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. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.118 2008/09/17 21:44:18 dillon Exp $ 41 */ 42 43 /* 44 * External virtual filesystem routines 45 */ 46 #include "opt_ddb.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/domain.h> 54 #include <sys/eventhandler.h> 55 #include <sys/fcntl.h> 56 #include <sys/file.h> 57 #include <sys/kernel.h> 58 #include <sys/kthread.h> 59 #include <sys/malloc.h> 60 #include <sys/mbuf.h> 61 #include <sys/mount.h> 62 #include <sys/priv.h> 63 #include <sys/proc.h> 64 #include <sys/reboot.h> 65 #include <sys/socket.h> 66 #include <sys/stat.h> 67 #include <sys/sysctl.h> 68 #include <sys/syslog.h> 69 #include <sys/unistd.h> 70 #include <sys/vmmeter.h> 71 #include <sys/vnode.h> 72 73 #include <machine/limits.h> 74 75 #include <vm/vm.h> 76 #include <vm/vm_object.h> 77 #include <vm/vm_extern.h> 78 #include <vm/vm_kern.h> 79 #include <vm/pmap.h> 80 #include <vm/vm_map.h> 81 #include <vm/vm_page.h> 82 #include <vm/vm_pager.h> 83 #include <vm/vnode_pager.h> 84 #include <vm/vm_zone.h> 85 86 #include <sys/buf2.h> 87 #include <sys/thread2.h> 88 #include <sys/sysref2.h> 89 #include <sys/mplock2.h> 90 91 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 92 93 int numvnodes; 94 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 95 "Number of vnodes allocated"); 96 97 enum vtype iftovt_tab[16] = { 98 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 99 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 100 }; 101 int vttoif_tab[9] = { 102 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 103 S_IFSOCK, S_IFIFO, S_IFMT, 104 }; 105 106 static int reassignbufcalls; 107 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, 108 &reassignbufcalls, 0, ""); 109 static int reassignbufloops; 110 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, 111 &reassignbufloops, 0, ""); 112 static int reassignbufsortgood; 113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, 114 &reassignbufsortgood, 0, ""); 115 static int reassignbufsortbad; 116 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, 117 &reassignbufsortbad, 0, ""); 118 static int reassignbufmethod = 1; 119 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, 120 &reassignbufmethod, 0, ""); 121 static int check_buf_overlap = 2; /* invasive check */ 122 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, 123 &check_buf_overlap, 0, ""); 124 125 int nfs_mount_type = -1; 126 static struct lwkt_token spechash_token; 127 struct nfs_public nfs_pub; /* publicly exported FS */ 128 129 int desiredvnodes; 130 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 131 &desiredvnodes, 0, "Maximum number of vnodes"); 132 133 static void vfs_free_addrlist (struct netexport *nep); 134 static int vfs_free_netcred (struct radix_node *rn, void *w); 135 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 136 const struct export_args *argp); 137 138 /* 139 * Red black tree functions 140 */ 141 static int rb_buf_compare(struct buf *b1, struct buf *b2); 142 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 143 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 144 145 static int 146 rb_buf_compare(struct buf *b1, struct buf *b2) 147 { 148 if (b1->b_loffset < b2->b_loffset) 149 return(-1); 150 if (b1->b_loffset > b2->b_loffset) 151 return(1); 152 return(0); 153 } 154 155 /* 156 * Returns non-zero if the vnode is a candidate for lazy msyncing. 157 * 158 * NOTE: v_object is not stable (this scan can race), however the 159 * mntvnodescan code holds vmobj_token so any VM object we 160 * do find will remain stable storage. 161 */ 162 static __inline int 163 vshouldmsync(struct vnode *vp) 164 { 165 vm_object_t object; 166 167 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0) 168 return (0); /* other holders */ 169 object = vp->v_object; 170 cpu_ccfence(); 171 if (object && (object->ref_count || object->resident_page_count)) 172 return(0); 173 return (1); 174 } 175 176 /* 177 * Initialize the vnode management data structures. 178 * 179 * Called from vfsinit() 180 */ 181 void 182 vfs_subr_init(void) 183 { 184 int factor1; 185 int factor2; 186 187 /* 188 * Desiredvnodes is kern.maxvnodes. We want to scale it 189 * according to available system memory but we may also have 190 * to limit it based on available KVM, which is capped on 32 bit 191 * systems. 192 * 193 * WARNING! For machines with 64-256M of ram we have to be sure 194 * that the default limit scales down well due to HAMMER 195 * taking up significantly more memory per-vnode vs UFS. 196 * We want around ~5800 on a 128M machine. 197 */ 198 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode)); 199 factor2 = 22 * (sizeof(struct vm_object) + sizeof(struct vnode)); 200 desiredvnodes = 201 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 202 KvaSize / factor2); 203 desiredvnodes = imax(desiredvnodes, maxproc * 8); 204 205 lwkt_token_init(&spechash_token, 1, "spechash"); 206 } 207 208 /* 209 * Knob to control the precision of file timestamps: 210 * 211 * 0 = seconds only; nanoseconds zeroed. 212 * 1 = seconds and nanoseconds, accurate within 1/HZ. 213 * 2 = seconds and nanoseconds, truncated to microseconds. 214 * >=3 = seconds and nanoseconds, maximum precision. 215 */ 216 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 217 218 static int timestamp_precision = TSP_SEC; 219 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 220 ×tamp_precision, 0, ""); 221 222 /* 223 * Get a current timestamp. 224 * 225 * MPSAFE 226 */ 227 void 228 vfs_timestamp(struct timespec *tsp) 229 { 230 struct timeval tv; 231 232 switch (timestamp_precision) { 233 case TSP_SEC: 234 tsp->tv_sec = time_second; 235 tsp->tv_nsec = 0; 236 break; 237 case TSP_HZ: 238 getnanotime(tsp); 239 break; 240 case TSP_USEC: 241 microtime(&tv); 242 TIMEVAL_TO_TIMESPEC(&tv, tsp); 243 break; 244 case TSP_NSEC: 245 default: 246 nanotime(tsp); 247 break; 248 } 249 } 250 251 /* 252 * Set vnode attributes to VNOVAL 253 */ 254 void 255 vattr_null(struct vattr *vap) 256 { 257 vap->va_type = VNON; 258 vap->va_size = VNOVAL; 259 vap->va_bytes = VNOVAL; 260 vap->va_mode = VNOVAL; 261 vap->va_nlink = VNOVAL; 262 vap->va_uid = VNOVAL; 263 vap->va_gid = VNOVAL; 264 vap->va_fsid = VNOVAL; 265 vap->va_fileid = VNOVAL; 266 vap->va_blocksize = VNOVAL; 267 vap->va_rmajor = VNOVAL; 268 vap->va_rminor = VNOVAL; 269 vap->va_atime.tv_sec = VNOVAL; 270 vap->va_atime.tv_nsec = VNOVAL; 271 vap->va_mtime.tv_sec = VNOVAL; 272 vap->va_mtime.tv_nsec = VNOVAL; 273 vap->va_ctime.tv_sec = VNOVAL; 274 vap->va_ctime.tv_nsec = VNOVAL; 275 vap->va_flags = VNOVAL; 276 vap->va_gen = VNOVAL; 277 vap->va_vaflags = 0; 278 /* va_*_uuid fields are only valid if related flags are set */ 279 } 280 281 /* 282 * Flush out and invalidate all buffers associated with a vnode. 283 * 284 * vp must be locked. 285 */ 286 static int vinvalbuf_bp(struct buf *bp, void *data); 287 288 struct vinvalbuf_bp_info { 289 struct vnode *vp; 290 int slptimeo; 291 int lkflags; 292 int flags; 293 int clean; 294 }; 295 296 int 297 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 298 { 299 struct vinvalbuf_bp_info info; 300 vm_object_t object; 301 int error; 302 303 lwkt_gettoken(&vp->v_token); 304 305 /* 306 * If we are being asked to save, call fsync to ensure that the inode 307 * is updated. 308 */ 309 if (flags & V_SAVE) { 310 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 311 if (error) 312 goto done; 313 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 314 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 315 goto done; 316 317 /* 318 * Dirty bufs may be left or generated via races 319 * in circumstances where vinvalbuf() is called on 320 * a vnode not undergoing reclamation. Only 321 * panic if we are trying to reclaim the vnode. 