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 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/domain.h> 54 #include <sys/eventhandler.h> 55 #include <sys/fcntl.h> 56 #include <sys/kernel.h> 57 #include <sys/kthread.h> 58 #include <sys/malloc.h> 59 #include <sys/mount.h> 60 #include <sys/namei.h> 61 #include <sys/proc.h> 62 #include <sys/reboot.h> 63 #include <sys/socket.h> 64 #include <sys/stat.h> 65 #include <sys/sysctl.h> 66 #include <sys/vmmeter.h> 67 #include <sys/vnode.h> 68 69 #include <machine/limits.h> 70 71 #include <vm/vm.h> 72 #include <vm/vm_object.h> 73 #include <vm/vm_extern.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_map.h> 76 #include <vm/vm_page.h> 77 #include <vm/vm_pager.h> 78 #include <vm/vnode_pager.h> 79 #include <vm/vm_zone.h> 80 81 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 82 83 static void insmntque __P((struct vnode *vp, struct mount *mp)); 84 static void vclean __P((struct vnode *vp, int flags, struct proc *p)); 85 static void vfree __P((struct vnode *)); 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 struct tobefreelist vnode_tobefree_list; /* vnode free list */ 100 101 static u_long wantfreevnodes = 25; 102 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 103 static u_long freevnodes = 0; 104 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 105 106 static int reassignbufcalls; 107 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 108 static int reassignbufloops; 109 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, ""); 110 static int reassignbufsortgood; 111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, ""); 112 static int reassignbufsortbad; 113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, ""); 114 static int reassignbufmethod = 1; 115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, ""); 116 117 #ifdef ENABLE_VFS_IOOPT 118 int vfs_ioopt = 0; 119 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 120 #endif 121 122 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */ 123 struct simplelock mountlist_slock; 124 struct simplelock mntvnode_slock; 125 int nfs_mount_type = -1; 126 #ifndef NULL_SIMPLELOCKS 127 static struct simplelock mntid_slock; 128 static struct simplelock vnode_free_list_slock; 129 static struct simplelock spechash_slock; 130 #endif 131 struct nfs_public nfs_pub; /* publicly exported FS */ 132 static vm_zone_t vnode_zone; 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 TAILQ_INIT(&vnode_tobefree_list); 177 simple_lock_init(&vnode_free_list_slock); 178 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 179 /* 180 * Initialize the filesystem syncer. 181 */ 182 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 183 &syncer_mask); 184 syncer_maxdelay = syncer_mask + 1; 185 } 186 187 /* 188 * Mark a mount point as busy. Used to synchronize access and to delay 189 * unmounting. Interlock is not released on failure. 190 */ 191 int 192 vfs_busy(mp, flags, interlkp, p) 193 struct mount *mp; 194 int flags; 195 struct simplelock *interlkp; 196 struct proc *p; 197 { 198 int lkflags; 199 200 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 201 if (flags & LK_NOWAIT) 202 return (ENOENT); 203 mp->mnt_kern_flag |= MNTK_MWAIT; 204 if (interlkp) { 205 simple_unlock(interlkp); 206 } 207 /* 208 * Since all busy locks are shared except the exclusive 209 * lock granted when unmounting, the only place that a 210 * wakeup needs to be done is at the release of the 211 * exclusive lock at the end of dounmount. 212 */ 213 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 214 if (interlkp) { 215 simple_lock(interlkp); 216 } 217 return (ENOENT); 218 } 219 lkflags = LK_SHARED | LK_NOPAUSE; 220 if (interlkp) 221 lkflags |= LK_INTERLOCK; 222 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 223 panic("vfs_busy: unexpected lock failure"); 224 return (0); 225 } 226 227 /* 228 * Free a busy filesystem. 229 */ 230 void 231 vfs_unbusy(mp, p) 232 struct mount *mp; 233 struct proc *p; 234 { 235 236 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 237 } 238 239 /* 240 * Lookup a filesystem type, and if found allocate and initialize 241 * a mount structure for it. 242 * 243 * Devname is usually updated by mount(8) after booting. 244 */ 245 int 246 vfs_rootmountalloc(fstypename, devname, mpp) 247 char *fstypename; 248 char *devname; 249 struct mount **mpp; 250 { 251 struct proc *p = curproc; /* XXX */ 252 struct vfsconf *vfsp; 253 struct mount *mp; 254 255 if (fstypename == NULL) 256 return (ENODEV); 257 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 258 if (!strcmp(vfsp->vfc_name, fstypename)) 259 break; 260 if (vfsp == NULL) 261 return (ENODEV); 262 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 263 bzero((char *)mp, (u_long)sizeof(struct mount)); 264 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); 265 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 266 LIST_INIT(&mp->mnt_vnodelist); 267 mp->mnt_vfc = vfsp; 268 mp->mnt_op = vfsp->vfc_vfsops; 269 mp->mnt_flag = MNT_RDONLY; 270 mp->mnt_vnodecovered = NULLVP; 271 vfsp->vfc_refcount++; 272 mp->mnt_iosize_max = DFLTPHYS; 273 mp->mnt_stat.f_type = vfsp->vfc_typenum; 274 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 275 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 276 mp->mnt_stat.f_mntonname[0] = '/'; 277 mp->mnt_stat.f_mntonname[1] = 0; 278 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 279 *mpp = mp; 280 return (0); 281 } 282 283 /* 284 * Find an appropriate filesystem to use for the root. If a filesystem 285 * has not been preselected, walk through the list of known filesystems 286 * trying those that have mountroot routines, and try them until one 287 * works or we have tried them all. 288 */ 289 #ifdef notdef /* XXX JH */ 290 int 291 lite2_vfs_mountroot() 292 { 293 struct vfsconf *vfsp; 294 extern int (*lite2_mountroot) __P((void)); 295 int error; 296 297 if (lite2_mountroot != NULL) 298 return ((*lite2_mountroot)()); 299 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 300 if (vfsp->vfc_mountroot == NULL) 301 continue; 302 if ((error = (*vfsp->vfc_mountroot)()) == 0) 303 return (0); 304 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 305 } 306 return (ENODEV); 307 } 308 #endif 309 310 /* 311 * Lookup a mount point by filesystem identifier. 312 */ 313 struct mount * 314 vfs_getvfs(fsid) 315 fsid_t *fsid; 316 { 317 register struct mount *mp; 318 319 simple_lock(&mountlist_slock); 320 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 321 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 322 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 323 simple_unlock(&mountlist_slock); 324 return (mp); 325 } 326 } 327 simple_unlock(&mountlist_slock); 328 return ((struct mount *) 0); 329 } 330 331 /* 332 * Get a new unique fsid. Try to make its val[0] unique, since this value 333 * will be used to create fake device numbers for stat(). Also try (but 334 * not so hard) make its val[0] unique mod 2^16, since some emulators only 335 * support 16-bit device numbers. We end up with unique val[0]'s for the 336 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 337 * 338 * Keep in mind that several mounts may be running in parallel. Starting 339 * the search one past where the previous search terminated is both a 340 * micro-optimization and a defense against returning the same fsid to 341 * different mounts. 342 */ 343 void 344 vfs_getnewfsid(mp) 345 struct mount *mp; 346 { 347 static u_int16_t mntid_base; 348 fsid_t tfsid; 349 int mtype; 350 351 simple_lock(&mntid_slock); 352 mtype = mp->mnt_vfc->vfc_typenum; 353 tfsid.val[1] = mtype; 354 mtype = (mtype & 0xFF) << 16; 355 for (;;) { 356 tfsid.val[0] = makeudev(255, mtype | mntid_base++); 357 if (vfs_getvfs(&tfsid) == NULL) 358 break; 359 } 360 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 361 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 362 simple_unlock(&mntid_slock); 363 } 364 365 /* 366 * Knob to control the precision of file timestamps: 367 * 368 * 0 = seconds only; nanoseconds zeroed. 369 * 1 = seconds and nanoseconds, accurate within 1/HZ. 370 * 2 = seconds and nanoseconds, truncated to microseconds. 371 * >=3 = seconds and nanoseconds, maximum precision. 372 */ 373 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 374 375 static int timestamp_precision = TSP_SEC; 376 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 377 ×tamp_precision, 0, ""); 378 379 /* 380 * Get a current timestamp. 381 */ 382 void 383 vfs_timestamp(tsp) 384 struct timespec *tsp; 385 { 386 struct timeval tv; 387 388 switch (timestamp_precision) { 389 case TSP_SEC: 390 tsp->tv_sec = time_second; 391 tsp->tv_nsec = 0; 392 break; 393 case TSP_HZ: 394 getnanotime(tsp); 395 break; 396 case TSP_USEC: 397 microtime(&tv); 398 TIMEVAL_TO_TIMESPEC(&tv, tsp); 399 break; 400 case TSP_NSEC: 401 default: 402 nanotime(tsp); 403 break; 404 } 405 } 406 407 /* 408 * Set vnode attributes to VNOVAL 409 */ 410 void 411 vattr_null(vap) 412 register struct vattr *vap; 413 { 414 415 vap->va_type = VNON; 416 vap->va_size = VNOVAL; 417 vap->va_bytes = VNOVAL; 418 vap->va_mode = VNOVAL; 419 vap->va_nlink = VNOVAL; 420 vap->va_uid = VNOVAL; 421 vap->va_gid = VNOVAL; 422 vap->va_fsid = VNOVAL; 423 vap->va_fileid = VNOVAL; 424 vap->va_blocksize = VNOVAL; 425 vap->va_rdev = VNOVAL; 426 vap->va_atime.tv_sec = VNOVAL; 427 vap->va_atime.tv_nsec = VNOVAL; 428 vap->va_mtime.tv_sec = VNOVAL; 429 vap->va_mtime.tv_nsec = VNOVAL; 430 vap->va_ctime.tv_sec = VNOVAL; 431 vap->va_ctime.tv_nsec = VNOVAL; 432 vap->va_flags = VNOVAL; 433 vap->va_gen = VNOVAL; 434 vap->va_vaflags = 0; 435 } 436 437 /* 438 * Routines having to do with the management of the vnode table. 