1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 35 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 36 */ 37 38 /* 39 * External virtual filesystem routines 40 */ 41 #include "opt_ddb.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/buf.h> 46 #include <sys/conf.h> 47 #include <sys/dirent.h> 48 #include <sys/domain.h> 49 #include <sys/eventhandler.h> 50 #include <sys/fcntl.h> 51 #include <sys/file.h> 52 #include <sys/kernel.h> 53 #include <sys/kthread.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/mount.h> 57 #include <sys/priv.h> 58 #include <sys/proc.h> 59 #include <sys/reboot.h> 60 #include <sys/socket.h> 61 #include <sys/stat.h> 62 #include <sys/sysctl.h> 63 #include <sys/syslog.h> 64 #include <sys/unistd.h> 65 #include <sys/vmmeter.h> 66 #include <sys/vnode.h> 67 68 #include <machine/limits.h> 69 70 #include <vm/vm.h> 71 #include <vm/vm_object.h> 72 #include <vm/vm_extern.h> 73 #include <vm/vm_kern.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_map.h> 76 #include <vm/vm_page.h> 77 #include <vm/vm_pager.h> 78 #include <vm/vnode_pager.h> 79 #include <vm/vm_zone.h> 80 81 #include <sys/buf2.h> 82 #include <sys/thread2.h> 83 #include <sys/sysref2.h> 84 #include <sys/mplock2.h> 85 86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 87 88 int numvnodes; 89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 90 "Number of vnodes allocated"); 91 int verbose_reclaims; 92 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0, 93 "Output filename of reclaimed vnode(s)"); 94 95 enum vtype iftovt_tab[16] = { 96 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 97 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 98 }; 99 int vttoif_tab[9] = { 100 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 101 S_IFSOCK, S_IFIFO, S_IFMT, 102 }; 103 104 static int reassignbufcalls; 105 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 106 0, "Number of times buffers have been reassigned to the proper list"); 107 108 static int check_buf_overlap = 2; /* invasive check */ 109 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap, 110 0, "Enable overlapping buffer checks"); 111 112 int nfs_mount_type = -1; 113 static struct lwkt_token spechash_token; 114 struct nfs_public nfs_pub; /* publicly exported FS */ 115 116 int desiredvnodes; 117 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 118 &desiredvnodes, 0, "Maximum number of vnodes"); 119 120 static void vfs_free_addrlist (struct netexport *nep); 121 static int vfs_free_netcred (struct radix_node *rn, void *w); 122 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 123 const struct export_args *argp); 124 125 /* 126 * Red black tree functions 127 */ 128 static int rb_buf_compare(struct buf *b1, struct buf *b2); 129 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 130 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 131 132 static int 133 rb_buf_compare(struct buf *b1, struct buf *b2) 134 { 135 if (b1->b_loffset < b2->b_loffset) 136 return(-1); 137 if (b1->b_loffset > b2->b_loffset) 138 return(1); 139 return(0); 140 } 141 142 /* 143 * Returns non-zero if the vnode is a candidate for lazy msyncing. 144 * 145 * NOTE: v_object is not stable (this scan can race), however the 146 * mntvnodescan code holds vmobj_token so any VM object we 147 * do find will remain stable storage. 148 */ 149 static __inline int 150 vshouldmsync(struct vnode *vp) 151 { 152 vm_object_t object; 153 154 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0) 155 return (0); /* other holders */ 156 object = vp->v_object; 157 cpu_ccfence(); 158 if (object && (object->ref_count || object->resident_page_count)) 159 return(0); 160 return (1); 161 } 162 163 /* 164 * Initialize the vnode management data structures. 165 * 166 * Called from vfsinit() 167 */ 168 void 169 vfs_subr_init(void) 170 { 171 int factor1; 172 int factor2; 173 174 /* 175 * Desiredvnodes is kern.maxvnodes. We want to scale it 176 * according to available system memory but we may also have 177 * to limit it based on available KVM, which is capped on 32 bit 178 * systems, to ~80K vnodes or so. 179 * 180 * WARNING! For machines with 64-256M of ram we have to be sure 181 * that the default limit scales down well due to HAMMER 182 * taking up significantly more memory per-vnode vs UFS. 183 * We want around ~5800 on a 128M machine. 184 */ 185 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode)); 186 factor2 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode)); 187 desiredvnodes = 188 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 189 KvaSize / factor2); 190 desiredvnodes = imax(desiredvnodes, maxproc * 8); 191 192 lwkt_token_init(&spechash_token, "spechash"); 193 } 194 195 /* 196 * Knob to control the precision of file timestamps: 197 * 198 * 0 = seconds only; nanoseconds zeroed. 199 * 1 = seconds and nanoseconds, accurate within 1/HZ. 200 * 2 = seconds and nanoseconds, truncated to microseconds. 201 * >=3 = seconds and nanoseconds, maximum precision. 202 */ 203 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 204 205 static int timestamp_precision = TSP_SEC; 206 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 207 ×tamp_precision, 0, "Precision of file timestamps"); 208 209 /* 210 * Get a current timestamp. 211 * 212 * MPSAFE 213 */ 214 void 215 vfs_timestamp(struct timespec *tsp) 216 { 217 struct timeval tv; 218 219 switch (timestamp_precision) { 220 case TSP_SEC: 221 tsp->tv_sec = time_second; 222 tsp->tv_nsec = 0; 223 break; 224 case TSP_HZ: 225 getnanotime(tsp); 226 break; 227 case TSP_USEC: 228 microtime(&tv); 229 TIMEVAL_TO_TIMESPEC(&tv, tsp); 230 break; 231 case TSP_NSEC: 232 default: 233 nanotime(tsp); 234 break; 235 } 236 } 237 238 /* 239 * Set vnode attributes to VNOVAL 240 */ 241 void 242 vattr_null(struct vattr *vap) 243 { 244 vap->va_type = VNON; 245 vap->va_size = VNOVAL; 246 vap->va_bytes = VNOVAL; 247 vap->va_mode = VNOVAL; 248 vap->va_nlink = VNOVAL; 249 vap->va_uid = VNOVAL; 250 vap->va_gid = VNOVAL; 251 vap->va_fsid = VNOVAL; 252 vap->va_fileid = VNOVAL; 253 vap->va_blocksize = VNOVAL; 254 vap->va_rmajor = VNOVAL; 255 vap->va_rminor = VNOVAL; 256 vap->va_atime.tv_sec = VNOVAL; 257 vap->va_atime.tv_nsec = VNOVAL; 258 vap->va_mtime.tv_sec = VNOVAL; 259 vap->va_mtime.tv_nsec = VNOVAL; 260 vap->va_ctime.tv_sec = VNOVAL; 261 vap->va_ctime.tv_nsec = VNOVAL; 262 vap->va_flags = VNOVAL; 263 vap->va_gen = VNOVAL; 264 vap->va_vaflags = 0; 265 /* va_*_uuid fields are only valid if related flags are set */ 266 } 267 268 /* 269 * Flush out and invalidate all buffers associated with a vnode. 270 * 271 * vp must be locked. 272 */ 273 static int vinvalbuf_bp(struct buf *bp, void *data); 274 275 struct vinvalbuf_bp_info { 276 struct vnode *vp; 277 int slptimeo; 278 int lkflags; 279 int flags; 280 int clean; 281 }; 282 283 int 284 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 285 { 286 struct vinvalbuf_bp_info info; 287 vm_object_t object; 288 int error; 289 290 lwkt_gettoken(&vp->v_token); 291 292 /* 293 * If we are being asked to save, call fsync to ensure that the inode 294 * is updated. 295 */ 296 if (flags & V_SAVE) { 297 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 298 if (error) 299 goto done; 300 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 301 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 302 goto done; 303 #if 0 304 /* 305 * Dirty bufs may be left or generated via races 306 * in circumstances where vinvalbuf() is called on 307 * a vnode not undergoing reclamation. Only 308 * panic if we are trying to reclaim the vnode. 