483 * VRECLAIMED but will not actually be reused until all existing refs and 1484 * holds go away. 1485 * 1486 * NOTE: This routine may be called on a vnode which has not yet been 1487 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1488 * already been reclaimed. 1489 * 1490 * This routine is not responsible for placing us back on the freelist. 1491 * Instead, it happens automatically when the caller releases the VX lock 1492 * (assuming there aren't any other references). 1493 */ 1494 void 1495 vgone_vxlocked(struct vnode *vp) 1496 { 1497 /* 1498 * assert that the VX lock is held. This is an absolute requirement 1499 * now for vgone_vxlocked() to be called. 1500 */ 1501 KKASSERT(lockinuse(&vp->v_lock)); 1502 1503 /* 1504 * Clean out the filesystem specific data and set the VRECLAIMED 1505 * bit. Also deactivate the vnode if necessary. 1506 * 1507 * The vnode should have automatically been removed from the syncer 1508 * list as syncer/dirty flags cleared during the cleaning. 1509 */ 1510 vclean_vxlocked(vp, DOCLOSE); 1511 1512 /* 1513 * Normally panic if the vnode is still dirty, unless we are doing 1514 * a forced unmount (tmpfs typically). 1515 */ 1516 if (vp->v_flag & VONWORKLST) { 1517 if (vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF) { 1518 /* force removal */ 1519 vn_syncer_remove(vp, 1); 1520 } else { 1521 panic("vp %p still dirty in vgone after flush", vp); 1522 } 1523 } 1524 1525 /* 1526 * Delete from old mount point vnode list, if on one. 1527 */ 1528 if (vp->v_mount != NULL) { 1529 KKASSERT(vp->v_data == NULL); 1530 insmntque(vp, NULL); 1531 } 1532 1533 /* 1534 * If special device, remove it from special device alias list 1535 * if it is on one. This should normally only occur if a vnode is 1536 * being revoked as the device should otherwise have been released 1537 * naturally. 1538 */ 1539 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1540 v_release_rdev(vp); 1541 } 1542 1543 /* 1544 * Set us to VBAD 1545 */ 1546 vp->v_type = VBAD; 1547 } 1548 1549 /* 1550 * Lookup a vnode by device number. 1551 * 1552 * Returns non-zero and *vpp set to a vref'd vnode on success. 1553 * Returns zero on failure. 1554 */ 1555 int 1556 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1557 { 1558 struct vnode *vp; 1559 1560 lwkt_gettoken(&spechash_token); 1561 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1562 if (type == vp->v_type) { 1563 *vpp = vp; 1564 vref(vp); 1565 lwkt_reltoken(&spechash_token); 1566 return (1); 1567 } 1568 } 1569 lwkt_reltoken(&spechash_token); 1570 return (0); 1571 } 1572 1573 /* 1574 * Calculate the total number of references to a special device. This 1575 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1576 * an overloaded field. Since udev2dev can now return NULL, we have 1577 * to check for a NULL v_rdev. 1578 */ 1579 int 1580 count_dev(cdev_t dev) 1581 { 1582 struct vnode *vp; 1583 int count = 0; 1584 1585 if (SLIST_FIRST(&dev->si_hlist)) { 1586 lwkt_gettoken(&spechash_token); 1587 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1588 count += vp->v_opencount; 1589 } 1590 lwkt_reltoken(&spechash_token); 1591 } 1592 return(count); 1593 } 1594 1595 int 1596 vcount(struct vnode *vp) 1597 { 1598 if (vp->v_rdev == NULL) 1599 return(0); 1600 return(count_dev(vp->v_rdev)); 1601 } 1602 1603 /* 1604 * Initialize VMIO for a vnode. This routine MUST be called before a 1605 * VFS can issue buffer cache ops on a vnode. It is typically called 1606 * when a vnode is initialized from its inode. 