439 */ 440 extern vop_t **dead_vnodeop_p; 441 442 /* 443 * Return the next vnode from the free list. 444 */ 445 int 446 getnewvnode(tag, mp, vops, vpp) 447 enum vtagtype tag; 448 struct mount *mp; 449 vop_t **vops; 450 struct vnode **vpp; 451 { 452 int s; 453 struct proc *p = curproc; /* XXX */ 454 struct vnode *vp, *tvp, *nvp; 455 vm_object_t object; 456 TAILQ_HEAD(freelst, vnode) vnode_tmp_list; 457 458 /* 459 * We take the least recently used vnode from the freelist 460 * if we can get it and it has no cached pages, and no 461 * namecache entries are relative to it. 462 * Otherwise we allocate a new vnode 463 */ 464 465 s = splbio(); 466 simple_lock(&vnode_free_list_slock); 467 TAILQ_INIT(&vnode_tmp_list); 468 469 for (vp = TAILQ_FIRST(&vnode_tobefree_list); vp; vp = nvp) { 470 nvp = TAILQ_NEXT(vp, v_freelist); 471 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 472 if (vp->v_flag & VAGE) { 473 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 474 } else { 475 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 476 } 477 vp->v_flag &= ~(VTBFREE|VAGE); 478 vp->v_flag |= VFREE; 479 if (vp->v_usecount) 480 panic("tobe free vnode isn't"); 481 freevnodes++; 482 } 483 484 if (wantfreevnodes && freevnodes < wantfreevnodes) { 485 vp = NULL; 486 } else if (!wantfreevnodes && freevnodes <= desiredvnodes) { 487 /* 488 * XXX: this is only here to be backwards compatible 489 */ 490 vp = NULL; 491 } else { 492 for (vp = TAILQ_FIRST(&vnode_free_list); vp; vp = nvp) { 493 nvp = TAILQ_NEXT(vp, v_freelist); 494 if (!simple_lock_try(&vp->v_interlock)) 495 continue; 496 if (vp->v_usecount) 497 panic("free vnode isn't"); 498 499 object = vp->v_object; 500 if (object && (object->resident_page_count || object->ref_count)) { 501 printf("object inconsistant state: RPC: %d, RC: %d\n", 502 object->resident_page_count, object->ref_count); 503 /* Don't recycle if it's caching some pages */ 504 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 505 TAILQ_INSERT_TAIL(&vnode_tmp_list, vp, v_freelist); 506 continue; 507 } else if (LIST_FIRST(&vp->v_cache_src)) { 508 /* Don't recycle if active in the namecache */ 509 simple_unlock(&vp->v_interlock); 510 continue; 511 } else { 512 break; 513 } 514 } 515 } 516 517 for (tvp = TAILQ_FIRST(&vnode_tmp_list); tvp; tvp = nvp) { 518 nvp = TAILQ_NEXT(tvp, v_freelist); 519 TAILQ_REMOVE(&vnode_tmp_list, tvp, v_freelist); 520 TAILQ_INSERT_TAIL(&vnode_free_list, tvp, v_freelist); 521 simple_unlock(&tvp->v_interlock); 522 } 523 524 if (vp) { 525 vp->v_flag |= VDOOMED; 526 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 527 freevnodes--; 528 simple_unlock(&vnode_free_list_slock); 529 cache_purge(vp); 530 vp->v_lease = NULL; 531 if (vp->v_type != VBAD) { 532 vgonel(vp, p); 533 } else { 534 simple_unlock(&vp->v_interlock); 535 } 536 537 #ifdef INVARIANTS 538 { 539 int s; 540 541 if (vp->v_data) 542 panic("cleaned vnode isn't"); 543 s = splbio(); 544 if (vp->v_numoutput) 545 panic("Clean vnode has pending I/O's"); 546 splx(s); 547 } 548 #endif 549 vp->v_flag = 0; 550 vp->v_lastw = 0; 551 vp->v_lasta = 0; 552 vp->v_cstart = 0; 553 vp->v_clen = 0; 554 vp->v_socket = 0; 555 vp->v_writecount = 0; /* XXX */ 556 } else { 557 simple_unlock(&vnode_free_list_slock); 558 vp = (struct vnode *) zalloc(vnode_zone); 559 bzero((char *) vp, sizeof *vp); 560 simple_lock_init(&vp->v_interlock); 561 vp->v_dd = vp; 562 cache_purge(vp); 563 LIST_INIT(&vp->v_cache_src); 564 TAILQ_INIT(&vp->v_cache_dst); 565 numvnodes++; 566 } 567 568 TAILQ_INIT(&vp->v_cleanblkhd); 569 TAILQ_INIT(&vp->v_dirtyblkhd); 570 vp->v_type = VNON; 571 vp->v_tag = tag; 572 vp->v_op = vops; 573 insmntque(vp, mp); 574 *vpp = vp; 575 vp->v_usecount = 1; 576 vp->v_data = 0; 577 splx(s); 578 579 vfs_object_create(vp, p, p->p_ucred); 580 return (0); 581 } 582 583 /* 584 * Move a vnode from one mount queue to another. 585 */ 586 static void 587 insmntque(vp, mp) 588 register struct vnode *vp; 589 register struct mount *mp; 590 { 591 592 simple_lock(&mntvnode_slock); 593 /* 594 * Delete from old mount point vnode list, if on one. 595 */ 596 if (vp->v_mount != NULL) 597 LIST_REMOVE(vp, v_mntvnodes); 598 /* 599 * Insert into list of vnodes for the new mount point, if available. 600 */ 601 if ((vp->v_mount = mp) == NULL) { 602 simple_unlock(&mntvnode_slock); 603 return; 604 } 605 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 606 simple_unlock(&mntvnode_slock); 607 } 608 609 /* 610 * Update outstanding I/O count and do wakeup if requested. 611 */ 612 void 613 vwakeup(bp) 614 register struct buf *bp; 615 { 616 register struct vnode *vp; 617 618 bp->b_flags &= ~B_WRITEINPROG; 619 if ((vp = bp->b_vp)) { 620 vp->v_numoutput--; 621 if (vp->v_numoutput < 0) 622 panic("vwakeup: neg numoutput"); 623 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 624 vp->v_flag &= ~VBWAIT; 625 wakeup((caddr_t) &vp->v_numoutput); 626 } 627 } 628 } 629 630 /* 631 * Flush out and invalidate all buffers associated with a vnode. 632 * Called with the underlying object locked. 633 */ 634 int 635 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 636 register struct vnode *vp; 637 int flags; 638 struct ucred *cred; 639 struct proc *p; 640 int slpflag, slptimeo; 641 { 642 register struct buf *bp; 643 struct buf *nbp, *blist; 644 int s, error; 645 vm_object_t object; 646 647 if (flags & V_SAVE) { 648 s = splbio(); 649 while (vp->v_numoutput) { 650 vp->v_flag |= VBWAIT; 651 error = tsleep((caddr_t)&vp->v_numoutput, 652 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo); 653 if (error) { 654 splx(s); 655 return (error); 656 } 657 } 658 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 659 splx(s); 660 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0) 661 return (error); 662 s = splbio(); 663 if (vp->v_numoutput > 0 || 664 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 665 panic("vinvalbuf: dirty bufs"); 666 } 667 splx(s); 668 } 669 s = splbio(); 670 for (;;) { 671 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 672 if (!blist) 673 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 674 if (!blist) 675 break; 676 677 for (bp = blist; bp; bp = nbp) { 678 nbp = TAILQ_NEXT(bp, b_vnbufs); 679 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 680 error = BUF_TIMELOCK(bp, 681 LK_EXCLUSIVE | LK_SLEEPFAIL, 682 "vinvalbuf", slpflag, slptimeo); 683 if (error == ENOLCK) 684 break; 685 splx(s); 686 return (error); 687 } 688 /* 689 * XXX Since there are no node locks for NFS, I 690 * believe there is a slight chance that a delayed 691 * write will occur while sleeping just above, so 692 * check for it. Note that vfs_bio_awrite expects 693 * buffers to reside on a queue, while VOP_BWRITE and 694 * brelse do not. 695 */ 696 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 697 (flags & V_SAVE)) { 698 699 if (bp->b_vp == vp) { 700 if (bp->b_flags & B_CLUSTEROK) { 701 BUF_UNLOCK(bp); 702 vfs_bio_awrite(bp); 703 } else { 704 bremfree(bp); 705 bp->b_flags |= B_ASYNC; 706 BUF_WRITE(bp); 707 } 708 } else { 709 bremfree(bp); 710 (void) BUF_WRITE(bp); 711 } 712 break; 713 } 714 bremfree(bp); 715 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 716 bp->b_flags &= ~B_ASYNC; 717 brelse(bp); 718 } 719 } 720 721 while (vp->v_numoutput > 0) { 722 vp->v_flag |= VBWAIT; 723 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 724 } 725 726 splx(s); 727 728 /* 729 * Destroy the copy in the VM cache, too. 730 */ 731 simple_lock(&vp->v_interlock); 732 object = vp->v_object; 733 if (object != NULL) { 734 vm_object_page_remove(object, 0, 0, 735 (flags & V_SAVE) ? TRUE : FALSE); 736 } 737 simple_unlock(&vp->v_interlock); 738 739 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 740 panic("vinvalbuf: flush failed"); 741 return (0); 742 } 743 744 /* 745 * Truncate a file's buffer and pages to a specified length. This 746 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 747 * sync activity. 748 */ 749 int 750 vtruncbuf(vp, cred, p, length, blksize) 751 register struct vnode *vp; 752 struct ucred *cred; 753 struct proc *p; 754 off_t length; 755 int blksize; 756 { 757 register struct buf *bp; 758 struct buf *nbp; 759 int s, anyfreed; 760 int trunclbn; 761 762 /* 763 * Round up to the *next* lbn. 764 */ 765 trunclbn = (length + blksize - 1) / blksize; 766 767 s = splbio(); 768 restart: 769 anyfreed = 1; 770 for (;anyfreed;) { 771 anyfreed = 0; 772 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 773 nbp = TAILQ_NEXT(bp, b_vnbufs); 774 if (bp->b_lblkno >= trunclbn) { 775 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 776 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 777 goto restart; 778 } else { 779 bremfree(bp); 780 bp->b_flags |= (B_INVAL | B_RELBUF); 781 bp->b_flags &= ~B_ASYNC; 782 brelse(bp); 783 anyfreed = 1; 784 } 785 if (nbp && 786 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 787 (nbp->b_vp != vp) || 788 (nbp->b_flags & B_DELWRI))) { 789 goto restart; 790 } 791 } 792 } 793 794 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 795 nbp = TAILQ_NEXT(bp, b_vnbufs); 796 if (bp->b_lblkno >= trunclbn) { 797 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 798 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 799 goto restart; 800 } else { 801 bremfree(bp); 802 bp->b_flags |= (B_INVAL | B_RELBUF); 803 bp->b_flags &= ~B_ASYNC; 804 brelse(bp); 805 anyfreed = 1; 806 } 807 if (nbp && 808 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 809 (nbp->b_vp != vp) || 810 (nbp->b_flags & B_DELWRI) == 0)) { 811 goto restart; 812 } 813 } 814 } 815 } 816 817 if (length > 0) { 818 restartsync: 819 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 820 nbp = TAILQ_NEXT(bp, b_vnbufs); 821 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 822 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 823 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 824 goto restart; 825 } else { 826 bremfree(bp); 827 if (bp->b_vp == vp) { 828 bp->b_flags |= B_ASYNC; 829 } else { 830 bp->b_flags &= ~B_ASYNC; 831 } 832 BUF_WRITE(bp); 833 } 834 goto restartsync; 835 } 836 837 } 838 } 839 840 while (vp->v_numoutput > 0) { 841 vp->v_flag |= VBWAIT; 842 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0); 843 } 844 845 splx(s); 846 847 vnode_pager_setsize(vp, length); 848 849 return (0); 850 } 851 852 /* 853 * Associate a buffer with a vnode. 