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 * $FreeBSD$ 40 */ 41 42 /* 43 * External virtual filesystem routines 44 */ 45 #include "opt_ddb.h" 46 #include "opt_ffs.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/bio.h> 51 #include <sys/buf.h> 52 #include <sys/conf.h> 53 #include <sys/dirent.h> 54 #include <sys/domain.h> 55 #include <sys/eventhandler.h> 56 #include <sys/fcntl.h> 57 #include <sys/kernel.h> 58 #include <sys/kthread.h> 59 #include <sys/malloc.h> 60 #include <sys/mount.h> 61 #include <sys/namei.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/vmmeter.h> 68 #include <sys/vnode.h> 69 70 #include <machine/limits.h> 71 72 #include <vm/vm.h> 73 #include <vm/vm_object.h> 74 #include <vm/vm_extern.h> 75 #include <vm/pmap.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_page.h> 78 #include <vm/vm_pager.h> 79 #include <vm/vnode_pager.h> 80 #include <vm/vm_zone.h> 81 82 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 83 84 static void insmntque __P((struct vnode *vp, struct mount *mp)); 85 static void vclean __P((struct vnode *vp, int flags, struct proc *p)); 86 static unsigned long numvnodes; 87 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 88 89 enum vtype iftovt_tab[16] = { 90 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 91 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 92 }; 93 int vttoif_tab[9] = { 94 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 95 S_IFSOCK, S_IFIFO, S_IFMT, 96 }; 97 98 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 99 100 static u_long wantfreevnodes = 25; 101 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 102 static u_long freevnodes = 0; 103 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 104 105 static int reassignbufcalls; 106 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 107 static int reassignbufloops; 108 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, ""); 109 static int reassignbufsortgood; 110 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, ""); 111 static int reassignbufsortbad; 112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, ""); 113 static int reassignbufmethod = 1; 114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, ""); 115 116 #ifdef ENABLE_VFS_IOOPT 117 int vfs_ioopt = 0; 118 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 119 #endif 120 121 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */ 122 struct simplelock mountlist_slock; 123 struct simplelock mntvnode_slock; 124 int nfs_mount_type = -1; 125 #ifndef NULL_SIMPLELOCKS 126 static struct simplelock mntid_slock; 127 static struct simplelock vnode_free_list_slock; 128 static struct simplelock spechash_slock; 129 #endif 130 struct nfs_public nfs_pub; /* publicly exported FS */ 131 static vm_zone_t vnode_zone; 132 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 133 134 /* 135 * The workitem queue. 136 */ 137 #define SYNCER_MAXDELAY 32 138 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 139 time_t syncdelay = 30; /* max time to delay syncing data */ 140 time_t filedelay = 30; /* time to delay syncing files */ 141 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 142 time_t dirdelay = 29; /* time to delay syncing directories */ 143 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 144 time_t metadelay = 28; /* time to delay syncing metadata */ 145 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 146 static int rushjob; /* number of slots to run ASAP */ 147 static int stat_rush_requests; /* number of times I/O speeded up */ 148 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 149 150 static int syncer_delayno = 0; 151 static long syncer_mask; 152 LIST_HEAD(synclist, vnode); 153 static struct synclist *syncer_workitem_pending; 154 155 int desiredvnodes; 156 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 157 &desiredvnodes, 0, "Maximum number of vnodes"); 158 159 static void vfs_free_addrlist __P((struct netexport *nep)); 160 static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 161 static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 162 struct export_args *argp)); 163 164 /* 165 * Initialize the vnode management data structures. 166 */ 167 void 168 vntblinit() 169 { 170 171 desiredvnodes = maxproc + cnt.v_page_count / 4; 172 simple_lock_init(&mntvnode_slock); 173 simple_lock_init(&mntid_slock); 174 simple_lock_init(&spechash_slock); 175 TAILQ_INIT(&vnode_free_list); 176 simple_lock_init(&vnode_free_list_slock); 177 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 178 /* 179 * Initialize the filesystem syncer. 180 */ 181 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 182 &syncer_mask); 183 syncer_maxdelay = syncer_mask + 1; 184 } 185 186 /* 187 * Mark a mount point as busy. Used to synchronize access and to delay 188 * unmounting. Interlock is not released on failure. 189 */ 190 int 191 vfs_busy(mp, flags, interlkp, p) 192 struct mount *mp; 193 int flags; 194 struct simplelock *interlkp; 195 struct proc *p; 196 { 197 int lkflags; 198 199 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 200 if (flags & LK_NOWAIT) 201 return (ENOENT); 202 mp->mnt_kern_flag |= MNTK_MWAIT; 203 if (interlkp) { 204 simple_unlock(interlkp); 205 } 206 /* 207 * Since all busy locks are shared except the exclusive 208 * lock granted when unmounting, the only place that a 209 * wakeup needs to be done is at the release of the 210 * exclusive lock at the end of dounmount. 211 */ 212 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 213 if (interlkp) { 214 simple_lock(interlkp); 215 } 216 return (ENOENT); 217 } 218 lkflags = LK_SHARED | LK_NOPAUSE; 219 if (interlkp) 220 lkflags |= LK_INTERLOCK; 221 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 222 panic("vfs_busy: unexpected lock failure"); 223 return (0); 224 } 225 226 /* 227 * Free a busy filesystem. 228 */ 229 void 230 vfs_unbusy(mp, p) 231 struct mount *mp; 232 struct proc *p; 233 { 234 235 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 236 } 237 238 /* 239 * Lookup a filesystem type, and if found allocate and initialize 240 * a mount structure for it. 241 * 242 * Devname is usually updated by mount(8) after booting. 243 */ 244 int 245 vfs_rootmountalloc(fstypename, devname, mpp) 246 char *fstypename; 247 char *devname; 248 struct mount **mpp; 249 { 250 struct proc *p = curproc; /* XXX */ 251 struct vfsconf *vfsp; 252 struct mount *mp; 253 254 if (fstypename == NULL) 255 return (ENODEV); 256 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 257 if (!strcmp(vfsp->vfc_name, fstypename)) 258 break; 259 if (vfsp == NULL) 260 return (ENODEV); 261 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 262 bzero((char *)mp, (u_long)sizeof(struct mount)); 263 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); 264 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 265 LIST_INIT(&mp->mnt_vnodelist); 266 mp->mnt_vfc = vfsp; 267 mp->mnt_op = vfsp->vfc_vfsops; 268 mp->mnt_flag = MNT_RDONLY; 269 mp->mnt_vnodecovered = NULLVP; 270 vfsp->vfc_refcount++; 271 mp->mnt_iosize_max = DFLTPHYS; 272 mp->mnt_stat.f_type = vfsp->vfc_typenum; 273 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 274 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 275 mp->mnt_stat.f_mntonname[0] = '/'; 276 mp->mnt_stat.f_mntonname[1] = 0; 277 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 278 *mpp = mp; 279 return (0); 280 } 281 282 /* 283 * Find an appropriate filesystem to use for the root. If a filesystem 284 * has not been preselected, walk through the list of known filesystems 285 * trying those that have mountroot routines, and try them until one 286 * works or we have tried them all. 287 */ 288 #ifdef notdef /* XXX JH */ 289 int 290 lite2_vfs_mountroot() 291 { 292 struct vfsconf *vfsp; 293 extern int (*lite2_mountroot) __P((void)); 294 int error; 295 296 if (lite2_mountroot != NULL) 297 return ((*lite2_mountroot)()); 298 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 299 if (vfsp->vfc_mountroot == NULL) 300 continue; 301 if ((error = (*vfsp->vfc_mountroot)()) == 0) 302 return (0); 303 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 304 } 305 return (ENODEV); 306 } 307 #endif 308 309 /* 310 * Lookup a mount point by filesystem identifier. 311 */ 312 struct mount * 313 vfs_getvfs(fsid) 314 fsid_t *fsid; 315 { 316 register struct mount *mp; 317 318 simple_lock(&mountlist_slock); 319 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 320 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 321 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 322 simple_unlock(&mountlist_slock); 323 return (mp); 324 } 325 } 326 simple_unlock(&mountlist_slock); 327 return ((struct mount *) 0); 328 } 329 330 /* 331 * Get a new unique fsid. Try to make its val[0] unique, since this value 332 * will be used to create fake device numbers for stat(). Also try (but 333 * not so hard) make its val[0] unique mod 2^16, since some emulators only 334 * support 16-bit device numbers. We end up with unique val[0]'s for the 335 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 336 * 337 * Keep in mind that several mounts may be running in parallel. Starting 338 * the search one past where the previous search terminated is both a 339 * micro-optimization and a defense against returning the same fsid to 340 * different mounts. 341 */ 342 void 343 vfs_getnewfsid(mp) 344 struct mount *mp; 345 { 346 static u_int16_t mntid_base; 347 fsid_t tfsid; 348 int mtype; 349 350 simple_lock(&mntid_slock); 351 mtype = mp->mnt_vfc->vfc_typenum; 352 tfsid.val[1] = mtype; 353 mtype = (mtype & 0xFF) << 24; 354 for (;;) { 355 tfsid.val[0] = makeudev(255, 356 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 357 mntid_base++; 358 if (vfs_getvfs(&tfsid) == NULL) 359 break; 360 } 361 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 362 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 363 simple_unlock(&mntid_slock); 364 } 365 366 /* 367 * Knob to control the precision of file timestamps: 368 * 369 * 0 = seconds only; nanoseconds zeroed. 