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