vfs_subr.c revision 131695
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 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 35 */ 36 37/* 38 * External virtual filesystem routines 39 */ 40 41#include <sys/cdefs.h> 42__FBSDID("$FreeBSD: head/sys/kern/vfs_subr.c 131695 2004-07-06 09:37:43Z alfred $"); 43 44#include "opt_ddb.h" 45#include "opt_mac.h" 46 47#include <sys/param.h> 48#include <sys/systm.h> 49#include <sys/bio.h> 50#include <sys/buf.h> 51#include <sys/conf.h> 52#include <sys/event.h> 53#include <sys/eventhandler.h> 54#include <sys/extattr.h> 55#include <sys/fcntl.h> 56#include <sys/kernel.h> 57#include <sys/kthread.h> 58#include <sys/mac.h> 59#include <sys/malloc.h> 60#include <sys/mount.h> 61#include <sys/namei.h> 62#include <sys/sleepqueue.h> 63#include <sys/stat.h> 64#include <sys/sysctl.h> 65#include <sys/syslog.h> 66#include <sys/vmmeter.h> 67#include <sys/vnode.h> 68 69#include <vm/vm.h> 70#include <vm/vm_object.h> 71#include <vm/vm_extern.h> 72#include <vm/pmap.h> 73#include <vm/vm_map.h> 74#include <vm/vm_page.h> 75#include <vm/vm_kern.h> 76#include <vm/uma.h> 77 78static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure"); 79 80static void addalias(struct vnode *vp, struct cdev *nvp_rdev); 81static void delmntque(struct vnode *vp); 82static void insmntque(struct vnode *vp, struct mount *mp); 83static void vclean(struct vnode *vp, int flags, struct thread *td); 84static void vlruvp(struct vnode *vp); 85static int flushbuflist(struct buf *blist, int flags, struct vnode *vp, 86 int slpflag, int slptimeo, int *errorp); 87static void syncer_shutdown(void *arg, int howto); 88static int vtryrecycle(struct vnode *vp); 89static void vx_lock(struct vnode *vp); 90static void vx_unlock(struct vnode *vp); 91static void vgonechrl(struct vnode *vp, struct thread *td); 92 93 94/* 95 * Number of vnodes in existence. Increased whenever getnewvnode() 96 * allocates a new vnode, never decreased. 97 */ 98static unsigned long numvnodes; 99 100SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 101 102/* 103 * Conversion tables for conversion from vnode types to inode formats 104 * and back. 105 */ 106enum vtype iftovt_tab[16] = { 107 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 108 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 109}; 110int vttoif_tab[9] = { 111 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 112 S_IFSOCK, S_IFIFO, S_IFMT, 113}; 114 115/* 116 * List of vnodes that are ready for recycling. 117 */ 118static TAILQ_HEAD(freelst, vnode) vnode_free_list; 119 120/* 121 * Minimum number of free vnodes. If there are fewer than this free vnodes, 122 * getnewvnode() will return a newly allocated vnode. 123 */ 124static u_long wantfreevnodes = 25; 125SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 126/* Number of vnodes in the free list. */ 127static u_long freevnodes; 128SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 129 130/* 131 * Various variables used for debugging the new implementation of 132 * reassignbuf(). 133 * XXX these are probably of (very) limited utility now. 134 */ 135static int reassignbufcalls; 136SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 137static int nameileafonly; 138SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, ""); 139 140/* 141 * Cache for the mount type id assigned to NFS. This is used for 142 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 143 */ 144int nfs_mount_type = -1; 145 146/* To keep more than one thread at a time from running vfs_getnewfsid */ 147static struct mtx mntid_mtx; 148 149/* 150 * Lock for any access to the following: 151 * vnode_free_list 152 * numvnodes 153 * freevnodes 154 */ 155static struct mtx vnode_free_list_mtx; 156 157/* 158 * For any iteration/modification of dev->si_hlist (linked through 159 * v_specnext) 160 */ 161static struct mtx spechash_mtx; 162 163/* Publicly exported FS */ 164struct nfs_public nfs_pub; 165 166/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 167static uma_zone_t vnode_zone; 168static uma_zone_t vnodepoll_zone; 169 170/* Set to 1 to print out reclaim of active vnodes */ 171int prtactive; 172 173/* 174 * The workitem queue. 175 * 176 * It is useful to delay writes of file data and filesystem metadata 177 * for tens of seconds so that quickly created and deleted files need 178 * not waste disk bandwidth being created and removed. To realize this, 179 * we append vnodes to a "workitem" queue. When running with a soft 180 * updates implementation, most pending metadata dependencies should 181 * not wait for more than a few seconds. Thus, mounted on block devices 182 * are delayed only about a half the time that file data is delayed. 183 * Similarly, directory updates are more critical, so are only delayed 184 * about a third the time that file data is delayed. Thus, there are 185 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 186 * one each second (driven off the filesystem syncer process). The 187 * syncer_delayno variable indicates the next queue that is to be processed. 188 * Items that need to be processed soon are placed in this queue: 189 * 190 * syncer_workitem_pending[syncer_delayno] 191 * 192 * A delay of fifteen seconds is done by placing the request fifteen 193 * entries later in the queue: 194 * 195 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 196 * 197 */ 198static int syncer_delayno; 199static long syncer_mask; 200LIST_HEAD(synclist, vnode); 201static struct synclist *syncer_workitem_pending; 202/* 203 * The sync_mtx protects: 204 * vp->v_synclist 205 * sync_vnode_count 206 * syncer_delayno 207 * syncer_state 208 * syncer_workitem_pending 209 * syncer_worklist_len 210 * rushjob 211 */ 212static struct mtx sync_mtx; 213 214#define SYNCER_MAXDELAY 32 215static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 216static int syncdelay = 30; /* max time to delay syncing data */ 217static int filedelay = 30; /* time to delay syncing files */ 218SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 219static int dirdelay = 29; /* time to delay syncing directories */ 220SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 221static int metadelay = 28; /* time to delay syncing metadata */ 222SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 223static int rushjob; /* number of slots to run ASAP */ 224static int stat_rush_requests; /* number of times I/O speeded up */ 225SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 226 227/* 228 * When shutting down the syncer, run it at four times normal speed. 