370 * 1 = seconds and nanoseconds, accurate within 1/HZ. 371 * 2 = seconds and nanoseconds, truncated to microseconds. 372 * >=3 = seconds and nanoseconds, maximum precision. 373 */ 374 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 375 376 static int timestamp_precision = TSP_SEC; 377 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 378 ×tamp_precision, 0, ""); 379 380 /* 381 * Get a current timestamp. 382 */ 383 void 384 vfs_timestamp(tsp) 385 struct timespec *tsp; 386 { 387 struct timeval tv; 388 389 switch (timestamp_precision) { 390 case TSP_SEC: 391 tsp->tv_sec = time_second; 392 tsp->tv_nsec = 0; 393 break; 394 case TSP_HZ: 395 getnanotime(tsp); 396 break; 397 case TSP_USEC: 398 microtime(&tv); 399 TIMEVAL_TO_TIMESPEC(&tv, tsp); 400 break; 401 case TSP_NSEC: 402 default: 403 nanotime(tsp); 404 break; 405 } 406 } 407 408 /* 409 * Set vnode attributes to VNOVAL 410 */ 411 void 412 vattr_null(vap) 413 register struct vattr *vap; 414 { 415 416 vap->va_type = VNON; 417 vap->va_size = VNOVAL; 418 vap->va_bytes = VNOVAL; 419 vap->va_mode = VNOVAL; 420 vap->va_nlink = VNOVAL; 421 vap->va_uid = VNOVAL; 422 vap->va_gid = VNOVAL; 423 vap->va_fsid = VNOVAL; 424 vap->va_fileid = VNOVAL; 425 vap->va_blocksize = VNOVAL; 426 vap->va_rdev = VNOVAL; 427 vap->va_atime.tv_sec = VNOVAL; 428 vap->va_atime.tv_nsec = VNOVAL; 429 vap->va_mtime.tv_sec = VNOVAL; 430 vap->va_mtime.tv_nsec = VNOVAL; 431 vap->va_ctime.tv_sec = VNOVAL; 432 vap->va_ctime.tv_nsec = VNOVAL; 433 vap->va_flags = VNOVAL; 434 vap->va_gen = VNOVAL; 435 vap->va_vaflags = 0; 436 } 437 438 /* 439 * Routines having to do with the management of the vnode table. 440 */ 441 extern vop_t **dead_vnodeop_p; 442 443 /* 444 * Return the next vnode from the free list. 445 */ 446 int 447 getnewvnode(tag, mp, vops, vpp) 448 enum vtagtype tag; 449 struct mount *mp; 450 vop_t **vops; 451 struct vnode **vpp; 452 { 453 int s, count; 454 struct proc *p = curproc; /* XXX */ 455 struct vnode *vp = NULL; 456 struct mount *vnmp; 457 vm_object_t object; 458 459 /* 460 * We take the least recently used vnode from the freelist 461 * if we can get it and it has no cached pages, and no 462 * namecache entries are relative to it. 463 * Otherwise we allocate a new vnode 464 */ 465 466 s = splbio(); 467 simple_lock(&vnode_free_list_slock); 468 469 if (wantfreevnodes && freevnodes < wantfreevnodes) { 470 vp = NULL; 471 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) { 472 /* 473 * XXX: this is only here to be backwards compatible 474 */ 475 vp = NULL; 476 } else for (count = 0; count < freevnodes; count++) { 477 vp = TAILQ_FIRST(&vnode_free_list); 478 if (vp == NULL || vp->v_usecount) 479 panic("getnewvnode: free vnode isn't"); 480 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 481 /* 482 * Don't recycle if active in the namecache or 483 * if it still has cached pages or we cannot get 484 * its interlock. 485 */ 486 object = vp->v_object; 487 if (LIST_FIRST(&vp->v_cache_src) != NULL || 488 (object && (object->resident_page_count || 489 object->ref_count)) || 490 !simple_lock_try(&vp->v_interlock)) { 491 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 492 vp = NULL; 493 continue; 494 } 495 /* 496 * Skip over it if its filesystem is being suspended. 497 */ 498 if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0) 499 break; 500 simple_unlock(&vp->v_interlock); 501 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 502 vp = NULL; 503 } 504 if (vp) { 505 vp->v_flag |= VDOOMED; 506 freevnodes--; 507 simple_unlock(&vnode_free_list_slock); 508 cache_purge(vp); 509 vp->v_lease = NULL; 510 if (vp->v_type != VBAD) { 511 vgonel(vp, p); 512 } else { 513 simple_unlock(&vp->v_interlock); 514 } 515 vn_finished_write(vnmp); 516 517 #ifdef INVARIANTS 518 { 519 int s; 520 521 if (vp->v_data) 522 panic("cleaned vnode isn't"); 523 s = splbio(); 524 if (vp->v_numoutput) 525 panic("Clean vnode has pending I/O's"); 526 splx(s); 527 if (vp->v_writecount != 0) 528 panic("Non-zero write count"); 529 } 530 #endif 531 vp->v_flag = 0; 532 vp->v_lastw = 0; 533 vp->v_lasta = 0; 534 vp->v_cstart = 0; 535 vp->v_clen = 0; 536 vp->v_socket = 0; 537 } else { 538 simple_unlock(&vnode_free_list_slock); 539 vp = (struct vnode *) zalloc(vnode_zone); 540 bzero((char *) vp, sizeof *vp); 541 simple_lock_init(&vp->v_interlock); 542 vp->v_dd = vp; 543 cache_purge(vp); 544 LIST_INIT(&vp->v_cache_src); 545 TAILQ_INIT(&vp->v_cache_dst); 546 numvnodes++; 547 } 548 549 TAILQ_INIT(&vp->v_cleanblkhd); 550 TAILQ_INIT(&vp->v_dirtyblkhd); 551 vp->v_type = VNON; 552 vp->v_tag = tag; 553 vp->v_op = vops; 554 insmntque(vp, mp); 555 *vpp = vp; 556 vp->v_usecount = 1; 557 vp->v_data = 0; 558 splx(s); 559 560 vfs_object_create(vp, p, p->p_ucred); 561 return (0); 562 } 563 564 /* 565 * Move a vnode from one mount queue to another. 566 */ 567 static void 568 insmntque(vp, mp) 569 register struct vnode *vp; 570 register struct mount *mp; 571 { 572 573 simple_lock(&mntvnode_slock); 574 /* 575 * Delete from old mount point vnode list, if on one. 576 */ 577 if (vp->v_mount != NULL) 578 LIST_REMOVE(vp, v_mntvnodes); 579 /* 580 * Insert into list of vnodes for the new mount point, if available. 581 */ 582 if ((vp->v_mount = mp) == NULL) { 583 simple_unlock(&mntvnode_slock); 584 return; 585 } 586 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 587 simple_unlock(&mntvnode_slock); 588 } 589 590 /* 591 * Update outstanding I/O count and do wakeup if requested. 592 */ 593 void 594 vwakeup(bp) 595 register struct buf *bp; 596 { 597 register struct vnode *vp; 598 599 bp->b_flags &= ~B_WRITEINPROG; 600 if ((vp = bp->b_vp)) { 601 vp->v_numoutput--; 602 if (vp->v_numoutput < 0) 603 panic("vwakeup: neg numoutput"); 604 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 605 vp->v_flag &= ~VBWAIT; 606 wakeup((caddr_t) &vp->v_numoutput); 607 } 608 } 609 } 610 611 /* 612 * Flush out and invalidate all buffers associated with a vnode. 613 * Called with the underlying object locked. 614 */ 615 int 616 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 617 register struct vnode *vp; 618 int flags; 619 struct ucred *cred; 620 struct proc *p; 621 int slpflag, slptimeo; 622 { 623 register struct buf *bp; 624 struct buf *nbp, *blist; 625 int s, error; 626 vm_object_t object; 627 628 if (flags & V_SAVE) { 629 s = splbio(); 630 while (vp->v_numoutput) { 631 vp->v_flag |= VBWAIT; 632 error = tsleep((caddr_t)&vp->v_numoutput, 633 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo); 634 if (error) { 635 splx(s); 636 return (error); 637 } 638 } 639 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 640 splx(s); 641 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0) 642 return (error); 643 s = splbio(); 644 if (vp->v_numoutput > 0 || 645 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 646 panic("vinvalbuf: dirty bufs"); 647 } 648 splx(s); 649 } 650 s = splbio(); 651 for (;;) { 652 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 653 if (!blist) 654 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 655 if (!blist) 656 break; 657 658 for (bp = blist; bp; bp = nbp) { 659 nbp = TAILQ_NEXT(bp, b_vnbufs); 660 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 661 error = BUF_TIMELOCK(bp, 662 LK_EXCLUSIVE | LK_SLEEPFAIL, 663 "vinvalbuf", slpflag, slptimeo); 664 if (error == ENOLCK) 665 break; 666 splx(s); 667 return (error); 668 } 669 /* 670 * XXX Since there are no node locks for NFS, I 671 * believe there is a slight chance that a delayed 672 * write will occur while sleeping just above, so 673 * check for it. Note that vfs_bio_awrite expects 674 * buffers to reside on a queue, while VOP_BWRITE and 675 * brelse do not. 676 */ 677 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 678 (flags & V_SAVE)) { 679 680 if (bp->b_vp == vp) { 681 if (bp->b_flags & B_CLUSTEROK) { 682 BUF_UNLOCK(bp); 683 vfs_bio_awrite(bp); 684 } else { 685 bremfree(bp); 686 bp->b_flags |= B_ASYNC; 687 BUF_WRITE(bp); 688 } 689 } else { 690 bremfree(bp); 691 (void) BUF_WRITE(bp); 692 } 693 break; 694 } 695 bremfree(bp); 696 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 697 bp->b_flags &= ~B_ASYNC; 698 brelse(bp); 699 } 700 } 701 702 while (vp->v_numoutput > 0) { 703 vp->v_flag |= VBWAIT; 704 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 705 } 706 707 splx(s); 708 709 /* 710 * Destroy the copy in the VM cache, too. 711 */ 712 simple_lock(&vp->v_interlock); 713 object = vp->v_object; 714 if (object != NULL) { 715 vm_object_page_remove(object, 0, 0, 716 (flags & V_SAVE) ? TRUE : FALSE); 717 } 718 simple_unlock(&vp->v_interlock); 719 720 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 721 panic("vinvalbuf: flush failed"); 722 return (0); 723 } 724 725 /* 726 * Truncate a file's buffer and pages to a specified length. This 727 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 728 * sync activity. 