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 1.25 2004/02/10 07:34:42 dillon 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/vm_kern.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_map.h> 79 #include <vm/vm_page.h> 80 #include <vm/vm_pager.h> 81 #include <vm/vnode_pager.h> 82 #include <vm/vm_zone.h> 83 84 #include <sys/buf2.h> 85 #include <sys/thread2.h> 86 87 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 88 89 static void insmntque (struct vnode *vp, struct mount *mp); 90 static void vclean (struct vnode *vp, int flags, struct thread *td); 91 static unsigned long numvnodes; 92 static void vlruvp(struct vnode *vp); 93 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 94 95 enum vtype iftovt_tab[16] = { 96 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 97 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 98 }; 99 int vttoif_tab[9] = { 100 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 101 S_IFSOCK, S_IFIFO, S_IFMT, 102 }; 103 104 static TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ 105 106 static u_long wantfreevnodes = 25; 107 SYSCTL_INT(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 108 static u_long freevnodes = 0; 109 SYSCTL_INT(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 110 111 static int reassignbufcalls; 112 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 113 static int reassignbufloops; 114 SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, ""); 115 static int reassignbufsortgood; 116 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, ""); 117 static int reassignbufsortbad; 118 SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, ""); 119 static int reassignbufmethod = 1; 120 SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, ""); 121 122 #ifdef ENABLE_VFS_IOOPT 123 int vfs_ioopt = 0; 124 SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 125 #endif 126 127 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */ 128 struct lwkt_token mountlist_token; 129 struct lwkt_token mntvnode_token; 130 int nfs_mount_type = -1; 131 static struct lwkt_token mntid_token; 132 static struct lwkt_token vnode_free_list_token; 133 static struct lwkt_token spechash_token; 134 struct nfs_public nfs_pub; /* publicly exported FS */ 135 static vm_zone_t vnode_zone; 136 137 /* 138 * The workitem queue. 139 */ 140 #define SYNCER_MAXDELAY 32 141 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 142 time_t syncdelay = 30; /* max time to delay syncing data */ 143 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW, &syncdelay, 0, 144 "VFS data synchronization delay"); 145 time_t filedelay = 30; /* time to delay syncing files */ 146 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, 147 "File synchronization delay"); 148 time_t dirdelay = 29; /* time to delay syncing directories */ 149 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, 150 "Directory synchronization delay"); 151 time_t metadelay = 28; /* time to delay syncing metadata */ 152 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, 153 "VFS metadata synchronization delay"); 154 static int rushjob; /* number of slots to run ASAP */ 155 static int stat_rush_requests; /* number of times I/O speeded up */ 156 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 157 158 static int syncer_delayno = 0; 159 static long syncer_mask; 160 LIST_HEAD(synclist, vnode); 161 static struct synclist *syncer_workitem_pending; 162 163 int desiredvnodes; 164 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 165 &desiredvnodes, 0, "Maximum number of vnodes"); 166 static int minvnodes; 167 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 168 &minvnodes, 0, "Minimum number of vnodes"); 169 static int vnlru_nowhere = 0; 170 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, 171 "Number of times the vnlru process ran without success"); 172 173 static void vfs_free_addrlist (struct netexport *nep); 174 static int vfs_free_netcred (struct radix_node *rn, void *w); 175 static int vfs_hang_addrlist (struct mount *mp, struct netexport *nep, 176 struct export_args *argp); 177 178 /* 179 * Initialize the vnode management data structures. 180 */ 181 void 182 vntblinit() 183 { 184 185 /* 186 * Desired vnodes is a result of the physical page count 187 * and the size of kernel's heap. It scales in proportion 188 * to the amount of available physical memory. This can 189 * cause trouble on 64-bit and large memory platforms. 190 */ 191 /* desiredvnodes = maxproc + vmstats.v_page_count / 4; */ 192 desiredvnodes = 193 min(maxproc + vmstats.v_page_count /4, 194 2 * (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / 195 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 196 197 minvnodes = desiredvnodes / 4; 198 lwkt_inittoken(&mountlist_token); 199 lwkt_inittoken(&mntvnode_token); 200 lwkt_inittoken(&mntid_token); 201 lwkt_inittoken(&spechash_token); 202 TAILQ_INIT(&vnode_free_list); 203 lwkt_inittoken(&vnode_free_list_token); 204 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 205 /* 206 * Initialize the filesystem syncer. 207 */ 208 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 209 &syncer_mask); 210 syncer_maxdelay = syncer_mask + 1; 211 } 212 213 /* 214 * Mark a mount point as busy. Used to synchronize access and to delay 215 * unmounting. Interlock is not released on failure. 216 */ 217 int 218 vfs_busy(struct mount *mp, int flags, struct lwkt_token *interlkp, 219 struct thread *td) 220 { 221 int lkflags; 222 223 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 224 if (flags & LK_NOWAIT) 225 return (ENOENT); 226 mp->mnt_kern_flag |= MNTK_MWAIT; 227 if (interlkp) { 228 lwkt_reltoken(interlkp); 229 } 230 /* 231 * Since all busy locks are shared except the exclusive 232 * lock granted when unmounting, the only place that a 233 * wakeup needs to be done is at the release of the 234 * exclusive lock at the end of dounmount. 235 */ 236 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 237 if (interlkp) { 238 lwkt_gettoken(interlkp); 239 } 240 return (ENOENT); 241 } 242 lkflags = LK_SHARED | LK_NOPAUSE; 243 if (interlkp) 244 lkflags |= LK_INTERLOCK; 245 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td)) 246 panic("vfs_busy: unexpected lock failure"); 247 return (0); 248 } 249 250 /* 251 * Free a busy filesystem. 252 */ 253 void 254 vfs_unbusy(struct mount *mp, struct thread *td) 255 { 256 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 257 } 258 259 /* 260 * Lookup a filesystem type, and if found allocate and initialize 261 * a mount structure for it. 262 * 263 * Devname is usually updated by mount(8) after booting. 264 */ 265 int 266 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 267 { 268 struct thread *td = curthread; /* XXX */ 269 struct vfsconf *vfsp; 270 struct mount *mp; 271 272 if (fstypename == NULL) 273 return (ENODEV); 274 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 275 if (!strcmp(vfsp->vfc_name, fstypename)) 276 break; 277 if (vfsp == NULL) 278 return (ENODEV); 279 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 280 bzero((char *)mp, (u_long)sizeof(struct mount)); 281 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE); 282 (void)vfs_busy(mp, LK_NOWAIT, 0, td); 283 TAILQ_INIT(&mp->mnt_nvnodelist); 284 TAILQ_INIT(&mp->mnt_reservedvnlist); 285 mp->mnt_nvnodelistsize = 0; 286 mp->mnt_vfc = vfsp; 287 mp->mnt_op = vfsp->vfc_vfsops; 288 mp->mnt_flag = MNT_RDONLY; 289 mp->mnt_vnodecovered = NULLVP; 290 vfsp->vfc_refcount++; 291 mp->mnt_iosize_max = DFLTPHYS; 292 mp->mnt_stat.f_type = vfsp->vfc_typenum; 293 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 294 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 295 mp->mnt_stat.f_mntonname[0] = '/'; 296 mp->mnt_stat.f_mntonname[1] = 0; 297 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 298 *mpp = mp; 299 return (0); 300 } 301 302 /* 303 * Find an appropriate filesystem to use for the root. If a filesystem 304 * has not been preselected, walk through the list of known filesystems 305 * trying those that have mountroot routines, and try them until one 306 * works or we have tried them all. 307 */ 308 #ifdef notdef /* XXX JH */ 309 int 310 lite2_vfs_mountroot() 311 { 312 struct vfsconf *vfsp; 313 extern int (*lite2_mountroot) (void); 314 int error; 315 316 if (lite2_mountroot != NULL) 317 return ((*lite2_mountroot)()); 318 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 319 if (vfsp->vfc_mountroot == NULL) 320 continue; 321 if ((error = (*vfsp->vfc_mountroot)()) == 0) 322 return (0); 323 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 324 } 325 return (ENODEV); 326 } 327 #endif 328 329 /* 330 * Lookup a mount point by filesystem identifier. 331 */ 332 struct mount * 333 vfs_getvfs(fsid) 334 fsid_t *fsid; 335 { 336 struct mount *mp; 337 338 lwkt_gettoken(&mountlist_token); 339 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 340 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 341 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 342 lwkt_reltoken(&mountlist_token); 343 return (mp); 344 } 345 } 346 lwkt_reltoken(&mountlist_token); 347 return ((struct mount *) 0); 348 } 349 350 /* 351 * Get a new unique fsid. Try to make its val[0] unique, since this value 352 * will be used to create fake device numbers for stat(). Also try (but 353 * not so hard) make its val[0] unique mod 2^16, since some emulators only 354 * support 16-bit device numbers. We end up with unique val[0]'s for the 355 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 356 * 357 * Keep in mind that several mounts may be running in parallel. Starting 358 * the search one past where the previous search terminated is both a 359 * micro-optimization and a defense against returning the same fsid to 360 * different mounts. 361 */ 362 void 363 vfs_getnewfsid(mp) 364 struct mount *mp; 365 { 366 static u_int16_t mntid_base; 367 fsid_t tfsid; 368 int mtype; 369 370 lwkt_gettoken(&mntid_token); 371 mtype = mp->mnt_vfc->vfc_typenum; 372 tfsid.val[1] = mtype; 373 mtype = (mtype & 0xFF) << 24; 374 for (;;) { 375 tfsid.val[0] = makeudev(255, 376 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 377 mntid_base++; 378 if (vfs_getvfs(&tfsid) == NULL) 379 break; 380 } 381 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 382 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 383 lwkt_reltoken(&mntid_token); 384 } 385 386 /* 387 * Knob to control the precision of file timestamps: 388 * 389 * 0 = seconds only; nanoseconds zeroed. 390 * 1 = seconds and nanoseconds, accurate within 1/HZ. 391 * 2 = seconds and nanoseconds, truncated to microseconds. 