1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 40 * $DragonFly: src/sys/kern/vfs_subr.c,v 1.15 2003/07/26 19:42:11 rob Exp $ 41 */ 42 43 /* 44 * External virtual filesystem routines 45 */ 46 #include "opt_ddb.h" 47 48 #include <sys/param.h> 49 #include <sys/systm.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/domain.h> 54 #include <sys/eventhandler.h> 55 #include <sys/fcntl.h> 56 #include <sys/kernel.h> 57 #include <sys/kthread.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/mount.h> 61 #include <sys/proc.h> 62 #include <sys/namei.h> 63 #include <sys/reboot.h> 64 #include <sys/socket.h> 65 #include <sys/stat.h> 66 #include <sys/sysctl.h> 67 #include <sys/syslog.h> 68 #include <sys/vmmeter.h> 69 #include <sys/vnode.h> 70 71 #include <machine/limits.h> 72 73 #include <vm/vm.h> 74 #include <vm/vm_object.h> 75 #include <vm/vm_extern.h> 76 #include <vm/pmap.h> 77 #include <vm/vm_map.h> 78 #include <vm/vm_page.h> 79 #include <vm/vm_pager.h> 80 #include <vm/vnode_pager.h> 81 #include <vm/vm_zone.h> 82 83 #include <sys/buf2.h> 84 #include <sys/thread2.h> 85 86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 87 88 static void insmntque __P((struct vnode *vp, struct mount *mp)); 89 static void vclean __P((struct vnode *vp, int flags, struct thread *td)); 90 static unsigned long numvnodes; 91 static void vlruvp(struct vnode *vp); 92 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 93 94 enum vtype iftovt_tab[16] = { 95 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 96 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 97 }; 98 int vttoif_tab[9] = { 99 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 100 S_IFSOCK, S_IFIFO, S_IFMT, 101 }; 102 103 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 104 105 static u_long wantfreevnodes = 25; 106 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 107 static u_long freevnodes = 0; 108 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 109 110 static int reassignbufcalls; 111 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 112 static int reassignbufloops; 113 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, ""); 114 static int reassignbufsortgood; 115 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, ""); 116 static int reassignbufsortbad; 117 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, ""); 118 static int reassignbufmethod = 1; 119 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, ""); 120 static int nameileafonly = 0; 121 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, ""); 122 123 #ifdef ENABLE_VFS_IOOPT 124 int vfs_ioopt = 0; 125 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 126 #endif 127 128 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */ 129 struct lwkt_token mountlist_token; 130 struct lwkt_token mntvnode_token; 131 int nfs_mount_type = -1; 132 static struct lwkt_token mntid_token; 133 static struct lwkt_token vnode_free_list_token; 134 static struct lwkt_token spechash_token; 135 struct nfs_public nfs_pub; /* publicly exported FS */ 136 static vm_zone_t vnode_zone; 137 138 /* 139 * The workitem queue. 140 */ 141 #define SYNCER_MAXDELAY 32 142 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 143 time_t syncdelay = 30; /* max time to delay syncing data */ 144 time_t filedelay = 30; /* time to delay syncing files */ 145 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 146 time_t dirdelay = 29; /* time to delay syncing directories */ 147 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 148 time_t metadelay = 28; /* time to delay syncing metadata */ 149 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 150 static int rushjob; /* number of slots to run ASAP */ 151 static int stat_rush_requests; /* number of times I/O speeded up */ 152 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 153 154 static int syncer_delayno = 0; 155 static long syncer_mask; 156 LIST_HEAD(synclist, vnode); 157 static struct synclist *syncer_workitem_pending; 158 159 int desiredvnodes; 160 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 161 &desiredvnodes, 0, "Maximum number of vnodes"); 162 static int minvnodes; 163 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 164 &minvnodes, 0, "Minimum number of vnodes"); 165 static int vnlru_nowhere = 0; 166 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, 167 "Number of times the vnlru process ran without success"); 168 169 static void vfs_free_addrlist __P((struct netexport *nep)); 170 static int vfs_free_netcred __P((struct radix_node *rn, void *w)); 171 static int vfs_hang_addrlist __P((struct mount *mp, struct netexport *nep, 172 struct export_args *argp)); 173 174 /* 175 * Initialize the vnode management data structures. 176 */ 177 void 178 vntblinit() 179 { 180 181 desiredvnodes = maxproc + vmstats.v_page_count / 4; 182 minvnodes = desiredvnodes / 4; 183 lwkt_inittoken(&mntvnode_token); 184 lwkt_inittoken(&mntid_token); 185 lwkt_inittoken(&spechash_token); 186 TAILQ_INIT(&vnode_free_list); 187 lwkt_inittoken(&vnode_free_list_token); 188 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 189 /* 190 * Initialize the filesystem syncer. 191 */ 192 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 193 &syncer_mask); 194 syncer_maxdelay = syncer_mask + 1; 195 } 196 197 /* 198 * Mark a mount point as busy. Used to synchronize access and to delay 199 * unmounting. Interlock is not released on failure. 200 */ 201 int 202 vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp, 203 struct thread *td) 204 { 205 int lkflags; 206 207 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 208 if (flags & LK_NOWAIT) 209 return (ENOENT); 210 mp->mnt_kern_flag |= MNTK_MWAIT; 211 if (interlkp) { 212 lwkt_reltoken(interlkp); 213 } 214 /* 215 * Since all busy locks are shared except the exclusive 216 * lock granted when unmounting, the only place that a 217 * wakeup needs to be done is at the release of the 218 * exclusive lock at the end of dounmount. 219 */ 220 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 221 if (interlkp) { 222 lwkt_gettoken(interlkp); 223 } 224 return (ENOENT); 225 } 226 lkflags = LK_SHARED | LK_NOPAUSE; 227 if (interlkp) 228 lkflags |= LK_INTERLOCK; 229 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td)) 230 panic("vfs_busy: unexpected lock failure"); 231 return (0); 232 } 233 234 /* 235 * Free a busy filesystem. 236 */ 237 void 238 vfs_unbusy(struct mount *mp, struct thread *td) 239 { 240 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 241 } 242 243 /* 244 * Lookup a filesystem type, and if found allocate and initialize 245 * a mount structure for it. 246 * 247 * Devname is usually updated by mount(8) after booting. 248 */ 249 int 250 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 251 { 252 struct thread *td = curthread; /* XXX */ 253 struct vfsconf *vfsp; 254 struct mount *mp; 255 256 if (fstypename == NULL) 257 return (ENODEV); 258 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 259 if (!strcmp(vfsp->vfc_name, fstypename)) 260 break; 261 if (vfsp == NULL) 262 return (ENODEV); 263 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 264 bzero((char *)mp, (u_long)sizeof(struct mount)); 265 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE); 266 (void)vfs_busy(mp, LK_NOWAIT, 0, td); 267 TAILQ_INIT(&mp->mnt_nvnodelist); 268 TAILQ_INIT(&mp->mnt_reservedvnlist); 269 mp->mnt_nvnodelistsize = 0; 270 mp->mnt_vfc = vfsp; 271 mp->mnt_op = vfsp->vfc_vfsops; 272 mp->mnt_flag = MNT_RDONLY; 273 mp->mnt_vnodecovered = NULLVP; 274 vfsp->vfc_refcount++; 275 mp->mnt_iosize_max = DFLTPHYS; 276 mp->mnt_stat.f_type = vfsp->vfc_typenum; 277 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 278 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 279 mp->mnt_stat.f_mntonname[0] = '/'; 280 mp->mnt_stat.f_mntonname[1] = 0; 281 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 282 *mpp = mp; 283 return (0); 284 } 285 286 /* 287 * Find an appropriate filesystem to use for the root. If a filesystem 288 * has not been preselected, walk through the list of known filesystems 289 * trying those that have mountroot routines, and try them until one 290 * works or we have tried them all. 291 */ 292 #ifdef notdef /* XXX JH */ 293 int 294 lite2_vfs_mountroot() 295 { 296 struct vfsconf *vfsp; 297 extern int (*lite2_mountroot) __P((void)); 298 int error; 299 300 if (lite2_mountroot != NULL) 301 return ((*lite2_mountroot)()); 302 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 303 if (vfsp->vfc_mountroot == NULL) 304 continue; 305 if ((error = (*vfsp->vfc_mountroot)()) == 0) 306 return (0); 307 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 308 } 309 return (ENODEV); 310 } 311 #endif 312 313 /* 314 * Lookup a mount point by filesystem identifier. 315 */ 316 struct mount * 317 vfs_getvfs(fsid) 318 fsid_t *fsid; 319 { 320 struct mount *mp; 321 322 lwkt_gettoken(&mountlist_token); 323 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 324 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 325 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 326 lwkt_reltoken(&mountlist_token); 327 return (mp); 328 } 329 } 330 lwkt_reltoken(&mountlist_token); 331 return ((struct mount *) 0); 332 } 333 334 /* 335 * Get a new unique fsid. Try to make its val[0] unique, since this value 336 * will be used to create fake device numbers for stat(). Also try (but 337 * not so hard) make its val[0] unique mod 2^16, since some emulators only 338 * support 16-bit device numbers. We end up with unique val[0]'s for the 339 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 340 * 341 * Keep in mind that several mounts may be running in parallel. Starting 342 * the search one past where the previous search terminated is both a 343 * micro-optimization and a defense against returning the same fsid to 344 * different mounts. 345 */ 346 void 347 vfs_getnewfsid(mp) 348 struct mount *mp; 349 { 350 static u_int16_t mntid_base; 351 fsid_t tfsid; 352 int mtype; 353 354 lwkt_gettoken(&mntid_token); 355 mtype = mp->mnt_vfc->vfc_typenum; 356 tfsid.val[1] = mtype; 357 mtype = (mtype & 0xFF) << 24; 358 for (;;) { 359 tfsid.val[0] = makeudev(255, 360 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 361 mntid_base++; 362 if (vfs_getvfs(&tfsid) == NULL) 363 break; 364 } 365 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 366 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 367 lwkt_reltoken(&mntid_token); 368 } 369 370 /* 371 * Knob to control the precision of file timestamps: 372 * 373 * 0 = seconds only; nanoseconds zeroed. 