854 */ 855 void 856 bgetvp(vp, bp) 857 register struct vnode *vp; 858 register struct buf *bp; 859 { 860 int s; 861 862 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 863 864 vhold(vp); 865 bp->b_vp = vp; 866 bp->b_dev = vn_todev(vp); 867 /* 868 * Insert onto list for new vnode. 869 */ 870 s = splbio(); 871 bp->b_xflags |= BX_VNCLEAN; 872 bp->b_xflags &= ~BX_VNDIRTY; 873 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 874 splx(s); 875 } 876 877 /* 878 * Disassociate a buffer from a vnode. 879 */ 880 void 881 brelvp(bp) 882 register struct buf *bp; 883 { 884 struct vnode *vp; 885 struct buflists *listheadp; 886 int s; 887 888 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 889 890 /* 891 * Delete from old vnode list, if on one. 892 */ 893 vp = bp->b_vp; 894 s = splbio(); 895 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 896 if (bp->b_xflags & BX_VNDIRTY) 897 listheadp = &vp->v_dirtyblkhd; 898 else 899 listheadp = &vp->v_cleanblkhd; 900 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 901 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 902 } 903 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 904 vp->v_flag &= ~VONWORKLST; 905 LIST_REMOVE(vp, v_synclist); 906 } 907 splx(s); 908 bp->b_vp = (struct vnode *) 0; 909 vdrop(vp); 910 } 911 912 /* 913 * The workitem queue. 914 * 915 * It is useful to delay writes of file data and filesystem metadata 916 * for tens of seconds so that quickly created and deleted files need 917 * not waste disk bandwidth being created and removed. To realize this, 918 * we append vnodes to a "workitem" queue. When running with a soft 919 * updates implementation, most pending metadata dependencies should 920 * not wait for more than a few seconds. Thus, mounted on block devices 921 * are delayed only about a half the time that file data is delayed. 922 * Similarly, directory updates are more critical, so are only delayed 923 * about a third the time that file data is delayed. Thus, there are 924 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 925 * one each second (driven off the filesystem syncer process). The 926 * syncer_delayno variable indicates the next queue that is to be processed. 927 * Items that need to be processed soon are placed in this queue: 928 * 929 * syncer_workitem_pending[syncer_delayno] 930 * 931 * A delay of fifteen seconds is done by placing the request fifteen 932 * entries later in the queue: 933 * 934 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 935 * 936 */ 937 938 /* 939 * Add an item to the syncer work queue. 940 */ 941 static void 942 vn_syncer_add_to_worklist(struct vnode *vp, int delay) 943 { 944 int s, slot; 945 946 s = splbio(); 947 948 if (vp->v_flag & VONWORKLST) { 949 LIST_REMOVE(vp, v_synclist); 950 } 951 952 if (delay > syncer_maxdelay - 2) 953 delay = syncer_maxdelay - 2; 954 slot = (syncer_delayno + delay) & syncer_mask; 955 956 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 957 vp->v_flag |= VONWORKLST; 958 splx(s); 959 } 960 961 struct proc *updateproc; 962 static void sched_sync __P((void)); 963 static struct kproc_desc up_kp = { 964 "syncer", 965 sched_sync, 966 &updateproc 967 }; 968 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 969 970 /* 971 * System filesystem synchronizer daemon. 972 */ 973 void 974 sched_sync(void) 975 { 976 struct synclist *slp; 977 struct vnode *vp; 978 long starttime; 979 int s; 980 struct proc *p = updateproc; 981 982 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, p, 983 SHUTDOWN_PRI_LAST); 984 985 for (;;) { 986 kproc_suspend_loop(p); 987 988 starttime = time_second; 989 990 /* 991 * Push files whose dirty time has expired. Be careful 992 * of interrupt race on slp queue. 993 */ 994 s = splbio(); 995 slp = &syncer_workitem_pending[syncer_delayno]; 996 syncer_delayno += 1; 997 if (syncer_delayno == syncer_maxdelay) 998 syncer_delayno = 0; 999 splx(s); 1000 1001 while ((vp = LIST_FIRST(slp)) != NULL) { 1002 if (VOP_ISLOCKED(vp, NULL) == 0) { 1003 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 1004 (void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p); 1005 VOP_UNLOCK(vp, 0, p); 1006 } 1007 s = splbio(); 1008 if (LIST_FIRST(slp) == vp) { 1009 /* 1010 * Note: v_tag VT_VFS vps can remain on the 1011 * worklist too with no dirty blocks, but 1012 * since sync_fsync() moves it to a different 1013 * slot we are safe. 1014 */ 1015 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1016 !vn_isdisk(vp, NULL)) 1017 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1018 /* 1019 * Put us back on the worklist. The worklist 1020 * routine will remove us from our current 1021 * position and then add us back in at a later 1022 * position. 1023 */ 1024 vn_syncer_add_to_worklist(vp, syncdelay); 1025 } 1026 splx(s); 1027 } 1028 1029 /* 1030 * Do soft update processing. 1031 */ 1032 softdep_process_worklist(NULL); 1033 1034 /* 1035 * The variable rushjob allows the kernel to speed up the 1036 * processing of the filesystem syncer process. A rushjob 1037 * value of N tells the filesystem syncer to process the next 1038 * N seconds worth of work on its queue ASAP. Currently rushjob 1039 * is used by the soft update code to speed up the filesystem 1040 * syncer process when the incore state is getting so far 1041 * ahead of the disk that the kernel memory pool is being 1042 * threatened with exhaustion. 1043 */ 1044 if (rushjob > 0) { 1045 rushjob -= 1; 1046 continue; 1047 } 1048 /* 1049 * If it has taken us less than a second to process the 1050 * current work, then wait. Otherwise start right over 1051 * again. We can still lose time if any single round 1052 * takes more than two seconds, but it does not really 1053 * matter as we are just trying to generally pace the 1054 * filesystem activity. 1055 */ 1056 if (time_second == starttime) 1057 tsleep(&lbolt, PPAUSE, "syncer", 0); 1058 } 1059 } 1060 1061 /* 1062 * Request the syncer daemon to speed up its work. 1063 * We never push it to speed up more than half of its 1064 * normal turn time, otherwise it could take over the cpu. 1065 */ 1066 int 1067 speedup_syncer() 1068 { 1069 int s; 1070 1071 s = splhigh(); 1072 if (updateproc->p_wchan == &lbolt) 1073 setrunnable(updateproc); 1074 splx(s); 1075 if (rushjob < syncdelay / 2) { 1076 rushjob += 1; 1077 stat_rush_requests += 1; 1078 return (1); 1079 } 1080 return(0); 1081 } 1082 1083 /* 1084 * Associate a p-buffer with a vnode. 1085 * 1086 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1087 * with the buffer. i.e. the bp has not been linked into the vnode or 1088 * ref-counted. 1089 */ 1090 void 1091 pbgetvp(vp, bp) 1092 register struct vnode *vp; 1093 register struct buf *bp; 1094 { 1095 1096 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1097 1098 bp->b_vp = vp; 1099 bp->b_flags |= B_PAGING; 1100 bp->b_dev = vn_todev(vp); 1101 } 1102 1103 /* 1104 * Disassociate a p-buffer from a vnode. 1105 */ 1106 void 1107 pbrelvp(bp) 1108 register struct buf *bp; 1109 { 1110 1111 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1112 1113 /* XXX REMOVE ME */ 1114 if (bp->b_vnbufs.tqe_next != NULL) { 1115 panic( 1116 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1117 bp, 1118 (int)bp->b_flags 1119 ); 1120 } 1121 bp->b_vp = (struct vnode *) 0; 1122 bp->b_flags &= ~B_PAGING; 1123 } 1124 1125 void 1126 pbreassignbuf(bp, newvp) 1127 struct buf *bp; 1128 struct vnode *newvp; 1129 { 1130 if ((bp->b_flags & B_PAGING) == 0) { 1131 panic( 1132 "pbreassignbuf() on non phys bp %p", 1133 bp 1134 ); 1135 } 1136 bp->b_vp = newvp; 1137 } 1138 1139 /* 1140 * Reassign a buffer from one vnode to another. 1141 * Used to assign file specific control information 1142 * (indirect blocks) to the vnode to which they belong. 1143 */ 1144 void 1145 reassignbuf(bp, newvp) 1146 register struct buf *bp; 1147 register struct vnode *newvp; 1148 { 1149 struct buflists *listheadp; 1150 int delay; 1151 int s; 1152 1153 if (newvp == NULL) { 1154 printf("reassignbuf: NULL"); 1155 return; 1156 } 1157 ++reassignbufcalls; 1158 1159 /* 1160 * B_PAGING flagged buffers cannot be reassigned because their vp 1161 * is not fully linked in. 1162 */ 1163 if (bp->b_flags & B_PAGING) 1164 panic("cannot reassign paging buffer"); 1165 1166 s = splbio(); 1167 /* 1168 * Delete from old vnode list, if on one. 1169 */ 1170 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1171 if (bp->b_xflags & BX_VNDIRTY) 1172 listheadp = &bp->b_vp->v_dirtyblkhd; 1173 else 1174 listheadp = &bp->b_vp->v_cleanblkhd; 1175 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1176 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1177 if (bp->b_vp != newvp) { 1178 vdrop(bp->b_vp); 1179 bp->b_vp = NULL; /* for clarification */ 1180 } 1181 } 1182 /* 1183 * If dirty, put on list of dirty buffers; otherwise insert onto list 1184 * of clean buffers. 