392 * >=3 = seconds and nanoseconds, maximum precision. 393 */ 394 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 395 396 static int timestamp_precision = TSP_SEC; 397 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 398 ×tamp_precision, 0, ""); 399 400 /* 401 * Get a current timestamp. 402 */ 403 void 404 vfs_timestamp(tsp) 405 struct timespec *tsp; 406 { 407 struct timeval tv; 408 409 switch (timestamp_precision) { 410 case TSP_SEC: 411 tsp->tv_sec = time_second; 412 tsp->tv_nsec = 0; 413 break; 414 case TSP_HZ: 415 getnanotime(tsp); 416 break; 417 case TSP_USEC: 418 microtime(&tv); 419 TIMEVAL_TO_TIMESPEC(&tv, tsp); 420 break; 421 case TSP_NSEC: 422 default: 423 nanotime(tsp); 424 break; 425 } 426 } 427 428 /* 429 * Set vnode attributes to VNOVAL 430 */ 431 void 432 vattr_null(vap) 433 struct vattr *vap; 434 { 435 436 vap->va_type = VNON; 437 vap->va_size = VNOVAL; 438 vap->va_bytes = VNOVAL; 439 vap->va_mode = VNOVAL; 440 vap->va_nlink = VNOVAL; 441 vap->va_uid = VNOVAL; 442 vap->va_gid = VNOVAL; 443 vap->va_fsid = VNOVAL; 444 vap->va_fileid = VNOVAL; 445 vap->va_blocksize = VNOVAL; 446 vap->va_rdev = VNOVAL; 447 vap->va_atime.tv_sec = VNOVAL; 448 vap->va_atime.tv_nsec = VNOVAL; 449 vap->va_mtime.tv_sec = VNOVAL; 450 vap->va_mtime.tv_nsec = VNOVAL; 451 vap->va_ctime.tv_sec = VNOVAL; 452 vap->va_ctime.tv_nsec = VNOVAL; 453 vap->va_flags = VNOVAL; 454 vap->va_gen = VNOVAL; 455 vap->va_vaflags = 0; 456 } 457 458 /* 459 * This routine is called when we have too many vnodes. It attempts 460 * to free <count> vnodes and will potentially free vnodes that still 461 * have VM backing store (VM backing store is typically the cause 462 * of a vnode blowout so we want to do this). Therefore, this operation 463 * is not considered cheap. 464 * 465 * A number of conditions may prevent a vnode from being reclaimed. 466 * the buffer cache may have references on the vnode, a directory 467 * vnode may still have references due to the namei cache representing 468 * underlying files, or the vnode may be in active use. It is not 469 * desireable to reuse such vnodes. These conditions may cause the 470 * number of vnodes to reach some minimum value regardless of what 471 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 472 */ 473 static int 474 vlrureclaim(struct mount *mp) 475 { 476 struct vnode *vp; 477 int done; 478 int trigger; 479 int usevnodes; 480 int count; 481 int gen; 482 483 /* 484 * Calculate the trigger point, don't allow user 485 * screwups to blow us up. This prevents us from 486 * recycling vnodes with lots of resident pages. We 487 * aren't trying to free memory, we are trying to 488 * free vnodes. 489 */ 490 usevnodes = desiredvnodes; 491 if (usevnodes <= 0) 492 usevnodes = 1; 493 trigger = vmstats.v_page_count * 2 / usevnodes; 494 495 done = 0; 496 gen = lwkt_gettoken(&mntvnode_token); 497 count = mp->mnt_nvnodelistsize / 10 + 1; 498 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) { 499 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 500 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 501 502 if (vp->v_type != VNON && 503 vp->v_type != VBAD && 504 VMIGHTFREE(vp) && /* critical path opt */ 505 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) 506 ) { 507 lwkt_gettoken(&vp->v_interlock); 508 if (lwkt_gentoken(&mntvnode_token, &gen) == 0) { 509 if (VMIGHTFREE(vp)) { 510 vgonel(vp, curthread); 511 done++; 512 } else { 513 lwkt_reltoken(&vp->v_interlock); 514 } 515 } else { 516 lwkt_reltoken(&vp->v_interlock); 517 } 518 } 519 --count; 520 } 521 lwkt_reltoken(&mntvnode_token); 522 return done; 523 } 524 525 /* 526 * Attempt to recycle vnodes in a context that is always safe to block. 527 * Calling vlrurecycle() from the bowels of file system code has some 528 * interesting deadlock problems. 529 */ 530 static struct thread *vnlruthread; 531 static int vnlruproc_sig; 532 533 static void 534 vnlru_proc(void) 535 { 536 struct mount *mp, *nmp; 537 int s; 538 int done; 539 struct thread *td = curthread; 540 541 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 542 SHUTDOWN_PRI_FIRST); 543 544 s = splbio(); 545 for (;;) { 546 kproc_suspend_loop(); 547 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) { 548 vnlruproc_sig = 0; 549 wakeup(&vnlruproc_sig); 550 tsleep(td, 0, "vlruwt", hz); 551 continue; 552 } 553 done = 0; 554 lwkt_gettoken(&mountlist_token); 555 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 556 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) { 557 nmp = TAILQ_NEXT(mp, mnt_list); 558 continue; 559 } 560 done += vlrureclaim(mp); 561 lwkt_gettoken(&mountlist_token); 562 nmp = TAILQ_NEXT(mp, mnt_list); 563 vfs_unbusy(mp, td); 564 } 565 lwkt_reltoken(&mountlist_token); 566 if (done == 0) { 567 vnlru_nowhere++; 568 tsleep(td, 0, "vlrup", hz * 3); 569 } 570 } 571 splx(s); 572 } 573 574 static struct kproc_desc vnlru_kp = { 575 "vnlru", 576 vnlru_proc, 577 &vnlruthread 578 }; 579 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 580 581 /* 582 * Routines having to do with the management of the vnode table. 583 */ 584 extern vop_t **dead_vnodeop_p; 585 586 /* 587 * Return the next vnode from the free list. 588 */ 589 int 590 getnewvnode(tag, mp, vops, vpp) 591 enum vtagtype tag; 592 struct mount *mp; 593 vop_t **vops; 594 struct vnode **vpp; 595 { 596 int s; 597 int gen; 598 int vgen; 599 struct thread *td = curthread; /* XXX */ 600 struct vnode *vp = NULL; 601 vm_object_t object; 602 603 s = splbio(); 604 605 /* 606 * Try to reuse vnodes if we hit the max. This situation only 607 * occurs in certain large-memory (2G+) situations. We cannot 608 * attempt to directly reclaim vnodes due to nasty recursion 609 * problems. 610 */ 611 while (numvnodes - freevnodes > desiredvnodes) { 612 if (vnlruproc_sig == 0) { 613 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 614 wakeup(vnlruthread); 615 } 616 tsleep(&vnlruproc_sig, 0, "vlruwk", hz); 617 } 618 619 620 /* 621 * Attempt to reuse a vnode already on the free list, allocating 622 * a new vnode if we can't find one or if we have not reached a 623 * good minimum for good LRU performance. 624 */ 625 gen = lwkt_gettoken(&vnode_free_list_token); 626 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) { 627 int count; 628 629 for (count = 0; count < freevnodes; count++) { 630 vp = TAILQ_FIRST(&vnode_free_list); 631 if (vp == NULL || vp->v_usecount) 632 panic("getnewvnode: free vnode isn't"); 633 634 /* 635 * Get the vnode's interlock, then re-obtain 636 * vnode_free_list_token in case we lost it. If we 637 * did lose it while getting the vnode interlock, 638 * even if we got it back again, then retry. 639 */ 640 vgen = lwkt_gettoken(&vp->v_interlock); 641 if (lwkt_gentoken(&vnode_free_list_token, &gen) != 0) { 642 --count; 643 lwkt_reltoken(&vp->v_interlock); 644 vp = NULL; 645 continue; 646 } 647 648 /* 649 * Whew! We have both tokens. Since we didn't lose 650 * the free list VFREE had better still be set. But 651 * we aren't out of the woods yet. We have to get 652 * the object (may block). If the vnode is not 653 * suitable then move it to the end of the list 654 * if we can. If we can't move it to the end of the 655 * list retry again. 656 */ 657 if ((VOP_GETVOBJECT(vp, &object) == 0 && 658 (object->resident_page_count || object->ref_count)) 659 ) { 660 if (lwkt_gentoken(&vp->v_interlock, &vgen) == 0 && 661 lwkt_gentoken(&vnode_free_list_token, &gen) == 0 662 ) { 663 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 664 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 665 } else { 666 --count; 667 } 668 lwkt_reltoken(&vp->v_interlock); 669 vp = NULL; 670 continue; 671 } 672 673 /* 674 * Still not out of the woods. VOBJECT might have 675 * blocked, if we did not retain our tokens we have 676 * to retry. 677 */ 678 if (lwkt_gentoken(&vp->v_interlock, &vgen) != 0 || 679 lwkt_gentoken(&vnode_free_list_token, &gen) != 0) { 680 --count; 681 vp = NULL; 682 continue; 683 } 684 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 685 KKASSERT(vp->v_flag & VFREE); 686 687 /* 688 * If we have children in the namecache we cannot 689 * reuse the vnode yet because it will break the 690 * namecache chain (YYY use nc_refs for the check?) 691 */ 692 if (TAILQ_FIRST(&vp->v_namecache)) { 693 if (cache_leaf_test(vp) < 0) { 694 lwkt_reltoken(&vp->v_interlock); 695 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 696 vp = NULL; 697 continue; 698 } 699 } 700 break; 701 } 702 } 703 704 if (vp) { 705 vp->v_flag |= VDOOMED; 706 vp->v_flag &= ~VFREE; 707 freevnodes--; 708 lwkt_reltoken(&vnode_free_list_token); 709 cache_purge(vp); /* YYY may block */ 710 vp->v_lease = NULL; 711 if (vp->v_type != VBAD) { 712 vgonel(vp, td); 713 } else { 714 lwkt_reltoken(&vp->v_interlock); 715 } 716 717 #ifdef INVARIANTS 718 { 719 int s; 720 721 if (vp->v_data) 722 panic("cleaned vnode isn't"); 723 s = splbio(); 724 if (vp->v_numoutput) 725 panic("Clean vnode has pending I/O's"); 726 splx(s); 727 } 728 #endif 729 vp->v_flag = 0; 730 vp->v_lastw = 0; 731 vp->v_lasta = 0; 732 vp->v_cstart = 0; 733 vp->v_clen = 0; 734 vp->v_socket = 0; 735 vp->v_writecount = 0; /* XXX */ 736 } else { 737 lwkt_reltoken(&vnode_free_list_token); 738 vp = (struct vnode *) zalloc(vnode_zone); 739 bzero((char *) vp, sizeof *vp); 740 lwkt_inittoken(&vp->v_interlock); 741 lwkt_inittoken(&vp->v_pollinfo.vpi_token); 742 vp->v_dd = vp; 743 cache_purge(vp); 744 TAILQ_INIT(&vp->v_namecache); 745 numvnodes++; 746 } 747 748 TAILQ_INIT(&vp->v_cleanblkhd); 749 TAILQ_INIT(&vp->v_dirtyblkhd); 750 vp->v_type = VNON; 751 vp->v_tag = tag; 752 vp->v_op = vops; 753 insmntque(vp, mp); 754 *vpp = vp; 755 vp->v_usecount = 1; 756 vp->v_data = 0; 757 splx(s); 758 759 vfs_object_create(vp, td); 760 return (0); 761 } 762 763 /* 764 * Move a vnode from one mount queue to another. 765 */ 766 static void 767 insmntque(vp, mp) 768 struct vnode *vp; 769 struct mount *mp; 770 { 771 772 lwkt_gettoken(&mntvnode_token); 773 /* 774 * Delete from old mount point vnode list, if on one. 775 */ 776 if (vp->v_mount != NULL) { 777 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0, 778 ("bad mount point vnode list size")); 779 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 780 vp->v_mount->mnt_nvnodelistsize--; 781 } 782 /* 783 * Insert into list of vnodes for the new mount point, if available. 784 */ 785 if ((vp->v_mount = mp) == NULL) { 786 lwkt_reltoken(&mntvnode_token); 787 return; 788 } 789 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 790 mp->mnt_nvnodelistsize++; 791 lwkt_reltoken(&mntvnode_token); 792 } 793 794 /* 795 * Update outstanding I/O count and do wakeup if requested. 