1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 40 */ 41 42 /* 43 * External virtual filesystem routines 44 */ 45 #include "opt_ddb.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/buf.h> 50 #include <sys/conf.h> 51 #include <sys/dirent.h> 52 #include <sys/domain.h> 53 #include <sys/eventhandler.h> 54 #include <sys/fcntl.h> 55 #include <sys/kernel.h> 56 #include <sys/kthread.h> 57 #include <sys/malloc.h> 58 #include <sys/mbuf.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/syslog.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 static void vlruvp(struct vnode *vp); 88 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 89 90 enum vtype iftovt_tab[16] = { 91 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 92 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 93 }; 94 int vttoif_tab[9] = { 95 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 96 S_IFSOCK, S_IFIFO, S_IFMT, 97 }; 98 99 static TAILQ_HEAD(freelst, vnode) vnode_free_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 static int nameileafonly = 0; 117 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, ""); 118 119 #ifdef ENABLE_VFS_IOOPT 120 int vfs_ioopt = 0; 121 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 122 #endif 123 124 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */ 125 struct simplelock mountlist_slock; 126 struct simplelock mntvnode_slock; 127 int nfs_mount_type = -1; 128 #ifndef NULL_SIMPLELOCKS 129 static struct simplelock mntid_slock; 130 static struct simplelock vnode_free_list_slock; 131 static struct simplelock spechash_slock; 132 #endif 133 struct nfs_public nfs_pub; /* publicly exported FS */ 134 static vm_zone_t vnode_zone; 135 136 /* 137 * The workitem queue. 138 */ 139 #define SYNCER_MAXDELAY 32 140 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 141 time_t syncdelay = 30; /* max time to delay syncing data */ 142 time_t filedelay = 30; /* time to delay syncing files */ 143 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 144 time_t dirdelay = 29; /* time to delay syncing directories */ 145 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 146 time_t metadelay = 28; /* time to delay syncing metadata */ 147 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 148 static int rushjob; /* number of slots to run ASAP */ 149 static int stat_rush_requests; /* number of times I/O speeded up */ 150 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 151 152 static int syncer_delayno = 0; 153 static long syncer_mask; 154 LIST_HEAD(synclist, vnode); 155 static struct synclist *syncer_workitem_pending; 156 157 int desiredvnodes; 158 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 159 &desiredvnodes, 0, "Maximum number of vnodes"); 160 static int minvnodes; 161 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 162 &minvnodes, 0, "Minimum number of vnodes"); 163 static int vnlru_nowhere = 0; 164 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, 165 "Number of times the vnlru process ran without success"); 166 167 static void vfs_free_addrlist __P((struct netexport *nep)); 168 static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 169 static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 170 struct export_args *argp)); 171 172 /* 173 * Initialize the vnode management data structures. 174 */ 175 void 176 vntblinit() 177 { 178 179 desiredvnodes = maxproc + cnt.v_page_count / 4; 180 minvnodes = desiredvnodes / 4; 181 simple_lock_init(&mntvnode_slock); 182 simple_lock_init(&mntid_slock); 183 simple_lock_init(&spechash_slock); 184 TAILQ_INIT(&vnode_free_list); 185 simple_lock_init(&vnode_free_list_slock); 186 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 187 /* 188 * Initialize the filesystem syncer. 189 */ 190 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 191 &syncer_mask); 192 syncer_maxdelay = syncer_mask + 1; 193 } 194 195 /* 196 * Mark a mount point as busy. Used to synchronize access and to delay 197 * unmounting. Interlock is not released on failure. 198 */ 199 int 200 vfs_busy(mp, flags, interlkp, p) 201 struct mount *mp; 202 int flags; 203 struct simplelock *interlkp; 204 struct proc *p; 205 { 206 int lkflags; 207 208 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 209 if (flags & LK_NOWAIT) 210 return (ENOENT); 211 mp->mnt_kern_flag |= MNTK_MWAIT; 212 if (interlkp) { 213 simple_unlock(interlkp); 214 } 215 /* 216 * Since all busy locks are shared except the exclusive 217 * lock granted when unmounting, the only place that a 218 * wakeup needs to be done is at the release of the 219 * exclusive lock at the end of dounmount. 220 */ 221 tsleep((caddr_t)mp, PVFS, "vfs_busy", 0); 222 if (interlkp) { 223 simple_lock(interlkp); 224 } 225 return (ENOENT); 226 } 227 lkflags = LK_SHARED | LK_NOPAUSE; 228 if (interlkp) 229 lkflags |= LK_INTERLOCK; 230 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, p)) 231 panic("vfs_busy: unexpected lock failure"); 232 return (0); 233 } 234 235 /* 236 * Free a busy filesystem. 237 */ 238 void 239 vfs_unbusy(mp, p) 240 struct mount *mp; 241 struct proc *p; 242 { 243 244 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, p); 245 } 246 247 /* 248 * Lookup a filesystem type, and if found allocate and initialize 249 * a mount structure for it. 250 * 251 * Devname is usually updated by mount(8) after booting. 252 */ 253 int 254 vfs_rootmountalloc(fstypename, devname, mpp) 255 char *fstypename; 256 char *devname; 257 struct mount **mpp; 258 { 259 struct proc *p = curproc; /* XXX */ 260 struct vfsconf *vfsp; 261 struct mount *mp; 262 263 if (fstypename == NULL) 264 return (ENODEV); 265 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 266 if (!strcmp(vfsp->vfc_name, fstypename)) 267 break; 268 if (vfsp == NULL) 269 return (ENODEV); 270 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 271 bzero((char *)mp, (u_long)sizeof(struct mount)); 272 lockinit(&mp->mnt_lock, PVFS, "vfslock", VLKTIMEOUT, LK_NOPAUSE); 273 (void)vfs_busy(mp, LK_NOWAIT, 0, p); 274 TAILQ_INIT(&mp->mnt_nvnodelist); 275 TAILQ_INIT(&mp->mnt_reservedvnlist); 276 mp->mnt_nvnodelistsize = 0; 277 mp->mnt_vfc = vfsp; 278 mp->mnt_op = vfsp->vfc_vfsops; 279 mp->mnt_flag = MNT_RDONLY; 280 mp->mnt_vnodecovered = NULLVP; 281 vfsp->vfc_refcount++; 282 mp->mnt_iosize_max = DFLTPHYS; 283 mp->mnt_stat.f_type = vfsp->vfc_typenum; 284 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 285 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 286 mp->mnt_stat.f_mntonname[0] = '/'; 287 mp->mnt_stat.f_mntonname[1] = 0; 288 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 289 *mpp = mp; 290 return (0); 291 } 292 293 /* 294 * Find an appropriate filesystem to use for the root. If a filesystem 295 * has not been preselected, walk through the list of known filesystems 296 * trying those that have mountroot routines, and try them until one 297 * works or we have tried them all. 298 */ 299 #ifdef notdef /* XXX JH */ 300 int 301 lite2_vfs_mountroot() 302 { 303 struct vfsconf *vfsp; 304 extern int (*lite2_mountroot) __P((void)); 305 int error; 306 307 if (lite2_mountroot != NULL) 308 return ((*lite2_mountroot)()); 309 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 310 if (vfsp->vfc_mountroot == NULL) 311 continue; 312 if ((error = (*vfsp->vfc_mountroot)()) == 0) 313 return (0); 314 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 315 } 316 return (ENODEV); 317 } 318 #endif 319 320 /* 321 * Lookup a mount point by filesystem identifier. 322 */ 323 struct mount * 324 vfs_getvfs(fsid) 325 fsid_t *fsid; 326 { 327 register struct mount *mp; 328 329 simple_lock(&mountlist_slock); 330 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 331 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 332 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 333 simple_unlock(&mountlist_slock); 334 return (mp); 335 } 336 } 337 simple_unlock(&mountlist_slock); 338 return ((struct mount *) 0); 339 } 340 341 /* 342 * Get a new unique fsid. Try to make its val[0] unique, since this value 343 * will be used to create fake device numbers for stat(). Also try (but 344 * not so hard) make its val[0] unique mod 2^16, since some emulators only 345 * support 16-bit device numbers. We end up with unique val[0]'s for the 346 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 347 * 348 * Keep in mind that several mounts may be running in parallel. Starting 349 * the search one past where the previous search terminated is both a 350 * micro-optimization and a defense against returning the same fsid to 351 * different mounts. 352 */ 353 void 354 vfs_getnewfsid(mp) 355 struct mount *mp; 356 { 357 static u_int16_t mntid_base; 358 fsid_t tfsid; 359 int mtype; 360 361 simple_lock(&mntid_slock); 362 mtype = mp->mnt_vfc->vfc_typenum; 363 tfsid.val[1] = mtype; 364 mtype = (mtype & 0xFF) << 24; 365 for (;;) { 366 tfsid.val[0] = makeudev(255, 367 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 368 mntid_base++; 369 if (vfs_getvfs(&tfsid) == NULL) 370 break; 371 } 372 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 373 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 374 simple_unlock(&mntid_slock); 375 } 376 377 /* 378 * Knob to control the precision of file timestamps: 379 * 380 * 0 = seconds only; nanoseconds zeroed. 381 * 1 = seconds and nanoseconds, accurate within 1/HZ. 382 * 2 = seconds and nanoseconds, truncated to microseconds. 383 * >=3 = seconds and nanoseconds, maximum precision. 384 */ 385 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 386 387 static int timestamp_precision = TSP_SEC; 388 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 389 ×tamp_precision, 0, ""); 390 391 /* 392 * Get a current timestamp. 