729 */ 730 int 731 vtruncbuf(vp, cred, p, length, blksize) 732 register struct vnode *vp; 733 struct ucred *cred; 734 struct proc *p; 735 off_t length; 736 int blksize; 737 { 738 register struct buf *bp; 739 struct buf *nbp; 740 int s, anyfreed; 741 int trunclbn; 742 743 /* 744 * Round up to the *next* lbn. 745 */ 746 trunclbn = (length + blksize - 1) / blksize; 747 748 s = splbio(); 749 restart: 750 anyfreed = 1; 751 for (;anyfreed;) { 752 anyfreed = 0; 753 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 754 nbp = TAILQ_NEXT(bp, b_vnbufs); 755 if (bp->b_lblkno >= trunclbn) { 756 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 757 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 758 goto restart; 759 } else { 760 bremfree(bp); 761 bp->b_flags |= (B_INVAL | B_RELBUF); 762 bp->b_flags &= ~B_ASYNC; 763 brelse(bp); 764 anyfreed = 1; 765 } 766 if (nbp && 767 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 768 (nbp->b_vp != vp) || 769 (nbp->b_flags & B_DELWRI))) { 770 goto restart; 771 } 772 } 773 } 774 775 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 776 nbp = TAILQ_NEXT(bp, b_vnbufs); 777 if (bp->b_lblkno >= trunclbn) { 778 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 779 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 780 goto restart; 781 } else { 782 bremfree(bp); 783 bp->b_flags |= (B_INVAL | B_RELBUF); 784 bp->b_flags &= ~B_ASYNC; 785 brelse(bp); 786 anyfreed = 1; 787 } 788 if (nbp && 789 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 790 (nbp->b_vp != vp) || 791 (nbp->b_flags & B_DELWRI) == 0)) { 792 goto restart; 793 } 794 } 795 } 796 } 797 798 if (length > 0) { 799 restartsync: 800 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 801 nbp = TAILQ_NEXT(bp, b_vnbufs); 802 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 803 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 804 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 805 goto restart; 806 } else { 807 bremfree(bp); 808 if (bp->b_vp == vp) { 809 bp->b_flags |= B_ASYNC; 810 } else { 811 bp->b_flags &= ~B_ASYNC; 812 } 813 BUF_WRITE(bp); 814 } 815 goto restartsync; 816 } 817 818 } 819 } 820 821 while (vp->v_numoutput > 0) { 822 vp->v_flag |= VBWAIT; 823 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0); 824 } 825 826 splx(s); 827 828 vnode_pager_setsize(vp, length); 829 830 return (0); 831 } 832 833 /* 834 * Associate a buffer with a vnode. 835 */ 836 void 837 bgetvp(vp, bp) 838 register struct vnode *vp; 839 register struct buf *bp; 840 { 841 int s; 842 843 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 844 845 vhold(vp); 846 bp->b_vp = vp; 847 bp->b_dev = vn_todev(vp); 848 /* 849 * Insert onto list for new vnode. 850 */ 851 s = splbio(); 852 bp->b_xflags |= BX_VNCLEAN; 853 bp->b_xflags &= ~BX_VNDIRTY; 854 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 855 splx(s); 856 } 857 858 /* 859 * Disassociate a buffer from a vnode. 860 */ 861 void 862 brelvp(bp) 863 register struct buf *bp; 864 { 865 struct vnode *vp; 866 struct buflists *listheadp; 867 int s; 868 869 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 870 871 /* 872 * Delete from old vnode list, if on one. 873 */ 874 vp = bp->b_vp; 875 s = splbio(); 876 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 877 if (bp->b_xflags & BX_VNDIRTY) 878 listheadp = &vp->v_dirtyblkhd; 879 else 880 listheadp = &vp->v_cleanblkhd; 881 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 882 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 883 } 884 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 885 vp->v_flag &= ~VONWORKLST; 886 LIST_REMOVE(vp, v_synclist); 887 } 888 splx(s); 889 bp->b_vp = (struct vnode *) 0; 890 vdrop(vp); 891 } 892 893 /* 894 * The workitem queue. 895 * 896 * It is useful to delay writes of file data and filesystem metadata 897 * for tens of seconds so that quickly created and deleted files need 898 * not waste disk bandwidth being created and removed. To realize this, 899 * we append vnodes to a "workitem" queue. When running with a soft 900 * updates implementation, most pending metadata dependencies should 901 * not wait for more than a few seconds. Thus, mounted on block devices 902 * are delayed only about a half the time that file data is delayed. 903 * Similarly, directory updates are more critical, so are only delayed 904 * about a third the time that file data is delayed. Thus, there are 905 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 906 * one each second (driven off the filesystem syncer process). The 907 * syncer_delayno variable indicates the next queue that is to be processed. 908 * Items that need to be processed soon are placed in this queue: 909 * 910 * syncer_workitem_pending[syncer_delayno] 911 * 912 * A delay of fifteen seconds is done by placing the request fifteen 913 * entries later in the queue: 914 * 915 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 916 * 917 */ 918 919 /* 920 * Add an item to the syncer work queue. 921 */ 922 static void 923 vn_syncer_add_to_worklist(struct vnode *vp, int delay) 924 { 925 int s, slot; 926 927 s = splbio(); 928 929 if (vp->v_flag & VONWORKLST) { 930 LIST_REMOVE(vp, v_synclist); 931 } 932 933 if (delay > syncer_maxdelay - 2) 934 delay = syncer_maxdelay - 2; 935 slot = (syncer_delayno + delay) & syncer_mask; 936 937 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 938 vp->v_flag |= VONWORKLST; 939 splx(s); 940 } 941 942 struct proc *updateproc; 943 static void sched_sync __P((void)); 944 static struct kproc_desc up_kp = { 945 "syncer", 946 sched_sync, 947 &updateproc 948 }; 949 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 950 951 /* 952 * System filesystem synchronizer daemon. 953 */ 954 void 955 sched_sync(void) 956 { 957 struct synclist *slp; 958 struct vnode *vp; 959 struct mount *mp; 960 long starttime; 961 int s; 962 struct proc *p = updateproc; 963 964 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, p, 965 SHUTDOWN_PRI_LAST); 966 967 for (;;) { 968 kproc_suspend_loop(p); 969 970 starttime = time_second; 971 972 /* 973 * Push files whose dirty time has expired. Be careful 974 * of interrupt race on slp queue. 975 */ 976 s = splbio(); 977 slp = &syncer_workitem_pending[syncer_delayno]; 978 syncer_delayno += 1; 979 if (syncer_delayno == syncer_maxdelay) 980 syncer_delayno = 0; 981 splx(s); 982 983 while ((vp = LIST_FIRST(slp)) != NULL) { 984 if (VOP_ISLOCKED(vp, NULL) == 0 && 985 vn_start_write(vp, &mp, V_NOWAIT) == 0) { 986 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 987 (void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p); 988 VOP_UNLOCK(vp, 0, p); 989 vn_finished_write(mp); 990 } 991 s = splbio(); 992 if (LIST_FIRST(slp) == vp) { 993 /* 994 * Note: v_tag VT_VFS vps can remain on the 995 * worklist too with no dirty blocks, but 996 * since sync_fsync() moves it to a different 997 * slot we are safe. 998 */ 999 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1000 !vn_isdisk(vp, NULL)) 1001 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1002 /* 1003 * Put us back on the worklist. The worklist 1004 * routine will remove us from our current 1005 * position and then add us back in at a later 1006 * position. 1007 */ 1008 vn_syncer_add_to_worklist(vp, syncdelay); 1009 } 1010 splx(s); 1011 } 1012 1013 /* 1014 * Do soft update processing. 1015 */ 1016 #ifdef SOFTUPDATES 1017 softdep_process_worklist(NULL); 1018 #endif 1019 1020 /* 1021 * The variable rushjob allows the kernel to speed up the 1022 * processing of the filesystem syncer process. A rushjob 1023 * value of N tells the filesystem syncer to process the next 1024 * N seconds worth of work on its queue ASAP. Currently rushjob 1025 * is used by the soft update code to speed up the filesystem 1026 * syncer process when the incore state is getting so far 1027 * ahead of the disk that the kernel memory pool is being 1028 * threatened with exhaustion. 1029 */ 1030 if (rushjob > 0) { 1031 rushjob -= 1; 1032 continue; 1033 } 1034 /* 1035 * If it has taken us less than a second to process the 1036 * current work, then wait. Otherwise start right over 1037 * again. We can still lose time if any single round 1038 * takes more than two seconds, but it does not really 1039 * matter as we are just trying to generally pace the 1040 * filesystem activity. 1041 */ 1042 if (time_second == starttime) 1043 tsleep(&lbolt, PPAUSE, "syncer", 0); 1044 } 1045 } 1046 1047 /* 1048 * Request the syncer daemon to speed up its work. 1049 * We never push it to speed up more than half of its 1050 * normal turn time, otherwise it could take over the cpu. 1051 */ 1052 int 1053 speedup_syncer() 1054 { 1055 int s; 1056 1057 s = splhigh(); 1058 if (updateproc->p_wchan == &lbolt) 1059 setrunnable(updateproc); 1060 splx(s); 1061 if (rushjob < syncdelay / 2) { 1062 rushjob += 1; 1063 stat_rush_requests += 1; 1064 return (1); 1065 } 1066 return(0); 1067 } 1068 1069 /* 1070 * Associate a p-buffer with a vnode. 1071 * 1072 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1073 * with the buffer. i.e. the bp has not been linked into the vnode or 1074 * ref-counted. 1075 */ 1076 void 1077 pbgetvp(vp, bp) 1078 register struct vnode *vp; 1079 register struct buf *bp; 1080 { 1081 1082 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1083 1084 bp->b_vp = vp; 1085 bp->b_flags |= B_PAGING; 1086 bp->b_dev = vn_todev(vp); 1087 } 1088 1089 /* 1090 * Disassociate a p-buffer from a vnode. 1091 */ 1092 void 1093 pbrelvp(bp) 1094 register struct buf *bp; 1095 { 1096 1097 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1098 1099 /* XXX REMOVE ME */ 1100 if (bp->b_vnbufs.tqe_next != NULL) { 1101 panic( 1102 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1103 bp, 1104 (int)bp->b_flags 1105 ); 1106 } 1107 bp->b_vp = (struct vnode *) 0; 1108 bp->b_flags &= ~B_PAGING; 1109 } 1110 1111 void 1112 pbreassignbuf(bp, newvp) 1113 struct buf *bp; 1114 struct vnode *newvp; 1115 { 1116 if ((bp->b_flags & B_PAGING) == 0) { 1117 panic( 1118 "pbreassignbuf() on non phys bp %p", 1119 bp 1120 ); 1121 } 1122 bp->b_vp = newvp; 1123 } 1124 1125 /* 1126 * Reassign a buffer from one vnode to another. 1127 * Used to assign file specific control information 1128 * (indirect blocks) to the vnode to which they belong. 