322 */ 323 if ((vp->v_flag & VRECLAIMED) && 324 (bio_track_active(&vp->v_track_write) || 325 !RB_EMPTY(&vp->v_rbdirty_tree))) { 326 panic("vinvalbuf: dirty bufs"); 327 } 328 } 329 } 330 info.slptimeo = slptimeo; 331 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 332 if (slpflag & PCATCH) 333 info.lkflags |= LK_PCATCH; 334 info.flags = flags; 335 info.vp = vp; 336 337 /* 338 * Flush the buffer cache until nothing is left. 339 */ 340 while (!RB_EMPTY(&vp->v_rbclean_tree) || 341 !RB_EMPTY(&vp->v_rbdirty_tree)) { 342 info.clean = 1; 343 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 344 vinvalbuf_bp, &info); 345 if (error == 0) { 346 info.clean = 0; 347 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 348 vinvalbuf_bp, &info); 349 } 350 } 351 352 /* 353 * Wait for I/O completion. We may block in the pip code so we have 354 * to re-check. 355 */ 356 do { 357 bio_track_wait(&vp->v_track_write, 0, 0); 358 if ((object = vp->v_object) != NULL) { 359 while (object->paging_in_progress) 360 vm_object_pip_sleep(object, "vnvlbx"); 361 } 362 } while (bio_track_active(&vp->v_track_write)); 363 364 /* 365 * Destroy the copy in the VM cache, too. 366 */ 367 if ((object = vp->v_object) != NULL) { 368 vm_object_page_remove(object, 0, 0, 369 (flags & V_SAVE) ? TRUE : FALSE); 370 } 371 372 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 373 panic("vinvalbuf: flush failed"); 374 if (!RB_EMPTY(&vp->v_rbhash_tree)) 375 panic("vinvalbuf: flush failed, buffers still present"); 376 error = 0; 377 done: 378 lwkt_reltoken(&vp->v_token); 379 return (error); 380 } 381 382 static int 383 vinvalbuf_bp(struct buf *bp, void *data) 384 { 385 struct vinvalbuf_bp_info *info = data; 386 int error; 387 388 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 389 atomic_add_int(&bp->b_refs, 1); 390 error = BUF_TIMELOCK(bp, info->lkflags, 391 "vinvalbuf", info->slptimeo); 392 atomic_subtract_int(&bp->b_refs, 1); 393 if (error == 0) { 394 BUF_UNLOCK(bp); 395 error = ENOLCK; 396 } 397 if (error == ENOLCK) 398 return(0); 399 return (-error); 400 } 401 KKASSERT(bp->b_vp == info->vp); 402 403 /* 404 * Must check clean/dirty status after successfully locking as 405 * it may race. 406 */ 407 if ((info->clean && (bp->b_flags & B_DELWRI)) || 408 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 409 BUF_UNLOCK(bp); 410 return(0); 411 } 412 413 /* 414 * Note that vfs_bio_awrite expects buffers to reside 415 * on a queue, while bwrite() and brelse() do not. 416 * 417 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 418 * check. This code will write out the buffer, period. 419 */ 420 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 421 (info->flags & V_SAVE)) { 422 if (bp->b_flags & B_CLUSTEROK) { 423 vfs_bio_awrite(bp); 424 } else { 425 bremfree(bp); 426 bawrite(bp); 427 } 428 } else if (info->flags & V_SAVE) { 429 /* 430 * Cannot set B_NOCACHE on a clean buffer as this will 431 * destroy the VM backing store which might actually 432 * be dirty (and unsynchronized). 433 */ 434 bremfree(bp); 435 bp->b_flags |= (B_INVAL | B_RELBUF); 436 brelse(bp); 437 } else { 438 bremfree(bp); 439 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 440 brelse(bp); 441 } 442 return(0); 443 } 444 445 /* 446 * Truncate a file's buffer and pages to a specified length. This 447 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 448 * sync activity. 449 * 450 * The vnode must be locked. 451 */ 452 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 453 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 454 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 455 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 456 457 struct vtruncbuf_info { 458 struct vnode *vp; 459 off_t truncloffset; 460 int clean; 461 }; 462 463 int 464 vtruncbuf(struct vnode *vp, off_t length, int blksize) 465 { 466 struct vtruncbuf_info info; 467 const char *filename; 468 int count; 469 470 /* 471 * Round up to the *next* block, then destroy the buffers in question. 472 * Since we are only removing some of the buffers we must rely on the 473 * scan count to determine whether a loop is necessary. 474 */ 475 if ((count = (int)(length % blksize)) != 0) 476 info.truncloffset = length + (blksize - count); 477 else 478 info.truncloffset = length; 479 info.vp = vp; 480 481 lwkt_gettoken(&vp->v_token); 482 do { 483 info.clean = 1; 484 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 485 vtruncbuf_bp_trunc_cmp, 486 vtruncbuf_bp_trunc, &info); 487 info.clean = 0; 488 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 489 vtruncbuf_bp_trunc_cmp, 490 vtruncbuf_bp_trunc, &info); 491 } while(count); 492 493 /* 494 * For safety, fsync any remaining metadata if the file is not being 495 * truncated to 0. Since the metadata does not represent the entire 496 * dirty list we have to rely on the hit count to ensure that we get 497 * all of it. 498 */ 499 if (length > 0) { 500 do { 501 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 502 vtruncbuf_bp_metasync_cmp, 503 vtruncbuf_bp_metasync, &info); 504 } while (count); 505 } 506 507 /* 508 * Clean out any left over VM backing store. 509 * 510 * It is possible to have in-progress I/O from buffers that were 511 * not part of the truncation. This should not happen if we 512 * are truncating to 0-length. 513 */ 514 vnode_pager_setsize(vp, length); 515 bio_track_wait(&vp->v_track_write, 0, 0); 516 517 /* 518 * Debugging only 519 */ 520 spin_lock(&vp->v_spinlock); 521 filename = TAILQ_FIRST(&vp->v_namecache) ? 522 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 523 spin_unlock(&vp->v_spinlock); 524 525 /* 526 * Make sure no buffers were instantiated while we were trying 527 * to clean out the remaining VM pages. This could occur due 528 * to busy dirty VM pages being flushed out to disk. 529 */ 530 do { 531 info.clean = 1; 532 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 533 vtruncbuf_bp_trunc_cmp, 534 vtruncbuf_bp_trunc, &info); 535 info.clean = 0; 536 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 537 vtruncbuf_bp_trunc_cmp, 538 vtruncbuf_bp_trunc, &info); 539 if (count) { 540 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 541 "left over buffers in %s\n", count, filename); 542 } 543 } while(count); 544 545 lwkt_reltoken(&vp->v_token); 546 547 return (0); 548 } 549 550 /* 551 * The callback buffer is beyond the new file EOF and must be destroyed. 552 * Note that the compare function must conform to the RB_SCAN's requirements. 553 */ 554 static 555 int 556 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 557 { 558 struct vtruncbuf_info *info = data; 559 560 if (bp->b_loffset >= info->truncloffset) 561 return(0); 562 return(-1); 563 } 564 565 static 566 int 567 vtruncbuf_bp_trunc(struct buf *bp, void *data) 568 { 569 struct vtruncbuf_info *info = data; 570 571 /* 572 * Do not try to use a buffer we cannot immediately lock, but sleep 573 * anyway to prevent a livelock. The code will loop until all buffers 574 * can be acted upon. 575 * 576 * We must always revalidate the buffer after locking it to deal 577 * with MP races. 578 */ 579 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 580 atomic_add_int(&bp->b_refs, 1); 581 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 582 BUF_UNLOCK(bp); 583 atomic_subtract_int(&bp->b_refs, 1); 584 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 585 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 586 bp->b_vp != info->vp || 587 vtruncbuf_bp_trunc_cmp(bp, data)) { 588 BUF_UNLOCK(bp); 589 } else { 590 bremfree(bp); 591 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 592 brelse(bp); 593 } 594 return(1); 595 } 596 597 /* 598 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 599 * blocks (with a negative loffset) are scanned. 