1607 */ 1608 int 1609 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1610 { 1611 vm_object_t object; 1612 int error = 0; 1613 1614 object = vp->v_object; 1615 if (object) { 1616 vm_object_hold(object); 1617 KKASSERT(vp->v_object == object); 1618 } 1619 1620 if (object == NULL) { 1621 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1622 1623 /* 1624 * Dereference the reference we just created. This assumes 1625 * that the object is associated with the vp. Allow it to 1626 * have zero refs. It cannot be destroyed as long as it 1627 * is associated with the vnode. 1628 */ 1629 vm_object_hold(object); 1630 atomic_add_int(&object->ref_count, -1); 1631 vrele(vp); 1632 } else { 1633 KKASSERT((object->flags & OBJ_DEAD) == 0); 1634 } 1635 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1636 vsetflags(vp, VOBJBUF); 1637 vm_object_drop(object); 1638 1639 return (error); 1640 } 1641 1642 1643 /* 1644 * Print out a description of a vnode. 1645 */ 1646 static char *typename[] = 1647 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1648 1649 void 1650 vprint(char *label, struct vnode *vp) 1651 { 1652 char buf[96]; 1653 1654 if (label != NULL) 1655 kprintf("%s: %p: ", label, (void *)vp); 1656 else 1657 kprintf("%p: ", (void *)vp); 1658 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,", 1659 typename[vp->v_type], 1660 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs); 1661 buf[0] = '\0'; 1662 if (vp->v_flag & VROOT) 1663 strcat(buf, "|VROOT"); 1664 if (vp->v_flag & VPFSROOT) 1665 strcat(buf, "|VPFSROOT"); 1666 if (vp->v_flag & VTEXT) 1667 strcat(buf, "|VTEXT"); 1668 if (vp->v_flag & VSYSTEM) 1669 strcat(buf, "|VSYSTEM"); 1670 if (vp->v_flag & VOBJBUF) 1671 strcat(buf, "|VOBJBUF"); 1672 if (buf[0] != '\0') 1673 kprintf(" flags (%s)", &buf[1]); 1674 if (vp->v_data == NULL) { 1675 kprintf("\n"); 1676 } else { 1677 kprintf("\n\t"); 1678 VOP_PRINT(vp); 1679 } 1680 } 1681 1682 /* 1683 * Do the usual access checking. 1684 * file_mode, uid and gid are from the vnode in question, 1685 * while acc_mode and cred are from the VOP_ACCESS parameter list 1686 */ 1687 int 1688 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1689 mode_t acc_mode, struct ucred *cred) 1690 { 1691 mode_t mask; 1692 int ismember; 1693 1694 /* 1695 * Super-user always gets read/write access, but execute access depends 1696 * on at least one execute bit being set. 1697 */ 1698 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1699 if ((acc_mode & VEXEC) && type != VDIR && 1700 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1701 return (EACCES); 1702 return (0); 1703 } 1704 1705 mask = 0; 1706 1707 /* Otherwise, check the owner. */ 1708 if (cred->cr_uid == uid) { 1709 if (acc_mode & VEXEC) 1710 mask |= S_IXUSR; 1711 if (acc_mode & VREAD) 1712 mask |= S_IRUSR; 1713 if (acc_mode & VWRITE) 1714 mask |= S_IWUSR; 1715 return ((file_mode & mask) == mask ? 0 : EACCES); 1716 } 1717 1718 /* Otherwise, check the groups. */ 1719 ismember = groupmember(gid, cred); 1720 if (cred->cr_svgid == gid || ismember) { 1721 if (acc_mode & VEXEC) 1722 mask |= S_IXGRP; 1723 if (acc_mode & VREAD) 1724 mask |= S_IRGRP; 1725 if (acc_mode & VWRITE) 1726 mask |= S_IWGRP; 1727 return ((file_mode & mask) == mask ? 0 : EACCES); 1728 } 1729 1730 /* Otherwise, check everyone else. */ 1731 if (acc_mode & VEXEC) 1732 mask |= S_IXOTH; 1733 if (acc_mode & VREAD) 1734 mask |= S_IROTH; 1735 if (acc_mode & VWRITE) 1736 mask |= S_IWOTH; 1737 return ((file_mode & mask) == mask ? 0 : EACCES); 1738 } 1739 1740 #ifdef DDB 1741 #include <ddb/ddb.h> 1742 1743 static int db_show_locked_vnodes(struct mount *mp, void *data); 1744 1745 /* 1746 * List all of the locked vnodes in the system. 1747 * Called when debugging the kernel. 