vfs_subr.c revision 90100
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: head/sys/kern/vfs_subr.c 90100 2002-02-02 01:49:18Z mckusick $ 40 */ 41 42/* 43 * External virtual filesystem routines 44 */ 45#include "opt_ddb.h" 46#include "opt_ffs.h" 47 48#include <sys/param.h> 49#include <sys/systm.h> 50#include <sys/bio.h> 51#include <sys/buf.h> 52#include <sys/conf.h> 53#include <sys/eventhandler.h> 54#include <sys/fcntl.h> 55#include <sys/kernel.h> 56#include <sys/kthread.h> 57#include <sys/malloc.h> 58#include <sys/mount.h> 59#include <sys/namei.h> 60#include <sys/stat.h> 61#include <sys/sysctl.h> 62#include <sys/syslog.h> 63#include <sys/vmmeter.h> 64#include <sys/vnode.h> 65 66#include <vm/vm.h> 67#include <vm/vm_object.h> 68#include <vm/vm_extern.h> 69#include <vm/pmap.h> 70#include <vm/vm_map.h> 71#include <vm/vm_page.h> 72#include <vm/vm_zone.h> 73 74static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 75 76static void addalias __P((struct vnode *vp, dev_t nvp_rdev)); 77static void insmntque __P((struct vnode *vp, struct mount *mp)); 78static void vclean __P((struct vnode *vp, int flags, struct thread *td)); 79static void vlruvp(struct vnode *vp); 80 81/* 82 * Number of vnodes in existence. Increased whenever getnewvnode() 83 * allocates a new vnode, never decreased. 84 */ 85static unsigned long numvnodes; 86 87SYSCTL_LONG(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 88 89/* 90 * Conversion tables for conversion from vnode types to inode formats 91 * and back. 92 */ 93enum vtype iftovt_tab[16] = { 94 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 95 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 96}; 97int vttoif_tab[9] = { 98 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 99 S_IFSOCK, S_IFIFO, S_IFMT, 100}; 101 102/* 103 * List of vnodes that are ready for recycling. 104 */ 105static TAILQ_HEAD(freelst, vnode) vnode_free_list; 106 107/* 108 * Minimum number of free vnodes. If there are fewer than this free vnodes, 109 * getnewvnode() will return a newly allocated vnode. 110 */ 111static u_long wantfreevnodes = 25; 112SYSCTL_LONG(_debug, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 113/* Number of vnodes in the free list. */ 114static u_long freevnodes; 115SYSCTL_LONG(_debug, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 116 117#if 0 118/* Number of vnode allocation. */ 119static u_long vnodeallocs; 120SYSCTL_LONG(_debug, OID_AUTO, vnodeallocs, CTLFLAG_RD, &vnodeallocs, 0, ""); 121/* Period of vnode recycle from namecache in vnode allocation times. */ 122static u_long vnoderecycleperiod = 1000; 123SYSCTL_LONG(_debug, OID_AUTO, vnoderecycleperiod, CTLFLAG_RW, &vnoderecycleperiod, 0, ""); 124/* Minimum number of total vnodes required to invoke vnode recycle from namecache. */ 125static u_long vnoderecyclemintotalvn = 2000; 126SYSCTL_LONG(_debug, OID_AUTO, vnoderecyclemintotalvn, CTLFLAG_RW, &vnoderecyclemintotalvn, 0, ""); 127/* Minimum number of free vnodes required to invoke vnode recycle from namecache. */ 128static u_long vnoderecycleminfreevn = 2000; 129SYSCTL_LONG(_debug, OID_AUTO, vnoderecycleminfreevn, CTLFLAG_RW, &vnoderecycleminfreevn, 0, ""); 130/* Number of vnodes attempted to recycle at a time. */ 131static u_long vnoderecyclenumber = 3000; 132SYSCTL_LONG(_debug, OID_AUTO, vnoderecyclenumber, CTLFLAG_RW, &vnoderecyclenumber, 0, ""); 133#endif 134 135/* 136 * Various variables used for debugging the new implementation of 137 * reassignbuf(). 138 * XXX these are probably of (very) limited utility now. 139 */ 140static int reassignbufcalls; 141SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 142static int reassignbufloops; 143SYSCTL_INT(_vfs, OID_AUTO, reassignbufloops, CTLFLAG_RW, &reassignbufloops, 0, ""); 144static int reassignbufsortgood; 145SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortgood, CTLFLAG_RW, &reassignbufsortgood, 0, ""); 146static int reassignbufsortbad; 147SYSCTL_INT(_vfs, OID_AUTO, reassignbufsortbad, CTLFLAG_RW, &reassignbufsortbad, 0, ""); 148/* Set to 0 for old insertion-sort based reassignbuf, 1 for modern method. */ 149static int reassignbufmethod = 1; 150SYSCTL_INT(_vfs, OID_AUTO, reassignbufmethod, CTLFLAG_RW, &reassignbufmethod, 0, ""); 151static int nameileafonly; 152SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, ""); 153 154#ifdef ENABLE_VFS_IOOPT 155/* See NOTES for a description of this setting. */ 156int vfs_ioopt; 157SYSCTL_INT(_vfs, OID_AUTO, ioopt, CTLFLAG_RW, &vfs_ioopt, 0, ""); 158#endif 159 160/* List of mounted filesystems. */ 161struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 162 163/* For any iteration/modification of mountlist */ 164struct mtx mountlist_mtx; 165 166/* For any iteration/modification of mnt_vnodelist */ 167struct mtx mntvnode_mtx; 168 169/* 170 * Cache for the mount type id assigned to NFS. This is used for 171 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 172 */ 173int nfs_mount_type = -1; 174 175/* To keep more than one thread at a time from running vfs_getnewfsid */ 176static struct mtx mntid_mtx; 177 178/* For any iteration/modification of vnode_free_list */ 179static struct mtx vnode_free_list_mtx; 180 181/* 182 * For any iteration/modification of dev->si_hlist (linked through 183 * v_specnext) 184 */ 185static struct mtx spechash_mtx; 186 187/* Publicly exported FS */ 188struct nfs_public nfs_pub; 189 190/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 191static vm_zone_t vnode_zone; 192 193/* Set to 1 to print out reclaim of active vnodes */ 194int prtactive; 195 196/* 197 * The workitem queue. 198 * 199 * It is useful to delay writes of file data and filesystem metadata 200 * for tens of seconds so that quickly created and deleted files need 201 * not waste disk bandwidth being created and removed. To realize this, 202 * we append vnodes to a "workitem" queue. When running with a soft 203 * updates implementation, most pending metadata dependencies should 204 * not wait for more than a few seconds. Thus, mounted on block devices 205 * are delayed only about a half the time that file data is delayed. 206 * Similarly, directory updates are more critical, so are only delayed 207 * about a third the time that file data is delayed. Thus, there are 208 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 209 * one each second (driven off the filesystem syncer process). The 210 * syncer_delayno variable indicates the next queue that is to be processed. 211 * Items that need to be processed soon are placed in this queue: 212 * 213 * syncer_workitem_pending[syncer_delayno] 214 * 215 * A delay of fifteen seconds is done by placing the request fifteen 216 * entries later in the queue: 217 * 218 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 219 * 220 */ 221static int syncer_delayno; 222static long syncer_mask; 223LIST_HEAD(synclist, vnode); 224static struct synclist *syncer_workitem_pending; 225 226#define SYNCER_MAXDELAY 32 227static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 228static int syncdelay = 30; /* max time to delay syncing data */ 229static int filedelay = 30; /* time to delay syncing files */ 230SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 231static int dirdelay = 29; /* time to delay syncing directories */ 232SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 233static int metadelay = 28; /* time to delay syncing metadata */ 234SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 235static int rushjob; /* number of slots to run ASAP */ 236static int stat_rush_requests; /* number of times I/O speeded up */ 237SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 238 239/* 240 * Number of vnodes we want to exist at any one time. This is mostly used 241 * to size hash tables in vnode-related code. It is normally not used in 242 * getnewvnode(), as wantfreevnodes is normally nonzero.) 243 * 244 * XXX desiredvnodes is historical cruft and should not exist. 245 */ 246int desiredvnodes; 247SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 248 &desiredvnodes, 0, "Maximum number of vnodes"); 249static int minvnodes; 250SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 251 &minvnodes, 0, "Minimum number of vnodes"); 252static int vnlru_nowhere; 253SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, 254 "Number of times the vnlru process ran without success"); 255 256/* 257 * Initialize the vnode management data structures. 258 */ 259static void 260vntblinit(void *dummy __unused) 261{ 262 263 desiredvnodes = maxproc + cnt.v_page_count / 4; 264 minvnodes = desiredvnodes / 4; 265 mtx_init(&mountlist_mtx, "mountlist", MTX_DEF); 266 mtx_init(&mntvnode_mtx, "mntvnode", MTX_DEF); 267 mtx_init(&mntid_mtx, "mntid", MTX_DEF); 268 mtx_init(&spechash_mtx, "spechash", MTX_DEF); 269 TAILQ_INIT(&vnode_free_list); 270 mtx_init(&vnode_free_list_mtx, "vnode_free_list", MTX_DEF); 271 vnode_zone = zinit("VNODE", sizeof (struct vnode), 0, 0, 5); 272 /* 273 * Initialize the filesystem syncer. 274 */ 275 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 276 &syncer_mask); 277 syncer_maxdelay = syncer_mask + 1; 278} 279SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL) 280 281 282/* 283 * Mark a mount point as busy. Used to synchronize access and to delay 284 * unmounting. Interlock is not released on failure. 285 */ 286int 287vfs_busy(mp, flags, interlkp, td) 288 struct mount *mp; 289 int flags; 290 struct mtx *interlkp; 291 struct thread *td; 292{ 293 int lkflags; 294 295 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 296 if (flags & LK_NOWAIT) 297 return (ENOENT); 298 mp->mnt_kern_flag |= MNTK_MWAIT; 299 /* 300 * Since all busy locks are shared except the exclusive 301 * lock granted when unmounting, the only place that a 302 * wakeup needs to be done is at the release of the 303 * exclusive lock at the end of dounmount. 304 */ 305 msleep((caddr_t)mp, interlkp, PVFS, "vfs_busy", 0); 306 return (ENOENT); 307 } 308 lkflags = LK_SHARED | LK_NOPAUSE; 309 if (interlkp) 310 lkflags |= LK_INTERLOCK; 311 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td)) 312 panic("vfs_busy: unexpected lock failure"); 313 return (0); 314} 315 316/* 317 * Free a busy filesystem. 318 */ 319void 320vfs_unbusy(mp, td) 321 struct mount *mp; 322 struct thread *td; 323{ 324 325 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 326} 327 328/* 329 * Lookup a filesystem type, and if found allocate and initialize 330 * a mount structure for it. 331 * 332 * Devname is usually updated by mount(8) after booting. 