600 * Note that the compare function must conform to the RB_SCAN's requirements. 601 */ 602 static int 603 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 604 { 605 if (bp->b_loffset < 0) 606 return(0); 607 return(1); 608 } 609 610 static int 611 vtruncbuf_bp_metasync(struct buf *bp, void *data) 612 { 613 struct vtruncbuf_info *info = data; 614 615 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 616 atomic_add_int(&bp->b_refs, 1); 617 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 618 BUF_UNLOCK(bp); 619 atomic_subtract_int(&bp->b_refs, 1); 620 } else if ((bp->b_flags & B_DELWRI) == 0 || 621 bp->b_vp != info->vp || 622 vtruncbuf_bp_metasync_cmp(bp, data)) { 623 BUF_UNLOCK(bp); 624 } else { 625 bremfree(bp); 626 if (bp->b_vp == info->vp) 627 bawrite(bp); 628 else 629 bwrite(bp); 630 } 631 return(1); 632 } 633 634 /* 635 * vfsync - implements a multipass fsync on a file which understands 636 * dependancies and meta-data. The passed vnode must be locked. The 637 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 638 * 639 * When fsyncing data asynchronously just do one consolidated pass starting 640 * with the most negative block number. This may not get all the data due 641 * to dependancies. 642 * 643 * When fsyncing data synchronously do a data pass, then a metadata pass, 644 * then do additional data+metadata passes to try to get all the data out. 645 */ 646 static int vfsync_wait_output(struct vnode *vp, 647 int (*waitoutput)(struct vnode *, struct thread *)); 648 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 649 static int vfsync_data_only_cmp(struct buf *bp, void *data); 650 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 651 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 652 static int vfsync_bp(struct buf *bp, void *data); 653 654 struct vfsync_info { 655 struct vnode *vp; 656 int synchronous; 657 int syncdeps; 658 int lazycount; 659 int lazylimit; 660 int skippedbufs; 661 int (*checkdef)(struct buf *); 662 int (*cmpfunc)(struct buf *, void *); 663 }; 664 665 int 666 vfsync(struct vnode *vp, int waitfor, int passes, 667 int (*checkdef)(struct buf *), 668 int (*waitoutput)(struct vnode *, struct thread *)) 669 { 670 struct vfsync_info info; 671 int error; 672 673 bzero(&info, sizeof(info)); 674 info.vp = vp; 675 if ((info.checkdef = checkdef) == NULL) 676 info.syncdeps = 1; 677 678 lwkt_gettoken(&vp->v_token); 679 680 switch(waitfor) { 681 case MNT_LAZY: 682 /* 683 * Lazy (filesystem syncer typ) Asynchronous plus limit the 684 * number of data (not meta) pages we try to flush to 1MB. 685 * A non-zero return means that lazy limit was reached. 686 */ 687 info.lazylimit = 1024 * 1024; 688 info.syncdeps = 1; 689 info.cmpfunc = vfsync_lazy_range_cmp; 690 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 691 vfsync_lazy_range_cmp, vfsync_bp, &info); 692 info.cmpfunc = vfsync_meta_only_cmp; 693 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 694 vfsync_meta_only_cmp, vfsync_bp, &info); 695 if (error == 0) 696 vp->v_lazyw = 0; 697 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 698 vn_syncer_add(vp, 1); 699 error = 0; 700 break; 701 case MNT_NOWAIT: 702 /* 703 * Asynchronous. Do a data-only pass and a meta-only pass. 704 */ 705 info.syncdeps = 1; 706 info.cmpfunc = vfsync_data_only_cmp; 707 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 708 vfsync_bp, &info); 709 info.cmpfunc = vfsync_meta_only_cmp; 710 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 711 vfsync_bp, &info); 712 error = 0; 713 break; 714 default: 715 /* 716 * Synchronous. Do a data-only pass, then a meta-data+data 717 * pass, then additional integrated passes to try to get 718 * all the dependancies flushed. 719 */ 720 info.cmpfunc = vfsync_data_only_cmp; 721 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 722 vfsync_bp, &info); 723 error = vfsync_wait_output(vp, waitoutput); 724 if (error == 0) { 725 info.skippedbufs = 0; 726 info.cmpfunc = vfsync_dummy_cmp; 727 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 728 vfsync_bp, &info); 729 error = vfsync_wait_output(vp, waitoutput); 730 if (info.skippedbufs) { 731 kprintf("Warning: vfsync skipped %d dirty " 732 "bufs in pass2!\n", info.skippedbufs); 733 } 734 } 735 while (error == 0 && passes > 0 && 736 !RB_EMPTY(&vp->v_rbdirty_tree) 737 ) { 738 if (--passes == 0) { 739 info.synchronous = 1; 740 info.syncdeps = 1; 741 } 742 info.cmpfunc = vfsync_dummy_cmp; 743 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 744 vfsync_bp, &info); 745 if (error < 0) 746 error = -error; 747 info.syncdeps = 1; 748 if (error == 0) 749 error = vfsync_wait_output(vp, waitoutput); 750 } 751 break; 752 } 753 lwkt_reltoken(&vp->v_token); 754 return(error); 755 } 756 757 static int 758 vfsync_wait_output(struct vnode *vp, 759 int (*waitoutput)(struct vnode *, struct thread *)) 760 { 761 int error; 762 763 error = bio_track_wait(&vp->v_track_write, 0, 0); 764 if (waitoutput) 765 error = waitoutput(vp, curthread); 766 return(error); 767 } 768 769 static int 770 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 771 { 772 return(0); 773 } 774 775 static int 776 vfsync_data_only_cmp(struct buf *bp, void *data) 777 { 778 if (bp->b_loffset < 0) 779 return(-1); 780 return(0); 781 } 782 783 static int 784 vfsync_meta_only_cmp(struct buf *bp, void *data) 785 { 786 if (bp->b_loffset < 0) 787 return(0); 788 return(1); 789 } 790 791 static int 792 vfsync_lazy_range_cmp(struct buf *bp, void *data) 793 { 794 struct vfsync_info *info = data; 795 796 if (bp->b_loffset < info->vp->v_lazyw) 797 return(-1); 798 return(0); 799 } 800 801 static int 802 vfsync_bp(struct buf *bp, void *data) 803 { 804 struct vfsync_info *info = data; 805 struct vnode *vp = info->vp; 806 int error; 807 808 /* 809 * Ignore buffers that we cannot immediately lock. 810 */ 811 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 812 ++info->skippedbufs; 813 return(0); 814 } 815 816 /* 817 * We must revalidate the buffer after locking. 818 */ 819 if ((bp->b_flags & B_DELWRI) == 0 || 820 bp->b_vp != info->vp || 821 info->cmpfunc(bp, data)) { 822 BUF_UNLOCK(bp); 823 return(0); 824 } 825 826 /* 827 * If syncdeps is not set we do not try to write buffers which have 828 * dependancies. 829 */ 830 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 831 BUF_UNLOCK(bp); 832 return(0); 833 } 834 835 /* 836 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 837 * has been written but an additional handshake with the device 838 * is required before we can dispose of the buffer. We have no idea 839 * how to do this so we have to skip these buffers. 840 */ 841 if (bp->b_flags & B_NEEDCOMMIT) { 842 BUF_UNLOCK(bp); 843 return(0); 844 } 845 846 /* 847 * Ask bioops if it is ok to sync. If not the VFS may have 848 * set B_LOCKED so we have to cycle the buffer. 849 */ 850 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 851 bremfree(bp); 852 brelse(bp); 853 return(0); 854 } 855 856 if (info->synchronous) { 857 /* 858 * Synchronous flushing. An error may be returned. 859 */ 860 bremfree(bp); 861 error = bwrite(bp); 862 } else { 863 /* 864 * Asynchronous flushing. A negative return value simply 865 * stops the scan and is not considered an error. We use 866 * this to support limited MNT_LAZY flushes. 867 */ 868 vp->v_lazyw = bp->b_loffset; 869 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 870 info->lazycount += vfs_bio_awrite(bp); 871 } else { 872 info->lazycount += bp->b_bufsize; 873 bremfree(bp); 874 bawrite(bp); 875 } 876 waitrunningbufspace(); 877 if (info->lazylimit && info->lazycount >= info->lazylimit) 878 error = 1; 879 else 880 error = 0; 881 } 882 return(-error); 883 } 884 885 /* 886 * Associate a buffer with a vnode. 