1748 */ 1749 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1750 { 1751 kprintf("Locked vnodes\n"); 1752 mountlist_scan(db_show_locked_vnodes, NULL, 1753 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1754 } 1755 1756 static int 1757 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1758 { 1759 struct vnode *vp; 1760 1761 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1762 if (vn_islocked(vp)) 1763 vprint(NULL, vp); 1764 } 1765 return(0); 1766 } 1767 #endif 1768 1769 /* 1770 * Top level filesystem related information gathering. 1771 */ 1772 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1773 1774 static int 1775 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1776 { 1777 int *name = (int *)arg1 - 1; /* XXX */ 1778 u_int namelen = arg2 + 1; /* XXX */ 1779 struct vfsconf *vfsp; 1780 int maxtypenum; 1781 1782 #if 1 || defined(COMPAT_PRELITE2) 1783 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1784 if (namelen == 1) 1785 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1786 #endif 1787 1788 #ifdef notyet 1789 /* all sysctl names at this level are at least name and field */ 1790 if (namelen < 2) 1791 return (ENOTDIR); /* overloaded */ 1792 if (name[0] != VFS_GENERIC) { 1793 vfsp = vfsconf_find_by_typenum(name[0]); 1794 if (vfsp == NULL) 1795 return (EOPNOTSUPP); 1796 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1797 oldp, oldlenp, newp, newlen, p)); 1798 } 1799 #endif 1800 switch (name[1]) { 1801 case VFS_MAXTYPENUM: 1802 if (namelen != 2) 1803 return (ENOTDIR); 1804 maxtypenum = vfsconf_get_maxtypenum(); 1805 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1806 case VFS_CONF: 1807 if (namelen != 3) 1808 return (ENOTDIR); /* overloaded */ 1809 vfsp = vfsconf_find_by_typenum(name[2]); 1810 if (vfsp == NULL) 1811 return (EOPNOTSUPP); 1812 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1813 } 1814 return (EOPNOTSUPP); 1815 } 1816 1817 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1818 "Generic filesystem"); 1819 1820 #if 1 || defined(COMPAT_PRELITE2) 1821 1822 static int 1823 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1824 { 1825 int error; 1826 struct ovfsconf ovfs; 1827 struct sysctl_req *req = (struct sysctl_req*) data; 1828 1829 bzero(&ovfs, sizeof(ovfs)); 1830 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1831 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1832 ovfs.vfc_index = vfsp->vfc_typenum; 1833 ovfs.vfc_refcount = vfsp->vfc_refcount; 1834 ovfs.vfc_flags = vfsp->vfc_flags; 1835 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1836 if (error) 1837 return error; /* abort iteration with error code */ 1838 else 1839 return 0; /* continue iterating with next element */ 1840 } 1841 1842 static int 1843 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1844 { 1845 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1846 } 1847 1848 #endif /* 1 || COMPAT_PRELITE2 */ 1849 1850 /* 1851 * Check to see if a filesystem is mounted on a block device. 1852 */ 1853 int 1854 vfs_mountedon(struct vnode *vp) 1855 { 1856 cdev_t dev; 1857 1858 if ((dev = vp->v_rdev) == NULL) { 1859 /* if (vp->v_type != VBLK) 1860 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1861 } 1862 if (dev != NULL && dev->si_mountpoint) 1863 return (EBUSY); 1864 return (0); 1865 } 1866 1867 /* 1868 * Unmount all filesystems. The list is traversed in reverse order 1869 * of mounting to avoid dependencies. 1870 * 1871 * We want the umountall to be able to break out of its loop if a 1872 * failure occurs, after scanning all possible mounts, so the callback 1873 * returns 0 on error. 1874 * 1875 * NOTE: Do not call mountlist_remove(mp) on error any more, this will 1876 * confuse mountlist_scan()'s unbusy check. 