229 */ 230#define SYNCER_SHUTDOWN_SPEEDUP 4 231static int sync_vnode_count; 232static int syncer_worklist_len; 233static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 234 syncer_state; 235 236/* 237 * Number of vnodes we want to exist at any one time. This is mostly used 238 * to size hash tables in vnode-related code. It is normally not used in 239 * getnewvnode(), as wantfreevnodes is normally nonzero.) 240 * 241 * XXX desiredvnodes is historical cruft and should not exist. 242 */ 243int desiredvnodes; 244SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 245 &desiredvnodes, 0, "Maximum number of vnodes"); 246static int minvnodes; 247SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 248 &minvnodes, 0, "Minimum number of vnodes"); 249static int vnlru_nowhere; 250SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 251 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 252 253/* Hook for calling soft updates. */ 254int (*softdep_process_worklist_hook)(struct mount *); 255 256/* 257 * Initialize the vnode management data structures. 258 */ 259static void 260vntblinit(void *dummy __unused) 261{ 262 263 /* 264 * Desiredvnodes is a function of the physical memory size and 265 * the kernel's heap size. Specifically, desiredvnodes scales 266 * in proportion to the physical memory size until two fifths 267 * of the kernel's heap size is consumed by vnodes and vm 268 * objects. 269 */ 270 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size / 271 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 272 minvnodes = desiredvnodes / 4; 273 mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF); 274 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 275 mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF); 276 TAILQ_INIT(&vnode_free_list); 277 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 278 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 279 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 280 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 281 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 282 /* 283 * Initialize the filesystem syncer. 284 */ 285 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 286 &syncer_mask); 287 syncer_maxdelay = syncer_mask + 1; 288 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 289} 290SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL) 291 292 293/* 294 * Mark a mount point as busy. Used to synchronize access and to delay 295 * unmounting. Interlock is not released on failure. 296 */ 297int 298vfs_busy(mp, flags, interlkp, td) 299 struct mount *mp; 300 int flags; 301 struct mtx *interlkp; 302 struct thread *td; 303{ 304 int lkflags; 305 306 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 307 if (flags & LK_NOWAIT) 308 return (ENOENT); 309 mp->mnt_kern_flag |= MNTK_MWAIT; 310 /* 311 * Since all busy locks are shared except the exclusive 312 * lock granted when unmounting, the only place that a 313 * wakeup needs to be done is at the release of the 314 * exclusive lock at the end of dounmount. 315 */ 316 msleep(mp, interlkp, PVFS, "vfs_busy", 0); 317 return (ENOENT); 318 } 319 lkflags = LK_SHARED | LK_NOPAUSE; 320 if (interlkp) 321 lkflags |= LK_INTERLOCK; 322 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td)) 323 panic("vfs_busy: unexpected lock failure"); 324 return (0); 325} 326 327/* 328 * Free a busy filesystem. 329 */ 330void 331vfs_unbusy(mp, td) 332 struct mount *mp; 333 struct thread *td; 334{ 335 336 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 337} 338 339/* 340 * Lookup a mount point by filesystem identifier. 341 */ 342struct mount * 343vfs_getvfs(fsid) 344 fsid_t *fsid; 345{ 346 register struct mount *mp; 347 348 mtx_lock(&mountlist_mtx); 349 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 350 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 351 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 352 mtx_unlock(&mountlist_mtx); 353 return (mp); 354 } 355 } 356 mtx_unlock(&mountlist_mtx); 357 return ((struct mount *) 0); 358} 359 360/* 361 * Check if a user can access priveledged mount options. 362 */ 363int 364vfs_suser(struct mount *mp, struct thread *td) 365{ 366 int error; 367 368 if ((mp->mnt_flag & MNT_USER) == 0 || 369 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 370 if ((error = suser(td)) != 0) 371 return (error); 372 } 373 return (0); 374} 375 376/* 377 * Get a new unique fsid. Try to make its val[0] unique, since this value 378 * will be used to create fake device numbers for stat(). Also try (but 379 * not so hard) make its val[0] unique mod 2^16, since some emulators only 380 * support 16-bit device numbers. We end up with unique val[0]'s for the 381 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 382 * 383 * Keep in mind that several mounts may be running in parallel. Starting 384 * the search one past where the previous search terminated is both a 385 * micro-optimization and a defense against returning the same fsid to 386 * different mounts. 387 */ 388void 389vfs_getnewfsid(mp) 390 struct mount *mp; 391{ 392 static u_int16_t mntid_base; 393 fsid_t tfsid; 394 int mtype; 395 396 mtx_lock(&mntid_mtx); 397 mtype = mp->mnt_vfc->vfc_typenum; 398 tfsid.val[1] = mtype; 399 mtype = (mtype & 0xFF) << 24; 400 for (;;) { 401 tfsid.val[0] = makedev(255, 402 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 403 mntid_base++; 404 if (vfs_getvfs(&tfsid) == NULL) 405 break; 406 } 407 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 408 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 409 mtx_unlock(&mntid_mtx); 410} 411 412/* 413 * Knob to control the precision of file timestamps: 414 * 415 * 0 = seconds only; nanoseconds zeroed. 416 * 1 = seconds and nanoseconds, accurate within 1/HZ. 417 * 2 = seconds and nanoseconds, truncated to microseconds. 418 * >=3 = seconds and nanoseconds, maximum precision. 419 */ 420enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 421 422static int timestamp_precision = TSP_SEC; 423SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 424 ×tamp_precision, 0, ""); 425 426/* 427 * Get a current timestamp. 