309 */ 310 if ((vp->v_flag & VRECLAIMED) && 311 (bio_track_active(&vp->v_track_write) || 312 !RB_EMPTY(&vp->v_rbdirty_tree))) { 313 panic("vinvalbuf: dirty bufs"); 314 } 315 #endif 316 } 317 } 318 info.slptimeo = slptimeo; 319 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 320 if (slpflag & PCATCH) 321 info.lkflags |= LK_PCATCH; 322 info.flags = flags; 323 info.vp = vp; 324 325 /* 326 * Flush the buffer cache until nothing is left, wait for all I/O 327 * to complete. At least one pass is required. We might block 328 * in the pip code so we have to re-check. Order is important. 329 */ 330 do { 331 /* 332 * Flush buffer cache 333 */ 334 if (!RB_EMPTY(&vp->v_rbclean_tree)) { 335 info.clean = 1; 336 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 337 NULL, vinvalbuf_bp, &info); 338 } 339 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 340 info.clean = 0; 341 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 342 NULL, vinvalbuf_bp, &info); 343 } 344 345 /* 346 * Wait for I/O completion. 347 */ 348 bio_track_wait(&vp->v_track_write, 0, 0); 349 if ((object = vp->v_object) != NULL) 350 refcount_wait(&object->paging_in_progress, "vnvlbx"); 351 } while (bio_track_active(&vp->v_track_write) || 352 !RB_EMPTY(&vp->v_rbclean_tree) || 353 !RB_EMPTY(&vp->v_rbdirty_tree)); 354 355 /* 356 * Destroy the copy in the VM cache, too. 357 */ 358 if ((object = vp->v_object) != NULL) { 359 vm_object_page_remove(object, 0, 0, 360 (flags & V_SAVE) ? TRUE : FALSE); 361 } 362 363 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 364 panic("vinvalbuf: flush failed"); 365 if (!RB_EMPTY(&vp->v_rbhash_tree)) 366 panic("vinvalbuf: flush failed, buffers still present"); 367 error = 0; 368 done: 369 lwkt_reltoken(&vp->v_token); 370 return (error); 371 } 372 373 static int 374 vinvalbuf_bp(struct buf *bp, void *data) 375 { 376 struct vinvalbuf_bp_info *info = data; 377 int error; 378 379 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 380 atomic_add_int(&bp->b_refs, 1); 381 error = BUF_TIMELOCK(bp, info->lkflags, 382 "vinvalbuf", info->slptimeo); 383 atomic_subtract_int(&bp->b_refs, 1); 384 if (error == 0) { 385 BUF_UNLOCK(bp); 386 error = ENOLCK; 387 } 388 if (error == ENOLCK) 389 return(0); 390 return (-error); 391 } 392 KKASSERT(bp->b_vp == info->vp); 393 394 /* 395 * Must check clean/dirty status after successfully locking as 396 * it may race. 397 */ 398 if ((info->clean && (bp->b_flags & B_DELWRI)) || 399 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 400 BUF_UNLOCK(bp); 401 return(0); 402 } 403 404 /* 405 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 406 * check. This code will write out the buffer, period. 407 */ 408 bremfree(bp); 409 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 410 (info->flags & V_SAVE)) { 411 cluster_awrite(bp); 412 } else if (info->flags & V_SAVE) { 413 /* 414 * Cannot set B_NOCACHE on a clean buffer as this will 415 * destroy the VM backing store which might actually 416 * be dirty (and unsynchronized). 417 */ 418 bp->b_flags |= (B_INVAL | B_RELBUF); 419 brelse(bp); 420 } else { 421 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 422 brelse(bp); 423 } 424 return(0); 425 } 426 427 /* 428 * Truncate a file's buffer and pages to a specified length. This 429 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 430 * sync activity. 431 * 432 * The vnode must be locked. 433 */ 434 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 435 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 436 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 437 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 438 439 struct vtruncbuf_info { 440 struct vnode *vp; 441 off_t truncloffset; 442 int clean; 443 }; 444 445 int 446 vtruncbuf(struct vnode *vp, off_t length, int blksize) 447 { 448 struct vtruncbuf_info info; 449 const char *filename; 450 int count; 451 452 /* 453 * Round up to the *next* block, then destroy the buffers in question. 454 * Since we are only removing some of the buffers we must rely on the 455 * scan count to determine whether a loop is necessary. 456 */ 457 if ((count = (int)(length % blksize)) != 0) 458 info.truncloffset = length + (blksize - count); 459 else 460 info.truncloffset = length; 461 info.vp = vp; 462 463 lwkt_gettoken(&vp->v_token); 464 do { 465 info.clean = 1; 466 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 467 vtruncbuf_bp_trunc_cmp, 468 vtruncbuf_bp_trunc, &info); 469 info.clean = 0; 470 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 471 vtruncbuf_bp_trunc_cmp, 472 vtruncbuf_bp_trunc, &info); 473 } while(count); 474 475 /* 476 * For safety, fsync any remaining metadata if the file is not being 477 * truncated to 0. Since the metadata does not represent the entire 478 * dirty list we have to rely on the hit count to ensure that we get 479 * all of it. 480 */ 481 if (length > 0) { 482 do { 483 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 484 vtruncbuf_bp_metasync_cmp, 485 vtruncbuf_bp_metasync, &info); 486 } while (count); 487 } 488 489 /* 490 * Clean out any left over VM backing store. 491 * 492 * It is possible to have in-progress I/O from buffers that were 493 * not part of the truncation. This should not happen if we 494 * are truncating to 0-length. 495 */ 496 vnode_pager_setsize(vp, length); 497 bio_track_wait(&vp->v_track_write, 0, 0); 498 499 /* 500 * Debugging only 501 */ 502 spin_lock(&vp->v_spin); 503 filename = TAILQ_FIRST(&vp->v_namecache) ? 504 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 505 spin_unlock(&vp->v_spin); 506 507 /* 508 * Make sure no buffers were instantiated while we were trying 509 * to clean out the remaining VM pages. This could occur due 510 * to busy dirty VM pages being flushed out to disk. 511 */ 512 do { 513 info.clean = 1; 514 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 515 vtruncbuf_bp_trunc_cmp, 516 vtruncbuf_bp_trunc, &info); 517 info.clean = 0; 518 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 519 vtruncbuf_bp_trunc_cmp, 520 vtruncbuf_bp_trunc, &info); 521 if (count) { 522 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 523 "left over buffers in %s\n", count, filename); 524 } 525 } while(count); 526 527 lwkt_reltoken(&vp->v_token); 528 529 return (0); 530 } 531 532 /* 533 * The callback buffer is beyond the new file EOF and must be destroyed. 534 * Note that the compare function must conform to the RB_SCAN's requirements. 535 */ 536 static 537 int 538 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 539 { 540 struct vtruncbuf_info *info = data; 541 542 if (bp->b_loffset >= info->truncloffset) 543 return(0); 544 return(-1); 545 } 546 547 static 548 int 549 vtruncbuf_bp_trunc(struct buf *bp, void *data) 550 { 551 struct vtruncbuf_info *info = data; 552 553 /* 554 * Do not try to use a buffer we cannot immediately lock, but sleep 555 * anyway to prevent a livelock. The code will loop until all buffers 556 * can be acted upon. 557 * 558 * We must always revalidate the buffer after locking it to deal 559 * with MP races. 560 */ 561 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 562 atomic_add_int(&bp->b_refs, 1); 563 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 564 BUF_UNLOCK(bp); 565 atomic_subtract_int(&bp->b_refs, 1); 566 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 567 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 568 bp->b_vp != info->vp || 569 vtruncbuf_bp_trunc_cmp(bp, data)) { 570 BUF_UNLOCK(bp); 571 } else { 572 bremfree(bp); 573 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 574 brelse(bp); 575 } 576 return(1); 577 } 578 579 /* 580 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 581 * blocks (with a negative loffset) are scanned. 582 * Note that the compare function must conform to the RB_SCAN's requirements. 