393 */ 394 void 395 vfs_timestamp(tsp) 396 struct timespec *tsp; 397 { 398 struct timeval tv; 399 400 switch (timestamp_precision) { 401 case TSP_SEC: 402 tsp->tv_sec = time_second; 403 tsp->tv_nsec = 0; 404 break; 405 case TSP_HZ: 406 getnanotime(tsp); 407 break; 408 case TSP_USEC: 409 microtime(&tv); 410 TIMEVAL_TO_TIMESPEC(&tv, tsp); 411 break; 412 case TSP_NSEC: 413 default: 414 nanotime(tsp); 415 break; 416 } 417 } 418 419 /* 420 * Set vnode attributes to VNOVAL 421 */ 422 void 423 vattr_null(vap) 424 register struct vattr *vap; 425 { 426 427 vap->va_type = VNON; 428 vap->va_size = VNOVAL; 429 vap->va_bytes = VNOVAL; 430 vap->va_mode = VNOVAL; 431 vap->va_nlink = VNOVAL; 432 vap->va_uid = VNOVAL; 433 vap->va_gid = VNOVAL; 434 vap->va_fsid = VNOVAL; 435 vap->va_fileid = VNOVAL; 436 vap->va_blocksize = VNOVAL; 437 vap->va_rdev = VNOVAL; 438 vap->va_atime.tv_sec = VNOVAL; 439 vap->va_atime.tv_nsec = VNOVAL; 440 vap->va_mtime.tv_sec = VNOVAL; 441 vap->va_mtime.tv_nsec = VNOVAL; 442 vap->va_ctime.tv_sec = VNOVAL; 443 vap->va_ctime.tv_nsec = VNOVAL; 444 vap->va_flags = VNOVAL; 445 vap->va_gen = VNOVAL; 446 vap->va_vaflags = 0; 447 } 448 449 /* 450 * This routine is called when we have too many vnodes. It attempts 451 * to free <count> vnodes and will potentially free vnodes that still 452 * have VM backing store (VM backing store is typically the cause 453 * of a vnode blowout so we want to do this). Therefore, this operation 454 * is not considered cheap. 455 * 456 * A number of conditions may prevent a vnode from being reclaimed. 457 * the buffer cache may have references on the vnode, a directory 458 * vnode may still have references due to the namei cache representing 459 * underlying files, or the vnode may be in active use. It is not 460 * desireable to reuse such vnodes. These conditions may cause the 461 * number of vnodes to reach some minimum value regardless of what 462 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 463 */ 464 static int 465 vlrureclaim(struct mount *mp) 466 { 467 struct vnode *vp; 468 int done; 469 int trigger; 470 int usevnodes; 471 int count; 472 473 /* 474 * Calculate the trigger point, don't allow user 475 * screwups to blow us up. This prevents us from 476 * recycling vnodes with lots of resident pages. We 477 * aren't trying to free memory, we are trying to 478 * free vnodes. 479 */ 480 usevnodes = desiredvnodes; 481 if (usevnodes <= 0) 482 usevnodes = 1; 483 trigger = cnt.v_page_count * 2 / usevnodes; 484 485 done = 0; 486 simple_lock(&mntvnode_slock); 487 count = mp->mnt_nvnodelistsize / 10 + 1; 488 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) { 489 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 490 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 491 492 if (vp->v_type != VNON && 493 vp->v_type != VBAD && 494 VMIGHTFREE(vp) && /* critical path opt */ 495 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) && 496 simple_lock_try(&vp->v_interlock) 497 ) { 498 simple_unlock(&mntvnode_slock); 499 if (VMIGHTFREE(vp)) { 500 vgonel(vp, curproc); 501 done++; 502 } else { 503 simple_unlock(&vp->v_interlock); 504 } 505 simple_lock(&mntvnode_slock); 506 } 507 --count; 508 } 509 simple_unlock(&mntvnode_slock); 510 return done; 511 } 512 513 /* 514 * Attempt to recycle vnodes in a context that is always safe to block. 515 * Calling vlrurecycle() from the bowels of file system code has some 516 * interesting deadlock problems. 517 */ 518 static struct proc *vnlruproc; 519 static int vnlruproc_sig; 520 521 static void 522 vnlru_proc(void) 523 { 524 struct mount *mp, *nmp; 525 int s; 526 int done; 527 struct proc *p = vnlruproc; 528 529 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, p, 530 SHUTDOWN_PRI_FIRST); 531 532 s = splbio(); 533 for (;;) { 534 kproc_suspend_loop(p); 535 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) { 536 vnlruproc_sig = 0; 537 wakeup(&vnlruproc_sig); 538 tsleep(vnlruproc, PVFS, "vlruwt", hz); 539 continue; 540 } 541 done = 0; 542 simple_lock(&mountlist_slock); 543 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 544 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 545 nmp = TAILQ_NEXT(mp, mnt_list); 546 continue; 547 } 548 done += vlrureclaim(mp); 549 simple_lock(&mountlist_slock); 550 nmp = TAILQ_NEXT(mp, mnt_list); 551 vfs_unbusy(mp, p); 552 } 553 simple_unlock(&mountlist_slock); 554 if (done == 0) { 555 vnlru_nowhere++; 556 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 557 } 558 } 559 splx(s); 560 } 561 562 static struct kproc_desc vnlru_kp = { 563 "vnlru", 564 vnlru_proc, 565 &vnlruproc 566 }; 567 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 568 569 /* 570 * Routines having to do with the management of the vnode table. 571 */ 572 extern vop_t **dead_vnodeop_p; 573 574 /* 575 * Return the next vnode from the free list. 576 */ 577 int 578 getnewvnode(tag, mp, vops, vpp) 579 enum vtagtype tag; 580 struct mount *mp; 581 vop_t **vops; 582 struct vnode **vpp; 583 { 584 int s; 585 struct proc *p = curproc; /* XXX */ 586 struct vnode *vp = NULL; 587 vm_object_t object; 588 589 s = splbio(); 590 591 /* 592 * Try to reuse vnodes if we hit the max. This situation only 593 * occurs in certain large-memory (2G+) situations. We cannot 594 * attempt to directly reclaim vnodes due to nasty recursion 595 * problems. 596 */ 597 while (numvnodes - freevnodes > desiredvnodes) { 598 if (vnlruproc_sig == 0) { 599 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 600 wakeup(vnlruproc); 601 } 602 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz); 603 } 604 605 606 /* 607 * Attempt to reuse a vnode already on the free list, allocating 608 * a new vnode if we can't find one or if we have not reached a 609 * good minimum for good LRU performance. 610 */ 611 simple_lock(&vnode_free_list_slock); 612 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) { 613 int count; 614 615 for (count = 0; count < freevnodes; count++) { 616 vp = TAILQ_FIRST(&vnode_free_list); 617 if (vp == NULL || vp->v_usecount) 618 panic("getnewvnode: free vnode isn't"); 619 620 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 621 if ((VOP_GETVOBJECT(vp, &object) == 0 && 622 (object->resident_page_count || object->ref_count)) || 623 !simple_lock_try(&vp->v_interlock)) { 624 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 625 vp = NULL; 626 continue; 627 } 628 if (LIST_FIRST(&vp->v_cache_src)) { 629 /* 630 * note: nameileafonly sysctl is temporary, 631 * for debugging only, and will eventually be 632 * removed. 633 */ 634 if (nameileafonly > 0) { 635 /* 636 * Do not reuse namei-cached directory 637 * vnodes that have cached 638 * subdirectories. 639 */ 640 if (cache_leaf_test(vp) < 0) { 641 simple_unlock(&vp->v_interlock); 642 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 643 vp = NULL; 644 continue; 645 } 646 } else if (nameileafonly < 0 || 647 vmiodirenable == 0) { 648 /* 649 * Do not reuse namei-cached directory 650 * vnodes if nameileafonly is -1 or 651 * if VMIO backing for directories is 652 * turned off (otherwise we reuse them 653 * too quickly). 654 */ 655 simple_unlock(&vp->v_interlock); 656 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 657 vp = NULL; 658 continue; 659 } 660 } 661 break; 662 } 663 } 664 665 if (vp) { 666 vp->v_flag |= VDOOMED; 667 vp->v_flag &= ~VFREE; 668 freevnodes--; 669 simple_unlock(&vnode_free_list_slock); 670 cache_purge(vp); 671 vp->v_lease = NULL; 672 if (vp->v_type != VBAD) { 673 vgonel(vp, p); 674 } else { 675 simple_unlock(&vp->v_interlock); 676 } 677 678 #ifdef INVARIANTS 679 { 680 int s; 681 682 if (vp->v_data) 683 panic("cleaned vnode isn't"); 684 s = splbio(); 685 if (vp->v_numoutput) 686 panic("Clean vnode has pending I/O's"); 687 splx(s); 688 } 689 #endif 690 vp->v_flag = 0; 691 vp->v_lastw = 0; 692 vp->v_lasta = 0; 693 vp->v_cstart = 0; 694 vp->v_clen = 0; 695 vp->v_socket = 0; 696 vp->v_writecount = 0; /* XXX */ 697 } else { 698 simple_unlock(&vnode_free_list_slock); 699 vp = (struct vnode *) zalloc(vnode_zone); 700 bzero((char *) vp, sizeof *vp); 701 simple_lock_init(&vp->v_interlock); 702 vp->v_dd = vp; 703 cache_purge(vp); 704 LIST_INIT(&vp->v_cache_src); 705 TAILQ_INIT(&vp->v_cache_dst); 706 numvnodes++; 707 } 708 709 TAILQ_INIT(&vp->v_cleanblkhd); 710 TAILQ_INIT(&vp->v_dirtyblkhd); 711 vp->v_type = VNON; 712 vp->v_tag = tag; 713 vp->v_op = vops; 714 insmntque(vp, mp); 715 *vpp = vp; 716 vp->v_usecount = 1; 717 vp->v_data = 0; 718 splx(s); 719 720 vfs_object_create(vp, p, p->p_ucred); 721 return (0); 722 } 723 724 /* 725 * Move a vnode from one mount queue to another. 726 */ 727 static void 728 insmntque(vp, mp) 729 register struct vnode *vp; 730 register struct mount *mp; 731 { 732 733 simple_lock(&mntvnode_slock); 734 /* 735 * Delete from old mount point vnode list, if on one. 736 */ 737 if (vp->v_mount != NULL) { 738 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0, 739 ("bad mount point vnode list size")); 740 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 741 vp->v_mount->mnt_nvnodelistsize--; 742 } 743 /* 744 * Insert into list of vnodes for the new mount point, if available. 745 */ 746 if ((vp->v_mount = mp) == NULL) { 747 simple_unlock(&mntvnode_slock); 748 return; 749 } 750 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 751 mp->mnt_nvnodelistsize++; 752 simple_unlock(&mntvnode_slock); 753 } 754 755 /* 756 * Update outstanding I/O count and do wakeup if requested. 757 */ 758 void 759 vwakeup(bp) 760 register struct buf *bp; 761 { 762 register struct vnode *vp; 763 764 bp->b_flags &= ~B_WRITEINPROG; 765 if ((vp = bp->b_vp)) { 766 vp->v_numoutput--; 767 if (vp->v_numoutput < 0) 768 panic("vwakeup: neg numoutput"); 769 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 770 vp->v_flag &= ~VBWAIT; 771 wakeup((caddr_t) &vp->v_numoutput); 772 } 773 } 774 } 775 776 /* 777 * Flush out and invalidate all buffers associated with a vnode. 778 * Called with the underlying object locked. 