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 */ 41 42 /* 43 * External virtual filesystem routines 44 */ 45 #include "opt_ddb.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/buf.h> 50 #include <sys/conf.h> 51 #include <sys/dirent.h> 52 #include <sys/domain.h> 53 #include <sys/eventhandler.h> 54 #include <sys/fcntl.h> 55 #include <sys/file.h> 56 #include <sys/kernel.h> 57 #include <sys/kthread.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/mount.h> 61 #include <sys/priv.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 #include <sys/mplock2.h> 89 90 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 91 92 int numvnodes; 93 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 94 "Number of vnodes allocated"); 95 int verbose_reclaims; 96 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0, 97 "Output filename of reclaimed vnode(s)"); 98 99 enum vtype iftovt_tab[16] = { 100 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 101 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 102 }; 103 int vttoif_tab[9] = { 104 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 105 S_IFSOCK, S_IFIFO, S_IFMT, 106 }; 107 108 static int reassignbufcalls; 109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 110 0, "Number of times buffers have been reassigned to the proper list"); 111 112 static int check_buf_overlap = 2; /* invasive check */ 113 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap, 114 0, "Enable overlapping buffer checks"); 115 116 int nfs_mount_type = -1; 117 static struct lwkt_token spechash_token; 118 struct nfs_public nfs_pub; /* publicly exported FS */ 119 120 int desiredvnodes; 121 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 122 &desiredvnodes, 0, "Maximum number of vnodes"); 123 124 static void vfs_free_addrlist (struct netexport *nep); 125 static int vfs_free_netcred (struct radix_node *rn, void *w); 126 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 127 const struct export_args *argp); 128 129 /* 130 * Red black tree functions 131 */ 132 static int rb_buf_compare(struct buf *b1, struct buf *b2); 133 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset); 134 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset); 135 136 static int 137 rb_buf_compare(struct buf *b1, struct buf *b2) 138 { 139 if (b1->b_loffset < b2->b_loffset) 140 return(-1); 141 if (b1->b_loffset > b2->b_loffset) 142 return(1); 143 return(0); 144 } 145 146 /* 147 * Returns non-zero if the vnode is a candidate for lazy msyncing. 148 * 149 * NOTE: v_object is not stable (this scan can race), however the 150 * mntvnodescan code holds vmobj_token so any VM object we 151 * do find will remain stable storage. 152 */ 153 static __inline int 154 vshouldmsync(struct vnode *vp) 155 { 156 vm_object_t object; 157 158 if (vp->v_auxrefs != 0 || vp->v_sysref.refcnt > 0) 159 return (0); /* other holders */ 160 object = vp->v_object; 161 cpu_ccfence(); 162 if (object && (object->ref_count || object->resident_page_count)) 163 return(0); 164 return (1); 165 } 166 167 /* 168 * Initialize the vnode management data structures. 169 * 170 * Called from vfsinit() 171 */ 172 void 173 vfs_subr_init(void) 174 { 175 int factor1; 176 int factor2; 177 178 /* 179 * Desiredvnodes is kern.maxvnodes. We want to scale it 180 * according to available system memory but we may also have 181 * to limit it based on available KVM, which is capped on 32 bit 182 * systems. 183 * 184 * WARNING! For machines with 64-256M of ram we have to be sure 185 * that the default limit scales down well due to HAMMER 186 * taking up significantly more memory per-vnode vs UFS. 187 * We want around ~5800 on a 128M machine. 188 */ 189 factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode)); 190 factor2 = 22 * (sizeof(struct vm_object) + sizeof(struct vnode)); 191 desiredvnodes = 192 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1, 193 KvaSize / factor2); 194 desiredvnodes = imax(desiredvnodes, maxproc * 8); 195 196 lwkt_token_init(&spechash_token, "spechash"); 197 } 198 199 /* 200 * Knob to control the precision of file timestamps: 201 * 202 * 0 = seconds only; nanoseconds zeroed. 203 * 1 = seconds and nanoseconds, accurate within 1/HZ. 204 * 2 = seconds and nanoseconds, truncated to microseconds. 205 * >=3 = seconds and nanoseconds, maximum precision. 206 */ 207 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 208 209 static int timestamp_precision = TSP_SEC; 210 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 211 ×tamp_precision, 0, "Precision of file timestamps"); 212 213 /* 214 * Get a current timestamp. 215 * 216 * MPSAFE 217 */ 218 void 219 vfs_timestamp(struct timespec *tsp) 220 { 221 struct timeval tv; 222 223 switch (timestamp_precision) { 224 case TSP_SEC: 225 tsp->tv_sec = time_second; 226 tsp->tv_nsec = 0; 227 break; 228 case TSP_HZ: 229 getnanotime(tsp); 230 break; 231 case TSP_USEC: 232 microtime(&tv); 233 TIMEVAL_TO_TIMESPEC(&tv, tsp); 234 break; 235 case TSP_NSEC: 236 default: 237 nanotime(tsp); 238 break; 239 } 240 } 241 242 /* 243 * Set vnode attributes to VNOVAL 244 */ 245 void 246 vattr_null(struct vattr *vap) 247 { 248 vap->va_type = VNON; 249 vap->va_size = VNOVAL; 250 vap->va_bytes = VNOVAL; 251 vap->va_mode = VNOVAL; 252 vap->va_nlink = VNOVAL; 253 vap->va_uid = VNOVAL; 254 vap->va_gid = VNOVAL; 255 vap->va_fsid = VNOVAL; 256 vap->va_fileid = VNOVAL; 257 vap->va_blocksize = VNOVAL; 258 vap->va_rmajor = VNOVAL; 259 vap->va_rminor = VNOVAL; 260 vap->va_atime.tv_sec = VNOVAL; 261 vap->va_atime.tv_nsec = VNOVAL; 262 vap->va_mtime.tv_sec = VNOVAL; 263 vap->va_mtime.tv_nsec = VNOVAL; 264 vap->va_ctime.tv_sec = VNOVAL; 265 vap->va_ctime.tv_nsec = VNOVAL; 266 vap->va_flags = VNOVAL; 267 vap->va_gen = VNOVAL; 268 vap->va_vaflags = 0; 269 /* va_*_uuid fields are only valid if related flags are set */ 270 } 271 272 /* 273 * Flush out and invalidate all buffers associated with a vnode. 274 * 275 * vp must be locked. 276 */ 277 static int vinvalbuf_bp(struct buf *bp, void *data); 278 279 struct vinvalbuf_bp_info { 280 struct vnode *vp; 281 int slptimeo; 282 int lkflags; 283 int flags; 284 int clean; 285 }; 286 287 int 288 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 289 { 290 struct vinvalbuf_bp_info info; 291 vm_object_t object; 292 int error; 293 294 lwkt_gettoken(&vp->v_token); 295 296 /* 297 * If we are being asked to save, call fsync to ensure that the inode 298 * is updated. 299 */ 300 if (flags & V_SAVE) { 301 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo); 302 if (error) 303 goto done; 304 if (!RB_EMPTY(&vp->v_rbdirty_tree)) { 305 if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0) 306 goto done; 307 308 /* 309 * Dirty bufs may be left or generated via races 310 * in circumstances where vinvalbuf() is called on 311 * a vnode not undergoing reclamation. Only 312 * panic if we are trying to reclaim the vnode. 313 */ 314 if ((vp->v_flag & VRECLAIMED) && 315 (bio_track_active(&vp->v_track_write) || 316 !RB_EMPTY(&vp->v_rbdirty_tree))) { 317 panic("vinvalbuf: dirty bufs"); 318 } 319 } 320 } 321 info.slptimeo = slptimeo; 322 info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL; 323 if (slpflag & PCATCH) 324 info.lkflags |= LK_PCATCH; 325 info.flags = flags; 326 info.vp = vp; 327 328 /* 329 * Flush the buffer cache until nothing is left. 330 */ 331 while (!RB_EMPTY(&vp->v_rbclean_tree) || 332 !RB_EMPTY(&vp->v_rbdirty_tree)) { 333 info.clean = 1; 334 error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, NULL, 335 vinvalbuf_bp, &info); 336 if (error == 0) { 337 info.clean = 0; 338 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 339 vinvalbuf_bp, &info); 340 } 341 } 342 343 /* 344 * Wait for I/O completion. We may block in the pip code so we have 345 * to re-check. 