583 */ 584 static int 585 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 586 { 587 if (bp->b_loffset < 0) 588 return(0); 589 return(1); 590 } 591 592 static int 593 vtruncbuf_bp_metasync(struct buf *bp, void *data) 594 { 595 struct vtruncbuf_info *info = data; 596 597 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 598 atomic_add_int(&bp->b_refs, 1); 599 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 600 BUF_UNLOCK(bp); 601 atomic_subtract_int(&bp->b_refs, 1); 602 } else if ((bp->b_flags & B_DELWRI) == 0 || 603 bp->b_vp != info->vp || 604 vtruncbuf_bp_metasync_cmp(bp, data)) { 605 BUF_UNLOCK(bp); 606 } else { 607 bremfree(bp); 608 if (bp->b_vp == info->vp) 609 bawrite(bp); 610 else 611 bwrite(bp); 612 } 613 return(1); 614 } 615 616 /* 617 * vfsync - implements a multipass fsync on a file which understands 618 * dependancies and meta-data. The passed vnode must be locked. The 619 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 620 * 621 * When fsyncing data asynchronously just do one consolidated pass starting 622 * with the most negative block number. This may not get all the data due 623 * to dependancies. 624 * 625 * When fsyncing data synchronously do a data pass, then a metadata pass, 626 * then do additional data+metadata passes to try to get all the data out. 627 */ 628 static int vfsync_wait_output(struct vnode *vp, 629 int (*waitoutput)(struct vnode *, struct thread *)); 630 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 631 static int vfsync_data_only_cmp(struct buf *bp, void *data); 632 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 633 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 634 static int vfsync_bp(struct buf *bp, void *data); 635 636 struct vfsync_info { 637 struct vnode *vp; 638 int synchronous; 639 int syncdeps; 640 int lazycount; 641 int lazylimit; 642 int skippedbufs; 643 int (*checkdef)(struct buf *); 644 int (*cmpfunc)(struct buf *, void *); 645 }; 646 647 int 648 vfsync(struct vnode *vp, int waitfor, int passes, 649 int (*checkdef)(struct buf *), 650 int (*waitoutput)(struct vnode *, struct thread *)) 651 { 652 struct vfsync_info info; 653 int error; 654 655 bzero(&info, sizeof(info)); 656 info.vp = vp; 657 if ((info.checkdef = checkdef) == NULL) 658 info.syncdeps = 1; 659 660 lwkt_gettoken(&vp->v_token); 661 662 switch(waitfor) { 663 case MNT_LAZY | MNT_NOWAIT: 664 case MNT_LAZY: 665 /* 666 * Lazy (filesystem syncer typ) Asynchronous plus limit the 667 * number of data (not meta) pages we try to flush to 1MB. 668 * A non-zero return means that lazy limit was reached. 669 */ 670 info.lazylimit = 1024 * 1024; 671 info.syncdeps = 1; 672 info.cmpfunc = vfsync_lazy_range_cmp; 673 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 674 vfsync_lazy_range_cmp, vfsync_bp, &info); 675 info.cmpfunc = vfsync_meta_only_cmp; 676 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 677 vfsync_meta_only_cmp, vfsync_bp, &info); 678 if (error == 0) 679 vp->v_lazyw = 0; 680 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 681 vn_syncer_add(vp, 1); 682 error = 0; 683 break; 684 case MNT_NOWAIT: 685 /* 686 * Asynchronous. Do a data-only pass and a meta-only pass. 687 */ 688 info.syncdeps = 1; 689 info.cmpfunc = vfsync_data_only_cmp; 690 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 691 vfsync_bp, &info); 692 info.cmpfunc = vfsync_meta_only_cmp; 693 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 694 vfsync_bp, &info); 695 error = 0; 696 break; 697 default: 698 /* 699 * Synchronous. Do a data-only pass, then a meta-data+data 700 * pass, then additional integrated passes to try to get 701 * all the dependancies flushed. 702 */ 703 info.cmpfunc = vfsync_data_only_cmp; 704 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 705 vfsync_bp, &info); 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 "bufs in pass2!\n", info.skippedbufs); 716 } 717 } 718 while (error == 0 && passes > 0 && 719 !RB_EMPTY(&vp->v_rbdirty_tree) 720 ) { 721 if (--passes == 0) { 722 info.synchronous = 1; 723 info.syncdeps = 1; 724 } 725 info.cmpfunc = vfsync_dummy_cmp; 726 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 727 vfsync_bp, &info); 728 if (error < 0) 729 error = -error; 730 info.syncdeps = 1; 731 if (error == 0) 732 error = vfsync_wait_output(vp, waitoutput); 733 } 734 break; 735 } 736 lwkt_reltoken(&vp->v_token); 737 return(error); 738 } 739 740 static int 741 vfsync_wait_output(struct vnode *vp, 742 int (*waitoutput)(struct vnode *, struct thread *)) 743 { 744 int error; 745 746 error = bio_track_wait(&vp->v_track_write, 0, 0); 747 if (waitoutput) 748 error = waitoutput(vp, curthread); 749 return(error); 750 } 751 752 static int 753 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 754 { 755 return(0); 756 } 757 758 static int 759 vfsync_data_only_cmp(struct buf *bp, void *data) 760 { 761 if (bp->b_loffset < 0) 762 return(-1); 763 return(0); 764 } 765 766 static int 767 vfsync_meta_only_cmp(struct buf *bp, void *data) 768 { 769 if (bp->b_loffset < 0) 770 return(0); 771 return(1); 772 } 773 774 static int 775 vfsync_lazy_range_cmp(struct buf *bp, void *data) 776 { 777 struct vfsync_info *info = data; 778 779 if (bp->b_loffset < info->vp->v_lazyw) 780 return(-1); 781 return(0); 782 } 783 784 static int 785 vfsync_bp(struct buf *bp, void *data) 786 { 787 struct vfsync_info *info = data; 788 struct vnode *vp = info->vp; 789 int error; 790 791 /* 792 * Ignore buffers that we cannot immediately lock. 793 */ 794 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 795 ++info->skippedbufs; 796 return(0); 797 } 798 799 /* 800 * We must revalidate the buffer after locking. 801 */ 802 if ((bp->b_flags & B_DELWRI) == 0 || 803 bp->b_vp != info->vp || 804 info->cmpfunc(bp, data)) { 805 BUF_UNLOCK(bp); 806 return(0); 807 } 808 809 /* 810 * If syncdeps is not set we do not try to write buffers which have 811 * dependancies. 812 */ 813 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 814 BUF_UNLOCK(bp); 815 return(0); 816 } 817 818 /* 819 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 820 * has been written but an additional handshake with the device 821 * is required before we can dispose of the buffer. We have no idea 822 * how to do this so we have to skip these buffers. 823 */ 824 if (bp->b_flags & B_NEEDCOMMIT) { 825 BUF_UNLOCK(bp); 826 return(0); 827 } 828 829 /* 830 * Ask bioops if it is ok to sync. If not the VFS may have 831 * set B_LOCKED so we have to cycle the buffer. 832 */ 833 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 834 bremfree(bp); 835 brelse(bp); 836 return(0); 837 } 838 839 if (info->synchronous) { 840 /* 841 * Synchronous flushing. An error may be returned. 842 */ 843 bremfree(bp); 844 error = bwrite(bp); 845 } else { 846 /* 847 * Asynchronous flushing. A negative return value simply 848 * stops the scan and is not considered an error. We use 849 * this to support limited MNT_LAZY flushes. 850 */ 851 vp->v_lazyw = bp->b_loffset; 852 bremfree(bp); 853 info->lazycount += cluster_awrite(bp); 854 waitrunningbufspace(); 855 vm_wait_nominal(); 856 if (info->lazylimit && info->lazycount >= info->lazylimit) 857 error = 1; 858 else 859 error = 0; 860 } 861 return(-error); 862 } 863 864 /* 865 * Associate a buffer with a vnode. 866 * 867 * MPSAFE 868 */ 869 int 870 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 871 { 872 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 873 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 874 875 /* 876 * Insert onto list for new vnode. 877 */ 878 lwkt_gettoken(&vp->v_token); 879 880 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 881 lwkt_reltoken(&vp->v_token); 882 return (EEXIST); 883 } 884 885 /* 886 * Diagnostics (mainly for HAMMER debugging). Check for 887 * overlapping buffers. 