428 */ 429void 430vfs_timestamp(tsp) 431 struct timespec *tsp; 432{ 433 struct timeval tv; 434 435 switch (timestamp_precision) { 436 case TSP_SEC: 437 tsp->tv_sec = time_second; 438 tsp->tv_nsec = 0; 439 break; 440 case TSP_HZ: 441 getnanotime(tsp); 442 break; 443 case TSP_USEC: 444 microtime(&tv); 445 TIMEVAL_TO_TIMESPEC(&tv, tsp); 446 break; 447 case TSP_NSEC: 448 default: 449 nanotime(tsp); 450 break; 451 } 452} 453 454/* 455 * Set vnode attributes to VNOVAL 456 */ 457void 458vattr_null(vap) 459 register struct vattr *vap; 460{ 461 462 vap->va_type = VNON; 463 vap->va_size = VNOVAL; 464 vap->va_bytes = VNOVAL; 465 vap->va_mode = VNOVAL; 466 vap->va_nlink = VNOVAL; 467 vap->va_uid = VNOVAL; 468 vap->va_gid = VNOVAL; 469 vap->va_fsid = VNOVAL; 470 vap->va_fileid = VNOVAL; 471 vap->va_blocksize = VNOVAL; 472 vap->va_rdev = VNOVAL; 473 vap->va_atime.tv_sec = VNOVAL; 474 vap->va_atime.tv_nsec = VNOVAL; 475 vap->va_mtime.tv_sec = VNOVAL; 476 vap->va_mtime.tv_nsec = VNOVAL; 477 vap->va_ctime.tv_sec = VNOVAL; 478 vap->va_ctime.tv_nsec = VNOVAL; 479 vap->va_birthtime.tv_sec = VNOVAL; 480 vap->va_birthtime.tv_nsec = VNOVAL; 481 vap->va_flags = VNOVAL; 482 vap->va_gen = VNOVAL; 483 vap->va_vaflags = 0; 484} 485 486/* 487 * This routine is called when we have too many vnodes. It attempts 488 * to free <count> vnodes and will potentially free vnodes that still 489 * have VM backing store (VM backing store is typically the cause 490 * of a vnode blowout so we want to do this). Therefore, this operation 491 * is not considered cheap. 492 * 493 * A number of conditions may prevent a vnode from being reclaimed. 494 * the buffer cache may have references on the vnode, a directory 495 * vnode may still have references due to the namei cache representing 496 * underlying files, or the vnode may be in active use. It is not 497 * desireable to reuse such vnodes. These conditions may cause the 498 * number of vnodes to reach some minimum value regardless of what 499 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 500 */ 501static int 502vlrureclaim(struct mount *mp) 503{ 504 struct vnode *vp; 505 int done; 506 int trigger; 507 int usevnodes; 508 int count; 509 510 /* 511 * Calculate the trigger point, don't allow user 512 * screwups to blow us up. This prevents us from 513 * recycling vnodes with lots of resident pages. We 514 * aren't trying to free memory, we are trying to 515 * free vnodes. 516 */ 517 usevnodes = desiredvnodes; 518 if (usevnodes <= 0) 519 usevnodes = 1; 520 trigger = cnt.v_page_count * 2 / usevnodes; 521 522 done = 0; 523 MNT_ILOCK(mp); 524 count = mp->mnt_nvnodelistsize / 10 + 1; 525 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) { 526 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 527 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 528 529 if (vp->v_type != VNON && 530 vp->v_type != VBAD && 531 VI_TRYLOCK(vp)) { 532 if (VMIGHTFREE(vp) && /* critical path opt */ 533 (vp->v_object == NULL || 534 vp->v_object->resident_page_count < trigger)) { 535 MNT_IUNLOCK(mp); 536 vgonel(vp, curthread); 537 done++; 538 MNT_ILOCK(mp); 539 } else 540 VI_UNLOCK(vp); 541 } 542 --count; 543 } 544 MNT_IUNLOCK(mp); 545 return done; 546} 547 548/* 549 * Attempt to recycle vnodes in a context that is always safe to block. 550 * Calling vlrurecycle() from the bowels of filesystem code has some 551 * interesting deadlock problems. 552 */ 553static struct proc *vnlruproc; 554static int vnlruproc_sig; 555 556static void 557vnlru_proc(void) 558{ 559 struct mount *mp, *nmp; 560 int done; 561 struct proc *p = vnlruproc; 562 struct thread *td = FIRST_THREAD_IN_PROC(p); 563 564 mtx_lock(&Giant); 565 566 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 567 SHUTDOWN_PRI_FIRST); 568 569 for (;;) { 570 kthread_suspend_check(p); 571 mtx_lock(&vnode_free_list_mtx); 572 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) { 573 mtx_unlock(&vnode_free_list_mtx); 574 vnlruproc_sig = 0; 575 wakeup(&vnlruproc_sig); 576 tsleep(vnlruproc, PVFS, "vlruwt", hz); 577 continue; 578 } 579 mtx_unlock(&vnode_free_list_mtx); 580 done = 0; 581 mtx_lock(&mountlist_mtx); 582 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 583 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 584 nmp = TAILQ_NEXT(mp, mnt_list); 585 continue; 586 } 587 done += vlrureclaim(mp); 588 mtx_lock(&mountlist_mtx); 589 nmp = TAILQ_NEXT(mp, mnt_list); 590 vfs_unbusy(mp, td); 591 } 592 mtx_unlock(&mountlist_mtx); 593 if (done == 0) { 594#if 0 595 /* These messages are temporary debugging aids */ 596 if (vnlru_nowhere < 5) 597 printf("vnlru process getting nowhere..\n"); 598 else if (vnlru_nowhere == 5) 599 printf("vnlru process messages stopped.\n"); 600#endif 601 vnlru_nowhere++; 602 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 603 } 604 } 605} 606 607static struct kproc_desc vnlru_kp = { 608 "vnlru", 609 vnlru_proc, 610 &vnlruproc 611}; 612SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 613 614 615/* 616 * Routines having to do with the management of the vnode table. 617 */ 618 619/* 620 * Check to see if a free vnode can be recycled. If it can, 621 * recycle it and return it with the vnode interlock held. 622 */ 623static int 624vtryrecycle(struct vnode *vp) 625{ 626 struct thread *td = curthread; 627 vm_object_t object; 628 struct mount *vnmp; 629 int error; 630 631 /* Don't recycle if we can't get the interlock */ 632 if (!VI_TRYLOCK(vp)) 633 return (EWOULDBLOCK); 634 /* 635 * This vnode may found and locked via some other list, if so we 636 * can't recycle it yet. 637 */ 638 if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT, td) != 0) 639 return (EWOULDBLOCK); 640 /* 641 * Don't recycle if its filesystem is being suspended. 642 */ 643 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 644 VOP_UNLOCK(vp, 0, td); 645 return (EBUSY); 646 } 647 648 /* 649 * Don't recycle if we still have cached pages. 650 */ 651 if (VOP_GETVOBJECT(vp, &object) == 0) { 652 VM_OBJECT_LOCK(object); 653 if (object->resident_page_count || 654 object->ref_count) { 655 VM_OBJECT_UNLOCK(object); 656 error = EBUSY; 657 goto done; 658 } 659 VM_OBJECT_UNLOCK(object); 660 } 661 if (LIST_FIRST(&vp->v_cache_src)) { 662 /* 663 * note: nameileafonly sysctl is temporary, 664 * for debugging only, and will eventually be 665 * removed. 666 */ 667 if (nameileafonly > 0) { 668 /* 669 * Do not reuse namei-cached directory 670 * vnodes that have cached 671 * subdirectories. 672 */ 673 if (cache_leaf_test(vp) < 0) { 674 error = EISDIR; 675 goto done; 676 } 677 } else if (nameileafonly < 0 || 678 vmiodirenable == 0) { 679 /* 680 * Do not reuse namei-cached directory 681 * vnodes if nameileafonly is -1 or 682 * if VMIO backing for directories is 683 * turned off (otherwise we reuse them 684 * too quickly). 