346 */ 347 do { 348 bio_track_wait(&vp->v_track_write, 0, 0); 349 if ((object = vp->v_object) != NULL) { 350 while (object->paging_in_progress) 351 vm_object_pip_sleep(object, "vnvlbx"); 352 } 353 } while (bio_track_active(&vp->v_track_write)); 354 355 /* 356 * Destroy the copy in the VM cache, too. 357 */ 358 if ((object = vp->v_object) != NULL) { 359 vm_object_page_remove(object, 0, 0, 360 (flags & V_SAVE) ? TRUE : FALSE); 361 } 362 363 if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree)) 364 panic("vinvalbuf: flush failed"); 365 if (!RB_EMPTY(&vp->v_rbhash_tree)) 366 panic("vinvalbuf: flush failed, buffers still present"); 367 error = 0; 368 done: 369 lwkt_reltoken(&vp->v_token); 370 return (error); 371 } 372 373 static int 374 vinvalbuf_bp(struct buf *bp, void *data) 375 { 376 struct vinvalbuf_bp_info *info = data; 377 int error; 378 379 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 380 atomic_add_int(&bp->b_refs, 1); 381 error = BUF_TIMELOCK(bp, info->lkflags, 382 "vinvalbuf", info->slptimeo); 383 atomic_subtract_int(&bp->b_refs, 1); 384 if (error == 0) { 385 BUF_UNLOCK(bp); 386 error = ENOLCK; 387 } 388 if (error == ENOLCK) 389 return(0); 390 return (-error); 391 } 392 KKASSERT(bp->b_vp == info->vp); 393 394 /* 395 * Must check clean/dirty status after successfully locking as 396 * it may race. 397 */ 398 if ((info->clean && (bp->b_flags & B_DELWRI)) || 399 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) { 400 BUF_UNLOCK(bp); 401 return(0); 402 } 403 404 /* 405 * Note that vfs_bio_awrite expects buffers to reside 406 * on a queue, while bwrite() and brelse() do not. 407 * 408 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite() 409 * check. This code will write out the buffer, period. 410 */ 411 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 412 (info->flags & V_SAVE)) { 413 if (bp->b_flags & B_CLUSTEROK) { 414 vfs_bio_awrite(bp); 415 } else { 416 bremfree(bp); 417 bawrite(bp); 418 } 419 } else if (info->flags & V_SAVE) { 420 /* 421 * Cannot set B_NOCACHE on a clean buffer as this will 422 * destroy the VM backing store which might actually 423 * be dirty (and unsynchronized). 424 */ 425 bremfree(bp); 426 bp->b_flags |= (B_INVAL | B_RELBUF); 427 brelse(bp); 428 } else { 429 bremfree(bp); 430 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 431 brelse(bp); 432 } 433 return(0); 434 } 435 436 /* 437 * Truncate a file's buffer and pages to a specified length. This 438 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 439 * sync activity. 440 * 441 * The vnode must be locked. 442 */ 443 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data); 444 static int vtruncbuf_bp_trunc(struct buf *bp, void *data); 445 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data); 446 static int vtruncbuf_bp_metasync(struct buf *bp, void *data); 447 448 struct vtruncbuf_info { 449 struct vnode *vp; 450 off_t truncloffset; 451 int clean; 452 }; 453 454 int 455 vtruncbuf(struct vnode *vp, off_t length, int blksize) 456 { 457 struct vtruncbuf_info info; 458 const char *filename; 459 int count; 460 461 /* 462 * Round up to the *next* block, then destroy the buffers in question. 463 * Since we are only removing some of the buffers we must rely on the 464 * scan count to determine whether a loop is necessary. 465 */ 466 if ((count = (int)(length % blksize)) != 0) 467 info.truncloffset = length + (blksize - count); 468 else 469 info.truncloffset = length; 470 info.vp = vp; 471 472 lwkt_gettoken(&vp->v_token); 473 do { 474 info.clean = 1; 475 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 476 vtruncbuf_bp_trunc_cmp, 477 vtruncbuf_bp_trunc, &info); 478 info.clean = 0; 479 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 480 vtruncbuf_bp_trunc_cmp, 481 vtruncbuf_bp_trunc, &info); 482 } while(count); 483 484 /* 485 * For safety, fsync any remaining metadata if the file is not being 486 * truncated to 0. Since the metadata does not represent the entire 487 * dirty list we have to rely on the hit count to ensure that we get 488 * all of it. 489 */ 490 if (length > 0) { 491 do { 492 count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 493 vtruncbuf_bp_metasync_cmp, 494 vtruncbuf_bp_metasync, &info); 495 } while (count); 496 } 497 498 /* 499 * Clean out any left over VM backing store. 500 * 501 * It is possible to have in-progress I/O from buffers that were 502 * not part of the truncation. This should not happen if we 503 * are truncating to 0-length. 504 */ 505 vnode_pager_setsize(vp, length); 506 bio_track_wait(&vp->v_track_write, 0, 0); 507 508 /* 509 * Debugging only 510 */ 511 spin_lock(&vp->v_spinlock); 512 filename = TAILQ_FIRST(&vp->v_namecache) ? 513 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 514 spin_unlock(&vp->v_spinlock); 515 516 /* 517 * Make sure no buffers were instantiated while we were trying 518 * to clean out the remaining VM pages. This could occur due 519 * to busy dirty VM pages being flushed out to disk. 520 */ 521 do { 522 info.clean = 1; 523 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 524 vtruncbuf_bp_trunc_cmp, 525 vtruncbuf_bp_trunc, &info); 526 info.clean = 0; 527 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 528 vtruncbuf_bp_trunc_cmp, 529 vtruncbuf_bp_trunc, &info); 530 if (count) { 531 kprintf("Warning: vtruncbuf(): Had to re-clean %d " 532 "left over buffers in %s\n", count, filename); 533 } 534 } while(count); 535 536 lwkt_reltoken(&vp->v_token); 537 538 return (0); 539 } 540 541 /* 542 * The callback buffer is beyond the new file EOF and must be destroyed. 543 * Note that the compare function must conform to the RB_SCAN's requirements. 544 */ 545 static 546 int 547 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data) 548 { 549 struct vtruncbuf_info *info = data; 550 551 if (bp->b_loffset >= info->truncloffset) 552 return(0); 553 return(-1); 554 } 555 556 static 557 int 558 vtruncbuf_bp_trunc(struct buf *bp, void *data) 559 { 560 struct vtruncbuf_info *info = data; 561 562 /* 563 * Do not try to use a buffer we cannot immediately lock, but sleep 564 * anyway to prevent a livelock. The code will loop until all buffers 565 * can be acted upon. 566 * 567 * We must always revalidate the buffer after locking it to deal 568 * with MP races. 569 */ 570 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 571 atomic_add_int(&bp->b_refs, 1); 572 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 573 BUF_UNLOCK(bp); 574 atomic_subtract_int(&bp->b_refs, 1); 575 } else if ((info->clean && (bp->b_flags & B_DELWRI)) || 576 (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) || 577 bp->b_vp != info->vp || 578 vtruncbuf_bp_trunc_cmp(bp, data)) { 579 BUF_UNLOCK(bp); 580 } else { 581 bremfree(bp); 582 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE); 583 brelse(bp); 584 } 585 return(1); 586 } 587 588 /* 589 * Fsync all meta-data after truncating a file to be non-zero. Only metadata 590 * blocks (with a negative loffset) are scanned. 591 * Note that the compare function must conform to the RB_SCAN's requirements. 