887 * 888 * MPSAFE 889 */ 890 int 891 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 892 { 893 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 894 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 895 896 /* 897 * Insert onto list for new vnode. 898 */ 899 lwkt_gettoken(&vp->v_token); 900 901 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 902 lwkt_reltoken(&vp->v_token); 903 return (EEXIST); 904 } 905 906 /* 907 * Diagnostics (mainly for HAMMER debugging). Check for 908 * overlapping buffers. 909 */ 910 if (check_buf_overlap) { 911 struct buf *bx; 912 bx = buf_rb_hash_RB_PREV(bp); 913 if (bx) { 914 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 915 kprintf("bgetvp: overlapl %016jx/%d %016jx " 916 "bx %p bp %p\n", 917 (intmax_t)bx->b_loffset, 918 bx->b_bufsize, 919 (intmax_t)bp->b_loffset, 920 bx, bp); 921 if (check_buf_overlap > 1) 922 panic("bgetvp - overlapping buffer"); 923 } 924 } 925 bx = buf_rb_hash_RB_NEXT(bp); 926 if (bx) { 927 if (bp->b_loffset + testsize > bx->b_loffset) { 928 kprintf("bgetvp: overlapr %016jx/%d %016jx " 929 "bp %p bx %p\n", 930 (intmax_t)bp->b_loffset, 931 testsize, 932 (intmax_t)bx->b_loffset, 933 bp, bx); 934 if (check_buf_overlap > 1) 935 panic("bgetvp - overlapping buffer"); 936 } 937 } 938 } 939 bp->b_vp = vp; 940 bp->b_flags |= B_HASHED; 941 bp->b_flags |= B_VNCLEAN; 942 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 943 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 944 vhold(vp); 945 lwkt_reltoken(&vp->v_token); 946 return(0); 947 } 948 949 /* 950 * Disassociate a buffer from a vnode. 951 * 952 * MPSAFE 953 */ 954 void 955 brelvp(struct buf *bp) 956 { 957 struct vnode *vp; 958 959 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 960 961 /* 962 * Delete from old vnode list, if on one. 963 */ 964 vp = bp->b_vp; 965 lwkt_gettoken(&vp->v_token); 966 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 967 if (bp->b_flags & B_VNDIRTY) 968 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 969 else 970 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 971 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 972 } 973 if (bp->b_flags & B_HASHED) { 974 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 975 bp->b_flags &= ~B_HASHED; 976 } 977 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) 978 vn_syncer_remove(vp); 979 bp->b_vp = NULL; 980 981 lwkt_reltoken(&vp->v_token); 982 983 vdrop(vp); 984 } 985 986 /* 987 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 988 * This routine is called when the state of the B_DELWRI bit is changed. 989 * 990 * Must be called with vp->v_token held. 991 * MPSAFE 992 */ 993 void 994 reassignbuf(struct buf *bp) 995 { 996 struct vnode *vp = bp->b_vp; 997 int delay; 998 999 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 1000 ++reassignbufcalls; 1001 1002 /* 1003 * B_PAGING flagged buffers cannot be reassigned because their vp 1004 * is not fully linked in. 1005 */ 1006 if (bp->b_flags & B_PAGING) 1007 panic("cannot reassign paging buffer"); 1008 1009 if (bp->b_flags & B_DELWRI) { 1010 /* 1011 * Move to the dirty list, add the vnode to the worklist 1012 */ 1013 if (bp->b_flags & B_VNCLEAN) { 1014 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1015 bp->b_flags &= ~B_VNCLEAN; 1016 } 1017 if ((bp->b_flags & B_VNDIRTY) == 0) { 1018 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 1019 panic("reassignbuf: dup lblk vp %p bp %p", 1020 vp, bp); 1021 } 1022 bp->b_flags |= B_VNDIRTY; 1023 } 1024 if ((vp->v_flag & VONWORKLST) == 0) { 1025 switch (vp->v_type) { 1026 case VDIR: 1027 delay = dirdelay; 1028 break; 1029 case VCHR: 1030 case VBLK: 1031 if (vp->v_rdev && 1032 vp->v_rdev->si_mountpoint != NULL) { 1033 delay = metadelay; 1034 break; 1035 } 1036 /* fall through */ 1037 default: 1038 delay = filedelay; 1039 } 1040 vn_syncer_add(vp, delay); 1041 } 1042 } else { 1043 /* 1044 * Move to the clean list, remove the vnode from the worklist 1045 * if no dirty blocks remain. 1046 */ 1047 if (bp->b_flags & B_VNDIRTY) { 1048 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1049 bp->b_flags &= ~B_VNDIRTY; 1050 } 1051 if ((bp->b_flags & B_VNCLEAN) == 0) { 1052 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1053 panic("reassignbuf: dup lblk vp %p bp %p", 1054 vp, bp); 1055 } 1056 bp->b_flags |= B_VNCLEAN; 1057 } 1058 if ((vp->v_flag & VONWORKLST) && 1059 RB_EMPTY(&vp->v_rbdirty_tree)) { 1060 vn_syncer_remove(vp); 1061 } 1062 } 1063 } 1064 1065 /* 1066 * Create a vnode for a block device. 1067 * Used for mounting the root file system. 1068 */ 1069 extern struct vop_ops *devfs_vnode_dev_vops_p; 1070 int 1071 bdevvp(cdev_t dev, struct vnode **vpp) 1072 { 1073 struct vnode *vp; 1074 struct vnode *nvp; 1075 int error; 1076 1077 if (dev == NULL) { 1078 *vpp = NULLVP; 1079 return (ENXIO); 1080 } 1081 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1082 &nvp, 0, 0); 1083 if (error) { 1084 *vpp = NULLVP; 1085 return (error); 1086 } 1087 vp = nvp; 1088 vp->v_type = VCHR; 1089 #if 0 1090 vp->v_rdev = dev; 1091 #endif 1092 v_associate_rdev(vp, dev); 1093 vp->v_umajor = dev->si_umajor; 1094 vp->v_uminor = dev->si_uminor; 1095 vx_unlock(vp); 1096 *vpp = vp; 1097 return (0); 1098 } 1099 1100 int 1101 v_associate_rdev(struct vnode *vp, cdev_t dev) 1102 { 1103 if (dev == NULL) 1104 return(ENXIO); 1105 if (dev_is_good(dev) == 0) 1106 return(ENXIO); 1107 KKASSERT(vp->v_rdev == NULL); 1108 vp->v_rdev = reference_dev(dev); 1109 lwkt_gettoken(&spechash_token); 1110 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1111 lwkt_reltoken(&spechash_token); 1112 return(0); 1113 } 1114 1115 void 1116 v_release_rdev(struct vnode *vp) 1117 { 1118 cdev_t dev; 1119 1120 if ((dev = vp->v_rdev) != NULL) { 1121 lwkt_gettoken(&spechash_token); 1122 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1123 vp->v_rdev = NULL; 1124 release_dev(dev); 1125 lwkt_reltoken(&spechash_token); 1126 } 1127 } 1128 1129 /* 1130 * Add a vnode to the alias list hung off the cdev_t. We only associate 1131 * the device number with the vnode. The actual device is not associated 1132 * until the vnode is opened (usually in spec_open()), and will be 1133 * disassociated on last close. 1134 */ 1135 void 1136 addaliasu(struct vnode *nvp, int x, int y) 1137 { 1138 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1139 panic("addaliasu on non-special vnode"); 1140 nvp->v_umajor = x; 1141 nvp->v_uminor = y; 1142 } 1143 1144 /* 1145 * Simple call that a filesystem can make to try to get rid of a 1146 * vnode. It will fail if anyone is referencing the vnode (including 1147 * the caller). 1148 * 1149 * The filesystem can check whether its in-memory inode structure still 1150 * references the vp on return. 1151 */ 1152 void 1153 vclean_unlocked(struct vnode *vp) 1154 { 1155 vx_get(vp); 1156 if (sysref_isactive(&vp->v_sysref) == 0) 1157 vgone_vxlocked(vp); 1158 vx_put(vp); 1159 } 1160 1161 /* 1162 * Disassociate a vnode from its underlying filesystem. 