685 */ 686 error = EBUSY; 687 goto done; 688 } 689 } 690 /* 691 * If we got this far, we need to acquire the interlock and see if 692 * anyone picked up this vnode from another list. If not, we will 693 * mark it with XLOCK via vgonel() so that anyone who does find it 694 * will skip over it. 695 */ 696 VI_LOCK(vp); 697 if (VSHOULDBUSY(vp) && (vp->v_iflag & VI_XLOCK) == 0) { 698 VI_UNLOCK(vp); 699 error = EBUSY; 700 goto done; 701 } 702 mtx_lock(&vnode_free_list_mtx); 703 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 704 vp->v_iflag &= ~VI_FREE; 705 mtx_unlock(&vnode_free_list_mtx); 706 vp->v_iflag |= VI_DOOMED; 707 if (vp->v_type != VBAD) { 708 VOP_UNLOCK(vp, 0, td); 709 vgonel(vp, td); 710 VI_LOCK(vp); 711 } else 712 VOP_UNLOCK(vp, 0, td); 713 vn_finished_write(vnmp); 714 return (0); 715done: 716 VOP_UNLOCK(vp, 0, td); 717 vn_finished_write(vnmp); 718 return (error); 719} 720 721/* 722 * Return the next vnode from the free list. 723 */ 724int 725getnewvnode(tag, mp, vops, vpp) 726 const char *tag; 727 struct mount *mp; 728 vop_t **vops; 729 struct vnode **vpp; 730{ 731 struct vnode *vp = NULL; 732 struct vpollinfo *pollinfo = NULL; 733 734 mtx_lock(&vnode_free_list_mtx); 735 736 /* 737 * Try to reuse vnodes if we hit the max. This situation only 738 * occurs in certain large-memory (2G+) situations. We cannot 739 * attempt to directly reclaim vnodes due to nasty recursion 740 * problems. 741 */ 742 while (numvnodes - freevnodes > desiredvnodes) { 743 if (vnlruproc_sig == 0) { 744 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 745 wakeup(vnlruproc); 746 } 747 mtx_unlock(&vnode_free_list_mtx); 748 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz); 749 mtx_lock(&vnode_free_list_mtx); 750 } 751 752 /* 753 * Attempt to reuse a vnode already on the free list, allocating 754 * a new vnode if we can't find one or if we have not reached a 755 * good minimum for good LRU performance. 756 */ 757 758 if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) { 759 int error; 760 int count; 761 762 for (count = 0; count < freevnodes; count++) { 763 vp = TAILQ_FIRST(&vnode_free_list); 764 765 KASSERT(vp->v_usecount == 0 && 766 (vp->v_iflag & VI_DOINGINACT) == 0, 767 ("getnewvnode: free vnode isn't")); 768 769 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 770 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 771 mtx_unlock(&vnode_free_list_mtx); 772 error = vtryrecycle(vp); 773 mtx_lock(&vnode_free_list_mtx); 774 if (error == 0) 775 break; 776 vp = NULL; 777 } 778 } 779 if (vp) { 780 freevnodes--; 781 mtx_unlock(&vnode_free_list_mtx); 782 783#ifdef INVARIANTS 784 { 785 if (vp->v_data) 786 panic("cleaned vnode isn't"); 787 if (vp->v_numoutput) 788 panic("Clean vnode has pending I/O's"); 789 if (vp->v_writecount != 0) 790 panic("Non-zero write count"); 791 } 792#endif 793 if ((pollinfo = vp->v_pollinfo) != NULL) { 794 /* 795 * To avoid lock order reversals, the call to 796 * uma_zfree() must be delayed until the vnode 797 * interlock is released. 798 */ 799 vp->v_pollinfo = NULL; 800 } 801#ifdef MAC 802 mac_destroy_vnode(vp); 803#endif 804 vp->v_iflag = 0; 805 vp->v_vflag = 0; 806 vp->v_lastw = 0; 807 vp->v_lasta = 0; 808 vp->v_cstart = 0; 809 vp->v_clen = 0; 810 vp->v_socket = 0; 811 lockdestroy(vp->v_vnlock); 812 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE); 813 KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0")); 814 KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL")); 815 KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0")); 816 KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL")); 817 } else { 818 numvnodes++; 819 mtx_unlock(&vnode_free_list_mtx); 820 821 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 822 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 823 VI_LOCK(vp); 824 vp->v_dd = vp; 825 vp->v_vnlock = &vp->v_lock; 826 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE); 827 cache_purge(vp); /* Sets up v_id. */ 828 LIST_INIT(&vp->v_cache_src); 829 TAILQ_INIT(&vp->v_cache_dst); 830 } 831 832 TAILQ_INIT(&vp->v_cleanblkhd); 833 TAILQ_INIT(&vp->v_dirtyblkhd); 834 vp->v_type = VNON; 835 vp->v_tag = tag; 836 vp->v_op = vops; 837 *vpp = vp; 838 vp->v_usecount = 1; 839 vp->v_data = 0; 840 vp->v_cachedid = -1; 841 VI_UNLOCK(vp); 842 if (pollinfo != NULL) { 843 mtx_destroy(&pollinfo->vpi_lock); 844 uma_zfree(vnodepoll_zone, pollinfo); 845 } 846#ifdef MAC 847 mac_init_vnode(vp); 848 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 849 mac_associate_vnode_singlelabel(mp, vp); 850#endif 851 delmntque(vp); 852 if (mp != NULL) { 853 insmntque(vp, mp); 854 vp->v_bsize = mp->mnt_stat.f_iosize; 855 } 856 857 return (0); 858} 859 860/* 861 * Delete from old mount point vnode list, if on one. 862 */ 863static void 864delmntque(struct vnode *vp) 865{ 866 struct mount *mp; 867 868 if (vp->v_mount == NULL) 869 return; 870 mp = vp->v_mount; 871 MNT_ILOCK(mp); 872 vp->v_mount = NULL; 873 KASSERT(mp->mnt_nvnodelistsize > 0, 874 ("bad mount point vnode list size")); 875 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 876 mp->mnt_nvnodelistsize--; 877 MNT_IUNLOCK(mp); 878} 879 880/* 881 * Insert into list of vnodes for the new mount point, if available. 882 */ 883static void 884insmntque(struct vnode *vp, struct mount *mp) 885{ 886 887 vp->v_mount = mp; 888 KASSERT(mp != NULL, ("Don't call insmntque(foo, NULL)")); 889 MNT_ILOCK(vp->v_mount); 890 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 891 mp->mnt_nvnodelistsize++; 892 MNT_IUNLOCK(vp->v_mount); 893} 894 895/* 896 * Update outstanding I/O count and do wakeup if requested. 897 */ 898void 899vwakeup(bp) 900 register struct buf *bp; 901{ 902 register struct vnode *vp; 903 904 bp->b_flags &= ~B_WRITEINPROG; 905 if ((vp = bp->b_vp)) { 906 VI_LOCK(vp); 907 vp->v_numoutput--; 908 if (vp->v_numoutput < 0) 909 panic("vwakeup: neg numoutput"); 910 if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) { 911 vp->v_iflag &= ~VI_BWAIT; 912 wakeup(&vp->v_numoutput); 913 } 914 VI_UNLOCK(vp); 915 } 916} 917 918/* 919 * Flush out and invalidate all buffers associated with a vnode. 920 * Called with the underlying object locked. 921 */ 922int 923vinvalbuf(vp, flags, cred, td, slpflag, slptimeo) 924 struct vnode *vp; 925 int flags; 926 struct ucred *cred; 927 struct thread *td; 928 int slpflag, slptimeo; 929{ 930 struct buf *blist; 931 int error; 932 vm_object_t object; 933 934 GIANT_REQUIRED; 935 936 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 937 938 VI_LOCK(vp); 939 if (flags & V_SAVE) { 940 while (vp->v_numoutput) { 941 vp->v_iflag |= VI_BWAIT; 942 error = msleep(&vp->v_numoutput, VI_MTX(vp), 943 slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo); 944 if (error) { 945 VI_UNLOCK(vp); 946 return (error); 947 } 948 } 949 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 950 VI_UNLOCK(vp); 951 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0) 952 return (error); 953 /* 954 * XXX We could save a lock/unlock if this was only 955 * enabled under INVARIANTS 956 */ 957 VI_LOCK(vp); 958 if (vp->v_numoutput > 0 || 959 !TAILQ_EMPTY(&vp->v_dirtyblkhd)) 960 panic("vinvalbuf: dirty bufs"); 961 } 962 } 963 /* 964 * If you alter this loop please notice that interlock is dropped and 965 * reacquired in flushbuflist. Special care is needed to ensure that 966 * no race conditions occur from this. 967 */ 968 for (error = 0;;) { 969 if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 && 970 flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) { 971 if (error) 972 break; 973 continue; 974 } 975 if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 && 976 flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) { 977 if (error) 978 break; 979 continue; 980 } 981 break; 982 } 983 if (error) { 984 VI_UNLOCK(vp); 985 return (error); 986 } 987 988 /* 989 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 990 * have write I/O in-progress but if there is a VM object then the 991 * VM object can also have read-I/O in-progress. 