592 */ 593 static int 594 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused) 595 { 596 if (bp->b_loffset < 0) 597 return(0); 598 return(1); 599 } 600 601 static int 602 vtruncbuf_bp_metasync(struct buf *bp, void *data) 603 { 604 struct vtruncbuf_info *info = data; 605 606 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 607 atomic_add_int(&bp->b_refs, 1); 608 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0) 609 BUF_UNLOCK(bp); 610 atomic_subtract_int(&bp->b_refs, 1); 611 } else if ((bp->b_flags & B_DELWRI) == 0 || 612 bp->b_vp != info->vp || 613 vtruncbuf_bp_metasync_cmp(bp, data)) { 614 BUF_UNLOCK(bp); 615 } else { 616 bremfree(bp); 617 if (bp->b_vp == info->vp) 618 bawrite(bp); 619 else 620 bwrite(bp); 621 } 622 return(1); 623 } 624 625 /* 626 * vfsync - implements a multipass fsync on a file which understands 627 * dependancies and meta-data. The passed vnode must be locked. The 628 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY. 629 * 630 * When fsyncing data asynchronously just do one consolidated pass starting 631 * with the most negative block number. This may not get all the data due 632 * to dependancies. 633 * 634 * When fsyncing data synchronously do a data pass, then a metadata pass, 635 * then do additional data+metadata passes to try to get all the data out. 636 */ 637 static int vfsync_wait_output(struct vnode *vp, 638 int (*waitoutput)(struct vnode *, struct thread *)); 639 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused); 640 static int vfsync_data_only_cmp(struct buf *bp, void *data); 641 static int vfsync_meta_only_cmp(struct buf *bp, void *data); 642 static int vfsync_lazy_range_cmp(struct buf *bp, void *data); 643 static int vfsync_bp(struct buf *bp, void *data); 644 645 struct vfsync_info { 646 struct vnode *vp; 647 int synchronous; 648 int syncdeps; 649 int lazycount; 650 int lazylimit; 651 int skippedbufs; 652 int (*checkdef)(struct buf *); 653 int (*cmpfunc)(struct buf *, void *); 654 }; 655 656 int 657 vfsync(struct vnode *vp, int waitfor, int passes, 658 int (*checkdef)(struct buf *), 659 int (*waitoutput)(struct vnode *, struct thread *)) 660 { 661 struct vfsync_info info; 662 int error; 663 664 bzero(&info, sizeof(info)); 665 info.vp = vp; 666 if ((info.checkdef = checkdef) == NULL) 667 info.syncdeps = 1; 668 669 lwkt_gettoken(&vp->v_token); 670 671 switch(waitfor) { 672 case MNT_LAZY | MNT_NOWAIT: 673 case MNT_LAZY: 674 /* 675 * Lazy (filesystem syncer typ) Asynchronous plus limit the 676 * number of data (not meta) pages we try to flush to 1MB. 677 * A non-zero return means that lazy limit was reached. 678 */ 679 info.lazylimit = 1024 * 1024; 680 info.syncdeps = 1; 681 info.cmpfunc = vfsync_lazy_range_cmp; 682 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 683 vfsync_lazy_range_cmp, vfsync_bp, &info); 684 info.cmpfunc = vfsync_meta_only_cmp; 685 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 686 vfsync_meta_only_cmp, vfsync_bp, &info); 687 if (error == 0) 688 vp->v_lazyw = 0; 689 else if (!RB_EMPTY(&vp->v_rbdirty_tree)) 690 vn_syncer_add(vp, 1); 691 error = 0; 692 break; 693 case MNT_NOWAIT: 694 /* 695 * Asynchronous. Do a data-only pass and a meta-only pass. 696 */ 697 info.syncdeps = 1; 698 info.cmpfunc = vfsync_data_only_cmp; 699 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 700 vfsync_bp, &info); 701 info.cmpfunc = vfsync_meta_only_cmp; 702 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 703 vfsync_bp, &info); 704 error = 0; 705 break; 706 default: 707 /* 708 * Synchronous. Do a data-only pass, then a meta-data+data 709 * pass, then additional integrated passes to try to get 710 * all the dependancies flushed. 711 */ 712 info.cmpfunc = vfsync_data_only_cmp; 713 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 714 vfsync_bp, &info); 715 error = vfsync_wait_output(vp, waitoutput); 716 if (error == 0) { 717 info.skippedbufs = 0; 718 info.cmpfunc = vfsync_dummy_cmp; 719 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 720 vfsync_bp, &info); 721 error = vfsync_wait_output(vp, waitoutput); 722 if (info.skippedbufs) { 723 kprintf("Warning: vfsync skipped %d dirty " 724 "bufs in pass2!\n", info.skippedbufs); 725 } 726 } 727 while (error == 0 && passes > 0 && 728 !RB_EMPTY(&vp->v_rbdirty_tree) 729 ) { 730 if (--passes == 0) { 731 info.synchronous = 1; 732 info.syncdeps = 1; 733 } 734 info.cmpfunc = vfsync_dummy_cmp; 735 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, 736 vfsync_bp, &info); 737 if (error < 0) 738 error = -error; 739 info.syncdeps = 1; 740 if (error == 0) 741 error = vfsync_wait_output(vp, waitoutput); 742 } 743 break; 744 } 745 lwkt_reltoken(&vp->v_token); 746 return(error); 747 } 748 749 static int 750 vfsync_wait_output(struct vnode *vp, 751 int (*waitoutput)(struct vnode *, struct thread *)) 752 { 753 int error; 754 755 error = bio_track_wait(&vp->v_track_write, 0, 0); 756 if (waitoutput) 757 error = waitoutput(vp, curthread); 758 return(error); 759 } 760 761 static int 762 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused) 763 { 764 return(0); 765 } 766 767 static int 768 vfsync_data_only_cmp(struct buf *bp, void *data) 769 { 770 if (bp->b_loffset < 0) 771 return(-1); 772 return(0); 773 } 774 775 static int 776 vfsync_meta_only_cmp(struct buf *bp, void *data) 777 { 778 if (bp->b_loffset < 0) 779 return(0); 780 return(1); 781 } 782 783 static int 784 vfsync_lazy_range_cmp(struct buf *bp, void *data) 785 { 786 struct vfsync_info *info = data; 787 788 if (bp->b_loffset < info->vp->v_lazyw) 789 return(-1); 790 return(0); 791 } 792 793 static int 794 vfsync_bp(struct buf *bp, void *data) 795 { 796 struct vfsync_info *info = data; 797 struct vnode *vp = info->vp; 798 int error; 799 800 /* 801 * Ignore buffers that we cannot immediately lock. 802 */ 803 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 804 ++info->skippedbufs; 805 return(0); 806 } 807 808 /* 809 * We must revalidate the buffer after locking. 810 */ 811 if ((bp->b_flags & B_DELWRI) == 0 || 812 bp->b_vp != info->vp || 813 info->cmpfunc(bp, data)) { 814 BUF_UNLOCK(bp); 815 return(0); 816 } 817 818 /* 819 * If syncdeps is not set we do not try to write buffers which have 820 * dependancies. 821 */ 822 if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) { 823 BUF_UNLOCK(bp); 824 return(0); 825 } 826 827 /* 828 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer 829 * has been written but an additional handshake with the device 830 * is required before we can dispose of the buffer. We have no idea 831 * how to do this so we have to skip these buffers. 832 */ 833 if (bp->b_flags & B_NEEDCOMMIT) { 834 BUF_UNLOCK(bp); 835 return(0); 836 } 837 838 /* 839 * Ask bioops if it is ok to sync. If not the VFS may have 840 * set B_LOCKED so we have to cycle the buffer. 841 */ 842 if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) { 843 bremfree(bp); 844 brelse(bp); 845 return(0); 846 } 847 848 if (info->synchronous) { 849 /* 850 * Synchronous flushing. An error may be returned. 851 */ 852 bremfree(bp); 853 error = bwrite(bp); 854 } else { 855 /* 856 * Asynchronous flushing. A negative return value simply 857 * stops the scan and is not considered an error. We use 858 * this to support limited MNT_LAZY flushes. 859 */ 860 vp->v_lazyw = bp->b_loffset; 861 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) { 862 info->lazycount += vfs_bio_awrite(bp); 863 } else { 864 info->lazycount += bp->b_bufsize; 865 bremfree(bp); 866 bawrite(bp); 867 } 868 waitrunningbufspace(); 869 if (info->lazylimit && info->lazycount >= info->lazylimit) 870 error = 1; 871 else 872 error = 0; 873 } 874 return(-error); 875 } 876 877 /* 878 * Associate a buffer with a vnode. 879 * 880 * MPSAFE 881 */ 882 int 883 bgetvp(struct vnode *vp, struct buf *bp, int testsize) 884 { 885 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 886 KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0); 887 888 /* 889 * Insert onto list for new vnode. 890 */ 891 lwkt_gettoken(&vp->v_token); 892 893 if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) { 894 lwkt_reltoken(&vp->v_token); 895 return (EEXIST); 896 } 897 898 /* 899 * Diagnostics (mainly for HAMMER debugging). Check for 900 * overlapping buffers. 