1129 */ 1130 void 1131 reassignbuf(bp, newvp) 1132 register struct buf *bp; 1133 register struct vnode *newvp; 1134 { 1135 struct buflists *listheadp; 1136 int delay; 1137 int s; 1138 1139 if (newvp == NULL) { 1140 printf("reassignbuf: NULL"); 1141 return; 1142 } 1143 ++reassignbufcalls; 1144 1145 /* 1146 * B_PAGING flagged buffers cannot be reassigned because their vp 1147 * is not fully linked in. 1148 */ 1149 if (bp->b_flags & B_PAGING) 1150 panic("cannot reassign paging buffer"); 1151 1152 s = splbio(); 1153 /* 1154 * Delete from old vnode list, if on one. 1155 */ 1156 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1157 if (bp->b_xflags & BX_VNDIRTY) 1158 listheadp = &bp->b_vp->v_dirtyblkhd; 1159 else 1160 listheadp = &bp->b_vp->v_cleanblkhd; 1161 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1162 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1163 if (bp->b_vp != newvp) { 1164 vdrop(bp->b_vp); 1165 bp->b_vp = NULL; /* for clarification */ 1166 } 1167 } 1168 /* 1169 * If dirty, put on list of dirty buffers; otherwise insert onto list 1170 * of clean buffers. 1171 */ 1172 if (bp->b_flags & B_DELWRI) { 1173 struct buf *tbp; 1174 1175 listheadp = &newvp->v_dirtyblkhd; 1176 if ((newvp->v_flag & VONWORKLST) == 0) { 1177 switch (newvp->v_type) { 1178 case VDIR: 1179 delay = dirdelay; 1180 break; 1181 case VCHR: 1182 case VBLK: 1183 if (newvp->v_specmountpoint != NULL) { 1184 delay = metadelay; 1185 break; 1186 } 1187 /* fall through */ 1188 default: 1189 delay = filedelay; 1190 } 1191 vn_syncer_add_to_worklist(newvp, delay); 1192 } 1193 bp->b_xflags |= BX_VNDIRTY; 1194 tbp = TAILQ_FIRST(listheadp); 1195 if (tbp == NULL || 1196 bp->b_lblkno == 0 || 1197 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1198 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1199 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1200 ++reassignbufsortgood; 1201 } else if (bp->b_lblkno < 0) { 1202 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1203 ++reassignbufsortgood; 1204 } else if (reassignbufmethod == 1) { 1205 /* 1206 * New sorting algorithm, only handle sequential case, 1207 * otherwise append to end (but before metadata) 1208 */ 1209 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1210 (tbp->b_xflags & BX_VNDIRTY)) { 1211 /* 1212 * Found the best place to insert the buffer 1213 */ 1214 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1215 ++reassignbufsortgood; 1216 } else { 1217 /* 1218 * Missed, append to end, but before meta-data. 1219 * We know that the head buffer in the list is 1220 * not meta-data due to prior conditionals. 1221 * 1222 * Indirect effects: NFS second stage write 1223 * tends to wind up here, giving maximum 1224 * distance between the unstable write and the 1225 * commit rpc. 1226 */ 1227 tbp = TAILQ_LAST(listheadp, buflists); 1228 while (tbp && tbp->b_lblkno < 0) 1229 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1230 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1231 ++reassignbufsortbad; 1232 } 1233 } else { 1234 /* 1235 * Old sorting algorithm, scan queue and insert 1236 */ 1237 struct buf *ttbp; 1238 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1239 (ttbp->b_lblkno < bp->b_lblkno)) { 1240 ++reassignbufloops; 1241 tbp = ttbp; 1242 } 1243 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1244 } 1245 } else { 1246 bp->b_xflags |= BX_VNCLEAN; 1247 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1248 if ((newvp->v_flag & VONWORKLST) && 1249 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1250 newvp->v_flag &= ~VONWORKLST; 1251 LIST_REMOVE(newvp, v_synclist); 1252 } 1253 } 1254 if (bp->b_vp != newvp) { 1255 bp->b_vp = newvp; 1256 vhold(bp->b_vp); 1257 } 1258 splx(s); 1259 } 1260 1261 /* 1262 * Create a vnode for a block device. 1263 * Used for mounting the root file system. 1264 * XXX: This now changed to a VCHR due to the block/char merging. 1265 */ 1266 int 1267 bdevvp(dev, vpp) 1268 dev_t dev; 1269 struct vnode **vpp; 1270 { 1271 register struct vnode *vp; 1272 struct vnode *nvp; 1273 int error; 1274 1275 if (dev == NODEV) { 1276 *vpp = NULLVP; 1277 return (ENXIO); 1278 } 1279 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1280 if (error) { 1281 *vpp = NULLVP; 1282 return (error); 1283 } 1284 vp = nvp; 1285 vp->v_type = VCHR; 1286 addalias(vp, dev); 1287 *vpp = vp; 1288 return (0); 1289 } 1290 1291 /* 1292 * Add vnode to the alias list hung off the dev_t. 1293 * 1294 * The reason for this gunk is that multiple vnodes can reference 1295 * the same physical device, so checking vp->v_usecount to see 1296 * how many users there are is inadequate; the v_usecount for 1297 * the vnodes need to be accumulated. vcount() does that. 1298 */ 1299 struct vnode * 1300 addaliasu(nvp, nvp_rdev) 1301 struct vnode *nvp; 1302 udev_t nvp_rdev; 1303 { 1304 struct vnode *ovp; 1305 vop_t **ops; 1306 dev_t dev; 1307 1308 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1309 panic("addaliasu on non-special vnode"); 1310 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0); 1311 /* 1312 * Check to see if we have a bdevvp vnode with no associated 1313 * filesystem. If so, we want to associate the filesystem of 1314 * the new newly instigated vnode with the bdevvp vnode and 1315 * discard the newly created vnode rather than leaving the 1316 * bdevvp vnode lying around with no associated filesystem. 1317 */ 1318 if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) { 1319 addalias(nvp, dev); 1320 return (nvp); 1321 } 1322 /* 1323 * Discard unneeded vnode, but save its node specific data. 1324 * Note that if there is a lock, it is carried over in the 1325 * node specific data to the replacement vnode. 1326 */ 1327 vref(ovp); 1328 ovp->v_data = nvp->v_data; 1329 ovp->v_tag = nvp->v_tag; 1330 nvp->v_data = NULL; 1331 ops = nvp->v_op; 1332 nvp->v_op = ovp->v_op; 1333 ovp->v_op = ops; 1334 insmntque(ovp, nvp->v_mount); 1335 vrele(nvp); 1336 vgone(nvp); 1337 return (ovp); 1338 } 1339 1340 void 1341 addalias(nvp, dev) 1342 struct vnode *nvp; 1343 dev_t dev; 1344 { 1345 1346 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1347 panic("addalias on non-special vnode"); 1348 1349 nvp->v_rdev = dev; 1350 simple_lock(&spechash_slock); 1351 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1352 simple_unlock(&spechash_slock); 1353 } 1354 1355 /* 1356 * Grab a particular vnode from the free list, increment its 1357 * reference count and lock it. The vnode lock bit is set if the 1358 * vnode is being eliminated in vgone. The process is awakened 1359 * when the transition is completed, and an error returned to 1360 * indicate that the vnode is no longer usable (possibly having 1361 * been changed to a new file system type). 1362 */ 1363 int 1364 vget(vp, flags, p) 1365 register struct vnode *vp; 1366 int flags; 1367 struct proc *p; 1368 { 1369 int error; 1370 1371 /* 1372 * If the vnode is in the process of being cleaned out for 1373 * another use, we wait for the cleaning to finish and then 1374 * return failure. Cleaning is determined by checking that 1375 * the VXLOCK flag is set. 1376 */ 1377 if ((flags & LK_INTERLOCK) == 0) { 1378 simple_lock(&vp->v_interlock); 1379 } 1380 if (vp->v_flag & VXLOCK) { 1381 vp->v_flag |= VXWANT; 1382 simple_unlock(&vp->v_interlock); 1383 tsleep((caddr_t)vp, PINOD, "vget", 0); 1384 return (ENOENT); 1385 } 1386 1387 vp->v_usecount++; 1388 1389 if (VSHOULDBUSY(vp)) 1390 vbusy(vp); 1391 if (flags & LK_TYPE_MASK) { 1392 if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) { 1393 /* 1394 * must expand vrele here because we do not want 1395 * to call VOP_INACTIVE if the reference count 1396 * drops back to zero since it was never really 1397 * active. We must remove it from the free list 1398 * before sleeping so that multiple processes do 1399 * not try to recycle it. 1400 */ 1401 simple_lock(&vp->v_interlock); 1402 vp->v_usecount--; 1403 if (VSHOULDFREE(vp)) 1404 vfree(vp); 1405 simple_unlock(&vp->v_interlock); 1406 } 1407 return (error); 1408 } 1409 simple_unlock(&vp->v_interlock); 1410 return (0); 1411 } 1412 1413 void 1414 vref(struct vnode *vp) 1415 { 1416 simple_lock(&vp->v_interlock); 1417 vp->v_usecount++; 1418 simple_unlock(&vp->v_interlock); 1419 } 1420 1421 /* 1422 * Vnode put/release. 1423 * If count drops to zero, call inactive routine and return to freelist. 1424 */ 1425 void 1426 vrele(vp) 1427 struct vnode *vp; 1428 { 1429 struct proc *p = curproc; /* XXX */ 1430 1431 KASSERT(vp != NULL, ("vrele: null vp")); 1432 KASSERT(vp->v_writecount < vp->v_usecount, ("vrele: missed vn_close")); 1433 1434 simple_lock(&vp->v_interlock); 1435 1436 if (vp->v_usecount > 1) { 1437 1438 vp->v_usecount--; 1439 simple_unlock(&vp->v_interlock); 1440 1441 return; 1442 } 1443 1444 if (vp->v_usecount == 1) { 1445 1446 vp->v_usecount--; 1447 if (VSHOULDFREE(vp)) 1448 vfree(vp); 1449 /* 1450 * If we are doing a vput, the node is already locked, and we must 1451 * call VOP_INACTIVE with the node locked. So, in the case of 1452 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1453 */ 1454 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 1455 VOP_INACTIVE(vp, p); 1456 } 1457 1458 } else { 1459 #ifdef DIAGNOSTIC 1460 vprint("vrele: negative ref count", vp); 1461 simple_unlock(&vp->v_interlock); 1462 #endif 1463 panic("vrele: negative ref cnt"); 1464 } 1465 } 1466 1467 void 1468 vput(vp) 1469 struct vnode *vp; 1470 { 1471 struct proc *p = curproc; /* XXX */ 1472 1473 KASSERT(vp != NULL, ("vput: null vp")); 1474 KASSERT(vp->v_writecount < vp->v_usecount, ("vput: missed vn_close")); 1475 1476 simple_lock(&vp->v_interlock); 1477 1478 if (vp->v_usecount > 1) { 1479 1480 vp->v_usecount--; 1481 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1482 return; 1483 1484 } 1485 1486 if (vp->v_usecount == 1) { 1487 1488 vp->v_usecount--; 1489 if (VSHOULDFREE(vp)) 1490 vfree(vp); 1491 /* 1492 * If we are doing a vput, the node is already locked, and we must 1493 * call VOP_INACTIVE with the node locked. So, in the case of 1494 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1495 */ 1496 simple_unlock(&vp->v_interlock); 1497 VOP_INACTIVE(vp, p); 1498 1499 } else { 1500 #ifdef DIAGNOSTIC 1501 vprint("vput: negative ref count", vp); 1502 #endif 1503 panic("vput: negative ref cnt"); 1504 } 1505 } 1506 1507 /* 1508 * Somebody doesn't want the vnode recycled. 