992 */ 993 do { 994 while (vp->v_numoutput > 0) { 995 vp->v_iflag |= VI_BWAIT; 996 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0); 997 } 998 VI_UNLOCK(vp); 999 if (VOP_GETVOBJECT(vp, &object) == 0) { 1000 VM_OBJECT_LOCK(object); 1001 vm_object_pip_wait(object, "vnvlbx"); 1002 VM_OBJECT_UNLOCK(object); 1003 } 1004 VI_LOCK(vp); 1005 } while (vp->v_numoutput > 0); 1006 VI_UNLOCK(vp); 1007 1008 /* 1009 * Destroy the copy in the VM cache, too. 1010 */ 1011 if (VOP_GETVOBJECT(vp, &object) == 0) { 1012 VM_OBJECT_LOCK(object); 1013 vm_object_page_remove(object, 0, 0, 1014 (flags & V_SAVE) ? TRUE : FALSE); 1015 VM_OBJECT_UNLOCK(object); 1016 } 1017 1018#ifdef INVARIANTS 1019 VI_LOCK(vp); 1020 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1021 (!TAILQ_EMPTY(&vp->v_dirtyblkhd) || 1022 !TAILQ_EMPTY(&vp->v_cleanblkhd))) 1023 panic("vinvalbuf: flush failed"); 1024 VI_UNLOCK(vp); 1025#endif 1026 return (0); 1027} 1028 1029/* 1030 * Flush out buffers on the specified list. 1031 * 1032 */ 1033static int 1034flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp) 1035 struct buf *blist; 1036 int flags; 1037 struct vnode *vp; 1038 int slpflag, slptimeo; 1039 int *errorp; 1040{ 1041 struct buf *bp, *nbp; 1042 int found, error; 1043 1044 ASSERT_VI_LOCKED(vp, "flushbuflist"); 1045 1046 for (found = 0, bp = blist; bp; bp = nbp) { 1047 nbp = TAILQ_NEXT(bp, b_vnbufs); 1048 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1049 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1050 continue; 1051 } 1052 found += 1; 1053 error = BUF_TIMELOCK(bp, 1054 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp), 1055 "flushbuf", slpflag, slptimeo); 1056 if (error) { 1057 if (error != ENOLCK) 1058 *errorp = error; 1059 goto done; 1060 } 1061 /* 1062 * XXX Since there are no node locks for NFS, I 1063 * believe there is a slight chance that a delayed 1064 * write will occur while sleeping just above, so 1065 * check for it. Note that vfs_bio_awrite expects 1066 * buffers to reside on a queue, while bwrite and 1067 * brelse do not. 1068 */ 1069 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1070 (flags & V_SAVE)) { 1071 1072 if (bp->b_vp == vp) { 1073 if (bp->b_flags & B_CLUSTEROK) { 1074 vfs_bio_awrite(bp); 1075 } else { 1076 bremfree(bp); 1077 bp->b_flags |= B_ASYNC; 1078 bwrite(bp); 1079 } 1080 } else { 1081 bremfree(bp); 1082 (void) bwrite(bp); 1083 } 1084 goto done; 1085 } 1086 bremfree(bp); 1087 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF); 1088 bp->b_flags &= ~B_ASYNC; 1089 brelse(bp); 1090 VI_LOCK(vp); 1091 } 1092 return (found); 1093done: 1094 VI_LOCK(vp); 1095 return (found); 1096} 1097 1098/* 1099 * Truncate a file's buffer and pages to a specified length. This 1100 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1101 * sync activity. 1102 */ 1103int 1104vtruncbuf(vp, cred, td, length, blksize) 1105 register struct vnode *vp; 1106 struct ucred *cred; 1107 struct thread *td; 1108 off_t length; 1109 int blksize; 1110{ 1111 register struct buf *bp; 1112 struct buf *nbp; 1113 int anyfreed; 1114 int trunclbn; 1115 1116 /* 1117 * Round up to the *next* lbn. 1118 */ 1119 trunclbn = (length + blksize - 1) / blksize; 1120 1121 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1122restart: 1123 VI_LOCK(vp); 1124 anyfreed = 1; 1125 for (;anyfreed;) { 1126 anyfreed = 0; 1127 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 1128 nbp = TAILQ_NEXT(bp, b_vnbufs); 1129 if (bp->b_lblkno >= trunclbn) { 1130 if (BUF_LOCK(bp, 1131 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1132 VI_MTX(vp)) == ENOLCK) 1133 goto restart; 1134 1135 bremfree(bp); 1136 bp->b_flags |= (B_INVAL | B_RELBUF); 1137 bp->b_flags &= ~B_ASYNC; 1138 brelse(bp); 1139 anyfreed = 1; 1140 1141 if (nbp && 1142 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1143 (nbp->b_vp != vp) || 1144 (nbp->b_flags & B_DELWRI))) { 1145 goto restart; 1146 } 1147 VI_LOCK(vp); 1148 } 1149 } 1150 1151 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1152 nbp = TAILQ_NEXT(bp, b_vnbufs); 1153 if (bp->b_lblkno >= trunclbn) { 1154 if (BUF_LOCK(bp, 1155 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1156 VI_MTX(vp)) == ENOLCK) 1157 goto restart; 1158 bremfree(bp); 1159 bp->b_flags |= (B_INVAL | B_RELBUF); 1160 bp->b_flags &= ~B_ASYNC; 1161 brelse(bp); 1162 anyfreed = 1; 1163 if (nbp && 1164 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1165 (nbp->b_vp != vp) || 1166 (nbp->b_flags & B_DELWRI) == 0)) { 1167 goto restart; 1168 } 1169 VI_LOCK(vp); 1170 } 1171 } 1172 } 1173 1174 if (length > 0) { 1175restartsync: 1176 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 1177 nbp = TAILQ_NEXT(bp, b_vnbufs); 1178 if (bp->b_lblkno > 0) 1179 continue; 1180 /* 1181 * Since we hold the vnode lock this should only 1182 * fail if we're racing with the buf daemon. 1183 */ 1184 if (BUF_LOCK(bp, 1185 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1186 VI_MTX(vp)) == ENOLCK) { 1187 goto restart; 1188 } 1189 KASSERT((bp->b_flags & B_DELWRI), 1190 ("buf(%p) on dirty queue without DELWRI", bp)); 1191 1192 bremfree(bp); 1193 bawrite(bp); 1194 VI_LOCK(vp); 1195 goto restartsync; 1196 } 1197 } 1198 1199 while (vp->v_numoutput > 0) { 1200 vp->v_iflag |= VI_BWAIT; 1201 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0); 1202 } 1203 VI_UNLOCK(vp); 1204 vnode_pager_setsize(vp, length); 1205 1206 return (0); 1207} 1208 1209/* 1210 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1211 * a vnode. 1212 * 1213 * NOTE: We have to deal with the special case of a background bitmap 1214 * buffer, a situation where two buffers will have the same logical 1215 * block offset. We want (1) only the foreground buffer to be accessed 1216 * in a lookup and (2) must differentiate between the foreground and 1217 * background buffer in the splay tree algorithm because the splay 1218 * tree cannot normally handle multiple entities with the same 'index'. 