901 */ 902 if (check_buf_overlap) { 903 struct buf *bx; 904 bx = buf_rb_hash_RB_PREV(bp); 905 if (bx) { 906 if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) { 907 kprintf("bgetvp: overlapl %016jx/%d %016jx " 908 "bx %p bp %p\n", 909 (intmax_t)bx->b_loffset, 910 bx->b_bufsize, 911 (intmax_t)bp->b_loffset, 912 bx, bp); 913 if (check_buf_overlap > 1) 914 panic("bgetvp - overlapping buffer"); 915 } 916 } 917 bx = buf_rb_hash_RB_NEXT(bp); 918 if (bx) { 919 if (bp->b_loffset + testsize > bx->b_loffset) { 920 kprintf("bgetvp: overlapr %016jx/%d %016jx " 921 "bp %p bx %p\n", 922 (intmax_t)bp->b_loffset, 923 testsize, 924 (intmax_t)bx->b_loffset, 925 bp, bx); 926 if (check_buf_overlap > 1) 927 panic("bgetvp - overlapping buffer"); 928 } 929 } 930 } 931 bp->b_vp = vp; 932 bp->b_flags |= B_HASHED; 933 bp->b_flags |= B_VNCLEAN; 934 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) 935 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp); 936 vhold(vp); 937 lwkt_reltoken(&vp->v_token); 938 return(0); 939 } 940 941 /* 942 * Disassociate a buffer from a vnode. 943 * 944 * MPSAFE 945 */ 946 void 947 brelvp(struct buf *bp) 948 { 949 struct vnode *vp; 950 951 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 952 953 /* 954 * Delete from old vnode list, if on one. 955 */ 956 vp = bp->b_vp; 957 lwkt_gettoken(&vp->v_token); 958 if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) { 959 if (bp->b_flags & B_VNDIRTY) 960 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 961 else 962 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 963 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN); 964 } 965 if (bp->b_flags & B_HASHED) { 966 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp); 967 bp->b_flags &= ~B_HASHED; 968 } 969 if ((vp->v_flag & VONWORKLST) && RB_EMPTY(&vp->v_rbdirty_tree)) 970 vn_syncer_remove(vp); 971 bp->b_vp = NULL; 972 973 lwkt_reltoken(&vp->v_token); 974 975 vdrop(vp); 976 } 977 978 /* 979 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI. 980 * This routine is called when the state of the B_DELWRI bit is changed. 981 * 982 * Must be called with vp->v_token held. 983 * MPSAFE 984 */ 985 void 986 reassignbuf(struct buf *bp) 987 { 988 struct vnode *vp = bp->b_vp; 989 int delay; 990 991 ASSERT_LWKT_TOKEN_HELD(&vp->v_token); 992 ++reassignbufcalls; 993 994 /* 995 * B_PAGING flagged buffers cannot be reassigned because their vp 996 * is not fully linked in. 997 */ 998 if (bp->b_flags & B_PAGING) 999 panic("cannot reassign paging buffer"); 1000 1001 if (bp->b_flags & B_DELWRI) { 1002 /* 1003 * Move to the dirty list, add the vnode to the worklist 1004 */ 1005 if (bp->b_flags & B_VNCLEAN) { 1006 buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp); 1007 bp->b_flags &= ~B_VNCLEAN; 1008 } 1009 if ((bp->b_flags & B_VNDIRTY) == 0) { 1010 if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) { 1011 panic("reassignbuf: dup lblk vp %p bp %p", 1012 vp, bp); 1013 } 1014 bp->b_flags |= B_VNDIRTY; 1015 } 1016 if ((vp->v_flag & VONWORKLST) == 0) { 1017 switch (vp->v_type) { 1018 case VDIR: 1019 delay = dirdelay; 1020 break; 1021 case VCHR: 1022 case VBLK: 1023 if (vp->v_rdev && 1024 vp->v_rdev->si_mountpoint != NULL) { 1025 delay = metadelay; 1026 break; 1027 } 1028 /* fall through */ 1029 default: 1030 delay = filedelay; 1031 } 1032 vn_syncer_add(vp, delay); 1033 } 1034 } else { 1035 /* 1036 * Move to the clean list, remove the vnode from the worklist 1037 * if no dirty blocks remain. 1038 */ 1039 if (bp->b_flags & B_VNDIRTY) { 1040 buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp); 1041 bp->b_flags &= ~B_VNDIRTY; 1042 } 1043 if ((bp->b_flags & B_VNCLEAN) == 0) { 1044 if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) { 1045 panic("reassignbuf: dup lblk vp %p bp %p", 1046 vp, bp); 1047 } 1048 bp->b_flags |= B_VNCLEAN; 1049 } 1050 if ((vp->v_flag & VONWORKLST) && 1051 RB_EMPTY(&vp->v_rbdirty_tree)) { 1052 vn_syncer_remove(vp); 1053 } 1054 } 1055 } 1056 1057 /* 1058 * Create a vnode for a block device. 1059 * Used for mounting the root file system. 1060 */ 1061 extern struct vop_ops *devfs_vnode_dev_vops_p; 1062 int 1063 bdevvp(cdev_t dev, struct vnode **vpp) 1064 { 1065 struct vnode *vp; 1066 struct vnode *nvp; 1067 int error; 1068 1069 if (dev == NULL) { 1070 *vpp = NULLVP; 1071 return (ENXIO); 1072 } 1073 error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p, 1074 &nvp, 0, 0); 1075 if (error) { 1076 *vpp = NULLVP; 1077 return (error); 1078 } 1079 vp = nvp; 1080 vp->v_type = VCHR; 1081 #if 0 1082 vp->v_rdev = dev; 1083 #endif 1084 v_associate_rdev(vp, dev); 1085 vp->v_umajor = dev->si_umajor; 1086 vp->v_uminor = dev->si_uminor; 1087 vx_unlock(vp); 1088 *vpp = vp; 1089 return (0); 1090 } 1091 1092 int 1093 v_associate_rdev(struct vnode *vp, cdev_t dev) 1094 { 1095 if (dev == NULL) 1096 return(ENXIO); 1097 if (dev_is_good(dev) == 0) 1098 return(ENXIO); 1099 KKASSERT(vp->v_rdev == NULL); 1100 vp->v_rdev = reference_dev(dev); 1101 lwkt_gettoken(&spechash_token); 1102 SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext); 1103 lwkt_reltoken(&spechash_token); 1104 return(0); 1105 } 1106 1107 void 1108 v_release_rdev(struct vnode *vp) 1109 { 1110 cdev_t dev; 1111 1112 if ((dev = vp->v_rdev) != NULL) { 1113 lwkt_gettoken(&spechash_token); 1114 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext); 1115 vp->v_rdev = NULL; 1116 release_dev(dev); 1117 lwkt_reltoken(&spechash_token); 1118 } 1119 } 1120 1121 /* 1122 * Add a vnode to the alias list hung off the cdev_t. We only associate 1123 * the device number with the vnode. The actual device is not associated 1124 * until the vnode is opened (usually in spec_open()), and will be 1125 * disassociated on last close. 1126 */ 1127 void 1128 addaliasu(struct vnode *nvp, int x, int y) 1129 { 1130 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1131 panic("addaliasu on non-special vnode"); 1132 nvp->v_umajor = x; 1133 nvp->v_uminor = y; 1134 } 1135 1136 /* 1137 * Simple call that a filesystem can make to try to get rid of a 1138 * vnode. It will fail if anyone is referencing the vnode (including 1139 * the caller). 1140 * 1141 * The filesystem can check whether its in-memory inode structure still 1142 * references the vp on return. 1143 */ 1144 void 1145 vclean_unlocked(struct vnode *vp) 1146 { 1147 vx_get(vp); 1148 if (sysref_isactive(&vp->v_sysref) == 0) 1149 vgone_vxlocked(vp); 1150 vx_put(vp); 1151 } 1152 1153 /* 1154 * Disassociate a vnode from its underlying filesystem. 1155 * 1156 * The vnode must be VX locked and referenced. In all normal situations 1157 * there are no active references. If vclean_vxlocked() is called while 1158 * there are active references, the vnode is being ripped out and we have 1159 * to call VOP_CLOSE() as appropriate before we can reclaim it. 1160 */ 1161 void 1162 vclean_vxlocked(struct vnode *vp, int flags) 1163 { 1164 int active; 1165 int n; 1166 vm_object_t object; 1167 struct namecache *ncp; 1168 1169 /* 1170 * If the vnode has already been reclaimed we have nothing to do. 1171 */ 1172 if (vp->v_flag & VRECLAIMED) 1173 return; 1174 vsetflags(vp, VRECLAIMED); 1175 1176 if (verbose_reclaims) { 1177 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL) 1178 kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name); 1179 } 1180 1181 /* 1182 * Scrap the vfs cache 1183 */ 1184 while (cache_inval_vp(vp, 0) != 0) { 1185 kprintf("Warning: vnode %p clean/cache_resolution " 1186 "race detected\n", vp); 1187 tsleep(vp, 0, "vclninv", 2); 1188 } 1189 1190 /* 1191 * Check to see if the vnode is in use. If so we have to reference it 1192 * before we clean it out so that its count cannot fall to zero and 1193 * generate a race against ourselves to recycle it. 1194 */ 1195 active = sysref_isactive(&vp->v_sysref); 1196 1197 /* 1198 * Clean out any buffers associated with the vnode and destroy its 1199 * object, if it has one. 1200 */ 1201 vinvalbuf(vp, V_SAVE, 0, 0); 1202 1203 /* 1204 * If purging an active vnode (typically during a forced unmount 1205 * or reboot), it must be closed and deactivated before being 1206 * reclaimed. This isn't really all that safe, but what can 1207 * we do? XXX. 1208 * 1209 * Note that neither of these routines unlocks the vnode. 