779 */ 780 int 781 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 782 register struct vnode *vp; 783 int flags; 784 struct ucred *cred; 785 struct proc *p; 786 int slpflag, slptimeo; 787 { 788 register struct buf *bp; 789 struct buf *nbp, *blist; 790 int s, error; 791 vm_object_t object; 792 793 if (flags & V_SAVE) { 794 s = splbio(); 795 while (vp->v_numoutput) { 796 vp->v_flag |= VBWAIT; 797 error = tsleep((caddr_t)&vp->v_numoutput, 798 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo); 799 if (error) { 800 splx(s); 801 return (error); 802 } 803 } 804 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 805 splx(s); 806 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0) 807 return (error); 808 s = splbio(); 809 if (vp->v_numoutput > 0 || 810 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 811 panic("vinvalbuf: dirty bufs"); 812 } 813 splx(s); 814 } 815 s = splbio(); 816 for (;;) { 817 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 818 if (!blist) 819 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 820 if (!blist) 821 break; 822 823 for (bp = blist; bp; bp = nbp) { 824 nbp = TAILQ_NEXT(bp, b_vnbufs); 825 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 826 error = BUF_TIMELOCK(bp, 827 LK_EXCLUSIVE | LK_SLEEPFAIL, 828 "vinvalbuf", slpflag, slptimeo); 829 if (error == ENOLCK) 830 break; 831 splx(s); 832 return (error); 833 } 834 /* 835 * XXX Since there are no node locks for NFS, I 836 * believe there is a slight chance that a delayed 837 * write will occur while sleeping just above, so 838 * check for it. Note that vfs_bio_awrite expects 839 * buffers to reside on a queue, while VOP_BWRITE and 840 * brelse do not. 841 */ 842 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 843 (flags & V_SAVE)) { 844 845 if (bp->b_vp == vp) { 846 if (bp->b_flags & B_CLUSTEROK) { 847 BUF_UNLOCK(bp); 848 vfs_bio_awrite(bp); 849 } else { 850 bremfree(bp); 851 bp->b_flags |= B_ASYNC; 852 VOP_BWRITE(bp->b_vp, bp); 853 } 854 } else { 855 bremfree(bp); 856 (void) VOP_BWRITE(bp->b_vp, bp); 857 } 858 break; 859 } 860 bremfree(bp); 861 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 862 bp->b_flags &= ~B_ASYNC; 863 brelse(bp); 864 } 865 } 866 867 /* 868 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 869 * have write I/O in-progress but if there is a VM object then the 870 * VM object can also have read-I/O in-progress. 871 */ 872 do { 873 while (vp->v_numoutput > 0) { 874 vp->v_flag |= VBWAIT; 875 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 876 } 877 if (VOP_GETVOBJECT(vp, &object) == 0) { 878 while (object->paging_in_progress) 879 vm_object_pip_sleep(object, "vnvlbx"); 880 } 881 } while (vp->v_numoutput > 0); 882 883 splx(s); 884 885 /* 886 * Destroy the copy in the VM cache, too. 887 */ 888 simple_lock(&vp->v_interlock); 889 if (VOP_GETVOBJECT(vp, &object) == 0) { 890 vm_object_page_remove(object, 0, 0, 891 (flags & V_SAVE) ? TRUE : FALSE); 892 } 893 simple_unlock(&vp->v_interlock); 894 895 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 896 panic("vinvalbuf: flush failed"); 897 return (0); 898 } 899 900 /* 901 * Truncate a file's buffer and pages to a specified length. This 902 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 903 * sync activity. 904 */ 905 int 906 vtruncbuf(vp, cred, p, length, blksize) 907 register struct vnode *vp; 908 struct ucred *cred; 909 struct proc *p; 910 off_t length; 911 int blksize; 912 { 913 register struct buf *bp; 914 struct buf *nbp; 915 int s, anyfreed; 916 int trunclbn; 917 918 /* 919 * Round up to the *next* lbn. 920 */ 921 trunclbn = (length + blksize - 1) / blksize; 922 923 s = splbio(); 924 restart: 925 anyfreed = 1; 926 for (;anyfreed;) { 927 anyfreed = 0; 928 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 929 nbp = TAILQ_NEXT(bp, b_vnbufs); 930 if (bp->b_lblkno >= trunclbn) { 931 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 932 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 933 goto restart; 934 } else { 935 bremfree(bp); 936 bp->b_flags |= (B_INVAL | B_RELBUF); 937 bp->b_flags &= ~B_ASYNC; 938 brelse(bp); 939 anyfreed = 1; 940 } 941 if (nbp && 942 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 943 (nbp->b_vp != vp) || 944 (nbp->b_flags & B_DELWRI))) { 945 goto restart; 946 } 947 } 948 } 949 950 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 951 nbp = TAILQ_NEXT(bp, b_vnbufs); 952 if (bp->b_lblkno >= trunclbn) { 953 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 954 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 955 goto restart; 956 } else { 957 bremfree(bp); 958 bp->b_flags |= (B_INVAL | B_RELBUF); 959 bp->b_flags &= ~B_ASYNC; 960 brelse(bp); 961 anyfreed = 1; 962 } 963 if (nbp && 964 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 965 (nbp->b_vp != vp) || 966 (nbp->b_flags & B_DELWRI) == 0)) { 967 goto restart; 968 } 969 } 970 } 971 } 972 973 if (length > 0) { 974 restartsync: 975 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 976 nbp = TAILQ_NEXT(bp, b_vnbufs); 977 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 978 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 979 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 980 goto restart; 981 } else { 982 bremfree(bp); 983 if (bp->b_vp == vp) { 984 bp->b_flags |= B_ASYNC; 985 } else { 986 bp->b_flags &= ~B_ASYNC; 987 } 988 VOP_BWRITE(bp->b_vp, bp); 989 } 990 goto restartsync; 991 } 992 993 } 994 } 995 996 while (vp->v_numoutput > 0) { 997 vp->v_flag |= VBWAIT; 998 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0); 999 } 1000 1001 splx(s); 1002 1003 vnode_pager_setsize(vp, length); 1004 1005 return (0); 1006 } 1007 1008 /* 1009 * Associate a buffer with a vnode. 1010 */ 1011 void 1012 bgetvp(vp, bp) 1013 register struct vnode *vp; 1014 register struct buf *bp; 1015 { 1016 int s; 1017 1018 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 1019 1020 vhold(vp); 1021 bp->b_vp = vp; 1022 bp->b_dev = vn_todev(vp); 1023 /* 1024 * Insert onto list for new vnode. 1025 */ 1026 s = splbio(); 1027 bp->b_xflags |= BX_VNCLEAN; 1028 bp->b_xflags &= ~BX_VNDIRTY; 1029 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 1030 splx(s); 1031 } 1032 1033 /* 1034 * Disassociate a buffer from a vnode. 1035 */ 1036 void 1037 brelvp(bp) 1038 register struct buf *bp; 1039 { 1040 struct vnode *vp; 1041 struct buflists *listheadp; 1042 int s; 1043 1044 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1045 1046 /* 1047 * Delete from old vnode list, if on one. 1048 */ 1049 vp = bp->b_vp; 1050 s = splbio(); 1051 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1052 if (bp->b_xflags & BX_VNDIRTY) 1053 listheadp = &vp->v_dirtyblkhd; 1054 else 1055 listheadp = &vp->v_cleanblkhd; 1056 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1057 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1058 } 1059 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1060 vp->v_flag &= ~VONWORKLST; 1061 LIST_REMOVE(vp, v_synclist); 1062 } 1063 splx(s); 1064 bp->b_vp = (struct vnode *) 0; 1065 vdrop(vp); 1066 } 1067 1068 /* 1069 * The workitem queue. 1070 * 1071 * It is useful to delay writes of file data and filesystem metadata 1072 * for tens of seconds so that quickly created and deleted files need 1073 * not waste disk bandwidth being created and removed. To realize this, 1074 * we append vnodes to a "workitem" queue. When running with a soft 1075 * updates implementation, most pending metadata dependencies should 1076 * not wait for more than a few seconds. Thus, mounted on block devices 1077 * are delayed only about a half the time that file data is delayed. 1078 * Similarly, directory updates are more critical, so are only delayed 1079 * about a third the time that file data is delayed. Thus, there are 1080 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 1081 * one each second (driven off the filesystem syncer process). The 1082 * syncer_delayno variable indicates the next queue that is to be processed. 1083 * Items that need to be processed soon are placed in this queue: 1084 * 1085 * syncer_workitem_pending[syncer_delayno] 1086 * 1087 * A delay of fifteen seconds is done by placing the request fifteen 1088 * entries later in the queue: 1089 * 1090 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 1091 * 1092 */ 1093 1094 /* 1095 * Add an item to the syncer work queue. 1096 */ 1097 static void 1098 vn_syncer_add_to_worklist(struct vnode *vp, int delay) 1099 { 1100 int s, slot; 1101 1102 s = splbio(); 1103 1104 if (vp->v_flag & VONWORKLST) { 1105 LIST_REMOVE(vp, v_synclist); 1106 } 1107 1108 if (delay > syncer_maxdelay - 2) 1109 delay = syncer_maxdelay - 2; 1110 slot = (syncer_delayno + delay) & syncer_mask; 1111 1112 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 1113 vp->v_flag |= VONWORKLST; 1114 splx(s); 1115 } 1116 1117 struct proc *updateproc; 1118 static void sched_sync __P((void)); 1119 static struct kproc_desc up_kp = { 1120 "syncer", 1121 sched_sync, 1122 &updateproc 1123 }; 1124 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1125 1126 /* 1127 * System filesystem synchronizer daemon. 1128 */ 1129 void 1130 sched_sync(void) 1131 { 1132 struct synclist *slp; 1133 struct vnode *vp; 1134 long starttime; 1135 int s; 1136 struct proc *p = updateproc; 1137 1138 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, p, 1139 SHUTDOWN_PRI_LAST); 1140 1141 for (;;) { 1142 kproc_suspend_loop(p); 1143 1144 starttime = time_second; 1145 1146 /* 1147 * Push files whose dirty time has expired. Be careful 1148 * of interrupt race on slp queue. 1149 */ 1150 s = splbio(); 1151 slp = &syncer_workitem_pending[syncer_delayno]; 1152 syncer_delayno += 1; 1153 if (syncer_delayno == syncer_maxdelay) 1154 syncer_delayno = 0; 1155 splx(s); 1156 1157 while ((vp = LIST_FIRST(slp)) != NULL) { 1158 if (VOP_ISLOCKED(vp, NULL) == 0) { 1159 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p); 1160 (void) VOP_FSYNC(vp, p->p_ucred, MNT_LAZY, p); 1161 VOP_UNLOCK(vp, 0, p); 1162 } 1163 s = splbio(); 1164 if (LIST_FIRST(slp) == vp) { 1165 /* 1166 * Note: v_tag VT_VFS vps can remain on the 1167 * worklist too with no dirty blocks, but 1168 * since sync_fsync() moves it to a different 1169 * slot we are safe. 1170 */ 1171 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1172 !vn_isdisk(vp, NULL)) 1173 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1174 /* 1175 * Put us back on the worklist. The worklist 1176 * routine will remove us from our current 1177 * position and then add us back in at a later 1178 * position. 1179 */ 1180 vn_syncer_add_to_worklist(vp, syncdelay); 1181 } 1182 splx(s); 1183 } 1184 1185 /* 1186 * Do soft update processing. 1187 */ 1188 if (bioops.io_sync) 1189 (*bioops.io_sync)(NULL); 1190 1191 /* 1192 * The variable rushjob allows the kernel to speed up the 1193 * processing of the filesystem syncer process. A rushjob 1194 * value of N tells the filesystem syncer to process the next 1195 * N seconds worth of work on its queue ASAP. Currently rushjob 1196 * is used by the soft update code to speed up the filesystem 1197 * syncer process when the incore state is getting so far 1198 * ahead of the disk that the kernel memory pool is being 1199 * threatened with exhaustion. 