796 */ 797 void 798 vwakeup(bp) 799 struct buf *bp; 800 { 801 struct vnode *vp; 802 803 bp->b_flags &= ~B_WRITEINPROG; 804 if ((vp = bp->b_vp)) { 805 vp->v_numoutput--; 806 if (vp->v_numoutput < 0) 807 panic("vwakeup: neg numoutput"); 808 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 809 vp->v_flag &= ~VBWAIT; 810 wakeup((caddr_t) &vp->v_numoutput); 811 } 812 } 813 } 814 815 /* 816 * Flush out and invalidate all buffers associated with a vnode. 817 * Called with the underlying object locked. 818 */ 819 int 820 vinvalbuf(struct vnode *vp, int flags, struct thread *td, 821 int slpflag, int slptimeo) 822 { 823 struct buf *bp; 824 struct buf *nbp, *blist; 825 int s, error; 826 vm_object_t object; 827 828 if (flags & V_SAVE) { 829 s = splbio(); 830 while (vp->v_numoutput) { 831 vp->v_flag |= VBWAIT; 832 error = tsleep((caddr_t)&vp->v_numoutput, 833 slpflag, "vinvlbuf", slptimeo); 834 if (error) { 835 splx(s); 836 return (error); 837 } 838 } 839 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 840 splx(s); 841 if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) 842 return (error); 843 s = splbio(); 844 if (vp->v_numoutput > 0 || 845 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 846 panic("vinvalbuf: dirty bufs"); 847 } 848 splx(s); 849 } 850 s = splbio(); 851 for (;;) { 852 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 853 if (!blist) 854 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 855 if (!blist) 856 break; 857 858 for (bp = blist; bp; bp = nbp) { 859 nbp = TAILQ_NEXT(bp, b_vnbufs); 860 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 861 error = BUF_TIMELOCK(bp, 862 LK_EXCLUSIVE | LK_SLEEPFAIL, 863 "vinvalbuf", slpflag, slptimeo); 864 if (error == ENOLCK) 865 break; 866 splx(s); 867 return (error); 868 } 869 /* 870 * XXX Since there are no node locks for NFS, I 871 * believe there is a slight chance that a delayed 872 * write will occur while sleeping just above, so 873 * check for it. Note that vfs_bio_awrite expects 874 * buffers to reside on a queue, while VOP_BWRITE and 875 * brelse do not. 876 */ 877 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 878 (flags & V_SAVE)) { 879 880 if (bp->b_vp == vp) { 881 if (bp->b_flags & B_CLUSTEROK) { 882 BUF_UNLOCK(bp); 883 vfs_bio_awrite(bp); 884 } else { 885 bremfree(bp); 886 bp->b_flags |= B_ASYNC; 887 VOP_BWRITE(bp->b_vp, bp); 888 } 889 } else { 890 bremfree(bp); 891 (void) VOP_BWRITE(bp->b_vp, bp); 892 } 893 break; 894 } 895 bremfree(bp); 896 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 897 bp->b_flags &= ~B_ASYNC; 898 brelse(bp); 899 } 900 } 901 902 /* 903 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 904 * have write I/O in-progress but if there is a VM object then the 905 * VM object can also have read-I/O in-progress. 906 */ 907 do { 908 while (vp->v_numoutput > 0) { 909 vp->v_flag |= VBWAIT; 910 tsleep(&vp->v_numoutput, 0, "vnvlbv", 0); 911 } 912 if (VOP_GETVOBJECT(vp, &object) == 0) { 913 while (object->paging_in_progress) 914 vm_object_pip_sleep(object, "vnvlbx"); 915 } 916 } while (vp->v_numoutput > 0); 917 918 splx(s); 919 920 /* 921 * Destroy the copy in the VM cache, too. 922 */ 923 lwkt_gettoken(&vp->v_interlock); 924 if (VOP_GETVOBJECT(vp, &object) == 0) { 925 vm_object_page_remove(object, 0, 0, 926 (flags & V_SAVE) ? TRUE : FALSE); 927 } 928 lwkt_reltoken(&vp->v_interlock); 929 930 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 931 panic("vinvalbuf: flush failed"); 932 return (0); 933 } 934 935 /* 936 * Truncate a file's buffer and pages to a specified length. This 937 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 938 * sync activity. 939 */ 940 int 941 vtruncbuf(struct vnode *vp, struct thread *td, off_t length, int blksize) 942 { 943 struct buf *bp; 944 struct buf *nbp; 945 int s, anyfreed; 946 int trunclbn; 947 948 /* 949 * Round up to the *next* lbn. 950 */ 951 trunclbn = (length + blksize - 1) / blksize; 952 953 s = splbio(); 954 restart: 955 anyfreed = 1; 956 for (;anyfreed;) { 957 anyfreed = 0; 958 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 959 nbp = TAILQ_NEXT(bp, b_vnbufs); 960 if (bp->b_lblkno >= trunclbn) { 961 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 962 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 963 goto restart; 964 } else { 965 bremfree(bp); 966 bp->b_flags |= (B_INVAL | B_RELBUF); 967 bp->b_flags &= ~B_ASYNC; 968 brelse(bp); 969 anyfreed = 1; 970 } 971 if (nbp && 972 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 973 (nbp->b_vp != vp) || 974 (nbp->b_flags & B_DELWRI))) { 975 goto restart; 976 } 977 } 978 } 979 980 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 981 nbp = TAILQ_NEXT(bp, b_vnbufs); 982 if (bp->b_lblkno >= trunclbn) { 983 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 984 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 985 goto restart; 986 } else { 987 bremfree(bp); 988 bp->b_flags |= (B_INVAL | B_RELBUF); 989 bp->b_flags &= ~B_ASYNC; 990 brelse(bp); 991 anyfreed = 1; 992 } 993 if (nbp && 994 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 995 (nbp->b_vp != vp) || 996 (nbp->b_flags & B_DELWRI) == 0)) { 997 goto restart; 998 } 999 } 1000 } 1001 } 1002 1003 if (length > 0) { 1004 restartsync: 1005 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1006 nbp = TAILQ_NEXT(bp, b_vnbufs); 1007 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 1008 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1009 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1010 goto restart; 1011 } else { 1012 bremfree(bp); 1013 if (bp->b_vp == vp) { 1014 bp->b_flags |= B_ASYNC; 1015 } else { 1016 bp->b_flags &= ~B_ASYNC; 1017 } 1018 VOP_BWRITE(bp->b_vp, bp); 1019 } 1020 goto restartsync; 1021 } 1022 1023 } 1024 } 1025 1026 while (vp->v_numoutput > 0) { 1027 vp->v_flag |= VBWAIT; 1028 tsleep(&vp->v_numoutput, 0, "vbtrunc", 0); 1029 } 1030 1031 splx(s); 1032 1033 vnode_pager_setsize(vp, length); 1034 1035 return (0); 1036 } 1037 1038 /* 1039 * Associate a buffer with a vnode. 1040 */ 1041 void 1042 bgetvp(vp, bp) 1043 struct vnode *vp; 1044 struct buf *bp; 1045 { 1046 int s; 1047 1048 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 1049 1050 vhold(vp); 1051 bp->b_vp = vp; 1052 bp->b_dev = vn_todev(vp); 1053 /* 1054 * Insert onto list for new vnode. 1055 */ 1056 s = splbio(); 1057 bp->b_xflags |= BX_VNCLEAN; 1058 bp->b_xflags &= ~BX_VNDIRTY; 1059 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 1060 splx(s); 1061 } 1062 1063 /* 1064 * Disassociate a buffer from a vnode. 1065 */ 1066 void 1067 brelvp(bp) 1068 struct buf *bp; 1069 { 1070 struct vnode *vp; 1071 struct buflists *listheadp; 1072 int s; 1073 1074 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1075 1076 /* 1077 * Delete from old vnode list, if on one. 1078 */ 1079 vp = bp->b_vp; 1080 s = splbio(); 1081 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1082 if (bp->b_xflags & BX_VNDIRTY) 1083 listheadp = &vp->v_dirtyblkhd; 1084 else 1085 listheadp = &vp->v_cleanblkhd; 1086 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1087 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1088 } 1089 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1090 vp->v_flag &= ~VONWORKLST; 1091 LIST_REMOVE(vp, v_synclist); 1092 } 1093 splx(s); 1094 bp->b_vp = (struct vnode *) 0; 1095 vdrop(vp); 1096 } 1097 1098 /* 1099 * The workitem queue. 1100 * 1101 * It is useful to delay writes of file data and filesystem metadata 1102 * for tens of seconds so that quickly created and deleted files need 1103 * not waste disk bandwidth being created and removed. To realize this, 1104 * we append vnodes to a "workitem" queue. When running with a soft 1105 * updates implementation, most pending metadata dependencies should 1106 * not wait for more than a few seconds. Thus, mounted on block devices 1107 * are delayed only about a half the time that file data is delayed. 1108 * Similarly, directory updates are more critical, so are only delayed 1109 * about a third the time that file data is delayed. Thus, there are 1110 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 1111 * one each second (driven off the filesystem syncer process). The 1112 * syncer_delayno variable indicates the next queue that is to be processed. 1113 * Items that need to be processed soon are placed in this queue: 1114 * 1115 * syncer_workitem_pending[syncer_delayno] 1116 * 1117 * A delay of fifteen seconds is done by placing the request fifteen 1118 * entries later in the queue: 1119 * 1120 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 1121 * 1122 */ 1123 1124 /* 1125 * Add an item to the syncer work queue. 1126 */ 1127 static void 1128 vn_syncer_add_to_worklist(struct vnode *vp, int delay) 1129 { 1130 int s, slot; 1131 1132 s = splbio(); 1133 1134 if (vp->v_flag & VONWORKLST) { 1135 LIST_REMOVE(vp, v_synclist); 1136 } 1137 1138 if (delay > syncer_maxdelay - 2) 1139 delay = syncer_maxdelay - 2; 1140 slot = (syncer_delayno + delay) & syncer_mask; 1141 1142 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 1143 vp->v_flag |= VONWORKLST; 1144 splx(s); 1145 } 1146 1147 struct thread *updatethread; 1148 static void sched_sync (void); 1149 static struct kproc_desc up_kp = { 1150 "syncer", 1151 sched_sync, 1152 &updatethread 1153 }; 1154 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1155 1156 /* 1157 * System filesystem synchronizer daemon. 1158 */ 1159 void 1160 sched_sync(void) 1161 { 1162 struct synclist *slp; 1163 struct vnode *vp; 1164 long starttime; 1165 int s; 1166 struct thread *td = curthread; 1167 1168 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 1169 SHUTDOWN_PRI_LAST); 1170 1171 for (;;) { 1172 kproc_suspend_loop(); 1173 1174 starttime = time_second; 1175 1176 /* 1177 * Push files whose dirty time has expired. Be careful 1178 * of interrupt race on slp queue. 1179 */ 1180 s = splbio(); 1181 slp = &syncer_workitem_pending[syncer_delayno]; 1182 syncer_delayno += 1; 1183 if (syncer_delayno == syncer_maxdelay) 1184 syncer_delayno = 0; 1185 splx(s); 1186 1187 while ((vp = LIST_FIRST(slp)) != NULL) { 1188 if (VOP_ISLOCKED(vp, NULL) == 0) { 1189 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1190 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1191 VOP_UNLOCK(vp, 0, td); 1192 } 1193 s = splbio(); 1194 if (LIST_FIRST(slp) == vp) { 1195 /* 1196 * Note: v_tag VT_VFS vps can remain on the 1197 * worklist too with no dirty blocks, but 1198 * since sync_fsync() moves it to a different 1199 * slot we are safe. 1200 */ 1201 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1202 !vn_isdisk(vp, NULL)) 1203 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1204 /* 1205 * Put us back on the worklist. The worklist 1206 * routine will remove us from our current 1207 * position and then add us back in at a later 1208 * position. 