333 */ 334int 335vfs_rootmountalloc(fstypename, devname, mpp) 336 char *fstypename; 337 char *devname; 338 struct mount **mpp; 339{ 340 struct thread *td = curthread; /* XXX */ 341 struct vfsconf *vfsp; 342 struct mount *mp; 343 344 if (fstypename == NULL) 345 return (ENODEV); 346 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 347 if (!strcmp(vfsp->vfc_name, fstypename)) 348 break; 349 if (vfsp == NULL) 350 return (ENODEV); 351 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 352 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, LK_NOPAUSE); 353 (void)vfs_busy(mp, LK_NOWAIT, 0, td); 354 TAILQ_INIT(&mp->mnt_nvnodelist); 355 TAILQ_INIT(&mp->mnt_reservedvnlist); 356 mp->mnt_vfc = vfsp; 357 mp->mnt_op = vfsp->vfc_vfsops; 358 mp->mnt_flag = MNT_RDONLY; 359 mp->mnt_vnodecovered = NULLVP; 360 vfsp->vfc_refcount++; 361 mp->mnt_iosize_max = DFLTPHYS; 362 mp->mnt_stat.f_type = vfsp->vfc_typenum; 363 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 364 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 365 mp->mnt_stat.f_mntonname[0] = '/'; 366 mp->mnt_stat.f_mntonname[1] = 0; 367 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 368 *mpp = mp; 369 return (0); 370} 371 372/* 373 * Find an appropriate filesystem to use for the root. If a filesystem 374 * has not been preselected, walk through the list of known filesystems 375 * trying those that have mountroot routines, and try them until one 376 * works or we have tried them all. 377 */ 378#ifdef notdef /* XXX JH */ 379int 380lite2_vfs_mountroot() 381{ 382 struct vfsconf *vfsp; 383 extern int (*lite2_mountroot) __P((void)); 384 int error; 385 386 if (lite2_mountroot != NULL) 387 return ((*lite2_mountroot)()); 388 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 389 if (vfsp->vfc_mountroot == NULL) 390 continue; 391 if ((error = (*vfsp->vfc_mountroot)()) == 0) 392 return (0); 393 printf("%s_mountroot failed: %d\n", vfsp->vfc_name, error); 394 } 395 return (ENODEV); 396} 397#endif 398 399/* 400 * Lookup a mount point by filesystem identifier. 401 */ 402struct mount * 403vfs_getvfs(fsid) 404 fsid_t *fsid; 405{ 406 register struct mount *mp; 407 408 mtx_lock(&mountlist_mtx); 409 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 410 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 411 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 412 mtx_unlock(&mountlist_mtx); 413 return (mp); 414 } 415 } 416 mtx_unlock(&mountlist_mtx); 417 return ((struct mount *) 0); 418} 419 420/* 421 * Get a new unique fsid. Try to make its val[0] unique, since this value 422 * will be used to create fake device numbers for stat(). Also try (but 423 * not so hard) make its val[0] unique mod 2^16, since some emulators only 424 * support 16-bit device numbers. We end up with unique val[0]'s for the 425 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 426 * 427 * Keep in mind that several mounts may be running in parallel. Starting 428 * the search one past where the previous search terminated is both a 429 * micro-optimization and a defense against returning the same fsid to 430 * different mounts. 431 */ 432void 433vfs_getnewfsid(mp) 434 struct mount *mp; 435{ 436 static u_int16_t mntid_base; 437 fsid_t tfsid; 438 int mtype; 439 440 mtx_lock(&mntid_mtx); 441 mtype = mp->mnt_vfc->vfc_typenum; 442 tfsid.val[1] = mtype; 443 mtype = (mtype & 0xFF) << 24; 444 for (;;) { 445 tfsid.val[0] = makeudev(255, 446 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 447 mntid_base++; 448 if (vfs_getvfs(&tfsid) == NULL) 449 break; 450 } 451 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 452 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 453 mtx_unlock(&mntid_mtx); 454} 455 456/* 457 * Knob to control the precision of file timestamps: 458 * 459 * 0 = seconds only; nanoseconds zeroed. 460 * 1 = seconds and nanoseconds, accurate within 1/HZ. 461 * 2 = seconds and nanoseconds, truncated to microseconds. 462 * >=3 = seconds and nanoseconds, maximum precision. 463 */ 464enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 465 466static int timestamp_precision = TSP_SEC; 467SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 468 ×tamp_precision, 0, ""); 469 470/* 471 * Get a current timestamp. 472 */ 473void 474vfs_timestamp(tsp) 475 struct timespec *tsp; 476{ 477 struct timeval tv; 478 479 switch (timestamp_precision) { 480 case TSP_SEC: 481 tsp->tv_sec = time_second; 482 tsp->tv_nsec = 0; 483 break; 484 case TSP_HZ: 485 getnanotime(tsp); 486 break; 487 case TSP_USEC: 488 microtime(&tv); 489 TIMEVAL_TO_TIMESPEC(&tv, tsp); 490 break; 491 case TSP_NSEC: 492 default: 493 nanotime(tsp); 494 break; 495 } 496} 497 498/* 499 * Set vnode attributes to VNOVAL 500 */ 501void 502vattr_null(vap) 503 register struct vattr *vap; 504{ 505 506 vap->va_type = VNON; 507 vap->va_size = VNOVAL; 508 vap->va_bytes = VNOVAL; 509 vap->va_mode = VNOVAL; 510 vap->va_nlink = VNOVAL; 511 vap->va_uid = VNOVAL; 512 vap->va_gid = VNOVAL; 513 vap->va_fsid = VNOVAL; 514 vap->va_fileid = VNOVAL; 515 vap->va_blocksize = VNOVAL; 516 vap->va_rdev = VNOVAL; 517 vap->va_atime.tv_sec = VNOVAL; 518 vap->va_atime.tv_nsec = VNOVAL; 519 vap->va_mtime.tv_sec = VNOVAL; 520 vap->va_mtime.tv_nsec = VNOVAL; 521 vap->va_ctime.tv_sec = VNOVAL; 522 vap->va_ctime.tv_nsec = VNOVAL; 523 vap->va_flags = VNOVAL; 524 vap->va_gen = VNOVAL; 525 vap->va_vaflags = 0; 526} 527 528/* 529 * This routine is called when we have too many vnodes. It attempts 530 * to free <count> vnodes and will potentially free vnodes that still 531 * have VM backing store (VM backing store is typically the cause 532 * of a vnode blowout so we want to do this). Therefore, this operation 533 * is not considered cheap. 534 * 535 * A number of conditions may prevent a vnode from being reclaimed. 536 * the buffer cache may have references on the vnode, a directory 537 * vnode may still have references due to the namei cache representing 538 * underlying files, or the vnode may be in active use. It is not 539 * desireable to reuse such vnodes. These conditions may cause the 540 * number of vnodes to reach some minimum value regardless of what 541 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 542 */ 543static int 544vlrureclaim(struct mount *mp, int count) 545{ 546 struct vnode *vp; 547 int done; 548 int trigger; 549 int usevnodes; 550 551 /* 552 * Calculate the trigger point, don't allow user 553 * screwups to blow us up. This prevents us from 554 * recycling vnodes with lots of resident pages. We 555 * aren't trying to free memory, we are trying to 556 * free vnodes. 557 */ 558 usevnodes = desiredvnodes; 559 if (usevnodes <= 0) 560 usevnodes = 1; 561 trigger = cnt.v_page_count * 2 / usevnodes; 562 563 done = 0; 564 mtx_lock(&mntvnode_mtx); 565 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) { 566 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 567 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 568 569 if (vp->v_type != VNON && 570 vp->v_type != VBAD && 571 VMIGHTFREE(vp) && /* critical path opt */ 572 (vp->v_object == NULL || vp->v_object->resident_page_count < trigger) && 573 mtx_trylock(&vp->v_interlock) 574 ) { 575 mtx_unlock(&mntvnode_mtx); 576 if (VMIGHTFREE(vp)) { 577 vgonel(vp, curthread); 578 done++; 579 } else { 580 mtx_unlock(&vp->v_interlock); 581 } 582 mtx_lock(&mntvnode_mtx); 583 } 584 --count; 585 } 586 mtx_unlock(&mntvnode_mtx); 587 return done; 588} 589 590/* 591 * Attempt to recycle vnodes in a context that is always safe to block. 592 * Calling vlrurecycle() from the bowels of file system code has some 593 * interesting deadlock problems. 594 */ 595static struct proc *vnlruproc; 596static int vnlruproc_sig; 597 598static void 599vnlru_proc(void) 600{ 601 struct mount *mp, *nmp; 602 int s; 603 int done; 604 struct proc *p = vnlruproc; 605 struct thread *td = &p->p_thread; /* XXXKSE */ 606 607 mtx_lock(&Giant); 608 609 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 610 SHUTDOWN_PRI_FIRST); 611 612 s = splbio(); 613 for (;;) { 614 kthread_suspend_check(p); 615 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) { 616 vnlruproc_sig = 0; 617 tsleep(vnlruproc, PVFS, "vlruwt", 0); 618 continue; 619 } 620 done = 0; 621 mtx_lock(&mountlist_mtx); 622 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 623 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 624 nmp = TAILQ_NEXT(mp, mnt_list); 625 continue; 626 } 627 done += vlrureclaim(mp, 10); 628 mtx_lock(&mountlist_mtx); 629 nmp = TAILQ_NEXT(mp, mnt_list); 630 vfs_unbusy(mp, td); 631 } 632 mtx_unlock(&mountlist_mtx); 633 if (done == 0) { 634#if 0 635 /* These messages are temporary debugging aids */ 636 if (vnlru_nowhere < 5) 637 printf("vnlru process getting nowhere..\n"); 638 else if (vnlru_nowhere == 5) 639 printf("vnlru process messages stopped.\n"); 640#endif 641 vnlru_nowhere++; 642 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 643 } 644 } 645 splx(s); 646} 647 648static struct kproc_desc vnlru_kp = { 649 "vnlru", 650 vnlru_proc, 651 &vnlruproc 652}; 653SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 654 655 656/* 657 * Routines having to do with the management of the vnode table. 658 */ 659 660/* 661 * Return the next vnode from the free list. 662 */ 663int 664getnewvnode(tag, mp, vops, vpp) 665 enum vtagtype tag; 666 struct mount *mp; 667 vop_t **vops; 668 struct vnode **vpp; 669{ 670 int s; 671 struct thread *td = curthread; /* XXX */ 672 struct vnode *vp = NULL; 673 struct mount *vnmp; 674 vm_object_t object; 675 676 s = splbio(); 677 /* 678 * Try to reuse vnodes if we hit the max. This situation only 679 * occurs in certain large-memory (2G+) situations. We cannot 680 * attempt to directly reclaim vnodes due to nasty recursion 681 * problems. 682 */ 683 if (vnlruproc_sig == 0 && numvnodes - freevnodes > desiredvnodes) { 684 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 685 wakeup(vnlruproc); 686 } 687 688 /* 689 * Attempt to reuse a vnode already on the free list, allocating 690 * a new vnode if we can't find one or if we have not reached a 691 * good minimum for good LRU performance. 692 */ 693 694 mtx_lock(&vnode_free_list_mtx); 695 696 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) { 697 int count; 698 699 for (count = 0; count < freevnodes; count++) { 700 vp = TAILQ_FIRST(&vnode_free_list); 701 if (vp == NULL || vp->v_usecount) 702 panic("getnewvnode: free vnode isn't"); 703 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 704 705 /* 706 * Don't recycle if we still have cached pages or if 707 * we cannot get the interlock. 708 */ 709 if ((VOP_GETVOBJECT(vp, &object) == 0 && 710 (object->resident_page_count || 711 object->ref_count)) || 712 !mtx_trylock(&vp->v_interlock)) { 713 TAILQ_INSERT_TAIL(&vnode_free_list, vp, 714 v_freelist); 715 vp = NULL; 716 continue; 717 } 718 if (LIST_FIRST(&vp->v_cache_src)) { 719 /* 720 * note: nameileafonly sysctl is temporary, 721 * for debugging only, and will eventually be 722 * removed. 723 */ 724 if (nameileafonly > 0) { 725 /* 726 * Do not reuse namei-cached directory 727 * vnodes that have cached 728 * subdirectories. 