1877 */ 1878 static int vfs_umountall_callback(struct mount *mp, void *data); 1879 1880 void 1881 vfs_unmountall(int halting) 1882 { 1883 int count; 1884 1885 do { 1886 count = mountlist_scan(vfs_umountall_callback, &halting, 1887 MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1888 } while (count); 1889 } 1890 1891 static 1892 int 1893 vfs_umountall_callback(struct mount *mp, void *data) 1894 { 1895 int error; 1896 int halting = *(int *)data; 1897 1898 /* 1899 * NOTE: When halting, dounmount will disconnect but leave 1900 * certain mount points intact. e.g. devfs. 1901 */ 1902 error = dounmount(mp, MNT_FORCE, halting); 1903 if (error) { 1904 kprintf("unmount of filesystem mounted from %s failed (", 1905 mp->mnt_stat.f_mntfromname); 1906 if (error == EBUSY) 1907 kprintf("BUSY)\n"); 1908 else 1909 kprintf("%d)\n", error); 1910 return 0; 1911 } else { 1912 return 1; 1913 } 1914 } 1915 1916 /* 1917 * Checks the mount flags for parameter mp and put the names comma-separated 1918 * into a string buffer buf with a size limit specified by len. 1919 * 1920 * It returns the number of bytes written into buf, and (*errorp) will be 1921 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1922 * not large enough). The buffer will be 0-terminated if len was not 0. 1923 */ 1924 size_t 1925 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1926 char *buf, size_t len, int *errorp) 1927 { 1928 static const struct mountctl_opt optnames[] = { 1929 { MNT_RDONLY, "read-only" }, 1930 { MNT_SYNCHRONOUS, "synchronous" }, 1931 { MNT_NOEXEC, "noexec" }, 1932 { MNT_NOSUID, "nosuid" }, 1933 { MNT_NODEV, "nodev" }, 1934 { MNT_AUTOMOUNTED, "automounted" }, 1935 { MNT_ASYNC, "asynchronous" }, 1936 { MNT_SUIDDIR, "suiddir" }, 1937 { MNT_SOFTDEP, "soft-updates" }, 1938 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1939 { MNT_TRIM, "trim" }, 1940 { MNT_NOATIME, "noatime" }, 1941 { MNT_NOCLUSTERR, "noclusterr" }, 1942 { MNT_NOCLUSTERW, "noclusterw" }, 1943 { MNT_EXRDONLY, "NFS read-only" }, 1944 { MNT_EXPORTED, "NFS exported" }, 1945 /* Remaining NFS flags could come here */ 1946 { MNT_LOCAL, "local" }, 1947 { MNT_QUOTA, "with-quotas" }, 1948 /* { MNT_ROOTFS, "rootfs" }, */ 1949 /* { MNT_IGNORE, "ignore" }, */ 1950 { 0, NULL} 1951 }; 1952 int bwritten; 1953 int bleft; 1954 int optlen; 1955 int actsize; 1956 1957 *errorp = 0; 1958 bwritten = 0; 1959 bleft = len - 1; /* leave room for trailing \0 */ 1960 1961 /* 1962 * Checks the size of the string. If it contains 1963 * any data, then we will append the new flags to 1964 * it. 1965 */ 1966 actsize = strlen(buf); 1967 if (actsize > 0) 1968 buf += actsize; 1969 1970 /* Default flags if no flags passed */ 1971 if (optp == NULL) 1972 optp = optnames; 1973 1974 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1975 *errorp = EINVAL; 1976 return(0); 1977 } 1978 1979 for (; flags && optp->o_opt; ++optp) { 1980 if ((flags & optp->o_opt) == 0) 1981 continue; 1982 optlen = strlen(optp->o_name); 1983 if (bwritten || actsize > 0) { 1984 if (bleft < 2) { 1985 *errorp = ENOSPC; 1986 break; 1987 } 1988 buf[bwritten++] = ','; 1989 buf[bwritten++] = ' '; 1990 bleft -= 2; 1991 } 1992 if (bleft < optlen) { 1993 *errorp = ENOSPC; 1994 break; 1995 } 1996 bcopy(optp->o_name, buf + bwritten, optlen); 1997 bwritten += optlen; 1998 bleft -= optlen; 1999 flags &= ~optp->o_opt; 2000 } 2001 2002 /* 2003 * Space already reserved for trailing \0 2004 */ 2005 buf[bwritten] = 0; 2006 return (bwritten); 2007 } 2008 2009 /* 2010 * Build hash lists of net addresses and hang them off the mount point. 2011 * Called by ufs_mount() to set up the lists of export addresses. 