1200 */ 1201 if (rushjob > 0) { 1202 rushjob -= 1; 1203 continue; 1204 } 1205 /* 1206 * If it has taken us less than a second to process the 1207 * current work, then wait. Otherwise start right over 1208 * again. We can still lose time if any single round 1209 * takes more than two seconds, but it does not really 1210 * matter as we are just trying to generally pace the 1211 * filesystem activity. 1212 */ 1213 if (time_second == starttime) 1214 tsleep(&lbolt, PPAUSE, "syncer", 0); 1215 } 1216 } 1217 1218 /* 1219 * Request the syncer daemon to speed up its work. 1220 * We never push it to speed up more than half of its 1221 * normal turn time, otherwise it could take over the cpu. 1222 */ 1223 int 1224 speedup_syncer() 1225 { 1226 int s; 1227 1228 s = splhigh(); 1229 if (updateproc->p_wchan == &lbolt) 1230 setrunnable(updateproc); 1231 splx(s); 1232 if (rushjob < syncdelay / 2) { 1233 rushjob += 1; 1234 stat_rush_requests += 1; 1235 return (1); 1236 } 1237 return(0); 1238 } 1239 1240 /* 1241 * Associate a p-buffer with a vnode. 1242 * 1243 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1244 * with the buffer. i.e. the bp has not been linked into the vnode or 1245 * ref-counted. 1246 */ 1247 void 1248 pbgetvp(vp, bp) 1249 register struct vnode *vp; 1250 register struct buf *bp; 1251 { 1252 1253 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1254 1255 bp->b_vp = vp; 1256 bp->b_flags |= B_PAGING; 1257 bp->b_dev = vn_todev(vp); 1258 } 1259 1260 /* 1261 * Disassociate a p-buffer from a vnode. 1262 */ 1263 void 1264 pbrelvp(bp) 1265 register struct buf *bp; 1266 { 1267 1268 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1269 1270 /* XXX REMOVE ME */ 1271 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 1272 panic( 1273 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1274 bp, 1275 (int)bp->b_flags 1276 ); 1277 } 1278 bp->b_vp = (struct vnode *) 0; 1279 bp->b_flags &= ~B_PAGING; 1280 } 1281 1282 void 1283 pbreassignbuf(bp, newvp) 1284 struct buf *bp; 1285 struct vnode *newvp; 1286 { 1287 if ((bp->b_flags & B_PAGING) == 0) { 1288 panic( 1289 "pbreassignbuf() on non phys bp %p", 1290 bp 1291 ); 1292 } 1293 bp->b_vp = newvp; 1294 } 1295 1296 /* 1297 * Reassign a buffer from one vnode to another. 1298 * Used to assign file specific control information 1299 * (indirect blocks) to the vnode to which they belong. 1300 */ 1301 void 1302 reassignbuf(bp, newvp) 1303 register struct buf *bp; 1304 register struct vnode *newvp; 1305 { 1306 struct buflists *listheadp; 1307 int delay; 1308 int s; 1309 1310 if (newvp == NULL) { 1311 printf("reassignbuf: NULL"); 1312 return; 1313 } 1314 ++reassignbufcalls; 1315 1316 /* 1317 * B_PAGING flagged buffers cannot be reassigned because their vp 1318 * is not fully linked in. 1319 */ 1320 if (bp->b_flags & B_PAGING) 1321 panic("cannot reassign paging buffer"); 1322 1323 s = splbio(); 1324 /* 1325 * Delete from old vnode list, if on one. 1326 */ 1327 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1328 if (bp->b_xflags & BX_VNDIRTY) 1329 listheadp = &bp->b_vp->v_dirtyblkhd; 1330 else 1331 listheadp = &bp->b_vp->v_cleanblkhd; 1332 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1333 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1334 if (bp->b_vp != newvp) { 1335 vdrop(bp->b_vp); 1336 bp->b_vp = NULL; /* for clarification */ 1337 } 1338 } 1339 /* 1340 * If dirty, put on list of dirty buffers; otherwise insert onto list 1341 * of clean buffers. 1342 */ 1343 if (bp->b_flags & B_DELWRI) { 1344 struct buf *tbp; 1345 1346 listheadp = &newvp->v_dirtyblkhd; 1347 if ((newvp->v_flag & VONWORKLST) == 0) { 1348 switch (newvp->v_type) { 1349 case VDIR: 1350 delay = dirdelay; 1351 break; 1352 case VCHR: 1353 case VBLK: 1354 if (newvp->v_specmountpoint != NULL) { 1355 delay = metadelay; 1356 break; 1357 } 1358 /* fall through */ 1359 default: 1360 delay = filedelay; 1361 } 1362 vn_syncer_add_to_worklist(newvp, delay); 1363 } 1364 bp->b_xflags |= BX_VNDIRTY; 1365 tbp = TAILQ_FIRST(listheadp); 1366 if (tbp == NULL || 1367 bp->b_lblkno == 0 || 1368 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1369 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1370 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1371 ++reassignbufsortgood; 1372 } else if (bp->b_lblkno < 0) { 1373 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1374 ++reassignbufsortgood; 1375 } else if (reassignbufmethod == 1) { 1376 /* 1377 * New sorting algorithm, only handle sequential case, 1378 * otherwise append to end (but before metadata) 1379 */ 1380 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1381 (tbp->b_xflags & BX_VNDIRTY)) { 1382 /* 1383 * Found the best place to insert the buffer 1384 */ 1385 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1386 ++reassignbufsortgood; 1387 } else { 1388 /* 1389 * Missed, append to end, but before meta-data. 1390 * We know that the head buffer in the list is 1391 * not meta-data due to prior conditionals. 1392 * 1393 * Indirect effects: NFS second stage write 1394 * tends to wind up here, giving maximum 1395 * distance between the unstable write and the 1396 * commit rpc. 1397 */ 1398 tbp = TAILQ_LAST(listheadp, buflists); 1399 while (tbp && tbp->b_lblkno < 0) 1400 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1401 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1402 ++reassignbufsortbad; 1403 } 1404 } else { 1405 /* 1406 * Old sorting algorithm, scan queue and insert 1407 */ 1408 struct buf *ttbp; 1409 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1410 (ttbp->b_lblkno < bp->b_lblkno)) { 1411 ++reassignbufloops; 1412 tbp = ttbp; 1413 } 1414 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1415 } 1416 } else { 1417 bp->b_xflags |= BX_VNCLEAN; 1418 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1419 if ((newvp->v_flag & VONWORKLST) && 1420 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1421 newvp->v_flag &= ~VONWORKLST; 1422 LIST_REMOVE(newvp, v_synclist); 1423 } 1424 } 1425 if (bp->b_vp != newvp) { 1426 bp->b_vp = newvp; 1427 vhold(bp->b_vp); 1428 } 1429 splx(s); 1430 } 1431 1432 /* 1433 * Create a vnode for a block device. 1434 * Used for mounting the root file system. 1435 */ 1436 int 1437 bdevvp(dev, vpp) 1438 dev_t dev; 1439 struct vnode **vpp; 1440 { 1441 register struct vnode *vp; 1442 struct vnode *nvp; 1443 int error; 1444 1445 if (dev == NODEV) { 1446 *vpp = NULLVP; 1447 return (ENXIO); 1448 } 1449 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1450 if (error) { 1451 *vpp = NULLVP; 1452 return (error); 1453 } 1454 vp = nvp; 1455 vp->v_type = VBLK; 1456 addalias(vp, dev); 1457 *vpp = vp; 1458 return (0); 1459 } 1460 1461 /* 1462 * Add vnode to the alias list hung off the dev_t. 1463 * 1464 * The reason for this gunk is that multiple vnodes can reference 1465 * the same physical device, so checking vp->v_usecount to see 1466 * how many users there are is inadequate; the v_usecount for 1467 * the vnodes need to be accumulated. vcount() does that. 1468 */ 1469 void 1470 addaliasu(nvp, nvp_rdev) 1471 struct vnode *nvp; 1472 udev_t nvp_rdev; 1473 { 1474 1475 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1476 panic("addaliasu on non-special vnode"); 1477 addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0)); 1478 } 1479 1480 void 1481 addalias(nvp, dev) 1482 struct vnode *nvp; 1483 dev_t dev; 1484 { 1485 1486 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1487 panic("addalias on non-special vnode"); 1488 1489 nvp->v_rdev = dev; 1490 simple_lock(&spechash_slock); 1491 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1492 simple_unlock(&spechash_slock); 1493 } 1494 1495 /* 1496 * Grab a particular vnode from the free list, increment its 1497 * reference count and lock it. The vnode lock bit is set if the 1498 * vnode is being eliminated in vgone. The process is awakened 1499 * when the transition is completed, and an error returned to 1500 * indicate that the vnode is no longer usable (possibly having 1501 * been changed to a new file system type). 1502 */ 1503 int 1504 vget(vp, flags, p) 1505 register struct vnode *vp; 1506 int flags; 1507 struct proc *p; 1508 { 1509 int error; 1510 1511 /* 1512 * If the vnode is in the process of being cleaned out for 1513 * another use, we wait for the cleaning to finish and then 1514 * return failure. Cleaning is determined by checking that 1515 * the VXLOCK flag is set. 1516 */ 1517 if ((flags & LK_INTERLOCK) == 0) { 1518 simple_lock(&vp->v_interlock); 1519 } 1520 if (vp->v_flag & VXLOCK) { 1521 if (vp->v_vxproc == curproc) { 1522 #if 0 1523 /* this can now occur in normal operation */ 1524 log(LOG_INFO, "VXLOCK interlock avoided\n"); 1525 #endif 1526 } else { 1527 vp->v_flag |= VXWANT; 1528 simple_unlock(&vp->v_interlock); 1529 tsleep((caddr_t)vp, PINOD, "vget", 0); 1530 return (ENOENT); 1531 } 1532 } 1533 1534 vp->v_usecount++; 1535 1536 if (VSHOULDBUSY(vp)) 1537 vbusy(vp); 1538 if (flags & LK_TYPE_MASK) { 1539 if ((error = vn_lock(vp, flags | LK_INTERLOCK, p)) != 0) { 1540 /* 1541 * must expand vrele here because we do not want 1542 * to call VOP_INACTIVE if the reference count 1543 * drops back to zero since it was never really 1544 * active. We must remove it from the free list 1545 * before sleeping so that multiple processes do 1546 * not try to recycle it. 1547 */ 1548 simple_lock(&vp->v_interlock); 1549 vp->v_usecount--; 1550 if (VSHOULDFREE(vp)) 1551 vfree(vp); 1552 else 1553 vlruvp(vp); 1554 simple_unlock(&vp->v_interlock); 1555 } 1556 return (error); 1557 } 1558 simple_unlock(&vp->v_interlock); 1559 return (0); 1560 } 1561 1562 void 1563 vref(struct vnode *vp) 1564 { 1565 simple_lock(&vp->v_interlock); 1566 vp->v_usecount++; 1567 simple_unlock(&vp->v_interlock); 1568 } 1569 1570 /* 1571 * Vnode put/release. 1572 * If count drops to zero, call inactive routine and return to freelist. 1573 */ 1574 void 1575 vrele(vp) 1576 struct vnode *vp; 1577 { 1578 struct proc *p = curproc; /* XXX */ 1579 1580 KASSERT(vp != NULL, ("vrele: null vp")); 1581 1582 simple_lock(&vp->v_interlock); 1583 1584 if (vp->v_usecount > 1) { 1585 1586 vp->v_usecount--; 1587 simple_unlock(&vp->v_interlock); 1588 1589 return; 1590 } 1591 1592 if (vp->v_usecount == 1) { 1593 vp->v_usecount--; 1594 /* 1595 * We must call VOP_INACTIVE with the node locked. 