374 * 1 = seconds and nanoseconds, accurate within 1/HZ. 375 * 2 = seconds and nanoseconds, truncated to microseconds. 376 * >=3 = seconds and nanoseconds, maximum precision. 377 */ 378 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 379 380 static int timestamp_precision = TSP_SEC; 381 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 382 ×tamp_precision, 0, ""); 383 384 /* 385 * Get a current timestamp. 386 */ 387 void 388 vfs_timestamp(tsp) 389 struct timespec *tsp; 390 { 391 struct timeval tv; 392 393 switch (timestamp_precision) { 394 case TSP_SEC: 395 tsp->tv_sec = time_second; 396 tsp->tv_nsec = 0; 397 break; 398 case TSP_HZ: 399 getnanotime(tsp); 400 break; 401 case TSP_USEC: 402 microtime(&tv); 403 TIMEVAL_TO_TIMESPEC(&tv, tsp); 404 break; 405 case TSP_NSEC: 406 default: 407 nanotime(tsp); 408 break; 409 } 410 } 411 412 /* 413 * Set vnode attributes to VNOVAL 414 */ 415 void 416 vattr_null(vap) 417 struct vattr *vap; 418 { 419 420 vap->va_type = VNON; 421 vap->va_size = VNOVAL; 422 vap->va_bytes = VNOVAL; 423 vap->va_mode = VNOVAL; 424 vap->va_nlink = VNOVAL; 425 vap->va_uid = VNOVAL; 426 vap->va_gid = VNOVAL; 427 vap->va_fsid = VNOVAL; 428 vap->va_fileid = VNOVAL; 429 vap->va_blocksize = VNOVAL; 430 vap->va_rdev = VNOVAL; 431 vap->va_atime.tv_sec = VNOVAL; 432 vap->va_atime.tv_nsec = VNOVAL; 433 vap->va_mtime.tv_sec = VNOVAL; 434 vap->va_mtime.tv_nsec = VNOVAL; 435 vap->va_ctime.tv_sec = VNOVAL; 436 vap->va_ctime.tv_nsec = VNOVAL; 437 vap->va_flags = VNOVAL; 438 vap->va_gen = VNOVAL; 439 vap->va_vaflags = 0; 440 } 441 442 /* 443 * This routine is called when we have too many vnodes. It attempts 444 * to free <count> vnodes and will potentially free vnodes that still 445 * have VM backing store (VM backing store is typically the cause 446 * of a vnode blowout so we want to do this). Therefore, this operation 447 * is not considered cheap. 448 * 449 * A number of conditions may prevent a vnode from being reclaimed. 450 * the buffer cache may have references on the vnode, a directory 451 * vnode may still have references due to the namei cache representing 452 * underlying files, or the vnode may be in active use. It is not 453 * desireable to reuse such vnodes. These conditions may cause the 454 * number of vnodes to reach some minimum value regardless of what 455 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 456 */ 457 static int 458 vlrureclaim(struct mount *mp) 459 { 460 struct vnode *vp; 461 int done; 462 int trigger; 463 int usevnodes; 464 int count; 465 int gen; 466 467 /* 468 * Calculate the trigger point, don't allow user 469 * screwups to blow us up. This prevents us from 470 * recycling vnodes with lots of resident pages. We 471 * aren't trying to free memory, we are trying to 472 * free vnodes. 473 */ 474 usevnodes = desiredvnodes; 475 if (usevnodes <= 0) 476 usevnodes = 1; 477 trigger = vmstats.v_page_count * 2 / usevnodes; 478 479 done = 0; 480 gen = lwkt_gettoken(&mntvnode_token); 481 count = mp->mnt_nvnodelistsize / 10 + 1; 482 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) { 483 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 484 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 485 486 if (vp->v_type != VNON && 487 vp->v_type != VBAD && 488 VMIGHTFREE(vp) && /* critical path opt */ 489 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) 490 ) { 491 lwkt_gettoken(&vp->v_interlock); 492 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) { 493 if (VMIGHTFREE(vp)) { 494 vgonel(vp, curthread); 495 done++; 496 } else { 497 lwkt_reltoken(&vp->v_interlock); 498 } 499 } else { 500 lwkt_reltoken(&vp->v_interlock); 501 } 502 } 503 --count; 504 } 505 lwkt_reltoken(&mntvnode_token); 506 return done; 507 } 508 509 /* 510 * Attempt to recycle vnodes in a context that is always safe to block. 511 * Calling vlrurecycle() from the bowels of file system code has some 512 * interesting deadlock problems. 513 */ 514 static struct thread *vnlruthread; 515 static int vnlruproc_sig; 516 517 static void 518 vnlru_proc(void) 519 { 520 struct mount *mp, *nmp; 521 int s; 522 int done; 523 struct thread *td = curthread; 524 525 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 526 SHUTDOWN_PRI_FIRST); 527 528 s = splbio(); 529 for (;;) { 530 kproc_suspend_loop(); 531 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) { 532 vnlruproc_sig = 0; 533 wakeup(&vnlruproc_sig); 534 tsleep(td, 0, "vlruwt", hz); 535 continue; 536 } 537 done = 0; 538 lwkt_gettoken(&mountlist_token); 539 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 540 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) { 541 nmp = TAILQ_NEXT(mp, mnt_list); 542 continue; 543 } 544 done += vlrureclaim(mp); 545 lwkt_gettoken(&mountlist_token); 546 nmp = TAILQ_NEXT(mp, mnt_list); 547 vfs_unbusy(mp, td); 548 } 549 lwkt_reltoken(&mountlist_token); 550 if (done == 0) { 551 vnlru_nowhere++; 552 tsleep(td, 0, "vlrup", hz * 3); 553 } 554 } 555 splx(s); 556 } 557 558 static struct kproc_desc vnlru_kp = { 559 "vnlru", 560 vnlru_proc, 561 &vnlruthread 562 }; 563 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 564 565 /* 566 * Routines having to do with the management of the vnode table. 567 */ 568 extern vop_t **dead_vnodeop_p; 569 570 /* 571 * Return the next vnode from the free list. 572 */ 573 int 574 getnewvnode(tag, mp, vops, vpp) 575 enum vtagtype tag; 576 struct mount *mp; 577 vop_t **vops; 578 struct vnode **vpp; 579 { 580 int s; 581 int gen; 582 int vgen; 583 struct thread *td = curthread; /* XXX */ 584 struct vnode *vp = NULL; 585 vm_object_t object; 586 587 s = splbio(); 588 589 /* 590 * Try to reuse vnodes if we hit the max. This situation only 591 * occurs in certain large-memory (2G+) situations. We cannot 592 * attempt to directly reclaim vnodes due to nasty recursion 593 * problems. 594 */ 595 while (numvnodes - freevnodes > desiredvnodes) { 596 if (vnlruproc_sig == 0) { 597 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 598 wakeup(vnlruthread); 599 } 600 tsleep(&vnlruproc_sig, 0, "vlruwk", hz); 601 } 602 603 604 /* 605 * Attempt to reuse a vnode already on the free list, allocating 606 * a new vnode if we can't find one or if we have not reached a 607 * good minimum for good LRU performance. 608 */ 609 gen = lwkt_gettoken(&vnode_free_list_token); 610 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) { 611 int count; 612 613 for (count = 0; count < freevnodes; count++) { 614 vp = TAILQ_FIRST(&vnode_free_list); 615 if (vp == NULL || vp->v_usecount) 616 panic("getnewvnode: free vnode isn't"); 617 618 /* 619 * Get the vnode's interlock, then re-obtain 620 * vnode_free_list_token in case we lost it. If we 621 * did lose it while getting the vnode interlock, 622 * even if we got it back again, then retry. 623 */ 624 vgen = lwkt_gettoken(&vp->v_interlock); 625 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) { 626 --count; 627 lwkt_reltoken(&vp->v_interlock); 628 vp = NULL; 629 continue; 630 } 631 632 /* 633 * Whew! We have both tokens. Since we didn't lose 634 * the free list VFREE had better still be set. But 635 * we aren't out of the woods yet. We have to get 636 * the object (may block). If the vnode is not 637 * suitable then move it to the end of the list 638 * if we can. If we can't move it to the end of the 639 * list retry again. 640 */ 641 if ((VOP_GETVOBJECT(vp, &object) == 0 && 642 (object->resident_page_count || object->ref_count)) 643 ) { 644 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 && 645 lwkt_gentoken(&vnode_free_list_token, &gen) == 0 646 ) { 647 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 648 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 649 } else { 650 --count; 651 } 652 lwkt_reltoken(&vp->v_interlock); 653 vp = NULL; 654 continue; 655 } 656 657 /* 658 * Still not out of the woods. VOBJECT might have 659 * blocked, if we did not retain our tokens we have 660 * to retry. 661 */ 662 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 || 663 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) { 664 --count; 665 vp = NULL; 666 continue; 667 } 668 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 669 KKASSERT(vp->v_flag & VFREE); 670 671 if (LIST_FIRST(&vp->v_cache_src)) { 672 /* 673 * note: nameileafonly sysctl is temporary, 674 * for debugging only, and will eventually be 675 * removed. 676 */ 677 if (nameileafonly > 0) { 678 /* 679 * Do not reuse namei-cached directory 680 * vnodes that have cached 681 * subdirectories. 682 */ 683 if (cache_leaf_test(vp) < 0) { 684 lwkt_reltoken(&vp->v_interlock); 685 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 686 vp = NULL; 687 continue; 688 } 689 } else if (nameileafonly < 0 || 690 vmiodirenable == 0) { 691 /* 692 * Do not reuse namei-cached directory 693 * vnodes if nameileafonly is -1 or 694 * if VMIO backing for directories is 695 * turned off (otherwise we reuse them 696 * too quickly). 697 */ 698 lwkt_reltoken(&vp->v_interlock); 699 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 700 vp = NULL; 701 continue; 702 } 703 } 704 break; 705 } 706 } 707 708 if (vp) { 709 vp->v_flag |= VDOOMED; 710 vp->v_flag &= ~VFREE; 711 freevnodes--; 712 lwkt_reltoken(&vnode_free_list_token); 713 cache_purge(vp); /* YYY may block */ 714 vp->v_lease = NULL; 715 if (vp->v_type != VBAD) { 716 vgonel(vp, td); 717 } else { 718 lwkt_reltoken(&vp->v_interlock); 719 } 720 721 #ifdef INVARIANTS 722 { 723 int s; 724 725 if (vp->v_data) 726 panic("cleaned vnode isn't"); 727 s = splbio(); 728 if (vp->v_numoutput) 729 panic("Clean vnode has pending I/O's"); 730 splx(s); 731 } 732 #endif 733 vp->v_flag = 0; 734 vp->v_lastw = 0; 735 vp->v_lasta = 0; 736 vp->v_cstart = 0; 737 vp->v_clen = 0; 738 vp->v_socket = 0; 739 vp->v_writecount = 0; /* XXX */ 740 } else { 741 lwkt_reltoken(&vnode_free_list_token); 742 vp = (struct vnode *) zalloc(vnode_zone); 743 bzero((char *) vp, sizeof *vp); 744 lwkt_inittoken(&vp->v_interlock); 745 vp->v_dd = vp; 746 cache_purge(vp); 747 LIST_INIT(&vp->v_cache_src); 748 TAILQ_INIT(&vp->v_cache_dst); 749 numvnodes++; 750 } 751 752 TAILQ_INIT(&vp->v_cleanblkhd); 753 TAILQ_INIT(&vp->v_dirtyblkhd); 754 vp->v_type = VNON; 755 vp->v_tag = tag; 756 vp->v_op = vops; 757 insmntque(vp, mp); 758 *vpp = vp; 759 vp->v_usecount = 1; 760 vp->v_data = 0; 761 splx(s); 762 763 vfs_object_create(vp, td); 764 return (0); 765 } 766 767 /* 768 * Move a vnode from one mount queue to another. 