729 */ 730 if (cache_leaf_test(vp) < 0) { 731 mtx_unlock(&vp->v_interlock); 732 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 733 vp = NULL; 734 continue; 735 } 736 } else if (nameileafonly < 0 || 737 vmiodirenable == 0) { 738 /* 739 * Do not reuse namei-cached directory 740 * vnodes if nameileafonly is -1 or 741 * if VMIO backing for directories is 742 * turned off (otherwise we reuse them 743 * too quickly). 744 */ 745 mtx_unlock(&vp->v_interlock); 746 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 747 vp = NULL; 748 continue; 749 } 750 } 751 /* 752 * Skip over it if its filesystem is being suspended. 753 */ 754 if (vn_start_write(vp, &vnmp, V_NOWAIT) == 0) 755 break; 756 mtx_unlock(&vp->v_interlock); 757 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 758 vp = NULL; 759 } 760 } 761 if (vp) { 762 vp->v_flag |= VDOOMED; 763 vp->v_flag &= ~VFREE; 764 freevnodes--; 765 mtx_unlock(&vnode_free_list_mtx); 766 cache_purge(vp); 767 vp->v_lease = NULL; 768 if (vp->v_type != VBAD) { 769 vgonel(vp, td); 770 } else { 771 mtx_unlock(&vp->v_interlock); 772 } 773 vn_finished_write(vnmp); 774 775#ifdef INVARIANTS 776 { 777 int s; 778 779 if (vp->v_data) 780 panic("cleaned vnode isn't"); 781 s = splbio(); 782 if (vp->v_numoutput) 783 panic("Clean vnode has pending I/O's"); 784 splx(s); 785 if (vp->v_writecount != 0) 786 panic("Non-zero write count"); 787 } 788#endif 789 vp->v_flag = 0; 790 vp->v_lastw = 0; 791 vp->v_lasta = 0; 792 vp->v_cstart = 0; 793 vp->v_clen = 0; 794 vp->v_socket = 0; 795 } else { 796 mtx_unlock(&vnode_free_list_mtx); 797 vp = (struct vnode *) zalloc(vnode_zone); 798 bzero((char *) vp, sizeof *vp); 799 mtx_init(&vp->v_interlock, "vnode interlock", MTX_DEF); 800 vp->v_dd = vp; 801 mtx_init(&vp->v_pollinfo.vpi_lock, "vnode pollinfo", MTX_DEF); 802 cache_purge(vp); 803 LIST_INIT(&vp->v_cache_src); 804 TAILQ_INIT(&vp->v_cache_dst); 805 numvnodes++; 806 } 807 808 TAILQ_INIT(&vp->v_cleanblkhd); 809 TAILQ_INIT(&vp->v_dirtyblkhd); 810 vp->v_type = VNON; 811 vp->v_tag = tag; 812 vp->v_op = vops; 813 lockinit(&vp->v_lock, PVFS, "vnlock", VLKTIMEOUT, LK_NOPAUSE); 814 insmntque(vp, mp); 815 *vpp = vp; 816 vp->v_usecount = 1; 817 vp->v_data = 0; 818 819 splx(s); 820 821 vfs_object_create(vp, td, td->td_proc->p_ucred); 822 823#if 0 824 vnodeallocs++; 825 if (vnodeallocs % vnoderecycleperiod == 0 && 826 freevnodes < vnoderecycleminfreevn && 827 vnoderecyclemintotalvn < numvnodes) { 828 /* Recycle vnodes. */ 829 cache_purgeleafdirs(vnoderecyclenumber); 830 } 831#endif 832 833 return (0); 834} 835 836/* 837 * Move a vnode from one mount queue to another. 838 */ 839static void 840insmntque(vp, mp) 841 register struct vnode *vp; 842 register struct mount *mp; 843{ 844 845 mtx_lock(&mntvnode_mtx); 846 /* 847 * Delete from old mount point vnode list, if on one. 848 */ 849 if (vp->v_mount != NULL) 850 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 851 /* 852 * Insert into list of vnodes for the new mount point, if available. 853 */ 854 if ((vp->v_mount = mp) == NULL) { 855 mtx_unlock(&mntvnode_mtx); 856 return; 857 } 858 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 859 mtx_unlock(&mntvnode_mtx); 860} 861 862/* 863 * Update outstanding I/O count and do wakeup if requested. 864 */ 865void 866vwakeup(bp) 867 register struct buf *bp; 868{ 869 register struct vnode *vp; 870 871 bp->b_flags &= ~B_WRITEINPROG; 872 if ((vp = bp->b_vp)) { 873 vp->v_numoutput--; 874 if (vp->v_numoutput < 0) 875 panic("vwakeup: neg numoutput"); 876 if ((vp->v_numoutput == 0) && (vp->v_flag & VBWAIT)) { 877 vp->v_flag &= ~VBWAIT; 878 wakeup((caddr_t) &vp->v_numoutput); 879 } 880 } 881} 882 883/* 884 * Flush out and invalidate all buffers associated with a vnode. 885 * Called with the underlying object locked. 886 */ 887int 888vinvalbuf(vp, flags, cred, td, slpflag, slptimeo) 889 register struct vnode *vp; 890 int flags; 891 struct ucred *cred; 892 struct thread *td; 893 int slpflag, slptimeo; 894{ 895 register struct buf *bp; 896 struct buf *nbp, *blist; 897 int s, error; 898 vm_object_t object; 899 900 GIANT_REQUIRED; 901 902 if (flags & V_SAVE) { 903 s = splbio(); 904 while (vp->v_numoutput) { 905 vp->v_flag |= VBWAIT; 906 error = tsleep((caddr_t)&vp->v_numoutput, 907 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo); 908 if (error) { 909 splx(s); 910 return (error); 911 } 912 } 913 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 914 splx(s); 915 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0) 916 return (error); 917 s = splbio(); 918 if (vp->v_numoutput > 0 || 919 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 920 panic("vinvalbuf: dirty bufs"); 921 } 922 splx(s); 923 } 924 s = splbio(); 925 for (;;) { 926 blist = TAILQ_FIRST(&vp->v_cleanblkhd); 927 if (!blist) 928 blist = TAILQ_FIRST(&vp->v_dirtyblkhd); 929 if (!blist) 930 break; 931 932 for (bp = blist; bp; bp = nbp) { 933 nbp = TAILQ_NEXT(bp, b_vnbufs); 934 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 935 error = BUF_TIMELOCK(bp, 936 LK_EXCLUSIVE | LK_SLEEPFAIL, 937 "vinvalbuf", slpflag, slptimeo); 938 if (error == ENOLCK) 939 break; 940 splx(s); 941 return (error); 942 } 943 /* 944 * XXX Since there are no node locks for NFS, I 945 * believe there is a slight chance that a delayed 946 * write will occur while sleeping just above, so 947 * check for it. Note that vfs_bio_awrite expects 948 * buffers to reside on a queue, while BUF_WRITE and 949 * brelse do not. 950 */ 951 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 952 (flags & V_SAVE)) { 953 954 if (bp->b_vp == vp) { 955 if (bp->b_flags & B_CLUSTEROK) { 956 BUF_UNLOCK(bp); 957 vfs_bio_awrite(bp); 958 } else { 959 bremfree(bp); 960 bp->b_flags |= B_ASYNC; 961 BUF_WRITE(bp); 962 } 963 } else { 964 bremfree(bp); 965 (void) BUF_WRITE(bp); 966 } 967 break; 968 } 969 bremfree(bp); 970 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 971 bp->b_flags &= ~B_ASYNC; 972 brelse(bp); 973 } 974 } 975 976 /* 977 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 978 * have write I/O in-progress but if there is a VM object then the 979 * VM object can also have read-I/O in-progress. 980 */ 981 do { 982 while (vp->v_numoutput > 0) { 983 vp->v_flag |= VBWAIT; 984 tsleep(&vp->v_numoutput, PVM, "vnvlbv", 0); 985 } 986 if (VOP_GETVOBJECT(vp, &object) == 0) { 987 while (object->paging_in_progress) 988 vm_object_pip_sleep(object, "vnvlbx"); 989 } 990 } while (vp->v_numoutput > 0); 991 992 splx(s); 993 994 /* 995 * Destroy the copy in the VM cache, too. 996 */ 997 mtx_lock(&vp->v_interlock); 998 if (VOP_GETVOBJECT(vp, &object) == 0) { 999 vm_object_page_remove(object, 0, 0, 1000 (flags & V_SAVE) ? TRUE : FALSE); 1001 } 1002 mtx_unlock(&vp->v_interlock); 1003 1004 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || !TAILQ_EMPTY(&vp->v_cleanblkhd)) 1005 panic("vinvalbuf: flush failed"); 1006 return (0); 1007} 1008 1009/* 1010 * Truncate a file's buffer and pages to a specified length. This 1011 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1012 * sync activity. 1013 */ 1014int 1015vtruncbuf(vp, cred, td, length, blksize) 1016 register struct vnode *vp; 1017 struct ucred *cred; 1018 struct thread *td; 1019 off_t length; 1020 int blksize; 1021{ 1022 register struct buf *bp; 1023 struct buf *nbp; 1024 int s, anyfreed; 1025 int trunclbn; 1026 1027 /* 1028 * Round up to the *next* lbn. 1029 */ 1030 trunclbn = (length + blksize - 1) / blksize; 1031 1032 s = splbio(); 1033restart: 1034 anyfreed = 1; 1035 for (;anyfreed;) { 1036 anyfreed = 0; 1037 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 1038 nbp = TAILQ_NEXT(bp, b_vnbufs); 1039 if (bp->b_lblkno >= trunclbn) { 1040 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1041 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1042 goto restart; 1043 } else { 1044 bremfree(bp); 1045 bp->b_flags |= (B_INVAL | B_RELBUF); 1046 bp->b_flags &= ~B_ASYNC; 1047 brelse(bp); 1048 anyfreed = 1; 1049 } 1050 if (nbp && 1051 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1052 (nbp->b_vp != vp) || 1053 (nbp->b_flags & B_DELWRI))) { 1054 goto restart; 1055 } 1056 } 1057 } 1058 1059 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1060 nbp = TAILQ_NEXT(bp, b_vnbufs); 1061 if (bp->b_lblkno >= trunclbn) { 1062 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1063 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1064 goto restart; 1065 } else { 1066 bremfree(bp); 1067 bp->b_flags |= (B_INVAL | B_RELBUF); 1068 bp->b_flags &= ~B_ASYNC; 1069 brelse(bp); 1070 anyfreed = 1; 1071 } 1072 if (nbp && 1073 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1074 (nbp->b_vp != vp) || 1075 (nbp->b_flags & B_DELWRI) == 0)) { 1076 goto restart; 1077 } 1078 } 1079 } 1080 } 1081 1082 if (length > 0) { 1083restartsync: 1084 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1085 nbp = TAILQ_NEXT(bp, b_vnbufs); 1086 if ((bp->b_flags & B_DELWRI) && (bp->b_lblkno < 0)) { 1087 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) { 1088 BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL); 1089 goto restart; 1090 } else { 1091 bremfree(bp); 1092 if (bp->b_vp == vp) { 1093 bp->b_flags |= B_ASYNC; 1094 } else { 1095 bp->b_flags &= ~B_ASYNC; 1096 } 1097 BUF_WRITE(bp); 1098 } 1099 goto restartsync; 1100 } 1101 1102 } 1103 } 1104 1105 while (vp->v_numoutput > 0) { 1106 vp->v_flag |= VBWAIT; 1107 tsleep(&vp->v_numoutput, PVM, "vbtrunc", 0); 1108 } 1109 1110 splx(s); 1111 1112 vnode_pager_setsize(vp, length); 1113 1114 return (0); 1115} 1116 1117/* 1118 * Associate a buffer with a vnode. 1119 */ 1120void 1121bgetvp(vp, bp) 1122 register struct vnode *vp; 1123 register struct buf *bp; 1124{ 1125 int s; 1126 1127 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 1128 1129 vhold(vp); 1130 bp->b_vp = vp; 1131 bp->b_dev = vn_todev(vp); 1132 /* 1133 * Insert onto list for new vnode. 1134 */ 1135 s = splbio(); 1136 bp->b_xflags |= BX_VNCLEAN; 1137 bp->b_xflags &= ~BX_VNDIRTY; 1138 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 1139 splx(s); 1140} 1141 1142/* 1143 * Disassociate a buffer from a vnode. 1144 */ 1145void 1146brelvp(bp) 1147 register struct buf *bp; 1148{ 1149 struct vnode *vp; 1150 struct buflists *listheadp; 1151 int s; 1152 1153 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1154 1155 /* 1156 * Delete from old vnode list, if on one. 1157 */ 1158 vp = bp->b_vp; 1159 s = splbio(); 1160 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1161 if (bp->b_xflags & BX_VNDIRTY) 1162 listheadp = &vp->v_dirtyblkhd; 1163 else 1164 listheadp = &vp->v_cleanblkhd; 1165 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1166 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1167 } 1168 if ((vp->v_flag & VONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1169 vp->v_flag &= ~VONWORKLST; 1170 LIST_REMOVE(vp, v_synclist); 1171 } 1172 splx(s); 1173 bp->b_vp = (struct vnode *) 0; 1174 vdrop(vp); 1175} 1176 1177/* 1178 * Add an item to the syncer work queue. 