1163 * 1164 * The vnode must be VX locked and referenced. In all normal situations 1165 * there are no active references. If vclean_vxlocked() is called while 1166 * there are active references, the vnode is being ripped out and we have 1167 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1168 */ 1169 void 1170 vclean_vxlocked(struct vnode *vp, int flags) 1171 { 1172 int active; 1173 int n; 1174 vm_object_t object; 1175 1176 /* 1177 * If the vnode has already been reclaimed we have nothing to do. 1178 */ 1179 if (vp->v_flag & VRECLAIMED) 1180 return; 1181 vsetflags(vp, VRECLAIMED); 1182 1183 /* 1184 * Scrap the vfs cache 1185 */ 1186 while (cache_inval_vp(vp, 0) != 0) { 1187 kprintf("Warning: vnode %p clean/cache_resolution race detected\n", vp); 1188 tsleep(vp, 0, "vclninv", 2); 1189 } 1190 1191 /* 1192 * Check to see if the vnode is in use. If so we have to reference it 1193 * before we clean it out so that its count cannot fall to zero and 1194 * generate a race against ourselves to recycle it. 1195 */ 1196 active = sysref_isactive(&vp->v_sysref); 1197 1198 /* 1199 * Clean out any buffers associated with the vnode and destroy its 1200 * object, if it has one. 1201 */ 1202 vinvalbuf(vp, V_SAVE, 0, 0); 1203 1204 /* 1205 * If purging an active vnode (typically during a forced unmount 1206 * or reboot), it must be closed and deactivated before being 1207 * reclaimed. This isn't really all that safe, but what can 1208 * we do? XXX. 1209 * 1210 * Note that neither of these routines unlocks the vnode. 1211 */ 1212 if (active && (flags & DOCLOSE)) { 1213 while ((n = vp->v_opencount) != 0) { 1214 if (vp->v_writecount) 1215 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1216 else 1217 VOP_CLOSE(vp, FNONBLOCK); 1218 if (vp->v_opencount == n) { 1219 kprintf("Warning: unable to force-close" 1220 " vnode %p\n", vp); 1221 break; 1222 } 1223 } 1224 } 1225 1226 /* 1227 * If the vnode has not been deactivated, deactivated it. Deactivation 1228 * can create new buffers and VM pages so we have to call vinvalbuf() 1229 * again to make sure they all get flushed. 1230 * 1231 * This can occur if a file with a link count of 0 needs to be 1232 * truncated. 1233 * 1234 * If the vnode is already dead don't try to deactivate it. 1235 */ 1236 if ((vp->v_flag & VINACTIVE) == 0) { 1237 vsetflags(vp, VINACTIVE); 1238 if (vp->v_mount) 1239 VOP_INACTIVE(vp); 1240 vinvalbuf(vp, V_SAVE, 0, 0); 1241 } 1242 1243 /* 1244 * If the vnode has an object, destroy it. 1245 */ 1246 lwkt_gettoken(&vmobj_token); 1247 if ((object = vp->v_object) != NULL) { 1248 KKASSERT(object == vp->v_object); 1249 if (object->ref_count == 0) { 1250 if ((object->flags & OBJ_DEAD) == 0) 1251 vm_object_terminate(object); 1252 } else { 1253 vm_pager_deallocate(object); 1254 } 1255 vclrflags(vp, VOBJBUF); 1256 } 1257 lwkt_reltoken(&vmobj_token); 1258 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1259 1260 /* 1261 * Reclaim the vnode if not already dead. 1262 */ 1263 if (vp->v_mount && VOP_RECLAIM(vp)) 1264 panic("vclean: cannot reclaim"); 1265 1266 /* 1267 * Done with purge, notify sleepers of the grim news. 1268 */ 1269 vp->v_ops = &dead_vnode_vops_p; 1270 vn_gone(vp); 1271 vp->v_tag = VT_NON; 1272 1273 /* 1274 * If we are destroying an active vnode, reactivate it now that 1275 * we have reassociated it with deadfs. This prevents the system 1276 * from crashing on the vnode due to it being unexpectedly marked 1277 * as inactive or reclaimed. 1278 */ 1279 if (active && (flags & DOCLOSE)) { 1280 vclrflags(vp, VINACTIVE | VRECLAIMED); 1281 } 1282 } 1283 1284 /* 1285 * Eliminate all activity associated with the requested vnode 1286 * and with all vnodes aliased to the requested vnode. 1287 * 1288 * The vnode must be referenced but should not be locked. 1289 */ 1290 int 1291 vrevoke(struct vnode *vp, struct ucred *cred) 1292 { 1293 struct vnode *vq; 1294 struct vnode *vqn; 1295 cdev_t dev; 1296 int error; 1297 1298 /* 1299 * If the vnode has a device association, scrap all vnodes associated 1300 * with the device. Don't let the device disappear on us while we 1301 * are scrapping the vnodes. 1302 * 1303 * The passed vp will probably show up in the list, do not VX lock 1304 * it twice! 1305 * 1306 * Releasing the vnode's rdev here can mess up specfs's call to 1307 * device close, so don't do it. The vnode has been disassociated 1308 * and the device will be closed after the last ref on the related 1309 * fp goes away (if not still open by e.g. the kernel). 1310 */ 1311 if (vp->v_type != VCHR) { 1312 error = fdrevoke(vp, DTYPE_VNODE, cred); 1313 return (error); 1314 } 1315 if ((dev = vp->v_rdev) == NULL) { 1316 return(0); 1317 } 1318 reference_dev(dev); 1319 lwkt_gettoken(&spechash_token); 1320 1321 vqn = SLIST_FIRST(&dev->si_hlist); 1322 if (vqn) 1323 vref(vqn); 1324 while ((vq = vqn) != NULL) { 1325 vqn = SLIST_NEXT(vqn, v_cdevnext); 1326 if (vqn) 1327 vref(vqn); 1328 fdrevoke(vq, DTYPE_VNODE, cred); 1329 /*v_release_rdev(vq);*/ 1330 vrele(vq); 1331 } 1332 lwkt_reltoken(&spechash_token); 1333 dev_drevoke(dev); 1334 release_dev(dev); 1335 return (0); 1336 } 1337 1338 /* 1339 * This is called when the object underlying a vnode is being destroyed, 1340 * such as in a remove(). Try to recycle the vnode immediately if the 1341 * only active reference is our reference. 1342 * 1343 * Directory vnodes in the namecache with children cannot be immediately 1344 * recycled because numerous VOP_N*() ops require them to be stable. 1345 * 1346 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1347 * function is a NOP if VRECLAIMED is already set. 1348 */ 1349 int 1350 vrecycle(struct vnode *vp) 1351 { 1352 if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1353 if (cache_inval_vp_nonblock(vp)) 1354 return(0); 1355 vgone_vxlocked(vp); 1356 return (1); 1357 } 1358 return (0); 1359 } 1360 1361 /* 1362 * Return the maximum I/O size allowed for strategy calls on VP. 1363 * 1364 * If vp is VCHR or VBLK we dive the device, otherwise we use 1365 * the vp's mount info. 1366 */ 1367 int 1368 vmaxiosize(struct vnode *vp) 1369 { 1370 if (vp->v_type == VBLK || vp->v_type == VCHR) { 1371 return(vp->v_rdev->si_iosize_max); 1372 } else { 1373 return(vp->v_mount->mnt_iosize_max); 1374 } 1375 } 1376 1377 /* 1378 * Eliminate all activity associated with a vnode in preparation for reuse. 1379 * 1380 * The vnode must be VX locked and refd and will remain VX locked and refd 1381 * on return. This routine may be called with the vnode in any state, as 1382 * long as it is VX locked. The vnode will be cleaned out and marked 1383 * VRECLAIMED but will not actually be reused until all existing refs and 1384 * holds go away. 1385 * 1386 * NOTE: This routine may be called on a vnode which has not yet been 1387 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1388 * already been reclaimed. 1389 * 1390 * This routine is not responsible for placing us back on the freelist. 1391 * Instead, it happens automatically when the caller releases the VX lock 1392 * (assuming there aren't any other references). 1393 */ 1394 void 1395 vgone_vxlocked(struct vnode *vp) 1396 { 1397 /* 1398 * assert that the VX lock is held. This is an absolute requirement 1399 * now for vgone_vxlocked() to be called. 1400 */ 1401 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1402 1403 get_mplock(); 1404 1405 /* 1406 * Clean out the filesystem specific data and set the VRECLAIMED 1407 * bit. Also deactivate the vnode if necessary. 1408 */ 1409 vclean_vxlocked(vp, DOCLOSE); 1410 1411 /* 1412 * Delete from old mount point vnode list, if on one. 1413 */ 1414 if (vp->v_mount != NULL) { 1415 KKASSERT(vp->v_data == NULL); 1416 insmntque(vp, NULL); 1417 } 1418 1419 /* 1420 * If special device, remove it from special device alias list 1421 * if it is on one. This should normally only occur if a vnode is 1422 * being revoked as the device should otherwise have been released 1423 * naturally. 1424 */ 1425 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1426 v_release_rdev(vp); 1427 } 1428 1429 /* 1430 * Set us to VBAD 1431 */ 1432 vp->v_type = VBAD; 1433 rel_mplock(); 1434 } 1435 1436 /* 1437 * Lookup a vnode by device number. 1438 * 1439 * Returns non-zero and *vpp set to a vref'd vnode on success. 1440 * Returns zero on failure. 