1596 * If we are doing a vpu, the node is already locked, 1597 * but, in the case of vrele, we must explicitly lock 1598 * the vnode before calling VOP_INACTIVE 1599 */ 1600 1601 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, p) == 0) 1602 VOP_INACTIVE(vp, p); 1603 if (VSHOULDFREE(vp)) 1604 vfree(vp); 1605 else 1606 vlruvp(vp); 1607 } else { 1608 #ifdef DIAGNOSTIC 1609 vprint("vrele: negative ref count", vp); 1610 simple_unlock(&vp->v_interlock); 1611 #endif 1612 panic("vrele: negative ref cnt"); 1613 } 1614 } 1615 1616 void 1617 vput(vp) 1618 struct vnode *vp; 1619 { 1620 struct proc *p = curproc; /* XXX */ 1621 1622 KASSERT(vp != NULL, ("vput: null vp")); 1623 1624 simple_lock(&vp->v_interlock); 1625 1626 if (vp->v_usecount > 1) { 1627 vp->v_usecount--; 1628 VOP_UNLOCK(vp, LK_INTERLOCK, p); 1629 return; 1630 } 1631 1632 if (vp->v_usecount == 1) { 1633 vp->v_usecount--; 1634 /* 1635 * We must call VOP_INACTIVE with the node locked. 1636 * If we are doing a vpu, the node is already locked, 1637 * so we just need to release the vnode mutex. 1638 */ 1639 simple_unlock(&vp->v_interlock); 1640 VOP_INACTIVE(vp, p); 1641 if (VSHOULDFREE(vp)) 1642 vfree(vp); 1643 else 1644 vlruvp(vp); 1645 } else { 1646 #ifdef DIAGNOSTIC 1647 vprint("vput: negative ref count", vp); 1648 #endif 1649 panic("vput: negative ref cnt"); 1650 } 1651 } 1652 1653 /* 1654 * Somebody doesn't want the vnode recycled. 1655 */ 1656 void 1657 vhold(vp) 1658 register struct vnode *vp; 1659 { 1660 int s; 1661 1662 s = splbio(); 1663 vp->v_holdcnt++; 1664 if (VSHOULDBUSY(vp)) 1665 vbusy(vp); 1666 splx(s); 1667 } 1668 1669 /* 1670 * One less who cares about this vnode. 1671 */ 1672 void 1673 vdrop(vp) 1674 register struct vnode *vp; 1675 { 1676 int s; 1677 1678 s = splbio(); 1679 if (vp->v_holdcnt <= 0) 1680 panic("vdrop: holdcnt"); 1681 vp->v_holdcnt--; 1682 if (VSHOULDFREE(vp)) 1683 vfree(vp); 1684 splx(s); 1685 } 1686 1687 /* 1688 * Remove any vnodes in the vnode table belonging to mount point mp. 1689 * 1690 * If FORCECLOSE is not specified, there should not be any active ones, 1691 * return error if any are found (nb: this is a user error, not a 1692 * system error). If FORCECLOSE is specified, detach any active vnodes 1693 * that are found. 1694 * 1695 * If WRITECLOSE is set, only flush out regular file vnodes open for 1696 * writing. 1697 * 1698 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 1699 * 1700 * `rootrefs' specifies the base reference count for the root vnode 1701 * of this filesystem. The root vnode is considered busy if its 1702 * v_usecount exceeds this value. On a successful return, vflush() 1703 * will call vrele() on the root vnode exactly rootrefs times. 1704 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 1705 * be zero. 1706 */ 1707 #ifdef DIAGNOSTIC 1708 static int busyprt = 0; /* print out busy vnodes */ 1709 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1710 #endif 1711 1712 int 1713 vflush(mp, rootrefs, flags) 1714 struct mount *mp; 1715 int rootrefs; 1716 int flags; 1717 { 1718 struct proc *p = curproc; /* XXX */ 1719 struct vnode *vp, *nvp, *rootvp = NULL; 1720 struct vattr vattr; 1721 int busy = 0, error; 1722 1723 if (rootrefs > 0) { 1724 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1725 ("vflush: bad args")); 1726 /* 1727 * Get the filesystem root vnode. We can vput() it 1728 * immediately, since with rootrefs > 0, it won't go away. 1729 */ 1730 if ((error = VFS_ROOT(mp, &rootvp)) != 0) 1731 return (error); 1732 vput(rootvp); 1733 } 1734 simple_lock(&mntvnode_slock); 1735 loop: 1736 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) { 1737 /* 1738 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1739 * Start over if it has (it won't be on the list anymore). 1740 */ 1741 if (vp->v_mount != mp) 1742 goto loop; 1743 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1744 1745 simple_lock(&vp->v_interlock); 1746 /* 1747 * Skip over a vnodes marked VSYSTEM. 1748 */ 1749 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1750 simple_unlock(&vp->v_interlock); 1751 continue; 1752 } 1753 /* 1754 * If WRITECLOSE is set, flush out unlinked but still open 1755 * files (even if open only for reading) and regular file 1756 * vnodes open for writing. 1757 */ 1758 if ((flags & WRITECLOSE) && 1759 (vp->v_type == VNON || 1760 (VOP_GETATTR(vp, &vattr, p->p_ucred, p) == 0 && 1761 vattr.va_nlink > 0)) && 1762 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1763 simple_unlock(&vp->v_interlock); 1764 continue; 1765 } 1766 1767 /* 1768 * With v_usecount == 0, all we need to do is clear out the 1769 * vnode data structures and we are done. 1770 */ 1771 if (vp->v_usecount == 0) { 1772 simple_unlock(&mntvnode_slock); 1773 vgonel(vp, p); 1774 simple_lock(&mntvnode_slock); 1775 continue; 1776 } 1777 1778 /* 1779 * If FORCECLOSE is set, forcibly close the vnode. For block 1780 * or character devices, revert to an anonymous device. For 1781 * all other files, just kill them. 1782 */ 1783 if (flags & FORCECLOSE) { 1784 simple_unlock(&mntvnode_slock); 1785 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1786 vgonel(vp, p); 1787 } else { 1788 vclean(vp, 0, p); 1789 vp->v_op = spec_vnodeop_p; 1790 insmntque(vp, (struct mount *) 0); 1791 } 1792 simple_lock(&mntvnode_slock); 1793 continue; 1794 } 1795 #ifdef DIAGNOSTIC 1796 if (busyprt) 1797 vprint("vflush: busy vnode", vp); 1798 #endif 1799 simple_unlock(&vp->v_interlock); 1800 busy++; 1801 } 1802 simple_unlock(&mntvnode_slock); 1803 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1804 /* 1805 * If just the root vnode is busy, and if its refcount 1806 * is equal to `rootrefs', then go ahead and kill it. 1807 */ 1808 simple_lock(&rootvp->v_interlock); 1809 KASSERT(busy > 0, ("vflush: not busy")); 1810 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs")); 1811 if (busy == 1 && rootvp->v_usecount == rootrefs) { 1812 vgonel(rootvp, p); 1813 busy = 0; 1814 } else 1815 simple_unlock(&rootvp->v_interlock); 1816 } 1817 if (busy) 1818 return (EBUSY); 1819 for (; rootrefs > 0; rootrefs--) 1820 vrele(rootvp); 1821 return (0); 1822 } 1823 1824 /* 1825 * We do not want to recycle the vnode too quickly. 1826 * 1827 * XXX we can't move vp's around the nvnodelist without really screwing 1828 * up the efficiency of filesystem SYNC and friends. This code is 1829 * disabled until we fix the syncing code's scanning algorithm. 1830 */ 1831 static void 1832 vlruvp(struct vnode *vp) 1833 { 1834 #if 0 1835 struct mount *mp; 1836 1837 if ((mp = vp->v_mount) != NULL) { 1838 simple_lock(&mntvnode_slock); 1839 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1840 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1841 simple_unlock(&mntvnode_slock); 1842 } 1843 #endif 1844 } 1845 1846 /* 1847 * Disassociate the underlying file system from a vnode. 1848 */ 1849 static void 1850 vclean(vp, flags, p) 1851 struct vnode *vp; 1852 int flags; 1853 struct proc *p; 1854 { 1855 int active; 1856 1857 /* 1858 * Check to see if the vnode is in use. If so we have to reference it 1859 * before we clean it out so that its count cannot fall to zero and 1860 * generate a race against ourselves to recycle it. 1861 */ 1862 if ((active = vp->v_usecount)) 1863 vp->v_usecount++; 1864 1865 /* 1866 * Prevent the vnode from being recycled or brought into use while we 1867 * clean it out. 1868 */ 1869 if (vp->v_flag & VXLOCK) 1870 panic("vclean: deadlock"); 1871 vp->v_flag |= VXLOCK; 1872 vp->v_vxproc = curproc; 1873 /* 1874 * Even if the count is zero, the VOP_INACTIVE routine may still 1875 * have the object locked while it cleans it out. The VOP_LOCK 1876 * ensures that the VOP_INACTIVE routine is done with its work. 1877 * For active vnodes, it ensures that no other activity can 1878 * occur while the underlying object is being cleaned out. 1879 */ 1880 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, p); 1881 1882 /* 1883 * Clean out any buffers associated with the vnode. 1884 */ 1885 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1886 1887 VOP_DESTROYVOBJECT(vp); 1888 1889 /* 1890 * If purging an active vnode, it must be closed and 1891 * deactivated before being reclaimed. Note that the 1892 * VOP_INACTIVE will unlock the vnode. 1893 */ 1894 if (active) { 1895 if (flags & DOCLOSE) 1896 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p); 1897 VOP_INACTIVE(vp, p); 1898 } else { 1899 /* 1900 * Any other processes trying to obtain this lock must first 1901 * wait for VXLOCK to clear, then call the new lock operation. 1902 */ 1903 VOP_UNLOCK(vp, 0, p); 1904 } 1905 /* 1906 * Reclaim the vnode. 1907 */ 1908 if (VOP_RECLAIM(vp, p)) 1909 panic("vclean: cannot reclaim"); 1910 1911 if (active) { 1912 /* 1913 * Inline copy of vrele() since VOP_INACTIVE 1914 * has already been called. 1915 */ 1916 simple_lock(&vp->v_interlock); 1917 if (--vp->v_usecount <= 0) { 1918 #ifdef DIAGNOSTIC 1919 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1920 vprint("vclean: bad ref count", vp); 1921 panic("vclean: ref cnt"); 1922 } 1923 #endif 1924 vfree(vp); 1925 } 1926 simple_unlock(&vp->v_interlock); 1927 } 1928 1929 cache_purge(vp); 1930 vp->v_vnlock = NULL; 1931 1932 if (VSHOULDFREE(vp)) 1933 vfree(vp); 1934 1935 /* 1936 * Done with purge, notify sleepers of the grim news. 1937 */ 1938 vp->v_op = dead_vnodeop_p; 1939 vn_pollgone(vp); 1940 vp->v_tag = VT_NON; 1941 vp->v_flag &= ~VXLOCK; 1942 vp->v_vxproc = NULL; 1943 if (vp->v_flag & VXWANT) { 1944 vp->v_flag &= ~VXWANT; 1945 wakeup((caddr_t) vp); 1946 } 1947 } 1948 1949 /* 1950 * Eliminate all activity associated with the requested vnode 1951 * and with all vnodes aliased to the requested vnode. 1952 */ 1953 int 1954 vop_revoke(ap) 1955 struct vop_revoke_args /* { 1956 struct vnode *a_vp; 1957 int a_flags; 1958 } */ *ap; 1959 { 1960 struct vnode *vp, *vq; 1961 dev_t dev; 1962 1963 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1964 1965 vp = ap->a_vp; 1966 /* 1967 * If a vgone (or vclean) is already in progress, 1968 * wait until it is done and return. 