1219 * We accomplish this by adding differentiating flags to the splay tree's 1220 * numerical domain. 1221 */ 1222static 1223struct buf * 1224buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1225{ 1226 struct buf dummy; 1227 struct buf *lefttreemax, *righttreemin, *y; 1228 1229 if (root == NULL) 1230 return (NULL); 1231 lefttreemax = righttreemin = &dummy; 1232 for (;;) { 1233 if (lblkno < root->b_lblkno || 1234 (lblkno == root->b_lblkno && 1235 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1236 if ((y = root->b_left) == NULL) 1237 break; 1238 if (lblkno < y->b_lblkno) { 1239 /* Rotate right. */ 1240 root->b_left = y->b_right; 1241 y->b_right = root; 1242 root = y; 1243 if ((y = root->b_left) == NULL) 1244 break; 1245 } 1246 /* Link into the new root's right tree. */ 1247 righttreemin->b_left = root; 1248 righttreemin = root; 1249 } else if (lblkno > root->b_lblkno || 1250 (lblkno == root->b_lblkno && 1251 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1252 if ((y = root->b_right) == NULL) 1253 break; 1254 if (lblkno > y->b_lblkno) { 1255 /* Rotate left. */ 1256 root->b_right = y->b_left; 1257 y->b_left = root; 1258 root = y; 1259 if ((y = root->b_right) == NULL) 1260 break; 1261 } 1262 /* Link into the new root's left tree. */ 1263 lefttreemax->b_right = root; 1264 lefttreemax = root; 1265 } else { 1266 break; 1267 } 1268 root = y; 1269 } 1270 /* Assemble the new root. */ 1271 lefttreemax->b_right = root->b_left; 1272 righttreemin->b_left = root->b_right; 1273 root->b_left = dummy.b_right; 1274 root->b_right = dummy.b_left; 1275 return (root); 1276} 1277 1278static 1279void 1280buf_vlist_remove(struct buf *bp) 1281{ 1282 struct vnode *vp = bp->b_vp; 1283 struct buf *root; 1284 1285 ASSERT_VI_LOCKED(vp, "buf_vlist_remove"); 1286 if (bp->b_xflags & BX_VNDIRTY) { 1287 if (bp != vp->v_dirtyblkroot) { 1288 root = buf_splay(bp->b_lblkno, bp->b_xflags, 1289 vp->v_dirtyblkroot); 1290 KASSERT(root == bp, 1291 ("splay lookup failed during dirty remove")); 1292 } 1293 if (bp->b_left == NULL) { 1294 root = bp->b_right; 1295 } else { 1296 root = buf_splay(bp->b_lblkno, bp->b_xflags, 1297 bp->b_left); 1298 root->b_right = bp->b_right; 1299 } 1300 vp->v_dirtyblkroot = root; 1301 TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs); 1302 vp->v_dirtybufcnt--; 1303 } else { 1304 /* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */ 1305 if (bp != vp->v_cleanblkroot) { 1306 root = buf_splay(bp->b_lblkno, bp->b_xflags, 1307 vp->v_cleanblkroot); 1308 KASSERT(root == bp, 1309 ("splay lookup failed during clean remove")); 1310 } 1311 if (bp->b_left == NULL) { 1312 root = bp->b_right; 1313 } else { 1314 root = buf_splay(bp->b_lblkno, bp->b_xflags, 1315 bp->b_left); 1316 root->b_right = bp->b_right; 1317 } 1318 vp->v_cleanblkroot = root; 1319 TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs); 1320 vp->v_cleanbufcnt--; 1321 } 1322 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1323} 1324 1325/* 1326 * Add the buffer to the sorted clean or dirty block list using a 1327 * splay tree algorithm. 1328 * 1329 * NOTE: xflags is passed as a constant, optimizing this inline function! 1330 */ 1331static 1332void 1333buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags) 1334{ 1335 struct buf *root; 1336 1337 ASSERT_VI_LOCKED(vp, "buf_vlist_add"); 1338 bp->b_xflags |= xflags; 1339 if (xflags & BX_VNDIRTY) { 1340 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot); 1341 if (root == NULL) { 1342 bp->b_left = NULL; 1343 bp->b_right = NULL; 1344 TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs); 1345 } else if (bp->b_lblkno < root->b_lblkno || 1346 (bp->b_lblkno == root->b_lblkno && 1347 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1348 bp->b_left = root->b_left; 1349 bp->b_right = root; 1350 root->b_left = NULL; 1351 TAILQ_INSERT_BEFORE(root, bp, b_vnbufs); 1352 } else { 1353 bp->b_right = root->b_right; 1354 bp->b_left = root; 1355 root->b_right = NULL; 1356 TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd, 1357 root, bp, b_vnbufs); 1358 } 1359 vp->v_dirtybufcnt++; 1360 vp->v_dirtyblkroot = bp; 1361 } else { 1362 /* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */ 1363 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot); 1364 if (root == NULL) { 1365 bp->b_left = NULL; 1366 bp->b_right = NULL; 1367 TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs); 1368 } else if (bp->b_lblkno < root->b_lblkno || 1369 (bp->b_lblkno == root->b_lblkno && 1370 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1371 bp->b_left = root->b_left; 1372 bp->b_right = root; 1373 root->b_left = NULL; 1374 TAILQ_INSERT_BEFORE(root, bp, b_vnbufs); 1375 } else { 1376 bp->b_right = root->b_right; 1377 bp->b_left = root; 1378 root->b_right = NULL; 1379 TAILQ_INSERT_AFTER(&vp->v_cleanblkhd, 1380 root, bp, b_vnbufs); 1381 } 1382 vp->v_cleanbufcnt++; 1383 vp->v_cleanblkroot = bp; 1384 } 1385} 1386 1387/* 1388 * Lookup a buffer using the splay tree. Note that we specifically avoid 1389 * shadow buffers used in background bitmap writes. 1390 * 1391 * This code isn't quite efficient as it could be because we are maintaining 1392 * two sorted lists and do not know which list the block resides in. 1393 * 1394 * During a "make buildworld" the desired buffer is found at one of 1395 * the roots more than 60% of the time. Thus, checking both roots 1396 * before performing either splay eliminates unnecessary splays on the 1397 * first tree splayed. 1398 */ 1399struct buf * 1400gbincore(struct vnode *vp, daddr_t lblkno) 1401{ 1402 struct buf *bp; 1403 1404 GIANT_REQUIRED; 1405 1406 ASSERT_VI_LOCKED(vp, "gbincore"); 1407 if ((bp = vp->v_cleanblkroot) != NULL && 1408 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1409 return (bp); 1410 if ((bp = vp->v_dirtyblkroot) != NULL && 1411 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1412 return (bp); 1413 if ((bp = vp->v_cleanblkroot) != NULL) { 1414 vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp); 1415 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1416 return (bp); 1417 } 1418 if ((bp = vp->v_dirtyblkroot) != NULL) { 1419 vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp); 1420 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1421 return (bp); 1422 } 1423 return (NULL); 1424} 1425 1426/* 1427 * Associate a buffer with a vnode. 1428 */ 1429void 1430bgetvp(vp, bp) 1431 register struct vnode *vp; 1432 register struct buf *bp; 1433{ 1434 1435 KASSERT(bp->b_vp == NULL, ("bgetvp: not free")); 1436 1437 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1438 ("bgetvp: bp already attached! %p", bp)); 1439 1440 ASSERT_VI_LOCKED(vp, "bgetvp"); 1441 vholdl(vp); 1442 bp->b_vp = vp; 1443 bp->b_dev = vn_todev(vp); 1444 /* 1445 * Insert onto list for new vnode. 