769 */ 770 static void 771 insmntque(vp, mp) 772 struct vnode *vp; 773 struct mount *mp; 774 { 775 776 lwkt_gettoken(&mntvnode_token); 777 /* 778 * Delete from old mount point vnode list, if on one. 779 */ 780 if (vp->v_mount != NULL) { 781 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0, 782 ("bad mount point vnode list size")); 783 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 784 vp->v_mount->mnt_nvnodelistsize--; 785 } 786 /* 787 * Insert into list of vnodes for the new mount point, if available. 788 */ 789 if ((vp->v_mount = mp) == NULL) { 790 lwkt_reltoken(&mntvnode_token); 791 return; 792 } 793 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 794 mp->mnt_nvnodelistsize++; 795 lwkt_reltoken(&mntvnode_token); 796 } 797 798 /* 799 * Update outstanding I/O count and do wakeup if requested. 800 */ 801 void 802 vwakeup(bp) 803 struct buf *bp; 804 { 805 struct vnode *vp; 806 807 bp->b_flags &= ~B_WRITEINPROG; 808 if ((vp = bp->b_vp)) { 809 vp->v_numoutput--; 810 if (vp->v_numoutput < 0) 811 panic("vwakeup: neg numoutput"); 812 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 813 vp->v_flag &= ~VBWAIT; 814 wakeup((caddr_t) &vp->v_numoutput); 815 } 816 } 817 } 818 819 /* 820 * Flush out and invalidate all buffers associated with a vnode. 821 * Called with the underlying object locked. 822 */ 823 int 824 vinvalbuf(struct vnode *vp, int flags, struct thread *td, 825 int slpflag, int slptimeo) 826 { 827 struct buf *bp; 828 struct buf *nbp, *blist; 829 int s, error; 830 vm_object_t object; 831 832 if (flags & V_SAVE) { 833 s = splbio(); 834 while (vp->v_numoutput) { 835 vp->v_flag |= VBWAIT; 836 error = tsleep((caddr_t)&vp->v_numoutput, 837 slpflag, "vinvlbuf", slptimeo); 838 if (error) { 839 splx(s); 840 return (error); 841 } 842 } 843 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 844 splx(s); 845 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) 846 return (error); 847 s = splbio(); 848 if (vp->v_numoutput > 0 || 849 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 850 panic("vinvalbuf: dirty bufs"); 851 } 852 splx(s); 853 } 854 s = splbio(); 855 for (;;) { 856 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 857 if (!blist) 858 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 859 if (!blist) 860 break; 861 862 for (bp = blist; bp; bp = nbp) { 863 nbp = TAILQ_NEXT(bp, b_vnbufs); 864 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 865 error = BUF_TIMELOCK(bp, 866 LK_EXCLUSIVE | LK_SLEEPFAIL, 867 "vinvalbuf", slpflag, slptimeo); 868 if (error == ENOLCK) 869 break; 870 splx(s); 871 return (error); 872 } 873 /* 874 * XXX Since there are no node locks for NFS, I 875 * believe there is a slight chance that a delayed 876 * write will occur while sleeping just above, so 877 * check for it. Note that vfs_bio_awrite expects 878 * buffers to reside on a queue, while VOP_BWRITE and 879 * brelse do not. 880 */ 881 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 882 (flags & V_SAVE)) { 883 884 if (bp->b_vp == vp) { 885 if (bp->b_flags & B_CLUSTEROK) { 886 BUF_UNLOCK(bp); 887 vfs_bio_awrite(bp); 888 } else { 889 bremfree(bp); 890 bp->b_flags |= B_ASYNC; 891 VOP_BWRITE(bp->b_vp, bp); 892 } 893 } else { 894 bremfree(bp); 895 (void) VOP_BWRITE(bp->b_vp, bp); 896 } 897 break; 898 } 899 bremfree(bp); 900 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 901 bp->b_flags &= ~B_ASYNC; 902 brelse(bp); 903 } 904 } 905 906 /* 907 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 908 * have write I/O in-progress but if there is a VM object then the 909 * VM object can also have read-I/O in-progress. 910 */ 911 do { 912 while (vp->v_numoutput > 0) { 913 vp->v_flag |= VBWAIT; 914 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0); 915 } 916 if (VOP_GETVOBJECT(vp, &object) == 0) { 917 while (object->paging_in_progress) 918 vm_object_pip_sleep(object, "vnvlbx"); 919 } 920 } while (vp->v_numoutput > 0); 921 922 splx(s); 923 924 /* 925 * Destroy the copy in the VM cache, too. 926 */ 927 lwkt_gettoken(&vp->v_interlock); 928 if (VOP_GETVOBJECT(vp, &object) == 0) { 929 vm_object_page_remove(object, 0, 0, 930 (flags & V_SAVE) ? TRUE : FALSE); 931 } 932 lwkt_reltoken(&vp->v_interlock); 933 934 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 935 panic("vinvalbuf: flush failed"); 936 return (0); 937 } 938 939 /* 940 * Truncate a file's buffer and pages to a specified length. This 941 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 942 * sync activity. 943 */ 944 int 945 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize) 946 { 947 struct buf *bp; 948 struct buf *nbp; 949 int s, anyfreed; 950 int trunclbn; 951 952 /* 953 * Round up to the *next* lbn. 954 */ 955 trunclbn = (length + blksize - 1) / blksize; 956 957 s = splbio(); 958 restart: 959 anyfreed = 1; 960 for (;anyfreed;) { 961 anyfreed = 0; 962 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 963 nbp = TAILQ_NEXT(bp, b_vnbufs); 964 if (bp->b_lblkno >= trunclbn) { 965 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 966 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 967 goto restart; 968 } else { 969 bremfree(bp); 970 bp->b_flags |= (B_INVAL | B_RELBUF); 971 bp->b_flags &= ~B_ASYNC; 972 brelse(bp); 973 anyfreed = 1; 974 } 975 if (nbp && 976 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 977 (nbp->b_vp != vp) || 978 (nbp->b_flags & B_DELWRI))) { 979 goto restart; 980 } 981 } 982 } 983 984 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 985 nbp = TAILQ_NEXT(bp, b_vnbufs); 986 if (bp->b_lblkno >= trunclbn) { 987 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 988 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 989 goto restart; 990 } else { 991 bremfree(bp); 992 bp->b_flags |= (B_INVAL | B_RELBUF); 993 bp->b_flags &= ~B_ASYNC; 994 brelse(bp); 995 anyfreed = 1; 996 } 997 if (nbp && 998 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 999 (nbp->b_vp != vp) || 1000 (nbp->b_flags & B_DELWRI) == 0)) { 1001 goto restart; 1002 } 1003 } 1004 } 1005 } 1006 1007 if (length > 0) { 1008 restartsync: 1009 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1010 nbp = TAILQ_NEXT(bp, b_vnbufs); 1011 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 1012 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1013 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1014 goto restart; 1015 } else { 1016 bremfree(bp); 1017 if (bp->b_vp == vp) { 1018 bp->b_flags |= B_ASYNC; 1019 } else { 1020 bp->b_flags &= ~B_ASYNC; 1021 } 1022 VOP_BWRITE(bp->b_vp, bp); 1023 } 1024 goto restartsync; 1025 } 1026 1027 } 1028 } 1029 1030 while (vp->v_numoutput > 0) { 1031 vp->v_flag |= VBWAIT; 1032 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0); 1033 } 1034 1035 splx(s); 1036 1037 vnode_pager_setsize(vp, length); 1038 1039 return (0); 1040 } 1041 1042 /* 1043 * Associate a buffer with a vnode. 1044 */ 1045 void 1046 bgetvp(vp, bp) 1047 struct vnode *vp; 1048 struct buf *bp; 1049 { 1050 int s; 1051 1052 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 1053 1054 vhold(vp); 1055 bp->b_vp = vp; 1056 bp->b_dev = vn_todev(vp); 1057 /* 1058 * Insert onto list for new vnode. 1059 */ 1060 s = splbio(); 1061 bp->b_xflags |= BX_VNCLEAN; 1062 bp->b_xflags &= ~BX_VNDIRTY; 1063 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 1064 splx(s); 1065 } 1066 1067 /* 1068 * Disassociate a buffer from a vnode. 1069 */ 1070 void 1071 brelvp(bp) 1072 struct buf *bp; 1073 { 1074 struct vnode *vp; 1075 struct buflists *listheadp; 1076 int s; 1077 1078 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1079 1080 /* 1081 * Delete from old vnode list, if on one. 1082 */ 1083 vp = bp->b_vp; 1084 s = splbio(); 1085 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1086 if (bp->b_xflags & BX_VNDIRTY) 1087 listheadp = &vp->v_dirtyblkhd; 1088 else 1089 listheadp = &vp->v_cleanblkhd; 1090 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1091 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1092 } 1093 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1094 vp->v_flag &= ~VONWORKLST; 1095 LIST_REMOVE(vp, v_synclist); 1096 } 1097 splx(s); 1098 bp->b_vp = (struct vnode *) 0; 1099 vdrop(vp); 1100 } 1101 1102 /* 1103 * The workitem queue. 1104 * 1105 * It is useful to delay writes of file data and filesystem metadata 1106 * for tens of seconds so that quickly created and deleted files need 1107 * not waste disk bandwidth being created and removed. To realize this, 1108 * we append vnodes to a "workitem" queue. When running with a soft 1109 * updates implementation, most pending metadata dependencies should 1110 * not wait for more than a few seconds. Thus, mounted on block devices 1111 * are delayed only about a half the time that file data is delayed. 1112 * Similarly, directory updates are more critical, so are only delayed 1113 * about a third the time that file data is delayed. Thus, there are 1114 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 1115 * one each second (driven off the filesystem syncer process). The 1116 * syncer_delayno variable indicates the next queue that is to be processed. 1117 * Items that need to be processed soon are placed in this queue: 1118 * 1119 * syncer_workitem_pending[syncer_delayno] 1120 * 1121 * A delay of fifteen seconds is done by placing the request fifteen 1122 * entries later in the queue: 1123 * 1124 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 1125 * 1126 */ 1127 1128 /* 1129 * Add an item to the syncer work queue. 1130 */ 1131 static void 1132 vn_syncer_add_to_worklist(struct vnode *vp, int delay) 1133 { 1134 int s, slot; 1135 1136 s = splbio(); 1137 1138 if (vp->v_flag & VONWORKLST) { 1139 LIST_REMOVE(vp, v_synclist); 1140 } 1141 1142 if (delay > syncer_maxdelay - 2) 1143 delay = syncer_maxdelay - 2; 1144 slot = (syncer_delayno + delay) & syncer_mask; 1145 1146 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 1147 vp->v_flag |= VONWORKLST; 1148 splx(s); 1149 } 1150 1151 struct thread *updatethread; 1152 static void sched_sync __P((void)); 1153 static struct kproc_desc up_kp = { 1154 "syncer", 1155 sched_sync, 1156 &updatethread 1157 }; 1158 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1159 1160 /* 1161 * System filesystem synchronizer daemon. 1162 */ 1163 void 1164 sched_sync(void) 1165 { 1166 struct synclist *slp; 1167 struct vnode *vp; 1168 long starttime; 1169 int s; 1170 struct thread *td = curthread; 1171 1172 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 1173 SHUTDOWN_PRI_LAST); 1174 1175 for (;;) { 1176 kproc_suspend_loop(); 1177 1178 starttime = time_second; 1179 1180 /* 1181 * Push files whose dirty time has expired. Be careful 1182 * of interrupt race on slp queue. 