1210 */ 1211 if (active && (flags & DOCLOSE)) { 1212 while ((n = vp->v_opencount) != 0) { 1213 if (vp->v_writecount) 1214 VOP_CLOSE(vp, FWRITE|FNONBLOCK); 1215 else 1216 VOP_CLOSE(vp, FNONBLOCK); 1217 if (vp->v_opencount == n) { 1218 kprintf("Warning: unable to force-close" 1219 " vnode %p\n", vp); 1220 break; 1221 } 1222 } 1223 } 1224 1225 /* 1226 * If the vnode has not been deactivated, deactivated it. Deactivation 1227 * can create new buffers and VM pages so we have to call vinvalbuf() 1228 * again to make sure they all get flushed. 1229 * 1230 * This can occur if a file with a link count of 0 needs to be 1231 * truncated. 1232 * 1233 * If the vnode is already dead don't try to deactivate it. 1234 */ 1235 if ((vp->v_flag & VINACTIVE) == 0) { 1236 vsetflags(vp, VINACTIVE); 1237 if (vp->v_mount) 1238 VOP_INACTIVE(vp); 1239 vinvalbuf(vp, V_SAVE, 0, 0); 1240 } 1241 1242 /* 1243 * If the vnode has an object, destroy it. 1244 */ 1245 lwkt_gettoken(&vmobj_token); 1246 object = vp->v_object; 1247 if (object != NULL) { 1248 /* 1249 * Use vm_object_lock() rather than vm_object_hold to avoid 1250 * creating an extra (self-)hold on the object. 1251 */ 1252 vm_object_lock(object); 1253 KKASSERT(object == vp->v_object); 1254 if (object->ref_count == 0) { 1255 if ((object->flags & OBJ_DEAD) == 0) 1256 vm_object_terminate(object); 1257 } else { 1258 vm_pager_deallocate(object); 1259 } 1260 vclrflags(vp, VOBJBUF); 1261 vm_object_unlock(object); 1262 } 1263 lwkt_reltoken(&vmobj_token); 1264 KKASSERT((vp->v_flag & VOBJBUF) == 0); 1265 1266 /* 1267 * Reclaim the vnode if not already dead. 1268 */ 1269 if (vp->v_mount && VOP_RECLAIM(vp)) 1270 panic("vclean: cannot reclaim"); 1271 1272 /* 1273 * Done with purge, notify sleepers of the grim news. 1274 */ 1275 vp->v_ops = &dead_vnode_vops_p; 1276 vn_gone(vp); 1277 vp->v_tag = VT_NON; 1278 1279 /* 1280 * If we are destroying an active vnode, reactivate it now that 1281 * we have reassociated it with deadfs. This prevents the system 1282 * from crashing on the vnode due to it being unexpectedly marked 1283 * as inactive or reclaimed. 1284 */ 1285 if (active && (flags & DOCLOSE)) { 1286 vclrflags(vp, VINACTIVE | VRECLAIMED); 1287 } 1288 } 1289 1290 /* 1291 * Eliminate all activity associated with the requested vnode 1292 * and with all vnodes aliased to the requested vnode. 1293 * 1294 * The vnode must be referenced but should not be locked. 1295 */ 1296 int 1297 vrevoke(struct vnode *vp, struct ucred *cred) 1298 { 1299 struct vnode *vq; 1300 struct vnode *vqn; 1301 cdev_t dev; 1302 int error; 1303 1304 /* 1305 * If the vnode has a device association, scrap all vnodes associated 1306 * with the device. Don't let the device disappear on us while we 1307 * are scrapping the vnodes. 1308 * 1309 * The passed vp will probably show up in the list, do not VX lock 1310 * it twice! 1311 * 1312 * Releasing the vnode's rdev here can mess up specfs's call to 1313 * device close, so don't do it. The vnode has been disassociated 1314 * and the device will be closed after the last ref on the related 1315 * fp goes away (if not still open by e.g. the kernel). 1316 */ 1317 if (vp->v_type != VCHR) { 1318 error = fdrevoke(vp, DTYPE_VNODE, cred); 1319 return (error); 1320 } 1321 if ((dev = vp->v_rdev) == NULL) { 1322 return(0); 1323 } 1324 reference_dev(dev); 1325 lwkt_gettoken(&spechash_token); 1326 1327 vqn = SLIST_FIRST(&dev->si_hlist); 1328 if (vqn) 1329 vref(vqn); 1330 while ((vq = vqn) != NULL) { 1331 vqn = SLIST_NEXT(vqn, v_cdevnext); 1332 if (vqn) 1333 vref(vqn); 1334 fdrevoke(vq, DTYPE_VNODE, cred); 1335 /*v_release_rdev(vq);*/ 1336 vrele(vq); 1337 } 1338 lwkt_reltoken(&spechash_token); 1339 dev_drevoke(dev); 1340 release_dev(dev); 1341 return (0); 1342 } 1343 1344 /* 1345 * This is called when the object underlying a vnode is being destroyed, 1346 * such as in a remove(). Try to recycle the vnode immediately if the 1347 * only active reference is our reference. 1348 * 1349 * Directory vnodes in the namecache with children cannot be immediately 1350 * recycled because numerous VOP_N*() ops require them to be stable. 1351 * 1352 * To avoid recursive recycling from VOP_INACTIVE implemenetations this 1353 * function is a NOP if VRECLAIMED is already set. 1354 */ 1355 int 1356 vrecycle(struct vnode *vp) 1357 { 1358 if (vp->v_sysref.refcnt <= 1 && (vp->v_flag & VRECLAIMED) == 0) { 1359 if (cache_inval_vp_nonblock(vp)) 1360 return(0); 1361 vgone_vxlocked(vp); 1362 return (1); 1363 } 1364 return (0); 1365 } 1366 1367 /* 1368 * Return the maximum I/O size allowed for strategy calls on VP. 1369 * 1370 * If vp is VCHR or VBLK we dive the device, otherwise we use 1371 * the vp's mount info. 1372 */ 1373 int 1374 vmaxiosize(struct vnode *vp) 1375 { 1376 if (vp->v_type == VBLK || vp->v_type == VCHR) { 1377 return(vp->v_rdev->si_iosize_max); 1378 } else { 1379 return(vp->v_mount->mnt_iosize_max); 1380 } 1381 } 1382 1383 /* 1384 * Eliminate all activity associated with a vnode in preparation for reuse. 1385 * 1386 * The vnode must be VX locked and refd and will remain VX locked and refd 1387 * on return. This routine may be called with the vnode in any state, as 1388 * long as it is VX locked. The vnode will be cleaned out and marked 1389 * VRECLAIMED but will not actually be reused until all existing refs and 1390 * holds go away. 1391 * 1392 * NOTE: This routine may be called on a vnode which has not yet been 1393 * already been deactivated (VOP_INACTIVE), or on a vnode which has 1394 * already been reclaimed. 1395 * 1396 * This routine is not responsible for placing us back on the freelist. 1397 * Instead, it happens automatically when the caller releases the VX lock 1398 * (assuming there aren't any other references). 1399 */ 1400 void 1401 vgone_vxlocked(struct vnode *vp) 1402 { 1403 /* 1404 * assert that the VX lock is held. This is an absolute requirement 1405 * now for vgone_vxlocked() to be called. 1406 */ 1407 KKASSERT(vp->v_lock.lk_exclusivecount == 1); 1408 1409 get_mplock(); 1410 1411 /* 1412 * Clean out the filesystem specific data and set the VRECLAIMED 1413 * bit. Also deactivate the vnode if necessary. 1414 */ 1415 vclean_vxlocked(vp, DOCLOSE); 1416 1417 /* 1418 * Delete from old mount point vnode list, if on one. 1419 */ 1420 if (vp->v_mount != NULL) { 1421 KKASSERT(vp->v_data == NULL); 1422 insmntque(vp, NULL); 1423 } 1424 1425 /* 1426 * If special device, remove it from special device alias list 1427 * if it is on one. This should normally only occur if a vnode is 1428 * being revoked as the device should otherwise have been released 1429 * naturally. 1430 */ 1431 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1432 v_release_rdev(vp); 1433 } 1434 1435 /* 1436 * Set us to VBAD 1437 */ 1438 vp->v_type = VBAD; 1439 rel_mplock(); 1440 } 1441 1442 /* 1443 * Lookup a vnode by device number. 1444 * 1445 * Returns non-zero and *vpp set to a vref'd vnode on success. 1446 * Returns zero on failure. 