1185 */ 1186 if (bp->b_flags & B_DELWRI) { 1187 struct buf *tbp; 1188 1189 listheadp = &newvp->v_dirtyblkhd; 1190 if ((newvp->v_flag & VONWORKLST) == 0) { 1191 switch (newvp->v_type) { 1192 case VDIR: 1193 delay = dirdelay; 1194 break; 1195 case VCHR: 1196 case VBLK: 1197 if (newvp->v_specmountpoint != NULL) { 1198 delay = metadelay; 1199 break; 1200 } 1201 /* fall through */ 1202 default: 1203 delay = filedelay; 1204 } 1205 vn_syncer_add_to_worklist(newvp, delay); 1206 } 1207 bp->b_xflags |= BX_VNDIRTY; 1208 tbp = TAILQ_FIRST(listheadp); 1209 if (tbp == NULL || 1210 bp->b_lblkno == 0 || 1211 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1212 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1213 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1214 ++reassignbufsortgood; 1215 } else if (bp->b_lblkno < 0) { 1216 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1217 ++reassignbufsortgood; 1218 } else if (reassignbufmethod == 1) { 1219 /* 1220 * New sorting algorithm, only handle sequential case, 1221 * otherwise append to end (but before metadata) 1222 */ 1223 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1224 (tbp->b_xflags & BX_VNDIRTY)) { 1225 /* 1226 * Found the best place to insert the buffer 1227 */ 1228 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1229 ++reassignbufsortgood; 1230 } else { 1231 /* 1232 * Missed, append to end, but before meta-data. 1233 * We know that the head buffer in the list is 1234 * not meta-data due to prior conditionals. 1235 * 1236 * Indirect effects: NFS second stage write 1237 * tends to wind up here, giving maximum 1238 * distance between the unstable write and the 1239 * commit rpc. 1240 */ 1241 tbp = TAILQ_LAST(listheadp, buflists); 1242 while (tbp && tbp->b_lblkno < 0) 1243 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1244 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1245 ++reassignbufsortbad; 1246 } 1247 } else { 1248 /* 1249 * Old sorting algorithm, scan queue and insert 1250 */ 1251 struct buf *ttbp; 1252 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1253 (ttbp->b_lblkno < bp->b_lblkno)) { 1254 ++reassignbufloops; 1255 tbp = ttbp; 1256 } 1257 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1258 } 1259 } else { 1260 bp->b_xflags |= BX_VNCLEAN; 1261 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1262 if ((newvp->v_flag & VONWORKLST) && 1263 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1264 newvp->v_flag &= ~VONWORKLST; 1265 LIST_REMOVE(newvp, v_synclist); 1266 } 1267 } 1268 if (bp->b_vp != newvp) { 1269 bp->b_vp = newvp; 1270 vhold(bp->b_vp); 1271 } 1272 splx(s); 1273 } 1274 1275 /* 1276 * Create a vnode for a block device. 1277 * Used for mounting the root file system. 1278 * XXX: This now changed to a VCHR due to the block/char merging. 1279 */ 1280 int 1281 bdevvp(dev, vpp) 1282 dev_t dev; 1283 struct vnode **vpp; 1284 { 1285 register struct vnode *vp; 1286 struct vnode *nvp; 1287 int error; 1288 1289 if (dev == NODEV) { 1290 *vpp = NULLVP; 1291 return (ENXIO); 1292 } 1293 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1294 if (error) { 1295 *vpp = NULLVP; 1296 return (error); 1297 } 1298 vp = nvp; 1299 vp->v_type = VCHR; 1300 addalias(vp, dev); 1301 *vpp = vp; 1302 return (0); 1303 } 1304 1305 /* 1306 * Add vnode to the alias list hung off the dev_t. 1307 * 1308 * The reason for this gunk is that multiple vnodes can reference 1309 * the same physical device, so checking vp->v_usecount to see 1310 * how many users there are is inadequate; the v_usecount for 1311 * the vnodes need to be accumulated. vcount() does that. 1312 */ 1313 void 1314 addaliasu(nvp, nvp_rdev) 1315 struct vnode *nvp; 1316 udev_t nvp_rdev; 1317 { 1318 1319 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1320 panic("addaliasu on non-special vnode"); 1321 addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0)); 1322 } 1323 1324 void 1325 addalias(nvp, dev) 1326 struct vnode *nvp; 1327 dev_t dev; 1328 { 1329 1330 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1331 panic("addalias on non-special vnode"); 1332 1333 nvp->v_rdev = dev; 1334 simple_lock(&spechash_slock); 1335 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1336 simple_unlock(&spechash_slock); 1337 } 1338 1339 /* 1340 * Grab a particular vnode from the free list, increment its 1341 * reference count and lock it. The vnode lock bit is set if the 1342 * vnode is being eliminated in vgone. The process is awakened 1343 * when the transition is completed, and an error returned to 1344 * indicate that the vnode is no longer usable (possibly having 1345 * been changed to a new file system type). 1346 */ 1347 int 1348 vget(vp, flags, p) 1349 register struct vnode *vp; 1350 int flags; 1351 struct proc *p; 1352 { 1353 int error; 1354 1355 /* 1356 * If the vnode is in the process of being cleaned out for 1357 * another use, we wait for the cleaning to finish and then 1358 * return failure. Cleaning is determined by checking that 1359 * the VXLOCK flag is set. 1360 */ 1361 if ((flags & LK_INTERLOCK) == 0) { 1362 simple_lock(&vp->v_interlock); 1363 } 1364 if (vp->v_flag & VXLOCK) { 1365 vp->v_flag |= VXWANT; 1366 simple_unlock(&vp->v_interlock); 1367 tsleep((caddr_t)vp, PINOD, "vget", 0); 1368 return (ENOENT); 1369 } 1370 1371 vp->v_usecount++; 1372 1373 if (VSHOULDBUSY(vp)) 1374 vbusy(vp); 1375 if (flags & LK_TYPE_MASK) { 1376 if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) { 1377 /* 1378 * must expand vrele here because we do not want 1379 * to call VOP_INACTIVE if the reference count 1380 * drops back to zero since it was never really 1381 * active. We must remove it from the free list 1382 * before sleeping so that multiple processes do 1383 * not try to recycle it. 1384 */ 1385 simple_lock(&vp->v_interlock); 1386 vp->v_usecount--; 1387 if (VSHOULDFREE(vp)) 1388 vfree(vp); 1389 simple_unlock(&vp->v_interlock); 1390 } 1391 return (error); 1392 } 1393 simple_unlock(&vp->v_interlock); 1394 return (0); 1395 } 1396 1397 void 1398 vref(struct vnode *vp) 1399 { 1400 simple_lock(&vp->v_interlock); 1401 vp->v_usecount++; 1402 simple_unlock(&vp->v_interlock); 1403 } 1404 1405 /* 1406 * Vnode put/release. 1407 * If count drops to zero, call inactive routine and return to freelist. 1408 */ 1409 void 1410 vrele(vp) 1411 struct vnode *vp; 1412 { 1413 struct proc *p = curproc; /* XXX */ 1414 1415 KASSERT(vp != NULL, ("vrele: null vp")); 1416 1417 simple_lock(&vp->v_interlock); 1418 1419 if (vp->v_usecount > 1) { 1420 1421 vp->v_usecount--; 1422 simple_unlock(&vp->v_interlock); 1423 1424 return; 1425 } 1426 1427 if (vp->v_usecount == 1) { 1428 1429 vp->v_usecount--; 1430 if (VSHOULDFREE(vp)) 1431 vfree(vp); 1432 /* 1433 * If we are doing a vput, the node is already locked, and we must 1434 * call VOP_INACTIVE with the node locked. So, in the case of 1435 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1436 */ 1437 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) { 1438 VOP_INACTIVE(vp, p); 1439 } 1440 1441 } else { 1442 #ifdef DIAGNOSTIC 1443 vprint("vrele: negative ref count", vp); 1444 simple_unlock(&vp->v_interlock); 1445 #endif 1446 panic("vrele: negative ref cnt"); 1447 } 1448 } 1449 1450 void 1451 vput(vp) 1452 struct vnode *vp; 1453 { 1454 struct proc *p = curproc; /* XXX */ 1455 1456 KASSERT(vp != NULL, ("vput: null vp")); 1457 1458 simple_lock(&vp->v_interlock); 1459 1460 if (vp->v_usecount > 1) { 1461 1462 vp->v_usecount--; 1463 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1464 return; 1465 1466 } 1467 1468 if (vp->v_usecount == 1) { 1469 1470 vp->v_usecount--; 1471 if (VSHOULDFREE(vp)) 1472 vfree(vp); 1473 /* 1474 * If we are doing a vput, the node is already locked, and we must 1475 * call VOP_INACTIVE with the node locked. So, in the case of 1476 * vrele, we explicitly lock the vnode before calling VOP_INACTIVE. 1477 */ 1478 simple_unlock(&vp->v_interlock); 1479 VOP_INACTIVE(vp, p); 1480 1481 } else { 1482 #ifdef DIAGNOSTIC 1483 vprint("vput: negative ref count", vp); 1484 #endif 1485 panic("vput: negative ref cnt"); 1486 } 1487 } 1488 1489 /* 1490 * Somebody doesn't want the vnode recycled. 1491 */ 1492 void 1493 vhold(vp) 1494 register struct vnode *vp; 1495 { 1496 int s; 1497 1498 s = splbio(); 1499 vp->v_holdcnt++; 1500 if (VSHOULDBUSY(vp)) 1501 vbusy(vp); 1502 splx(s); 1503 } 1504 1505 /* 1506 * One less who cares about this vnode. 1507 */ 1508 void 1509 vdrop(vp) 1510 register struct vnode *vp; 1511 { 1512 int s; 1513 1514 s = splbio(); 1515 if (vp->v_holdcnt <= 0) 1516 panic("vdrop: holdcnt"); 1517 vp->v_holdcnt--; 1518 if (VSHOULDFREE(vp)) 1519 vfree(vp); 1520 splx(s); 1521 } 1522 1523 /* 1524 * Remove any vnodes in the vnode table belonging to mount point mp. 1525 * 1526 * If MNT_NOFORCE is specified, there should not be any active ones, 1527 * return error if any are found (nb: this is a user error, not a 1528 * system error). If MNT_FORCE is specified, detach any active vnodes 1529 * that are found. 