1183 */ 1184 s = splbio(); 1185 slp = &syncer_workitem_pending[syncer_delayno]; 1186 syncer_delayno += 1; 1187 if (syncer_delayno == syncer_maxdelay) 1188 syncer_delayno = 0; 1189 splx(s); 1190 1191 while ((vp = LIST_FIRST(slp)) != NULL) { 1192 if (VOP_ISLOCKED(vp, NULL) == 0) { 1193 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1194 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1195 VOP_UNLOCK(vp, 0, td); 1196 } 1197 s = splbio(); 1198 if (LIST_FIRST(slp) == vp) { 1199 /* 1200 * Note: v_tag VT_VFS vps can remain on the 1201 * worklist too with no dirty blocks, but 1202 * since sync_fsync() moves it to a different 1203 * slot we are safe. 1204 */ 1205 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1206 !vn_isdisk(vp, NULL)) 1207 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1208 /* 1209 * Put us back on the worklist. The worklist 1210 * routine will remove us from our current 1211 * position and then add us back in at a later 1212 * position. 1213 */ 1214 vn_syncer_add_to_worklist(vp, syncdelay); 1215 } 1216 splx(s); 1217 } 1218 1219 /* 1220 * Do soft update processing. 1221 */ 1222 if (bioops.io_sync) 1223 (*bioops.io_sync)(NULL); 1224 1225 /* 1226 * The variable rushjob allows the kernel to speed up the 1227 * processing of the filesystem syncer process. A rushjob 1228 * value of N tells the filesystem syncer to process the next 1229 * N seconds worth of work on its queue ASAP. Currently rushjob 1230 * is used by the soft update code to speed up the filesystem 1231 * syncer process when the incore state is getting so far 1232 * ahead of the disk that the kernel memory pool is being 1233 * threatened with exhaustion. 1234 */ 1235 if (rushjob > 0) { 1236 rushjob -= 1; 1237 continue; 1238 } 1239 /* 1240 * If it has taken us less than a second to process the 1241 * current work, then wait. Otherwise start right over 1242 * again. We can still lose time if any single round 1243 * takes more than two seconds, but it does not really 1244 * matter as we are just trying to generally pace the 1245 * filesystem activity. 1246 */ 1247 if (time_second == starttime) 1248 tsleep(&lbolt, 0, "syncer", 0); 1249 } 1250 } 1251 1252 /* 1253 * Request the syncer daemon to speed up its work. 1254 * We never push it to speed up more than half of its 1255 * normal turn time, otherwise it could take over the cpu. 1256 * 1257 * YYY wchan field protected by the BGL. 1258 */ 1259 int 1260 speedup_syncer() 1261 { 1262 crit_enter(); 1263 if (updatethread->td_wchan == &lbolt) { /* YYY */ 1264 unsleep(updatethread); 1265 lwkt_schedule(updatethread); 1266 } 1267 crit_exit(); 1268 if (rushjob < syncdelay / 2) { 1269 rushjob += 1; 1270 stat_rush_requests += 1; 1271 return (1); 1272 } 1273 return(0); 1274 } 1275 1276 /* 1277 * Associate a p-buffer with a vnode. 1278 * 1279 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1280 * with the buffer. i.e. the bp has not been linked into the vnode or 1281 * ref-counted. 1282 */ 1283 void 1284 pbgetvp(vp, bp) 1285 struct vnode *vp; 1286 struct buf *bp; 1287 { 1288 1289 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1290 1291 bp->b_vp = vp; 1292 bp->b_flags |= B_PAGING; 1293 bp->b_dev = vn_todev(vp); 1294 } 1295 1296 /* 1297 * Disassociate a p-buffer from a vnode. 1298 */ 1299 void 1300 pbrelvp(bp) 1301 struct buf *bp; 1302 { 1303 1304 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1305 1306 /* XXX REMOVE ME */ 1307 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 1308 panic( 1309 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1310 bp, 1311 (int)bp->b_flags 1312 ); 1313 } 1314 bp->b_vp = (struct vnode *) 0; 1315 bp->b_flags &= ~B_PAGING; 1316 } 1317 1318 void 1319 pbreassignbuf(bp, newvp) 1320 struct buf *bp; 1321 struct vnode *newvp; 1322 { 1323 if ((bp->b_flags & B_PAGING) == 0) { 1324 panic( 1325 "pbreassignbuf() on non phys bp %p", 1326 bp 1327 ); 1328 } 1329 bp->b_vp = newvp; 1330 } 1331 1332 /* 1333 * Reassign a buffer from one vnode to another. 1334 * Used to assign file specific control information 1335 * (indirect blocks) to the vnode to which they belong. 1336 */ 1337 void 1338 reassignbuf(bp, newvp) 1339 struct buf *bp; 1340 struct vnode *newvp; 1341 { 1342 struct buflists *listheadp; 1343 int delay; 1344 int s; 1345 1346 if (newvp == NULL) { 1347 printf("reassignbuf: NULL"); 1348 return; 1349 } 1350 ++reassignbufcalls; 1351 1352 /* 1353 * B_PAGING flagged buffers cannot be reassigned because their vp 1354 * is not fully linked in. 1355 */ 1356 if (bp->b_flags & B_PAGING) 1357 panic("cannot reassign paging buffer"); 1358 1359 s = splbio(); 1360 /* 1361 * Delete from old vnode list, if on one. 1362 */ 1363 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1364 if (bp->b_xflags & BX_VNDIRTY) 1365 listheadp = &bp->b_vp->v_dirtyblkhd; 1366 else 1367 listheadp = &bp->b_vp->v_cleanblkhd; 1368 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1369 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1370 if (bp->b_vp != newvp) { 1371 vdrop(bp->b_vp); 1372 bp->b_vp = NULL; /* for clarification */ 1373 } 1374 } 1375 /* 1376 * If dirty, put on list of dirty buffers; otherwise insert onto list 1377 * of clean buffers. 1378 */ 1379 if (bp->b_flags & B_DELWRI) { 1380 struct buf *tbp; 1381 1382 listheadp = &newvp->v_dirtyblkhd; 1383 if ((newvp->v_flag & VONWORKLST) == 0) { 1384 switch (newvp->v_type) { 1385 case VDIR: 1386 delay = dirdelay; 1387 break; 1388 case VCHR: 1389 case VBLK: 1390 if (newvp->v_specmountpoint != NULL) { 1391 delay = metadelay; 1392 break; 1393 } 1394 /* fall through */ 1395 default: 1396 delay = filedelay; 1397 } 1398 vn_syncer_add_to_worklist(newvp, delay); 1399 } 1400 bp->b_xflags |= BX_VNDIRTY; 1401 tbp = TAILQ_FIRST(listheadp); 1402 if (tbp == NULL || 1403 bp->b_lblkno == 0 || 1404 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1405 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1406 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1407 ++reassignbufsortgood; 1408 } else if (bp->b_lblkno < 0) { 1409 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1410 ++reassignbufsortgood; 1411 } else if (reassignbufmethod == 1) { 1412 /* 1413 * New sorting algorithm, only handle sequential case, 1414 * otherwise append to end (but before metadata) 1415 */ 1416 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1417 (tbp->b_xflags & BX_VNDIRTY)) { 1418 /* 1419 * Found the best place to insert the buffer 1420 */ 1421 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1422 ++reassignbufsortgood; 1423 } else { 1424 /* 1425 * Missed, append to end, but before meta-data. 1426 * We know that the head buffer in the list is 1427 * not meta-data due to prior conditionals. 1428 * 1429 * Indirect effects: NFS second stage write 1430 * tends to wind up here, giving maximum 1431 * distance between the unstable write and the 1432 * commit rpc. 1433 */ 1434 tbp = TAILQ_LAST(listheadp, buflists); 1435 while (tbp && tbp->b_lblkno < 0) 1436 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1437 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1438 ++reassignbufsortbad; 1439 } 1440 } else { 1441 /* 1442 * Old sorting algorithm, scan queue and insert 1443 */ 1444 struct buf *ttbp; 1445 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1446 (ttbp->b_lblkno < bp->b_lblkno)) { 1447 ++reassignbufloops; 1448 tbp = ttbp; 1449 } 1450 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1451 } 1452 } else { 1453 bp->b_xflags |= BX_VNCLEAN; 1454 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1455 if ((newvp->v_flag & VONWORKLST) && 1456 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1457 newvp->v_flag &= ~VONWORKLST; 1458 LIST_REMOVE(newvp, v_synclist); 1459 } 1460 } 1461 if (bp->b_vp != newvp) { 1462 bp->b_vp = newvp; 1463 vhold(bp->b_vp); 1464 } 1465 splx(s); 1466 } 1467 1468 /* 1469 * Create a vnode for a block device. 1470 * Used for mounting the root file system. 1471 */ 1472 int 1473 bdevvp(dev, vpp) 1474 dev_t dev; 1475 struct vnode **vpp; 1476 { 1477 struct vnode *vp; 1478 struct vnode *nvp; 1479 int error; 1480 1481 if (dev == NODEV) { 1482 *vpp = NULLVP; 1483 return (ENXIO); 1484 } 1485 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1486 if (error) { 1487 *vpp = NULLVP; 1488 return (error); 1489 } 1490 vp = nvp; 1491 vp->v_type = VBLK; 1492 addalias(vp, dev); 1493 *vpp = vp; 1494 return (0); 1495 } 1496 1497 /* 1498 * Add vnode to the alias list hung off the dev_t. 1499 * 1500 * The reason for this gunk is that multiple vnodes can reference 1501 * the same physical device, so checking vp->v_usecount to see 1502 * how many users there are is inadequate; the v_usecount for 1503 * the vnodes need to be accumulated. vcount() does that. 1504 */ 1505 void 1506 addaliasu(nvp, nvp_rdev) 1507 struct vnode *nvp; 1508 udev_t nvp_rdev; 1509 { 1510 1511 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1512 panic("addaliasu on non-special vnode"); 1513 addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0)); 1514 } 1515 1516 void 1517 addalias(nvp, dev) 1518 struct vnode *nvp; 1519 dev_t dev; 1520 { 1521 1522 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1523 panic("addalias on non-special vnode"); 1524 1525 nvp->v_rdev = dev; 1526 lwkt_gettoken(&spechash_token); 1527 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1528 lwkt_reltoken(&spechash_token); 1529 } 1530 1531 /* 1532 * Grab a particular vnode from the free list, increment its 1533 * reference count and lock it. The vnode lock bit is set if the 1534 * vnode is being eliminated in vgone. The process is awakened 1535 * when the transition is completed, and an error returned to 1536 * indicate that the vnode is no longer usable (possibly having 1537 * been changed to a new file system type). 