1446 */ 1447 buf_vlist_add(bp, vp, BX_VNCLEAN); 1448} 1449 1450/* 1451 * Disassociate a buffer from a vnode. 1452 */ 1453void 1454brelvp(bp) 1455 register struct buf *bp; 1456{ 1457 struct vnode *vp; 1458 1459 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1460 1461 /* 1462 * Delete from old vnode list, if on one. 1463 */ 1464 vp = bp->b_vp; 1465 VI_LOCK(vp); 1466 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1467 buf_vlist_remove(bp); 1468 if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) { 1469 vp->v_iflag &= ~VI_ONWORKLST; 1470 mtx_lock(&sync_mtx); 1471 LIST_REMOVE(vp, v_synclist); 1472 syncer_worklist_len--; 1473 mtx_unlock(&sync_mtx); 1474 } 1475 vdropl(vp); 1476 bp->b_vp = (struct vnode *) 0; 1477 if (bp->b_object) 1478 bp->b_object = NULL; 1479 VI_UNLOCK(vp); 1480} 1481 1482/* 1483 * Add an item to the syncer work queue. 1484 */ 1485static void 1486vn_syncer_add_to_worklist(struct vnode *vp, int delay) 1487{ 1488 int slot; 1489 1490 ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist"); 1491 1492 mtx_lock(&sync_mtx); 1493 if (vp->v_iflag & VI_ONWORKLST) 1494 LIST_REMOVE(vp, v_synclist); 1495 else { 1496 vp->v_iflag |= VI_ONWORKLST; 1497 syncer_worklist_len++; 1498 } 1499 1500 if (delay > syncer_maxdelay - 2) 1501 delay = syncer_maxdelay - 2; 1502 slot = (syncer_delayno + delay) & syncer_mask; 1503 1504 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist); 1505 mtx_unlock(&sync_mtx); 1506} 1507 1508static int 1509sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1510{ 1511 int error, len; 1512 1513 mtx_lock(&sync_mtx); 1514 len = syncer_worklist_len - sync_vnode_count; 1515 mtx_unlock(&sync_mtx); 1516 error = SYSCTL_OUT(req, &len, sizeof(len)); 1517 return (error); 1518} 1519 1520SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1521 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1522 1523struct proc *updateproc; 1524static void sched_sync(void); 1525static struct kproc_desc up_kp = { 1526 "syncer", 1527 sched_sync, 1528 &updateproc 1529}; 1530SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1531 1532/* 1533 * System filesystem synchronizer daemon. 1534 */ 1535static void 1536sched_sync(void) 1537{ 1538 struct synclist *next; 1539 struct synclist *slp; 1540 struct vnode *vp; 1541 struct mount *mp; 1542 long starttime; 1543 struct thread *td = FIRST_THREAD_IN_PROC(updateproc); 1544 static int dummychan; 1545 int last_work_seen; 1546 int net_worklist_len; 1547 int syncer_final_iter; 1548 1549 mtx_lock(&Giant); 1550 last_work_seen = 0; 1551 syncer_final_iter = 0; 1552 syncer_state = SYNCER_RUNNING; 1553 starttime = time_second; 1554 1555 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1556 SHUTDOWN_PRI_LAST); 1557 1558 for (;;) { 1559 mtx_lock(&sync_mtx); 1560 if (syncer_state == SYNCER_FINAL_DELAY && 1561 syncer_final_iter == 0) { 1562 mtx_unlock(&sync_mtx); 1563 kthread_suspend_check(td->td_proc); 1564 mtx_lock(&sync_mtx); 1565 } 1566 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1567 if (syncer_state != SYNCER_RUNNING && starttime != time_second) 1568 printf("%d ", net_worklist_len); 1569 starttime = time_second; 1570 1571 /* 1572 * Push files whose dirty time has expired. Be careful 1573 * of interrupt race on slp queue. 1574 * 1575 * Skip over empty worklist slots when shutting down. 1576 */ 1577 do { 1578 slp = &syncer_workitem_pending[syncer_delayno]; 1579 syncer_delayno += 1; 1580 if (syncer_delayno == syncer_maxdelay) 1581 syncer_delayno = 0; 1582 next = &syncer_workitem_pending[syncer_delayno]; 1583 /* 1584 * If the worklist has wrapped since the 1585 * it was emptied of all but syncer vnodes, 1586 * switch to the FINAL_DELAY state and run 1587 * for one more second. 1588 */ 1589 if (syncer_state == SYNCER_SHUTTING_DOWN && 1590 net_worklist_len == 0 && 1591 last_work_seen == syncer_delayno) { 1592 syncer_state = SYNCER_FINAL_DELAY; 1593 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1594 } 1595 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1596 syncer_worklist_len > 0); 1597 1598 /* 1599 * Keep track of the last time there was anything 1600 * on the worklist other than syncer vnodes. 1601 * Return to the SHUTTING_DOWN state if any 1602 * new work appears. 1603 */ 1604 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1605 last_work_seen = syncer_delayno; 1606 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1607 syncer_state = SYNCER_SHUTTING_DOWN; 1608 while ((vp = LIST_FIRST(slp)) != NULL) { 1609 if (VOP_ISLOCKED(vp, NULL) != 0 || 1610 vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1611 LIST_REMOVE(vp, v_synclist); 1612 LIST_INSERT_HEAD(next, vp, v_synclist); 1613 continue; 1614 } 1615 if (VI_TRYLOCK(vp) == 0) { 1616 LIST_REMOVE(vp, v_synclist); 1617 LIST_INSERT_HEAD(next, vp, v_synclist); 1618 vn_finished_write(mp); 1619 continue; 1620 } 1621 /* 1622 * We use vhold in case the vnode does not 1623 * successfully sync. vhold prevents the vnode from 1624 * going away when we unlock the sync_mtx so that 1625 * we can acquire the vnode interlock. 1626 */ 1627 vholdl(vp); 1628 mtx_unlock(&sync_mtx); 1629 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td); 1630 (void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td); 1631 VOP_UNLOCK(vp, 0, td); 1632 vn_finished_write(mp); 1633 VI_LOCK(vp); 1634 if ((vp->v_iflag & VI_ONWORKLST) != 0) { 1635 /* 1636 * Put us back on the worklist. The worklist 1637 * routine will remove us from our current 1638 * position and then add us back in at a later 1639 * position. 1640 */ 1641 vn_syncer_add_to_worklist(vp, syncdelay); 1642 } 1643 vdropl(vp); 1644 VI_UNLOCK(vp); 1645 mtx_lock(&sync_mtx); 1646 } 1647 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1648 syncer_final_iter--; 1649 mtx_unlock(&sync_mtx); 1650 1651 /* 1652 * Do soft update processing. 1653 */ 1654 if (softdep_process_worklist_hook != NULL) 1655 (*softdep_process_worklist_hook)(NULL); 1656 1657 /* 1658 * The variable rushjob allows the kernel to speed up the 1659 * processing of the filesystem syncer process. A rushjob 1660 * value of N tells the filesystem syncer to process the next 1661 * N seconds worth of work on its queue ASAP. Currently rushjob 1662 * is used by the soft update code to speed up the filesystem 1663 * syncer process when the incore state is getting so far 1664 * ahead of the disk that the kernel memory pool is being 1665 * threatened with exhaustion. 1666 */ 1667 mtx_lock(&sync_mtx); 1668 if (rushjob > 0) { 1669 rushjob -= 1; 1670 mtx_unlock(&sync_mtx); 1671 continue; 1672 } 1673 mtx_unlock(&sync_mtx); 1674 /* 1675 * Just sleep for a short period if time between 1676 * iterations when shutting down to allow some I/O 1677 * to happen. 