1209 */ 1210 vn_syncer_add_to_worklist(vp, syncdelay); 1211 } 1212 splx(s); 1213 } 1214 1215 /* 1216 * Do soft update processing. 1217 */ 1218 if (bioops.io_sync) 1219 (*bioops.io_sync)(NULL); 1220 1221 /* 1222 * The variable rushjob allows the kernel to speed up the 1223 * processing of the filesystem syncer process. A rushjob 1224 * value of N tells the filesystem syncer to process the next 1225 * N seconds worth of work on its queue ASAP. Currently rushjob 1226 * is used by the soft update code to speed up the filesystem 1227 * syncer process when the incore state is getting so far 1228 * ahead of the disk that the kernel memory pool is being 1229 * threatened with exhaustion. 1230 */ 1231 if (rushjob > 0) { 1232 rushjob -= 1; 1233 continue; 1234 } 1235 /* 1236 * If it has taken us less than a second to process the 1237 * current work, then wait. Otherwise start right over 1238 * again. We can still lose time if any single round 1239 * takes more than two seconds, but it does not really 1240 * matter as we are just trying to generally pace the 1241 * filesystem activity. 1242 */ 1243 if (time_second == starttime) 1244 tsleep(&lbolt, 0, "syncer", 0); 1245 } 1246 } 1247 1248 /* 1249 * Request the syncer daemon to speed up its work. 1250 * We never push it to speed up more than half of its 1251 * normal turn time, otherwise it could take over the cpu. 1252 * 1253 * YYY wchan field protected by the BGL. 1254 */ 1255 int 1256 speedup_syncer() 1257 { 1258 crit_enter(); 1259 if (updatethread->td_wchan == &lbolt) { /* YYY */ 1260 unsleep(updatethread); 1261 lwkt_schedule(updatethread); 1262 } 1263 crit_exit(); 1264 if (rushjob < syncdelay / 2) { 1265 rushjob += 1; 1266 stat_rush_requests += 1; 1267 return (1); 1268 } 1269 return(0); 1270 } 1271 1272 /* 1273 * Associate a p-buffer with a vnode. 1274 * 1275 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1276 * with the buffer. i.e. the bp has not been linked into the vnode or 1277 * ref-counted. 1278 */ 1279 void 1280 pbgetvp(vp, bp) 1281 struct vnode *vp; 1282 struct buf *bp; 1283 { 1284 1285 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1286 1287 bp->b_vp = vp; 1288 bp->b_flags |= B_PAGING; 1289 bp->b_dev = vn_todev(vp); 1290 } 1291 1292 /* 1293 * Disassociate a p-buffer from a vnode. 1294 */ 1295 void 1296 pbrelvp(bp) 1297 struct buf *bp; 1298 { 1299 1300 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1301 1302 /* XXX REMOVE ME */ 1303 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 1304 panic( 1305 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1306 bp, 1307 (int)bp->b_flags 1308 ); 1309 } 1310 bp->b_vp = (struct vnode *) 0; 1311 bp->b_flags &= ~B_PAGING; 1312 } 1313 1314 void 1315 pbreassignbuf(bp, newvp) 1316 struct buf *bp; 1317 struct vnode *newvp; 1318 { 1319 if ((bp->b_flags & B_PAGING) == 0) { 1320 panic( 1321 "pbreassignbuf() on non phys bp %p", 1322 bp 1323 ); 1324 } 1325 bp->b_vp = newvp; 1326 } 1327 1328 /* 1329 * Reassign a buffer from one vnode to another. 1330 * Used to assign file specific control information 1331 * (indirect blocks) to the vnode to which they belong. 1332 */ 1333 void 1334 reassignbuf(bp, newvp) 1335 struct buf *bp; 1336 struct vnode *newvp; 1337 { 1338 struct buflists *listheadp; 1339 int delay; 1340 int s; 1341 1342 if (newvp == NULL) { 1343 printf("reassignbuf: NULL"); 1344 return; 1345 } 1346 ++reassignbufcalls; 1347 1348 /* 1349 * B_PAGING flagged buffers cannot be reassigned because their vp 1350 * is not fully linked in. 1351 */ 1352 if (bp->b_flags & B_PAGING) 1353 panic("cannot reassign paging buffer"); 1354 1355 s = splbio(); 1356 /* 1357 * Delete from old vnode list, if on one. 1358 */ 1359 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1360 if (bp->b_xflags & BX_VNDIRTY) 1361 listheadp = &bp->b_vp->v_dirtyblkhd; 1362 else 1363 listheadp = &bp->b_vp->v_cleanblkhd; 1364 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1365 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1366 if (bp->b_vp != newvp) { 1367 vdrop(bp->b_vp); 1368 bp->b_vp = NULL; /* for clarification */ 1369 } 1370 } 1371 /* 1372 * If dirty, put on list of dirty buffers; otherwise insert onto list 1373 * of clean buffers. 1374 */ 1375 if (bp->b_flags & B_DELWRI) { 1376 struct buf *tbp; 1377 1378 listheadp = &newvp->v_dirtyblkhd; 1379 if ((newvp->v_flag & VONWORKLST) == 0) { 1380 switch (newvp->v_type) { 1381 case VDIR: 1382 delay = dirdelay; 1383 break; 1384 case VCHR: 1385 case VBLK: 1386 if (newvp->v_specmountpoint != NULL) { 1387 delay = metadelay; 1388 break; 1389 } 1390 /* fall through */ 1391 default: 1392 delay = filedelay; 1393 } 1394 vn_syncer_add_to_worklist(newvp, delay); 1395 } 1396 bp->b_xflags |= BX_VNDIRTY; 1397 tbp = TAILQ_FIRST(listheadp); 1398 if (tbp == NULL || 1399 bp->b_lblkno == 0 || 1400 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1401 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1402 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1403 ++reassignbufsortgood; 1404 } else if (bp->b_lblkno < 0) { 1405 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1406 ++reassignbufsortgood; 1407 } else if (reassignbufmethod == 1) { 1408 /* 1409 * New sorting algorithm, only handle sequential case, 1410 * otherwise append to end (but before metadata) 1411 */ 1412 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1413 (tbp->b_xflags & BX_VNDIRTY)) { 1414 /* 1415 * Found the best place to insert the buffer 1416 */ 1417 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1418 ++reassignbufsortgood; 1419 } else { 1420 /* 1421 * Missed, append to end, but before meta-data. 1422 * We know that the head buffer in the list is 1423 * not meta-data due to prior conditionals. 1424 * 1425 * Indirect effects: NFS second stage write 1426 * tends to wind up here, giving maximum 1427 * distance between the unstable write and the 1428 * commit rpc. 1429 */ 1430 tbp = TAILQ_LAST(listheadp, buflists); 1431 while (tbp && tbp->b_lblkno < 0) 1432 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1433 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1434 ++reassignbufsortbad; 1435 } 1436 } else { 1437 /* 1438 * Old sorting algorithm, scan queue and insert 1439 */ 1440 struct buf *ttbp; 1441 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1442 (ttbp->b_lblkno < bp->b_lblkno)) { 1443 ++reassignbufloops; 1444 tbp = ttbp; 1445 } 1446 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1447 } 1448 } else { 1449 bp->b_xflags |= BX_VNCLEAN; 1450 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1451 if ((newvp->v_flag & VONWORKLST) && 1452 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1453 newvp->v_flag &= ~VONWORKLST; 1454 LIST_REMOVE(newvp, v_synclist); 1455 } 1456 } 1457 if (bp->b_vp != newvp) { 1458 bp->b_vp = newvp; 1459 vhold(bp->b_vp); 1460 } 1461 splx(s); 1462 } 1463 1464 /* 1465 * Create a vnode for a block device. 1466 * Used for mounting the root file system. 1467 */ 1468 int 1469 bdevvp(dev, vpp) 1470 dev_t dev; 1471 struct vnode **vpp; 1472 { 1473 struct vnode *vp; 1474 struct vnode *nvp; 1475 int error; 1476 1477 if (dev == NODEV) { 1478 *vpp = NULLVP; 1479 return (ENXIO); 1480 } 1481 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1482 if (error) { 1483 *vpp = NULLVP; 1484 return (error); 1485 } 1486 vp = nvp; 1487 vp->v_type = VBLK; 1488 addalias(vp, dev); 1489 *vpp = vp; 1490 return (0); 1491 } 1492 1493 /* 1494 * Add a vnode to the alias list hung off the dev_t. 1495 * 1496 * The reason for this gunk is that multiple vnodes can reference 1497 * the same physical device, so checking vp->v_usecount to see 1498 * how many users there are is inadequate; the v_usecount for 1499 * the vnodes need to be accumulated. vcount() does that. 1500 */ 1501 void 1502 addaliasu(struct vnode *nvp, udev_t nvp_rdev) 1503 { 1504 dev_t dev; 1505 1506 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1507 panic("addaliasu on non-special vnode"); 1508 dev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0); 1509 if (dev != NODEV) { 1510 nvp->v_rdev = dev; 1511 addalias(nvp, dev); 1512 } else 1513 nvp->v_rdev = NULL; 1514 } 1515 1516 void 1517 addalias(struct vnode *nvp, dev_t dev) 1518 { 1519 1520 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1521 panic("addalias on non-special vnode"); 1522 1523 nvp->v_rdev = dev; 1524 lwkt_gettoken(&spechash_token); 1525 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1526 lwkt_reltoken(&spechash_token); 1527 } 1528 1529 /* 1530 * Grab a particular vnode from the free list, increment its 1531 * reference count and lock it. The vnode lock bit is set if the 1532 * vnode is being eliminated in vgone. The process is awakened 1533 * when the transition is completed, and an error returned to 1534 * indicate that the vnode is no longer usable (possibly having 1535 * been changed to a new file system type). 1536 */ 1537 int 1538 vget(vp, flags, td) 1539 struct vnode *vp; 1540 int flags; 1541 struct thread *td; 1542 { 1543 int error; 1544 1545 /* 1546 * If the vnode is in the process of being cleaned out for 1547 * another use, we wait for the cleaning to finish and then 1548 * return failure. Cleaning is determined by checking that 1549 * the VXLOCK flag is set. 1550 */ 1551 if (vp->v_flag & VXLOCK) { 1552 if (vp->v_vxproc == curproc) { 1553 #if 0 1554 /* this can now occur in normal operation */ 1555 log(LOG_INFO, "VXLOCK interlock avoided\n"); 1556 #endif 1557 } else { 1558 vp->v_flag |= VXWANT; 1559 tsleep((caddr_t)vp, 0, "vget", 0); 1560 return (ENOENT); 1561 } 1562 } 1563 1564 /* 1565 * Bump v_usecount to prevent the vnode from being cleaned. If the 1566 * vnode gets cleaned unexpectedly we could wind up calling lockmgr 1567 * on a lock embedded in an inode which is then ripped out from 1568 * it. 1569 */ 1570 vp->v_usecount++; /* XXX MP */ 1571 1572 if ((flags & LK_INTERLOCK) == 0) { 1573 lwkt_gettoken(&vp->v_interlock); 1574 } 1575 1576 if (VSHOULDBUSY(vp)) 1577 vbusy(vp); 1578 if (flags & LK_TYPE_MASK) { 1579 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) { 1580 /* 1581 * must expand vrele here because we do not want 1582 * to call VOP_INACTIVE if the reference count 1583 * drops back to zero since it was never really 1584 * active. We must remove it from the free list 1585 * before sleeping so that multiple processes do 1586 * not try to recycle it. 1587 */ 1588 lwkt_gettoken(&vp->v_interlock); 1589 vp->v_usecount--; 1590 if (VSHOULDFREE(vp)) 1591 vfree(vp); 1592 else 1593 vlruvp(vp); 1594 lwkt_reltoken(&vp->v_interlock); 1595 } 1596 return (error); 1597 } 1598 lwkt_reltoken(&vp->v_interlock); 1599 return (0); 1600 } 1601 1602 void 1603 vref(struct vnode *vp) 1604 { 1605 vp->v_usecount++; /* XXX MP */ 1606 } 1607 1608 /* 1609 * Vnode put/release. 