1447 */ 1448 int 1449 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp) 1450 { 1451 struct vnode *vp; 1452 1453 lwkt_gettoken(&spechash_token); 1454 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1455 if (type == vp->v_type) { 1456 *vpp = vp; 1457 vref(vp); 1458 lwkt_reltoken(&spechash_token); 1459 return (1); 1460 } 1461 } 1462 lwkt_reltoken(&spechash_token); 1463 return (0); 1464 } 1465 1466 /* 1467 * Calculate the total number of references to a special device. This 1468 * routine may only be called for VBLK and VCHR vnodes since v_rdev is 1469 * an overloaded field. Since udev2dev can now return NULL, we have 1470 * to check for a NULL v_rdev. 1471 */ 1472 int 1473 count_dev(cdev_t dev) 1474 { 1475 struct vnode *vp; 1476 int count = 0; 1477 1478 if (SLIST_FIRST(&dev->si_hlist)) { 1479 lwkt_gettoken(&spechash_token); 1480 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) { 1481 count += vp->v_opencount; 1482 } 1483 lwkt_reltoken(&spechash_token); 1484 } 1485 return(count); 1486 } 1487 1488 int 1489 vcount(struct vnode *vp) 1490 { 1491 if (vp->v_rdev == NULL) 1492 return(0); 1493 return(count_dev(vp->v_rdev)); 1494 } 1495 1496 /* 1497 * Initialize VMIO for a vnode. This routine MUST be called before a 1498 * VFS can issue buffer cache ops on a vnode. It is typically called 1499 * when a vnode is initialized from its inode. 1500 */ 1501 int 1502 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff) 1503 { 1504 vm_object_t object; 1505 int error = 0; 1506 1507 lwkt_gettoken(&vmobj_token); 1508 retry: 1509 if ((object = vp->v_object) == NULL) { 1510 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff); 1511 /* 1512 * Dereference the reference we just created. This assumes 1513 * that the object is associated with the vp. 1514 */ 1515 object->ref_count--; 1516 vrele(vp); 1517 } else { 1518 if (object->flags & OBJ_DEAD) { 1519 vn_unlock(vp); 1520 if (vp->v_object == object) 1521 vm_object_dead_sleep(object, "vodead"); 1522 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1523 goto retry; 1524 } 1525 } 1526 KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object")); 1527 vsetflags(vp, VOBJBUF); 1528 lwkt_reltoken(&vmobj_token); 1529 1530 return (error); 1531 } 1532 1533 1534 /* 1535 * Print out a description of a vnode. 1536 */ 1537 static char *typename[] = 1538 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1539 1540 void 1541 vprint(char *label, struct vnode *vp) 1542 { 1543 char buf[96]; 1544 1545 if (label != NULL) 1546 kprintf("%s: %p: ", label, (void *)vp); 1547 else 1548 kprintf("%p: ", (void *)vp); 1549 kprintf("type %s, sysrefs %d, writecount %d, holdcnt %d,", 1550 typename[vp->v_type], 1551 vp->v_sysref.refcnt, vp->v_writecount, vp->v_auxrefs); 1552 buf[0] = '\0'; 1553 if (vp->v_flag & VROOT) 1554 strcat(buf, "|VROOT"); 1555 if (vp->v_flag & VPFSROOT) 1556 strcat(buf, "|VPFSROOT"); 1557 if (vp->v_flag & VTEXT) 1558 strcat(buf, "|VTEXT"); 1559 if (vp->v_flag & VSYSTEM) 1560 strcat(buf, "|VSYSTEM"); 1561 if (vp->v_flag & VFREE) 1562 strcat(buf, "|VFREE"); 1563 if (vp->v_flag & VOBJBUF) 1564 strcat(buf, "|VOBJBUF"); 1565 if (buf[0] != '\0') 1566 kprintf(" flags (%s)", &buf[1]); 1567 if (vp->v_data == NULL) { 1568 kprintf("\n"); 1569 } else { 1570 kprintf("\n\t"); 1571 VOP_PRINT(vp); 1572 } 1573 } 1574 1575 /* 1576 * Do the usual access checking. 1577 * file_mode, uid and gid are from the vnode in question, 1578 * while acc_mode and cred are from the VOP_ACCESS parameter list 1579 */ 1580 int 1581 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1582 mode_t acc_mode, struct ucred *cred) 1583 { 1584 mode_t mask; 1585 int ismember; 1586 1587 /* 1588 * Super-user always gets read/write access, but execute access depends 1589 * on at least one execute bit being set. 1590 */ 1591 if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) { 1592 if ((acc_mode & VEXEC) && type != VDIR && 1593 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1594 return (EACCES); 1595 return (0); 1596 } 1597 1598 mask = 0; 1599 1600 /* Otherwise, check the owner. */ 1601 if (cred->cr_uid == uid) { 1602 if (acc_mode & VEXEC) 1603 mask |= S_IXUSR; 1604 if (acc_mode & VREAD) 1605 mask |= S_IRUSR; 1606 if (acc_mode & VWRITE) 1607 mask |= S_IWUSR; 1608 return ((file_mode & mask) == mask ? 0 : EACCES); 1609 } 1610 1611 /* Otherwise, check the groups. */ 1612 ismember = groupmember(gid, cred); 1613 if (cred->cr_svgid == gid || ismember) { 1614 if (acc_mode & VEXEC) 1615 mask |= S_IXGRP; 1616 if (acc_mode & VREAD) 1617 mask |= S_IRGRP; 1618 if (acc_mode & VWRITE) 1619 mask |= S_IWGRP; 1620 return ((file_mode & mask) == mask ? 0 : EACCES); 1621 } 1622 1623 /* Otherwise, check everyone else. */ 1624 if (acc_mode & VEXEC) 1625 mask |= S_IXOTH; 1626 if (acc_mode & VREAD) 1627 mask |= S_IROTH; 1628 if (acc_mode & VWRITE) 1629 mask |= S_IWOTH; 1630 return ((file_mode & mask) == mask ? 0 : EACCES); 1631 } 1632 1633 #ifdef DDB 1634 #include <ddb/ddb.h> 1635 1636 static int db_show_locked_vnodes(struct mount *mp, void *data); 1637 1638 /* 1639 * List all of the locked vnodes in the system. 1640 * Called when debugging the kernel. 1641 */ 1642 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 1643 { 1644 kprintf("Locked vnodes\n"); 1645 mountlist_scan(db_show_locked_vnodes, NULL, 1646 MNTSCAN_FORWARD|MNTSCAN_NOBUSY); 1647 } 1648 1649 static int 1650 db_show_locked_vnodes(struct mount *mp, void *data __unused) 1651 { 1652 struct vnode *vp; 1653 1654 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 1655 if (vn_islocked(vp)) 1656 vprint(NULL, vp); 1657 } 1658 return(0); 1659 } 1660 #endif 1661 1662 /* 1663 * Top level filesystem related information gathering. 1664 */ 1665 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 1666 1667 static int 1668 vfs_sysctl(SYSCTL_HANDLER_ARGS) 1669 { 1670 int *name = (int *)arg1 - 1; /* XXX */ 1671 u_int namelen = arg2 + 1; /* XXX */ 1672 struct vfsconf *vfsp; 1673 int maxtypenum; 1674 1675 #if 1 || defined(COMPAT_PRELITE2) 1676 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 1677 if (namelen == 1) 1678 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 1679 #endif 1680 1681 #ifdef notyet 1682 /* all sysctl names at this level are at least name and field */ 1683 if (namelen < 2) 1684 return (ENOTDIR); /* overloaded */ 1685 if (name[0] != VFS_GENERIC) { 1686 vfsp = vfsconf_find_by_typenum(name[0]); 1687 if (vfsp == NULL) 1688 return (EOPNOTSUPP); 1689 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1690 oldp, oldlenp, newp, newlen, p)); 1691 } 1692 #endif 1693 switch (name[1]) { 1694 case VFS_MAXTYPENUM: 1695 if (namelen != 2) 1696 return (ENOTDIR); 1697 maxtypenum = vfsconf_get_maxtypenum(); 1698 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum))); 1699 case VFS_CONF: 1700 if (namelen != 3) 1701 return (ENOTDIR); /* overloaded */ 1702 vfsp = vfsconf_find_by_typenum(name[2]); 1703 if (vfsp == NULL) 1704 return (EOPNOTSUPP); 1705 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 1706 } 1707 return (EOPNOTSUPP); 1708 } 1709 1710 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 1711 "Generic filesystem"); 1712 1713 #if 1 || defined(COMPAT_PRELITE2) 1714 1715 static int 1716 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data) 1717 { 1718 int error; 1719 struct ovfsconf ovfs; 1720 struct sysctl_req *req = (struct sysctl_req*) data; 1721 1722 bzero(&ovfs, sizeof(ovfs)); 1723 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 1724 strcpy(ovfs.vfc_name, vfsp->vfc_name); 1725 ovfs.vfc_index = vfsp->vfc_typenum; 1726 ovfs.vfc_refcount = vfsp->vfc_refcount; 1727 ovfs.vfc_flags = vfsp->vfc_flags; 1728 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 1729 if (error) 1730 return error; /* abort iteration with error code */ 1731 else 1732 return 0; /* continue iterating with next element */ 1733 } 1734 1735 static int 1736 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 1737 { 1738 return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req); 1739 } 1740 1741 #endif /* 1 || COMPAT_PRELITE2 */ 1742 1743 /* 1744 * Check to see if a filesystem is mounted on a block device. 