888 */ 889 if (check_buf_overlap) { 890 struct buf *bx; 891 bx = buf_rb_hash_RB_PREV(bp); 892 if (bx) { 893 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 894 kprintf("bgetvp: overlapl %016jx/%d %016jx " 895 "bx %p bp %p\n", 896 (intmax_t)bx->b_loffset, 897 bx->b_bufsize, 898 (intmax_t)bp->b_loffset, 899 bx, bp); 900 if (check_buf_overlap > 1) 901 panic("bgetvp - overlapping buffer"); 902 } 903 } 904 bx = buf_rb_hash_RB_NEXT(bp); 905 if (bx) { 906 if (bp->b_loffset + testsize > bx->b_loffset) { 907 kprintf("bgetvp: overlapr %016jx/%d %016jx " 908 "bp %p bx %p\n", 909 (intmax_t)bp->b_loffset, 910 testsize, 911 (intmax_t)bx->b_loffset, 912 bp, bx); 913 if (check_buf_overlap > 1) 914 panic("bgetvp - overlapping buffer"); 915 } 916 } 917 } 918 bp->b_vp = vp; 919 bp->b_flags |= B_HASHED; 920 bp->b_flags |= B_VNCLEAN; 921 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 922 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 923 /*vhold(vp);*/ 924 lwkt_reltoken(&vp->v_token); 925 return(0); 926 } 927 928 /* 929 * Disassociate a buffer from a vnode. 930 * 931 * MPSAFE 932 */ 933 void 934 brelvp(struct buf *bp) 935 { 936 struct vnode *vp; 937 938 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 939 940 /* 941 * Delete from old vnode list, if on one. 942 */ 943 vp = bp->b_vp; 944 lwkt_gettoken(&vp->v_token); 945 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 946 if (bp->b_flags & B_VNDIRTY) 947 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 948 else 949 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 950 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 951 } 952 if (bp->b_flags & B_HASHED) { 953 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 954 bp->b_flags &= ~B_HASHED; 955 } 956 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) 957 vn_syncer_remove(vp); 958 bp->b_vp = NULL; 959 960 lwkt_reltoken(&vp->v_token); 961 962 /*vdrop(vp);*/ 963 } 964 965 /* 966 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 967 * This routine is called when the state of the B_DELWRI bit is changed. 968 * 969 * Must be called with vp->v_token held. 970 * MPSAFE 971 */ 972 void 973 reassignbuf(struct buf *bp) 974 { 975 struct vnode *vp = bp->b_vp; 976 int delay; 977 978 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 979 ++reassignbufcalls; 980 981 /* 982 * B_PAGING flagged buffers cannot be reassigned because their vp 983 * is not fully linked in. 984 */ 985 if (bp->b_flags & B_PAGING) 986 panic("cannot reassign paging buffer"); 987 988 if (bp->b_flags & B_DELWRI) { 989 /* 990 * Move to the dirty list, add the vnode to the worklist 991 */ 992 if (bp->b_flags & B_VNCLEAN) { 993 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 994 bp->b_flags &= ~B_VNCLEAN; 995 } 996 if ((bp->b_flags & B_VNDIRTY) == 0) { 997 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 998 panic("reassignbuf: dup lblk vp %p bp %p", 999 vp, bp); 1000 } 1001 bp->b_flags |= B_VNDIRTY; 1002 } 1003 if ((vp->v_flag & VONWORKLST) == 0) { 1004 switch (vp->v_type) { 1005 case VDIR: 1006 delay = dirdelay; 1007 break; 1008 case VCHR: 1009 case VBLK: 1010 if (vp->v_rdev && 1011 vp->v_rdev->si_mountpoint != NULL) { 1012 delay = metadelay; 1013 break; 1014 } 1015 /* fall through */ 1016 default: 1017 delay = filedelay; 1018 } 1019 vn_syncer_add(vp, delay); 1020 } 1021 } else { 1022 /* 1023 * Move to the clean list, remove the vnode from the worklist 1024 * if no dirty blocks remain. 1025 */ 1026 if (bp->b_flags & B_VNDIRTY) { 1027 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1028 bp->b_flags &= ~B_VNDIRTY; 1029 } 1030 if ((bp->b_flags & B_VNCLEAN) == 0) { 1031 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1032 panic("reassignbuf: dup lblk vp %p bp %p", 1033 vp, bp); 1034 } 1035 bp->b_flags |= B_VNCLEAN; 1036 } 1037 if ((vp->v_flag & VONWORKLST) && 1038 RB_EMPTY(&vp->v_rbdirty_tree)) { 1039 vn_syncer_remove(vp); 1040 } 1041 } 1042 } 1043 1044 /* 1045 * Create a vnode for a block device. Used for mounting the root file 1046 * system. 1047 * 1048 * A vref()'d vnode is returned. 1049 */ 1050 extern struct vop_ops *devfs_vnode_dev_vops_p; 1051 int 1052 bdevvp(cdev_t dev, struct vnode **vpp) 1053 { 1054 struct vnode *vp; 1055 struct vnode *nvp; 1056 int error; 1057 1058 if (dev == NULL) { 1059 *vpp = NULLVP; 1060 return (ENXIO); 1061 } 1062 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1063 &nvp, 0, 0); 1064 if (error) { 1065 *vpp = NULLVP; 1066 return (error); 1067 } 1068 vp = nvp; 1069 vp->v_type = VCHR; 1070 #if 0 1071 vp->v_rdev = dev; 1072 #endif 1073 v_associate_rdev(vp, dev); 1074 vp->v_umajor = dev->si_umajor; 1075 vp->v_uminor = dev->si_uminor; 1076 vx_unlock(vp); 1077 *vpp = vp; 1078 return (0); 1079 } 1080 1081 int 1082 v_associate_rdev(struct vnode *vp, cdev_t dev) 1083 { 1084 if (dev == NULL) 1085 return(ENXIO); 1086 if (dev_is_good(dev) == 0) 1087 return(ENXIO); 1088 KKASSERT(vp->v_rdev == NULL); 1089 vp->v_rdev = reference_dev(dev); 1090 lwkt_gettoken(&spechash_token); 1091 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1092 lwkt_reltoken(&spechash_token); 1093 return(0); 1094 } 1095 1096 void 1097 v_release_rdev(struct vnode *vp) 1098 { 1099 cdev_t dev; 1100 1101 if ((dev = vp->v_rdev) != NULL) { 1102 lwkt_gettoken(&spechash_token); 1103 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1104 vp->v_rdev = NULL; 1105 release_dev(dev); 1106 lwkt_reltoken(&spechash_token); 1107 } 1108 } 1109 1110 /* 1111 * Add a vnode to the alias list hung off the cdev_t. We only associate 1112 * the device number with the vnode. The actual device is not associated 1113 * until the vnode is opened (usually in spec_open()), and will be 1114 * disassociated on last close. 1115 */ 1116 void 1117 addaliasu(struct vnode *nvp, int x, int y) 1118 { 1119 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1120 panic("addaliasu on non-special vnode"); 1121 nvp->v_umajor = x; 1122 nvp->v_uminor = y; 1123 } 1124 1125 /* 1126 * Simple call that a filesystem can make to try to get rid of a 1127 * vnode. It will fail if anyone is referencing the vnode (including 1128 * the caller). 1129 * 1130 * The filesystem can check whether its in-memory inode structure still 1131 * references the vp on return. 1132 */ 1133 void 1134 vclean_unlocked(struct vnode *vp) 1135 { 1136 vx_get(vp); 1137 if (sysref_isactive(&vp->v_sysref) == 0) 1138 vgone_vxlocked(vp); 1139 vx_put(vp); 1140 } 1141 1142 /* 1143 * Disassociate a vnode from its underlying filesystem. 1144 * 1145 * The vnode must be VX locked and referenced. In all normal situations 1146 * there are no active references. If vclean_vxlocked() is called while 1147 * there are active references, the vnode is being ripped out and we have 1148 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1149 */ 1150 void 1151 vclean_vxlocked(struct vnode *vp, int flags) 1152 { 1153 int active; 1154 int n; 1155 vm_object_t object; 1156 struct namecache *ncp; 1157 1158 /* 1159 * If the vnode has already been reclaimed we have nothing to do. 1160 */ 1161 if (vp->v_flag & VRECLAIMED) 1162 return; 1163 vsetflags(vp, VRECLAIMED); 1164 1165 if (verbose_reclaims) { 1166 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1167 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1168 } 1169 1170 /* 1171 * Scrap the vfs cache 1172 */ 1173 while (cache_inval_vp(vp, 0) != 0) { 1174 kprintf("Warning: vnode %p clean/cache_resolution " 1175 "race detected\n", vp); 1176 tsleep(vp, 0, "vclninv", 2); 1177 } 1178 1179 /* 1180 * Check to see if the vnode is in use. If so we have to reference it 1181 * before we clean it out so that its count cannot fall to zero and 1182 * generate a race against ourselves to recycle it. 1183 */ 1184 active = sysref_isactive(&vp->v_sysref); 1185 1186 /* 1187 * Clean out any buffers associated with the vnode and destroy its 1188 * object, if it has one. 1189 */ 1190 vinvalbuf(vp, V_SAVE, 0, 0); 1191 1192 /* 1193 * If purging an active vnode (typically during a forced unmount 1194 * or reboot), it must be closed and deactivated before being 1195 * reclaimed. This isn't really all that safe, but what can 1196 * we do? XXX. 1197 * 1198 * Note that neither of these routines unlocks the vnode. 