1441 */ 1442 int 1443 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1444 { 1445 struct vnode *vp; 1446 1447 lwkt_gettoken(&spechash_token); 1448 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1449 if (type == vp->v_type) { 1450 *vpp = vp; 1451 vref(vp); 1452 lwkt_reltoken(&spechash_token); 1453 return (1); 1454 } 1455 } 1456 lwkt_reltoken(&spechash_token); 1457 return (0); 1458 } 1459 1460 /* 1461 * Calculate the total number of references to a special device. This 1462 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1463 * an overloaded field. Since udev2dev can now return NULL, we have 1464 * to check for a NULL v_rdev. 1465 */ 1466 int 1467 count_dev(cdev_t dev) 1468 { 1469 struct vnode *vp; 1470 int count = 0; 1471 1472 if (SLIST_FIRST(&dev->si_hlist)) { 1473 lwkt_gettoken(&spechash_token); 1474 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1475 count += vp->v_opencount; 1476 } 1477 lwkt_reltoken(&spechash_token); 1478 } 1479 return(count); 1480 } 1481 1482 int 1483 vcount(struct vnode *vp) 1484 { 1485 if (vp->v_rdev == NULL) 1486 return(0); 1487 return(count_dev(vp->v_rdev)); 1488 } 1489 1490 /* 1491 * Initialize VMIO for a vnode. This routine MUST be called before a 1492 * VFS can issue buffer cache ops on a vnode. It is typically called 1493 * when a vnode is initialized from its inode. 1494 */ 1495 int 1496 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1497 { 1498 vm_object_t object; 1499 int error = 0; 1500 1501 lwkt_gettoken(&vmobj_token); 1502 retry: 1503 if ((object = vp->v_object) == NULL) { 1504 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1505 /* 1506 * Dereference the reference we just created. This assumes 1507 * that the object is associated with the vp. 1508 */ 1509 object->ref_count--; 1510 vrele(vp); 1511 } else { 1512 if (object->flags & OBJ_DEAD) { 1513 vn_unlock(vp); 1514 if (vp->v_object == object) 1515 vm_object_dead_sleep(object, "vodead"); 1516 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1517 goto retry; 1518 } 1519 } 1520 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1521 vsetflags(vp, VOBJBUF); 1522 lwkt_reltoken(&vmobj_token); 1523 1524 return (error); 1525 } 1526 1527 1528 /* 1529 * Print out a description of a vnode. 1530 */ 1531 static char *typename[] = 1532 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1533 1534 void 1535 vprint(char *label, struct vnode *vp) 1536 { 1537 char buf[96]; 1538 1539 if (label != NULL) 1540 kprintf("%s: %p: ", label, (void *)vp); 1541 else 1542 kprintf("%p: ", (void *)vp); 1543 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1544 typename[vp->v_type], 1545 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1546 buf[0] = '\0'; 1547 if (vp->v_flag & VROOT) 1548 strcat(buf, "|VROOT"); 1549 if (vp->v_flag & VPFSROOT) 1550 strcat(buf, "|VPFSROOT"); 1551 if (vp->v_flag & VTEXT) 1552 strcat(buf, "|VTEXT"); 1553 if (vp->v_flag & VSYSTEM) 1554 strcat(buf, "|VSYSTEM"); 1555 if (vp->v_flag & VFREE) 1556 strcat(buf, "|VFREE"); 1557 if (vp->v_flag & VOBJBUF) 1558 strcat(buf, "|VOBJBUF"); 1559 if (buf[0] != '\0') 1560 kprintf(" flags (%s)", &buf[1]); 1561 if (vp->v_data == NULL) { 1562 kprintf("\n"); 1563 } else { 1564 kprintf("\n\t"); 1565 VOP_PRINT(vp); 1566 } 1567 } 1568 1569 /* 1570 * Do the usual access checking. 1571 * file_mode, uid and gid are from the vnode in question, 1572 * while acc_mode and cred are from the VOP_ACCESS parameter list 1573 */ 1574 int 1575 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1576 mode_t acc_mode, struct ucred *cred) 1577 { 1578 mode_t mask; 1579 int ismember; 1580 1581 /* 1582 * Super-user always gets read/write access, but execute access depends 1583 * on at least one execute bit being set. 1584 */ 1585 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1586 if ((acc_mode & VEXEC) && type != VDIR && 1587 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1588 return (EACCES); 1589 return (0); 1590 } 1591 1592 mask = 0; 1593 1594 /* Otherwise, check the owner. */ 1595 if (cred->cr_uid == uid) { 1596 if (acc_mode & VEXEC) 1597 mask |= S_IXUSR; 1598 if (acc_mode & VREAD) 1599 mask |= S_IRUSR; 1600 if (acc_mode & VWRITE) 1601 mask |= S_IWUSR; 1602 return ((file_mode & mask) == mask ? 0 : EACCES); 1603 } 1604 1605 /* Otherwise, check the groups. */ 1606 ismember = groupmember(gid, cred); 1607 if (cred->cr_svgid == gid || ismember) { 1608 if (acc_mode & VEXEC) 1609 mask |= S_IXGRP; 1610 if (acc_mode & VREAD) 1611 mask |= S_IRGRP; 1612 if (acc_mode & VWRITE) 1613 mask |= S_IWGRP; 1614 return ((file_mode & mask) == mask ? 0 : EACCES); 1615 } 1616 1617 /* Otherwise, check everyone else. */ 1618 if (acc_mode & VEXEC) 1619 mask |= S_IXOTH; 1620 if (acc_mode & VREAD) 1621 mask |= S_IROTH; 1622 if (acc_mode & VWRITE) 1623 mask |= S_IWOTH; 1624 return ((file_mode & mask) == mask ? 0 : EACCES); 1625 } 1626 1627 #ifdef DDB 1628 #include <ddb/ddb.h> 1629 1630 static int db_show_locked_vnodes(struct mount *mp, void *data); 1631 1632 /* 1633 * List all of the locked vnodes in the system. 1634 * Called when debugging the kernel. 1635 */ 1636 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1637 { 1638 kprintf("Locked vnodes\n"); 1639 mountlist_scan(db_show_locked_vnodes, NULL, 1640 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1641 } 1642 1643 static int 1644 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1645 { 1646 struct vnode *vp; 1647 1648 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1649 if (vn_islocked(vp)) 1650 vprint(NULL, vp); 1651 } 1652 return(0); 1653 } 1654 #endif 1655 1656 /* 1657 * Top level filesystem related information gathering. 1658 */ 1659 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1660 1661 static int 1662 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1663 { 1664 int *name = (int *)arg1 - 1; /* XXX */ 1665 u_int namelen = arg2 + 1; /* XXX */ 1666 struct vfsconf *vfsp; 1667 int maxtypenum; 1668 1669 #if 1 || defined(COMPAT_PRELITE2) 1670 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1671 if (namelen == 1) 1672 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1673 #endif 1674 1675 #ifdef notyet 1676 /* all sysctl names at this level are at least name and field */ 1677 if (namelen < 2) 1678 return (ENOTDIR); /* overloaded */ 1679 if (name[0] != VFS_GENERIC) { 1680 vfsp = vfsconf_find_by_typenum(name[0]); 1681 if (vfsp == NULL) 1682 return (EOPNOTSUPP); 1683 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1684 oldp, oldlenp, newp, newlen, p)); 1685 } 1686 #endif 1687 switch (name[1]) { 1688 case VFS_MAXTYPENUM: 1689 if (namelen != 2) 1690 return (ENOTDIR); 1691 maxtypenum = vfsconf_get_maxtypenum(); 1692 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1693 case VFS_CONF: 1694 if (namelen != 3) 1695 return (ENOTDIR); /* overloaded */ 1696 vfsp = vfsconf_find_by_typenum(name[2]); 1697 if (vfsp == NULL) 1698 return (EOPNOTSUPP); 1699 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1700 } 1701 return (EOPNOTSUPP); 1702 } 1703 1704 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1705 "Generic filesystem"); 1706 1707 #if 1 || defined(COMPAT_PRELITE2) 1708 1709 static int 1710 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1711 { 1712 int error; 1713 struct ovfsconf ovfs; 1714 struct sysctl_req *req = (struct sysctl_req*) data; 1715 1716 bzero(&ovfs, sizeof(ovfs)); 1717 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1718 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1719 ovfs.vfc_index = vfsp->vfc_typenum; 1720 ovfs.vfc_refcount = vfsp->vfc_refcount; 1721 ovfs.vfc_flags = vfsp->vfc_flags; 1722 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1723 if (error) 1724 return error; /* abort iteration with error code */ 1725 else 1726 return 0; /* continue iterating with next element */ 1727 } 1728 1729 static int 1730 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1731 { 1732 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1733 } 1734 1735 #endif /* 1 || COMPAT_PRELITE2 */ 1736 1737 /* 1738 * Check to see if a filesystem is mounted on a block device. 