1745 */ 1746 int 1747 vfs_mountedon(struct vnode *vp) 1748 { 1749 cdev_t dev; 1750 1751 if ((dev = vp->v_rdev) == NULL) { 1752 /* if (vp->v_type != VBLK) 1753 dev = get_dev(vp->v_uminor, vp->v_umajor); */ 1754 } 1755 if (dev != NULL && dev->si_mountpoint) 1756 return (EBUSY); 1757 return (0); 1758 } 1759 1760 /* 1761 * Unmount all filesystems. The list is traversed in reverse order 1762 * of mounting to avoid dependencies. 1763 */ 1764 1765 static int vfs_umountall_callback(struct mount *mp, void *data); 1766 1767 void 1768 vfs_unmountall(void) 1769 { 1770 int count; 1771 1772 do { 1773 count = mountlist_scan(vfs_umountall_callback, 1774 NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY); 1775 } while (count); 1776 } 1777 1778 static 1779 int 1780 vfs_umountall_callback(struct mount *mp, void *data) 1781 { 1782 int error; 1783 1784 error = dounmount(mp, MNT_FORCE); 1785 if (error) { 1786 mountlist_remove(mp); 1787 kprintf("unmount of filesystem mounted from %s failed (", 1788 mp->mnt_stat.f_mntfromname); 1789 if (error == EBUSY) 1790 kprintf("BUSY)\n"); 1791 else 1792 kprintf("%d)\n", error); 1793 } 1794 return(1); 1795 } 1796 1797 /* 1798 * Checks the mount flags for parameter mp and put the names comma-separated 1799 * into a string buffer buf with a size limit specified by len. 1800 * 1801 * It returns the number of bytes written into buf, and (*errorp) will be 1802 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was 1803 * not large enough). The buffer will be 0-terminated if len was not 0. 1804 */ 1805 size_t 1806 vfs_flagstostr(int flags, const struct mountctl_opt *optp, 1807 char *buf, size_t len, int *errorp) 1808 { 1809 static const struct mountctl_opt optnames[] = { 1810 { MNT_ASYNC, "asynchronous" }, 1811 { MNT_EXPORTED, "NFS exported" }, 1812 { MNT_LOCAL, "local" }, 1813 { MNT_NOATIME, "noatime" }, 1814 { MNT_NODEV, "nodev" }, 1815 { MNT_NOEXEC, "noexec" }, 1816 { MNT_NOSUID, "nosuid" }, 1817 { MNT_NOSYMFOLLOW, "nosymfollow" }, 1818 { MNT_QUOTA, "with-quotas" }, 1819 { MNT_RDONLY, "read-only" }, 1820 { MNT_SYNCHRONOUS, "synchronous" }, 1821 { MNT_UNION, "union" }, 1822 { MNT_NOCLUSTERR, "noclusterr" }, 1823 { MNT_NOCLUSTERW, "noclusterw" }, 1824 { MNT_SUIDDIR, "suiddir" }, 1825 { MNT_SOFTDEP, "soft-updates" }, 1826 { MNT_IGNORE, "ignore" }, 1827 { 0, NULL} 1828 }; 1829 int bwritten; 1830 int bleft; 1831 int optlen; 1832 int actsize; 1833 1834 *errorp = 0; 1835 bwritten = 0; 1836 bleft = len - 1; /* leave room for trailing \0 */ 1837 1838 /* 1839 * Checks the size of the string. If it contains 1840 * any data, then we will append the new flags to 1841 * it. 1842 */ 1843 actsize = strlen(buf); 1844 if (actsize > 0) 1845 buf += actsize; 1846 1847 /* Default flags if no flags passed */ 1848 if (optp == NULL) 1849 optp = optnames; 1850 1851 if (bleft < 0) { /* degenerate case, 0-length buffer */ 1852 *errorp = EINVAL; 1853 return(0); 1854 } 1855 1856 for (; flags && optp->o_opt; ++optp) { 1857 if ((flags & optp->o_opt) == 0) 1858 continue; 1859 optlen = strlen(optp->o_name); 1860 if (bwritten || actsize > 0) { 1861 if (bleft < 2) { 1862 *errorp = ENOSPC; 1863 break; 1864 } 1865 buf[bwritten++] = ','; 1866 buf[bwritten++] = ' '; 1867 bleft -= 2; 1868 } 1869 if (bleft < optlen) { 1870 *errorp = ENOSPC; 1871 break; 1872 } 1873 bcopy(optp->o_name, buf + bwritten, optlen); 1874 bwritten += optlen; 1875 bleft -= optlen; 1876 flags &= ~optp->o_opt; 1877 } 1878 1879 /* 1880 * Space already reserved for trailing \0 1881 */ 1882 buf[bwritten] = 0; 1883 return (bwritten); 1884 } 1885 1886 /* 1887 * Build hash lists of net addresses and hang them off the mount point. 1888 * Called by ufs_mount() to set up the lists of export addresses. 1889 */ 1890 static int 1891 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1892 const struct export_args *argp) 1893 { 1894 struct netcred *np; 1895 struct radix_node_head *rnh; 1896 int i; 1897 struct radix_node *rn; 1898 struct sockaddr *saddr, *smask = 0; 1899 struct domain *dom; 1900 int error; 1901 1902 if (argp->ex_addrlen == 0) { 1903 if (mp->mnt_flag & MNT_DEFEXPORTED) 1904 return (EPERM); 1905 np = &nep->ne_defexported; 1906 np->netc_exflags = argp->ex_flags; 1907 np->netc_anon = argp->ex_anon; 1908 np->netc_anon.cr_ref = 1; 1909 mp->mnt_flag |= MNT_DEFEXPORTED; 1910 return (0); 1911 } 1912 1913 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 1914 return (EINVAL); 1915 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 1916 return (EINVAL); 1917 1918 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1919 np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO); 1920 saddr = (struct sockaddr *) (np + 1); 1921 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 1922 goto out; 1923 if (saddr->sa_len > argp->ex_addrlen) 1924 saddr->sa_len = argp->ex_addrlen; 1925 if (argp->ex_masklen) { 1926 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1927 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 1928 if (error) 1929 goto out; 1930 if (smask->sa_len > argp->ex_masklen) 1931 smask->sa_len = argp->ex_masklen; 1932 } 1933 i = saddr->sa_family; 1934 if ((rnh = nep->ne_rtable[i]) == 0) { 1935 /* 1936 * Seems silly to initialize every AF when most are not used, 1937 * do so on demand here 1938 */ 1939 SLIST_FOREACH(dom, &domains, dom_next) 1940 if (dom->dom_family == i && dom->dom_rtattach) { 1941 dom->dom_rtattach((void **) &nep->ne_rtable[i], 1942 dom->dom_rtoffset); 1943 break; 1944 } 1945 if ((rnh = nep->ne_rtable[i]) == 0) { 1946 error = ENOBUFS; 1947 goto out; 1948 } 1949 } 1950 rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh, 1951 np->netc_rnodes); 1952 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 1953 error = EPERM; 1954 goto out; 1955 } 1956 np->netc_exflags = argp->ex_flags; 1957 np->netc_anon = argp->ex_anon; 1958 np->netc_anon.cr_ref = 1; 1959 return (0); 1960 out: 1961 kfree(np, M_NETADDR); 1962 return (error); 1963 } 1964 1965 /* ARGSUSED */ 1966 static int 1967 vfs_free_netcred(struct radix_node *rn, void *w) 1968 { 1969 struct radix_node_head *rnh = (struct radix_node_head *) w; 1970 1971 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 1972 kfree((caddr_t) rn, M_NETADDR); 1973 return (0); 1974 } 1975 1976 /* 1977 * Free the net address hash lists that are hanging off the mount points. 