1509 */ 1510 void 1511 vhold(vp) 1512 register struct vnode *vp; 1513 { 1514 int s; 1515 1516 s = splbio(); 1517 vp->v_holdcnt++; 1518 if (VSHOULDBUSY(vp)) 1519 vbusy(vp); 1520 splx(s); 1521 } 1522 1523 /* 1524 * One less who cares about this vnode. 1525 */ 1526 void 1527 vdrop(vp) 1528 register struct vnode *vp; 1529 { 1530 int s; 1531 1532 s = splbio(); 1533 if (vp->v_holdcnt <= 0) 1534 panic("vdrop: holdcnt"); 1535 vp->v_holdcnt--; 1536 if (VSHOULDFREE(vp)) 1537 vfree(vp); 1538 splx(s); 1539 } 1540 1541 /* 1542 * Remove any vnodes in the vnode table belonging to mount point mp. 1543 * 1544 * If MNT_NOFORCE is specified, there should not be any active ones, 1545 * return error if any are found (nb: this is a user error, not a 1546 * system error). If MNT_FORCE is specified, detach any active vnodes 1547 * that are found. 1548 */ 1549 #ifdef DIAGNOSTIC 1550 static int busyprt = 0; /* print out busy vnodes */ 1551 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1552 #endif 1553 1554 int 1555 vflush(mp, skipvp, flags) 1556 struct mount *mp; 1557 struct vnode *skipvp; 1558 int flags; 1559 { 1560 struct proc *p = curproc; /* XXX */ 1561 struct vnode *vp, *nvp; 1562 int busy = 0; 1563 1564 simple_lock(&mntvnode_slock); 1565 loop: 1566 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) { 1567 /* 1568 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1569 * Start over if it has (it won't be on the list anymore). 1570 */ 1571 if (vp->v_mount != mp) 1572 goto loop; 1573 nvp = LIST_NEXT(vp, v_mntvnodes); 1574 /* 1575 * Skip over a selected vnode. 1576 */ 1577 if (vp == skipvp) 1578 continue; 1579 1580 simple_lock(&vp->v_interlock); 1581 /* 1582 * Skip over a vnodes marked VSYSTEM. 1583 */ 1584 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1585 simple_unlock(&vp->v_interlock); 1586 continue; 1587 } 1588 /* 1589 * If WRITECLOSE is set, only flush out regular file vnodes 1590 * open for writing. 1591 */ 1592 if ((flags & WRITECLOSE) && 1593 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1594 simple_unlock(&vp->v_interlock); 1595 continue; 1596 } 1597 1598 /* 1599 * With v_usecount == 0, all we need to do is clear out the 1600 * vnode data structures and we are done. 1601 */ 1602 if (vp->v_usecount == 0) { 1603 simple_unlock(&mntvnode_slock); 1604 vgonel(vp, p); 1605 simple_lock(&mntvnode_slock); 1606 continue; 1607 } 1608 1609 /* 1610 * If FORCECLOSE is set, forcibly close the vnode. For block 1611 * or character devices, revert to an anonymous device. For 1612 * all other files, just kill them. 1613 */ 1614 if (flags & FORCECLOSE) { 1615 simple_unlock(&mntvnode_slock); 1616 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1617 vgonel(vp, p); 1618 } else { 1619 vclean(vp, 0, p); 1620 vp->v_op = spec_vnodeop_p; 1621 insmntque(vp, (struct mount *) 0); 1622 } 1623 simple_lock(&mntvnode_slock); 1624 continue; 1625 } 1626 #ifdef DIAGNOSTIC 1627 if (busyprt) 1628 vprint("vflush: busy vnode", vp); 1629 #endif 1630 simple_unlock(&vp->v_interlock); 1631 busy++; 1632 } 1633 simple_unlock(&mntvnode_slock); 1634 if (busy) 1635 return (EBUSY); 1636 return (0); 1637 } 1638 1639 /* 1640 * Disassociate the underlying file system from a vnode. 1641 */ 1642 static void 1643 vclean(vp, flags, p) 1644 struct vnode *vp; 1645 int flags; 1646 struct proc *p; 1647 { 1648 int active; 1649 vm_object_t obj; 1650 1651 /* 1652 * Check to see if the vnode is in use. If so we have to reference it 1653 * before we clean it out so that its count cannot fall to zero and 1654 * generate a race against ourselves to recycle it. 1655 */ 1656 if ((active = vp->v_usecount)) 1657 vp->v_usecount++; 1658 1659 /* 1660 * Prevent the vnode from being recycled or brought into use while we 1661 * clean it out. 1662 */ 1663 if (vp->v_flag & VXLOCK) 1664 panic("vclean: deadlock"); 1665 vp->v_flag |= VXLOCK; 1666 /* 1667 * Even if the count is zero, the VOP_INACTIVE routine may still 1668 * have the object locked while it cleans it out. The VOP_LOCK 1669 * ensures that the VOP_INACTIVE routine is done with its work. 1670 * For active vnodes, it ensures that no other activity can 1671 * occur while the underlying object is being cleaned out. 1672 */ 1673 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1674 1675 /* 1676 * Clean out any buffers associated with the vnode. 1677 * If the flush fails, just toss the buffers. 1678 */ 1679 if (flags & DOCLOSE) { 1680 if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL) 1681 (void) vn_write_suspend_wait(vp, NULL, V_WAIT); 1682 if (vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0) != 0) 1683 vinvalbuf(vp, 0, NOCRED, p, 0, 0); 1684 } 1685 1686 if ((obj = vp->v_object) != NULL) { 1687 if (obj->ref_count == 0) { 1688 /* 1689 * vclean() may be called twice. The first time 1690 * removes the primary reference to the object, 1691 * the second time goes one further and is a 1692 * special-case to terminate the object. 1693 */ 1694 vm_object_terminate(obj); 1695 } else { 1696 /* 1697 * Woe to the process that tries to page now :-). 1698 */ 1699 vm_pager_deallocate(obj); 1700 } 1701 } 1702 1703 /* 1704 * If purging an active vnode, it must be closed and 1705 * deactivated before being reclaimed. Note that the 1706 * VOP_INACTIVE will unlock the vnode. 1707 */ 1708 if (active) { 1709 if (flags & DOCLOSE) 1710 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p); 1711 VOP_INACTIVE(vp, p); 1712 } else { 1713 /* 1714 * Any other processes trying to obtain this lock must first 1715 * wait for VXLOCK to clear, then call the new lock operation. 1716 */ 1717 VOP_UNLOCK(vp, 0, p); 1718 } 1719 /* 1720 * Reclaim the vnode. 1721 */ 1722 if (VOP_RECLAIM(vp, p)) 1723 panic("vclean: cannot reclaim"); 1724 1725 if (active) { 1726 /* 1727 * Inline copy of vrele() since VOP_INACTIVE 1728 * has already been called. 1729 */ 1730 simple_lock(&vp->v_interlock); 1731 if (--vp->v_usecount <= 0) { 1732 #ifdef DIAGNOSTIC 1733 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1734 vprint("vclean: bad ref count", vp); 1735 panic("vclean: ref cnt"); 1736 } 1737 #endif 1738 vfree(vp); 1739 } 1740 simple_unlock(&vp->v_interlock); 1741 } 1742 1743 cache_purge(vp); 1744 if (vp->v_vnlock) { 1745 FREE(vp->v_vnlock, M_VNODE); 1746 vp->v_vnlock = NULL; 1747 } 1748 1749 if (VSHOULDFREE(vp)) 1750 vfree(vp); 1751 1752 /* 1753 * Done with purge, notify sleepers of the grim news. 1754 */ 1755 vp->v_op = dead_vnodeop_p; 1756 vn_pollgone(vp); 1757 vp->v_tag = VT_NON; 1758 vp->v_flag &= ~VXLOCK; 1759 if (vp->v_flag & VXWANT) { 1760 vp->v_flag &= ~VXWANT; 1761 wakeup((caddr_t) vp); 1762 } 1763 } 1764 1765 /* 1766 * Eliminate all activity associated with the requested vnode 1767 * and with all vnodes aliased to the requested vnode. 1768 */ 1769 int 1770 vop_revoke(ap) 1771 struct vop_revoke_args /* { 1772 struct vnode *a_vp; 1773 int a_flags; 1774 } */ *ap; 1775 { 1776 struct vnode *vp, *vq; 1777 dev_t dev; 1778 1779 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1780 1781 vp = ap->a_vp; 1782 /* 1783 * If a vgone (or vclean) is already in progress, 1784 * wait until it is done and return. 1785 */ 1786 if (vp->v_flag & VXLOCK) { 1787 vp->v_flag |= VXWANT; 1788 simple_unlock(&vp->v_interlock); 1789 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1790 return (0); 1791 } 1792 dev = vp->v_rdev; 1793 for (;;) { 1794 simple_lock(&spechash_slock); 1795 vq = SLIST_FIRST(&dev->si_hlist); 1796 simple_unlock(&spechash_slock); 1797 if (!vq) 1798 break; 1799 vgone(vq); 1800 } 1801 return (0); 1802 } 1803 1804 /* 1805 * Recycle an unused vnode to the front of the free list. 1806 * Release the passed interlock if the vnode will be recycled. 1807 */ 1808 int 1809 vrecycle(vp, inter_lkp, p) 1810 struct vnode *vp; 1811 struct simplelock *inter_lkp; 1812 struct proc *p; 1813 { 1814 1815 simple_lock(&vp->v_interlock); 1816 if (vp->v_usecount == 0) { 1817 if (inter_lkp) { 1818 simple_unlock(inter_lkp); 1819 } 1820 vgonel(vp, p); 1821 return (1); 1822 } 1823 simple_unlock(&vp->v_interlock); 1824 return (0); 1825 } 1826 1827 /* 1828 * Eliminate all activity associated with a vnode 1829 * in preparation for reuse. 1830 */ 1831 void 1832 vgone(vp) 1833 register struct vnode *vp; 1834 { 1835 struct proc *p = curproc; /* XXX */ 1836 1837 simple_lock(&vp->v_interlock); 1838 vgonel(vp, p); 1839 } 1840 1841 /* 1842 * vgone, with the vp interlock held. 1843 */ 1844 void 1845 vgonel(vp, p) 1846 struct vnode *vp; 1847 struct proc *p; 1848 { 1849 int s; 1850 1851 /* 1852 * If a vgone (or vclean) is already in progress, 1853 * wait until it is done and return. 1854 */ 1855 if (vp->v_flag & VXLOCK) { 1856 vp->v_flag |= VXWANT; 1857 simple_unlock(&vp->v_interlock); 1858 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1859 return; 1860 } 1861 1862 /* 1863 * Clean out the filesystem specific data. 1864 */ 1865 vclean(vp, DOCLOSE, p); 1866 simple_lock(&vp->v_interlock); 1867 1868 /* 1869 * Delete from old mount point vnode list, if on one. 1870 */ 1871 if (vp->v_mount != NULL) 1872 insmntque(vp, (struct mount *)0); 1873 /* 1874 * If special device, remove it from special device alias list 1875 * if it is on one. 1876 */ 1877 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1878 simple_lock(&spechash_slock); 1879 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext); 1880 freedev(vp->v_rdev); 1881 simple_unlock(&spechash_slock); 1882 vp->v_rdev = NULL; 1883 } 1884 1885 /* 1886 * If it is on the freelist and not already at the head, 1887 * move it to the head of the list. The test of the 1888 * VDOOMED flag and the reference count of zero is because 1889 * it will be removed from the free list by getnewvnode, 1890 * but will not have its reference count incremented until 1891 * after calling vgone. If the reference count were 1892 * incremented first, vgone would (incorrectly) try to 1893 * close the previous instance of the underlying object. 