1530 */ 1531 #ifdef DIAGNOSTIC 1532 static int busyprt = 0; /* print out busy vnodes */ 1533 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1534 #endif 1535 1536 int 1537 vflush(mp, skipvp, flags) 1538 struct mount *mp; 1539 struct vnode *skipvp; 1540 int flags; 1541 { 1542 struct proc *p = curproc; /* XXX */ 1543 struct vnode *vp, *nvp; 1544 int busy = 0; 1545 1546 simple_lock(&mntvnode_slock); 1547 loop: 1548 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) { 1549 /* 1550 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1551 * Start over if it has (it won't be on the list anymore). 1552 */ 1553 if (vp->v_mount != mp) 1554 goto loop; 1555 nvp = LIST_NEXT(vp, v_mntvnodes); 1556 /* 1557 * Skip over a selected vnode. 1558 */ 1559 if (vp == skipvp) 1560 continue; 1561 1562 simple_lock(&vp->v_interlock); 1563 /* 1564 * Skip over a vnodes marked VSYSTEM. 1565 */ 1566 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1567 simple_unlock(&vp->v_interlock); 1568 continue; 1569 } 1570 /* 1571 * If WRITECLOSE is set, only flush out regular file vnodes 1572 * open for writing. 1573 */ 1574 if ((flags & WRITECLOSE) && 1575 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1576 simple_unlock(&vp->v_interlock); 1577 continue; 1578 } 1579 1580 /* 1581 * With v_usecount == 0, all we need to do is clear out the 1582 * vnode data structures and we are done. 1583 */ 1584 if (vp->v_usecount == 0) { 1585 simple_unlock(&mntvnode_slock); 1586 vgonel(vp, p); 1587 simple_lock(&mntvnode_slock); 1588 continue; 1589 } 1590 1591 /* 1592 * If FORCECLOSE is set, forcibly close the vnode. For block 1593 * or character devices, revert to an anonymous device. For 1594 * all other files, just kill them. 1595 */ 1596 if (flags & FORCECLOSE) { 1597 simple_unlock(&mntvnode_slock); 1598 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1599 vgonel(vp, p); 1600 } else { 1601 vclean(vp, 0, p); 1602 vp->v_op = spec_vnodeop_p; 1603 insmntque(vp, (struct mount *) 0); 1604 } 1605 simple_lock(&mntvnode_slock); 1606 continue; 1607 } 1608 #ifdef DIAGNOSTIC 1609 if (busyprt) 1610 vprint("vflush: busy vnode", vp); 1611 #endif 1612 simple_unlock(&vp->v_interlock); 1613 busy++; 1614 } 1615 simple_unlock(&mntvnode_slock); 1616 if (busy) 1617 return (EBUSY); 1618 return (0); 1619 } 1620 1621 /* 1622 * Disassociate the underlying file system from a vnode. 1623 */ 1624 static void 1625 vclean(vp, flags, p) 1626 struct vnode *vp; 1627 int flags; 1628 struct proc *p; 1629 { 1630 int active; 1631 vm_object_t obj; 1632 1633 /* 1634 * Check to see if the vnode is in use. If so we have to reference it 1635 * before we clean it out so that its count cannot fall to zero and 1636 * generate a race against ourselves to recycle it. 1637 */ 1638 if ((active = vp->v_usecount)) 1639 vp->v_usecount++; 1640 1641 /* 1642 * Prevent the vnode from being recycled or brought into use while we 1643 * clean it out. 1644 */ 1645 if (vp->v_flag & VXLOCK) 1646 panic("vclean: deadlock"); 1647 vp->v_flag |= VXLOCK; 1648 /* 1649 * Even if the count is zero, the VOP_INACTIVE routine may still 1650 * have the object locked while it cleans it out. The VOP_LOCK 1651 * ensures that the VOP_INACTIVE routine is done with its work. 1652 * For active vnodes, it ensures that no other activity can 1653 * occur while the underlying object is being cleaned out. 1654 */ 1655 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1656 1657 /* 1658 * Clean out any buffers associated with the vnode. 1659 */ 1660 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1661 if ((obj = vp->v_object) != NULL) { 1662 if (obj->ref_count == 0) { 1663 /* 1664 * vclean() may be called twice. The first time removes the 1665 * primary reference to the object, the second time goes 1666 * one further and is a special-case to terminate the object. 1667 */ 1668 vm_object_terminate(obj); 1669 } else { 1670 /* 1671 * Woe to the process that tries to page now :-). 1672 */ 1673 vm_pager_deallocate(obj); 1674 } 1675 } 1676 1677 /* 1678 * If purging an active vnode, it must be closed and 1679 * deactivated before being reclaimed. Note that the 1680 * VOP_INACTIVE will unlock the vnode. 1681 */ 1682 if (active) { 1683 if (flags & DOCLOSE) 1684 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p); 1685 VOP_INACTIVE(vp, p); 1686 } else { 1687 /* 1688 * Any other processes trying to obtain this lock must first 1689 * wait for VXLOCK to clear, then call the new lock operation. 1690 */ 1691 VOP_UNLOCK(vp, 0, p); 1692 } 1693 /* 1694 * Reclaim the vnode. 1695 */ 1696 if (VOP_RECLAIM(vp, p)) 1697 panic("vclean: cannot reclaim"); 1698 1699 if (active) { 1700 /* 1701 * Inline copy of vrele() since VOP_INACTIVE 1702 * has already been called. 1703 */ 1704 simple_lock(&vp->v_interlock); 1705 if (--vp->v_usecount <= 0) { 1706 #ifdef DIAGNOSTIC 1707 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1708 vprint("vclean: bad ref count", vp); 1709 panic("vclean: ref cnt"); 1710 } 1711 #endif 1712 vfree(vp); 1713 } 1714 simple_unlock(&vp->v_interlock); 1715 } 1716 1717 cache_purge(vp); 1718 if (vp->v_vnlock) { 1719 FREE(vp->v_vnlock, M_VNODE); 1720 vp->v_vnlock = NULL; 1721 } 1722 1723 if (VSHOULDFREE(vp)) 1724 vfree(vp); 1725 1726 /* 1727 * Done with purge, notify sleepers of the grim news. 1728 */ 1729 vp->v_op = dead_vnodeop_p; 1730 vn_pollgone(vp); 1731 vp->v_tag = VT_NON; 1732 vp->v_flag &= ~VXLOCK; 1733 if (vp->v_flag & VXWANT) { 1734 vp->v_flag &= ~VXWANT; 1735 wakeup((caddr_t) vp); 1736 } 1737 } 1738 1739 /* 1740 * Eliminate all activity associated with the requested vnode 1741 * and with all vnodes aliased to the requested vnode. 1742 */ 1743 int 1744 vop_revoke(ap) 1745 struct vop_revoke_args /* { 1746 struct vnode *a_vp; 1747 int a_flags; 1748 } */ *ap; 1749 { 1750 struct vnode *vp, *vq; 1751 dev_t dev; 1752 1753 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1754 1755 vp = ap->a_vp; 1756 /* 1757 * If a vgone (or vclean) is already in progress, 1758 * wait until it is done and return. 1759 */ 1760 if (vp->v_flag & VXLOCK) { 1761 vp->v_flag |= VXWANT; 1762 simple_unlock(&vp->v_interlock); 1763 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1764 return (0); 1765 } 1766 dev = vp->v_rdev; 1767 for (;;) { 1768 simple_lock(&spechash_slock); 1769 vq = SLIST_FIRST(&dev->si_hlist); 1770 simple_unlock(&spechash_slock); 1771 if (!vq) 1772 break; 1773 vgone(vq); 1774 } 1775 return (0); 1776 } 1777 1778 /* 1779 * Recycle an unused vnode to the front of the free list. 1780 * Release the passed interlock if the vnode will be recycled. 1781 */ 1782 int 1783 vrecycle(vp, inter_lkp, p) 1784 struct vnode *vp; 1785 struct simplelock *inter_lkp; 1786 struct proc *p; 1787 { 1788 1789 simple_lock(&vp->v_interlock); 1790 if (vp->v_usecount == 0) { 1791 if (inter_lkp) { 1792 simple_unlock(inter_lkp); 1793 } 1794 vgonel(vp, p); 1795 return (1); 1796 } 1797 simple_unlock(&vp->v_interlock); 1798 return (0); 1799 } 1800 1801 /* 1802 * Eliminate all activity associated with a vnode 1803 * in preparation for reuse. 1804 */ 1805 void 1806 vgone(vp) 1807 register struct vnode *vp; 1808 { 1809 struct proc *p = curproc; /* XXX */ 1810 1811 simple_lock(&vp->v_interlock); 1812 vgonel(vp, p); 1813 } 1814 1815 /* 1816 * vgone, with the vp interlock held. 1817 */ 1818 void 1819 vgonel(vp, p) 1820 struct vnode *vp; 1821 struct proc *p; 1822 { 1823 int s; 1824 1825 /* 1826 * If a vgone (or vclean) is already in progress, 1827 * wait until it is done and return. 1828 */ 1829 if (vp->v_flag & VXLOCK) { 1830 vp->v_flag |= VXWANT; 1831 simple_unlock(&vp->v_interlock); 1832 tsleep((caddr_t)vp, PINOD, "vgone", 0); 1833 return; 1834 } 1835 1836 /* 1837 * Clean out the filesystem specific data. 1838 */ 1839 vclean(vp, DOCLOSE, p); 1840 simple_lock(&vp->v_interlock); 1841 1842 /* 1843 * Delete from old mount point vnode list, if on one. 1844 */ 1845 if (vp->v_mount != NULL) 1846 insmntque(vp, (struct mount *)0); 1847 /* 1848 * If special device, remove it from special device alias list 1849 * if it is on one. 1850 */ 1851 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 1852 simple_lock(&spechash_slock); 1853 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext); 1854 freedev(vp->v_rdev); 1855 simple_unlock(&spechash_slock); 1856 vp->v_rdev = NULL; 1857 } 1858 1859 /* 1860 * If it is on the freelist and not already at the head, 1861 * move it to the head of the list. The test of the back 1862 * pointer and the reference count of zero is because 1863 * it will be removed from the free list by getnewvnode, 1864 * but will not have its reference count incremented until 1865 * after calling vgone. If the reference count were 1866 * incremented first, vgone would (incorrectly) try to 1867 * close the previous instance of the underlying object. 1868 */ 1869 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 1870 s = splbio(); 1871 simple_lock(&vnode_free_list_slock); 1872 if (vp->v_flag & VFREE) { 1873 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1874 } else if (vp->v_flag & VTBFREE) { 1875 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 1876 vp->v_flag &= ~VTBFREE; 1877 freevnodes++; 1878 } else 1879 freevnodes++; 1880 vp->v_flag |= VFREE; 1881 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1882 simple_unlock(&vnode_free_list_slock); 1883 splx(s); 1884 } 1885 1886 vp->v_type = VBAD; 1887 simple_unlock(&vp->v_interlock); 1888 } 1889 1890 /* 1891 * Lookup a vnode by device number. 