1969 */ 1970 if (vp->v_flag & VXLOCK) { 1971 vp->v_flag |= VXWANT; 1972 simple_unlock(&vp->v_interlock); 1973 tsleep((caddr_t)vp, PINOD, "vop_revokeall", 0); 1974 return (0); 1975 } 1976 dev = vp->v_rdev; 1977 for (;;) { 1978 simple_lock(&spechash_slock); 1979 vq = SLIST_FIRST(&dev->si_hlist); 1980 simple_unlock(&spechash_slock); 1981 if (!vq) 1982 break; 1983 vgone(vq); 1984 } 1985 return (0); 1986 } 1987 1988 /* 1989 * Recycle an unused vnode to the front of the free list. 1990 * Release the passed interlock if the vnode will be recycled. 1991 */ 1992 int 1993 vrecycle(vp, inter_lkp, p) 1994 struct vnode *vp; 1995 struct simplelock *inter_lkp; 1996 struct proc *p; 1997 { 1998 1999 simple_lock(&vp->v_interlock); 2000 if (vp->v_usecount == 0) { 2001 if (inter_lkp) { 2002 simple_unlock(inter_lkp); 2003 } 2004 vgonel(vp, p); 2005 return (1); 2006 } 2007 simple_unlock(&vp->v_interlock); 2008 return (0); 2009 } 2010 2011 /* 2012 * Eliminate all activity associated with a vnode 2013 * in preparation for reuse. 2014 */ 2015 void 2016 vgone(vp) 2017 register struct vnode *vp; 2018 { 2019 struct proc *p = curproc; /* XXX */ 2020 2021 simple_lock(&vp->v_interlock); 2022 vgonel(vp, p); 2023 } 2024 2025 /* 2026 * vgone, with the vp interlock held. 2027 */ 2028 void 2029 vgonel(vp, p) 2030 struct vnode *vp; 2031 struct proc *p; 2032 { 2033 int s; 2034 2035 /* 2036 * If a vgone (or vclean) is already in progress, 2037 * wait until it is done and return. 2038 */ 2039 if (vp->v_flag & VXLOCK) { 2040 vp->v_flag |= VXWANT; 2041 simple_unlock(&vp->v_interlock); 2042 tsleep((caddr_t)vp, PINOD, "vgone", 0); 2043 return; 2044 } 2045 2046 /* 2047 * Clean out the filesystem specific data. 2048 */ 2049 vclean(vp, DOCLOSE, p); 2050 simple_lock(&vp->v_interlock); 2051 2052 /* 2053 * Delete from old mount point vnode list, if on one. 2054 */ 2055 if (vp->v_mount != NULL) 2056 insmntque(vp, (struct mount *)0); 2057 /* 2058 * If special device, remove it from special device alias list 2059 * if it is on one. 2060 */ 2061 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 2062 simple_lock(&spechash_slock); 2063 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext); 2064 freedev(vp->v_rdev); 2065 simple_unlock(&spechash_slock); 2066 vp->v_rdev = NULL; 2067 } 2068 2069 /* 2070 * If it is on the freelist and not already at the head, 2071 * move it to the head of the list. The test of the 2072 * VDOOMED flag and the reference count of zero is because 2073 * it will be removed from the free list by getnewvnode, 2074 * but will not have its reference count incremented until 2075 * after calling vgone. If the reference count were 2076 * incremented first, vgone would (incorrectly) try to 2077 * close the previous instance of the underlying object. 2078 */ 2079 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 2080 s = splbio(); 2081 simple_lock(&vnode_free_list_slock); 2082 if (vp->v_flag & VFREE) 2083 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2084 else 2085 freevnodes++; 2086 vp->v_flag |= VFREE; 2087 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2088 simple_unlock(&vnode_free_list_slock); 2089 splx(s); 2090 } 2091 2092 vp->v_type = VBAD; 2093 simple_unlock(&vp->v_interlock); 2094 } 2095 2096 /* 2097 * Lookup a vnode by device number. 2098 */ 2099 int 2100 vfinddev(dev, type, vpp) 2101 dev_t dev; 2102 enum vtype type; 2103 struct vnode **vpp; 2104 { 2105 struct vnode *vp; 2106 2107 simple_lock(&spechash_slock); 2108 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 2109 if (type == vp->v_type) { 2110 *vpp = vp; 2111 simple_unlock(&spechash_slock); 2112 return (1); 2113 } 2114 } 2115 simple_unlock(&spechash_slock); 2116 return (0); 2117 } 2118 2119 /* 2120 * Calculate the total number of references to a special device. 2121 */ 2122 int 2123 vcount(vp) 2124 struct vnode *vp; 2125 { 2126 struct vnode *vq; 2127 int count; 2128 2129 count = 0; 2130 simple_lock(&spechash_slock); 2131 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext) 2132 count += vq->v_usecount; 2133 simple_unlock(&spechash_slock); 2134 return (count); 2135 } 2136 2137 /* 2138 * Same as above, but using the dev_t as argument 2139 */ 2140 2141 int 2142 count_dev(dev) 2143 dev_t dev; 2144 { 2145 struct vnode *vp; 2146 2147 vp = SLIST_FIRST(&dev->si_hlist); 2148 if (vp == NULL) 2149 return (0); 2150 return(vcount(vp)); 2151 } 2152 2153 /* 2154 * Print out a description of a vnode. 2155 */ 2156 static char *typename[] = 2157 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 2158 2159 void 2160 vprint(label, vp) 2161 char *label; 2162 struct vnode *vp; 2163 { 2164 char buf[96]; 2165 2166 if (label != NULL) 2167 printf("%s: %p: ", label, (void *)vp); 2168 else 2169 printf("%p: ", (void *)vp); 2170 printf("type %s, usecount %d, writecount %d, refcount %d,", 2171 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 2172 vp->v_holdcnt); 2173 buf[0] = '\0'; 2174 if (vp->v_flag & VROOT) 2175 strcat(buf, "|VROOT"); 2176 if (vp->v_flag & VTEXT) 2177 strcat(buf, "|VTEXT"); 2178 if (vp->v_flag & VSYSTEM) 2179 strcat(buf, "|VSYSTEM"); 2180 if (vp->v_flag & VXLOCK) 2181 strcat(buf, "|VXLOCK"); 2182 if (vp->v_flag & VXWANT) 2183 strcat(buf, "|VXWANT"); 2184 if (vp->v_flag & VBWAIT) 2185 strcat(buf, "|VBWAIT"); 2186 if (vp->v_flag & VDOOMED) 2187 strcat(buf, "|VDOOMED"); 2188 if (vp->v_flag & VFREE) 2189 strcat(buf, "|VFREE"); 2190 if (vp->v_flag & VOBJBUF) 2191 strcat(buf, "|VOBJBUF"); 2192 if (buf[0] != '\0') 2193 printf(" flags (%s)", &buf[1]); 2194 if (vp->v_data == NULL) { 2195 printf("\n"); 2196 } else { 2197 printf("\n\t"); 2198 VOP_PRINT(vp); 2199 } 2200 } 2201 2202 #ifdef DDB 2203 #include <ddb/ddb.h> 2204 /* 2205 * List all of the locked vnodes in the system. 2206 * Called when debugging the kernel. 2207 */ 2208 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 2209 { 2210 struct proc *p = curproc; /* XXX */ 2211 struct mount *mp, *nmp; 2212 struct vnode *vp; 2213 2214 printf("Locked vnodes\n"); 2215 simple_lock(&mountlist_slock); 2216 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2217 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 2218 nmp = TAILQ_NEXT(mp, mnt_list); 2219 continue; 2220 } 2221 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2222 if (VOP_ISLOCKED(vp, NULL)) 2223 vprint((char *)0, vp); 2224 } 2225 simple_lock(&mountlist_slock); 2226 nmp = TAILQ_NEXT(mp, mnt_list); 2227 vfs_unbusy(mp, p); 2228 } 2229 simple_unlock(&mountlist_slock); 2230 } 2231 #endif 2232 2233 /* 2234 * Top level filesystem related information gathering. 2235 */ 2236 static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS)); 2237 2238 static int 2239 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2240 { 2241 int *name = (int *)arg1 - 1; /* XXX */ 2242 u_int namelen = arg2 + 1; /* XXX */ 2243 struct vfsconf *vfsp; 2244 2245 #if 1 || defined(COMPAT_PRELITE2) 2246 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2247 if (namelen == 1) 2248 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2249 #endif 2250 2251 #ifdef notyet 2252 /* all sysctl names at this level are at least name and field */ 2253 if (namelen < 2) 2254 return (ENOTDIR); /* overloaded */ 2255 if (name[0] != VFS_GENERIC) { 2256 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2257 if (vfsp->vfc_typenum == name[0]) 2258 break; 2259 if (vfsp == NULL) 2260 return (EOPNOTSUPP); 2261 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2262 oldp, oldlenp, newp, newlen, p)); 2263 } 2264 #endif 2265 switch (name[1]) { 2266 case VFS_MAXTYPENUM: 2267 if (namelen != 2) 2268 return (ENOTDIR); 2269 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2270 case VFS_CONF: 2271 if (namelen != 3) 2272 return (ENOTDIR); /* overloaded */ 2273 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2274 if (vfsp->vfc_typenum == name[2]) 2275 break; 2276 if (vfsp == NULL) 2277 return (EOPNOTSUPP); 2278 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2279 } 2280 return (EOPNOTSUPP); 2281 } 2282 2283 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2284 "Generic filesystem"); 2285 2286 #if 1 || defined(COMPAT_PRELITE2) 2287 2288 static int 2289 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2290 { 2291 int error; 2292 struct vfsconf *vfsp; 2293 struct ovfsconf ovfs; 2294 2295 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2296 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2297 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2298 ovfs.vfc_index = vfsp->vfc_typenum; 2299 ovfs.vfc_refcount = vfsp->vfc_refcount; 2300 ovfs.vfc_flags = vfsp->vfc_flags; 2301 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2302 if (error) 2303 return error; 2304 } 2305 return 0; 2306 } 2307 2308 #endif /* 1 || COMPAT_PRELITE2 */ 2309 2310 #if 0 2311 #define KINFO_VNODESLOP 10 2312 /* 2313 * Dump vnode list (via sysctl). 2314 * Copyout address of vnode followed by vnode. 2315 */ 2316 /* ARGSUSED */ 2317 static int 2318 sysctl_vnode(SYSCTL_HANDLER_ARGS) 2319 { 2320 struct proc *p = curproc; /* XXX */ 2321 struct mount *mp, *nmp; 2322 struct vnode *nvp, *vp; 2323 int error; 2324 2325 #define VPTRSZ sizeof (struct vnode *) 2326 #define VNODESZ sizeof (struct vnode) 2327 2328 req->lock = 0; 2329 if (!req->oldptr) /* Make an estimate */ 2330 return (SYSCTL_OUT(req, 0, 2331 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2332 2333 simple_lock(&mountlist_slock); 2334 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2335 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock, p)) { 2336 nmp = TAILQ_NEXT(mp, mnt_list); 2337 continue; 2338 } 2339 again: 2340 simple_lock(&mntvnode_slock); 2341 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2342 vp != NULL; 2343 vp = nvp) { 2344 /* 2345 * Check that the vp is still associated with 2346 * this filesystem. RACE: could have been 2347 * recycled onto the same filesystem. 2348 */ 2349 if (vp->v_mount != mp) { 2350 simple_unlock(&mntvnode_slock); 2351 goto again; 2352 } 2353 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2354 simple_unlock(&mntvnode_slock); 2355 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2356 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2357 return (error); 2358 simple_lock(&mntvnode_slock); 2359 } 2360 simple_unlock(&mntvnode_slock); 2361 simple_lock(&mountlist_slock); 2362 nmp = TAILQ_NEXT(mp, mnt_list); 2363 vfs_unbusy(mp, p); 2364 } 2365 simple_unlock(&mountlist_slock); 2366 2367 return (0); 2368 } 2369 #endif 2370 2371 /* 2372 * XXX 2373 * Exporting the vnode list on large systems causes them to crash. 