1199 */ 1200 if (active && (flags & DOCLOSE)) { 1201 while ((n = vp->v_opencount) != 0) { 1202 if (vp->v_writecount) 1203 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1204 else 1205 VOP_CLOSE(vp, FNONBLOCK); 1206 if (vp->v_opencount == n) { 1207 kprintf("Warning: unable to force-close" 1208 " vnode %p\n", vp); 1209 break; 1210 } 1211 } 1212 } 1213 1214 /* 1215 * If the vnode has not been deactivated, deactivated it. Deactivation 1216 * can create new buffers and VM pages so we have to call vinvalbuf() 1217 * again to make sure they all get flushed. 1218 * 1219 * This can occur if a file with a link count of 0 needs to be 1220 * truncated. 1221 * 1222 * If the vnode is already dead don't try to deactivate it. 1223 */ 1224 if ((vp->v_flag & VINACTIVE) == 0) { 1225 vsetflags(vp, VINACTIVE); 1226 if (vp->v_mount) 1227 VOP_INACTIVE(vp); 1228 vinvalbuf(vp, V_SAVE, 0, 0); 1229 } 1230 1231 /* 1232 * If the vnode has an object, destroy it. 1233 */ 1234 while ((object = vp->v_object) != NULL) { 1235 vm_object_hold(object); 1236 if (object == vp->v_object) 1237 break; 1238 vm_object_drop(object); 1239 } 1240 1241 if (object != NULL) { 1242 if (object->ref_count == 0) { 1243 if ((object->flags & OBJ_DEAD) == 0) 1244 vm_object_terminate(object); 1245 vm_object_drop(object); 1246 vclrflags(vp, VOBJBUF); 1247 } else { 1248 vm_pager_deallocate(object); 1249 vclrflags(vp, VOBJBUF); 1250 vm_object_drop(object); 1251 } 1252 } 1253 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1254 1255 /* 1256 * Reclaim the vnode if not already dead. 1257 */ 1258 if (vp->v_mount && VOP_RECLAIM(vp)) 1259 panic("vclean: cannot reclaim"); 1260 1261 /* 1262 * Done with purge, notify sleepers of the grim news. 1263 */ 1264 vp->v_ops = &dead_vnode_vops_p; 1265 vn_gone(vp); 1266 vp->v_tag = VT_NON; 1267 1268 /* 1269 * If we are destroying an active vnode, reactivate it now that 1270 * we have reassociated it with deadfs. This prevents the system 1271 * from crashing on the vnode due to it being unexpectedly marked 1272 * as inactive or reclaimed. 1273 */ 1274 if (active && (flags & DOCLOSE)) { 1275 vclrflags(vp, VINACTIVE | VRECLAIMED); 1276 } 1277 } 1278 1279 /* 1280 * Eliminate all activity associated with the requested vnode 1281 * and with all vnodes aliased to the requested vnode. 1282 * 1283 * The vnode must be referenced but should not be locked. 1284 */ 1285 int 1286 vrevoke(struct vnode *vp, struct ucred *cred) 1287 { 1288 struct vnode *vq; 1289 struct vnode *vqn; 1290 cdev_t dev; 1291 int error; 1292 1293 /* 1294 * If the vnode has a device association, scrap all vnodes associated 1295 * with the device. Don't let the device disappear on us while we 1296 * are scrapping the vnodes. 1297 * 1298 * The passed vp will probably show up in the list, do not VX lock 1299 * it twice! 1300 * 1301 * Releasing the vnode's rdev here can mess up specfs's call to 1302 * device close, so don't do it. The vnode has been disassociated 1303 * and the device will be closed after the last ref on the related 1304 * fp goes away (if not still open by e.g. the kernel). 1305 */ 1306 if (vp->v_type != VCHR) { 1307 error = fdrevoke(vp, DTYPE_VNODE, cred); 1308 return (error); 1309 } 1310 if ((dev = vp->v_rdev) == NULL) { 1311 return(0); 1312 } 1313 reference_dev(dev); 1314 lwkt_gettoken(&spechash_token); 1315 1316 restart: 1317 vqn = SLIST_FIRST(&dev->si_hlist); 1318 if (vqn) 1319 vhold(vqn); 1320 while ((vq = vqn) != NULL) { 1321 if (sysref_isactive(&vq->v_sysref)) { 1322 vref(vq); 1323 fdrevoke(vq, DTYPE_VNODE, cred); 1324 /*v_release_rdev(vq);*/ 1325 vrele(vq); 1326 if (vq->v_rdev != dev) { 1327 vdrop(vq); 1328 goto restart; 1329 } 1330 } 1331 vqn = SLIST_NEXT(vq, v_cdevnext); 1332 if (vqn) 1333 vhold(vqn); 1334 vdrop(vq); 1335 } 1336 lwkt_reltoken(&spechash_token); 1337 dev_drevoke(dev); 1338 release_dev(dev); 1339 return (0); 1340 } 1341 1342 /* 1343 * This is called when the object underlying a vnode is being destroyed, 1344 * such as in a remove(). Try to recycle the vnode immediately if the 1345 * only active reference is our reference. 1346 * 1347 * Directory vnodes in the namecache with children cannot be immediately 1348 * recycled because numerous VOP_N*() ops require them to be stable. 1349 * 1350 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1351 * function is a NOP if VRECLAIMED is already set. 1352 */ 1353 int 1354 vrecycle(struct vnode *vp) 1355 { 1356 if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1357 if (cache_inval_vp_nonblock(vp)) 1358 return(0); 1359 vgone_vxlocked(vp); 1360 return (1); 1361 } 1362 return (0); 1363 } 1364 1365 /* 1366 * Return the maximum I/O size allowed for strategy calls on VP. 1367 * 1368 * If vp is VCHR or VBLK we dive the device, otherwise we use 1369 * the vp's mount info. 1370 * 1371 * The returned value is clamped at MAXPHYS as most callers cannot use 1372 * buffers larger than that size. 1373 */ 1374 int 1375 vmaxiosize(struct vnode *vp) 1376 { 1377 int maxiosize; 1378 1379 if (vp->v_type == VBLK || vp->v_type == VCHR) 1380 maxiosize = vp->v_rdev->si_iosize_max; 1381 else 1382 maxiosize = vp->v_mount->mnt_iosize_max; 1383 1384 if (maxiosize > MAXPHYS) 1385 maxiosize = MAXPHYS; 1386 return (maxiosize); 1387 } 1388 1389 /* 1390 * Eliminate all activity associated with a vnode in preparation for reuse. 1391 * 1392 * The vnode must be VX locked and refd and will remain VX locked and refd 1393 * on return. This routine may be called with the vnode in any state, as 1394 * long as it is VX locked. The vnode will be cleaned out and marked 1395 * VRECLAIMED but will not actually be reused until all existing refs and 1396 * holds go away. 1397 * 1398 * NOTE: This routine may be called on a vnode which has not yet been 1399 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1400 * already been reclaimed. 1401 * 1402 * This routine is not responsible for placing us back on the freelist. 1403 * Instead, it happens automatically when the caller releases the VX lock 1404 * (assuming there aren't any other references). 1405 */ 1406 void 1407 vgone_vxlocked(struct vnode *vp) 1408 { 1409 /* 1410 * assert that the VX lock is held. This is an absolute requirement 1411 * now for vgone_vxlocked() to be called. 1412 */ 1413 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1414 1415 /* 1416 * Clean out the filesystem specific data and set the VRECLAIMED 1417 * bit. Also deactivate the vnode if necessary. 1418 */ 1419 vclean_vxlocked(vp, DOCLOSE); 1420 1421 /* 1422 * Delete from old mount point vnode list, if on one. 1423 */ 1424 if (vp->v_mount != NULL) { 1425 KKASSERT(vp->v_data == NULL); 1426 insmntque(vp, NULL); 1427 } 1428 1429 /* 1430 * If special device, remove it from special device alias list 1431 * if it is on one. This should normally only occur if a vnode is 1432 * being revoked as the device should otherwise have been released 1433 * naturally. 1434 */ 1435 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1436 v_release_rdev(vp); 1437 } 1438 1439 /* 1440 * Set us to VBAD 1441 */ 1442 vp->v_type = VBAD; 1443 } 1444 1445 /* 1446 * Lookup a vnode by device number. 1447 * 1448 * Returns non-zero and *vpp set to a vref'd vnode on success. 1449 * Returns zero on failure. 