1179 */ 1180static void 1181vn_syncer_add_to_worklist(struct vnode *vp, int delay) 1182{ 1183 int s, slot; 1184 1185 s = splbio(); 1186 1187 if (vp->v_flag & VONWORKLST) { 1188 LIST_REMOVE(vp, v_synclist); 1189 } 1190 1191 if (delay > syncer_maxdelay - 2) 1192 delay = syncer_maxdelay - 2; 1193 slot = (syncer_delayno + delay) & syncer_mask; 1194 1195 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 1196 vp->v_flag |= VONWORKLST; 1197 splx(s); 1198} 1199 1200struct proc *updateproc; 1201static void sched_sync __P((void)); 1202static struct kproc_desc up_kp = { 1203 "syncer", 1204 sched_sync, 1205 &updateproc 1206}; 1207SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1208 1209/* 1210 * System filesystem synchronizer daemon. 1211 */ 1212void 1213sched_sync(void) 1214{ 1215 struct synclist *slp; 1216 struct vnode *vp; 1217 struct mount *mp; 1218 long starttime; 1219 int s; 1220 struct thread *td = &updateproc->p_thread; /* XXXKSE */ 1221 1222 mtx_lock(&Giant); 1223 1224 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc, 1225 SHUTDOWN_PRI_LAST); 1226 1227 for (;;) { 1228 kthread_suspend_check(td->td_proc); 1229 1230 starttime = time_second; 1231 1232 /* 1233 * Push files whose dirty time has expired. Be careful 1234 * of interrupt race on slp queue. 1235 */ 1236 s = splbio(); 1237 slp = &syncer_workitem_pending[syncer_delayno]; 1238 syncer_delayno += 1; 1239 if (syncer_delayno == syncer_maxdelay) 1240 syncer_delayno = 0; 1241 splx(s); 1242 1243 while ((vp = LIST_FIRST(slp)) != NULL) { 1244 if (VOP_ISLOCKED(vp, NULL) == 0 && 1245 vn_start_write(vp, &mp, V_NOWAIT) == 0) { 1246 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1247 (void) VOP_FSYNC(vp, td->td_proc->p_ucred, MNT_LAZY, td); 1248 VOP_UNLOCK(vp, 0, td); 1249 vn_finished_write(mp); 1250 } 1251 s = splbio(); 1252 if (LIST_FIRST(slp) == vp) { 1253 /* 1254 * Note: v_tag VT_VFS vps can remain on the 1255 * worklist too with no dirty blocks, but 1256 * since sync_fsync() moves it to a different 1257 * slot we are safe. 1258 */ 1259 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) && 1260 !vn_isdisk(vp, NULL)) 1261 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag); 1262 /* 1263 * Put us back on the worklist. The worklist 1264 * routine will remove us from our current 1265 * position and then add us back in at a later 1266 * position. 1267 */ 1268 vn_syncer_add_to_worklist(vp, syncdelay); 1269 } 1270 splx(s); 1271 } 1272 1273 /* 1274 * Do soft update processing. 1275 */ 1276#ifdef SOFTUPDATES 1277 softdep_process_worklist(NULL); 1278#endif 1279 1280 /* 1281 * The variable rushjob allows the kernel to speed up the 1282 * processing of the filesystem syncer process. A rushjob 1283 * value of N tells the filesystem syncer to process the next 1284 * N seconds worth of work on its queue ASAP. Currently rushjob 1285 * is used by the soft update code to speed up the filesystem 1286 * syncer process when the incore state is getting so far 1287 * ahead of the disk that the kernel memory pool is being 1288 * threatened with exhaustion. 1289 */ 1290 if (rushjob > 0) { 1291 rushjob -= 1; 1292 continue; 1293 } 1294 /* 1295 * If it has taken us less than a second to process the 1296 * current work, then wait. Otherwise start right over 1297 * again. We can still lose time if any single round 1298 * takes more than two seconds, but it does not really 1299 * matter as we are just trying to generally pace the 1300 * filesystem activity. 1301 */ 1302 if (time_second == starttime) 1303 tsleep(&lbolt, PPAUSE, "syncer", 0); 1304 } 1305} 1306 1307/* 1308 * Request the syncer daemon to speed up its work. 1309 * We never push it to speed up more than half of its 1310 * normal turn time, otherwise it could take over the cpu. 1311 * XXXKSE only one update? 1312 */ 1313int 1314speedup_syncer() 1315{ 1316 1317 mtx_lock_spin(&sched_lock); 1318 if (updateproc->p_thread.td_wchan == &lbolt) /* XXXKSE */ 1319 setrunnable(&updateproc->p_thread); 1320 mtx_unlock_spin(&sched_lock); 1321 if (rushjob < syncdelay / 2) { 1322 rushjob += 1; 1323 stat_rush_requests += 1; 1324 return (1); 1325 } 1326 return(0); 1327} 1328 1329/* 1330 * Associate a p-buffer with a vnode. 1331 * 1332 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1333 * with the buffer. i.e. the bp has not been linked into the vnode or 1334 * ref-counted. 1335 */ 1336void 1337pbgetvp(vp, bp) 1338 register struct vnode *vp; 1339 register struct buf *bp; 1340{ 1341 1342 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1343 1344 bp->b_vp = vp; 1345 bp->b_flags |= B_PAGING; 1346 bp->b_dev = vn_todev(vp); 1347} 1348 1349/* 1350 * Disassociate a p-buffer from a vnode. 1351 */ 1352void 1353pbrelvp(bp) 1354 register struct buf *bp; 1355{ 1356 1357 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1358 1359 /* XXX REMOVE ME */ 1360 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 1361 panic( 1362 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1363 bp, 1364 (int)bp->b_flags 1365 ); 1366 } 1367 bp->b_vp = (struct vnode *) 0; 1368 bp->b_flags &= ~B_PAGING; 1369} 1370 1371/* 1372 * Change the vnode a pager buffer is associated with. 1373 */ 1374void 1375pbreassignbuf(bp, newvp) 1376 struct buf *bp; 1377 struct vnode *newvp; 1378{ 1379 1380 KASSERT(bp->b_flags & B_PAGING, 1381 ("pbreassignbuf() on non phys bp %p", bp)); 1382 bp->b_vp = newvp; 1383} 1384 1385/* 1386 * Reassign a buffer from one vnode to another. 1387 * Used to assign file specific control information 1388 * (indirect blocks) to the vnode to which they belong. 1389 */ 1390void 1391reassignbuf(bp, newvp) 1392 register struct buf *bp; 1393 register struct vnode *newvp; 1394{ 1395 struct buflists *listheadp; 1396 int delay; 1397 int s; 1398 1399 if (newvp == NULL) { 1400 printf("reassignbuf: NULL"); 1401 return; 1402 } 1403 ++reassignbufcalls; 1404 1405 /* 1406 * B_PAGING flagged buffers cannot be reassigned because their vp 1407 * is not fully linked in. 1408 */ 1409 if (bp->b_flags & B_PAGING) 1410 panic("cannot reassign paging buffer"); 1411 1412 s = splbio(); 1413 /* 1414 * Delete from old vnode list, if on one. 1415 */ 1416 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1417 if (bp->b_xflags & BX_VNDIRTY) 1418 listheadp = &bp->b_vp->v_dirtyblkhd; 1419 else 1420 listheadp = &bp->b_vp->v_cleanblkhd; 1421 TAILQ_REMOVE(listheadp, bp, b_vnbufs); 1422 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1423 if (bp->b_vp != newvp) { 1424 vdrop(bp->b_vp); 1425 bp->b_vp = NULL; /* for clarification */ 1426 } 1427 } 1428 /* 1429 * If dirty, put on list of dirty buffers; otherwise insert onto list 1430 * of clean buffers. 1431 */ 1432 if (bp->b_flags & B_DELWRI) { 1433 struct buf *tbp; 1434 1435 listheadp = &newvp->v_dirtyblkhd; 1436 if ((newvp->v_flag & VONWORKLST) == 0) { 1437 switch (newvp->v_type) { 1438 case VDIR: 1439 delay = dirdelay; 1440 break; 1441 case VCHR: 1442 if (newvp->v_rdev->si_mountpoint != NULL) { 1443 delay = metadelay; 1444 break; 1445 } 1446 /* fall through */ 1447 default: 1448 delay = filedelay; 1449 } 1450 vn_syncer_add_to_worklist(newvp, delay); 1451 } 1452 bp->b_xflags |= BX_VNDIRTY; 1453 tbp = TAILQ_FIRST(listheadp); 1454 if (tbp == NULL || 1455 bp->b_lblkno == 0 || 1456 (bp->b_lblkno > 0 && tbp->b_lblkno < 0) || 1457 (bp->b_lblkno > 0 && bp->b_lblkno < tbp->b_lblkno)) { 1458 TAILQ_INSERT_HEAD(listheadp, bp, b_vnbufs); 1459 ++reassignbufsortgood; 1460 } else if (bp->b_lblkno < 0) { 1461 TAILQ_INSERT_TAIL(listheadp, bp, b_vnbufs); 1462 ++reassignbufsortgood; 1463 } else if (reassignbufmethod == 1) { 1464 /* 1465 * New sorting algorithm, only handle sequential case, 1466 * otherwise append to end (but before metadata) 1467 */ 1468 if ((tbp = gbincore(newvp, bp->b_lblkno - 1)) != NULL && 1469 (tbp->b_xflags & BX_VNDIRTY)) { 1470 /* 1471 * Found the best place to insert the buffer 1472 */ 1473 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1474 ++reassignbufsortgood; 1475 } else { 1476 /* 1477 * Missed, append to end, but before meta-data. 1478 * We know that the head buffer in the list is 1479 * not meta-data due to prior conditionals. 1480 * 1481 * Indirect effects: NFS second stage write 1482 * tends to wind up here, giving maximum 1483 * distance between the unstable write and the 1484 * commit rpc. 1485 */ 1486 tbp = TAILQ_LAST(listheadp, buflists); 1487 while (tbp && tbp->b_lblkno < 0) 1488 tbp = TAILQ_PREV(tbp, buflists, b_vnbufs); 1489 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1490 ++reassignbufsortbad; 1491 } 1492 } else { 1493 /* 1494 * Old sorting algorithm, scan queue and insert 1495 */ 1496 struct buf *ttbp; 1497 while ((ttbp = TAILQ_NEXT(tbp, b_vnbufs)) && 1498 (ttbp->b_lblkno < bp->b_lblkno)) { 1499 ++reassignbufloops; 1500 tbp = ttbp; 1501 } 1502 TAILQ_INSERT_AFTER(listheadp, tbp, bp, b_vnbufs); 1503 } 1504 } else { 1505 bp->b_xflags |= BX_VNCLEAN; 1506 TAILQ_INSERT_TAIL(&newvp->v_cleanblkhd, bp, b_vnbufs); 1507 if ((newvp->v_flag & VONWORKLST) && 1508 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1509 newvp->v_flag &= ~VONWORKLST; 1510 LIST_REMOVE(newvp, v_synclist); 1511 } 1512 } 1513 if (bp->b_vp != newvp) { 1514 bp->b_vp = newvp; 1515 vhold(bp->b_vp); 1516 } 1517 splx(s); 1518} 1519 1520/* 1521 * Create a vnode for a device. 1522 * Used for mounting the root file system. 1523 */ 1524int 1525bdevvp(dev, vpp) 1526 dev_t dev; 1527 struct vnode **vpp; 1528{ 1529 register struct vnode *vp; 1530 struct vnode *nvp; 1531 int error; 1532 1533 if (dev == NODEV) { 1534 *vpp = NULLVP; 1535 return (ENXIO); 1536 } 1537 if (vfinddev(dev, VCHR, vpp)) 1538 return (0); 1539 error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp); 1540 if (error) { 1541 *vpp = NULLVP; 1542 return (error); 1543 } 1544 vp = nvp; 1545 vp->v_type = VCHR; 1546 addalias(vp, dev); 1547 *vpp = vp; 1548 return (0); 1549} 1550 1551/* 1552 * Add vnode to the alias list hung off the dev_t. 1553 * 1554 * The reason for this gunk is that multiple vnodes can reference 1555 * the same physical device, so checking vp->v_usecount to see 1556 * how many users there are is inadequate; the v_usecount for 1557 * the vnodes need to be accumulated. vcount() does that. 1558 */ 1559struct vnode * 1560addaliasu(nvp, nvp_rdev) 1561 struct vnode *nvp; 1562 udev_t nvp_rdev; 1563{ 1564 struct vnode *ovp; 1565 vop_t **ops; 1566 dev_t dev; 1567 1568 if (nvp->v_type == VBLK) 1569 return (nvp); 1570 if (nvp->v_type != VCHR) 1571 panic("addaliasu on non-special vnode"); 1572 dev = udev2dev(nvp_rdev, 0); 1573 /* 1574 * Check to see if we have a bdevvp vnode with no associated 1575 * filesystem. If so, we want to associate the filesystem of 1576 * the new newly instigated vnode with the bdevvp vnode and 1577 * discard the newly created vnode rather than leaving the 1578 * bdevvp vnode lying around with no associated filesystem. 