2012 */ 2013 static int 2014 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 2015 const struct export_args *argp) 2016 { 2017 struct netcred *np; 2018 struct radix_node_head *rnh; 2019 int i; 2020 struct radix_node *rn; 2021 struct sockaddr *saddr, *smask = NULL; 2022 int error; 2023 2024 if (argp->ex_addrlen == 0) { 2025 if (mp->mnt_flag & MNT_DEFEXPORTED) 2026 return (EPERM); 2027 np = &nep->ne_defexported; 2028 np->netc_exflags = argp->ex_flags; 2029 np->netc_anon = argp->ex_anon; 2030 np->netc_anon.cr_ref = 1; 2031 mp->mnt_flag |= MNT_DEFEXPORTED; 2032 return (0); 2033 } 2034 2035 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 2036 return (EINVAL); 2037 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 2038 return (EINVAL); 2039 2040 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2041 np = (struct netcred *)kmalloc(i, M_NETCRED, M_WAITOK | M_ZERO); 2042 saddr = (struct sockaddr *) (np + 1); 2043 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2044 goto out; 2045 if (saddr->sa_len > argp->ex_addrlen) 2046 saddr->sa_len = argp->ex_addrlen; 2047 if (argp->ex_masklen) { 2048 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2049 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2050 if (error) 2051 goto out; 2052 if (smask->sa_len > argp->ex_masklen) 2053 smask->sa_len = argp->ex_masklen; 2054 } 2055 NE_LOCK(nep); 2056 if (nep->ne_maskhead == NULL) { 2057 if (!rn_inithead((void **)&nep->ne_maskhead, NULL, 0)) { 2058 error = ENOBUFS; 2059 goto out; 2060 } 2061 } 2062 if ((rnh = vfs_create_addrlist_af(saddr->sa_family, nep)) == NULL) { 2063 error = ENOBUFS; 2064 goto out; 2065 } 2066 rn = (*rnh->rnh_addaddr)((char *)saddr, (char *)smask, rnh, 2067 np->netc_rnodes); 2068 NE_UNLOCK(nep); 2069 if (rn == NULL || np != (struct netcred *)rn) { /* already exists */ 2070 error = EPERM; 2071 goto out; 2072 } 2073 np->netc_exflags = argp->ex_flags; 2074 np->netc_anon = argp->ex_anon; 2075 np->netc_anon.cr_ref = 1; 2076 return (0); 2077 2078 out: 2079 kfree(np, M_NETCRED); 2080 return (error); 2081 } 2082 2083 /* 2084 * Free netcred structures installed in the netexport 2085 */ 2086 static int 2087 vfs_free_netcred(struct radix_node *rn, void *w) 2088 { 2089 struct radix_node_head *rnh = (struct radix_node_head *)w; 2090 2091 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2092 kfree(rn, M_NETCRED); 2093 2094 return (0); 2095 } 2096 2097 /* 2098 * callback to free an element of the mask table installed in the 2099 * netexport. These may be created indirectly and are not netcred 2100 * structures. 2101 */ 2102 static int 2103 vfs_free_netcred_mask(struct radix_node *rn, void *w) 2104 { 2105 struct radix_node_head *rnh = (struct radix_node_head *)w; 2106 2107 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2108 kfree(rn, M_RTABLE); 2109 2110 return (0); 2111 } 2112 2113 static struct radix_node_head * 2114 vfs_create_addrlist_af(int af, struct netexport *nep) 2115 { 2116 struct radix_node_head *rnh = NULL; 2117 #if defined(INET) || defined(INET6) 2118 struct radix_node_head *maskhead = nep->ne_maskhead; 2119 int off; 2120 #endif 2121 2122 NE_ASSERT_LOCKED(nep); 2123 #if defined(INET) || defined(INET6) 2124 KKASSERT(maskhead != NULL); 2125 #endif 2126 switch (af) { 2127 #ifdef INET 2128 case AF_INET: 2129 if ((rnh = nep->ne_inethead) == NULL) { 2130 off = offsetof(struct sockaddr_in, sin_addr) << 3; 2131 if (!rn_inithead((void **)&rnh, maskhead, off)) 2132 return (NULL); 2133 nep->ne_inethead = rnh; 2134 } 2135 break; 2136 #endif 2137 #ifdef INET6 2138 case AF_INET6: 2139 if ((rnh = nep->ne_inet6head) == NULL) { 2140 off = offsetof(struct sockaddr_in6, sin6_addr) << 