1894 */ 1895 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1896 s = splbio(); 1897 simple_lock(&vnode_free_list_slock); 1898 if (vp->v_flag & VFREE) 1899 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1900 else 1901 freevnodes++; 1902 vp->v_flag |= VFREE; 1903 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1904 simple_unlock(&vnode_free_list_slock); 1905 splx(s); 1906 } 1907 1908 vp->v_type = VBAD; 1909 simple_unlock(&vp->v_interlock); 1910 } 1911 1912 /* 1913 * Lookup a vnode by device number. 1914 */ 1915 int 1916 vfinddev(dev, type, vpp) 1917 dev_t dev; 1918 enum vtype type; 1919 struct vnode **vpp; 1920 { 1921 struct vnode *vp; 1922 1923 simple_lock(&spechash_slock); 1924 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1925 if (type == vp->v_type) { 1926 *vpp = vp; 1927 simple_unlock(&spechash_slock); 1928 return (1); 1929 } 1930 } 1931 simple_unlock(&spechash_slock); 1932 return (0); 1933 } 1934 1935 /* 1936 * Calculate the total number of references to a special device. 1937 */ 1938 int 1939 vcount(vp) 1940 struct vnode *vp; 1941 { 1942 struct vnode *vq; 1943 int count; 1944 1945 count = 0; 1946 simple_lock(&spechash_slock); 1947 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext) 1948 count += vq->v_usecount; 1949 simple_unlock(&spechash_slock); 1950 return (count); 1951 } 1952 1953 /* 1954 * Same as above, but using the dev_t as argument 1955 */ 1956 1957 int 1958 count_dev(dev) 1959 dev_t dev; 1960 { 1961 struct vnode *vp; 1962 1963 vp = SLIST_FIRST(&dev->si_hlist); 1964 if (vp == NULL) 1965 return (0); 1966 return(vcount(vp)); 1967 } 1968 1969 /* 1970 * Print out a description of a vnode. 1971 */ 1972 static char *typename[] = 1973 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1974 1975 void 1976 vprint(label, vp) 1977 char *label; 1978 struct vnode *vp; 1979 { 1980 char buf[96]; 1981 1982 if (label != NULL) 1983 printf("%s: %p: ", label, (void *)vp); 1984 else 1985 printf("%p: ", (void *)vp); 1986 printf("type %s, usecount %d, writecount %d, refcount %d,", 1987 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1988 vp->v_holdcnt); 1989 buf[0] = '\0'; 1990 if (vp->v_flag & VROOT) 1991 strcat(buf, "|VROOT"); 1992 if (vp->v_flag & VTEXT) 1993 strcat(buf, "|VTEXT"); 1994 if (vp->v_flag & VSYSTEM) 1995 strcat(buf, "|VSYSTEM"); 1996 if (vp->v_flag & VXLOCK) 1997 strcat(buf, "|VXLOCK"); 1998 if (vp->v_flag & VXWANT) 1999 strcat(buf, "|VXWANT"); 2000 if (vp->v_flag & VBWAIT) 2001 strcat(buf, "|VBWAIT"); 2002 if (vp->v_flag & VDOOMED) 2003 strcat(buf, "|VDOOMED"); 2004 if (vp->v_flag & VFREE) 2005 strcat(buf, "|VFREE"); 2006 if (vp->v_flag & VOBJBUF) 2007 strcat(buf, "|VOBJBUF"); 2008 if (buf[0] != '\0') 2009 printf(" flags (%s)", &buf[1]); 2010 if (vp->v_data == NULL) { 2011 printf("\n"); 2012 } else { 2013 printf("\n\t"); 2014 VOP_PRINT(vp); 2015 } 2016 } 2017 2018 #ifdef DDB 2019 #include <ddb/ddb.h> 2020 /* 2021 * List all of the locked vnodes in the system. 2022 * Called when debugging the kernel. 2023 */ 2024 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 2025 { 2026 struct proc *p = curproc; /* XXX */ 2027 struct mount *mp, *nmp; 2028 struct vnode *vp; 2029 2030 printf("Locked vnodes\n"); 2031 simple_lock(&mountlist_slock); 2032 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2033 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 2034 nmp = TAILQ_NEXT(mp, mnt_list); 2035 continue; 2036 } 2037 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2038 if (VOP_ISLOCKED(vp, NULL)) 2039 vprint((char *)0, vp); 2040 } 2041 simple_lock(&mountlist_slock); 2042 nmp = TAILQ_NEXT(mp, mnt_list); 2043 vfs_unbusy(mp, p); 2044 } 2045 simple_unlock(&mountlist_slock); 2046 } 2047 #endif 2048 2049 /* 2050 * Top level filesystem related information gathering. 2051 */ 2052 static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS)); 2053 2054 static int 2055 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2056 { 2057 int *name = (int *)arg1 - 1; /* XXX */ 2058 u_int namelen = arg2 + 1; /* XXX */ 2059 struct vfsconf *vfsp; 2060 2061 #if 1 || defined(COMPAT_PRELITE2) 2062 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2063 if (namelen == 1) 2064 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2065 #endif 2066 2067 #ifdef notyet 2068 /* all sysctl names at this level are at least name and field */ 2069 if (namelen < 2) 2070 return (ENOTDIR); /* overloaded */ 2071 if (name[0] != VFS_GENERIC) { 2072 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2073 if (vfsp->vfc_typenum == name[0]) 2074 break; 2075 if (vfsp == NULL) 2076 return (EOPNOTSUPP); 2077 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2078 oldp, oldlenp, newp, newlen, p)); 2079 } 2080 #endif 2081 switch (name[1]) { 2082 case VFS_MAXTYPENUM: 2083 if (namelen != 2) 2084 return (ENOTDIR); 2085 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2086 case VFS_CONF: 2087 if (namelen != 3) 2088 return (ENOTDIR); /* overloaded */ 2089 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2090 if (vfsp->vfc_typenum == name[2]) 2091 break; 2092 if (vfsp == NULL) 2093 return (EOPNOTSUPP); 2094 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2095 } 2096 return (EOPNOTSUPP); 2097 } 2098 2099 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2100 "Generic filesystem"); 2101 2102 #if 1 || defined(COMPAT_PRELITE2) 2103 2104 static int 2105 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2106 { 2107 int error; 2108 struct vfsconf *vfsp; 2109 struct ovfsconf ovfs; 2110 2111 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2112 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2113 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2114 ovfs.vfc_index = vfsp->vfc_typenum; 2115 ovfs.vfc_refcount = vfsp->vfc_refcount; 2116 ovfs.vfc_flags = vfsp->vfc_flags; 2117 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2118 if (error) 2119 return error; 2120 } 2121 return 0; 2122 } 2123 2124 #endif /* 1 || COMPAT_PRELITE2 */ 2125 2126 #if 0 2127 #define KINFO_VNODESLOP 10 2128 /* 2129 * Dump vnode list (via sysctl). 2130 * Copyout address of vnode followed by vnode. 2131 */ 2132 /* ARGSUSED */ 2133 static int 2134 sysctl_vnode(SYSCTL_HANDLER_ARGS) 2135 { 2136 struct proc *p = curproc; /* XXX */ 2137 struct mount *mp, *nmp; 2138 struct vnode *nvp, *vp; 2139 int error; 2140 2141 #define VPTRSZ sizeof (struct vnode *) 2142 #define VNODESZ sizeof (struct vnode) 2143 2144 req->lock = 0; 2145 if (!req->oldptr) /* Make an estimate */ 2146 return (SYSCTL_OUT(req, 0, 2147 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2148 2149 simple_lock(&mountlist_slock); 2150 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2151 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 2152 nmp = TAILQ_NEXT(mp, mnt_list); 2153 continue; 2154 } 2155 again: 2156 simple_lock(&mntvnode_slock); 2157 for (vp = LIST_FIRST(&mp->mnt_vnodelist); 2158 vp != NULL; 2159 vp = nvp) { 2160 /* 2161 * Check that the vp is still associated with 2162 * this filesystem. RACE: could have been 2163 * recycled onto the same filesystem. 2164 */ 2165 if (vp->v_mount != mp) { 2166 simple_unlock(&mntvnode_slock); 2167 goto again; 2168 } 2169 nvp = LIST_NEXT(vp, v_mntvnodes); 2170 simple_unlock(&mntvnode_slock); 2171 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2172 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2173 return (error); 2174 simple_lock(&mntvnode_slock); 2175 } 2176 simple_unlock(&mntvnode_slock); 2177 simple_lock(&mountlist_slock); 2178 nmp = TAILQ_NEXT(mp, mnt_list); 2179 vfs_unbusy(mp, p); 2180 } 2181 simple_unlock(&mountlist_slock); 2182 2183 return (0); 2184 } 2185 #endif 2186 2187 /* 2188 * XXX 2189 * Exporting the vnode list on large systems causes them to crash. 2190 * Exporting the vnode list on medium systems causes sysctl to coredump. 2191 */ 2192 #if 0 2193 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2194 0, 0, sysctl_vnode, "S,vnode", ""); 2195 #endif 2196 2197 /* 2198 * Check to see if a filesystem is mounted on a block device. 2199 */ 2200 int 2201 vfs_mountedon(vp) 2202 struct vnode *vp; 2203 { 2204 2205 if (vp->v_specmountpoint != NULL) 2206 return (EBUSY); 2207 return (0); 2208 } 2209 2210 /* 2211 * Unmount all filesystems. The list is traversed in reverse order 2212 * of mounting to avoid dependencies. 2213 */ 2214 void 2215 vfs_unmountall() 2216 { 2217 struct mount *mp; 2218 struct proc *p; 2219 int error; 2220 2221 if (curproc != NULL) 2222 p = curproc; 2223 else 2224 p = initproc; /* XXX XXX should this be proc0? */ 2225 /* 2226 * Since this only runs when rebooting, it is not interlocked. 2227 */ 2228 while(!TAILQ_EMPTY(&mountlist)) { 2229 mp = TAILQ_LAST(&mountlist, mntlist); 2230 error = dounmount(mp, MNT_FORCE, p); 2231 if (error) { 2232 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2233 printf("unmount of %s failed (", 2234 mp->mnt_stat.f_mntonname); 2235 if (error == EBUSY) 2236 printf("BUSY)\n"); 2237 else 2238 printf("%d)\n", error); 2239 } else { 2240 /* The unmount has removed mp from the mountlist */ 2241 } 2242 } 2243 } 2244 2245 /* 2246 * Build hash lists of net addresses and hang them off the mount point. 2247 * Called by ufs_mount() to set up the lists of export addresses. 