1739 */ 1740 int 1741 vfs_mountedon(struct vnode *vp) 1742 { 1743 cdev_t dev; 1744 1745 if ((dev = vp->v_rdev) == NULL) { 1746 /* if (vp->v_type != VBLK) 1747 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1748 } 1749 if (dev != NULL && dev->si_mountpoint) 1750 return (EBUSY); 1751 return (0); 1752 } 1753 1754 /* 1755 * Unmount all filesystems. The list is traversed in reverse order 1756 * of mounting to avoid dependencies. 1757 */ 1758 1759 static int vfs_umountall_callback(struct mount *mp, void *data); 1760 1761 void 1762 vfs_unmountall(void) 1763 { 1764 int count; 1765 1766 do { 1767 count = mountlist_scan(vfs_umountall_callback, 1768 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1769 } while (count); 1770 } 1771 1772 static 1773 int 1774 vfs_umountall_callback(struct mount *mp, void *data) 1775 { 1776 int error; 1777 1778 error = dounmount(mp, MNT_FORCE); 1779 if (error) { 1780 mountlist_remove(mp); 1781 kprintf("unmount of filesystem mounted from %s failed (", 1782 mp->mnt_stat.f_mntfromname); 1783 if (error == EBUSY) 1784 kprintf("BUSY)\n"); 1785 else 1786 kprintf("%d)\n", error); 1787 } 1788 return(1); 1789 } 1790 1791 /* 1792 * Checks the mount flags for parameter mp and put the names comma-separated 1793 * into a string buffer buf with a size limit specified by len. 1794 * 1795 * It returns the number of bytes written into buf, and (*errorp) will be 1796 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1797 * not large enough). The buffer will be 0-terminated if len was not 0. 1798 */ 1799 size_t 1800 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1801 char *buf, size_t len, int *errorp) 1802 { 1803 static const struct mountctl_opt optnames[] = { 1804 { MNT_ASYNC, "asynchronous" }, 1805 { MNT_EXPORTED, "NFS exported" }, 1806 { MNT_LOCAL, "local" }, 1807 { MNT_NOATIME, "noatime" }, 1808 { MNT_NODEV, "nodev" }, 1809 { MNT_NOEXEC, "noexec" }, 1810 { MNT_NOSUID, "nosuid" }, 1811 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1812 { MNT_QUOTA, "with-quotas" }, 1813 { MNT_RDONLY, "read-only" }, 1814 { MNT_SYNCHRONOUS, "synchronous" }, 1815 { MNT_UNION, "union" }, 1816 { MNT_NOCLUSTERR, "noclusterr" }, 1817 { MNT_NOCLUSTERW, "noclusterw" }, 1818 { MNT_SUIDDIR, "suiddir" }, 1819 { MNT_SOFTDEP, "soft-updates" }, 1820 { MNT_IGNORE, "ignore" }, 1821 { 0, NULL} 1822 }; 1823 int bwritten; 1824 int bleft; 1825 int optlen; 1826 int actsize; 1827 1828 *errorp = 0; 1829 bwritten = 0; 1830 bleft = len - 1; /* leave room for trailing \0 */ 1831 1832 /* 1833 * Checks the size of the string. If it contains 1834 * any data, then we will append the new flags to 1835 * it. 1836 */ 1837 actsize = strlen(buf); 1838 if (actsize > 0) 1839 buf += actsize; 1840 1841 /* Default flags if no flags passed */ 1842 if (optp == NULL) 1843 optp = optnames; 1844 1845 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1846 *errorp = EINVAL; 1847 return(0); 1848 } 1849 1850 for (; flags && optp->o_opt; ++optp) { 1851 if ((flags & optp->o_opt) == 0) 1852 continue; 1853 optlen = strlen(optp->o_name); 1854 if (bwritten || actsize > 0) { 1855 if (bleft < 2) { 1856 *errorp = ENOSPC; 1857 break; 1858 } 1859 buf[bwritten++] = ','; 1860 buf[bwritten++] = ' '; 1861 bleft -= 2; 1862 } 1863 if (bleft < optlen) { 1864 *errorp = ENOSPC; 1865 break; 1866 } 1867 bcopy(optp->o_name, buf + bwritten, optlen); 1868 bwritten += optlen; 1869 bleft -= optlen; 1870 flags &= ~optp->o_opt; 1871 } 1872 1873 /* 1874 * Space already reserved for trailing \0 1875 */ 1876 buf[bwritten] = 0; 1877 return (bwritten); 1878 } 1879 1880 /* 1881 * Build hash lists of net addresses and hang them off the mount point. 1882 * Called by ufs_mount() to set up the lists of export addresses. 1883 */ 1884 static int 1885 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1886 const struct export_args *argp) 1887 { 1888 struct netcred *np; 1889 struct radix_node_head *rnh; 1890 int i; 1891 struct radix_node *rn; 1892 struct sockaddr *saddr, *smask = 0; 1893 struct domain *dom; 1894 int error; 1895 1896 if (argp->ex_addrlen == 0) { 1897 if (mp->mnt_flag & MNT_DEFEXPORTED) 1898 return (EPERM); 1899 np = &nep->ne_defexported; 1900 np->netc_exflags = argp->ex_flags; 1901 np->netc_anon = argp->ex_anon; 1902 np->netc_anon.cr_ref = 1; 1903 mp->mnt_flag |= MNT_DEFEXPORTED; 1904 return (0); 1905 } 1906 1907 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1908 return (EINVAL); 1909 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1910 return (EINVAL); 1911 1912 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1913 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1914 saddr = (struct sockaddr *) (np + 1); 1915 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1916 goto out; 1917 if (saddr->sa_len > argp->ex_addrlen) 1918 saddr->sa_len = argp->ex_addrlen; 1919 if (argp->ex_masklen) { 1920 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1921 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1922 if (error) 1923 goto out; 1924 if (smask->sa_len > argp->ex_masklen) 1925 smask->sa_len = argp->ex_masklen; 1926 } 1927 i = saddr->sa_family; 1928 if ((rnh = nep->ne_rtable[i]) == 0) { 1929 /* 1930 * Seems silly to initialize every AF when most are not used, 1931 * do so on demand here 1932 */ 1933 SLIST_FOREACH(dom, &domains, dom_next) 1934 if (dom->dom_family == i && dom->dom_rtattach) { 1935 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1936 dom->dom_rtoffset); 1937 break; 1938 } 1939 if ((rnh = nep->ne_rtable[i]) == 0) { 1940 error = ENOBUFS; 1941 goto out; 1942 } 1943 } 1944 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1945 np->netc_rnodes); 1946 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1947 error = EPERM; 1948 goto out; 1949 } 1950 np->netc_exflags = argp->ex_flags; 1951 np->netc_anon = argp->ex_anon; 1952 np->netc_anon.cr_ref = 1; 1953 return (0); 1954 out: 1955 kfree(np, M_NETADDR); 1956 return (error); 1957 } 1958 1959 /* ARGSUSED */ 1960 static int 1961 vfs_free_netcred(struct radix_node *rn, void *w) 1962 { 1963 struct radix_node_head *rnh = (struct radix_node_head *) w; 1964 1965 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1966 kfree((caddr_t) rn, M_NETADDR); 1967 return (0); 1968 } 1969 1970 /* 1971 * Free the net address hash lists that are hanging off the mount points. 