1892 */ 1893 int 1894 vfinddev(dev, type, vpp) 1895 dev_t dev; 1896 enum vtype type; 1897 struct vnode **vpp; 1898 { 1899 struct vnode *vp; 1900 1901 simple_lock(&spechash_slock); 1902 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 1903 if (type == vp->v_type) { 1904 *vpp = vp; 1905 simple_unlock(&spechash_slock); 1906 return (1); 1907 } 1908 } 1909 simple_unlock(&spechash_slock); 1910 return (0); 1911 } 1912 1913 /* 1914 * Calculate the total number of references to a special device. 1915 */ 1916 int 1917 vcount(vp) 1918 struct vnode *vp; 1919 { 1920 struct vnode *vq; 1921 int count; 1922 1923 count = 0; 1924 simple_lock(&spechash_slock); 1925 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext) 1926 count += vq->v_usecount; 1927 simple_unlock(&spechash_slock); 1928 return (count); 1929 } 1930 1931 /* 1932 * Same as above, but using the dev_t as argument 1933 */ 1934 1935 int 1936 count_dev(dev) 1937 dev_t dev; 1938 { 1939 struct vnode *vp; 1940 1941 vp = SLIST_FIRST(&dev->si_hlist); 1942 if (vp == NULL) 1943 return (0); 1944 return(vcount(vp)); 1945 } 1946 1947 /* 1948 * Print out a description of a vnode. 1949 */ 1950 static char *typename[] = 1951 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 1952 1953 void 1954 vprint(label, vp) 1955 char *label; 1956 struct vnode *vp; 1957 { 1958 char buf[96]; 1959 1960 if (label != NULL) 1961 printf("%s: %p: ", label, (void *)vp); 1962 else 1963 printf("%p: ", (void *)vp); 1964 printf("type %s, usecount %d, writecount %d, refcount %d,", 1965 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1966 vp->v_holdcnt); 1967 buf[0] = '\0'; 1968 if (vp->v_flag & VROOT) 1969 strcat(buf, "|VROOT"); 1970 if (vp->v_flag & VTEXT) 1971 strcat(buf, "|VTEXT"); 1972 if (vp->v_flag & VSYSTEM) 1973 strcat(buf, "|VSYSTEM"); 1974 if (vp->v_flag & VXLOCK) 1975 strcat(buf, "|VXLOCK"); 1976 if (vp->v_flag & VXWANT) 1977 strcat(buf, "|VXWANT"); 1978 if (vp->v_flag & VBWAIT) 1979 strcat(buf, "|VBWAIT"); 1980 if (vp->v_flag & VDOOMED) 1981 strcat(buf, "|VDOOMED"); 1982 if (vp->v_flag & VFREE) 1983 strcat(buf, "|VFREE"); 1984 if (vp->v_flag & VOBJBUF) 1985 strcat(buf, "|VOBJBUF"); 1986 if (buf[0] != '\0') 1987 printf(" flags (%s)", &buf[1]); 1988 if (vp->v_data == NULL) { 1989 printf("\n"); 1990 } else { 1991 printf("\n\t"); 1992 VOP_PRINT(vp); 1993 } 1994 } 1995 1996 #ifdef DDB 1997 #include <ddb/ddb.h> 1998 /* 1999 * List all of the locked vnodes in the system. 2000 * Called when debugging the kernel. 2001 */ 2002 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 2003 { 2004 struct proc *p = curproc; /* XXX */ 2005 struct mount *mp, *nmp; 2006 struct vnode *vp; 2007 2008 printf("Locked vnodes\n"); 2009 simple_lock(&mountlist_slock); 2010 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2011 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 2012 nmp = TAILQ_NEXT(mp, mnt_list); 2013 continue; 2014 } 2015 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2016 if (VOP_ISLOCKED(vp, NULL)) 2017 vprint((char *)0, vp); 2018 } 2019 simple_lock(&mountlist_slock); 2020 nmp = TAILQ_NEXT(mp, mnt_list); 2021 vfs_unbusy(mp, p); 2022 } 2023 simple_unlock(&mountlist_slock); 2024 } 2025 #endif 2026 2027 /* 2028 * Top level filesystem related information gathering. 2029 */ 2030 static int sysctl_ovfs_conf __P(SYSCTL_HANDLER_ARGS); 2031 2032 static int 2033 vfs_sysctl SYSCTL_HANDLER_ARGS 2034 { 2035 int *name = (int *)arg1 - 1; /* XXX */ 2036 u_int namelen = arg2 + 1; /* XXX */ 2037 struct vfsconf *vfsp; 2038 2039 #if 1 || defined(COMPAT_PRELITE2) 2040 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2041 if (namelen == 1) 2042 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2043 #endif 2044 2045 #ifdef notyet 2046 /* all sysctl names at this level are at least name and field */ 2047 if (namelen < 2) 2048 return (ENOTDIR); /* overloaded */ 2049 if (name[0] != VFS_GENERIC) { 2050 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2051 if (vfsp->vfc_typenum == name[0]) 2052 break; 2053 if (vfsp == NULL) 2054 return (EOPNOTSUPP); 2055 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2056 oldp, oldlenp, newp, newlen, p)); 2057 } 2058 #endif 2059 switch (name[1]) { 2060 case VFS_MAXTYPENUM: 2061 if (namelen != 2) 2062 return (ENOTDIR); 2063 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2064 case VFS_CONF: 2065 if (namelen != 3) 2066 return (ENOTDIR); /* overloaded */ 2067 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2068 if (vfsp->vfc_typenum == name[2]) 2069 break; 2070 if (vfsp == NULL) 2071 return (EOPNOTSUPP); 2072 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2073 } 2074 return (EOPNOTSUPP); 2075 } 2076 2077 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2078 "Generic filesystem"); 2079 2080 #if 1 || defined(COMPAT_PRELITE2) 2081 2082 static int 2083 sysctl_ovfs_conf SYSCTL_HANDLER_ARGS 2084 { 2085 int error; 2086 struct vfsconf *vfsp; 2087 struct ovfsconf ovfs; 2088 2089 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2090 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2091 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2092 ovfs.vfc_index = vfsp->vfc_typenum; 2093 ovfs.vfc_refcount = vfsp->vfc_refcount; 2094 ovfs.vfc_flags = vfsp->vfc_flags; 2095 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2096 if (error) 2097 return error; 2098 } 2099 return 0; 2100 } 2101 2102 #endif /* 1 || COMPAT_PRELITE2 */ 2103 2104 #if 0 2105 #define KINFO_VNODESLOP 10 2106 /* 2107 * Dump vnode list (via sysctl). 2108 * Copyout address of vnode followed by vnode. 2109 */ 2110 /* ARGSUSED */ 2111 static int 2112 sysctl_vnode SYSCTL_HANDLER_ARGS 2113 { 2114 struct proc *p = curproc; /* XXX */ 2115 struct mount *mp, *nmp; 2116 struct vnode *nvp, *vp; 2117 int error; 2118 2119 #define VPTRSZ sizeof (struct vnode *) 2120 #define VNODESZ sizeof (struct vnode) 2121 2122 req->lock = 0; 2123 if (!req->oldptr) /* Make an estimate */ 2124 return (SYSCTL_OUT(req, 0, 2125 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2126 2127 simple_lock(&mountlist_slock); 2128 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2129 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 2130 nmp = TAILQ_NEXT(mp, mnt_list); 2131 continue; 2132 } 2133 again: 2134 simple_lock(&mntvnode_slock); 2135 for (vp = LIST_FIRST(&mp->mnt_vnodelist); 2136 vp != NULL; 2137 vp = nvp) { 2138 /* 2139 * Check that the vp is still associated with 2140 * this filesystem. RACE: could have been 2141 * recycled onto the same filesystem. 2142 */ 2143 if (vp->v_mount != mp) { 2144 simple_unlock(&mntvnode_slock); 2145 goto again; 2146 } 2147 nvp = LIST_NEXT(vp, v_mntvnodes); 2148 simple_unlock(&mntvnode_slock); 2149 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2150 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2151 return (error); 2152 simple_lock(&mntvnode_slock); 2153 } 2154 simple_unlock(&mntvnode_slock); 2155 simple_lock(&mountlist_slock); 2156 nmp = TAILQ_NEXT(mp, mnt_list); 2157 vfs_unbusy(mp, p); 2158 } 2159 simple_unlock(&mountlist_slock); 2160 2161 return (0); 2162 } 2163 #endif 2164 2165 /* 2166 * XXX 2167 * Exporting the vnode list on large systems causes them to crash. 2168 * Exporting the vnode list on medium systems causes sysctl to coredump. 2169 */ 2170 #if 0 2171 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2172 0, 0, sysctl_vnode, "S,vnode", ""); 2173 #endif 2174 2175 /* 2176 * Check to see if a filesystem is mounted on a block device. 2177 */ 2178 int 2179 vfs_mountedon(vp) 2180 struct vnode *vp; 2181 { 2182 2183 if (vp->v_specmountpoint != NULL) 2184 return (EBUSY); 2185 return (0); 2186 } 2187 2188 /* 2189 * Unmount all filesystems. The list is traversed in reverse order 2190 * of mounting to avoid dependencies. 2191 */ 2192 void 2193 vfs_unmountall() 2194 { 2195 struct mount *mp; 2196 struct proc *p; 2197 int error; 2198 2199 if (curproc != NULL) 2200 p = curproc; 2201 else 2202 p = initproc; /* XXX XXX should this be proc0? */ 2203 /* 2204 * Since this only runs when rebooting, it is not interlocked. 2205 */ 2206 while(!TAILQ_EMPTY(&mountlist)) { 2207 mp = TAILQ_LAST(&mountlist, mntlist); 2208 error = dounmount(mp, MNT_FORCE, p); 2209 if (error) { 2210 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2211 printf("unmount of %s failed (", 2212 mp->mnt_stat.f_mntonname); 2213 if (error == EBUSY) 2214 printf("BUSY)\n"); 2215 else 2216 printf("%d)\n", error); 2217 } else { 2218 /* The unmount has removed mp from the mountlist */ 2219 } 2220 } 2221 } 2222 2223 /* 2224 * Build hash lists of net addresses and hang them off the mount point. 2225 * Called by ufs_mount() to set up the lists of export addresses. 