3; 2141 if (!rn_inithead((void **)&rnh, maskhead, off)) 2142 return (NULL); 2143 nep->ne_inet6head = rnh; 2144 } 2145 break; 2146 #endif 2147 } 2148 return (rnh); 2149 } 2150 2151 /* 2152 * helper function for freeing netcred elements 2153 */ 2154 static void 2155 vfs_free_addrlist_af(struct radix_node_head **prnh) 2156 { 2157 struct radix_node_head *rnh = *prnh; 2158 2159 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, rnh); 2160 kfree(rnh, M_RTABLE); 2161 *prnh = NULL; 2162 } 2163 2164 /* 2165 * helper function for freeing mask elements 2166 */ 2167 static void 2168 vfs_free_addrlist_masks(struct radix_node_head **prnh) 2169 { 2170 struct radix_node_head *rnh = *prnh; 2171 2172 (*rnh->rnh_walktree) (rnh, vfs_free_netcred_mask, rnh); 2173 kfree(rnh, M_RTABLE); 2174 *prnh = NULL; 2175 } 2176 2177 /* 2178 * Free the net address hash lists that are hanging off the mount points. 2179 */ 2180 static void 2181 vfs_free_addrlist(struct netexport *nep) 2182 { 2183 NE_LOCK(nep); 2184 if (nep->ne_inethead != NULL) 2185 vfs_free_addrlist_af(&nep->ne_inethead); 2186 if (nep->ne_inet6head != NULL) 2187 vfs_free_addrlist_af(&nep->ne_inet6head); 2188 if (nep->ne_maskhead) 2189 vfs_free_addrlist_masks(&nep->ne_maskhead); 2190 NE_UNLOCK(nep); 2191 } 2192 2193 int 2194 vfs_export(struct mount *mp, struct netexport *nep, 2195 const struct export_args *argp) 2196 { 2197 int error; 2198 2199 if (argp->ex_flags & MNT_DELEXPORT) { 2200 if (mp->mnt_flag & MNT_EXPUBLIC) { 2201 vfs_setpublicfs(NULL, NULL, NULL); 2202 mp->mnt_flag &= ~MNT_EXPUBLIC; 2203 } 2204 vfs_free_addrlist(nep); 2205 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2206 } 2207 if (argp->ex_flags & MNT_EXPORTED) { 2208 if (argp->ex_flags & MNT_EXPUBLIC) { 2209 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2210 return (error); 2211 mp->mnt_flag |= MNT_EXPUBLIC; 2212 } 2213 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2214 return (error); 2215 mp->mnt_flag |= MNT_EXPORTED; 2216 } 2217 return (0); 2218 } 2219 2220 2221 /* 2222 * Set the publicly exported filesystem (WebNFS). Currently, only 2223 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2224 */ 2225 int 2226 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2227 const struct export_args *argp) 2228 { 2229 int error; 2230 struct vnode *rvp; 2231 char *cp; 2232 2233 /* 2234 * mp == NULL -> invalidate the current info, the FS is 2235 * no longer exported. May be called from either vfs_export 2236 * or unmount, so check if it hasn't already been done. 2237 */ 2238 if (mp == NULL) { 2239 if (nfs_pub.np_valid) { 2240 nfs_pub.np_valid = 0; 2241 if (nfs_pub.np_index != NULL) { 2242 kfree(nfs_pub.np_index, M_TEMP); 2243 nfs_pub.np_index = NULL; 2244 } 2245 } 2246 return (0); 2247 } 2248 2249 /* 2250 * Only one allowed at a time. 2251 */ 2252 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2253 return (EBUSY); 2254 2255 /* 2256 * Get real filehandle for root of exported FS. 2257 */ 2258 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2259 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2260 2261 if ((error = VFS_ROOT(mp, &rvp))) 2262 return (error); 2263 2264 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2265 return (error); 2266 2267 vput(rvp); 2268 2269 /* 2270 * If an indexfile was specified, pull it in. 2271 */ 2272 if (argp->ex_indexfile != NULL) { 2273 int namelen; 2274 2275 error = vn_get_namelen(rvp, &namelen); 2276 if (error) 2277 return (error); 2278 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2279 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2280 namelen, NULL); 2281 if (!error) { 2282 /* 2283 * Check for illegal filenames. 