1978 */ 1979 static void 1980 vfs_free_addrlist(struct netexport *nep) 1981 { 1982 int i; 1983 struct radix_node_head *rnh; 1984 1985 for (i = 0; i <= AF_MAX; i++) 1986 if ((rnh = nep->ne_rtable[i])) { 1987 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 1988 (caddr_t) rnh); 1989 kfree((caddr_t) rnh, M_RTABLE); 1990 nep->ne_rtable[i] = 0; 1991 } 1992 } 1993 1994 int 1995 vfs_export(struct mount *mp, struct netexport *nep, 1996 const struct export_args *argp) 1997 { 1998 int error; 1999 2000 if (argp->ex_flags & MNT_DELEXPORT) { 2001 if (mp->mnt_flag & MNT_EXPUBLIC) { 2002 vfs_setpublicfs(NULL, NULL, NULL); 2003 mp->mnt_flag &= ~MNT_EXPUBLIC; 2004 } 2005 vfs_free_addrlist(nep); 2006 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2007 } 2008 if (argp->ex_flags & MNT_EXPORTED) { 2009 if (argp->ex_flags & MNT_EXPUBLIC) { 2010 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2011 return (error); 2012 mp->mnt_flag |= MNT_EXPUBLIC; 2013 } 2014 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2015 return (error); 2016 mp->mnt_flag |= MNT_EXPORTED; 2017 } 2018 return (0); 2019 } 2020 2021 2022 /* 2023 * Set the publicly exported filesystem (WebNFS). Currently, only 2024 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2025 */ 2026 int 2027 vfs_setpublicfs(struct mount *mp, struct netexport *nep, 2028 const struct export_args *argp) 2029 { 2030 int error; 2031 struct vnode *rvp; 2032 char *cp; 2033 2034 /* 2035 * mp == NULL -> invalidate the current info, the FS is 2036 * no longer exported. May be called from either vfs_export 2037 * or unmount, so check if it hasn't already been done. 2038 */ 2039 if (mp == NULL) { 2040 if (nfs_pub.np_valid) { 2041 nfs_pub.np_valid = 0; 2042 if (nfs_pub.np_index != NULL) { 2043 FREE(nfs_pub.np_index, M_TEMP); 2044 nfs_pub.np_index = NULL; 2045 } 2046 } 2047 return (0); 2048 } 2049 2050 /* 2051 * Only one allowed at a time. 2052 */ 2053 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2054 return (EBUSY); 2055 2056 /* 2057 * Get real filehandle for root of exported FS. 2058 */ 2059 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2060 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2061 2062 if ((error = VFS_ROOT(mp, &rvp))) 2063 return (error); 2064 2065 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2066 return (error); 2067 2068 vput(rvp); 2069 2070 /* 2071 * If an indexfile was specified, pull it in. 2072 */ 2073 if (argp->ex_indexfile != NULL) { 2074 int namelen; 2075 2076 error = vn_get_namelen(rvp, &namelen); 2077 if (error) 2078 return (error); 2079 MALLOC(nfs_pub.np_index, char *, namelen, M_TEMP, 2080 M_WAITOK); 2081 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2082 namelen, NULL); 2083 if (!error) { 2084 /* 2085 * Check for illegal filenames. 2086 */ 2087 for (cp = nfs_pub.np_index; *cp; cp++) { 2088 if (*cp == '/') { 2089 error = EINVAL; 2090 break; 2091 } 2092 } 2093 } 2094 if (error) { 2095 FREE(nfs_pub.np_index, M_TEMP); 2096 return (error); 2097 } 2098 } 2099 2100 nfs_pub.np_mount = mp; 2101 nfs_pub.np_valid = 1; 2102 return (0); 2103 } 2104 2105 struct netcred * 2106 vfs_export_lookup(struct mount *mp, struct netexport *nep, 2107 struct sockaddr *nam) 2108 { 2109 struct netcred *np; 2110 struct radix_node_head *rnh; 2111 struct sockaddr *saddr; 2112 2113 np = NULL; 2114 if (mp->mnt_flag & MNT_EXPORTED) { 2115 /* 2116 * Lookup in the export list first. 2117 */ 2118 if (nam != NULL) { 2119 saddr = nam; 2120 rnh = nep->ne_rtable[saddr->sa_family]; 2121 if (rnh != NULL) { 2122 np = (struct netcred *) 2123 (*rnh->rnh_matchaddr)((char *)saddr, 2124 rnh); 2125 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2126 np = NULL; 2127 } 2128 } 2129 /* 2130 * If no address match, use the default if it exists. 2131 */ 2132 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2133 np = &nep->ne_defexported; 2134 } 2135 return (np); 2136 } 2137 2138 /* 2139 * perform msync on all vnodes under a mount point. The mount point must 2140 * be locked. This code is also responsible for lazy-freeing unreferenced 2141 * vnodes whos VM objects no longer contain pages. 2142 * 2143 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state. 2144 * 2145 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked, 2146 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it 2147 * way up in this high level function. 2148 */ 2149 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data); 2150 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data); 2151 2152 void 2153 vfs_msync(struct mount *mp, int flags) 2154 { 2155 int vmsc_flags; 2156 2157 /* 2158 * tmpfs sets this flag to prevent msync(), sync, and the 2159 * filesystem periodic syncer from trying to flush VM pages 2160 * to swap. Only pure memory pressure flushes tmpfs VM pages 2161 * to swap. 2162 */ 2163 if (mp->mnt_kern_flag & MNTK_NOMSYNC) 2164 return; 2165 2166 /* 2167 * Ok, scan the vnodes for work. 2168 */ 2169 vmsc_flags = VMSC_GETVP; 2170 if (flags != MNT_WAIT) 2171 vmsc_flags |= VMSC_NOWAIT; 2172 vmntvnodescan(mp, vmsc_flags, vfs_msync_scan1, vfs_msync_scan2, 2173 (void *)(intptr_t)flags); 2174 } 2175 2176 /* 2177 * scan1 is a fast pre-check. There could be hundreds of thousands of 2178 * vnodes, we cannot afford to do anything heavy weight until we have a 2179 * fairly good indication that there is work to do. 2180 */ 2181 static 2182 int 2183 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data) 2184 { 2185 int flags = (int)(intptr_t)data; 2186 2187 if ((vp->v_flag & VRECLAIMED) == 0) { 2188 if (vshouldmsync(vp)) 2189 return(0); /* call scan2 */ 2190 if ((mp->mnt_flag & MNT_RDONLY) == 0 && 2191 (vp->v_flag & VOBJDIRTY) && 2192 (flags == MNT_WAIT || vn_islocked(vp) == 0)) { 2193 return(0); /* call scan2 */ 2194 } 2195 } 2196 2197 /* 2198 * do not call scan2, continue the loop 2199 */ 2200 return(-1); 2201 } 2202 2203 /* 2204 * This callback is handed a locked vnode. 2205 */ 2206 static 2207 int 2208 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data) 2209 { 2210 vm_object_t obj; 2211 int flags = (int)(intptr_t)data; 2212 2213 if (vp->v_flag & VRECLAIMED) 2214 return(0); 2215 2216 if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) { 2217 if ((obj = vp->v_object) != NULL) { 2218 vm_object_page_clean(obj, 0, 0, 2219 flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2220 } 2221 } 2222 return(0); 2223 } 2224 2225 /* 2226 * Wake up anyone interested in vp because it is being revoked. 2227 */ 2228 void 2229 vn_gone(struct vnode *vp) 2230 { 2231 lwkt_gettoken(&vp->v_token); 2232 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE); 2233 lwkt_reltoken(&vp->v_token); 2234 } 2235 2236 /* 2237 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened 2238 * (or v_rdev might be NULL). 2239 */ 2240 cdev_t 2241 vn_todev(struct vnode *vp) 2242 { 2243 if (vp->v_type != VBLK && vp->v_type != VCHR) 2244 return (NULL); 2245 KKASSERT(vp->v_rdev != NULL); 2246 return (vp->v_rdev); 2247 } 2248 2249 /* 2250 * Check if vnode represents a disk device. The vnode does not need to be 2251 * opened. 2252 * 2253 * MPALMOSTSAFE 2254 */ 2255 int 2256 vn_isdisk(struct vnode *vp, int *errp) 2257 { 2258 cdev_t dev; 2259 2260 if (vp->v_type != VCHR) { 2261 if (errp != NULL) 2262 *errp = ENOTBLK; 2263 return (0); 2264 } 2265 2266 dev = vp->v_rdev; 2267 2268 if (dev == NULL) { 2269 if (errp != NULL) 2270 *errp = ENXIO; 2271 return (0); 2272 } 2273 if (dev_is_good(dev) == 0) { 2274 if (errp != NULL) 2275 *errp = ENXIO; 2276 return (0); 2277 } 2278 if ((dev_dflags(dev) & D_DISK) == 0) { 2279 if (errp != NULL) 2280 *errp = ENOTBLK; 2281 return (0); 2282 } 2283 if (errp != NULL) 2284 *errp = 0; 2285 return (1); 2286 } 2287 2288 int 2289 vn_get_namelen(struct vnode *vp, int *namelen) 2290 { 2291 int error; 2292 register_t retval[2]; 2293 2294 error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval); 2295 if (error) 2296 return (error); 2297 *namelen = (int)retval[0]; 2298 return (0); 2299 } 2300 2301 int 2302 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 2303 uint16_t d_namlen, const char *d_name) 2304 { 2305 struct dirent *dp; 2306 size_t len; 2307 2308 len = _DIRENT_RECLEN(d_namlen); 2309 if (len > uio->uio_resid) 2310 return(1); 2311 2312 dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO); 2313 2314 dp->d_ino = d_ino; 2315 dp->d_namlen = d_namlen; 2316 dp->d_type = d_type; 2317 bcopy(d_name, dp->d_name, d_namlen); 2318 2319 *error = uiomove((caddr_t)dp, len, uio); 2320 2321 kfree(dp, M_TEMP); 2322 2323 return(0); 2324 } 2325 2326 void 2327 vn_mark_atime(struct vnode *vp, struct thread *td) 2328 { 2329 struct proc *p = td->td_proc; 2330 struct ucred *cred = p ? p->p_ucred : proc0.p_ucred; 2331 2332 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 2333 VOP_MARKATIME(vp, cred); 2334 } 2335 } 2336