2248 */ 2249 static int 2250 vfs_hang_addrlist(mp, nep, argp) 2251 struct mount *mp; 2252 struct netexport *nep; 2253 struct export_args *argp; 2254 { 2255 register struct netcred *np; 2256 register struct radix_node_head *rnh; 2257 register int i; 2258 struct radix_node *rn; 2259 struct sockaddr *saddr, *smask = 0; 2260 struct domain *dom; 2261 int error; 2262 2263 if (argp->ex_addrlen == 0) { 2264 if (mp->mnt_flag & MNT_DEFEXPORTED) 2265 return (EPERM); 2266 np = &nep->ne_defexported; 2267 np->netc_exflags = argp->ex_flags; 2268 np->netc_anon = argp->ex_anon; 2269 np->netc_anon.cr_ref = 1; 2270 mp->mnt_flag |= MNT_DEFEXPORTED; 2271 return (0); 2272 } 2273 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2274 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 2275 bzero((caddr_t) np, i); 2276 saddr = (struct sockaddr *) (np + 1); 2277 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2278 goto out; 2279 if (saddr->sa_len > argp->ex_addrlen) 2280 saddr->sa_len = argp->ex_addrlen; 2281 if (argp->ex_masklen) { 2282 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 2283 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 2284 if (error) 2285 goto out; 2286 if (smask->sa_len > argp->ex_masklen) 2287 smask->sa_len = argp->ex_masklen; 2288 } 2289 i = saddr->sa_family; 2290 if ((rnh = nep->ne_rtable[i]) == 0) { 2291 /* 2292 * Seems silly to initialize every AF when most are not used, 2293 * do so on demand here 2294 */ 2295 for (dom = domains; dom; dom = dom->dom_next) 2296 if (dom->dom_family == i && dom->dom_rtattach) { 2297 dom->dom_rtattach((void **) &nep->ne_rtable[i], 2298 dom->dom_rtoffset); 2299 break; 2300 } 2301 if ((rnh = nep->ne_rtable[i]) == 0) { 2302 error = ENOBUFS; 2303 goto out; 2304 } 2305 } 2306 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 2307 np->netc_rnodes); 2308 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 2309 error = EPERM; 2310 goto out; 2311 } 2312 np->netc_exflags = argp->ex_flags; 2313 np->netc_anon = argp->ex_anon; 2314 np->netc_anon.cr_ref = 1; 2315 return (0); 2316 out: 2317 free(np, M_NETADDR); 2318 return (error); 2319 } 2320 2321 /* ARGSUSED */ 2322 static int 2323 vfs_free_netcred(rn, w) 2324 struct radix_node *rn; 2325 void *w; 2326 { 2327 register struct radix_node_head *rnh = (struct radix_node_head *) w; 2328 2329 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2330 free((caddr_t) rn, M_NETADDR); 2331 return (0); 2332 } 2333 2334 /* 2335 * Free the net address hash lists that are hanging off the mount points. 2336 */ 2337 static void 2338 vfs_free_addrlist(nep) 2339 struct netexport *nep; 2340 { 2341 register int i; 2342 register struct radix_node_head *rnh; 2343 2344 for (i = 0; i <= AF_MAX; i++) 2345 if ((rnh = nep->ne_rtable[i])) { 2346 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2347 (caddr_t) rnh); 2348 free((caddr_t) rnh, M_RTABLE); 2349 nep->ne_rtable[i] = 0; 2350 } 2351 } 2352 2353 int 2354 vfs_export(mp, nep, argp) 2355 struct mount *mp; 2356 struct netexport *nep; 2357 struct export_args *argp; 2358 { 2359 int error; 2360 2361 if (argp->ex_flags & MNT_DELEXPORT) { 2362 if (mp->mnt_flag & MNT_EXPUBLIC) { 2363 vfs_setpublicfs(NULL, NULL, NULL); 2364 mp->mnt_flag &= ~MNT_EXPUBLIC; 2365 } 2366 vfs_free_addrlist(nep); 2367 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2368 } 2369 if (argp->ex_flags & MNT_EXPORTED) { 2370 if (argp->ex_flags & MNT_EXPUBLIC) { 2371 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2372 return (error); 2373 mp->mnt_flag |= MNT_EXPUBLIC; 2374 } 2375 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2376 return (error); 2377 mp->mnt_flag |= MNT_EXPORTED; 2378 } 2379 return (0); 2380 } 2381 2382 2383 /* 2384 * Set the publicly exported filesystem (WebNFS). Currently, only 2385 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2386 */ 2387 int 2388 vfs_setpublicfs(mp, nep, argp) 2389 struct mount *mp; 2390 struct netexport *nep; 2391 struct export_args *argp; 2392 { 2393 int error; 2394 struct vnode *rvp; 2395 char *cp; 2396 2397 /* 2398 * mp == NULL -> invalidate the current info, the FS is 2399 * no longer exported. May be called from either vfs_export 2400 * or unmount, so check if it hasn't already been done. 2401 */ 2402 if (mp == NULL) { 2403 if (nfs_pub.np_valid) { 2404 nfs_pub.np_valid = 0; 2405 if (nfs_pub.np_index != NULL) { 2406 FREE(nfs_pub.np_index, M_TEMP); 2407 nfs_pub.np_index = NULL; 2408 } 2409 } 2410 return (0); 2411 } 2412 2413 /* 2414 * Only one allowed at a time. 2415 */ 2416 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2417 return (EBUSY); 2418 2419 /* 2420 * Get real filehandle for root of exported FS. 2421 */ 2422 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2423 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2424 2425 if ((error = VFS_ROOT(mp, &rvp))) 2426 return (error); 2427 2428 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2429 return (error); 2430 2431 vput(rvp); 2432 2433 /* 2434 * If an indexfile was specified, pull it in. 2435 */ 2436 if (argp->ex_indexfile != NULL) { 2437 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2438 M_WAITOK); 2439 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2440 MAXNAMLEN, (size_t *)0); 2441 if (!error) { 2442 /* 2443 * Check for illegal filenames. 2444 */ 2445 for (cp = nfs_pub.np_index; *cp; cp++) { 2446 if (*cp == '/') { 2447 error = EINVAL; 2448 break; 2449 } 2450 } 2451 } 2452 if (error) { 2453 FREE(nfs_pub.np_index, M_TEMP); 2454 return (error); 2455 } 2456 } 2457 2458 nfs_pub.np_mount = mp; 2459 nfs_pub.np_valid = 1; 2460 return (0); 2461 } 2462 2463 struct netcred * 2464 vfs_export_lookup(mp, nep, nam) 2465 register struct mount *mp; 2466 struct netexport *nep; 2467 struct sockaddr *nam; 2468 { 2469 register struct netcred *np; 2470 register struct radix_node_head *rnh; 2471 struct sockaddr *saddr; 2472 2473 np = NULL; 2474 if (mp->mnt_flag & MNT_EXPORTED) { 2475 /* 2476 * Lookup in the export list first. 2477 */ 2478 if (nam != NULL) { 2479 saddr = nam; 2480 rnh = nep->ne_rtable[saddr->sa_family]; 2481 if (rnh != NULL) { 2482 np = (struct netcred *) 2483 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2484 rnh); 2485 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2486 np = NULL; 2487 } 2488 } 2489 /* 2490 * If no address match, use the default if it exists. 2491 */ 2492 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2493 np = &nep->ne_defexported; 2494 } 2495 return (np); 2496 } 2497 2498 /* 2499 * perform msync on all vnodes under a mount point 2500 * the mount point must be locked. 2501 */ 2502 void 2503 vfs_msync(struct mount *mp, int flags) { 2504 struct vnode *vp, *nvp; 2505 struct vm_object *obj; 2506 int anyio, tries; 2507 2508 tries = 5; 2509 loop: 2510 anyio = 0; 2511 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) { 2512 2513 nvp = LIST_NEXT(vp, v_mntvnodes); 2514 2515 if (vp->v_mount != mp) { 2516 goto loop; 2517 } 2518 2519 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2520 continue; 2521 2522 if (flags != MNT_WAIT) { 2523 obj = vp->v_object; 2524 if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0) 2525 continue; 2526 if (VOP_ISLOCKED(vp, NULL)) 2527 continue; 2528 } 2529 2530 simple_lock(&vp->v_interlock); 2531 if (vp->v_object && 2532 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2533 if (!vget(vp, 2534 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) { 2535 if (vp->v_object) { 2536 vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2537 anyio = 1; 2538 } 2539 vput(vp); 2540 } 2541 } else { 2542 simple_unlock(&vp->v_interlock); 2543 } 2544 } 2545 if (anyio && (--tries > 0)) 2546 goto loop; 2547 } 2548 2549 /* 2550 * Create the VM object needed for VMIO and mmap support. This 2551 * is done for all VREG files in the system. Some filesystems might 2552 * afford the additional metadata buffering capability of the 2553 * VMIO code by making the device node be VMIO mode also. 2554 * 2555 * vp must be locked when vfs_object_create is called. 2556 */ 2557 int 2558 vfs_object_create(vp, p, cred) 2559 struct vnode *vp; 2560 struct proc *p; 2561 struct ucred *cred; 2562 { 2563 struct vattr vat; 2564 vm_object_t object; 2565 int error = 0; 2566 2567 if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE) 2568 return 0; 2569 2570 retry: 2571 if ((object = vp->v_object) == NULL) { 2572 if (vp->v_type == VREG || vp->v_type == VDIR) { 2573 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2574 goto retn; 2575 object = vnode_pager_alloc(vp, vat.va_size, 0, 0); 2576 } else if (devsw(vp->v_rdev) != NULL) { 2577 /* 2578 * This simply allocates the biggest object possible 2579 * for a disk vnode. This should be fixed, but doesn't 2580 * cause any problems (yet). 2581 */ 2582 object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0); 2583 } else { 2584 goto retn; 2585 } 2586 /* 2587 * Dereference the reference we just created. This assumes 2588 * that the object is associated with the vp. 2589 */ 2590 object->ref_count--; 2591 vp->v_usecount--; 2592 } else { 2593 if (object->flags & OBJ_DEAD) { 2594 VOP_UNLOCK(vp, 0, p); 2595 tsleep(object, PVM, "vodead", 0); 2596 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2597 goto retry; 2598 } 2599 } 2600 2601 KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object")); 2602 vp->v_flag |= VOBJBUF; 2603 2604 retn: 2605 return error; 2606 } 2607 2608 void 2609 vfree(vp) 2610 struct vnode *vp; 2611 { 2612 int s; 2613 2614 s = splbio(); 2615 simple_lock(&vnode_free_list_slock); 2616 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free")); 2617 if (vp->v_flag & VAGE) { 2618 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2619 } else { 2620 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2621 } 2622 freevnodes++; 2623 simple_unlock(&vnode_free_list_slock); 2624 vp->v_flag &= ~VAGE; 2625 vp->v_flag |= VFREE; 2626 splx(s); 2627 } 2628 2629 void 2630 vbusy(vp) 2631 struct vnode *vp; 2632 { 2633 int s; 2634 2635 s = splbio(); 2636 simple_lock(&vnode_free_list_slock); 2637 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free")); 2638 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2639 freevnodes--; 2640 simple_unlock(&vnode_free_list_slock); 2641 vp->v_flag &= ~(VFREE|VAGE); 2642 splx(s); 2643 } 2644 2645 /* 2646 * Record a process's interest in events which might happen to 2647 * a vnode. Because poll uses the historic select-style interface 2648 * internally, this routine serves as both the ``check for any 2649 * pending events'' and the ``record my interest in future events'' 2650 * functions. (These are done together, while the lock is held, 2651 * to avoid race conditions.) 