2284 */ 2285 for (cp = nfs_pub.np_index; *cp; cp++) { 2286 if (*cp == '/') { 2287 error = EINVAL; 2288 break; 2289 } 2290 } 2291 } 2292 if (error) { 2293 kfree(nfs_pub.np_index, M_TEMP); 2294 return (error); 2295 } 2296 } 2297 2298 nfs_pub.np_mount = mp; 2299 nfs_pub.np_valid = 1; 2300 return (0); 2301 } 2302 2303 struct netcred * 2304 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2305 struct sockaddr *nam) 2306 { 2307 struct netcred *np; 2308 struct radix_node_head *rnh; 2309 struct sockaddr *saddr; 2310 2311 np = NULL; 2312 if (mp->mnt_flag & MNT_EXPORTED) { 2313 /* 2314 * Lookup in the export list first. 2315 */ 2316 NE_LOCK(nep); 2317 if (nam != NULL) { 2318 saddr = nam; 2319 switch (saddr->sa_family) { 2320 #ifdef INET 2321 case AF_INET: 2322 rnh = nep->ne_inethead; 2323 break; 2324 #endif 2325 #ifdef INET6 2326 case AF_INET6: 2327 rnh = nep->ne_inet6head; 2328 break; 2329 #endif 2330 default: 2331 rnh = NULL; 2332 } 2333 if (rnh != NULL) { 2334 np = (struct netcred *) 2335 (*rnh->rnh_matchaddr)((char *)saddr, 2336 rnh); 2337 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2338 np = NULL; 2339 } 2340 } 2341 NE_UNLOCK(nep); 2342 /* 2343 * If no address match, use the default if it exists. 2344 */ 2345 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2346 np = &nep->ne_defexported; 2347 } 2348 return (np); 2349 } 2350 2351 /* 2352 * perform msync on all vnodes under a mount point. The mount point must 2353 * be locked. This code is also responsible for lazy-freeing unreferenced 2354 * vnodes whos VM objects no longer contain pages. 2355 * 2356 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2357 * 2358 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2359 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2360 * way up in this high level function. 2361 */ 2362 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2363 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2364 2365 void 2366 vfs_msync(struct mount *mp, int flags) 2367 { 2368 int vmsc_flags; 2369 2370 /* 2371 * tmpfs sets this flag to prevent msync(), sync, and the 2372 * filesystem periodic syncer from trying to flush VM pages 2373 * to swap. Only pure memory pressure flushes tmpfs VM pages 2374 * to swap. 2375 */ 2376 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2377 return; 2378 2379 /* 2380 * Ok, scan the vnodes for work. If the filesystem is using the 2381 * syncer thread feature we can use vsyncscan() instead of 2382 * vmntvnodescan(), which is much faster. 2383 */ 2384 vmsc_flags = VMSC_GETVP; 2385 if (flags != MNT_WAIT) 2386 vmsc_flags |= VMSC_NOWAIT; 2387 2388 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 2389 vsyncscan(mp, vmsc_flags, vfs_msync_scan2, 2390 (void *)(intptr_t)flags); 2391 } else { 2392 vmntvnodescan(mp, vmsc_flags, 2393 vfs_msync_scan1, vfs_msync_scan2, 2394 (void *)(intptr_t)flags); 2395 } 2396 } 2397 2398 /* 2399 * scan1 is a fast pre-check. There could be hundreds of thousands of 2400 * vnodes, we cannot afford to do anything heavy weight until we have a 2401 * fairly good indication that there is work to do. 2402 */ 2403 static 2404 int 2405 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2406 { 2407 int flags = (int)(intptr_t)data; 2408 2409 if ((vp->v_flag & VRECLAIMED) == 0) { 2410 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 && 2411 vp->v_object) { 2412 return(0); /* call scan2 */ 2413 } 2414 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2415 (vp->v_flag & VOBJDIRTY) && 2416 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2417 return(0); /* call scan2 */ 2418 } 2419 } 2420 2421 /* 2422 * do not call scan2, continue the loop 2423 */ 2424 return(-1); 2425 } 2426 2427 /* 2428 * This callback is handed a locked vnode. 