1538 */ 1539 int 1540 vget(vp, flags, td) 1541 struct vnode *vp; 1542 int flags; 1543 struct thread *td; 1544 { 1545 int error; 1546 1547 /* 1548 * If the vnode is in the process of being cleaned out for 1549 * another use, we wait for the cleaning to finish and then 1550 * return failure. Cleaning is determined by checking that 1551 * the VXLOCK flag is set. 1552 */ 1553 if ((flags & LK_INTERLOCK) == 0) { 1554 lwkt_gettoken(&vp->v_interlock); 1555 } 1556 if (vp->v_flag & VXLOCK) { 1557 if (vp->v_vxproc == curproc) { 1558 #if 0 1559 /* this can now occur in normal operation */ 1560 log(LOG_INFO, "VXLOCK interlock avoided\n"); 1561 #endif 1562 } else { 1563 vp->v_flag |= VXWANT; 1564 lwkt_reltoken(&vp->v_interlock); 1565 tsleep((caddr_t)vp, 0, "vget", 0); 1566 return (ENOENT); 1567 } 1568 } 1569 1570 vp->v_usecount++; 1571 1572 if (VSHOULDBUSY(vp)) 1573 vbusy(vp); 1574 if (flags & LK_TYPE_MASK) { 1575 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) { 1576 /* 1577 * must expand vrele here because we do not want 1578 * to call VOP_INACTIVE if the reference count 1579 * drops back to zero since it was never really 1580 * active. We must remove it from the free list 1581 * before sleeping so that multiple processes do 1582 * not try to recycle it. 1583 */ 1584 lwkt_gettoken(&vp->v_interlock); 1585 vp->v_usecount--; 1586 if (VSHOULDFREE(vp)) 1587 vfree(vp); 1588 else 1589 vlruvp(vp); 1590 lwkt_reltoken(&vp->v_interlock); 1591 } 1592 return (error); 1593 } 1594 lwkt_reltoken(&vp->v_interlock); 1595 return (0); 1596 } 1597 1598 void 1599 vref(struct vnode *vp) 1600 { 1601 lwkt_gettoken(&vp->v_interlock); 1602 vp->v_usecount++; 1603 lwkt_reltoken(&vp->v_interlock); 1604 } 1605 1606 /* 1607 * Vnode put/release. 1608 * If count drops to zero, call inactive routine and return to freelist. 1609 */ 1610 void 1611 vrele(struct vnode *vp) 1612 { 1613 struct thread *td = curthread; /* XXX */ 1614 1615 KASSERT(vp != NULL, ("vrele: null vp")); 1616 1617 lwkt_gettoken(&vp->v_interlock); 1618 1619 if (vp->v_usecount > 1) { 1620 1621 vp->v_usecount--; 1622 lwkt_reltoken(&vp->v_interlock); 1623 1624 return; 1625 } 1626 1627 if (vp->v_usecount == 1) { 1628 vp->v_usecount--; 1629 /* 1630 * We must call VOP_INACTIVE with the node locked. 1631 * If we are doing a vpu, the node is already locked, 1632 * but, in the case of vrele, we must explicitly lock 1633 * the vnode before calling VOP_INACTIVE 1634 */ 1635 1636 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) 1637 VOP_INACTIVE(vp, td); 1638 if (VSHOULDFREE(vp)) 1639 vfree(vp); 1640 else 1641 vlruvp(vp); 1642 } else { 1643 #ifdef DIAGNOSTIC 1644 vprint("vrele: negative ref count", vp); 1645 lwkt_reltoken(&vp->v_interlock); 1646 #endif 1647 panic("vrele: negative ref cnt"); 1648 } 1649 } 1650 1651 void 1652 vput(struct vnode *vp) 1653 { 1654 struct thread *td = curthread; /* XXX */ 1655 1656 KASSERT(vp != NULL, ("vput: null vp")); 1657 1658 lwkt_gettoken(&vp->v_interlock); 1659 1660 if (vp->v_usecount > 1) { 1661 vp->v_usecount--; 1662 VOP_UNLOCK(vp, LK_INTERLOCK, td); 1663 return; 1664 } 1665 1666 if (vp->v_usecount == 1) { 1667 vp->v_usecount--; 1668 /* 1669 * We must call VOP_INACTIVE with the node locked. 1670 * If we are doing a vpu, the node is already locked, 1671 * so we just need to release the vnode mutex. 1672 */ 1673 lwkt_reltoken(&vp->v_interlock); 1674 VOP_INACTIVE(vp, td); 1675 if (VSHOULDFREE(vp)) 1676 vfree(vp); 1677 else 1678 vlruvp(vp); 1679 } else { 1680 #ifdef DIAGNOSTIC 1681 vprint("vput: negative ref count", vp); 1682 #endif 1683 panic("vput: negative ref cnt"); 1684 } 1685 } 1686 1687 /* 1688 * Somebody doesn't want the vnode recycled. 1689 */ 1690 void 1691 vhold(vp) 1692 struct vnode *vp; 1693 { 1694 int s; 1695 1696 s = splbio(); 1697 vp->v_holdcnt++; 1698 if (VSHOULDBUSY(vp)) 1699 vbusy(vp); 1700 splx(s); 1701 } 1702 1703 /* 1704 * One less who cares about this vnode. 1705 */ 1706 void 1707 vdrop(vp) 1708 struct vnode *vp; 1709 { 1710 int s; 1711 1712 s = splbio(); 1713 if (vp->v_holdcnt <= 0) 1714 panic("vdrop: holdcnt"); 1715 vp->v_holdcnt--; 1716 if (VSHOULDFREE(vp)) 1717 vfree(vp); 1718 splx(s); 1719 } 1720 1721 /* 1722 * Remove any vnodes in the vnode table belonging to mount point mp. 1723 * 1724 * If FORCECLOSE is not specified, there should not be any active ones, 1725 * return error if any are found (nb: this is a user error, not a 1726 * system error). If FORCECLOSE is specified, detach any active vnodes 1727 * that are found. 1728 * 1729 * If WRITECLOSE is set, only flush out regular file vnodes open for 1730 * writing. 1731 * 1732 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 1733 * 1734 * `rootrefs' specifies the base reference count for the root vnode 1735 * of this filesystem. The root vnode is considered busy if its 1736 * v_usecount exceeds this value. On a successful return, vflush() 1737 * will call vrele() on the root vnode exactly rootrefs times. 1738 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 1739 * be zero. 1740 */ 1741 #ifdef DIAGNOSTIC 1742 static int busyprt = 0; /* print out busy vnodes */ 1743 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1744 #endif 1745 1746 int 1747 vflush(mp, rootrefs, flags) 1748 struct mount *mp; 1749 int rootrefs; 1750 int flags; 1751 { 1752 struct thread *td = curthread; /* XXX */ 1753 struct vnode *vp, *nvp, *rootvp = NULL; 1754 struct vattr vattr; 1755 int busy = 0, error; 1756 1757 if (rootrefs > 0) { 1758 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1759 ("vflush: bad args")); 1760 /* 1761 * Get the filesystem root vnode. We can vput() it 1762 * immediately, since with rootrefs > 0, it won't go away. 1763 */ 1764 if ((error = VFS_ROOT(mp, &rootvp)) != 0) 1765 return (error); 1766 vput(rootvp); 1767 } 1768 lwkt_gettoken(&mntvnode_token); 1769 loop: 1770 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) { 1771 /* 1772 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1773 * Start over if it has (it won't be on the list anymore). 1774 */ 1775 if (vp->v_mount != mp) 1776 goto loop; 1777 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1778 1779 lwkt_gettoken(&vp->v_interlock); 1780 /* 1781 * Skip over a vnodes marked VSYSTEM. 1782 */ 1783 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1784 lwkt_reltoken(&vp->v_interlock); 1785 continue; 1786 } 1787 /* 1788 * If WRITECLOSE is set, flush out unlinked but still open 1789 * files (even if open only for reading) and regular file 1790 * vnodes open for writing. 1791 */ 1792 if ((flags & WRITECLOSE) && 1793 (vp->v_type == VNON || 1794 (VOP_GETATTR(vp, &vattr, td) == 0 && 1795 vattr.va_nlink > 0)) && 1796 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1797 lwkt_reltoken(&vp->v_interlock); 1798 continue; 1799 } 1800 1801 /* 1802 * With v_usecount == 0, all we need to do is clear out the 1803 * vnode data structures and we are done. 1804 */ 1805 if (vp->v_usecount == 0) { 1806 lwkt_reltoken(&mntvnode_token); 1807 vgonel(vp, td); 1808 lwkt_gettoken(&mntvnode_token); 1809 continue; 1810 } 1811 1812 /* 1813 * If FORCECLOSE is set, forcibly close the vnode. For block 1814 * or character devices, revert to an anonymous device. For 1815 * all other files, just kill them. 1816 */ 1817 if (flags & FORCECLOSE) { 1818 lwkt_reltoken(&mntvnode_token); 1819 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1820 vgonel(vp, td); 1821 } else { 1822 vclean(vp, 0, td); 1823 vp->v_op = spec_vnodeop_p; 1824 insmntque(vp, (struct mount *) 0); 1825 } 1826 lwkt_gettoken(&mntvnode_token); 1827 continue; 1828 } 1829 #ifdef DIAGNOSTIC 1830 if (busyprt) 1831 vprint("vflush: busy vnode", vp); 1832 #endif 1833 lwkt_reltoken(&vp->v_interlock); 1834 busy++; 1835 } 1836 lwkt_reltoken(&mntvnode_token); 1837 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1838 /* 1839 * If just the root vnode is busy, and if its refcount 1840 * is equal to `rootrefs', then go ahead and kill it. 1841 */ 1842 lwkt_gettoken(&rootvp->v_interlock); 1843 KASSERT(busy > 0, ("vflush: not busy")); 1844 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs")); 1845 if (busy == 1 && rootvp->v_usecount == rootrefs) { 1846 vgonel(rootvp, td); 1847 busy = 0; 1848 } else 1849 lwkt_reltoken(&rootvp->v_interlock); 1850 } 1851 if (busy) 1852 return (EBUSY); 1853 for (; rootrefs > 0; rootrefs--) 1854 vrele(rootvp); 1855 return (0); 1856 } 1857 1858 /* 1859 * We do not want to recycle the vnode too quickly. 1860 * 1861 * XXX we can't move vp's around the nvnodelist without really screwing 1862 * up the efficiency of filesystem SYNC and friends. This code is 1863 * disabled until we fix the syncing code's scanning algorithm. 1864 */ 1865 static void 1866 vlruvp(struct vnode *vp) 1867 { 1868 #if 0 1869 struct mount *mp; 1870 1871 if ((mp = vp->v_mount) != NULL) { 1872 lwkt_gettoken(&mntvnode_token); 1873 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1874 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1875 lwkt_reltoken(&mntvnode_token); 1876 } 1877 #endif 1878 } 1879 1880 /* 1881 * Disassociate the underlying file system from a vnode. 1882 */ 1883 static void 1884 vclean(struct vnode *vp, int flags, struct thread *td) 1885 { 1886 int active; 1887 1888 /* 1889 * Check to see if the vnode is in use. If so we have to reference it 1890 * before we clean it out so that its count cannot fall to zero and 1891 * generate a race against ourselves to recycle it. 1892 */ 1893 if ((active = vp->v_usecount)) 1894 vp->v_usecount++; 1895 1896 /* 1897 * Prevent the vnode from being recycled or brought into use while we 1898 * clean it out. 1899 */ 1900 if (vp->v_flag & VXLOCK) 1901 panic("vclean: deadlock"); 1902 vp->v_flag |= VXLOCK; 1903 vp->v_vxproc = curproc; 1904 /* 1905 * Even if the count is zero, the VOP_INACTIVE routine may still 1906 * have the object locked while it cleans it out. The VOP_LOCK 1907 * ensures that the VOP_INACTIVE routine is done with its work. 1908 * For active vnodes, it ensures that no other activity can 1909 * occur while the underlying object is being cleaned out. 1910 */ 1911 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td); 1912 1913 /* 1914 * Clean out any buffers associated with the vnode. 1915 */ 1916 vinvalbuf(vp, V_SAVE, td, 0, 0); 1917 1918 VOP_DESTROYVOBJECT(vp); 1919 1920 /* 1921 * If purging an active vnode, it must be closed and 1922 * deactivated before being reclaimed. Note that the 1923 * VOP_INACTIVE will unlock the vnode. 1924 */ 1925 if (active) { 1926 if (flags & DOCLOSE) 1927 VOP_CLOSE(vp, FNONBLOCK, td); 1928 VOP_INACTIVE(vp, td); 1929 } else { 1930 /* 1931 * Any other processes trying to obtain this lock must first 1932 * wait for VXLOCK to clear, then call the new lock operation. 1933 */ 1934 VOP_UNLOCK(vp, 0, td); 1935 } 1936 /* 1937 * Reclaim the vnode. 1938 */ 1939 if (VOP_RECLAIM(vp, td)) 1940 panic("vclean: cannot reclaim"); 1941 1942 if (active) { 1943 /* 1944 * Inline copy of vrele() since VOP_INACTIVE 1945 * has already been called. 