1610 * If count drops to zero, call inactive routine and return to freelist. 1611 */ 1612 void 1613 vrele(struct vnode *vp) 1614 { 1615 struct thread *td = curthread; /* XXX */ 1616 1617 KASSERT(vp != NULL, ("vrele: null vp")); 1618 1619 lwkt_gettoken(&vp->v_interlock); 1620 1621 if (vp->v_usecount > 1) { 1622 1623 vp->v_usecount--; 1624 lwkt_reltoken(&vp->v_interlock); 1625 1626 return; 1627 } 1628 1629 if (vp->v_usecount == 1) { 1630 vp->v_usecount--; 1631 /* 1632 * We must call VOP_INACTIVE with the node locked. 1633 * If we are doing a vpu, the node is already locked, 1634 * but, in the case of vrele, we must explicitly lock 1635 * the vnode before calling VOP_INACTIVE 1636 */ 1637 1638 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) 1639 VOP_INACTIVE(vp, td); 1640 if (VSHOULDFREE(vp)) 1641 vfree(vp); 1642 else 1643 vlruvp(vp); 1644 } else { 1645 #ifdef DIAGNOSTIC 1646 vprint("vrele: negative ref count", vp); 1647 lwkt_reltoken(&vp->v_interlock); 1648 #endif 1649 panic("vrele: negative ref cnt"); 1650 } 1651 } 1652 1653 void 1654 vput(struct vnode *vp) 1655 { 1656 struct thread *td = curthread; /* XXX */ 1657 1658 KASSERT(vp != NULL, ("vput: null vp")); 1659 1660 lwkt_gettoken(&vp->v_interlock); 1661 1662 if (vp->v_usecount > 1) { 1663 vp->v_usecount--; 1664 VOP_UNLOCK(vp, LK_INTERLOCK, td); 1665 return; 1666 } 1667 1668 if (vp->v_usecount == 1) { 1669 vp->v_usecount--; 1670 /* 1671 * We must call VOP_INACTIVE with the node locked. 1672 * If we are doing a vpu, the node is already locked, 1673 * so we just need to release the vnode mutex. 1674 */ 1675 lwkt_reltoken(&vp->v_interlock); 1676 VOP_INACTIVE(vp, td); 1677 if (VSHOULDFREE(vp)) 1678 vfree(vp); 1679 else 1680 vlruvp(vp); 1681 } else { 1682 #ifdef DIAGNOSTIC 1683 vprint("vput: negative ref count", vp); 1684 #endif 1685 panic("vput: negative ref cnt"); 1686 } 1687 } 1688 1689 /* 1690 * Somebody doesn't want the vnode recycled. 1691 */ 1692 void 1693 vhold(vp) 1694 struct vnode *vp; 1695 { 1696 int s; 1697 1698 s = splbio(); 1699 vp->v_holdcnt++; 1700 if (VSHOULDBUSY(vp)) 1701 vbusy(vp); 1702 splx(s); 1703 } 1704 1705 /* 1706 * One less who cares about this vnode. 1707 */ 1708 void 1709 vdrop(vp) 1710 struct vnode *vp; 1711 { 1712 int s; 1713 1714 s = splbio(); 1715 if (vp->v_holdcnt <= 0) 1716 panic("vdrop: holdcnt"); 1717 vp->v_holdcnt--; 1718 if (VSHOULDFREE(vp)) 1719 vfree(vp); 1720 splx(s); 1721 } 1722 1723 /* 1724 * Remove any vnodes in the vnode table belonging to mount point mp. 1725 * 1726 * If FORCECLOSE is not specified, there should not be any active ones, 1727 * return error if any are found (nb: this is a user error, not a 1728 * system error). If FORCECLOSE is specified, detach any active vnodes 1729 * that are found. 1730 * 1731 * If WRITECLOSE is set, only flush out regular file vnodes open for 1732 * writing. 1733 * 1734 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 1735 * 1736 * `rootrefs' specifies the base reference count for the root vnode 1737 * of this filesystem. The root vnode is considered busy if its 1738 * v_usecount exceeds this value. On a successful return, vflush() 1739 * will call vrele() on the root vnode exactly rootrefs times. 1740 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 1741 * be zero. 1742 */ 1743 #ifdef DIAGNOSTIC 1744 static int busyprt = 0; /* print out busy vnodes */ 1745 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1746 #endif 1747 1748 int 1749 vflush(mp, rootrefs, flags) 1750 struct mount *mp; 1751 int rootrefs; 1752 int flags; 1753 { 1754 struct thread *td = curthread; /* XXX */ 1755 struct vnode *vp, *nvp, *rootvp = NULL; 1756 struct vattr vattr; 1757 int busy = 0, error; 1758 1759 if (rootrefs > 0) { 1760 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1761 ("vflush: bad args")); 1762 /* 1763 * Get the filesystem root vnode. We can vput() it 1764 * immediately, since with rootrefs > 0, it won't go away. 1765 */ 1766 if ((error = VFS_ROOT(mp, &rootvp)) != 0) 1767 return (error); 1768 vput(rootvp); 1769 } 1770 lwkt_gettoken(&mntvnode_token); 1771 loop: 1772 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) { 1773 /* 1774 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1775 * Start over if it has (it won't be on the list anymore). 1776 */ 1777 if (vp->v_mount != mp) 1778 goto loop; 1779 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1780 1781 lwkt_gettoken(&vp->v_interlock); 1782 /* 1783 * Skip over a vnodes marked VSYSTEM. 1784 */ 1785 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1786 lwkt_reltoken(&vp->v_interlock); 1787 continue; 1788 } 1789 /* 1790 * If WRITECLOSE is set, flush out unlinked but still open 1791 * files (even if open only for reading) and regular file 1792 * vnodes open for writing. 1793 */ 1794 if ((flags & WRITECLOSE) && 1795 (vp->v_type == VNON || 1796 (VOP_GETATTR(vp, &vattr, td) == 0 && 1797 vattr.va_nlink > 0)) && 1798 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1799 lwkt_reltoken(&vp->v_interlock); 1800 continue; 1801 } 1802 1803 /* 1804 * With v_usecount == 0, all we need to do is clear out the 1805 * vnode data structures and we are done. 1806 */ 1807 if (vp->v_usecount == 0) { 1808 lwkt_reltoken(&mntvnode_token); 1809 vgonel(vp, td); 1810 lwkt_gettoken(&mntvnode_token); 1811 continue; 1812 } 1813 1814 /* 1815 * If FORCECLOSE is set, forcibly close the vnode. For block 1816 * or character devices, revert to an anonymous device. For 1817 * all other files, just kill them. 1818 */ 1819 if (flags & FORCECLOSE) { 1820 lwkt_reltoken(&mntvnode_token); 1821 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1822 vgonel(vp, td); 1823 } else { 1824 vclean(vp, 0, td); 1825 vp->v_op = spec_vnodeop_p; 1826 insmntque(vp, (struct mount *) 0); 1827 } 1828 lwkt_gettoken(&mntvnode_token); 1829 continue; 1830 } 1831 #ifdef DIAGNOSTIC 1832 if (busyprt) 1833 vprint("vflush: busy vnode", vp); 1834 #endif 1835 lwkt_reltoken(&vp->v_interlock); 1836 busy++; 1837 } 1838 lwkt_reltoken(&mntvnode_token); 1839 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1840 /* 1841 * If just the root vnode is busy, and if its refcount 1842 * is equal to `rootrefs', then go ahead and kill it. 1843 */ 1844 lwkt_gettoken(&rootvp->v_interlock); 1845 KASSERT(busy > 0, ("vflush: not busy")); 1846 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs")); 1847 if (busy == 1 && rootvp->v_usecount == rootrefs) { 1848 vgonel(rootvp, td); 1849 busy = 0; 1850 } else 1851 lwkt_reltoken(&rootvp->v_interlock); 1852 } 1853 if (busy) 1854 return (EBUSY); 1855 for (; rootrefs > 0; rootrefs--) 1856 vrele(rootvp); 1857 return (0); 1858 } 1859 1860 /* 1861 * We do not want to recycle the vnode too quickly. 1862 * 1863 * XXX we can't move vp's around the nvnodelist without really screwing 1864 * up the efficiency of filesystem SYNC and friends. This code is 1865 * disabled until we fix the syncing code's scanning algorithm. 1866 */ 1867 static void 1868 vlruvp(struct vnode *vp) 1869 { 1870 #if 0 1871 struct mount *mp; 1872 1873 if ((mp = vp->v_mount) != NULL) { 1874 lwkt_gettoken(&mntvnode_token); 1875 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1876 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1877 lwkt_reltoken(&mntvnode_token); 1878 } 1879 #endif 1880 } 1881 1882 /* 1883 * Disassociate the underlying file system from a vnode. 1884 */ 1885 static void 1886 vclean(struct vnode *vp, int flags, struct thread *td) 1887 { 1888 int active; 1889 1890 /* 1891 * Check to see if the vnode is in use. If so we have to reference it 1892 * before we clean it out so that its count cannot fall to zero and 1893 * generate a race against ourselves to recycle it. 1894 */ 1895 if ((active = vp->v_usecount)) 1896 vp->v_usecount++; 1897 1898 /* 1899 * Prevent the vnode from being recycled or brought into use while we 1900 * clean it out. 1901 */ 1902 if (vp->v_flag & VXLOCK) 1903 panic("vclean: deadlock"); 1904 vp->v_flag |= VXLOCK; 1905 vp->v_vxproc = curproc; 1906 /* 1907 * Even if the count is zero, the VOP_INACTIVE routine may still 1908 * have the object locked while it cleans it out. The VOP_LOCK 1909 * ensures that the VOP_INACTIVE routine is done with its work. 1910 * For active vnodes, it ensures that no other activity can 1911 * occur while the underlying object is being cleaned out. 1912 */ 1913 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td); 1914 1915 /* 1916 * Clean out any buffers associated with the vnode. 1917 */ 1918 vinvalbuf(vp, V_SAVE, td, 0, 0); 1919 1920 VOP_DESTROYVOBJECT(vp); 1921 1922 /* 1923 * If purging an active vnode, it must be closed and 1924 * deactivated before being reclaimed. Note that the 1925 * VOP_INACTIVE will unlock the vnode. 1926 */ 1927 if (active) { 1928 if (flags & DOCLOSE) 1929 VOP_CLOSE(vp, FNONBLOCK, td); 1930 VOP_INACTIVE(vp, td); 1931 } else { 1932 /* 1933 * Any other processes trying to obtain this lock must first 1934 * wait for VXLOCK to clear, then call the new lock operation. 1935 */ 1936 VOP_UNLOCK(vp, 0, td); 1937 } 1938 /* 1939 * Reclaim the vnode. 1940 */ 1941 if (VOP_RECLAIM(vp, td)) 1942 panic("vclean: cannot reclaim"); 1943 1944 if (active) { 1945 /* 1946 * Inline copy of vrele() since VOP_INACTIVE 1947 * has already been called. 1948 */ 1949 lwkt_gettoken(&vp->v_interlock); 1950 if (--vp->v_usecount <= 0) { 1951 #ifdef DIAGNOSTIC 1952 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1953 vprint("vclean: bad ref count", vp); 1954 panic("vclean: ref cnt"); 1955 } 1956 #endif 1957 vfree(vp); 1958 } 1959 lwkt_reltoken(&vp->v_interlock); 1960 } 1961 1962 cache_purge(vp); 1963 vp->v_vnlock = NULL; 1964 1965 if (VSHOULDFREE(vp)) 1966 vfree(vp); 1967 1968 /* 1969 * Done with purge, notify sleepers of the grim news. 1970 */ 1971 vp->v_op = dead_vnodeop_p; 1972 vn_pollgone(vp); 1973 vp->v_tag = VT_NON; 1974 vp->v_flag &= ~VXLOCK; 1975 vp->v_vxproc = NULL; 1976 if (vp->v_flag & VXWANT) { 1977 vp->v_flag &= ~VXWANT; 1978 wakeup((caddr_t) vp); 1979 } 1980 } 1981 1982 /* 1983 * Eliminate all activity associated with the requested vnode 1984 * and with all vnodes aliased to the requested vnode. 