1579 */ 1580 if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) { 1581 addalias(nvp, dev); 1582 return (nvp); 1583 } 1584 /* 1585 * Discard unneeded vnode, but save its node specific data. 1586 * Note that if there is a lock, it is carried over in the 1587 * node specific data to the replacement vnode. 1588 */ 1589 vref(ovp); 1590 ovp->v_data = nvp->v_data; 1591 ovp->v_tag = nvp->v_tag; 1592 nvp->v_data = NULL; 1593 lockinit(&ovp->v_lock, PVFS, nvp->v_lock.lk_wmesg, 1594 nvp->v_lock.lk_timo, nvp->v_lock.lk_flags & LK_EXTFLG_MASK); 1595 if (nvp->v_vnlock) 1596 ovp->v_vnlock = &ovp->v_lock; 1597 ops = ovp->v_op; 1598 ovp->v_op = nvp->v_op; 1599 if (VOP_ISLOCKED(nvp, curthread)) { 1600 VOP_UNLOCK(nvp, 0, curthread); 1601 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread); 1602 } 1603 nvp->v_op = ops; 1604 insmntque(ovp, nvp->v_mount); 1605 vrele(nvp); 1606 vgone(nvp); 1607 return (ovp); 1608} 1609 1610/* This is a local helper function that do the same as addaliasu, but for a 1611 * dev_t instead of an udev_t. */ 1612static void 1613addalias(nvp, dev) 1614 struct vnode *nvp; 1615 dev_t dev; 1616{ 1617 1618 KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode")); 1619 nvp->v_rdev = dev; 1620 mtx_lock(&spechash_mtx); 1621 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1622 mtx_unlock(&spechash_mtx); 1623} 1624 1625/* 1626 * Grab a particular vnode from the free list, increment its 1627 * reference count and lock it. The vnode lock bit is set if the 1628 * vnode is being eliminated in vgone. The process is awakened 1629 * when the transition is completed, and an error returned to 1630 * indicate that the vnode is no longer usable (possibly having 1631 * been changed to a new file system type). 1632 */ 1633int 1634vget(vp, flags, td) 1635 register struct vnode *vp; 1636 int flags; 1637 struct thread *td; 1638{ 1639 int error; 1640 1641 /* 1642 * If the vnode is in the process of being cleaned out for 1643 * another use, we wait for the cleaning to finish and then 1644 * return failure. Cleaning is determined by checking that 1645 * the VXLOCK flag is set. 1646 */ 1647 if ((flags & LK_INTERLOCK) == 0) 1648 mtx_lock(&vp->v_interlock); 1649 if (vp->v_flag & VXLOCK) { 1650 if (vp->v_vxproc == curthread) { 1651#if 0 1652 /* this can now occur in normal operation */ 1653 log(LOG_INFO, "VXLOCK interlock avoided\n"); 1654#endif 1655 } else { 1656 vp->v_flag |= VXWANT; 1657 msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP, 1658 "vget", 0); 1659 return (ENOENT); 1660 } 1661 } 1662 1663 vp->v_usecount++; 1664 1665 if (VSHOULDBUSY(vp)) 1666 vbusy(vp); 1667 if (flags & LK_TYPE_MASK) { 1668 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) { 1669 /* 1670 * must expand vrele here because we do not want 1671 * to call VOP_INACTIVE if the reference count 1672 * drops back to zero since it was never really 1673 * active. We must remove it from the free list 1674 * before sleeping so that multiple processes do 1675 * not try to recycle it. 1676 */ 1677 mtx_lock(&vp->v_interlock); 1678 vp->v_usecount--; 1679 if (VSHOULDFREE(vp)) 1680 vfree(vp); 1681 else 1682 vlruvp(vp); 1683 mtx_unlock(&vp->v_interlock); 1684 } 1685 return (error); 1686 } 1687 mtx_unlock(&vp->v_interlock); 1688 return (0); 1689} 1690 1691/* 1692 * Increase the reference count of a vnode. 1693 */ 1694void 1695vref(struct vnode *vp) 1696{ 1697 mtx_lock(&vp->v_interlock); 1698 vp->v_usecount++; 1699 mtx_unlock(&vp->v_interlock); 1700} 1701 1702/* 1703 * Vnode put/release. 1704 * If count drops to zero, call inactive routine and return to freelist. 1705 */ 1706void 1707vrele(vp) 1708 struct vnode *vp; 1709{ 1710 struct thread *td = curthread; /* XXX */ 1711 1712 KASSERT(vp != NULL, ("vrele: null vp")); 1713 1714 mtx_lock(&vp->v_interlock); 1715 1716 /* Skip this v_writecount check if we're going to panic below. */ 1717 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, 1718 ("vrele: missed vn_close")); 1719 1720 if (vp->v_usecount > 1) { 1721 1722 vp->v_usecount--; 1723 mtx_unlock(&vp->v_interlock); 1724 1725 return; 1726 } 1727 1728 if (vp->v_usecount == 1) { 1729 vp->v_usecount--; 1730 /* 1731 * We must call VOP_INACTIVE with the node locked. 1732 * If we are doing a vput, the node is already locked, 1733 * but, in the case of vrele, we must explicitly lock 1734 * the vnode before calling VOP_INACTIVE. 1735 */ 1736 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) 1737 VOP_INACTIVE(vp, td); 1738 if (VSHOULDFREE(vp)) 1739 vfree(vp); 1740 else 1741 vlruvp(vp); 1742 1743 } else { 1744#ifdef DIAGNOSTIC 1745 vprint("vrele: negative ref count", vp); 1746 mtx_unlock(&vp->v_interlock); 1747#endif 1748 panic("vrele: negative ref cnt"); 1749 } 1750} 1751 1752/* 1753 * Release an already locked vnode. This give the same effects as 1754 * unlock+vrele(), but takes less time and avoids releasing and 1755 * re-aquiring the lock (as vrele() aquires the lock internally.) 1756 */ 1757void 1758vput(vp) 1759 struct vnode *vp; 1760{ 1761 struct thread *td = curthread; /* XXX */ 1762 1763 GIANT_REQUIRED; 1764 1765 KASSERT(vp != NULL, ("vput: null vp")); 1766 mtx_lock(&vp->v_interlock); 1767 /* Skip this v_writecount check if we're going to panic below. */ 1768 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, 1769 ("vput: missed vn_close")); 1770 1771 if (vp->v_usecount > 1) { 1772 vp->v_usecount--; 1773 VOP_UNLOCK(vp, LK_INTERLOCK, td); 1774 return; 1775 } 1776 1777 if (vp->v_usecount == 1) { 1778 vp->v_usecount--; 1779 /* 1780 * We must call VOP_INACTIVE with the node locked. 1781 * If we are doing a vput, the node is already locked, 1782 * so we just need to release the vnode mutex. 1783 */ 1784 mtx_unlock(&vp->v_interlock); 1785 VOP_INACTIVE(vp, td); 1786 if (VSHOULDFREE(vp)) 1787 vfree(vp); 1788 else 1789 vlruvp(vp); 1790 1791 } else { 1792#ifdef DIAGNOSTIC 1793 vprint("vput: negative ref count", vp); 1794#endif 1795 panic("vput: negative ref cnt"); 1796 } 1797} 1798 1799/* 1800 * Somebody doesn't want the vnode recycled. 1801 */ 1802void 1803vhold(vp) 1804 register struct vnode *vp; 1805{ 1806 int s; 1807 1808 s = splbio(); 1809 vp->v_holdcnt++; 1810 if (VSHOULDBUSY(vp)) 1811 vbusy(vp); 1812 splx(s); 1813} 1814 1815/* 1816 * Note that there is one less who cares about this vnode. vdrop() is the 1817 * opposite of vhold(). 1818 */ 1819void 1820vdrop(vp) 1821 register struct vnode *vp; 1822{ 1823 int s; 1824 1825 s = splbio(); 1826 if (vp->v_holdcnt <= 0) 1827 panic("vdrop: holdcnt"); 1828 vp->v_holdcnt--; 1829 if (VSHOULDFREE(vp)) 1830 vfree(vp); 1831 else 1832 vlruvp(vp); 1833 splx(s); 1834} 1835 1836/* 1837 * Remove any vnodes in the vnode table belonging to mount point mp. 1838 * 1839 * If FORCECLOSE is not specified, there should not be any active ones, 1840 * return error if any are found (nb: this is a user error, not a 1841 * system error). If FORCECLOSE is specified, detach any active vnodes 1842 * that are found. 1843 * 1844 * If WRITECLOSE is set, only flush out regular file vnodes open for 1845 * writing. 1846 * 1847 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 1848 * 1849 * `rootrefs' specifies the base reference count for the root vnode 1850 * of this filesystem. The root vnode is considered busy if its 1851 * v_usecount exceeds this value. On a successful return, vflush() 1852 * will call vrele() on the root vnode exactly rootrefs times. 1853 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 1854 * be zero. 1855 */ 1856#ifdef DIAGNOSTIC 1857static int busyprt = 0; /* print out busy vnodes */ 1858SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 1859#endif 1860 1861int 1862vflush(mp, rootrefs, flags) 1863 struct mount *mp; 1864 int rootrefs; 1865 int flags; 1866{ 1867 struct thread *td = curthread; /* XXX */ 1868 struct vnode *vp, *nvp, *rootvp = NULL; 1869 struct vattr vattr; 1870 int busy = 0, error; 1871 1872 if (rootrefs > 0) { 1873 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1874 ("vflush: bad args")); 1875 /* 1876 * Get the filesystem root vnode. We can vput() it 1877 * immediately, since with rootrefs > 0, it won't go away. 1878 */ 1879 if ((error = VFS_ROOT(mp, &rootvp)) != 0) 1880 return (error); 1881 vput(rootvp); 1882 } 1883 mtx_lock(&mntvnode_mtx); 1884loop: 1885 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) { 1886 /* 1887 * Make sure this vnode wasn't reclaimed in getnewvnode(). 1888 * Start over if it has (it won't be on the list anymore). 1889 */ 1890 if (vp->v_mount != mp) 1891 goto loop; 1892 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 1893 1894 mtx_unlock(&mntvnode_mtx); 1895 mtx_lock(&vp->v_interlock); 1896 /* 1897 * Skip over a vnodes marked VSYSTEM. 1898 */ 1899 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1900 mtx_unlock(&vp->v_interlock); 1901 mtx_lock(&mntvnode_mtx); 1902 continue; 1903 } 1904 /* 1905 * If WRITECLOSE is set, flush out unlinked but still open 1906 * files (even if open only for reading) and regular file 1907 * vnodes open for writing. 1908 */ 1909 if ((flags & WRITECLOSE) && 1910 (vp->v_type == VNON || 1911 (VOP_GETATTR(vp, &vattr, td->td_proc->p_ucred, td) == 0 && 1912 vattr.va_nlink > 0)) && 1913 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1914 mtx_unlock(&vp->v_interlock); 1915 mtx_lock(&mntvnode_mtx); 1916 continue; 1917 } 1918 1919 /* 1920 * With v_usecount == 0, all we need to do is clear out the 1921 * vnode data structures and we are done. 1922 */ 1923 if (vp->v_usecount == 0) { 1924 vgonel(vp, td); 1925 mtx_lock(&mntvnode_mtx); 1926 continue; 1927 } 1928 1929 /* 1930 * If FORCECLOSE is set, forcibly close the vnode. For block 1931 * or character devices, revert to an anonymous device. For 1932 * all other files, just kill them. 1933 */ 1934 if (flags & FORCECLOSE) { 1935 if (vp->v_type != VCHR) { 1936 vgonel(vp, td); 1937 } else { 1938 vclean(vp, 0, td); 1939 vp->v_op = spec_vnodeop_p; 1940 insmntque(vp, (struct mount *) 0); 1941 } 1942 mtx_lock(&mntvnode_mtx); 1943 continue; 1944 } 1945#ifdef DIAGNOSTIC 1946 if (busyprt) 1947 vprint("vflush: busy vnode", vp); 1948#endif 1949 mtx_unlock(&vp->v_interlock); 1950 mtx_lock(&mntvnode_mtx); 1951 busy++; 1952 } 1953 mtx_unlock(&mntvnode_mtx); 1954 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1955 /* 1956 * If just the root vnode is busy, and if its refcount 1957 * is equal to `rootrefs', then go ahead and kill it. 1958 */ 1959 mtx_lock(&rootvp->v_interlock); 1960 KASSERT(busy > 0, ("vflush: not busy")); 1961 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs")); 1962 if (busy == 1 && rootvp->v_usecount == rootrefs) { 1963 vgonel(rootvp, td); 1964 busy = 0; 1965 } else 1966 mtx_unlock(&rootvp->v_interlock); 1967 } 1968 if (busy) 1969 return (EBUSY); 1970 for (; rootrefs > 0; rootrefs--) 1971 vrele(rootvp); 1972 return (0); 1973} 1974 1975/* 1976 * This moves a now (likely recyclable) vnode to the end of the 1977 * mountlist. XXX However, it is temporarily disabled until we 1978 * can clean up ffs_sync() and friends, which have loop restart 1979 * conditions which this code causes to operate O(N^2). 