2226 */ 2227 static int 2228 vfs_hang_addrlist(mp, nep, argp) 2229 struct mount *mp; 2230 struct netexport *nep; 2231 struct export_args *argp; 2232 { 2233 register struct netcred *np; 2234 register struct radix_node_head *rnh; 2235 register int i; 2236 struct radix_node *rn; 2237 struct sockaddr *saddr, *smask = 0; 2238 struct domain *dom; 2239 int error; 2240 2241 if (argp->ex_addrlen == 0) { 2242 if (mp->mnt_flag & MNT_DEFEXPORTED) 2243 return (EPERM); 2244 np = &nep->ne_defexported; 2245 np->netc_exflags = argp->ex_flags; 2246 np->netc_anon = argp->ex_anon; 2247 np->netc_anon.cr_ref = 1; 2248 mp->mnt_flag |= MNT_DEFEXPORTED; 2249 return (0); 2250 } 2251 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2252 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 2253 bzero((caddr_t) np, i); 2254 saddr = (struct sockaddr *) (np + 1); 2255 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2256 goto out; 2257 if (saddr->sa_len > argp->ex_addrlen) 2258 saddr->sa_len = argp->ex_addrlen; 2259 if (argp->ex_masklen) { 2260 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 2261 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 2262 if (error) 2263 goto out; 2264 if (smask->sa_len > argp->ex_masklen) 2265 smask->sa_len = argp->ex_masklen; 2266 } 2267 i = saddr->sa_family; 2268 if ((rnh = nep->ne_rtable[i]) == 0) { 2269 /* 2270 * Seems silly to initialize every AF when most are not used, 2271 * do so on demand here 2272 */ 2273 for (dom = domains; dom; dom = dom->dom_next) 2274 if (dom->dom_family == i && dom->dom_rtattach) { 2275 dom->dom_rtattach((void **) &nep->ne_rtable[i], 2276 dom->dom_rtoffset); 2277 break; 2278 } 2279 if ((rnh = nep->ne_rtable[i]) == 0) { 2280 error = ENOBUFS; 2281 goto out; 2282 } 2283 } 2284 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 2285 np->netc_rnodes); 2286 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 2287 error = EPERM; 2288 goto out; 2289 } 2290 np->netc_exflags = argp->ex_flags; 2291 np->netc_anon = argp->ex_anon; 2292 np->netc_anon.cr_ref = 1; 2293 return (0); 2294 out: 2295 free(np, M_NETADDR); 2296 return (error); 2297 } 2298 2299 /* ARGSUSED */ 2300 static int 2301 vfs_free_netcred(rn, w) 2302 struct radix_node *rn; 2303 void *w; 2304 { 2305 register struct radix_node_head *rnh = (struct radix_node_head *) w; 2306 2307 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2308 free((caddr_t) rn, M_NETADDR); 2309 return (0); 2310 } 2311 2312 /* 2313 * Free the net address hash lists that are hanging off the mount points. 2314 */ 2315 static void 2316 vfs_free_addrlist(nep) 2317 struct netexport *nep; 2318 { 2319 register int i; 2320 register struct radix_node_head *rnh; 2321 2322 for (i = 0; i <= AF_MAX; i++) 2323 if ((rnh = nep->ne_rtable[i])) { 2324 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2325 (caddr_t) rnh); 2326 free((caddr_t) rnh, M_RTABLE); 2327 nep->ne_rtable[i] = 0; 2328 } 2329 } 2330 2331 int 2332 vfs_export(mp, nep, argp) 2333 struct mount *mp; 2334 struct netexport *nep; 2335 struct export_args *argp; 2336 { 2337 int error; 2338 2339 if (argp->ex_flags & MNT_DELEXPORT) { 2340 if (mp->mnt_flag & MNT_EXPUBLIC) { 2341 vfs_setpublicfs(NULL, NULL, NULL); 2342 mp->mnt_flag &= ~MNT_EXPUBLIC; 2343 } 2344 vfs_free_addrlist(nep); 2345 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2346 } 2347 if (argp->ex_flags & MNT_EXPORTED) { 2348 if (argp->ex_flags & MNT_EXPUBLIC) { 2349 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2350 return (error); 2351 mp->mnt_flag |= MNT_EXPUBLIC; 2352 } 2353 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2354 return (error); 2355 mp->mnt_flag |= MNT_EXPORTED; 2356 } 2357 return (0); 2358 } 2359 2360 2361 /* 2362 * Set the publicly exported filesystem (WebNFS). Currently, only 2363 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2364 */ 2365 int 2366 vfs_setpublicfs(mp, nep, argp) 2367 struct mount *mp; 2368 struct netexport *nep; 2369 struct export_args *argp; 2370 { 2371 int error; 2372 struct vnode *rvp; 2373 char *cp; 2374 2375 /* 2376 * mp == NULL -> invalidate the current info, the FS is 2377 * no longer exported. May be called from either vfs_export 2378 * or unmount, so check if it hasn't already been done. 2379 */ 2380 if (mp == NULL) { 2381 if (nfs_pub.np_valid) { 2382 nfs_pub.np_valid = 0; 2383 if (nfs_pub.np_index != NULL) { 2384 FREE(nfs_pub.np_index, M_TEMP); 2385 nfs_pub.np_index = NULL; 2386 } 2387 } 2388 return (0); 2389 } 2390 2391 /* 2392 * Only one allowed at a time. 2393 */ 2394 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2395 return (EBUSY); 2396 2397 /* 2398 * Get real filehandle for root of exported FS. 2399 */ 2400 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2401 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2402 2403 if ((error = VFS_ROOT(mp, &rvp))) 2404 return (error); 2405 2406 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2407 return (error); 2408 2409 vput(rvp); 2410 2411 /* 2412 * If an indexfile was specified, pull it in. 2413 */ 2414 if (argp->ex_indexfile != NULL) { 2415 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2416 M_WAITOK); 2417 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2418 MAXNAMLEN, (size_t *)0); 2419 if (!error) { 2420 /* 2421 * Check for illegal filenames. 2422 */ 2423 for (cp = nfs_pub.np_index; *cp; cp++) { 2424 if (*cp == '/') { 2425 error = EINVAL; 2426 break; 2427 } 2428 } 2429 } 2430 if (error) { 2431 FREE(nfs_pub.np_index, M_TEMP); 2432 return (error); 2433 } 2434 } 2435 2436 nfs_pub.np_mount = mp; 2437 nfs_pub.np_valid = 1; 2438 return (0); 2439 } 2440 2441 struct netcred * 2442 vfs_export_lookup(mp, nep, nam) 2443 register struct mount *mp; 2444 struct netexport *nep; 2445 struct sockaddr *nam; 2446 { 2447 register struct netcred *np; 2448 register struct radix_node_head *rnh; 2449 struct sockaddr *saddr; 2450 2451 np = NULL; 2452 if (mp->mnt_flag & MNT_EXPORTED) { 2453 /* 2454 * Lookup in the export list first. 2455 */ 2456 if (nam != NULL) { 2457 saddr = nam; 2458 rnh = nep->ne_rtable[saddr->sa_family]; 2459 if (rnh != NULL) { 2460 np = (struct netcred *) 2461 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2462 rnh); 2463 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2464 np = NULL; 2465 } 2466 } 2467 /* 2468 * If no address match, use the default if it exists. 2469 */ 2470 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2471 np = &nep->ne_defexported; 2472 } 2473 return (np); 2474 } 2475 2476 /* 2477 * perform msync on all vnodes under a mount point 2478 * the mount point must be locked. 2479 */ 2480 void 2481 vfs_msync(struct mount *mp, int flags) { 2482 struct vnode *vp, *nvp; 2483 struct vm_object *obj; 2484 int anyio, tries; 2485 2486 tries = 5; 2487 loop: 2488 anyio = 0; 2489 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp != NULL; vp = nvp) { 2490 2491 nvp = LIST_NEXT(vp, v_mntvnodes); 2492 2493 if (vp->v_mount != mp) { 2494 goto loop; 2495 } 2496 2497 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2498 continue; 2499 2500 if (flags != MNT_WAIT) { 2501 obj = vp->v_object; 2502 if (obj == NULL || (obj->flags & OBJ_MIGHTBEDIRTY) == 0) 2503 continue; 2504 if (VOP_ISLOCKED(vp, NULL)) 2505 continue; 2506 } 2507 2508 simple_lock(&vp->v_interlock); 2509 if (vp->v_object && 2510 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) { 2511 if (!vget(vp, 2512 LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) { 2513 if (vp->v_object) { 2514 vm_object_page_clean(vp->v_object, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2515 anyio = 1; 2516 } 2517 vput(vp); 2518 } 2519 } else { 2520 simple_unlock(&vp->v_interlock); 2521 } 2522 } 2523 if (anyio && (--tries > 0)) 2524 goto loop; 2525 } 2526 2527 /* 2528 * Create the VM object needed for VMIO and mmap support. This 2529 * is done for all VREG files in the system. Some filesystems might 2530 * afford the additional metadata buffering capability of the 2531 * VMIO code by making the device node be VMIO mode also. 2532 * 2533 * vp must be locked when vfs_object_create is called. 2534 */ 2535 int 2536 vfs_object_create(vp, p, cred) 2537 struct vnode *vp; 2538 struct proc *p; 2539 struct ucred *cred; 2540 { 2541 struct vattr vat; 2542 vm_object_t object; 2543 int error = 0; 2544 2545 if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE) 2546 return 0; 2547 2548 retry: 2549 if ((object = vp->v_object) == NULL) { 2550 if (vp->v_type == VREG || vp->v_type == VDIR) { 2551 if ((error = VOP_GETATTR(vp, &vat, cred, p)) != 0) 2552 goto retn; 2553 object = vnode_pager_alloc(vp, vat.va_size, 0, 0); 2554 } else if (devsw(vp->v_rdev) != NULL) { 2555 /* 2556 * This simply allocates the biggest object possible 2557 * for a disk vnode. This should be fixed, but doesn't 2558 * cause any problems (yet). 2559 */ 2560 object = vnode_pager_alloc(vp, IDX_TO_OFF(INT_MAX), 0, 0); 2561 } else { 2562 goto retn; 2563 } 2564 /* 2565 * Dereference the reference we just created. This assumes 2566 * that the object is associated with the vp. 2567 */ 2568 object->ref_count--; 2569 vp->v_usecount--; 2570 } else { 2571 if (object->flags & OBJ_DEAD) { 2572 VOP_UNLOCK(vp, 0, p); 2573 tsleep(object, PVM, "vodead", 0); 2574 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 2575 goto retry; 2576 } 2577 } 2578 2579 KASSERT(vp->v_object != NULL, ("vfs_object_create: NULL object")); 2580 vp->v_flag |= VOBJBUF; 2581 2582 retn: 2583 return error; 2584 } 2585 2586 static void 2587 vfree(vp) 2588 struct vnode *vp; 2589 { 2590 int s; 2591 2592 s = splbio(); 2593 simple_lock(&vnode_free_list_slock); 2594 if (vp->v_flag & VTBFREE) { 2595 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 2596 vp->v_flag &= ~VTBFREE; 2597 } 2598 if (vp->v_flag & VAGE) { 2599 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2600 } else { 2601 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2602 } 2603 freevnodes++; 2604 simple_unlock(&vnode_free_list_slock); 2605 vp->v_flag &= ~VAGE; 2606 vp->v_flag |= VFREE; 2607 splx(s); 2608 } 2609 2610 void 2611 vbusy(vp) 2612 struct vnode *vp; 2613 { 2614 int s; 2615 2616 s = splbio(); 2617 simple_lock(&vnode_free_list_slock); 2618 if (vp->v_flag & VTBFREE) { 2619 TAILQ_REMOVE(&vnode_tobefree_list, vp, v_freelist); 2620 vp->v_flag &= ~VTBFREE; 2621 } else { 2622 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2623 freevnodes--; 2624 } 2625 simple_unlock(&vnode_free_list_slock); 2626 vp->v_flag &= ~(VFREE|VAGE); 2627 splx(s); 2628 } 2629 2630 /* 2631 * Record a process's interest in events which might happen to 2632 * a vnode. Because poll uses the historic select-style interface 2633 * internally, this routine serves as both the ``check for any 2634 * pending events'' and the ``record my interest in future events'' 2635 * functions. (These are done together, while the lock is held, 2636 * to avoid race conditions.) 