1985 */ 1986 int 1987 vop_revoke(ap) 1988 struct vop_revoke_args /* { 1989 struct vnode *a_vp; 1990 int a_flags; 1991 } */ *ap; 1992 { 1993 struct vnode *vp, *vq; 1994 dev_t dev; 1995 1996 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 1997 1998 vp = ap->a_vp; 1999 /* 2000 * If a vgone (or vclean) is already in progress, 2001 * wait until it is done and return. 2002 */ 2003 if (vp->v_flag & VXLOCK) { 2004 vp->v_flag |= VXWANT; 2005 lwkt_reltoken(&vp->v_interlock); 2006 tsleep((caddr_t)vp, 0, "vop_revokeall", 0); 2007 return (0); 2008 } 2009 dev = vp->v_rdev; 2010 for (;;) { 2011 lwkt_gettoken(&spechash_token); 2012 vq = SLIST_FIRST(&dev->si_hlist); 2013 lwkt_reltoken(&spechash_token); 2014 if (!vq) 2015 break; 2016 vgone(vq); 2017 } 2018 return (0); 2019 } 2020 2021 /* 2022 * Recycle an unused vnode to the front of the free list. 2023 * Release the passed interlock if the vnode will be recycled. 2024 */ 2025 int 2026 vrecycle(struct vnode *vp, struct lwkt_token *inter_lkp, struct thread *td) 2027 { 2028 lwkt_gettoken(&vp->v_interlock); 2029 if (vp->v_usecount == 0) { 2030 if (inter_lkp) { 2031 lwkt_reltoken(inter_lkp); 2032 } 2033 vgonel(vp, td); 2034 return (1); 2035 } 2036 lwkt_reltoken(&vp->v_interlock); 2037 return (0); 2038 } 2039 2040 /* 2041 * Eliminate all activity associated with a vnode 2042 * in preparation for reuse. 2043 */ 2044 void 2045 vgone(struct vnode *vp) 2046 { 2047 struct thread *td = curthread; /* XXX */ 2048 2049 lwkt_gettoken(&vp->v_interlock); 2050 vgonel(vp, td); 2051 } 2052 2053 /* 2054 * vgone, with the vp interlock held. 2055 */ 2056 void 2057 vgonel(struct vnode *vp, struct thread *td) 2058 { 2059 int s; 2060 2061 /* 2062 * If a vgone (or vclean) is already in progress, 2063 * wait until it is done and return. 2064 */ 2065 if (vp->v_flag & VXLOCK) { 2066 vp->v_flag |= VXWANT; 2067 lwkt_reltoken(&vp->v_interlock); 2068 tsleep((caddr_t)vp, 0, "vgone", 0); 2069 return; 2070 } 2071 2072 /* 2073 * Clean out the filesystem specific data. 2074 */ 2075 vclean(vp, DOCLOSE, td); 2076 lwkt_gettoken(&vp->v_interlock); 2077 2078 /* 2079 * Delete from old mount point vnode list, if on one. 2080 */ 2081 if (vp->v_mount != NULL) 2082 insmntque(vp, (struct mount *)0); 2083 /* 2084 * If special device, remove it from special device alias list 2085 * if it is on one. 2086 */ 2087 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) { 2088 lwkt_gettoken(&spechash_token); 2089 SLIST_REMOVE(&vp->v_hashchain, vp, vnode, v_specnext); 2090 freedev(vp->v_rdev); 2091 lwkt_reltoken(&spechash_token); 2092 vp->v_rdev = NULL; 2093 } 2094 2095 /* 2096 * If it is on the freelist and not already at the head, 2097 * move it to the head of the list. The test of the 2098 * VDOOMED flag and the reference count of zero is because 2099 * it will be removed from the free list by getnewvnode, 2100 * but will not have its reference count incremented until 2101 * after calling vgone. If the reference count were 2102 * incremented first, vgone would (incorrectly) try to 2103 * close the previous instance of the underlying object. 2104 */ 2105 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 2106 s = splbio(); 2107 lwkt_gettoken(&vnode_free_list_token); 2108 if (vp->v_flag & VFREE) 2109 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2110 else 2111 freevnodes++; 2112 vp->v_flag |= VFREE; 2113 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2114 lwkt_reltoken(&vnode_free_list_token); 2115 splx(s); 2116 } 2117 2118 vp->v_type = VBAD; 2119 lwkt_reltoken(&vp->v_interlock); 2120 } 2121 2122 /* 2123 * Lookup a vnode by device number. 2124 */ 2125 int 2126 vfinddev(dev, type, vpp) 2127 dev_t dev; 2128 enum vtype type; 2129 struct vnode **vpp; 2130 { 2131 struct vnode *vp; 2132 2133 lwkt_gettoken(&spechash_token); 2134 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 2135 if (type == vp->v_type) { 2136 *vpp = vp; 2137 lwkt_reltoken(&spechash_token); 2138 return (1); 2139 } 2140 } 2141 lwkt_reltoken(&spechash_token); 2142 return (0); 2143 } 2144 2145 /* 2146 * Calculate the total number of references to a special device. 2147 */ 2148 int 2149 vcount(vp) 2150 struct vnode *vp; 2151 { 2152 struct vnode *vq; 2153 int count; 2154 2155 count = 0; 2156 lwkt_gettoken(&spechash_token); 2157 SLIST_FOREACH(vq, &vp->v_hashchain, v_specnext) 2158 count += vq->v_usecount; 2159 lwkt_reltoken(&spechash_token); 2160 return (count); 2161 } 2162 2163 /* 2164 * Same as above, but using the dev_t as argument 2165 */ 2166 2167 int 2168 count_dev(dev) 2169 dev_t dev; 2170 { 2171 struct vnode *vp; 2172 2173 vp = SLIST_FIRST(&dev->si_hlist); 2174 if (vp == NULL) 2175 return (0); 2176 return(vcount(vp)); 2177 } 2178 2179 /* 2180 * Print out a description of a vnode. 2181 */ 2182 static char *typename[] = 2183 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 2184 2185 void 2186 vprint(label, vp) 2187 char *label; 2188 struct vnode *vp; 2189 { 2190 char buf[96]; 2191 2192 if (label != NULL) 2193 printf("%s: %p: ", label, (void *)vp); 2194 else 2195 printf("%p: ", (void *)vp); 2196 printf("type %s, usecount %d, writecount %d, refcount %d,", 2197 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 2198 vp->v_holdcnt); 2199 buf[0] = '\0'; 2200 if (vp->v_flag & VROOT) 2201 strcat(buf, "|VROOT"); 2202 if (vp->v_flag & VTEXT) 2203 strcat(buf, "|VTEXT"); 2204 if (vp->v_flag & VSYSTEM) 2205 strcat(buf, "|VSYSTEM"); 2206 if (vp->v_flag & VXLOCK) 2207 strcat(buf, "|VXLOCK"); 2208 if (vp->v_flag & VXWANT) 2209 strcat(buf, "|VXWANT"); 2210 if (vp->v_flag & VBWAIT) 2211 strcat(buf, "|VBWAIT"); 2212 if (vp->v_flag & VDOOMED) 2213 strcat(buf, "|VDOOMED"); 2214 if (vp->v_flag & VFREE) 2215 strcat(buf, "|VFREE"); 2216 if (vp->v_flag & VOBJBUF) 2217 strcat(buf, "|VOBJBUF"); 2218 if (buf[0] != '\0') 2219 printf(" flags (%s)", &buf[1]); 2220 if (vp->v_data == NULL) { 2221 printf("\n"); 2222 } else { 2223 printf("\n\t"); 2224 VOP_PRINT(vp); 2225 } 2226 } 2227 2228 #ifdef DDB 2229 #include <ddb/ddb.h> 2230 /* 2231 * List all of the locked vnodes in the system. 2232 * Called when debugging the kernel. 2233 */ 2234 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 2235 { 2236 struct thread *td = curthread; /* XXX */ 2237 struct mount *mp, *nmp; 2238 struct vnode *vp; 2239 2240 printf("Locked vnodes\n"); 2241 lwkt_gettoken(&mountlist_token); 2242 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2243 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, td)) { 2244 nmp = TAILQ_NEXT(mp, mnt_list); 2245 continue; 2246 } 2247 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2248 if (VOP_ISLOCKED(vp, NULL)) 2249 vprint((char *)0, vp); 2250 } 2251 lwkt_gettoken(&mountlist_token); 2252 nmp = TAILQ_NEXT(mp, mnt_list); 2253 vfs_unbusy(mp, td); 2254 } 2255 lwkt_reltoken(&mountlist_token); 2256 } 2257 #endif 2258 2259 /* 2260 * Top level filesystem related information gathering. 2261 */ 2262 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS); 2263 2264 static int 2265 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2266 { 2267 int *name = (int *)arg1 - 1; /* XXX */ 2268 u_int namelen = arg2 + 1; /* XXX */ 2269 struct vfsconf *vfsp; 2270 2271 #if 1 || defined(COMPAT_PRELITE2) 2272 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2273 if (namelen == 1) 2274 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2275 #endif 2276 2277 #ifdef notyet 2278 /* all sysctl names at this level are at least name and field */ 2279 if (namelen < 2) 2280 return (ENOTDIR); /* overloaded */ 2281 if (name[0] != VFS_GENERIC) { 2282 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2283 if (vfsp->vfc_typenum == name[0]) 2284 break; 2285 if (vfsp == NULL) 2286 return (EOPNOTSUPP); 2287 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2288 oldp, oldlenp, newp, newlen, p)); 2289 } 2290 #endif 2291 switch (name[1]) { 2292 case VFS_MAXTYPENUM: 2293 if (namelen != 2) 2294 return (ENOTDIR); 2295 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2296 case VFS_CONF: 2297 if (namelen != 3) 2298 return (ENOTDIR); /* overloaded */ 2299 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2300 if (vfsp->vfc_typenum == name[2]) 2301 break; 2302 if (vfsp == NULL) 2303 return (EOPNOTSUPP); 2304 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2305 } 2306 return (EOPNOTSUPP); 2307 } 2308 2309 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2310 "Generic filesystem"); 2311 2312 #if 1 || defined(COMPAT_PRELITE2) 2313 2314 static int 2315 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2316 { 2317 int error; 2318 struct vfsconf *vfsp; 2319 struct ovfsconf ovfs; 2320 2321 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2322 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2323 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2324 ovfs.vfc_index = vfsp->vfc_typenum; 2325 ovfs.vfc_refcount = vfsp->vfc_refcount; 2326 ovfs.vfc_flags = vfsp->vfc_flags; 2327 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2328 if (error) 2329 return error; 2330 } 2331 return 0; 2332 } 2333 2334 #endif /* 1 || COMPAT_PRELITE2 */ 2335 2336 #if 0 2337 #define KINFO_VNODESLOP 10 2338 /* 2339 * Dump vnode list (via sysctl). 2340 * Copyout address of vnode followed by vnode. 2341 */ 2342 /* ARGSUSED */ 2343 static int 2344 sysctl_vnode(SYSCTL_HANDLER_ARGS) 2345 { 2346 struct proc *p = curproc; /* XXX */ 2347 struct mount *mp, *nmp; 2348 struct vnode *nvp, *vp; 2349 int error; 2350 2351 #define VPTRSZ sizeof (struct vnode *) 2352 #define VNODESZ sizeof (struct vnode) 2353 2354 req->lock = 0; 2355 if (!req->oldptr) /* Make an estimate */ 2356 return (SYSCTL_OUT(req, 0, 2357 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2358 2359 lwkt_gettoken(&mountlist_token); 2360 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2361 if (vfs_busy(mp, LK_NOWAIT, &mountlist_token, p)) { 2362 nmp = TAILQ_NEXT(mp, mnt_list); 2363 continue; 2364 } 2365 again: 2366 lwkt_gettoken(&mntvnode_token); 2367 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2368 vp != NULL; 2369 vp = nvp) { 2370 /* 2371 * Check that the vp is still associated with 2372 * this filesystem. RACE: could have been 2373 * recycled onto the same filesystem. 