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