1946 */ 1947 lwkt_gettoken(&vp->v_interlock); 1948 if (--vp->v_usecount <= 0) { 1949 #ifdef DIAGNOSTIC 1950 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1951 vprint("vclean: bad ref count", vp); 1952 panic("vclean: ref cnt"); 1953 } 1954 #endif 1955 vfree(vp); 1956 } 1957 lwkt_reltoken(&vp->v_interlock); 1958 } 1959 1960 cache_purge(vp); 1961 vp->v_vnlock = NULL; 1962 1963 if (VSHOULDFREE(vp)) 1964 vfree(vp); 1965 1966 /* 1967 * Done with purge, notify sleepers of the grim news. 1968 */ 1969 vp->v_op = dead_vnodeop_p; 1970 vn_pollgone(vp); 1971 vp->v_tag = VT_NON; 1972 vp->v_flag &= ~VXLOCK; 1973 vp->v_vxproc = NULL; 1974 if (vp->v_flag & VXWANT) { 1975 vp->v_flag &= ~VXWANT; 1976 wakeup((caddr_t) vp); 1977 } 1978 } 1979 1980 /* 1981 * Eliminate all activity associated with the requested vnode 1982 * and with all vnodes aliased to the requested vnode. 1983 */ 1984 int 1985 vop_revoke(ap) 1986 struct vop_revoke_args /* { 1987 struct vnode *a_vp; 1988 int a_flags; 1989 } */ *ap; 1990 { 1991 struct vnode *vp, *vq; 1992 dev_t dev; 1993 1994 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1995 1996 vp = ap->a_vp; 1997 /* 1998 * If a vgone (or vclean) is already in progress, 1999 * wait until it is done and return. 2000 */ 2001 if (vp->v_flag & VXLOCK) { 2002 vp->v_flag |= VXWANT; 2003 lwkt_reltoken(&vp->v_interlock); 2004 tsleep((caddr_t)vp, 0, "vop_revokeall", 0); 2005 return (0); 2006 } 2007 dev = vp->v_rdev; 2008 for (;;) { 2009 lwkt_gettoken(&spechash_token); 2010 vq = SLIST_FIRST(&dev->si_hlist); 2011 lwkt_reltoken(&spechash_token); 2012 if (!vq) 2013 break; 2014 vgone(vq); 2015 } 2016 return (0); 2017 } 2018 2019 /* 2020 * Recycle an unused vnode to the front of the free list. 2021 * Release the passed interlock if the vnode will be recycled. 2022 */ 2023 int 2024 vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td) 2025 { 2026 lwkt_gettoken(&vp->v_interlock); 2027 if (vp->v_usecount == 0) { 2028 if (inter_lkp) { 2029 lwkt_reltoken(inter_lkp); 2030 } 2031 vgonel(vp, td); 2032 return (1); 2033 } 2034 lwkt_reltoken(&vp->v_interlock); 2035 return (0); 2036 } 2037 2038 /* 2039 * Eliminate all activity associated with a vnode 2040 * in preparation for reuse. 2041 */ 2042 void 2043 vgone(struct vnode *vp) 2044 { 2045 struct thread *td = curthread; /* XXX */ 2046 2047 lwkt_gettoken(&vp->v_interlock); 2048 vgonel(vp, td); 2049 } 2050 2051 /* 2052 * vgone, with the vp interlock held. 2053 */ 2054 void 2055 vgonel(struct vnode *vp, struct thread *td) 2056 { 2057 int s; 2058 2059 /* 2060 * If a vgone (or vclean) is already in progress, 2061 * wait until it is done and return. 2062 */ 2063 if (vp->v_flag & VXLOCK) { 2064 vp->v_flag |= VXWANT; 2065 lwkt_reltoken(&vp->v_interlock); 2066 tsleep((caddr_t)vp, 0, "vgone", 0); 2067 return; 2068 } 2069 2070 /* 2071 * Clean out the filesystem specific data. 2072 */ 2073 vclean(vp, DOCLOSE, td); 2074 lwkt_gettoken(&vp->v_interlock); 2075 2076 /* 2077 * Delete from old mount point vnode list, if on one. 2078 */ 2079 if (vp->v_mount != NULL) 2080 insmntque(vp, (struct mount *)0); 2081 /* 2082 * If special device, remove it from special device alias list 2083 * if it is on one. 2084 */ 2085 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 2086 lwkt_gettoken(&spechash_token); 2087 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext); 2088 freedev(vp->v_rdev); 2089 lwkt_reltoken(&spechash_token); 2090 vp->v_rdev = NULL; 2091 } 2092 2093 /* 2094 * If it is on the freelist and not already at the head, 2095 * move it to the head of the list. The test of the 2096 * VDOOMED flag and the reference count of zero is because 2097 * it will be removed from the free list by getnewvnode, 2098 * but will not have its reference count incremented until 2099 * after calling vgone. If the reference count were 2100 * incremented first, vgone would (incorrectly) try to 2101 * close the previous instance of the underlying object. 2102 */ 2103 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 2104 s = splbio(); 2105 lwkt_gettoken(&vnode_free_list_token); 2106 if (vp->v_flag & VFREE) 2107 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2108 else 2109 freevnodes++; 2110 vp->v_flag |= VFREE; 2111 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2112 lwkt_reltoken(&vnode_free_list_token); 2113 splx(s); 2114 } 2115 2116 vp->v_type = VBAD; 2117 lwkt_reltoken(&vp->v_interlock); 2118 } 2119 2120 /* 2121 * Lookup a vnode by device number. 2122 */ 2123 int 2124 vfinddev(dev, type, vpp) 2125 dev_t dev; 2126 enum vtype type; 2127 struct vnode **vpp; 2128 { 2129 struct vnode *vp; 2130 2131 lwkt_gettoken(&spechash_token); 2132 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 2133 if (type == vp->v_type) { 2134 *vpp = vp; 2135 lwkt_reltoken(&spechash_token); 2136 return (1); 2137 } 2138 } 2139 lwkt_reltoken(&spechash_token); 2140 return (0); 2141 } 2142 2143 /* 2144 * Calculate the total number of references to a special device. 2145 */ 2146 int 2147 vcount(vp) 2148 struct vnode *vp; 2149 { 2150 struct vnode *vq; 2151 int count; 2152 2153 count = 0; 2154 lwkt_gettoken(&spechash_token); 2155 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext) 2156 count += vq->v_usecount; 2157 lwkt_reltoken(&spechash_token); 2158 return (count); 2159 } 2160 2161 /* 2162 * Same as above, but using the dev_t as argument 2163 */ 2164 2165 int 2166 count_dev(dev) 2167 dev_t dev; 2168 { 2169 struct vnode *vp; 2170 2171 vp = SLIST_FIRST(&dev->si_hlist); 2172 if (vp == NULL) 2173 return (0); 2174 return(vcount(vp)); 2175 } 2176 2177 /* 2178 * Print out a description of a vnode. 2179 */ 2180 static char *typename[] = 2181 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 2182 2183 void 2184 vprint(label, vp) 2185 char *label; 2186 struct vnode *vp; 2187 { 2188 char buf[96]; 2189 2190 if (label != NULL) 2191 printf("%s: %p: ", label, (void *)vp); 2192 else 2193 printf("%p: ", (void *)vp); 2194 printf("type %s, usecount %d, writecount %d, refcount %d,", 2195 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 2196 vp->v_holdcnt); 2197 buf[0] = '\0'; 2198 if (vp->v_flag & VROOT) 2199 strcat(buf, "|VROOT"); 2200 if (vp->v_flag & VTEXT) 2201 strcat(buf, "|VTEXT"); 2202 if (vp->v_flag & VSYSTEM) 2203 strcat(buf, "|VSYSTEM"); 2204 if (vp->v_flag & VXLOCK) 2205 strcat(buf, "|VXLOCK"); 2206 if (vp->v_flag & VXWANT) 2207 strcat(buf, "|VXWANT"); 2208 if (vp->v_flag & VBWAIT) 2209 strcat(buf, "|VBWAIT"); 2210 if (vp->v_flag & VDOOMED) 2211 strcat(buf, "|VDOOMED"); 2212 if (vp->v_flag & VFREE) 2213 strcat(buf, "|VFREE"); 2214 if (vp->v_flag & VOBJBUF) 2215 strcat(buf, "|VOBJBUF"); 2216 if (buf[0] != '\0') 2217 printf(" flags (%s)", &buf[1]); 2218 if (vp->v_data == NULL) { 2219 printf("\n"); 2220 } else { 2221 printf("\n\t"); 2222 VOP_PRINT(vp); 2223 } 2224 } 2225 2226 #ifdef DDB 2227 #include <ddb/ddb.h> 2228 /* 2229 * List all of the locked vnodes in the system. 2230 * Called when debugging the kernel. 2231 */ 2232 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 2233 { 2234 struct thread *td = curthread; /* XXX */ 2235 struct mount *mp, *nmp; 2236 struct vnode *vp; 2237 2238 printf("Locked vnodes\n"); 2239 lwkt_gettoken(&mountlist_token); 2240 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2241 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) { 2242 nmp = TAILQ_NEXT(mp, mnt_list); 2243 continue; 2244 } 2245 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2246 if (VOP_ISLOCKED(vp, NULL)) 2247 vprint((char *)0, vp); 2248 } 2249 lwkt_gettoken(&mountlist_token); 2250 nmp = TAILQ_NEXT(mp, mnt_list); 2251 vfs_unbusy(mp, td); 2252 } 2253 lwkt_reltoken(&mountlist_token); 2254 } 2255 #endif 2256 2257 /* 2258 * Top level filesystem related information gathering. 2259 */ 2260 static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS)); 2261 2262 static int 2263 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2264 { 2265 int *name = (int *)arg1 - 1; /* XXX */ 2266 u_int namelen = arg2 + 1; /* XXX */ 2267 struct vfsconf *vfsp; 2268 2269 #if 1 || defined(COMPAT_PRELITE2) 2270 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2271 if (namelen == 1) 2272 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2273 #endif 2274 2275 #ifdef notyet 2276 /* all sysctl names at this level are at least name and field */ 2277 if (namelen < 2) 2278 return (ENOTDIR); /* overloaded */ 2279 if (name[0] != VFS_GENERIC) { 2280 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2281 if (vfsp->vfc_typenum == name[0]) 2282 break; 2283 if (vfsp == NULL) 2284 return (EOPNOTSUPP); 2285 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2286 oldp, oldlenp, newp, newlen, p)); 2287 } 2288 #endif 2289 switch (name[1]) { 2290 case VFS_MAXTYPENUM: 2291 if (namelen != 2) 2292 return (ENOTDIR); 2293 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2294 case VFS_CONF: 2295 if (namelen != 3) 2296 return (ENOTDIR); /* overloaded */ 2297 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2298 if (vfsp->vfc_typenum == name[2]) 2299 break; 2300 if (vfsp == NULL) 2301 return (EOPNOTSUPP); 2302 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2303 } 2304 return (EOPNOTSUPP); 2305 } 2306 2307 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2308 "Generic filesystem"); 2309 2310 #if 1 || defined(COMPAT_PRELITE2) 2311 2312 static int 2313 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2314 { 2315 int error; 2316 struct vfsconf *vfsp; 2317 struct ovfsconf ovfs; 2318 2319 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2320 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2321 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2322 ovfs.vfc_index = vfsp->vfc_typenum; 2323 ovfs.vfc_refcount = vfsp->vfc_refcount; 2324 ovfs.vfc_flags = vfsp->vfc_flags; 2325 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2326 if (error) 2327 return error; 2328 } 2329 return 0; 2330 } 2331 2332 #endif /* 1 || COMPAT_PRELITE2 */ 2333 2334 #if 0 2335 #define KINFO_VNODESLOP 10 2336 /* 2337 * Dump vnode list (via sysctl). 2338 * Copyout address of vnode followed by vnode. 2339 */ 2340 /* ARGSUSED */ 2341 static int 2342 sysctl_vnode(SYSCTL_HANDLER_ARGS) 2343 { 2344 struct proc *p = curproc; /* XXX */ 2345 struct mount *mp, *nmp; 2346 struct vnode *nvp, *vp; 2347 int error; 2348 2349 #define VPTRSZ sizeof (struct vnode *) 2350 #define VNODESZ sizeof (struct vnode) 2351 2352 req->lock = 0; 2353 if (!req->oldptr) /* Make an estimate */ 2354 return (SYSCTL_OUT(req, 0, 2355 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2356 2357 lwkt_gettoken(&mountlist_token); 2358 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2359 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) { 2360 nmp = TAILQ_NEXT(mp, mnt_list); 2361 continue; 2362 } 2363 again: 2364 lwkt_gettoken(&mntvnode_token); 2365 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2366 vp != NULL; 2367 vp = nvp) { 2368 /* 2369 * Check that the vp is still associated with 2370 * this filesystem. RACE: could have been 2371 * recycled onto the same filesystem. 