2374 */ 2375 if (vp->v_mount != mp) { 2376 lwkt_reltoken(&mntvnode_token); 2377 goto again; 2378 } 2379 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2380 lwkt_reltoken(&mntvnode_token); 2381 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2382 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2383 return (error); 2384 lwkt_gettoken(&mntvnode_token); 2385 } 2386 lwkt_reltoken(&mntvnode_token); 2387 lwkt_gettoken(&mountlist_token); 2388 nmp = TAILQ_NEXT(mp, mnt_list); 2389 vfs_unbusy(mp, p); 2390 } 2391 lwkt_reltoken(&mountlist_token); 2392 2393 return (0); 2394 } 2395 #endif 2396 2397 /* 2398 * XXX 2399 * Exporting the vnode list on large systems causes them to crash. 2400 * Exporting the vnode list on medium systems causes sysctl to coredump. 2401 */ 2402 #if 0 2403 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2404 0, 0, sysctl_vnode, "S,vnode", ""); 2405 #endif 2406 2407 /* 2408 * Check to see if a filesystem is mounted on a block device. 2409 */ 2410 int 2411 vfs_mountedon(vp) 2412 struct vnode *vp; 2413 { 2414 2415 if (vp->v_specmountpoint != NULL) 2416 return (EBUSY); 2417 return (0); 2418 } 2419 2420 /* 2421 * Unmount all filesystems. The list is traversed in reverse order 2422 * of mounting to avoid dependencies. 2423 */ 2424 void 2425 vfs_unmountall() 2426 { 2427 struct mount *mp; 2428 struct thread *td = curthread; 2429 int error; 2430 2431 if (td->td_proc == NULL) 2432 td = initproc->p_thread; /* XXX XXX use proc0 instead? */ 2433 2434 /* 2435 * Since this only runs when rebooting, it is not interlocked. 2436 */ 2437 while(!TAILQ_EMPTY(&mountlist)) { 2438 mp = TAILQ_LAST(&mountlist, mntlist); 2439 error = dounmount(mp, MNT_FORCE, td); 2440 if (error) { 2441 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2442 printf("unmount of %s failed (", 2443 mp->mnt_stat.f_mntonname); 2444 if (error == EBUSY) 2445 printf("BUSY)\n"); 2446 else 2447 printf("%d)\n", error); 2448 } else { 2449 /* The unmount has removed mp from the mountlist */ 2450 } 2451 } 2452 } 2453 2454 /* 2455 * Build hash lists of net addresses and hang them off the mount point. 2456 * Called by ufs_mount() to set up the lists of export addresses. 2457 */ 2458 static int 2459 vfs_hang_addrlist(mp, nep, argp) 2460 struct mount *mp; 2461 struct netexport *nep; 2462 struct export_args *argp; 2463 { 2464 struct netcred *np; 2465 struct radix_node_head *rnh; 2466 int i; 2467 struct radix_node *rn; 2468 struct sockaddr *saddr, *smask = 0; 2469 struct domain *dom; 2470 int error; 2471 2472 if (argp->ex_addrlen == 0) { 2473 if (mp->mnt_flag & MNT_DEFEXPORTED) 2474 return (EPERM); 2475 np = &nep->ne_defexported; 2476 np->netc_exflags = argp->ex_flags; 2477 np->netc_anon = argp->ex_anon; 2478 np->netc_anon.cr_ref = 1; 2479 mp->mnt_flag |= MNT_DEFEXPORTED; 2480 return (0); 2481 } 2482 2483 if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN) 2484 return (EINVAL); 2485 if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN) 2486 return (EINVAL); 2487 2488 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2489 np = (struct netcred *) malloc(i, M_NETADDR, M_WAITOK); 2490 bzero((caddr_t) np, i); 2491 saddr = (struct sockaddr *) (np + 1); 2492 if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen))) 2493 goto out; 2494 if (saddr->sa_len > argp->ex_addrlen) 2495 saddr->sa_len = argp->ex_addrlen; 2496 if (argp->ex_masklen) { 2497 smask = (struct sockaddr *) ((caddr_t) saddr + argp->ex_addrlen); 2498 error = copyin(argp->ex_mask, (caddr_t) smask, argp->ex_masklen); 2499 if (error) 2500 goto out; 2501 if (smask->sa_len > argp->ex_masklen) 2502 smask->sa_len = argp->ex_masklen; 2503 } 2504 i = saddr->sa_family; 2505 if ((rnh = nep->ne_rtable[i]) == 0) { 2506 /* 2507 * Seems silly to initialize every AF when most are not used, 2508 * do so on demand here 2509 */ 2510 for (dom = domains; dom; dom = dom->dom_next) 2511 if (dom->dom_family == i && dom->dom_rtattach) { 2512 dom->dom_rtattach((void **) &nep->ne_rtable[i], 2513 dom->dom_rtoffset); 2514 break; 2515 } 2516 if ((rnh = nep->ne_rtable[i]) == 0) { 2517 error = ENOBUFS; 2518 goto out; 2519 } 2520 } 2521 rn = (*rnh->rnh_addaddr) ((caddr_t) saddr, (caddr_t) smask, rnh, 2522 np->netc_rnodes); 2523 if (rn == 0 || np != (struct netcred *) rn) { /* already exists */ 2524 error = EPERM; 2525 goto out; 2526 } 2527 np->netc_exflags = argp->ex_flags; 2528 np->netc_anon = argp->ex_anon; 2529 np->netc_anon.cr_ref = 1; 2530 return (0); 2531 out: 2532 free(np, M_NETADDR); 2533 return (error); 2534 } 2535 2536 /* ARGSUSED */ 2537 static int 2538 vfs_free_netcred(rn, w) 2539 struct radix_node *rn; 2540 void *w; 2541 { 2542 struct radix_node_head *rnh = (struct radix_node_head *) w; 2543 2544 (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh); 2545 free((caddr_t) rn, M_NETADDR); 2546 return (0); 2547 } 2548 2549 /* 2550 * Free the net address hash lists that are hanging off the mount points. 2551 */ 2552 static void 2553 vfs_free_addrlist(nep) 2554 struct netexport *nep; 2555 { 2556 int i; 2557 struct radix_node_head *rnh; 2558 2559 for (i = 0; i <= AF_MAX; i++) 2560 if ((rnh = nep->ne_rtable[i])) { 2561 (*rnh->rnh_walktree) (rnh, vfs_free_netcred, 2562 (caddr_t) rnh); 2563 free((caddr_t) rnh, M_RTABLE); 2564 nep->ne_rtable[i] = 0; 2565 } 2566 } 2567 2568 int 2569 vfs_export(mp, nep, argp) 2570 struct mount *mp; 2571 struct netexport *nep; 2572 struct export_args *argp; 2573 { 2574 int error; 2575 2576 if (argp->ex_flags & MNT_DELEXPORT) { 2577 if (mp->mnt_flag & MNT_EXPUBLIC) { 2578 vfs_setpublicfs(NULL, NULL, NULL); 2579 mp->mnt_flag &= ~MNT_EXPUBLIC; 2580 } 2581 vfs_free_addrlist(nep); 2582 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2583 } 2584 if (argp->ex_flags & MNT_EXPORTED) { 2585 if (argp->ex_flags & MNT_EXPUBLIC) { 2586 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2587 return (error); 2588 mp->mnt_flag |= MNT_EXPUBLIC; 2589 } 2590 if ((error = vfs_hang_addrlist(mp, nep, argp))) 2591 return (error); 2592 mp->mnt_flag |= MNT_EXPORTED; 2593 } 2594 return (0); 2595 } 2596 2597 2598 /* 2599 * Set the publicly exported filesystem (WebNFS). Currently, only 2600 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2601 */ 2602 int 2603 vfs_setpublicfs(mp, nep, argp) 2604 struct mount *mp; 2605 struct netexport *nep; 2606 struct export_args *argp; 2607 { 2608 int error; 2609 struct vnode *rvp; 2610 char *cp; 2611 2612 /* 2613 * mp == NULL -> invalidate the current info, the FS is 2614 * no longer exported. May be called from either vfs_export 2615 * or unmount, so check if it hasn't already been done. 2616 */ 2617 if (mp == NULL) { 2618 if (nfs_pub.np_valid) { 2619 nfs_pub.np_valid = 0; 2620 if (nfs_pub.np_index != NULL) { 2621 FREE(nfs_pub.np_index, M_TEMP); 2622 nfs_pub.np_index = NULL; 2623 } 2624 } 2625 return (0); 2626 } 2627 2628 /* 2629 * Only one allowed at a time. 2630 */ 2631 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2632 return (EBUSY); 2633 2634 /* 2635 * Get real filehandle for root of exported FS. 2636 */ 2637 bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle)); 2638 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid; 2639 2640 if ((error = VFS_ROOT(mp, &rvp))) 2641 return (error); 2642 2643 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2644 return (error); 2645 2646 vput(rvp); 2647 2648 /* 2649 * If an indexfile was specified, pull it in. 2650 */ 2651 if (argp->ex_indexfile != NULL) { 2652 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2653 M_WAITOK); 2654 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2655 MAXNAMLEN, (size_t *)0); 2656 if (!error) { 2657 /* 2658 * Check for illegal filenames. 2659 */ 2660 for (cp = nfs_pub.np_index; *cp; cp++) { 2661 if (*cp == '/') { 2662 error = EINVAL; 2663 break; 2664 } 2665 } 2666 } 2667 if (error) { 2668 FREE(nfs_pub.np_index, M_TEMP); 2669 return (error); 2670 } 2671 } 2672 2673 nfs_pub.np_mount = mp; 2674 nfs_pub.np_valid = 1; 2675 return (0); 2676 } 2677 2678 struct netcred * 2679 vfs_export_lookup(mp, nep, nam) 2680 struct mount *mp; 2681 struct netexport *nep; 2682 struct sockaddr *nam; 2683 { 2684 struct netcred *np; 2685 struct radix_node_head *rnh; 2686 struct sockaddr *saddr; 2687 2688 np = NULL; 2689 if (mp->mnt_flag & MNT_EXPORTED) { 2690 /* 2691 * Lookup in the export list first. 2692 */ 2693 if (nam != NULL) { 2694 saddr = nam; 2695 rnh = nep->ne_rtable[saddr->sa_family]; 2696 if (rnh != NULL) { 2697 np = (struct netcred *) 2698 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2699 rnh); 2700 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2701 np = NULL; 2702 } 2703 } 2704 /* 2705 * If no address match, use the default if it exists. 2706 */ 2707 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2708 np = &nep->ne_defexported; 2709 } 2710 return (np); 2711 } 2712 2713 /* 2714 * perform msync on all vnodes under a mount point 2715 * the mount point must be locked. 2716 */ 2717 void 2718 vfs_msync(struct mount *mp, int flags) 2719 { 2720 struct thread *td = curthread; /* XXX */ 2721 struct vnode *vp, *nvp; 2722 struct vm_object *obj; 2723 int tries; 2724 2725 tries = 5; 2726 lwkt_gettoken(&mntvnode_token); 2727 loop: 2728 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) { 2729 if (vp->v_mount != mp) { 2730 if (--tries > 0) 2731 goto loop; 2732 break; 2733 } 2734 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2735 2736 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2737 continue; 2738 2739 /* 2740 * There could be hundreds of thousands of vnodes, we cannot 2741 * afford to do anything heavy-weight until we have a fairly 2742 * good indication that there is something to do. 2743 */ 2744 if ((vp->v_flag & VOBJDIRTY) && 2745 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 2746 lwkt_reltoken(&mntvnode_token); 2747 if (!vget(vp, 2748 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, td)) { 2749 if (VOP_GETVOBJECT(vp, &obj) == 0) { 2750 vm_object_page_clean(obj, 0, 0, flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC); 2751 } 2752 vput(vp); 2753 } 2754 lwkt_gettoken(&mntvnode_token); 2755 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) { 2756 if (--tries > 0) 2757 goto loop; 2758 break; 2759 } 2760 } 2761 } 2762 lwkt_reltoken(&mntvnode_token); 2763 } 2764 2765 /* 2766 * Create the VM object needed for VMIO and mmap support. This 2767 * is done for all VREG files in the system. Some filesystems might 2768 * afford the additional metadata buffering capability of the 2769 * VMIO code by making the device node be VMIO mode also. 2770 * 2771 * vp must be locked when vfs_object_create is called. 