1972 */ 1973 static void 1974 vfs_free_addrlist(struct netexport *nep) 1975 { 1976 int i; 1977 struct radix_node_head *rnh; 1978 1979 for (i = 0; i <= AF_MAX; i++) 1980 if ((rnh = nep->ne_rtable[i])) { 1981 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1982 (caddr_t) rnh); 1983 kfree((caddr_t) rnh, M_RTABLE); 1984 nep->ne_rtable[i] = 0; 1985 } 1986 } 1987 1988 int 1989 vfs_export(struct mount *mp, struct netexport *nep, 1990 const struct export_args *argp) 1991 { 1992 int error; 1993 1994 if (argp->ex_flags & MNT_DELEXPORT) { 1995 if (mp->mnt_flag & MNT_EXPUBLIC) { 1996 vfs_setpublicfs(NULL, NULL, NULL); 1997 mp->mnt_flag &= ~MNT_EXPUBLIC; 1998 } 1999 vfs_free_addrlist(nep); 2000 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2001 } 2002 if (argp->ex_flags & MNT_EXPORTED) { 2003 if (argp->ex_flags & MNT_EXPUBLIC) { 2004 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2005 return (error); 2006 mp->mnt_flag |= MNT_EXPUBLIC; 2007 } 2008 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2009 return (error); 2010 mp->mnt_flag |= MNT_EXPORTED; 2011 } 2012 return (0); 2013 } 2014 2015 2016 /* 2017 * Set the publicly exported filesystem (WebNFS). Currently, only 2018 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2019 */ 2020 int 2021 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2022 const struct export_args *argp) 2023 { 2024 int error; 2025 struct vnode *rvp; 2026 char *cp; 2027 2028 /* 2029 * mp == NULL -> invalidate the current info, the FS is 2030 * no longer exported. May be called from either vfs_export 2031 * or unmount, so check if it hasn't already been done. 2032 */ 2033 if (mp == NULL) { 2034 if (nfs_pub.np_valid) { 2035 nfs_pub.np_valid = 0; 2036 if (nfs_pub.np_index != NULL) { 2037 FREE(nfs_pub.np_index, M_TEMP); 2038 nfs_pub.np_index = NULL; 2039 } 2040 } 2041 return (0); 2042 } 2043 2044 /* 2045 * Only one allowed at a time. 2046 */ 2047 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2048 return (EBUSY); 2049 2050 /* 2051 * Get real filehandle for root of exported FS. 2052 */ 2053 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2054 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2055 2056 if ((error = VFS_ROOT(mp, &rvp))) 2057 return (error); 2058 2059 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2060 return (error); 2061 2062 vput(rvp); 2063 2064 /* 2065 * If an indexfile was specified, pull it in. 2066 */ 2067 if (argp->ex_indexfile != NULL) { 2068 int namelen; 2069 2070 error = vn_get_namelen(rvp, &namelen); 2071 if (error) 2072 return (error); 2073 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 2074 M_WAITOK); 2075 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2076 namelen, NULL); 2077 if (!error) { 2078 /* 2079 * Check for illegal filenames. 2080 */ 2081 for (cp = nfs_pub.np_index; *cp; cp++) { 2082 if (*cp == '/') { 2083 error = EINVAL; 2084 break; 2085 } 2086 } 2087 } 2088 if (error) { 2089 FREE(nfs_pub.np_index, M_TEMP); 2090 return (error); 2091 } 2092 } 2093 2094 nfs_pub.np_mount = mp; 2095 nfs_pub.np_valid = 1; 2096 return (0); 2097 } 2098 2099 struct netcred * 2100 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2101 struct sockaddr *nam) 2102 { 2103 struct netcred *np; 2104 struct radix_node_head *rnh; 2105 struct sockaddr *saddr; 2106 2107 np = NULL; 2108 if (mp->mnt_flag & MNT_EXPORTED) { 2109 /* 2110 * Lookup in the export list first. 2111 */ 2112 if (nam != NULL) { 2113 saddr = nam; 2114 rnh = nep->ne_rtable[saddr->sa_family]; 2115 if (rnh != NULL) { 2116 np = (struct netcred *) 2117 (*rnh->rnh_matchaddr)((char *)saddr, 2118 rnh); 2119 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2120 np = NULL; 2121 } 2122 } 2123 /* 2124 * If no address match, use the default if it exists. 2125 */ 2126 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2127 np = &nep->ne_defexported; 2128 } 2129 return (np); 2130 } 2131 2132 /* 2133 * perform msync on all vnodes under a mount point. The mount point must 2134 * be locked. This code is also responsible for lazy-freeing unreferenced 2135 * vnodes whos VM objects no longer contain pages. 2136 * 2137 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2138 * 2139 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2140 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2141 * way up in this high level function. 2142 */ 2143 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2144 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2145 2146 void 2147 vfs_msync(struct mount *mp, int flags) 2148 { 2149 int vmsc_flags; 2150 2151 /* 2152 * tmpfs sets this flag to prevent msync(), sync, and the 2153 * filesystem periodic syncer from trying to flush VM pages 2154 * to swap. Only pure memory pressure flushes tmpfs VM pages 2155 * to swap. 2156 */ 2157 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2158 return; 2159 2160 /* 2161 * Ok, scan the vnodes for work. 2162 */ 2163 vmsc_flags = VMSC_GETVP; 2164 if (flags != MNT_WAIT) 2165 vmsc_flags |= VMSC_NOWAIT; 2166 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 2167 (void *)(intptr_t)flags); 2168 } 2169 2170 /* 2171 * scan1 is a fast pre-check. There could be hundreds of thousands of 2172 * vnodes, we cannot afford to do anything heavy weight until we have a 2173 * fairly good indication that there is work to do. 2174 */ 2175 static 2176 int 2177 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2178 { 2179 int flags = (int)(intptr_t)data; 2180 2181 if ((vp->v_flag & VRECLAIMED) == 0) { 2182 if (vshouldmsync(vp)) 2183 return(0); /* call scan2 */ 2184 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2185 (vp->v_flag & VOBJDIRTY) && 2186 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2187 return(0); /* call scan2 */ 2188 } 2189 } 2190 2191 /* 2192 * do not call scan2, continue the loop 2193 */ 2194 return(-1); 2195 } 2196 2197 /* 2198 * This callback is handed a locked vnode. 2199 */ 2200 static 2201 int 2202 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2203 { 2204 vm_object_t obj; 2205 int flags = (int)(intptr_t)data; 2206 2207 if (vp->v_flag & VRECLAIMED) 2208 return(0); 2209 2210 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2211 if ((obj = vp->v_object) != NULL) { 2212 vm_object_page_clean(obj, 0, 0, 2213 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2214 } 2215 } 2216 return(0); 2217 } 2218 2219 /* 2220 * Wake up anyone interested in vp because it is being revoked. 2221 */ 2222 void 2223 vn_gone(struct vnode *vp) 2224 { 2225 lwkt_gettoken(&vp->v_token); 2226 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2227 lwkt_reltoken(&vp->v_token); 2228 } 2229 2230 /* 2231 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2232 * (or v_rdev might be NULL). 2233 */ 2234 cdev_t 2235 vn_todev(struct vnode *vp) 2236 { 2237 if (vp->v_type != VBLK && vp->v_type != VCHR) 2238 return (NULL); 2239 KKASSERT(vp->v_rdev != NULL); 2240 return (vp->v_rdev); 2241 } 2242 2243 /* 2244 * Check if vnode represents a disk device. The vnode does not need to be 2245 * opened. 2246 * 2247 * MPALMOSTSAFE 2248 */ 2249 int 2250 vn_isdisk(struct vnode *vp, int *errp) 2251 { 2252 cdev_t dev; 2253 2254 if (vp->v_type != VCHR) { 2255 if (errp != NULL) 2256 *errp = ENOTBLK; 2257 return (0); 2258 } 2259 2260 dev = vp->v_rdev; 2261 2262 if (dev == NULL) { 2263 if (errp != NULL) 2264 *errp = ENXIO; 2265 return (0); 2266 } 2267 if (dev_is_good(dev) == 0) { 2268 if (errp != NULL) 2269 *errp = ENXIO; 2270 return (0); 2271 } 2272 if ((dev_dflags(dev) & D_DISK) == 0) { 2273 if (errp != NULL) 2274 *errp = ENOTBLK; 2275 return (0); 2276 } 2277 if (errp != NULL) 2278 *errp = 0; 2279 return (1); 2280 } 2281 2282 int 2283 vn_get_namelen(struct vnode *vp, int *namelen) 2284 { 2285 int error; 2286 register_t retval[2]; 2287 2288 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2289 if (error) 2290 return (error); 2291 *namelen = (int)retval[0]; 2292 return (0); 2293 } 2294 2295 int 2296 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2297 uint16_t d_namlen, const char *d_name) 2298 { 2299 struct dirent *dp; 2300 size_t len; 2301 2302 len = _DIRENT_RECLEN(d_namlen); 2303 if (len > uio->uio_resid) 2304 return(1); 2305 2306 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2307 2308 dp->d_ino = d_ino; 2309 dp->d_namlen = d_namlen; 2310 dp->d_type = d_type; 2311 bcopy(d_name, dp->d_name, d_namlen); 2312 2313 *error = uiomove((caddr_t)dp, len, uio); 2314 2315 kfree(dp, M_TEMP); 2316 2317 return(0); 2318 } 2319 2320 void 2321 vn_mark_atime(struct vnode *vp, struct thread *td) 2322 { 2323 struct proc *p = td->td_proc; 2324 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2325 2326 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2327 VOP_MARKATIME(vp, cred); 2328 } 2329 } 2330