2372 */ 2373 if (vp->v_mount != mp) { 2374 lwkt_reltoken(&mntvnode_token); 2375 goto again; 2376 } 2377 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2378 lwkt_reltoken(&mntvnode_token); 2379 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2380 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2381 return (error); 2382 lwkt_gettoken(&mntvnode_token); 2383 } 2384 lwkt_reltoken(&mntvnode_token); 2385 lwkt_gettoken(&mountlist_token); 2386 nmp = TAILQ_NEXT(mp, mnt_list); 2387 vfs_unbusy(mp, p); 2388 } 2389 lwkt_reltoken(&mountlist_token); 2390 2391 return (0); 2392 } 2393 #endif 2394 2395 /* 2396 * XXX 2397 * Exporting the vnode list on large systems causes them to crash. 2398 * Exporting the vnode list on medium systems causes sysctl to coredump. 2399 */ 2400 #if 0 2401 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2402 0, 0, sysctl_vnode, "S,vnode", ""); 2403 #endif 2404 2405 /* 2406 * Check to see if a filesystem is mounted on a block device. 2407 */ 2408 int 2409 vfs_mountedon(vp) 2410 struct vnode *vp; 2411 { 2412 2413 if (vp->v_specmountpoint != NULL) 2414 return (EBUSY); 2415 return (0); 2416 } 2417 2418 /* 2419 * Unmount all filesystems. The list is traversed in reverse order 2420 * of mounting to avoid dependencies. 2421 */ 2422 void 2423 vfs_unmountall() 2424 { 2425 struct mount *mp; 2426 struct thread *td = curthread; 2427 int error; 2428 2429 if (td->td_proc == NULL) 2430 td = initproc->p_thread; /* XXX XXX use proc0 instead? */ 2431 2432 /* 2433 * Since this only runs when rebooting, it is not interlocked. 2434 */ 2435 while(!TAILQ_EMPTY(&mountlist)) { 2436 mp = TAILQ_LAST(&mountlist, mntlist); 2437 error = dounmount(mp, MNT_FORCE, td); 2438 if (error) { 2439 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2440 printf("unmount of %s failed (", 2441 mp->mnt_stat.f_mntonname); 2442 if (error == EBUSY) 2443 printf("BUSY)\n"); 2444 else 2445 printf("%d)\n", error); 2446 } else { 2447 /* The unmount has removed mp from the mountlist */ 2448 } 2449 } 2450 } 2451 2452 /* 2453 * Build hash lists of net addresses and hang them off the mount point. 2454 * Called by ufs_mount() to set up the lists of export addresses. 2455 */ 2456 static int 2457 vfs_hang_addrlist(mp, nep, argp) 2458 struct mount *mp; 2459 struct netexport *nep; 2460 struct export_args *argp; 2461 { 2462 struct netcred *np; 2463 struct radix_node_head *rnh; 2464 int i; 2465 struct radix_node *rn; 2466 struct sockaddr *saddr, *smask = 0; 2467 struct domain *dom; 2468 int error; 2469 2470 if (argp->ex_addrlen == 0) { 2471 if (mp->mnt_flag & MNT_DEFEXPORTED) 2472 return (EPERM); 2473 np = &nep->ne_defexported; 2474 np->netc_exflags = argp->ex_flags; 2475 np->netc_anon = argp->ex_anon; 2476 np->netc_anon.cr_ref = 1; 2477 mp->mnt_flag |= MNT_DEFEXPORTED; 2478 return (0); 2479 } 2480 2481 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 2482 return (EINVAL); 2483 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 2484 return (EINVAL); 2485 2486 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2487 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 2488 bzero((caddr_t) np, i); 2489 saddr = (struct sockaddr *) (np + 1); 2490 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2491 goto out; 2492 if (saddr->sa_len > argp->ex_addrlen) 2493 saddr->sa_len = argp->ex_addrlen; 2494 if (argp->ex_masklen) { 2495 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 2496 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 2497 if (error) 2498 goto out; 2499 if (smask->sa_len > argp->ex_masklen) 2500 smask->sa_len = argp->ex_masklen; 2501 } 2502 i = saddr->sa_family; 2503 if ((rnh = nep->ne_rtable[i]) == 0) { 2504 /* 2505 * Seems silly to initialize every AF when most are not used, 2506 * do so on demand here 2507 */ 2508 for (dom = domains; dom; dom = dom->dom_next) 2509 if (dom->dom_family == i && dom->dom_rtattach) { 2510 dom->dom_rtattach((void **) &nep->ne_rtable[i], 2511 dom->dom_rtoffset); 2512 break; 2513 } 2514 if ((rnh = nep->ne_rtable[i]) == 0) { 2515 error = ENOBUFS; 2516 goto out; 2517 } 2518 } 2519 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 2520 np->netc_rnodes); 2521 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 2522 error = EPERM; 2523 goto out; 2524 } 2525 np->netc_exflags = argp->ex_flags; 2526 np->netc_anon = argp->ex_anon; 2527 np->netc_anon.cr_ref = 1; 2528 return (0); 2529 out: 2530 free(np, M_NETADDR); 2531 return (error); 2532 } 2533 2534 /* ARGSUSED */ 2535 static int 2536 vfs_free_netcred(rn, w) 2537 struct radix_node *rn; 2538 void *w; 2539 { 2540 struct radix_node_head *rnh = (struct radix_node_head *) w; 2541 2542 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2543 free((caddr_t) rn, M_NETADDR); 2544 return (0); 2545 } 2546 2547 /* 2548 * Free the net address hash lists that are hanging off the mount points. 2549 */ 2550 static void 2551 vfs_free_addrlist(nep) 2552 struct netexport *nep; 2553 { 2554 int i; 2555 struct radix_node_head *rnh; 2556 2557 for (i = 0; i <= AF_MAX; i++) 2558 if ((rnh = nep->ne_rtable[i])) { 2559 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2560 (caddr_t) rnh); 2561 free((caddr_t) rnh, M_RTABLE); 2562 nep->ne_rtable[i] = 0; 2563 } 2564 } 2565 2566 int 2567 vfs_export(mp, nep, argp) 2568 struct mount *mp; 2569 struct netexport *nep; 2570 struct export_args *argp; 2571 { 2572 int error; 2573 2574 if (argp->ex_flags & MNT_DELEXPORT) { 2575 if (mp->mnt_flag & MNT_EXPUBLIC) { 2576 vfs_setpublicfs(NULL, NULL, NULL); 2577 mp->mnt_flag &= ~MNT_EXPUBLIC; 2578 } 2579 vfs_free_addrlist(nep); 2580 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2581 } 2582 if (argp->ex_flags & MNT_EXPORTED) { 2583 if (argp->ex_flags & MNT_EXPUBLIC) { 2584 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2585 return (error); 2586 mp->mnt_flag |= MNT_EXPUBLIC; 2587 } 2588 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2589 return (error); 2590 mp->mnt_flag |= MNT_EXPORTED; 2591 } 2592 return (0); 2593 } 2594 2595 2596 /* 2597 * Set the publicly exported filesystem (WebNFS). Currently, only 2598 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2599 */ 2600 int 2601 vfs_setpublicfs(mp, nep, argp) 2602 struct mount *mp; 2603 struct netexport *nep; 2604 struct export_args *argp; 2605 { 2606 int error; 2607 struct vnode *rvp; 2608 char *cp; 2609 2610 /* 2611 * mp == NULL -> invalidate the current info, the FS is 2612 * no longer exported. May be called from either vfs_export 2613 * or unmount, so check if it hasn't already been done. 2614 */ 2615 if (mp == NULL) { 2616 if (nfs_pub.np_valid) { 2617 nfs_pub.np_valid = 0; 2618 if (nfs_pub.np_index != NULL) { 2619 FREE(nfs_pub.np_index, M_TEMP); 2620 nfs_pub.np_index = NULL; 2621 } 2622 } 2623 return (0); 2624 } 2625 2626 /* 2627 * Only one allowed at a time. 2628 */ 2629 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2630 return (EBUSY); 2631 2632 /* 2633 * Get real filehandle for root of exported FS. 2634 */ 2635 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2636 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2637 2638 if ((error = VFS_ROOT(mp, &rvp))) 2639 return (error); 2640 2641 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2642 return (error); 2643 2644 vput(rvp); 2645 2646 /* 2647 * If an indexfile was specified, pull it in. 2648 */ 2649 if (argp->ex_indexfile != NULL) { 2650 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2651 M_WAITOK); 2652 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2653 MAXNAMLEN, (size_t *)0); 2654 if (!error) { 2655 /* 2656 * Check for illegal filenames. 2657 */ 2658 for (cp = nfs_pub.np_index; *cp; cp++) { 2659 if (*cp == '/') { 2660 error = EINVAL; 2661 break; 2662 } 2663 } 2664 } 2665 if (error) { 2666 FREE(nfs_pub.np_index, M_TEMP); 2667 return (error); 2668 } 2669 } 2670 2671 nfs_pub.np_mount = mp; 2672 nfs_pub.np_valid = 1; 2673 return (0); 2674 } 2675 2676 struct netcred * 2677 vfs_export_lookup(mp, nep, nam) 2678 struct mount *mp; 2679 struct netexport *nep; 2680 struct sockaddr *nam; 2681 { 2682 struct netcred *np; 2683 struct radix_node_head *rnh; 2684 struct sockaddr *saddr; 2685 2686 np = NULL; 2687 if (mp->mnt_flag & MNT_EXPORTED) { 2688 /* 2689 * Lookup in the export list first. 2690 */ 2691 if (nam != NULL) { 2692 saddr = nam; 2693 rnh = nep->ne_rtable[saddr->sa_family]; 2694 if (rnh != NULL) { 2695 np = (struct netcred *) 2696 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2697 rnh); 2698 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2699 np = NULL; 2700 } 2701 } 2702 /* 2703 * If no address match, use the default if it exists. 2704 */ 2705 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2706 np = &nep->ne_defexported; 2707 } 2708 return (np); 2709 } 2710 2711 /* 2712 * perform msync on all vnodes under a mount point 2713 * the mount point must be locked. 2714 */ 2715 void 2716 vfs_msync(struct mount *mp, int flags) 2717 { 2718 struct thread *td = curthread; /* XXX */ 2719 struct vnode *vp, *nvp; 2720 struct vm_object *obj; 2721 int tries; 2722 2723 tries = 5; 2724 lwkt_gettoken(&mntvnode_token); 2725 loop: 2726 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) { 2727 if (vp->v_mount != mp) { 2728 if (--tries > 0) 2729 goto loop; 2730 break; 2731 } 2732 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2733 2734 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2735 continue; 2736 2737 /* 2738 * There could be hundreds of thousands of vnodes, we cannot 2739 * afford to do anything heavy-weight until we have a fairly 2740 * good indication that there is something to do. 2741 */ 2742 if ((vp->v_flag & VOBJDIRTY) && 2743 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 2744 lwkt_reltoken(&mntvnode_token); 2745 if (!vget(vp, 2746 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) { 2747 if (VOP_GETVOBJECT(vp, &obj) == 0) { 2748 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2749 } 2750 vput(vp); 2751 } 2752 lwkt_gettoken(&mntvnode_token); 2753 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) { 2754 if (--tries > 0) 2755 goto loop; 2756 break; 2757 } 2758 } 2759 } 2760 lwkt_reltoken(&mntvnode_token); 2761 } 2762 2763 /* 2764 * Create the VM object needed for VMIO and mmap support. This 2765 * is done for all VREG files in the system. Some filesystems might 2766 * afford the additional metadata buffering capability of the 2767 * VMIO code by making the device node be VMIO mode also. 2768 * 2769 * vp must be locked when vfs_object_create is called. 