2772 */ 2773 int 2774 vfs_object_create(struct vnode *vp, struct thread *td) 2775 { 2776 return (VOP_CREATEVOBJECT(vp, td)); 2777 } 2778 2779 void 2780 vfree(vp) 2781 struct vnode *vp; 2782 { 2783 int s; 2784 2785 s = splbio(); 2786 lwkt_gettoken(&vnode_free_list_token); 2787 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free")); 2788 if (vp->v_flag & VAGE) { 2789 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2790 } else { 2791 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2792 } 2793 freevnodes++; 2794 lwkt_reltoken(&vnode_free_list_token); 2795 vp->v_flag &= ~VAGE; 2796 vp->v_flag |= VFREE; 2797 splx(s); 2798 } 2799 2800 void 2801 vbusy(vp) 2802 struct vnode *vp; 2803 { 2804 int s; 2805 2806 s = splbio(); 2807 lwkt_gettoken(&vnode_free_list_token); 2808 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free")); 2809 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2810 freevnodes--; 2811 lwkt_reltoken(&vnode_free_list_token); 2812 vp->v_flag &= ~(VFREE|VAGE); 2813 splx(s); 2814 } 2815 2816 /* 2817 * Record a process's interest in events which might happen to 2818 * a vnode. Because poll uses the historic select-style interface 2819 * internally, this routine serves as both the ``check for any 2820 * pending events'' and the ``record my interest in future events'' 2821 * functions. (These are done together, while the lock is held, 2822 * to avoid race conditions.) 2823 */ 2824 int 2825 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 2826 { 2827 lwkt_gettoken(&vp->v_pollinfo.vpi_token); 2828 if (vp->v_pollinfo.vpi_revents & events) { 2829 /* 2830 * This leaves events we are not interested 2831 * in available for the other process which 2832 * which presumably had requested them 2833 * (otherwise they would never have been 2834 * recorded). 2835 */ 2836 events &= vp->v_pollinfo.vpi_revents; 2837 vp->v_pollinfo.vpi_revents &= ~events; 2838 2839 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2840 return events; 2841 } 2842 vp->v_pollinfo.vpi_events |= events; 2843 selrecord(td, &vp->v_pollinfo.vpi_selinfo); 2844 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2845 return 0; 2846 } 2847 2848 /* 2849 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2850 * it is possible for us to miss an event due to race conditions, but 2851 * that condition is expected to be rare, so for the moment it is the 2852 * preferred interface. 2853 */ 2854 void 2855 vn_pollevent(vp, events) 2856 struct vnode *vp; 2857 short events; 2858 { 2859 lwkt_gettoken(&vp->v_pollinfo.vpi_token); 2860 if (vp->v_pollinfo.vpi_events & events) { 2861 /* 2862 * We clear vpi_events so that we don't 2863 * call selwakeup() twice if two events are 2864 * posted before the polling process(es) is 2865 * awakened. This also ensures that we take at 2866 * most one selwakeup() if the polling process 2867 * is no longer interested. However, it does 2868 * mean that only one event can be noticed at 2869 * a time. (Perhaps we should only clear those 2870 * event bits which we note?) XXX 2871 */ 2872 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2873 vp->v_pollinfo.vpi_revents |= events; 2874 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2875 } 2876 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2877 } 2878 2879 /* 2880 * Wake up anyone polling on vp because it is being revoked. 2881 * This depends on dead_poll() returning POLLHUP for correct 2882 * behavior. 2883 */ 2884 void 2885 vn_pollgone(vp) 2886 struct vnode *vp; 2887 { 2888 lwkt_gettoken(&vp->v_pollinfo.vpi_token); 2889 if (vp->v_pollinfo.vpi_events) { 2890 vp->v_pollinfo.vpi_events = 0; 2891 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2892 } 2893 lwkt_reltoken(&vp->v_pollinfo.vpi_token); 2894 } 2895 2896 2897 2898 /* 2899 * Routine to create and manage a filesystem syncer vnode. 2900 */ 2901 #define sync_close ((int (*) (struct vop_close_args *))nullop) 2902 static int sync_fsync (struct vop_fsync_args *); 2903 static int sync_inactive (struct vop_inactive_args *); 2904 static int sync_reclaim (struct vop_reclaim_args *); 2905 #define sync_lock ((int (*) (struct vop_lock_args *))vop_nolock) 2906 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_nounlock) 2907 static int sync_print (struct vop_print_args *); 2908 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_noislocked) 2909 2910 static vop_t **sync_vnodeop_p; 2911 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2912 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2913 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2914 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2915 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2916 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2917 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2918 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2919 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2920 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2921 { NULL, NULL } 2922 }; 2923 static struct vnodeopv_desc sync_vnodeop_opv_desc = 2924 { &sync_vnodeop_p, sync_vnodeop_entries }; 2925 2926 VNODEOP_SET(sync_vnodeop_opv_desc); 2927 2928 /* 2929 * Create a new filesystem syncer vnode for the specified mount point. 2930 */ 2931 int 2932 vfs_allocate_syncvnode(mp) 2933 struct mount *mp; 2934 { 2935 struct vnode *vp; 2936 static long start, incr, next; 2937 int error; 2938 2939 /* Allocate a new vnode */ 2940 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2941 mp->mnt_syncer = NULL; 2942 return (error); 2943 } 2944 vp->v_type = VNON; 2945 /* 2946 * Place the vnode onto the syncer worklist. We attempt to 2947 * scatter them about on the list so that they will go off 2948 * at evenly distributed times even if all the filesystems 2949 * are mounted at once. 2950 */ 2951 next += incr; 2952 if (next == 0 || next > syncer_maxdelay) { 2953 start /= 2; 2954 incr /= 2; 2955 if (start == 0) { 2956 start = syncer_maxdelay / 2; 2957 incr = syncer_maxdelay; 2958 } 2959 next = start; 2960 } 2961 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2962 mp->mnt_syncer = vp; 2963 return (0); 2964 } 2965 2966 /* 2967 * Do a lazy sync of the filesystem. 2968 */ 2969 static int 2970 sync_fsync(ap) 2971 struct vop_fsync_args /* { 2972 struct vnode *a_vp; 2973 struct ucred *a_cred; 2974 int a_waitfor; 2975 struct thread *a_td; 2976 } */ *ap; 2977 { 2978 struct vnode *syncvp = ap->a_vp; 2979 struct mount *mp = syncvp->v_mount; 2980 struct thread *td = ap->a_td; 2981 int asyncflag; 2982 2983 /* 2984 * We only need to do something if this is a lazy evaluation. 2985 */ 2986 if (ap->a_waitfor != MNT_LAZY) 2987 return (0); 2988 2989 /* 2990 * Move ourselves to the back of the sync list. 2991 */ 2992 vn_syncer_add_to_worklist(syncvp, syncdelay); 2993 2994 /* 2995 * Walk the list of vnodes pushing all that are dirty and 2996 * not already on the sync list. 2997 */ 2998 lwkt_gettoken(&mountlist_token); 2999 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_token, td) != 0) { 3000 lwkt_reltoken(&mountlist_token); 3001 return (0); 3002 } 3003 asyncflag = mp->mnt_flag & MNT_ASYNC; 3004 mp->mnt_flag &= ~MNT_ASYNC; 3005 vfs_msync(mp, MNT_NOWAIT); 3006 VFS_SYNC(mp, MNT_LAZY, td); 3007 if (asyncflag) 3008 mp->mnt_flag |= MNT_ASYNC; 3009 vfs_unbusy(mp, td); 3010 return (0); 3011 } 3012 3013 /* 3014 * The syncer vnode is no referenced. 3015 */ 3016 static int 3017 sync_inactive(ap) 3018 struct vop_inactive_args /* { 3019 struct vnode *a_vp; 3020 struct proc *a_p; 3021 } */ *ap; 3022 { 3023 3024 vgone(ap->a_vp); 3025 return (0); 3026 } 3027 3028 /* 3029 * The syncer vnode is no longer needed and is being decommissioned. 3030 * 3031 * Modifications to the worklist must be protected at splbio(). 3032 */ 3033 static int 3034 sync_reclaim(ap) 3035 struct vop_reclaim_args /* { 3036 struct vnode *a_vp; 3037 } */ *ap; 3038 { 3039 struct vnode *vp = ap->a_vp; 3040 int s; 3041 3042 s = splbio(); 3043 vp->v_mount->mnt_syncer = NULL; 3044 if (vp->v_flag & VONWORKLST) { 3045 LIST_REMOVE(vp, v_synclist); 3046 vp->v_flag &= ~VONWORKLST; 3047 } 3048 splx(s); 3049 3050 return (0); 3051 } 3052 3053 /* 3054 * Print out a syncer vnode. 3055 */ 3056 static int 3057 sync_print(ap) 3058 struct vop_print_args /* { 3059 struct vnode *a_vp; 3060 } */ *ap; 3061 { 3062 struct vnode *vp = ap->a_vp; 3063 3064 printf("syncer vnode"); 3065 if (vp->v_vnlock != NULL) 3066 lockmgr_printinfo(vp->v_vnlock); 3067 printf("\n"); 3068 return (0); 3069 } 3070 3071 /* 3072 * extract the dev_t from a VBLK or VCHR 3073 */ 3074 dev_t 3075 vn_todev(vp) 3076 struct vnode *vp; 3077 { 3078 if (vp->v_type != VBLK && vp->v_type != VCHR) 3079 return (NODEV); 3080 return (vp->v_rdev); 3081 } 3082 3083 /* 3084 * Check if vnode represents a disk device 3085 */ 3086 int 3087 vn_isdisk(vp, errp) 3088 struct vnode *vp; 3089 int *errp; 3090 { 3091 if (vp->v_type != VBLK && vp->v_type != VCHR) { 3092 if (errp != NULL) 3093 *errp = ENOTBLK; 3094 return (0); 3095 } 3096 if (vp->v_rdev == NULL) { 3097 if (errp != NULL) 3098 *errp = ENXIO; 3099 return (0); 3100 } 3101 if (!dev_dport(vp->v_rdev)) { 3102 if (errp != NULL) 3103 *errp = ENXIO; 3104 return (0); 3105 } 3106 if (!(dev_dflags(vp->v_rdev) & D_DISK)) { 3107 if (errp != NULL) 3108 *errp = ENOTBLK; 3109 return (0); 3110 } 3111 if (errp != NULL) 3112 *errp = 0; 3113 return (1); 3114 } 3115 3116 void 3117 NDFREE(ndp, flags) 3118 struct nameidata *ndp; 3119 const uint flags; 3120 { 3121 if (!(flags & NDF_NO_FREE_PNBUF) && 3122 (ndp->ni_cnd.cn_flags & CNP_HASBUF)) { 3123 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 3124 ndp->ni_cnd.cn_flags &= ~CNP_HASBUF; 3125 } 3126 if (!(flags & NDF_NO_DNCP_RELE) && 3127 (ndp->ni_cnd.cn_flags & CNP_WANTDNCP) && 3128 ndp->ni_dncp) { 3129 cache_drop(ndp->ni_dncp); 3130 ndp->ni_dncp = NULL; 3131 } 3132 if (!(flags & NDF_NO_NCP_RELE) && 3133 (ndp->ni_cnd.cn_flags & CNP_WANTNCP) && 3134 ndp->ni_ncp) { 3135 cache_drop(ndp->ni_ncp); 3136 ndp->ni_ncp = NULL; 3137 } 3138 if (!(flags & NDF_NO_DVP_UNLOCK) && 3139 (ndp->ni_cnd.cn_flags & CNP_LOCKPARENT) && 3140 ndp->ni_dvp != ndp->ni_vp) { 3141 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_td); 3142 } 3143 if (!(flags & NDF_NO_DVP_RELE) && 3144 (ndp->ni_cnd.cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT))) { 3145 vrele(ndp->ni_dvp); 3146 ndp->ni_dvp = NULL; 3147 } 3148 if (!(flags & NDF_NO_VP_UNLOCK) && 3149 (ndp->ni_cnd.cn_flags & CNP_LOCKLEAF) && ndp->ni_vp) { 3150 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_td); 3151 } 3152 if (!(flags & NDF_NO_VP_RELE) && 3153 ndp->ni_vp) { 3154 vrele(ndp->ni_vp); 3155 ndp->ni_vp = NULL; 3156 } 3157 if (!(flags & NDF_NO_STARTDIR_RELE) && 3158 (ndp->ni_cnd.cn_flags & CNP_SAVESTART)) { 3159 vrele(ndp->ni_startdir); 3160 ndp->ni_startdir = NULL; 3161 } 3162 } 3163 3164