2374 * Exporting the vnode list on medium systems causes sysctl to coredump. 2375 */ 2376 #if 0 2377 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2378 0, 0, sysctl_vnode, "S,vnode", ""); 2379 #endif 2380 2381 /* 2382 * Check to see if a filesystem is mounted on a block device. 2383 */ 2384 int 2385 vfs_mountedon(vp) 2386 struct vnode *vp; 2387 { 2388 2389 if (vp->v_specmountpoint != NULL) 2390 return (EBUSY); 2391 return (0); 2392 } 2393 2394 /* 2395 * Unmount all filesystems. The list is traversed in reverse order 2396 * of mounting to avoid dependencies. 2397 */ 2398 void 2399 vfs_unmountall() 2400 { 2401 struct mount *mp; 2402 struct proc *p; 2403 int error; 2404 2405 if (curproc != NULL) 2406 p = curproc; 2407 else 2408 p = initproc; /* XXX XXX should this be proc0? */ 2409 /* 2410 * Since this only runs when rebooting, it is not interlocked. 2411 */ 2412 while(!TAILQ_EMPTY(&mountlist)) { 2413 mp = TAILQ_LAST(&mountlist, mntlist); 2414 error = dounmount(mp, MNT_FORCE, p); 2415 if (error) { 2416 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2417 printf("unmount of %s failed (", 2418 mp->mnt_stat.f_mntonname); 2419 if (error == EBUSY) 2420 printf("BUSY)\n"); 2421 else 2422 printf("%d)\n", error); 2423 } else { 2424 /* The unmount has removed mp from the mountlist */ 2425 } 2426 } 2427 } 2428 2429 /* 2430 * Build hash lists of net addresses and hang them off the mount point. 2431 * Called by ufs_mount() to set up the lists of export addresses. 2432 */ 2433 static int 2434 vfs_hang_addrlist(mp, nep, argp) 2435 struct mount *mp; 2436 struct netexport *nep; 2437 struct export_args *argp; 2438 { 2439 register struct netcred *np; 2440 register struct radix_node_head *rnh; 2441 register int i; 2442 struct radix_node *rn; 2443 struct sockaddr *saddr, *smask = 0; 2444 struct domain *dom; 2445 int error; 2446 2447 if (argp->ex_addrlen == 0) { 2448 if (mp->mnt_flag & MNT_DEFEXPORTED) 2449 return (EPERM); 2450 np = &nep->ne_defexported; 2451 np->netc_exflags = argp->ex_flags; 2452 np->netc_anon = argp->ex_anon; 2453 np->netc_anon.cr_ref = 1; 2454 mp->mnt_flag |= MNT_DEFEXPORTED; 2455 return (0); 2456 } 2457 2458 if (argp->ex_addrlen > MLEN) 2459 return (EINVAL); 2460 2461 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2462 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 2463 bzero((caddr_t) np, i); 2464 saddr = (struct sockaddr *) (np + 1); 2465 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2466 goto out; 2467 if (saddr->sa_len > argp->ex_addrlen) 2468 saddr->sa_len = argp->ex_addrlen; 2469 if (argp->ex_masklen) { 2470 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 2471 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 2472 if (error) 2473 goto out; 2474 if (smask->sa_len > argp->ex_masklen) 2475 smask->sa_len = argp->ex_masklen; 2476 } 2477 i = saddr->sa_family; 2478 if ((rnh = nep->ne_rtable[i]) == 0) { 2479 /* 2480 * Seems silly to initialize every AF when most are not used, 2481 * do so on demand here 2482 */ 2483 for (dom = domains; dom; dom = dom->dom_next) 2484 if (dom->dom_family == i && dom->dom_rtattach) { 2485 dom->dom_rtattach((void **) &nep->ne_rtable[i], 2486 dom->dom_rtoffset); 2487 break; 2488 } 2489 if ((rnh = nep->ne_rtable[i]) == 0) { 2490 error = ENOBUFS; 2491 goto out; 2492 } 2493 } 2494 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 2495 np->netc_rnodes); 2496 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 2497 error = EPERM; 2498 goto out; 2499 } 2500 np->netc_exflags = argp->ex_flags; 2501 np->netc_anon = argp->ex_anon; 2502 np->netc_anon.cr_ref = 1; 2503 return (0); 2504 out: 2505 free(np, M_NETADDR); 2506 return (error); 2507 } 2508 2509 /* ARGSUSED */ 2510 static int 2511 vfs_free_netcred(rn, w) 2512 struct radix_node *rn; 2513 void *w; 2514 { 2515 register struct radix_node_head *rnh = (struct radix_node_head *) w; 2516 2517 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2518 free((caddr_t) rn, M_NETADDR); 2519 return (0); 2520 } 2521 2522 /* 2523 * Free the net address hash lists that are hanging off the mount points. 2524 */ 2525 static void 2526 vfs_free_addrlist(nep) 2527 struct netexport *nep; 2528 { 2529 register int i; 2530 register struct radix_node_head *rnh; 2531 2532 for (i = 0; i <= AF_MAX; i++) 2533 if ((rnh = nep->ne_rtable[i])) { 2534 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2535 (caddr_t) rnh); 2536 free((caddr_t) rnh, M_RTABLE); 2537 nep->ne_rtable[i] = 0; 2538 } 2539 } 2540 2541 int 2542 vfs_export(mp, nep, argp) 2543 struct mount *mp; 2544 struct netexport *nep; 2545 struct export_args *argp; 2546 { 2547 int error; 2548 2549 if (argp->ex_flags & MNT_DELEXPORT) { 2550 if (mp->mnt_flag & MNT_EXPUBLIC) { 2551 vfs_setpublicfs(NULL, NULL, NULL); 2552 mp->mnt_flag &= ~MNT_EXPUBLIC; 2553 } 2554 vfs_free_addrlist(nep); 2555 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2556 } 2557 if (argp->ex_flags & MNT_EXPORTED) { 2558 if (argp->ex_flags & MNT_EXPUBLIC) { 2559 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2560 return (error); 2561 mp->mnt_flag |= MNT_EXPUBLIC; 2562 } 2563 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2564 return (error); 2565 mp->mnt_flag |= MNT_EXPORTED; 2566 } 2567 return (0); 2568 } 2569 2570 2571 /* 2572 * Set the publicly exported filesystem (WebNFS). Currently, only 2573 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2574 */ 2575 int 2576 vfs_setpublicfs(mp, nep, argp) 2577 struct mount *mp; 2578 struct netexport *nep; 2579 struct export_args *argp; 2580 { 2581 int error; 2582 struct vnode *rvp; 2583 char *cp; 2584 2585 /* 2586 * mp == NULL -> invalidate the current info, the FS is 2587 * no longer exported. May be called from either vfs_export 2588 * or unmount, so check if it hasn't already been done. 2589 */ 2590 if (mp == NULL) { 2591 if (nfs_pub.np_valid) { 2592 nfs_pub.np_valid = 0; 2593 if (nfs_pub.np_index != NULL) { 2594 FREE(nfs_pub.np_index, M_TEMP); 2595 nfs_pub.np_index = NULL; 2596 } 2597 } 2598 return (0); 2599 } 2600 2601 /* 2602 * Only one allowed at a time. 2603 */ 2604 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2605 return (EBUSY); 2606 2607 /* 2608 * Get real filehandle for root of exported FS. 2609 */ 2610 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2611 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2612 2613 if ((error = VFS_ROOT(mp, &rvp))) 2614 return (error); 2615 2616 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2617 return (error); 2618 2619 vput(rvp); 2620 2621 /* 2622 * If an indexfile was specified, pull it in. 2623 */ 2624 if (argp->ex_indexfile != NULL) { 2625 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2626 M_WAITOK); 2627 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2628 MAXNAMLEN, (size_t *)0); 2629 if (!error) { 2630 /* 2631 * Check for illegal filenames. 2632 */ 2633 for (cp = nfs_pub.np_index; *cp; cp++) { 2634 if (*cp == '/') { 2635 error = EINVAL; 2636 break; 2637 } 2638 } 2639 } 2640 if (error) { 2641 FREE(nfs_pub.np_index, M_TEMP); 2642 return (error); 2643 } 2644 } 2645 2646 nfs_pub.np_mount = mp; 2647 nfs_pub.np_valid = 1; 2648 return (0); 2649 } 2650 2651 struct netcred * 2652 vfs_export_lookup(mp, nep, nam) 2653 register struct mount *mp; 2654 struct netexport *nep; 2655 struct sockaddr *nam; 2656 { 2657 register struct netcred *np; 2658 register struct radix_node_head *rnh; 2659 struct sockaddr *saddr; 2660 2661 np = NULL; 2662 if (mp->mnt_flag & MNT_EXPORTED) { 2663 /* 2664 * Lookup in the export list first. 2665 */ 2666 if (nam != NULL) { 2667 saddr = nam; 2668 rnh = nep->ne_rtable[saddr->sa_family]; 2669 if (rnh != NULL) { 2670 np = (struct netcred *) 2671 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2672 rnh); 2673 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2674 np = NULL; 2675 } 2676 } 2677 /* 2678 * If no address match, use the default if it exists. 2679 */ 2680 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2681 np = &nep->ne_defexported; 2682 } 2683 return (np); 2684 } 2685 2686 /* 2687 * perform msync on all vnodes under a mount point 2688 * the mount point must be locked. 2689 */ 2690 void 2691 vfs_msync(struct mount *mp, int flags) 2692 { 2693 struct vnode *vp, *nvp; 2694 struct vm_object *obj; 2695 int tries; 2696 2697 tries = 5; 2698 simple_lock(&mntvnode_slock); 2699 loop: 2700 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) { 2701 if (vp->v_mount != mp) { 2702 if (--tries > 0) 2703 goto loop; 2704 break; 2705 } 2706 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2707 2708 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2709 continue; 2710 2711 /* 2712 * There could be hundreds of thousands of vnodes, we cannot 2713 * afford to do anything heavy-weight until we have a fairly 2714 * good indication that there is something to do. 2715 */ 2716 if ((vp->v_flag & VOBJDIRTY) && 2717 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 2718 simple_unlock(&mntvnode_slock); 2719 if (!vget(vp, 2720 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curproc)) { 2721 if (VOP_GETVOBJECT(vp, &obj) == 0) { 2722 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2723 } 2724 vput(vp); 2725 } 2726 simple_lock(&mntvnode_slock); 2727 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) { 2728 if (--tries > 0) 2729 goto loop; 2730 break; 2731 } 2732 } 2733 } 2734 simple_unlock(&mntvnode_slock); 2735 } 2736 2737 /* 2738 * Create the VM object needed for VMIO and mmap support. This 2739 * is done for all VREG files in the system. Some filesystems might 2740 * afford the additional metadata buffering capability of the 2741 * VMIO code by making the device node be VMIO mode also. 2742 * 2743 * vp must be locked when vfs_object_create is called. 