1980 */ 1981static void 1982vlruvp(struct vnode *vp) 1983{ 1984#if 0 1985 struct mount *mp; 1986 1987 if ((mp = vp->v_mount) != NULL) { 1988 mtx_lock(&mntvnode_mtx); 1989 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1990 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1991 mtx_unlock(&mntvnode_mtx); 1992 } 1993#endif 1994} 1995 1996/* 1997 * Disassociate the underlying file system from a vnode. 1998 */ 1999static void 2000vclean(vp, flags, td) 2001 struct vnode *vp; 2002 int flags; 2003 struct thread *td; 2004{ 2005 int active; 2006 2007 /* 2008 * Check to see if the vnode is in use. If so we have to reference it 2009 * before we clean it out so that its count cannot fall to zero and 2010 * generate a race against ourselves to recycle it. 2011 */ 2012 if ((active = vp->v_usecount)) 2013 vp->v_usecount++; 2014 2015 /* 2016 * Prevent the vnode from being recycled or brought into use while we 2017 * clean it out. 2018 */ 2019 if (vp->v_flag & VXLOCK) 2020 panic("vclean: deadlock"); 2021 vp->v_flag |= VXLOCK; 2022 vp->v_vxproc = curthread; 2023 /* 2024 * Even if the count is zero, the VOP_INACTIVE routine may still 2025 * have the object locked while it cleans it out. The VOP_LOCK 2026 * ensures that the VOP_INACTIVE routine is done with its work. 2027 * For active vnodes, it ensures that no other activity can 2028 * occur while the underlying object is being cleaned out. 2029 */ 2030 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td); 2031 2032 /* 2033 * Clean out any buffers associated with the vnode. 2034 * If the flush fails, just toss the buffers. 2035 */ 2036 if (flags & DOCLOSE) { 2037 if (TAILQ_FIRST(&vp->v_dirtyblkhd) != NULL) 2038 (void) vn_write_suspend_wait(vp, NULL, V_WAIT); 2039 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0) 2040 vinvalbuf(vp, 0, NOCRED, td, 0, 0); 2041 } 2042 2043 VOP_DESTROYVOBJECT(vp); 2044 2045 /* 2046 * If purging an active vnode, it must be closed and 2047 * deactivated before being reclaimed. Note that the 2048 * VOP_INACTIVE will unlock the vnode. 2049 */ 2050 if (active) { 2051 if (flags & DOCLOSE) 2052 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2053 VOP_INACTIVE(vp, td); 2054 } else { 2055 /* 2056 * Any other processes trying to obtain this lock must first 2057 * wait for VXLOCK to clear, then call the new lock operation. 2058 */ 2059 VOP_UNLOCK(vp, 0, td); 2060 } 2061 /* 2062 * Reclaim the vnode. 2063 */ 2064 if (VOP_RECLAIM(vp, td)) 2065 panic("vclean: cannot reclaim"); 2066 2067 if (active) { 2068 /* 2069 * Inline copy of vrele() since VOP_INACTIVE 2070 * has already been called. 2071 */ 2072 mtx_lock(&vp->v_interlock); 2073 if (--vp->v_usecount <= 0) { 2074#ifdef DIAGNOSTIC 2075 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 2076 vprint("vclean: bad ref count", vp); 2077 panic("vclean: ref cnt"); 2078 } 2079#endif 2080 vfree(vp); 2081 } 2082 mtx_unlock(&vp->v_interlock); 2083 } 2084 2085 cache_purge(vp); 2086 vp->v_vnlock = NULL; 2087 lockdestroy(&vp->v_lock); 2088 2089 if (VSHOULDFREE(vp)) 2090 vfree(vp); 2091 2092 /* 2093 * Done with purge, notify sleepers of the grim news. 2094 */ 2095 vp->v_op = dead_vnodeop_p; 2096 vn_pollgone(vp); 2097 vp->v_tag = VT_NON; 2098 vp->v_flag &= ~VXLOCK; 2099 vp->v_vxproc = NULL; 2100 if (vp->v_flag & VXWANT) { 2101 vp->v_flag &= ~VXWANT; 2102 wakeup((caddr_t) vp); 2103 } 2104} 2105 2106/* 2107 * Eliminate all activity associated with the requested vnode 2108 * and with all vnodes aliased to the requested vnode. 2109 */ 2110int 2111vop_revoke(ap) 2112 struct vop_revoke_args /* { 2113 struct vnode *a_vp; 2114 int a_flags; 2115 } */ *ap; 2116{ 2117 struct vnode *vp, *vq; 2118 dev_t dev; 2119 2120 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 2121 2122 vp = ap->a_vp; 2123 /* 2124 * If a vgone (or vclean) is already in progress, 2125 * wait until it is done and return. 2126 */ 2127 if (vp->v_flag & VXLOCK) { 2128 vp->v_flag |= VXWANT; 2129 msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP, 2130 "vop_revokeall", 0); 2131 return (0); 2132 } 2133 dev = vp->v_rdev; 2134 for (;;) { 2135 mtx_lock(&spechash_mtx); 2136 vq = SLIST_FIRST(&dev->si_hlist); 2137 mtx_unlock(&spechash_mtx); 2138 if (!vq) 2139 break; 2140 vgone(vq); 2141 } 2142 return (0); 2143} 2144 2145/* 2146 * Recycle an unused vnode to the front of the free list. 2147 * Release the passed interlock if the vnode will be recycled. 2148 */ 2149int 2150vrecycle(vp, inter_lkp, td) 2151 struct vnode *vp; 2152 struct mtx *inter_lkp; 2153 struct thread *td; 2154{ 2155 2156 mtx_lock(&vp->v_interlock); 2157 if (vp->v_usecount == 0) { 2158 if (inter_lkp) { 2159 mtx_unlock(inter_lkp); 2160 } 2161 vgonel(vp, td); 2162 return (1); 2163 } 2164 mtx_unlock(&vp->v_interlock); 2165 return (0); 2166} 2167 2168/* 2169 * Eliminate all activity associated with a vnode 2170 * in preparation for reuse. 2171 */ 2172void 2173vgone(vp) 2174 register struct vnode *vp; 2175{ 2176 struct thread *td = curthread; /* XXX */ 2177 2178 mtx_lock(&vp->v_interlock); 2179 vgonel(vp, td); 2180} 2181 2182/* 2183 * vgone, with the vp interlock held. 2184 */ 2185void 2186vgonel(vp, td) 2187 struct vnode *vp; 2188 struct thread *td; 2189{ 2190 int s; 2191 2192 /* 2193 * If a vgone (or vclean) is already in progress, 2194 * wait until it is done and return. 2195 */ 2196 if (vp->v_flag & VXLOCK) { 2197 vp->v_flag |= VXWANT; 2198 msleep((caddr_t)vp, &vp->v_interlock, PINOD | PDROP, 2199 "vgone", 0); 2200 return; 2201 } 2202 2203 /* 2204 * Clean out the filesystem specific data. 2205 */ 2206 vclean(vp, DOCLOSE, td); 2207 mtx_lock(&vp->v_interlock); 2208 2209 /* 2210 * Delete from old mount point vnode list, if on one. 2211 */ 2212 if (vp->v_mount != NULL) 2213 insmntque(vp, (struct mount *)0); 2214 /* 2215 * If special device, remove it from special device alias list 2216 * if it is on one. 2217 */ 2218 if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) { 2219 mtx_lock(&spechash_mtx); 2220 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext); 2221 freedev(vp->v_rdev); 2222 mtx_unlock(&spechash_mtx); 2223 vp->v_rdev = NULL; 2224 } 2225 2226 /* 2227 * If it is on the freelist and not already at the head, 2228 * move it to the head of the list. The test of the 2229 * VDOOMED flag and the reference count of zero is because 2230 * it will be removed from the free list by getnewvnode, 2231 * but will not have its reference count incremented until 2232 * after calling vgone. If the reference count were 2233 * incremented first, vgone would (incorrectly) try to 2234 * close the previous instance of the underlying object. 2235 */ 2236 if (vp->v_usecount == 0 && !(vp->v_flag & VDOOMED)) { 2237 s = splbio(); 2238 mtx_lock(&vnode_free_list_mtx); 2239 if (vp->v_flag & VFREE) 2240 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2241 else 2242 freevnodes++; 2243 vp->v_flag |= VFREE; 2244 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2245 mtx_unlock(&vnode_free_list_mtx); 2246 splx(s); 2247 } 2248 2249 vp->v_type = VBAD; 2250 mtx_unlock(&vp->v_interlock); 2251} 2252 2253/* 2254 * Lookup a vnode by device number. 2255 */ 2256int 2257vfinddev(dev, type, vpp) 2258 dev_t dev; 2259 enum vtype type; 2260 struct vnode **vpp; 2261{ 2262 struct vnode *vp; 2263 2264 mtx_lock(&spechash_mtx); 2265 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 2266 if (type == vp->v_type) { 2267 *vpp = vp; 2268 mtx_unlock(&spechash_mtx); 2269 return (1); 2270 } 2271 } 2272 mtx_unlock(&spechash_mtx); 2273 return (0); 2274} 2275 2276/* 2277 * Calculate the total number of references to a special device. 2278 */ 2279int 2280vcount(vp) 2281 struct vnode *vp; 2282{ 2283 struct vnode *vq; 2284 int count; 2285 2286 count = 0; 2287 mtx_lock(&spechash_mtx); 2288 SLIST_FOREACH(vq, &vp->v_rdev->si_hlist, v_specnext) 2289 count += vq->v_usecount; 2290 mtx_unlock(&spechash_mtx); 2291 return (count); 2292} 2293 2294/* 2295 * Same as above, but using the dev_t as argument 2296 */ 2297int 2298count_dev(dev) 2299 dev_t dev; 2300{ 2301 struct vnode *vp; 2302 2303 vp = SLIST_FIRST(&dev->si_hlist); 2304 if (vp == NULL) 2305 return (0); 2306 return(vcount(vp)); 2307} 2308 2309/* 2310 * Print out a description of a vnode. 2311 */ 2312static char *typename[] = 2313{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 2314 2315void 2316vprint(label, vp) 2317 char *label; 2318 struct vnode *vp; 2319{ 2320 char buf[96]; 2321 2322 if (label != NULL) 2323 printf("%s: %p: ", label, (void *)vp); 2324 else 2325 printf("%p: ", (void *)vp); 2326 printf("type %s, usecount %d, writecount %d, refcount %d,", 2327 typename[vp->v_type], vp->v_usecount, vp->v_writecount, 2328 vp->v_holdcnt); 2329 buf[0] = '\0'; 2330 if (vp->v_flag & VROOT) 2331 strcat(buf, "|VROOT"); 2332 if (vp->v_flag & VTEXT) 2333 strcat(buf, "|VTEXT"); 2334 if (vp->v_flag & VSYSTEM) 2335 strcat(buf, "|VSYSTEM"); 2336 if (vp->v_flag & VXLOCK) 2337 strcat(buf, "|VXLOCK"); 2338 if (vp->v_flag & VXWANT) 2339 strcat(buf, "|VXWANT"); 2340 if (vp->v_flag & VBWAIT) 2341 strcat(buf, "|VBWAIT"); 2342 if (vp->v_flag & VDOOMED) 2343 strcat(buf, "|VDOOMED"); 2344 if (vp->v_flag & VFREE) 2345 strcat(buf, "|VFREE"); 2346 if (vp->v_flag & VOBJBUF) 2347 strcat(buf, "|VOBJBUF"); 2348 if (buf[0] != '\0') 2349 printf(" flags (%s)", &buf[1]); 2350 if (vp->v_data == NULL) { 2351 printf("\n"); 2352 } else { 2353 printf("\n\t"); 2354 VOP_PRINT(vp); 2355 } 2356} 2357 2358#ifdef DDB 2359#include <ddb/ddb.h> 2360/* 2361 * List all of the locked vnodes in the system. 2362 * Called when debugging the kernel. 2363 */ 2364DB_SHOW_COMMAND(lockedvnodes, lockedvnodes) 2365{ 2366 struct thread *td = curthread; /* XXX */ 2367 struct mount *mp, *nmp; 2368 struct vnode *vp; 2369 2370 printf("Locked vnodes\n"); 2371 mtx_lock(&mountlist_mtx); 2372 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2373 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 2374 nmp = TAILQ_NEXT(mp, mnt_list); 2375 continue; 2376 } 2377 mtx_lock(&mntvnode_mtx); 2378 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2379 if (VOP_ISLOCKED(vp, NULL)) 2380 vprint((char *)0, vp); 2381 } 2382 mtx_unlock(&mntvnode_mtx); 2383 mtx_lock(&mountlist_mtx); 2384 nmp = TAILQ_NEXT(mp, mnt_list); 2385 vfs_unbusy(mp, td); 2386 } 2387 mtx_unlock(&mountlist_mtx); 2388} 2389#endif 2390 2391/* 2392 * Top level filesystem related information gathering. 