1450 */ 1451 int 1452 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1453 { 1454 struct vnode *vp; 1455 1456 lwkt_gettoken(&spechash_token); 1457 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1458 if (type == vp->v_type) { 1459 *vpp = vp; 1460 vref(vp); 1461 lwkt_reltoken(&spechash_token); 1462 return (1); 1463 } 1464 } 1465 lwkt_reltoken(&spechash_token); 1466 return (0); 1467 } 1468 1469 /* 1470 * Calculate the total number of references to a special device. This 1471 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1472 * an overloaded field. Since udev2dev can now return NULL, we have 1473 * to check for a NULL v_rdev. 1474 */ 1475 int 1476 count_dev(cdev_t dev) 1477 { 1478 struct vnode *vp; 1479 int count = 0; 1480 1481 if (SLIST_FIRST(&dev->si_hlist)) { 1482 lwkt_gettoken(&spechash_token); 1483 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1484 count += vp->v_opencount; 1485 } 1486 lwkt_reltoken(&spechash_token); 1487 } 1488 return(count); 1489 } 1490 1491 int 1492 vcount(struct vnode *vp) 1493 { 1494 if (vp->v_rdev == NULL) 1495 return(0); 1496 return(count_dev(vp->v_rdev)); 1497 } 1498 1499 /* 1500 * Initialize VMIO for a vnode. This routine MUST be called before a 1501 * VFS can issue buffer cache ops on a vnode. It is typically called 1502 * when a vnode is initialized from its inode. 1503 */ 1504 int 1505 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1506 { 1507 vm_object_t object; 1508 int error = 0; 1509 1510 retry: 1511 while ((object = vp->v_object) != NULL) { 1512 vm_object_hold(object); 1513 if (object == vp->v_object) 1514 break; 1515 vm_object_drop(object); 1516 } 1517 1518 if (object == NULL) { 1519 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1520 1521 /* 1522 * Dereference the reference we just created. This assumes 1523 * that the object is associated with the vp. 1524 */ 1525 vm_object_hold(object); 1526 object->ref_count--; 1527 vrele(vp); 1528 } else { 1529 if (object->flags & OBJ_DEAD) { 1530 vn_unlock(vp); 1531 if (vp->v_object == object) 1532 vm_object_dead_sleep(object, "vodead"); 1533 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1534 vm_object_drop(object); 1535 goto retry; 1536 } 1537 } 1538 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1539 vsetflags(vp, VOBJBUF); 1540 vm_object_drop(object); 1541 1542 return (error); 1543 } 1544 1545 1546 /* 1547 * Print out a description of a vnode. 1548 */ 1549 static char *typename[] = 1550 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1551 1552 void 1553 vprint(char *label, struct vnode *vp) 1554 { 1555 char buf[96]; 1556 1557 if (label != NULL) 1558 kprintf("%s: %p: ", label, (void *)vp); 1559 else 1560 kprintf("%p: ", (void *)vp); 1561 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1562 typename[vp->v_type], 1563 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1564 buf[0] = '\0'; 1565 if (vp->v_flag & VROOT) 1566 strcat(buf, "|VROOT"); 1567 if (vp->v_flag & VPFSROOT) 1568 strcat(buf, "|VPFSROOT"); 1569 if (vp->v_flag & VTEXT) 1570 strcat(buf, "|VTEXT"); 1571 if (vp->v_flag & VSYSTEM) 1572 strcat(buf, "|VSYSTEM"); 1573 if (vp->v_flag & VFREE) 1574 strcat(buf, "|VFREE"); 1575 if (vp->v_flag & VOBJBUF) 1576 strcat(buf, "|VOBJBUF"); 1577 if (buf[0] != '\0') 1578 kprintf(" flags (%s)", &buf[1]); 1579 if (vp->v_data == NULL) { 1580 kprintf("\n"); 1581 } else { 1582 kprintf("\n\t"); 1583 VOP_PRINT(vp); 1584 } 1585 } 1586 1587 /* 1588 * Do the usual access checking. 1589 * file_mode, uid and gid are from the vnode in question, 1590 * while acc_mode and cred are from the VOP_ACCESS parameter list 1591 */ 1592 int 1593 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1594 mode_t acc_mode, struct ucred *cred) 1595 { 1596 mode_t mask; 1597 int ismember; 1598 1599 /* 1600 * Super-user always gets read/write access, but execute access depends 1601 * on at least one execute bit being set. 1602 */ 1603 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1604 if ((acc_mode & VEXEC) && type != VDIR && 1605 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1606 return (EACCES); 1607 return (0); 1608 } 1609 1610 mask = 0; 1611 1612 /* Otherwise, check the owner. */ 1613 if (cred->cr_uid == uid) { 1614 if (acc_mode & VEXEC) 1615 mask |= S_IXUSR; 1616 if (acc_mode & VREAD) 1617 mask |= S_IRUSR; 1618 if (acc_mode & VWRITE) 1619 mask |= S_IWUSR; 1620 return ((file_mode & mask) == mask ? 0 : EACCES); 1621 } 1622 1623 /* Otherwise, check the groups. */ 1624 ismember = groupmember(gid, cred); 1625 if (cred->cr_svgid == gid || ismember) { 1626 if (acc_mode & VEXEC) 1627 mask |= S_IXGRP; 1628 if (acc_mode & VREAD) 1629 mask |= S_IRGRP; 1630 if (acc_mode & VWRITE) 1631 mask |= S_IWGRP; 1632 return ((file_mode & mask) == mask ? 0 : EACCES); 1633 } 1634 1635 /* Otherwise, check everyone else. */ 1636 if (acc_mode & VEXEC) 1637 mask |= S_IXOTH; 1638 if (acc_mode & VREAD) 1639 mask |= S_IROTH; 1640 if (acc_mode & VWRITE) 1641 mask |= S_IWOTH; 1642 return ((file_mode & mask) == mask ? 0 : EACCES); 1643 } 1644 1645 #ifdef DDB 1646 #include <ddb/ddb.h> 1647 1648 static int db_show_locked_vnodes(struct mount *mp, void *data); 1649 1650 /* 1651 * List all of the locked vnodes in the system. 1652 * Called when debugging the kernel. 1653 */ 1654 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1655 { 1656 kprintf("Locked vnodes\n"); 1657 mountlist_scan(db_show_locked_vnodes, NULL, 1658 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1659 } 1660 1661 static int 1662 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1663 { 1664 struct vnode *vp; 1665 1666 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1667 if (vn_islocked(vp)) 1668 vprint(NULL, vp); 1669 } 1670 return(0); 1671 } 1672 #endif 1673 1674 /* 1675 * Top level filesystem related information gathering. 1676 */ 1677 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1678 1679 static int 1680 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1681 { 1682 int *name = (int *)arg1 - 1; /* XXX */ 1683 u_int namelen = arg2 + 1; /* XXX */ 1684 struct vfsconf *vfsp; 1685 int maxtypenum; 1686 1687 #if 1 || defined(COMPAT_PRELITE2) 1688 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1689 if (namelen == 1) 1690 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1691 #endif 1692 1693 #ifdef notyet 1694 /* all sysctl names at this level are at least name and field */ 1695 if (namelen < 2) 1696 return (ENOTDIR); /* overloaded */ 1697 if (name[0] != VFS_GENERIC) { 1698 vfsp = vfsconf_find_by_typenum(name[0]); 1699 if (vfsp == NULL) 1700 return (EOPNOTSUPP); 1701 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1702 oldp, oldlenp, newp, newlen, p)); 1703 } 1704 #endif 1705 switch (name[1]) { 1706 case VFS_MAXTYPENUM: 1707 if (namelen != 2) 1708 return (ENOTDIR); 1709 maxtypenum = vfsconf_get_maxtypenum(); 1710 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1711 case VFS_CONF: 1712 if (namelen != 3) 1713 return (ENOTDIR); /* overloaded */ 1714 vfsp = vfsconf_find_by_typenum(name[2]); 1715 if (vfsp == NULL) 1716 return (EOPNOTSUPP); 1717 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1718 } 1719 return (EOPNOTSUPP); 1720 } 1721 1722 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1723 "Generic filesystem"); 1724 1725 #if 1 || defined(COMPAT_PRELITE2) 1726 1727 static int 1728 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1729 { 1730 int error; 1731 struct ovfsconf ovfs; 1732 struct sysctl_req *req = (struct sysctl_req*) data; 1733 1734 bzero(&ovfs, sizeof(ovfs)); 1735 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1736 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1737 ovfs.vfc_index = vfsp->vfc_typenum; 1738 ovfs.vfc_refcount = vfsp->vfc_refcount; 1739 ovfs.vfc_flags = vfsp->vfc_flags; 1740 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1741 if (error) 1742 return error; /* abort iteration with error code */ 1743 else 1744 return 0; /* continue iterating with next element */ 1745 } 1746 1747 static int 1748 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1749 { 1750 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1751 } 1752 1753 #endif /* 1 || COMPAT_PRELITE2 */ 1754 1755 /* 1756 * Check to see if a filesystem is mounted on a block device. 