2744 */ 2745 int 2746 vfs_object_create(vp, p, cred) 2747 struct vnode *vp; 2748 struct proc *p; 2749 struct ucred *cred; 2750 { 2751 return (VOP_CREATEVOBJECT(vp, cred, p)); 2752 } 2753 2754 void 2755 vfree(vp) 2756 struct vnode *vp; 2757 { 2758 int s; 2759 2760 s = splbio(); 2761 simple_lock(&vnode_free_list_slock); 2762 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free")); 2763 if (vp->v_flag & VAGE) { 2764 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2765 } else { 2766 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2767 } 2768 freevnodes++; 2769 simple_unlock(&vnode_free_list_slock); 2770 vp->v_flag &= ~VAGE; 2771 vp->v_flag |= VFREE; 2772 splx(s); 2773 } 2774 2775 void 2776 vbusy(vp) 2777 struct vnode *vp; 2778 { 2779 int s; 2780 2781 s = splbio(); 2782 simple_lock(&vnode_free_list_slock); 2783 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free")); 2784 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2785 freevnodes--; 2786 simple_unlock(&vnode_free_list_slock); 2787 vp->v_flag &= ~(VFREE|VAGE); 2788 splx(s); 2789 } 2790 2791 /* 2792 * Record a process's interest in events which might happen to 2793 * a vnode. Because poll uses the historic select-style interface 2794 * internally, this routine serves as both the ``check for any 2795 * pending events'' and the ``record my interest in future events'' 2796 * functions. (These are done together, while the lock is held, 2797 * to avoid race conditions.) 2798 */ 2799 int 2800 vn_pollrecord(vp, p, events) 2801 struct vnode *vp; 2802 struct proc *p; 2803 short events; 2804 { 2805 simple_lock(&vp->v_pollinfo.vpi_lock); 2806 if (vp->v_pollinfo.vpi_revents & events) { 2807 /* 2808 * This leaves events we are not interested 2809 * in available for the other process which 2810 * which presumably had requested them 2811 * (otherwise they would never have been 2812 * recorded). 2813 */ 2814 events &= vp->v_pollinfo.vpi_revents; 2815 vp->v_pollinfo.vpi_revents &= ~events; 2816 2817 simple_unlock(&vp->v_pollinfo.vpi_lock); 2818 return events; 2819 } 2820 vp->v_pollinfo.vpi_events |= events; 2821 selrecord(p, &vp->v_pollinfo.vpi_selinfo); 2822 simple_unlock(&vp->v_pollinfo.vpi_lock); 2823 return 0; 2824 } 2825 2826 /* 2827 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2828 * it is possible for us to miss an event due to race conditions, but 2829 * that condition is expected to be rare, so for the moment it is the 2830 * preferred interface. 2831 */ 2832 void 2833 vn_pollevent(vp, events) 2834 struct vnode *vp; 2835 short events; 2836 { 2837 simple_lock(&vp->v_pollinfo.vpi_lock); 2838 if (vp->v_pollinfo.vpi_events & events) { 2839 /* 2840 * We clear vpi_events so that we don't 2841 * call selwakeup() twice if two events are 2842 * posted before the polling process(es) is 2843 * awakened. This also ensures that we take at 2844 * most one selwakeup() if the polling process 2845 * is no longer interested. However, it does 2846 * mean that only one event can be noticed at 2847 * a time. (Perhaps we should only clear those 2848 * event bits which we note?) XXX 2849 */ 2850 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2851 vp->v_pollinfo.vpi_revents |= events; 2852 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2853 } 2854 simple_unlock(&vp->v_pollinfo.vpi_lock); 2855 } 2856 2857 /* 2858 * Wake up anyone polling on vp because it is being revoked. 2859 * This depends on dead_poll() returning POLLHUP for correct 2860 * behavior. 2861 */ 2862 void 2863 vn_pollgone(vp) 2864 struct vnode *vp; 2865 { 2866 simple_lock(&vp->v_pollinfo.vpi_lock); 2867 if (vp->v_pollinfo.vpi_events) { 2868 vp->v_pollinfo.vpi_events = 0; 2869 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2870 } 2871 simple_unlock(&vp->v_pollinfo.vpi_lock); 2872 } 2873 2874 2875 2876 /* 2877 * Routine to create and manage a filesystem syncer vnode. 2878 */ 2879 #define sync_close ((int (*) __P((struct vop_close_args *)))nullop) 2880 static int sync_fsync __P((struct vop_fsync_args *)); 2881 static int sync_inactive __P((struct vop_inactive_args *)); 2882 static int sync_reclaim __P((struct vop_reclaim_args *)); 2883 #define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock) 2884 #define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock) 2885 static int sync_print __P((struct vop_print_args *)); 2886 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked) 2887 2888 static vop_t **sync_vnodeop_p; 2889 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2890 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2891 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2892 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2893 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2894 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2895 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2896 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2897 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2898 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2899 { NULL, NULL } 2900 }; 2901 static struct vnodeopv_desc sync_vnodeop_opv_desc = 2902 { &sync_vnodeop_p, sync_vnodeop_entries }; 2903 2904 VNODEOP_SET(sync_vnodeop_opv_desc); 2905 2906 /* 2907 * Create a new filesystem syncer vnode for the specified mount point. 2908 */ 2909 int 2910 vfs_allocate_syncvnode(mp) 2911 struct mount *mp; 2912 { 2913 struct vnode *vp; 2914 static long start, incr, next; 2915 int error; 2916 2917 /* Allocate a new vnode */ 2918 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2919 mp->mnt_syncer = NULL; 2920 return (error); 2921 } 2922 vp->v_type = VNON; 2923 /* 2924 * Place the vnode onto the syncer worklist. We attempt to 2925 * scatter them about on the list so that they will go off 2926 * at evenly distributed times even if all the filesystems 2927 * are mounted at once. 2928 */ 2929 next += incr; 2930 if (next == 0 || next > syncer_maxdelay) { 2931 start /= 2; 2932 incr /= 2; 2933 if (start == 0) { 2934 start = syncer_maxdelay / 2; 2935 incr = syncer_maxdelay; 2936 } 2937 next = start; 2938 } 2939 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2940 mp->mnt_syncer = vp; 2941 return (0); 2942 } 2943 2944 /* 2945 * Do a lazy sync of the filesystem. 2946 */ 2947 static int 2948 sync_fsync(ap) 2949 struct vop_fsync_args /* { 2950 struct vnode *a_vp; 2951 struct ucred *a_cred; 2952 int a_waitfor; 2953 struct proc *a_p; 2954 } */ *ap; 2955 { 2956 struct vnode *syncvp = ap->a_vp; 2957 struct mount *mp = syncvp->v_mount; 2958 struct proc *p = ap->a_p; 2959 int asyncflag; 2960 2961 /* 2962 * We only need to do something if this is a lazy evaluation. 2963 */ 2964 if (ap->a_waitfor != MNT_LAZY) 2965 return (0); 2966 2967 /* 2968 * Move ourselves to the back of the sync list. 2969 */ 2970 vn_syncer_add_to_worklist(syncvp, syncdelay); 2971 2972 /* 2973 * Walk the list of vnodes pushing all that are dirty and 2974 * not already on the sync list. 2975 */ 2976 simple_lock(&mountlist_slock); 2977 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_slock, p) != 0) { 2978 simple_unlock(&mountlist_slock); 2979 return (0); 2980 } 2981 asyncflag = mp->mnt_flag & MNT_ASYNC; 2982 mp->mnt_flag &= ~MNT_ASYNC; 2983 vfs_msync(mp, MNT_NOWAIT); 2984 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, p); 2985 if (asyncflag) 2986 mp->mnt_flag |= MNT_ASYNC; 2987 vfs_unbusy(mp, p); 2988 return (0); 2989 } 2990 2991 /* 2992 * The syncer vnode is no referenced. 2993 */ 2994 static int 2995 sync_inactive(ap) 2996 struct vop_inactive_args /* { 2997 struct vnode *a_vp; 2998 struct proc *a_p; 2999 } */ *ap; 3000 { 3001 3002 vgone(ap->a_vp); 3003 return (0); 3004 } 3005 3006 /* 3007 * The syncer vnode is no longer needed and is being decommissioned. 3008 * 3009 * Modifications to the worklist must be protected at splbio(). 3010 */ 3011 static int 3012 sync_reclaim(ap) 3013 struct vop_reclaim_args /* { 3014 struct vnode *a_vp; 3015 } */ *ap; 3016 { 3017 struct vnode *vp = ap->a_vp; 3018 int s; 3019 3020 s = splbio(); 3021 vp->v_mount->mnt_syncer = NULL; 3022 if (vp->v_flag & VONWORKLST) { 3023 LIST_REMOVE(vp, v_synclist); 3024 vp->v_flag &= ~VONWORKLST; 3025 } 3026 splx(s); 3027 3028 return (0); 3029 } 3030 3031 /* 3032 * Print out a syncer vnode. 3033 */ 3034 static int 3035 sync_print(ap) 3036 struct vop_print_args /* { 3037 struct vnode *a_vp; 3038 } */ *ap; 3039 { 3040 struct vnode *vp = ap->a_vp; 3041 3042 printf("syncer vnode"); 3043 if (vp->v_vnlock != NULL) 3044 lockmgr_printinfo(vp->v_vnlock); 3045 printf("\n"); 3046 return (0); 3047 } 3048 3049 /* 3050 * extract the dev_t from a VBLK or VCHR 3051 */ 3052 dev_t 3053 vn_todev(vp) 3054 struct vnode *vp; 3055 { 3056 if (vp->v_type != VBLK && vp->v_type != VCHR) 3057 return (NODEV); 3058 return (vp->v_rdev); 3059 } 3060 3061 /* 3062 * Check if vnode represents a disk device 3063 */ 3064 int 3065 vn_isdisk(vp, errp) 3066 struct vnode *vp; 3067 int *errp; 3068 { 3069 if (vp->v_type != VBLK && vp->v_type != VCHR) { 3070 if (errp != NULL) 3071 *errp = ENOTBLK; 3072 return (0); 3073 } 3074 if (vp->v_rdev == NULL) { 3075 if (errp != NULL) 3076 *errp = ENXIO; 3077 return (0); 3078 } 3079 if (!devsw(vp->v_rdev)) { 3080 if (errp != NULL) 3081 *errp = ENXIO; 3082 return (0); 3083 } 3084 if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) { 3085 if (errp != NULL) 3086 *errp = ENOTBLK; 3087 return (0); 3088 } 3089 if (errp != NULL) 3090 *errp = 0; 3091 return (1); 3092 } 3093 3094 void 3095 NDFREE(ndp, flags) 3096 struct nameidata *ndp; 3097 const uint flags; 3098 { 3099 if (!(flags & NDF_NO_FREE_PNBUF) && 3100 (ndp->ni_cnd.cn_flags & HASBUF)) { 3101 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 3102 ndp->ni_cnd.cn_flags &= ~HASBUF; 3103 } 3104 if (!(flags & NDF_NO_DVP_UNLOCK) && 3105 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 3106 ndp->ni_dvp != ndp->ni_vp) 3107 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_proc); 3108 if (!(flags & NDF_NO_DVP_RELE) && 3109 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 3110 vrele(ndp->ni_dvp); 3111 ndp->ni_dvp = NULL; 3112 } 3113 if (!(flags & NDF_NO_VP_UNLOCK) && 3114 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 3115 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_proc); 3116 if (!(flags & NDF_NO_VP_RELE) && 3117 ndp->ni_vp) { 3118 vrele(ndp->ni_vp); 3119 ndp->ni_vp = NULL; 3120 } 3121 if (!(flags & NDF_NO_STARTDIR_RELE) && 3122 (ndp->ni_cnd.cn_flags & SAVESTART)) { 3123 vrele(ndp->ni_startdir); 3124 ndp->ni_startdir = NULL; 3125 } 3126 } 3127