2393 */ 2394static int sysctl_ovfs_conf __P((SYSCTL_HANDLER_ARGS)); 2395 2396static int 2397vfs_sysctl(SYSCTL_HANDLER_ARGS) 2398{ 2399 int *name = (int *)arg1 - 1; /* XXX */ 2400 u_int namelen = arg2 + 1; /* XXX */ 2401 struct vfsconf *vfsp; 2402 2403#if 1 || defined(COMPAT_PRELITE2) 2404 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2405 if (namelen == 1) 2406 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2407#endif 2408 2409 /* XXX the below code does not compile; vfs_sysctl does not exist. */ 2410#ifdef notyet 2411 /* all sysctl names at this level are at least name and field */ 2412 if (namelen < 2) 2413 return (ENOTDIR); /* overloaded */ 2414 if (name[0] != VFS_GENERIC) { 2415 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2416 if (vfsp->vfc_typenum == name[0]) 2417 break; 2418 if (vfsp == NULL) 2419 return (EOPNOTSUPP); 2420 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 2421 oldp, oldlenp, newp, newlen, td)); 2422 } 2423#endif 2424 switch (name[1]) { 2425 case VFS_MAXTYPENUM: 2426 if (namelen != 2) 2427 return (ENOTDIR); 2428 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2429 case VFS_CONF: 2430 if (namelen != 3) 2431 return (ENOTDIR); /* overloaded */ 2432 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2433 if (vfsp->vfc_typenum == name[2]) 2434 break; 2435 if (vfsp == NULL) 2436 return (EOPNOTSUPP); 2437 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp)); 2438 } 2439 return (EOPNOTSUPP); 2440} 2441 2442SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl, 2443 "Generic filesystem"); 2444 2445#if 1 || defined(COMPAT_PRELITE2) 2446 2447static int 2448sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2449{ 2450 int error; 2451 struct vfsconf *vfsp; 2452 struct ovfsconf ovfs; 2453 2454 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2455 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2456 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2457 ovfs.vfc_index = vfsp->vfc_typenum; 2458 ovfs.vfc_refcount = vfsp->vfc_refcount; 2459 ovfs.vfc_flags = vfsp->vfc_flags; 2460 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2461 if (error) 2462 return error; 2463 } 2464 return 0; 2465} 2466 2467#endif /* 1 || COMPAT_PRELITE2 */ 2468 2469#if COMPILING_LINT 2470#define KINFO_VNODESLOP 10 2471/* 2472 * Dump vnode list (via sysctl). 2473 * Copyout address of vnode followed by vnode. 2474 */ 2475/* ARGSUSED */ 2476static int 2477sysctl_vnode(SYSCTL_HANDLER_ARGS) 2478{ 2479 struct thread *td = curthread; /* XXX */ 2480 struct mount *mp, *nmp; 2481 struct vnode *nvp, *vp; 2482 int error; 2483 2484#define VPTRSZ sizeof (struct vnode *) 2485#define VNODESZ sizeof (struct vnode) 2486 2487 req->lock = 0; 2488 if (!req->oldptr) /* Make an estimate */ 2489 return (SYSCTL_OUT(req, 0, 2490 (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ))); 2491 2492 mtx_lock(&mountlist_mtx); 2493 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2494 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 2495 nmp = TAILQ_NEXT(mp, mnt_list); 2496 continue; 2497 } 2498 mtx_lock(&mntvnode_mtx); 2499again: 2500 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 2501 vp != NULL; 2502 vp = nvp) { 2503 /* 2504 * Check that the vp is still associated with 2505 * this filesystem. RACE: could have been 2506 * recycled onto the same filesystem. 2507 */ 2508 if (vp->v_mount != mp) 2509 goto again; 2510 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2511 mtx_unlock(&mntvnode_mtx); 2512 if ((error = SYSCTL_OUT(req, &vp, VPTRSZ)) || 2513 (error = SYSCTL_OUT(req, vp, VNODESZ))) 2514 return (error); 2515 mtx_lock(&mntvnode_mtx); 2516 } 2517 mtx_unlock(&mntvnode_mtx); 2518 mtx_lock(&mountlist_mtx); 2519 nmp = TAILQ_NEXT(mp, mnt_list); 2520 vfs_unbusy(mp, td); 2521 } 2522 mtx_unlock(&mountlist_mtx); 2523 2524 return (0); 2525} 2526 2527/* 2528 * XXX 2529 * Exporting the vnode list on large systems causes them to crash. 2530 * Exporting the vnode list on medium systems causes sysctl to coredump. 2531 */ 2532SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2533 0, 0, sysctl_vnode, "S,vnode", ""); 2534#endif 2535 2536/* 2537 * Check to see if a filesystem is mounted on a block device. 2538 */ 2539int 2540vfs_mountedon(vp) 2541 struct vnode *vp; 2542{ 2543 2544 if (vp->v_rdev->si_mountpoint != NULL) 2545 return (EBUSY); 2546 return (0); 2547} 2548 2549/* 2550 * Unmount all filesystems. The list is traversed in reverse order 2551 * of mounting to avoid dependencies. 2552 */ 2553void 2554vfs_unmountall() 2555{ 2556 struct mount *mp; 2557 struct thread *td; 2558 int error; 2559 2560 if (curthread != NULL) 2561 td = curthread; 2562 else 2563 td = &initproc->p_thread; /* XXX XXX should this be proc0? */ 2564 /* 2565 * Since this only runs when rebooting, it is not interlocked. 2566 */ 2567 while(!TAILQ_EMPTY(&mountlist)) { 2568 mp = TAILQ_LAST(&mountlist, mntlist); 2569 error = dounmount(mp, MNT_FORCE, td); 2570 if (error) { 2571 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2572 printf("unmount of %s failed (", 2573 mp->mnt_stat.f_mntonname); 2574 if (error == EBUSY) 2575 printf("BUSY)\n"); 2576 else 2577 printf("%d)\n", error); 2578 } else { 2579 /* The unmount has removed mp from the mountlist */ 2580 } 2581 } 2582} 2583 2584/* 2585 * perform msync on all vnodes under a mount point 2586 * the mount point must be locked. 2587 */ 2588void 2589vfs_msync(struct mount *mp, int flags) 2590{ 2591 struct vnode *vp, *nvp; 2592 struct vm_object *obj; 2593 int tries; 2594 2595 GIANT_REQUIRED; 2596 2597 tries = 5; 2598 mtx_lock(&mntvnode_mtx); 2599loop: 2600 for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) { 2601 if (vp->v_mount != mp) { 2602 if (--tries > 0) 2603 goto loop; 2604 break; 2605 } 2606 nvp = TAILQ_NEXT(vp, v_nmntvnodes); 2607 2608 if (vp->v_flag & VXLOCK) /* XXX: what if MNT_WAIT? */ 2609 continue; 2610 2611 if (vp->v_flag & VNOSYNC) /* unlinked, skip it */ 2612 continue; 2613 2614 if ((vp->v_flag & VOBJDIRTY) && 2615 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 2616 mtx_unlock(&mntvnode_mtx); 2617 if (!vget(vp, 2618 LK_EXCLUSIVE | LK_RETRY | LK_NOOBJ, curthread)) { 2619 if (VOP_GETVOBJECT(vp, &obj) == 0) { 2620 vm_object_page_clean(obj, 0, 0, 2621 flags == MNT_WAIT ? 2622 OBJPC_SYNC : OBJPC_NOSYNC); 2623 } 2624 vput(vp); 2625 } 2626 mtx_lock(&mntvnode_mtx); 2627 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) { 2628 if (--tries > 0) 2629 goto loop; 2630 break; 2631 } 2632 } 2633 } 2634 mtx_unlock(&mntvnode_mtx); 2635} 2636 2637/* 2638 * Create the VM object needed for VMIO and mmap support. This 2639 * is done for all VREG files in the system. Some filesystems might 2640 * afford the additional metadata buffering capability of the 2641 * VMIO code by making the device node be VMIO mode also. 2642 * 2643 * vp must be locked when vfs_object_create is called. 2644 */ 2645int 2646vfs_object_create(vp, td, cred) 2647 struct vnode *vp; 2648 struct thread *td; 2649 struct ucred *cred; 2650{ 2651 GIANT_REQUIRED; 2652 return (VOP_CREATEVOBJECT(vp, cred, td)); 2653} 2654 2655/* 2656 * Mark a vnode as free, putting it up for recycling. 2657 */ 2658void 2659vfree(vp) 2660 struct vnode *vp; 2661{ 2662 int s; 2663 2664 s = splbio(); 2665 mtx_lock(&vnode_free_list_mtx); 2666 KASSERT((vp->v_flag & VFREE) == 0, ("vnode already free")); 2667 if (vp->v_flag & VAGE) { 2668 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2669 } else { 2670 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2671 } 2672 freevnodes++; 2673 mtx_unlock(&vnode_free_list_mtx); 2674 vp->v_flag &= ~VAGE; 2675 vp->v_flag |= VFREE; 2676 splx(s); 2677} 2678 2679/* 2680 * Opposite of vfree() - mark a vnode as in use. 2681 */ 2682void 2683vbusy(vp) 2684 struct vnode *vp; 2685{ 2686 int s; 2687 2688 s = splbio(); 2689 mtx_lock(&vnode_free_list_mtx); 2690 KASSERT((vp->v_flag & VFREE) != 0, ("vnode not free")); 2691 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2692 freevnodes--; 2693 mtx_unlock(&vnode_free_list_mtx); 2694 vp->v_flag &= ~(VFREE|VAGE); 2695 splx(s); 2696} 2697 2698/* 2699 * Record a process's interest in events which might happen to 2700 * a vnode. Because poll uses the historic select-style interface 2701 * internally, this routine serves as both the ``check for any 2702 * pending events'' and the ``record my interest in future events'' 2703 * functions. (These are done together, while the lock is held, 2704 * to avoid race conditions.) 2705 */ 2706int 2707vn_pollrecord(vp, td, events) 2708 struct vnode *vp; 2709 struct thread *td; 2710 short events; 2711{ 2712 mtx_lock(&vp->v_pollinfo.vpi_lock); 2713 if (vp->v_pollinfo.vpi_revents & events) { 2714 /* 2715 * This leaves events we are not interested 2716 * in available for the other process which 2717 * which presumably had requested them 2718 * (otherwise they would never have been 2719 * recorded). 2720 */ 2721 events &= vp->v_pollinfo.vpi_revents; 2722 vp->v_pollinfo.vpi_revents &= ~events; 2723 2724 mtx_unlock(&vp->v_pollinfo.vpi_lock); 2725 return events; 2726 } 2727 vp->v_pollinfo.vpi_events |= events; 2728 selrecord(td, &vp->v_pollinfo.vpi_selinfo); 2729 mtx_unlock(&vp->v_pollinfo.vpi_lock); 2730 return 0; 2731} 2732 2733/* 2734 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 2735 * it is possible for us to miss an event due to race conditions, but 2736 * that condition is expected to be rare, so for the moment it is the 2737 * preferred interface. 2738 */ 2739void 2740vn_pollevent(vp, events) 2741 struct vnode *vp; 2742 short events; 2743{ 2744 mtx_lock(&vp->v_pollinfo.vpi_lock); 2745 if (vp->v_pollinfo.vpi_events & events) { 2746 /* 2747 * We clear vpi_events so that we don't 2748 * call selwakeup() twice if two events are 2749 * posted before the polling process(es) is 2750 * awakened. This also ensures that we take at 2751 * most one selwakeup() if the polling process 2752 * is no longer interested. However, it does 2753 * mean that only one event can be noticed at 2754 * a time. (Perhaps we should only clear those 2755 * event bits which we note?) XXX 2756 */ 2757 vp->v_pollinfo.vpi_events = 0; /* &= ~events ??? */ 2758 vp->v_pollinfo.vpi_revents |= events; 2759 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2760 } 2761 mtx_unlock(&vp->v_pollinfo.vpi_lock); 2762} 2763 2764#define VN_KNOTE(vp, b) \ 2765 KNOTE((struct klist *)&vp->v_pollinfo.vpi_selinfo.si_note, (b)) 2766 2767/* 2768 * Wake up anyone polling on vp because it is being revoked. 2769 * This depends on dead_poll() returning POLLHUP for correct 2770 * behavior. 