2652 */ 2653 int 2654 vn_pollrecord(vp, p, events) 2655 struct vnode *vp; 2656 struct proc *p; 2657 short events; 2658 { 2659 simple_lock(&vp->v_pollinfo.vpi_lock); 2660 if (vp->v_pollinfo.vpi_revents & events) { 2661 /* 2662 * This leaves events we are not interested 2663 * in available for the other process which 2664 * which presumably had requested them 2665 * (otherwise they would never have been 2666 * recorded). 2667 */ 2668 events &= vp->v_pollinfo.vpi_revents; 2669 vp->v_pollinfo.vpi_revents &= ~events; 2670 2671 simple_unlock(&vp->v_pollinfo.vpi_lock); 2672 return events; 2673 } 2674 vp->v_pollinfo.vpi_events |= events; 2675 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2676 simple_unlock(&vp->v_pollinfo.vpi_lock); 2677 return 0; 2678 } 2679 2680 /* 2681 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2682 * it is possible for us to miss an event due to race conditions, but 2683 * that condition is expected to be rare, so for the moment it is the 2684 * preferred interface. 2685 */ 2686 void 2687 vn_pollevent(vp, events) 2688 struct vnode *vp; 2689 short events; 2690 { 2691 simple_lock(&vp->v_pollinfo.vpi_lock); 2692 if (vp->v_pollinfo.vpi_events & events) { 2693 /* 2694 * We clear vpi_events so that we don't 2695 * call selwakeup() twice if two events are 2696 * posted before the polling process(es) is 2697 * awakened. This also ensures that we take at 2698 * most one selwakeup() if the polling process 2699 * is no longer interested. However, it does 2700 * mean that only one event can be noticed at 2701 * a time. (Perhaps we should only clear those 2702 * event bits which we note?) XXX 2703 */ 2704 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2705 vp->v_pollinfo.vpi_revents |= events; 2706 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2707 } 2708 simple_unlock(&vp->v_pollinfo.vpi_lock); 2709 } 2710 2711 /* 2712 * Wake up anyone polling on vp because it is being revoked. 2713 * This depends on dead_poll() returning POLLHUP for correct 2714 * behavior. 2715 */ 2716 void 2717 vn_pollgone(vp) 2718 struct vnode *vp; 2719 { 2720 simple_lock(&vp->v_pollinfo.vpi_lock); 2721 if (vp->v_pollinfo.vpi_events) { 2722 vp->v_pollinfo.vpi_events = 0; 2723 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2724 } 2725 simple_unlock(&vp->v_pollinfo.vpi_lock); 2726 } 2727 2728 2729 2730 /* 2731 * Routine to create and manage a filesystem syncer vnode. 2732 */ 2733 #define sync_close ((int (*) __P((struct vop_close_args *)))nullop) 2734 static int sync_fsync __P((struct vop_fsync_args *)); 2735 static int sync_inactive __P((struct vop_inactive_args *)); 2736 static int sync_reclaim __P((struct vop_reclaim_args *)); 2737 #define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock) 2738 #define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock) 2739 static int sync_print __P((struct vop_print_args *)); 2740 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked) 2741 2742 static vop_t **sync_vnodeop_p; 2743 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2744 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2745 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2746 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2747 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2748 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2749 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2750 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2751 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2752 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2753 { NULL, NULL } 2754 }; 2755 static struct vnodeopv_desc sync_vnodeop_opv_desc = 2756 { &sync_vnodeop_p, sync_vnodeop_entries }; 2757 2758 VNODEOP_SET(sync_vnodeop_opv_desc); 2759 2760 /* 2761 * Create a new filesystem syncer vnode for the specified mount point. 2762 */ 2763 int 2764 vfs_allocate_syncvnode(mp) 2765 struct mount *mp; 2766 { 2767 struct vnode *vp; 2768 static long start, incr, next; 2769 int error; 2770 2771 /* Allocate a new vnode */ 2772 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2773 mp->mnt_syncer = NULL; 2774 return (error); 2775 } 2776 vp->v_type = VNON; 2777 /* 2778 * Place the vnode onto the syncer worklist. We attempt to 2779 * scatter them about on the list so that they will go off 2780 * at evenly distributed times even if all the filesystems 2781 * are mounted at once. 2782 */ 2783 next += incr; 2784 if (next == 0 || next > syncer_maxdelay) { 2785 start /= 2; 2786 incr /= 2; 2787 if (start == 0) { 2788 start = syncer_maxdelay / 2; 2789 incr = syncer_maxdelay; 2790 } 2791 next = start; 2792 } 2793 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2794 mp->mnt_syncer = vp; 2795 return (0); 2796 } 2797 2798 /* 2799 * Do a lazy sync of the filesystem. 2800 */ 2801 static int 2802 sync_fsync(ap) 2803 struct vop_fsync_args /* { 2804 struct vnode *a_vp; 2805 struct ucred *a_cred; 2806 int a_waitfor; 2807 struct proc *a_p; 2808 } */ *ap; 2809 { 2810 struct vnode *syncvp = ap->a_vp; 2811 struct mount *mp = syncvp->v_mount; 2812 struct proc *p = ap->a_p; 2813 int asyncflag; 2814 2815 /* 2816 * We only need to do something if this is a lazy evaluation. 2817 */ 2818 if (ap->a_waitfor != MNT_LAZY) 2819 return (0); 2820 2821 /* 2822 * Move ourselves to the back of the sync list. 2823 */ 2824 vn_syncer_add_to_worklist(syncvp, syncdelay); 2825 2826 /* 2827 * Walk the list of vnodes pushing all that are dirty and 2828 * not already on the sync list. 2829 */ 2830 simple_lock(&mountlist_slock); 2831 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) { 2832 simple_unlock(&mountlist_slock); 2833 return (0); 2834 } 2835 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 2836 vfs_unbusy(mp, p); 2837 simple_unlock(&mountlist_slock); 2838 return (0); 2839 } 2840 asyncflag = mp->mnt_flag & MNT_ASYNC; 2841 mp->mnt_flag &= ~MNT_ASYNC; 2842 vfs_msync(mp, MNT_NOWAIT); 2843 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p); 2844 if (asyncflag) 2845 mp->mnt_flag |= MNT_ASYNC; 2846 vn_finished_write(mp); 2847 vfs_unbusy(mp, p); 2848 return (0); 2849 } 2850 2851 /* 2852 * The syncer vnode is no referenced. 2853 */ 2854 static int 2855 sync_inactive(ap) 2856 struct vop_inactive_args /* { 2857 struct vnode *a_vp; 2858 struct proc *a_p; 2859 } */ *ap; 2860 { 2861 2862 vgone(ap->a_vp); 2863 return (0); 2864 } 2865 2866 /* 2867 * The syncer vnode is no longer needed and is being decommissioned. 2868 * 2869 * Modifications to the worklist must be protected at splbio(). 2870 */ 2871 static int 2872 sync_reclaim(ap) 2873 struct vop_reclaim_args /* { 2874 struct vnode *a_vp; 2875 } */ *ap; 2876 { 2877 struct vnode *vp = ap->a_vp; 2878 int s; 2879 2880 s = splbio(); 2881 vp->v_mount->mnt_syncer = NULL; 2882 if (vp->v_flag & VONWORKLST) { 2883 LIST_REMOVE(vp, v_synclist); 2884 vp->v_flag &= ~VONWORKLST; 2885 } 2886 splx(s); 2887 2888 return (0); 2889 } 2890 2891 /* 2892 * Print out a syncer vnode. 2893 */ 2894 static int 2895 sync_print(ap) 2896 struct vop_print_args /* { 2897 struct vnode *a_vp; 2898 } */ *ap; 2899 { 2900 struct vnode *vp = ap->a_vp; 2901 2902 printf("syncer vnode"); 2903 if (vp->v_vnlock != NULL) 2904 lockmgr_printinfo(vp->v_vnlock); 2905 printf("\n"); 2906 return (0); 2907 } 2908 2909 /* 2910 * extract the dev_t from a VBLK or VCHR 2911 */ 2912 dev_t 2913 vn_todev(vp) 2914 struct vnode *vp; 2915 { 2916 if (vp->v_type != VBLK && vp->v_type != VCHR) 2917 return (NODEV); 2918 return (vp->v_rdev); 2919 } 2920 2921 /* 2922 * Check if vnode represents a disk device 2923 */ 2924 int 2925 vn_isdisk(vp, errp) 2926 struct vnode *vp; 2927 int *errp; 2928 { 2929 if (vp->v_type != VBLK && vp->v_type != VCHR) { 2930 if (errp != NULL) 2931 *errp = ENOTBLK; 2932 return (0); 2933 } 2934 if (vp->v_rdev == NULL) { 2935 if (errp != NULL) 2936 *errp = ENXIO; 2937 return (0); 2938 } 2939 if (!devsw(vp->v_rdev)) { 2940 if (errp != NULL) 2941 *errp = ENXIO; 2942 return (0); 2943 } 2944 if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) { 2945 if (errp != NULL) 2946 *errp = ENOTBLK; 2947 return (0); 2948 } 2949 if (errp != NULL) 2950 *errp = 0; 2951 return (1); 2952 } 2953 2954 void 2955 NDFREE(ndp, flags) 2956 struct nameidata *ndp; 2957 const uint flags; 2958 { 2959 if (!(flags & NDF_NO_FREE_PNBUF) && 2960 (ndp->ni_cnd.cn_flags & HASBUF)) { 2961 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 2962 ndp->ni_cnd.cn_flags &= ~HASBUF; 2963 } 2964 if (!(flags & NDF_NO_DVP_UNLOCK) && 2965 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 2966 ndp->ni_dvp != ndp->ni_vp) 2967 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc); 2968 if (!(flags & NDF_NO_DVP_RELE) && 2969 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 2970 vrele(ndp->ni_dvp); 2971 ndp->ni_dvp = NULL; 2972 } 2973 if (!(flags & NDF_NO_VP_UNLOCK) && 2974 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 2975 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc); 2976 if (!(flags & NDF_NO_VP_RELE) && 2977 ndp->ni_vp) { 2978 vrele(ndp->ni_vp); 2979 ndp->ni_vp = NULL; 2980 } 2981 if (!(flags & NDF_NO_STARTDIR_RELE) && 2982 (ndp->ni_cnd.cn_flags & SAVESTART)) { 2983 vrele(ndp->ni_startdir); 2984 ndp->ni_startdir = NULL; 2985 } 2986 } 2987