2637 */ 2638 int 2639 vn_pollrecord(vp, p, events) 2640 struct vnode *vp; 2641 struct proc *p; 2642 short events; 2643 { 2644 simple_lock(&vp->v_pollinfo.vpi_lock); 2645 if (vp->v_pollinfo.vpi_revents & events) { 2646 /* 2647 * This leaves events we are not interested 2648 * in available for the other process which 2649 * which presumably had requested them 2650 * (otherwise they would never have been 2651 * recorded). 2652 */ 2653 events &= vp->v_pollinfo.vpi_revents; 2654 vp->v_pollinfo.vpi_revents &= ~events; 2655 2656 simple_unlock(&vp->v_pollinfo.vpi_lock); 2657 return events; 2658 } 2659 vp->v_pollinfo.vpi_events |= events; 2660 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2661 simple_unlock(&vp->v_pollinfo.vpi_lock); 2662 return 0; 2663 } 2664 2665 /* 2666 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2667 * it is possible for us to miss an event due to race conditions, but 2668 * that condition is expected to be rare, so for the moment it is the 2669 * preferred interface. 2670 */ 2671 void 2672 vn_pollevent(vp, events) 2673 struct vnode *vp; 2674 short events; 2675 { 2676 simple_lock(&vp->v_pollinfo.vpi_lock); 2677 if (vp->v_pollinfo.vpi_events & events) { 2678 /* 2679 * We clear vpi_events so that we don't 2680 * call selwakeup() twice if two events are 2681 * posted before the polling process(es) is 2682 * awakened. This also ensures that we take at 2683 * most one selwakeup() if the polling process 2684 * is no longer interested. However, it does 2685 * mean that only one event can be noticed at 2686 * a time. (Perhaps we should only clear those 2687 * event bits which we note?) XXX 2688 */ 2689 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2690 vp->v_pollinfo.vpi_revents |= events; 2691 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2692 } 2693 simple_unlock(&vp->v_pollinfo.vpi_lock); 2694 } 2695 2696 /* 2697 * Wake up anyone polling on vp because it is being revoked. 2698 * This depends on dead_poll() returning POLLHUP for correct 2699 * behavior. 2700 */ 2701 void 2702 vn_pollgone(vp) 2703 struct vnode *vp; 2704 { 2705 simple_lock(&vp->v_pollinfo.vpi_lock); 2706 if (vp->v_pollinfo.vpi_events) { 2707 vp->v_pollinfo.vpi_events = 0; 2708 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2709 } 2710 simple_unlock(&vp->v_pollinfo.vpi_lock); 2711 } 2712 2713 2714 2715 /* 2716 * Routine to create and manage a filesystem syncer vnode. 2717 */ 2718 #define sync_close ((int (*) __P((struct vop_close_args *)))nullop) 2719 static int sync_fsync __P((struct vop_fsync_args *)); 2720 static int sync_inactive __P((struct vop_inactive_args *)); 2721 static int sync_reclaim __P((struct vop_reclaim_args *)); 2722 #define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock) 2723 #define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock) 2724 static int sync_print __P((struct vop_print_args *)); 2725 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked) 2726 2727 static vop_t **sync_vnodeop_p; 2728 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2729 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2730 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2731 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2732 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2733 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2734 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2735 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2736 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2737 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2738 { NULL, NULL } 2739 }; 2740 static struct vnodeopv_desc sync_vnodeop_opv_desc = 2741 { &sync_vnodeop_p, sync_vnodeop_entries }; 2742 2743 VNODEOP_SET(sync_vnodeop_opv_desc); 2744 2745 /* 2746 * Create a new filesystem syncer vnode for the specified mount point. 2747 */ 2748 int 2749 vfs_allocate_syncvnode(mp) 2750 struct mount *mp; 2751 { 2752 struct vnode *vp; 2753 static long start, incr, next; 2754 int error; 2755 2756 /* Allocate a new vnode */ 2757 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2758 mp->mnt_syncer = NULL; 2759 return (error); 2760 } 2761 vp->v_type = VNON; 2762 /* 2763 * Place the vnode onto the syncer worklist. We attempt to 2764 * scatter them about on the list so that they will go off 2765 * at evenly distributed times even if all the filesystems 2766 * are mounted at once. 2767 */ 2768 next += incr; 2769 if (next == 0 || next > syncer_maxdelay) { 2770 start /= 2; 2771 incr /= 2; 2772 if (start == 0) { 2773 start = syncer_maxdelay / 2; 2774 incr = syncer_maxdelay; 2775 } 2776 next = start; 2777 } 2778 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2779 mp->mnt_syncer = vp; 2780 return (0); 2781 } 2782 2783 /* 2784 * Do a lazy sync of the filesystem. 2785 */ 2786 static int 2787 sync_fsync(ap) 2788 struct vop_fsync_args /* { 2789 struct vnode *a_vp; 2790 struct ucred *a_cred; 2791 int a_waitfor; 2792 struct proc *a_p; 2793 } */ *ap; 2794 { 2795 struct vnode *syncvp = ap->a_vp; 2796 struct mount *mp = syncvp->v_mount; 2797 struct proc *p = ap->a_p; 2798 int asyncflag; 2799 2800 /* 2801 * We only need to do something if this is a lazy evaluation. 2802 */ 2803 if (ap->a_waitfor != MNT_LAZY) 2804 return (0); 2805 2806 /* 2807 * Move ourselves to the back of the sync list. 2808 */ 2809 vn_syncer_add_to_worklist(syncvp, syncdelay); 2810 2811 /* 2812 * Walk the list of vnodes pushing all that are dirty and 2813 * not already on the sync list. 2814 */ 2815 simple_lock(&mountlist_slock); 2816 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) { 2817 simple_unlock(&mountlist_slock); 2818 return (0); 2819 } 2820 asyncflag = mp->mnt_flag & MNT_ASYNC; 2821 mp->mnt_flag &= ~MNT_ASYNC; 2822 vfs_msync(mp, MNT_NOWAIT); 2823 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p); 2824 if (asyncflag) 2825 mp->mnt_flag |= MNT_ASYNC; 2826 vfs_unbusy(mp, p); 2827 return (0); 2828 } 2829 2830 /* 2831 * The syncer vnode is no referenced. 2832 */ 2833 static int 2834 sync_inactive(ap) 2835 struct vop_inactive_args /* { 2836 struct vnode *a_vp; 2837 struct proc *a_p; 2838 } */ *ap; 2839 { 2840 2841 vgone(ap->a_vp); 2842 return (0); 2843 } 2844 2845 /* 2846 * The syncer vnode is no longer needed and is being decommissioned. 2847 * 2848 * Modifications to the worklist must be protected at splbio(). 2849 */ 2850 static int 2851 sync_reclaim(ap) 2852 struct vop_reclaim_args /* { 2853 struct vnode *a_vp; 2854 } */ *ap; 2855 { 2856 struct vnode *vp = ap->a_vp; 2857 int s; 2858 2859 s = splbio(); 2860 vp->v_mount->mnt_syncer = NULL; 2861 if (vp->v_flag & VONWORKLST) { 2862 LIST_REMOVE(vp, v_synclist); 2863 vp->v_flag &= ~VONWORKLST; 2864 } 2865 splx(s); 2866 2867 return (0); 2868 } 2869 2870 /* 2871 * Print out a syncer vnode. 2872 */ 2873 static int 2874 sync_print(ap) 2875 struct vop_print_args /* { 2876 struct vnode *a_vp; 2877 } */ *ap; 2878 { 2879 struct vnode *vp = ap->a_vp; 2880 2881 printf("syncer vnode"); 2882 if (vp->v_vnlock != NULL) 2883 lockmgr_printinfo(vp->v_vnlock); 2884 printf("\n"); 2885 return (0); 2886 } 2887 2888 /* 2889 * extract the dev_t from a VBLK or VCHR 2890 */ 2891 dev_t 2892 vn_todev(vp) 2893 struct vnode *vp; 2894 { 2895 if (vp->v_type != VBLK && vp->v_type != VCHR) 2896 return (NODEV); 2897 return (vp->v_rdev); 2898 } 2899 2900 /* 2901 * Check if vnode represents a disk device 2902 */ 2903 int 2904 vn_isdisk(vp, errp) 2905 struct vnode *vp; 2906 int *errp; 2907 { 2908 if (vp->v_type != VBLK && vp->v_type != VCHR) { 2909 if (errp != NULL) 2910 *errp = ENOTBLK; 2911 return (0); 2912 } 2913 if (vp->v_rdev == NULL) { 2914 if (errp != NULL) 2915 *errp = ENXIO; 2916 return (0); 2917 } 2918 if (!devsw(vp->v_rdev)) { 2919 if (errp != NULL) 2920 *errp = ENXIO; 2921 return (0); 2922 } 2923 if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) { 2924 if (errp != NULL) 2925 *errp = ENOTBLK; 2926 return (0); 2927 } 2928 if (errp != NULL) 2929 *errp = 0; 2930 return (1); 2931 } 2932 2933 void 2934 NDFREE(ndp, flags) 2935 struct nameidata *ndp; 2936 const uint flags; 2937 { 2938 if (!(flags & NDF_NO_FREE_PNBUF) && 2939 (ndp->ni_cnd.cn_flags & HASBUF)) { 2940 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 2941 ndp->ni_cnd.cn_flags &= ~HASBUF; 2942 } 2943 if (!(flags & NDF_NO_DVP_UNLOCK) && 2944 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 2945 ndp->ni_dvp != ndp->ni_vp) 2946 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc); 2947 if (!(flags & NDF_NO_DVP_RELE) && 2948 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 2949 vrele(ndp->ni_dvp); 2950 ndp->ni_dvp = NULL; 2951 } 2952 if (!(flags & NDF_NO_VP_UNLOCK) && 2953 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 2954 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc); 2955 if (!(flags & NDF_NO_VP_RELE) && 2956 ndp->ni_vp) { 2957 vrele(ndp->ni_vp); 2958 ndp->ni_vp = NULL; 2959 } 2960 if (!(flags & NDF_NO_STARTDIR_RELE) && 2961 (ndp->ni_cnd.cn_flags & SAVESTART)) { 2962 vrele(ndp->ni_startdir); 2963 ndp->ni_startdir = NULL; 2964 } 2965 } 2966