2771 */ 2772void 2773vn_pollgone(vp) 2774 struct vnode *vp; 2775{ 2776 mtx_lock(&vp->v_pollinfo.vpi_lock); 2777 VN_KNOTE(vp, NOTE_REVOKE); 2778 if (vp->v_pollinfo.vpi_events) { 2779 vp->v_pollinfo.vpi_events = 0; 2780 selwakeup(&vp->v_pollinfo.vpi_selinfo); 2781 } 2782 mtx_unlock(&vp->v_pollinfo.vpi_lock); 2783} 2784 2785 2786 2787/* 2788 * Routine to create and manage a filesystem syncer vnode. 2789 */ 2790#define sync_close ((int (*) __P((struct vop_close_args *)))nullop) 2791static int sync_fsync __P((struct vop_fsync_args *)); 2792static int sync_inactive __P((struct vop_inactive_args *)); 2793static int sync_reclaim __P((struct vop_reclaim_args *)); 2794#define sync_lock ((int (*) __P((struct vop_lock_args *)))vop_nolock) 2795#define sync_unlock ((int (*) __P((struct vop_unlock_args *)))vop_nounlock) 2796static int sync_print __P((struct vop_print_args *)); 2797#define sync_islocked ((int(*) __P((struct vop_islocked_args *)))vop_noislocked) 2798 2799static vop_t **sync_vnodeop_p; 2800static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 2801 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 2802 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 2803 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 2804 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 2805 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 2806 { &vop_lock_desc, (vop_t *) sync_lock }, /* lock */ 2807 { &vop_unlock_desc, (vop_t *) sync_unlock }, /* unlock */ 2808 { &vop_print_desc, (vop_t *) sync_print }, /* print */ 2809 { &vop_islocked_desc, (vop_t *) sync_islocked }, /* islocked */ 2810 { NULL, NULL } 2811}; 2812static struct vnodeopv_desc sync_vnodeop_opv_desc = 2813 { &sync_vnodeop_p, sync_vnodeop_entries }; 2814 2815VNODEOP_SET(sync_vnodeop_opv_desc); 2816 2817/* 2818 * Create a new filesystem syncer vnode for the specified mount point. 2819 */ 2820int 2821vfs_allocate_syncvnode(mp) 2822 struct mount *mp; 2823{ 2824 struct vnode *vp; 2825 static long start, incr, next; 2826 int error; 2827 2828 /* Allocate a new vnode */ 2829 if ((error = getnewvnode(VT_VFS, mp, sync_vnodeop_p, &vp)) != 0) { 2830 mp->mnt_syncer = NULL; 2831 return (error); 2832 } 2833 vp->v_type = VNON; 2834 /* 2835 * Place the vnode onto the syncer worklist. We attempt to 2836 * scatter them about on the list so that they will go off 2837 * at evenly distributed times even if all the filesystems 2838 * are mounted at once. 2839 */ 2840 next += incr; 2841 if (next == 0 || next > syncer_maxdelay) { 2842 start /= 2; 2843 incr /= 2; 2844 if (start == 0) { 2845 start = syncer_maxdelay / 2; 2846 incr = syncer_maxdelay; 2847 } 2848 next = start; 2849 } 2850 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 2851 mp->mnt_syncer = vp; 2852 return (0); 2853} 2854 2855/* 2856 * Do a lazy sync of the filesystem. 2857 */ 2858static int 2859sync_fsync(ap) 2860 struct vop_fsync_args /* { 2861 struct vnode *a_vp; 2862 struct ucred *a_cred; 2863 int a_waitfor; 2864 struct thread *a_td; 2865 } */ *ap; 2866{ 2867 struct vnode *syncvp = ap->a_vp; 2868 struct mount *mp = syncvp->v_mount; 2869 struct thread *td = ap->a_td; 2870 int asyncflag; 2871 2872 /* 2873 * We only need to do something if this is a lazy evaluation. 2874 */ 2875 if (ap->a_waitfor != MNT_LAZY) 2876 return (0); 2877 2878 /* 2879 * Move ourselves to the back of the sync list. 2880 */ 2881 vn_syncer_add_to_worklist(syncvp, syncdelay); 2882 2883 /* 2884 * Walk the list of vnodes pushing all that are dirty and 2885 * not already on the sync list. 2886 */ 2887 mtx_lock(&mountlist_mtx); 2888 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) { 2889 mtx_unlock(&mountlist_mtx); 2890 return (0); 2891 } 2892 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 2893 vfs_unbusy(mp, td); 2894 return (0); 2895 } 2896 asyncflag = mp->mnt_flag & MNT_ASYNC; 2897 mp->mnt_flag &= ~MNT_ASYNC; 2898 vfs_msync(mp, MNT_NOWAIT); 2899 VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td); 2900 if (asyncflag) 2901 mp->mnt_flag |= MNT_ASYNC; 2902 vn_finished_write(mp); 2903 vfs_unbusy(mp, td); 2904 return (0); 2905} 2906 2907/* 2908 * The syncer vnode is no referenced. 2909 */ 2910static int 2911sync_inactive(ap) 2912 struct vop_inactive_args /* { 2913 struct vnode *a_vp; 2914 struct thread *a_td; 2915 } */ *ap; 2916{ 2917 2918 vgone(ap->a_vp); 2919 return (0); 2920} 2921 2922/* 2923 * The syncer vnode is no longer needed and is being decommissioned. 2924 * 2925 * Modifications to the worklist must be protected at splbio(). 2926 */ 2927static int 2928sync_reclaim(ap) 2929 struct vop_reclaim_args /* { 2930 struct vnode *a_vp; 2931 } */ *ap; 2932{ 2933 struct vnode *vp = ap->a_vp; 2934 int s; 2935 2936 s = splbio(); 2937 vp->v_mount->mnt_syncer = NULL; 2938 if (vp->v_flag & VONWORKLST) { 2939 LIST_REMOVE(vp, v_synclist); 2940 vp->v_flag &= ~VONWORKLST; 2941 } 2942 splx(s); 2943 2944 return (0); 2945} 2946 2947/* 2948 * Print out a syncer vnode. 2949 */ 2950static int 2951sync_print(ap) 2952 struct vop_print_args /* { 2953 struct vnode *a_vp; 2954 } */ *ap; 2955{ 2956 struct vnode *vp = ap->a_vp; 2957 2958 printf("syncer vnode"); 2959 if (vp->v_vnlock != NULL) 2960 lockmgr_printinfo(vp->v_vnlock); 2961 printf("\n"); 2962 return (0); 2963} 2964 2965/* 2966 * extract the dev_t from a VCHR 2967 */ 2968dev_t 2969vn_todev(vp) 2970 struct vnode *vp; 2971{ 2972 if (vp->v_type != VCHR) 2973 return (NODEV); 2974 return (vp->v_rdev); 2975} 2976 2977/* 2978 * Check if vnode represents a disk device 2979 */ 2980int 2981vn_isdisk(vp, errp) 2982 struct vnode *vp; 2983 int *errp; 2984{ 2985 struct cdevsw *cdevsw; 2986 2987 if (vp->v_type != VCHR) { 2988 if (errp != NULL) 2989 *errp = ENOTBLK; 2990 return (0); 2991 } 2992 if (vp->v_rdev == NULL) { 2993 if (errp != NULL) 2994 *errp = ENXIO; 2995 return (0); 2996 } 2997 cdevsw = devsw(vp->v_rdev); 2998 if (cdevsw == NULL) { 2999 if (errp != NULL) 3000 *errp = ENXIO; 3001 return (0); 3002 } 3003 if (!(cdevsw->d_flags & D_DISK)) { 3004 if (errp != NULL) 3005 *errp = ENOTBLK; 3006 return (0); 3007 } 3008 if (errp != NULL) 3009 *errp = 0; 3010 return (1); 3011} 3012 3013/* 3014 * Free data allocated by namei(); see namei(9) for details. 3015 */ 3016void 3017NDFREE(ndp, flags) 3018 struct nameidata *ndp; 3019 const uint flags; 3020{ 3021 if (!(flags & NDF_NO_FREE_PNBUF) && 3022 (ndp->ni_cnd.cn_flags & HASBUF)) { 3023 zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 3024 ndp->ni_cnd.cn_flags &= ~HASBUF; 3025 } 3026 if (!(flags & NDF_NO_DVP_UNLOCK) && 3027 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 3028 ndp->ni_dvp != ndp->ni_vp) 3029 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread); 3030 if (!(flags & NDF_NO_DVP_RELE) && 3031 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 3032 vrele(ndp->ni_dvp); 3033 ndp->ni_dvp = NULL; 3034 } 3035 if (!(flags & NDF_NO_VP_UNLOCK) && 3036 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 3037 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread); 3038 if (!(flags & NDF_NO_VP_RELE) && 3039 ndp->ni_vp) { 3040 vrele(ndp->ni_vp); 3041 ndp->ni_vp = NULL; 3042 } 3043 if (!(flags & NDF_NO_STARTDIR_RELE) && 3044 (ndp->ni_cnd.cn_flags & SAVESTART)) { 3045 vrele(ndp->ni_startdir); 3046 ndp->ni_startdir = NULL; 3047 } 3048} 3049 3050/* 3051 * Common file system object access control check routine. Accepts a 3052 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3053 * and optional call-by-reference privused argument allowing vaccess() 3054 * to indicate to the caller whether privilege was used to satisfy the 3055 * request. Returns 0 on success, or an errno on failure. 3056 */ 3057int 3058vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused) 3059 enum vtype type; 3060 mode_t file_mode; 3061 uid_t file_uid; 3062 gid_t file_gid; 3063 mode_t acc_mode; 3064 struct ucred *cred; 3065 int *privused; 3066{ 3067 mode_t dac_granted; 3068#ifdef CAPABILITIES 3069 mode_t cap_granted; 3070#endif 3071 3072 /* 3073 * Look for a normal, non-privileged way to access the file/directory 3074 * as requested. If it exists, go with that. 3075 */ 3076 3077 if (privused != NULL) 3078 *privused = 0; 3079 3080 dac_granted = 0; 3081 3082 /* Check the owner. */ 3083 if (cred->cr_uid == file_uid) { 3084 dac_granted |= VADMIN; 3085 if (file_mode & S_IXUSR) 3086 dac_granted |= VEXEC; 3087 if (file_mode & S_IRUSR) 3088 dac_granted |= VREAD; 3089 if (file_mode & S_IWUSR) 3090 dac_granted |= VWRITE; 3091 3092 if ((acc_mode & dac_granted) == acc_mode) 3093 return (0); 3094 3095 goto privcheck; 3096 } 3097 3098 /* Otherwise, check the groups (first match) */ 3099 if (groupmember(file_gid, cred)) { 3100 if (file_mode & S_IXGRP) 3101 dac_granted |= VEXEC; 3102 if (file_mode & S_IRGRP) 3103 dac_granted |= VREAD; 3104 if (file_mode & S_IWGRP) 3105 dac_granted |= VWRITE; 3106 3107 if ((acc_mode & dac_granted) == acc_mode) 3108 return (0); 3109 3110 goto privcheck; 3111 } 3112 3113 /* Otherwise, check everyone else. */ 3114 if (file_mode & S_IXOTH) 3115 dac_granted |= VEXEC; 3116 if (file_mode & S_IROTH) 3117 dac_granted |= VREAD; 3118 if (file_mode & S_IWOTH) 3119 dac_granted |= VWRITE; 3120 if ((acc_mode & dac_granted) == acc_mode) 3121 return (0); 3122 3123privcheck: 3124 if (!suser_xxx(cred, NULL, PRISON_ROOT)) { 3125 /* XXX audit: privilege used */ 3126 if (privused != NULL) 3127 *privused = 1; 3128 return (0); 3129 } 3130 3131#ifdef CAPABILITIES 3132 /* 3133 * Build a capability mask to determine if the set of capabilities 3134 * satisfies the requirements when combined with the granted mask 3135 * from above. 3136 * For each capability, if the capability is required, bitwise 3137 * or the request type onto the cap_granted mask. 3138 */ 3139 cap_granted = 0; 3140 3141 if (type == VDIR) { 3142 /* 3143 * For directories, use CAP_DAC_READ_SEARCH to satisfy 3144 * VEXEC requests, instead of CAP_DAC_EXECUTE. 3145 */ 3146 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3147 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT)) 3148 cap_granted |= VEXEC; 3149 } else { 3150 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3151 !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT)) 3152 cap_granted |= VEXEC; 3153 } 3154 3155 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) && 3156 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT)) 3157 cap_granted |= VREAD; 3158 3159 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3160 !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT)) 3161 cap_granted |= VWRITE; 3162 3163 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3164 !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT)) 3165 cap_granted |= VADMIN; 3166 3167 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) { 3168 /* XXX audit: privilege used */ 3169 if (privused != NULL) 3170 *privused = 1; 3171 return (0); 3172 } 3173#endif 3174 3175 return ((acc_mode & VADMIN) ? EPERM : EACCES); 3176} 3177 3178