1757 */ 1758 int 1759 vfs_mountedon(struct vnode *vp) 1760 { 1761 cdev_t dev; 1762 1763 if ((dev = vp->v_rdev) == NULL) { 1764 /* if (vp->v_type != VBLK) 1765 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1766 } 1767 if (dev != NULL && dev->si_mountpoint) 1768 return (EBUSY); 1769 return (0); 1770 } 1771 1772 /* 1773 * Unmount all filesystems. The list is traversed in reverse order 1774 * of mounting to avoid dependencies. 1775 */ 1776 1777 static int vfs_umountall_callback(struct mount *mp, void *data); 1778 1779 void 1780 vfs_unmountall(void) 1781 { 1782 int count; 1783 1784 do { 1785 count = mountlist_scan(vfs_umountall_callback, 1786 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1787 } while (count); 1788 } 1789 1790 static 1791 int 1792 vfs_umountall_callback(struct mount *mp, void *data) 1793 { 1794 int error; 1795 1796 error = dounmount(mp, MNT_FORCE); 1797 if (error) { 1798 mountlist_remove(mp); 1799 kprintf("unmount of filesystem mounted from %s failed (", 1800 mp->mnt_stat.f_mntfromname); 1801 if (error == EBUSY) 1802 kprintf("BUSY)\n"); 1803 else 1804 kprintf("%d)\n", error); 1805 } 1806 return(1); 1807 } 1808 1809 /* 1810 * Checks the mount flags for parameter mp and put the names comma-separated 1811 * into a string buffer buf with a size limit specified by len. 1812 * 1813 * It returns the number of bytes written into buf, and (*errorp) will be 1814 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1815 * not large enough). The buffer will be 0-terminated if len was not 0. 1816 */ 1817 size_t 1818 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1819 char *buf, size_t len, int *errorp) 1820 { 1821 static const struct mountctl_opt optnames[] = { 1822 { MNT_ASYNC, "asynchronous" }, 1823 { MNT_EXPORTED, "NFS exported" }, 1824 { MNT_LOCAL, "local" }, 1825 { MNT_NOATIME, "noatime" }, 1826 { MNT_NODEV, "nodev" }, 1827 { MNT_NOEXEC, "noexec" }, 1828 { MNT_NOSUID, "nosuid" }, 1829 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1830 { MNT_QUOTA, "with-quotas" }, 1831 { MNT_RDONLY, "read-only" }, 1832 { MNT_SYNCHRONOUS, "synchronous" }, 1833 { MNT_UNION, "union" }, 1834 { MNT_NOCLUSTERR, "noclusterr" }, 1835 { MNT_NOCLUSTERW, "noclusterw" }, 1836 { MNT_SUIDDIR, "suiddir" }, 1837 { MNT_SOFTDEP, "soft-updates" }, 1838 { MNT_IGNORE, "ignore" }, 1839 { 0, NULL} 1840 }; 1841 int bwritten; 1842 int bleft; 1843 int optlen; 1844 int actsize; 1845 1846 *errorp = 0; 1847 bwritten = 0; 1848 bleft = len - 1; /* leave room for trailing \0 */ 1849 1850 /* 1851 * Checks the size of the string. If it contains 1852 * any data, then we will append the new flags to 1853 * it. 1854 */ 1855 actsize = strlen(buf); 1856 if (actsize > 0) 1857 buf += actsize; 1858 1859 /* Default flags if no flags passed */ 1860 if (optp == NULL) 1861 optp = optnames; 1862 1863 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1864 *errorp = EINVAL; 1865 return(0); 1866 } 1867 1868 for (; flags && optp->o_opt; ++optp) { 1869 if ((flags & optp->o_opt) == 0) 1870 continue; 1871 optlen = strlen(optp->o_name); 1872 if (bwritten || actsize > 0) { 1873 if (bleft < 2) { 1874 *errorp = ENOSPC; 1875 break; 1876 } 1877 buf[bwritten++] = ','; 1878 buf[bwritten++] = ' '; 1879 bleft -= 2; 1880 } 1881 if (bleft < optlen) { 1882 *errorp = ENOSPC; 1883 break; 1884 } 1885 bcopy(optp->o_name, buf + bwritten, optlen); 1886 bwritten += optlen; 1887 bleft -= optlen; 1888 flags &= ~optp->o_opt; 1889 } 1890 1891 /* 1892 * Space already reserved for trailing \0 1893 */ 1894 buf[bwritten] = 0; 1895 return (bwritten); 1896 } 1897 1898 /* 1899 * Build hash lists of net addresses and hang them off the mount point. 1900 * Called by ufs_mount() to set up the lists of export addresses. 1901 */ 1902 static int 1903 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1904 const struct export_args *argp) 1905 { 1906 struct netcred *np; 1907 struct radix_node_head *rnh; 1908 int i; 1909 struct radix_node *rn; 1910 struct sockaddr *saddr, *smask = NULL; 1911 struct domain *dom; 1912 int error; 1913 1914 if (argp->ex_addrlen == 0) { 1915 if (mp->mnt_flag & MNT_DEFEXPORTED) 1916 return (EPERM); 1917 np = &nep->ne_defexported; 1918 np->netc_exflags = argp->ex_flags; 1919 np->netc_anon = argp->ex_anon; 1920 np->netc_anon.cr_ref = 1; 1921 mp->mnt_flag |= MNT_DEFEXPORTED; 1922 return (0); 1923 } 1924 1925 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1926 return (EINVAL); 1927 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1928 return (EINVAL); 1929 1930 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1931 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1932 saddr = (struct sockaddr *) (np + 1); 1933 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1934 goto out; 1935 if (saddr->sa_len > argp->ex_addrlen) 1936 saddr->sa_len = argp->ex_addrlen; 1937 if (argp->ex_masklen) { 1938 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1939 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1940 if (error) 1941 goto out; 1942 if (smask->sa_len > argp->ex_masklen) 1943 smask->sa_len = argp->ex_masklen; 1944 } 1945 i = saddr->sa_family; 1946 if ((rnh = nep->ne_rtable[i]) == NULL) { 1947 /* 1948 * Seems silly to initialize every AF when most are not used, 1949 * do so on demand here 1950 */ 1951 SLIST_FOREACH(dom, &domains, dom_next) 1952 if (dom->dom_family == i && dom->dom_rtattach) { 1953 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1954 dom->dom_rtoffset); 1955 break; 1956 } 1957 if ((rnh = nep->ne_rtable[i]) == NULL) { 1958 error = ENOBUFS; 1959 goto out; 1960 } 1961 } 1962 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1963 np->netc_rnodes); 1964 if (rn == NULL || np != (struct netcred *) rn) { /* already exists */ 1965 error = EPERM; 1966 goto out; 1967 } 1968 np->netc_exflags = argp->ex_flags; 1969 np->netc_anon = argp->ex_anon; 1970 np->netc_anon.cr_ref = 1; 1971 return (0); 1972 out: 1973 kfree(np, M_NETADDR); 1974 return (error); 1975 } 1976 1977 /* ARGSUSED */ 1978 static int 1979 vfs_free_netcred(struct radix_node *rn, void *w) 1980 { 1981 struct radix_node_head *rnh = (struct radix_node_head *) w; 1982 1983 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1984 kfree((caddr_t) rn, M_NETADDR); 1985 return (0); 1986 } 1987 1988 /* 1989 * Free the net address hash lists that are hanging off the mount points. 