2429 */ 2430 static 2431 int 2432 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2433 { 2434 vm_object_t obj; 2435 int flags = (int)(intptr_t)data; 2436 2437 if (vp->v_flag & VRECLAIMED) 2438 return(0); 2439 2440 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2441 if ((obj = vp->v_object) != NULL) { 2442 vm_object_page_clean(obj, 0, 0, 2443 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2444 } 2445 } 2446 return(0); 2447 } 2448 2449 /* 2450 * Wake up anyone interested in vp because it is being revoked. 2451 */ 2452 void 2453 vn_gone(struct vnode *vp) 2454 { 2455 lwkt_gettoken(&vp->v_token); 2456 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2457 lwkt_reltoken(&vp->v_token); 2458 } 2459 2460 /* 2461 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2462 * (or v_rdev might be NULL). 2463 */ 2464 cdev_t 2465 vn_todev(struct vnode *vp) 2466 { 2467 if (vp->v_type != VBLK && vp->v_type != VCHR) 2468 return (NULL); 2469 KKASSERT(vp->v_rdev != NULL); 2470 return (vp->v_rdev); 2471 } 2472 2473 /* 2474 * Check if vnode represents a disk device. The vnode does not need to be 2475 * opened. 2476 * 2477 * MPALMOSTSAFE 2478 */ 2479 int 2480 vn_isdisk(struct vnode *vp, int *errp) 2481 { 2482 cdev_t dev; 2483 2484 if (vp->v_type != VCHR) { 2485 if (errp != NULL) 2486 *errp = ENOTBLK; 2487 return (0); 2488 } 2489 2490 dev = vp->v_rdev; 2491 2492 if (dev == NULL) { 2493 if (errp != NULL) 2494 *errp = ENXIO; 2495 return (0); 2496 } 2497 if (dev_is_good(dev) == 0) { 2498 if (errp != NULL) 2499 *errp = ENXIO; 2500 return (0); 2501 } 2502 if ((dev_dflags(dev) & D_DISK) == 0) { 2503 if (errp != NULL) 2504 *errp = ENOTBLK; 2505 return (0); 2506 } 2507 if (errp != NULL) 2508 *errp = 0; 2509 return (1); 2510 } 2511 2512 int 2513 vn_get_namelen(struct vnode *vp, int *namelen) 2514 { 2515 int error; 2516 register_t retval[2]; 2517 2518 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2519 if (error) 2520 return (error); 2521 *namelen = (int)retval[0]; 2522 return (0); 2523 } 2524 2525 int 2526 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2527 uint16_t d_namlen, const char *d_name) 2528 { 2529 struct dirent *dp; 2530 size_t len; 2531 2532 len = _DIRENT_RECLEN(d_namlen); 2533 if (len > uio->uio_resid) 2534 return(1); 2535 2536 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2537 2538 dp->d_ino = d_ino; 2539 dp->d_namlen = d_namlen; 2540 dp->d_type = d_type; 2541 bcopy(d_name, dp->d_name, d_namlen); 2542 2543 *error = uiomove((caddr_t)dp, len, uio); 2544 2545 kfree(dp, M_TEMP); 2546 2547 return(0); 2548 } 2549 2550 void 2551 vn_mark_atime(struct vnode *vp, struct thread *td) 2552 { 2553 struct proc *p = td->td_proc; 2554 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2555 2556 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2557 VOP_MARKATIME(vp, cred); 2558 } 2559 } 2560 2561 /* 2562 * Calculate the number of entries in an inode-related chained hash table. 2563 * With today's memory sizes, maxvnodes can wind up being a very large 2564 * number. There is no reason to waste memory, so tolerate some stacking. 2565 */ 2566 int 2567 vfs_inodehashsize(void) 2568 { 2569 int hsize; 2570 2571 hsize = 32; 2572 while (hsize < maxvnodes) 2573 hsize <<= 1; 2574 while (hsize > maxvnodes * 2) 2575 hsize >>= 1; /* nominal 2x stacking */ 2576 2577 if (maxvnodes > 1024 * 1024) 2578 hsize >>= 1; /* nominal 8x stacking */ 2579 2580 if (maxvnodes > 128 * 1024) 2581 hsize >>= 1; /* nominal 4x stacking */ 2582 2583 if (hsize < 16) 2584 hsize = 16; 2585 2586 return hsize; 2587 } 2588