2770 */ 2771 int 2772 vfs_object_create(struct vnode *vp, struct thread *td) 2773 { 2774 return (VOP_CREATEVOBJECT(vp, td)); 2775 } 2776 2777 void 2778 vfree(vp) 2779 struct vnode *vp; 2780 { 2781 int s; 2782 2783 s = splbio(); 2784 lwkt_gettoken(&vnode_free_list_token); 2785 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free")); 2786 if (vp->v_flag & VAGE) { 2787 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2788 } else { 2789 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2790 } 2791 freevnodes++; 2792 lwkt_reltoken(&vnode_free_list_token); 2793 vp->v_flag &= ~VAGE; 2794 vp->v_flag |= VFREE; 2795 splx(s); 2796 } 2797 2798 void 2799 vbusy(vp) 2800 struct vnode *vp; 2801 { 2802 int s; 2803 2804 s = splbio(); 2805 lwkt_gettoken(&vnode_free_list_token); 2806 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free")); 2807 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2808 freevnodes--; 2809 lwkt_reltoken(&vnode_free_list_token); 2810 vp->v_flag &= ~(VFREE|VAGE); 2811 splx(s); 2812 } 2813 2814 /* 2815 * Record a process's interest in events which might happen to 2816 * a vnode. Because poll uses the historic select-style interface 2817 * internally, this routine serves as both the ``check for any 2818 * pending events'' and the ``record my interest in future events'' 2819 * functions. (These are done together, while the lock is held, 2820 * to avoid race conditions.) 2821 */ 2822 int 2823 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 2824 { 2825 lwkt_gettoken(&vp->v_pollinfo.vpi_token); 2826 if (vp->v_pollinfo.vpi_revents & events) { 2827 /* 2828 * This leaves events we are not interested 2829 * in available for the other process which 2830 * which presumably had requested them 2831 * (otherwise they would never have been 2832 * recorded). 2833 */ 2834 events &= vp->v_pollinfo.vpi_revents; 2835 vp->v_pollinfo.vpi_revents &= ~events; 2836 2837 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2838 return events; 2839 } 2840 vp->v_pollinfo.vpi_events |= events; 2841 selrecord(td, &vp->v_pollinfo.vpi_selinfo); 2842 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2843 return 0; 2844 } 2845 2846 /* 2847 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2848 * it is possible for us to miss an event due to race conditions, but 2849 * that condition is expected to be rare, so for the moment it is the 2850 * preferred interface. 2851 */ 2852 void 2853 vn_pollevent(vp, events) 2854 struct vnode *vp; 2855 short events; 2856 { 2857 lwkt_gettoken(&vp->v_pollinfo.vpi_token); 2858 if (vp->v_pollinfo.vpi_events & events) { 2859 /* 2860 * We clear vpi_events so that we don't 2861 * call selwakeup() twice if two events are 2862 * posted before the polling process(es) is 2863 * awakened. This also ensures that we take at 2864 * most one selwakeup() if the polling process 2865 * is no longer interested. However, it does 2866 * mean that only one event can be noticed at 2867 * a time. (Perhaps we should only clear those 2868 * event bits which we note?) XXX 2869 */ 2870 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2871 vp->v_pollinfo.vpi_revents |= events; 2872 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2873 } 2874 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2875 } 2876 2877 /* 2878 * Wake up anyone polling on vp because it is being revoked. 2879 * This depends on dead_poll() returning POLLHUP for correct 2880 * behavior. 2881 */ 2882 void 2883 vn_pollgone(vp) 2884 struct vnode *vp; 2885 { 2886 lwkt_gettoken(&vp->v_pollinfo.vpi_token); 2887 if (vp->v_pollinfo.vpi_events) { 2888 vp->v_pollinfo.vpi_events = 0; 2889 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2890 } 2891 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2892 } 2893 2894 2895 2896 /* 2897 * Routine to create and manage a filesystem syncer vnode. 2898 */ 2899 #define sync_close ((int (*) __P((struct vop_close_args *)))nullop) 2900 static int sync_fsync __P((struct vop_fsync_args *)); 2901 static int sync_inactive __P((struct vop_inactive_args *)); 2902 static int sync_reclaim __P((struct vop_reclaim_args *)); 2903 #define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock) 2904 #define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock) 2905 static int sync_print __P((struct vop_print_args *)); 2906 #define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked) 2907 2908 static vop_t **sync_vnodeop_p; 2909 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2910 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2911 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2912 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2913 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2914 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2915 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2916 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2917 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2918 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2919 { NULL, NULL } 2920 }; 2921 static struct vnodeopv_desc sync_vnodeop_opv_desc = 2922 { &sync_vnodeop_p, sync_vnodeop_entries }; 2923 2924 VNODEOP_SET(sync_vnodeop_opv_desc); 2925 2926 /* 2927 * Create a new filesystem syncer vnode for the specified mount point. 2928 */ 2929 int 2930 vfs_allocate_syncvnode(mp) 2931 struct mount *mp; 2932 { 2933 struct vnode *vp; 2934 static long start, incr, next; 2935 int error; 2936 2937 /* Allocate a new vnode */ 2938 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2939 mp->mnt_syncer = NULL; 2940 return (error); 2941 } 2942 vp->v_type = VNON; 2943 /* 2944 * Place the vnode onto the syncer worklist. We attempt to 2945 * scatter them about on the list so that they will go off 2946 * at evenly distributed times even if all the filesystems 2947 * are mounted at once. 2948 */ 2949 next += incr; 2950 if (next == 0 || next > syncer_maxdelay) { 2951 start /= 2; 2952 incr /= 2; 2953 if (start == 0) { 2954 start = syncer_maxdelay / 2; 2955 incr = syncer_maxdelay; 2956 } 2957 next = start; 2958 } 2959 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2960 mp->mnt_syncer = vp; 2961 return (0); 2962 } 2963 2964 /* 2965 * Do a lazy sync of the filesystem. 2966 */ 2967 static int 2968 sync_fsync(ap) 2969 struct vop_fsync_args /* { 2970 struct vnode *a_vp; 2971 struct ucred *a_cred; 2972 int a_waitfor; 2973 struct thread *a_td; 2974 } */ *ap; 2975 { 2976 struct vnode *syncvp = ap->a_vp; 2977 struct mount *mp = syncvp->v_mount; 2978 struct thread *td = ap->a_td; 2979 int asyncflag; 2980 2981 /* 2982 * We only need to do something if this is a lazy evaluation. 2983 */ 2984 if (ap->a_waitfor != MNT_LAZY) 2985 return (0); 2986 2987 /* 2988 * Move ourselves to the back of the sync list. 2989 */ 2990 vn_syncer_add_to_worklist(syncvp, syncdelay); 2991 2992 /* 2993 * Walk the list of vnodes pushing all that are dirty and 2994 * not already on the sync list. 2995 */ 2996 lwkt_gettoken(&mountlist_token); 2997 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) { 2998 lwkt_reltoken(&mountlist_token); 2999 return (0); 3000 } 3001 asyncflag = mp->mnt_flag & MNT_ASYNC; 3002 mp->mnt_flag &= ~MNT_ASYNC; 3003 vfs_msync(mp, MNT_NOWAIT); 3004 VFS_SYNC(mp, MNT_LAZY, td); 3005 if (asyncflag) 3006 mp->mnt_flag |= MNT_ASYNC; 3007 vfs_unbusy(mp, td); 3008 return (0); 3009 } 3010 3011 /* 3012 * The syncer vnode is no referenced. 3013 */ 3014 static int 3015 sync_inactive(ap) 3016 struct vop_inactive_args /* { 3017 struct vnode *a_vp; 3018 struct proc *a_p; 3019 } */ *ap; 3020 { 3021 3022 vgone(ap->a_vp); 3023 return (0); 3024 } 3025 3026 /* 3027 * The syncer vnode is no longer needed and is being decommissioned. 3028 * 3029 * Modifications to the worklist must be protected at splbio(). 3030 */ 3031 static int 3032 sync_reclaim(ap) 3033 struct vop_reclaim_args /* { 3034 struct vnode *a_vp; 3035 } */ *ap; 3036 { 3037 struct vnode *vp = ap->a_vp; 3038 int s; 3039 3040 s = splbio(); 3041 vp->v_mount->mnt_syncer = NULL; 3042 if (vp->v_flag & VONWORKLST) { 3043 LIST_REMOVE(vp, v_synclist); 3044 vp->v_flag &= ~VONWORKLST; 3045 } 3046 splx(s); 3047 3048 return (0); 3049 } 3050 3051 /* 3052 * Print out a syncer vnode. 3053 */ 3054 static int 3055 sync_print(ap) 3056 struct vop_print_args /* { 3057 struct vnode *a_vp; 3058 } */ *ap; 3059 { 3060 struct vnode *vp = ap->a_vp; 3061 3062 printf("syncer vnode"); 3063 if (vp->v_vnlock != NULL) 3064 lockmgr_printinfo(vp->v_vnlock); 3065 printf("\n"); 3066 return (0); 3067 } 3068 3069 /* 3070 * extract the dev_t from a VBLK or VCHR 3071 */ 3072 dev_t 3073 vn_todev(vp) 3074 struct vnode *vp; 3075 { 3076 if (vp->v_type != VBLK && vp->v_type != VCHR) 3077 return (NODEV); 3078 return (vp->v_rdev); 3079 } 3080 3081 /* 3082 * Check if vnode represents a disk device 3083 */ 3084 int 3085 vn_isdisk(vp, errp) 3086 struct vnode *vp; 3087 int *errp; 3088 { 3089 if (vp->v_type != VBLK && vp->v_type != VCHR) { 3090 if (errp != NULL) 3091 *errp = ENOTBLK; 3092 return (0); 3093 } 3094 if (vp->v_rdev == NULL) { 3095 if (errp != NULL) 3096 *errp = ENXIO; 3097 return (0); 3098 } 3099 if (!dev_dport(vp->v_rdev)) { 3100 if (errp != NULL) 3101 *errp = ENXIO; 3102 return (0); 3103 } 3104 if (!(dev_dflags(vp->v_rdev) & D_DISK)) { 3105 if (errp != NULL) 3106 *errp = ENOTBLK; 3107 return (0); 3108 } 3109 if (errp != NULL) 3110 *errp = 0; 3111 return (1); 3112 } 3113 3114 void 3115 NDFREE(ndp, flags) 3116 struct nameidata *ndp; 3117 const uint flags; 3118 { 3119 if (!(flags & NDF_NO_FREE_PNBUF) && 3120 (ndp->ni_cnd.cn_flags & HASBUF)) { 3121 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 3122 ndp->ni_cnd.cn_flags &= ~HASBUF; 3123 } 3124 if (!(flags & NDF_NO_DVP_UNLOCK) && 3125 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 3126 ndp->ni_dvp != ndp->ni_vp) 3127 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td); 3128 if (!(flags & NDF_NO_DVP_RELE) && 3129 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 3130 vrele(ndp->ni_dvp); 3131 ndp->ni_dvp = NULL; 3132 } 3133 if (!(flags & NDF_NO_VP_UNLOCK) && 3134 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 3135 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td); 3136 if (!(flags & NDF_NO_VP_RELE) && 3137 ndp->ni_vp) { 3138 vrele(ndp->ni_vp); 3139 ndp->ni_vp = NULL; 3140 } 3141 if (!(flags & NDF_NO_STARTDIR_RELE) && 3142 (ndp->ni_cnd.cn_flags & SAVESTART)) { 3143 vrele(ndp->ni_startdir); 3144 ndp->ni_startdir = NULL; 3145 } 3146 } 3147