1990 */ 1991 static void 1992 vfs_free_addrlist(struct netexport *nep) 1993 { 1994 int i; 1995 struct radix_node_head *rnh; 1996 1997 for (i = 0; i <= AF_MAX; i++) 1998 if ((rnh = nep->ne_rtable[i])) { 1999 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2000 (caddr_t) rnh); 2001 kfree((caddr_t) rnh, M_RTABLE); 2002 nep->ne_rtable[i] = 0; 2003 } 2004 } 2005 2006 int 2007 vfs_export(struct mount *mp, struct netexport *nep, 2008 const struct export_args *argp) 2009 { 2010 int error; 2011 2012 if (argp->ex_flags & MNT_DELEXPORT) { 2013 if (mp->mnt_flag & MNT_EXPUBLIC) { 2014 vfs_setpublicfs(NULL, NULL, NULL); 2015 mp->mnt_flag &= ~MNT_EXPUBLIC; 2016 } 2017 vfs_free_addrlist(nep); 2018 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2019 } 2020 if (argp->ex_flags & MNT_EXPORTED) { 2021 if (argp->ex_flags & MNT_EXPUBLIC) { 2022 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2023 return (error); 2024 mp->mnt_flag |= MNT_EXPUBLIC; 2025 } 2026 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2027 return (error); 2028 mp->mnt_flag |= MNT_EXPORTED; 2029 } 2030 return (0); 2031 } 2032 2033 2034 /* 2035 * Set the publicly exported filesystem (WebNFS). Currently, only 2036 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2037 */ 2038 int 2039 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2040 const struct export_args *argp) 2041 { 2042 int error; 2043 struct vnode *rvp; 2044 char *cp; 2045 2046 /* 2047 * mp == NULL -> invalidate the current info, the FS is 2048 * no longer exported. May be called from either vfs_export 2049 * or unmount, so check if it hasn't already been done. 2050 */ 2051 if (mp == NULL) { 2052 if (nfs_pub.np_valid) { 2053 nfs_pub.np_valid = 0; 2054 if (nfs_pub.np_index != NULL) { 2055 kfree(nfs_pub.np_index, M_TEMP); 2056 nfs_pub.np_index = NULL; 2057 } 2058 } 2059 return (0); 2060 } 2061 2062 /* 2063 * Only one allowed at a time. 2064 */ 2065 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2066 return (EBUSY); 2067 2068 /* 2069 * Get real filehandle for root of exported FS. 2070 */ 2071 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2072 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2073 2074 if ((error = VFS_ROOT(mp, &rvp))) 2075 return (error); 2076 2077 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2078 return (error); 2079 2080 vput(rvp); 2081 2082 /* 2083 * If an indexfile was specified, pull it in. 2084 */ 2085 if (argp->ex_indexfile != NULL) { 2086 int namelen; 2087 2088 error = vn_get_namelen(rvp, &namelen); 2089 if (error) 2090 return (error); 2091 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK); 2092 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2093 namelen, NULL); 2094 if (!error) { 2095 /* 2096 * Check for illegal filenames. 2097 */ 2098 for (cp = nfs_pub.np_index; *cp; cp++) { 2099 if (*cp == '/') { 2100 error = EINVAL; 2101 break; 2102 } 2103 } 2104 } 2105 if (error) { 2106 kfree(nfs_pub.np_index, M_TEMP); 2107 return (error); 2108 } 2109 } 2110 2111 nfs_pub.np_mount = mp; 2112 nfs_pub.np_valid = 1; 2113 return (0); 2114 } 2115 2116 struct netcred * 2117 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2118 struct sockaddr *nam) 2119 { 2120 struct netcred *np; 2121 struct radix_node_head *rnh; 2122 struct sockaddr *saddr; 2123 2124 np = NULL; 2125 if (mp->mnt_flag & MNT_EXPORTED) { 2126 /* 2127 * Lookup in the export list first. 2128 */ 2129 if (nam != NULL) { 2130 saddr = nam; 2131 rnh = nep->ne_rtable[saddr->sa_family]; 2132 if (rnh != NULL) { 2133 np = (struct netcred *) 2134 (*rnh->rnh_matchaddr)((char *)saddr, 2135 rnh); 2136 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2137 np = NULL; 2138 } 2139 } 2140 /* 2141 * If no address match, use the default if it exists. 2142 */ 2143 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2144 np = &nep->ne_defexported; 2145 } 2146 return (np); 2147 } 2148 2149 /* 2150 * perform msync on all vnodes under a mount point. The mount point must 2151 * be locked. This code is also responsible for lazy-freeing unreferenced 2152 * vnodes whos VM objects no longer contain pages. 2153 * 2154 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2155 * 2156 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2157 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2158 * way up in this high level function. 2159 */ 2160 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2161 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2162 2163 void 2164 vfs_msync(struct mount *mp, int flags) 2165 { 2166 int vmsc_flags; 2167 2168 /* 2169 * tmpfs sets this flag to prevent msync(), sync, and the 2170 * filesystem periodic syncer from trying to flush VM pages 2171 * to swap. Only pure memory pressure flushes tmpfs VM pages 2172 * to swap. 2173 */ 2174 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2175 return; 2176 2177 /* 2178 * Ok, scan the vnodes for work. 2179 */ 2180 vmsc_flags = VMSC_GETVP; 2181 if (flags != MNT_WAIT) 2182 vmsc_flags |= VMSC_NOWAIT; 2183 vmntvnodescan(mp, vmsc_flags, 2184 vfs_msync_scan1, vfs_msync_scan2, 2185 (void *)(intptr_t)flags); 2186 } 2187 2188 /* 2189 * scan1 is a fast pre-check. There could be hundreds of thousands of 2190 * vnodes, we cannot afford to do anything heavy weight until we have a 2191 * fairly good indication that there is work to do. 2192 */ 2193 static 2194 int 2195 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2196 { 2197 int flags = (int)(intptr_t)data; 2198 2199 if ((vp->v_flag & VRECLAIMED) == 0) { 2200 if (vshouldmsync(vp)) 2201 return(0); /* call scan2 */ 2202 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2203 (vp->v_flag & VOBJDIRTY) && 2204 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2205 return(0); /* call scan2 */ 2206 } 2207 } 2208 2209 /* 2210 * do not call scan2, continue the loop 2211 */ 2212 return(-1); 2213 } 2214 2215 /* 2216 * This callback is handed a locked vnode. 2217 */ 2218 static 2219 int 2220 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2221 { 2222 vm_object_t obj; 2223 int flags = (int)(intptr_t)data; 2224 2225 if (vp->v_flag & VRECLAIMED) 2226 return(0); 2227 2228 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2229 if ((obj = vp->v_object) != NULL) { 2230 vm_object_page_clean(obj, 0, 0, 2231 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2232 } 2233 } 2234 return(0); 2235 } 2236 2237 /* 2238 * Wake up anyone interested in vp because it is being revoked. 2239 */ 2240 void 2241 vn_gone(struct vnode *vp) 2242 { 2243 lwkt_gettoken(&vp->v_token); 2244 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2245 lwkt_reltoken(&vp->v_token); 2246 } 2247 2248 /* 2249 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2250 * (or v_rdev might be NULL). 2251 */ 2252 cdev_t 2253 vn_todev(struct vnode *vp) 2254 { 2255 if (vp->v_type != VBLK && vp->v_type != VCHR) 2256 return (NULL); 2257 KKASSERT(vp->v_rdev != NULL); 2258 return (vp->v_rdev); 2259 } 2260 2261 /* 2262 * Check if vnode represents a disk device. The vnode does not need to be 2263 * opened. 2264 * 2265 * MPALMOSTSAFE 2266 */ 2267 int 2268 vn_isdisk(struct vnode *vp, int *errp) 2269 { 2270 cdev_t dev; 2271 2272 if (vp->v_type != VCHR) { 2273 if (errp != NULL) 2274 *errp = ENOTBLK; 2275 return (0); 2276 } 2277 2278 dev = vp->v_rdev; 2279 2280 if (dev == NULL) { 2281 if (errp != NULL) 2282 *errp = ENXIO; 2283 return (0); 2284 } 2285 if (dev_is_good(dev) == 0) { 2286 if (errp != NULL) 2287 *errp = ENXIO; 2288 return (0); 2289 } 2290 if ((dev_dflags(dev) & D_DISK) == 0) { 2291 if (errp != NULL) 2292 *errp = ENOTBLK; 2293 return (0); 2294 } 2295 if (errp != NULL) 2296 *errp = 0; 2297 return (1); 2298 } 2299 2300 int 2301 vn_get_namelen(struct vnode *vp, int *namelen) 2302 { 2303 int error; 2304 register_t retval[2]; 2305 2306 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2307 if (error) 2308 return (error); 2309 *namelen = (int)retval[0]; 2310 return (0); 2311 } 2312 2313 int 2314 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2315 uint16_t d_namlen, const char *d_name) 2316 { 2317 struct dirent *dp; 2318 size_t len; 2319 2320 len = _DIRENT_RECLEN(d_namlen); 2321 if (len > uio->uio_resid) 2322 return(1); 2323 2324 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2325 2326 dp->d_ino = d_ino; 2327 dp->d_namlen = d_namlen; 2328 dp->d_type = d_type; 2329 bcopy(d_name, dp->d_name, d_namlen); 2330 2331 *error = uiomove((caddr_t)dp, len, uio); 2332 2333 kfree(dp, M_TEMP); 2334 2335 return(0); 2336 } 2337 2338 void 2339 vn_mark_atime(struct vnode *vp, struct thread *td) 2340 { 2341 struct proc *p = td->td_proc; 2342 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2343 2344 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2345 VOP_MARKATIME(vp, cred); 2346 } 2347 } 2348