1678 * 1679 * If it has taken us less than a second to process the 1680 * current work, then wait. Otherwise start right over 1681 * again. We can still lose time if any single round 1682 * takes more than two seconds, but it does not really 1683 * matter as we are just trying to generally pace the 1684 * filesystem activity. 1685 */ 1686 if (syncer_state != SYNCER_RUNNING) 1687 tsleep(&dummychan, PPAUSE, "syncfnl", 1688 hz / SYNCER_SHUTDOWN_SPEEDUP); 1689 else if (time_second == starttime) 1690 tsleep(&lbolt, PPAUSE, "syncer", 0); 1691 } 1692} 1693 1694/* 1695 * Request the syncer daemon to speed up its work. 1696 * We never push it to speed up more than half of its 1697 * normal turn time, otherwise it could take over the cpu. 1698 */ 1699int 1700speedup_syncer() 1701{ 1702 struct thread *td; 1703 int ret = 0; 1704 1705 td = FIRST_THREAD_IN_PROC(updateproc); 1706 sleepq_remove(td, &lbolt); 1707 mtx_lock(&sync_mtx); 1708 if (rushjob < syncdelay / 2) { 1709 rushjob += 1; 1710 stat_rush_requests += 1; 1711 ret = 1; 1712 } 1713 mtx_unlock(&sync_mtx); 1714 return (ret); 1715} 1716 1717/* 1718 * Tell the syncer to speed up its work and run though its work 1719 * list several times, then tell it to shut down. 1720 */ 1721static void 1722syncer_shutdown(void *arg, int howto) 1723{ 1724 struct thread *td; 1725 1726 td = FIRST_THREAD_IN_PROC(updateproc); 1727 sleepq_remove(td, &lbolt); 1728 mtx_lock(&sync_mtx); 1729 syncer_state = SYNCER_SHUTTING_DOWN; 1730 rushjob = 0; 1731 mtx_unlock(&sync_mtx); 1732 kproc_shutdown(arg, howto); 1733} 1734 1735/* 1736 * Associate a p-buffer with a vnode. 1737 * 1738 * Also sets B_PAGING flag to indicate that vnode is not fully associated 1739 * with the buffer. i.e. the bp has not been linked into the vnode or 1740 * ref-counted. 1741 */ 1742void 1743pbgetvp(vp, bp) 1744 register struct vnode *vp; 1745 register struct buf *bp; 1746{ 1747 1748 KASSERT(bp->b_vp == NULL, ("pbgetvp: not free")); 1749 1750 bp->b_vp = vp; 1751 bp->b_object = vp->v_object; 1752 bp->b_flags |= B_PAGING; 1753 bp->b_dev = vn_todev(vp); 1754} 1755 1756/* 1757 * Disassociate a p-buffer from a vnode. 1758 */ 1759void 1760pbrelvp(bp) 1761 register struct buf *bp; 1762{ 1763 1764 KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL")); 1765 1766 /* XXX REMOVE ME */ 1767 VI_LOCK(bp->b_vp); 1768 if (TAILQ_NEXT(bp, b_vnbufs) != NULL) { 1769 panic( 1770 "relpbuf(): b_vp was probably reassignbuf()d %p %x", 1771 bp, 1772 (int)bp->b_flags 1773 ); 1774 } 1775 VI_UNLOCK(bp->b_vp); 1776 bp->b_vp = (struct vnode *) 0; 1777 bp->b_object = NULL; 1778 bp->b_flags &= ~B_PAGING; 1779} 1780 1781/* 1782 * Reassign a buffer from one vnode to another. 1783 * Used to assign file specific control information 1784 * (indirect blocks) to the vnode to which they belong. 1785 */ 1786void 1787reassignbuf(bp, newvp) 1788 register struct buf *bp; 1789 register struct vnode *newvp; 1790{ 1791 struct vnode *vp; 1792 int delay; 1793 1794 if (newvp == NULL) { 1795 printf("reassignbuf: NULL"); 1796 return; 1797 } 1798 vp = bp->b_vp; 1799 ++reassignbufcalls; 1800 1801 /* 1802 * B_PAGING flagged buffers cannot be reassigned because their vp 1803 * is not fully linked in. 1804 */ 1805 if (bp->b_flags & B_PAGING) 1806 panic("cannot reassign paging buffer"); 1807 1808 /* 1809 * Delete from old vnode list, if on one. 1810 */ 1811 VI_LOCK(vp); 1812 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) { 1813 buf_vlist_remove(bp); 1814 if (vp != newvp) { 1815 vdropl(bp->b_vp); 1816 bp->b_vp = NULL; /* for clarification */ 1817 } 1818 } 1819 if (vp != newvp) { 1820 VI_UNLOCK(vp); 1821 VI_LOCK(newvp); 1822 } 1823 /* 1824 * If dirty, put on list of dirty buffers; otherwise insert onto list 1825 * of clean buffers. 1826 */ 1827 if (bp->b_flags & B_DELWRI) { 1828 if ((newvp->v_iflag & VI_ONWORKLST) == 0) { 1829 switch (newvp->v_type) { 1830 case VDIR: 1831 delay = dirdelay; 1832 break; 1833 case VCHR: 1834 delay = metadelay; 1835 break; 1836 default: 1837 delay = filedelay; 1838 } 1839 vn_syncer_add_to_worklist(newvp, delay); 1840 } 1841 buf_vlist_add(bp, newvp, BX_VNDIRTY); 1842 } else { 1843 buf_vlist_add(bp, newvp, BX_VNCLEAN); 1844 1845 if ((newvp->v_iflag & VI_ONWORKLST) && 1846 TAILQ_EMPTY(&newvp->v_dirtyblkhd)) { 1847 mtx_lock(&sync_mtx); 1848 LIST_REMOVE(newvp, v_synclist); 1849 syncer_worklist_len--; 1850 mtx_unlock(&sync_mtx); 1851 newvp->v_iflag &= ~VI_ONWORKLST; 1852 } 1853 } 1854 if (bp->b_vp != newvp) { 1855 bp->b_vp = newvp; 1856 vholdl(bp->b_vp); 1857 } 1858 VI_UNLOCK(newvp); 1859} 1860 1861/* 1862 * Create a vnode for a device. 1863 * Used for mounting the root filesystem. 1864 */ 1865int 1866bdevvp(dev, vpp) 1867 struct cdev *dev; 1868 struct vnode **vpp; 1869{ 1870 register struct vnode *vp; 1871 struct vnode *nvp; 1872 int error; 1873 1874 if (dev == NULL) { 1875 *vpp = NULLVP; 1876 return (ENXIO); 1877 } 1878 if (vfinddev(dev, vpp)) 1879 return (0); 1880 1881 error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp); 1882 if (error) { 1883 *vpp = NULLVP; 1884 return (error); 1885 } 1886 vp = nvp; 1887 vp->v_type = VCHR; 1888 vp->v_bsize = DEV_BSIZE; 1889 addalias(vp, dev); 1890 *vpp = vp; 1891 return (0); 1892} 1893 1894static void 1895v_incr_usecount(struct vnode *vp, int delta) 1896{ 1897 1898 vp->v_usecount += delta; 1899 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1900 mtx_lock(&spechash_mtx); 1901 vp->v_rdev->si_usecount += delta; 1902 mtx_unlock(&spechash_mtx); 1903 } 1904} 1905 1906/* 1907 * Add vnode to the alias list hung off the struct cdev *. 1908 * 1909 * The reason for this gunk is that multiple vnodes can reference 1910 * the same physical device, so checking vp->v_usecount to see 1911 * how many users there are is inadequate; the v_usecount for 1912 * the vnodes need to be accumulated. vcount() does that. 1913 */ 1914struct vnode * 1915addaliasu(nvp, nvp_rdev) 1916 struct vnode *nvp; 1917 dev_t nvp_rdev; 1918{ 1919 struct vnode *ovp; 1920 vop_t **ops; 1921 struct cdev *dev; 1922 1923 if (nvp->v_type == VBLK) 1924 return (nvp); 1925 if (nvp->v_type != VCHR) 1926 panic("addaliasu on non-special vnode"); 1927 dev = findcdev(nvp_rdev); 1928 if (dev == NULL) 1929 return (nvp); 1930 /* 1931 * Check to see if we have a bdevvp vnode with no associated 1932 * filesystem. If so, we want to associate the filesystem of 1933 * the new newly instigated vnode with the bdevvp vnode and 1934 * discard the newly created vnode rather than leaving the 1935 * bdevvp vnode lying around with no associated filesystem. 1936 */ 1937 if (vfinddev(dev, &ovp) == 0 || ovp->v_data != NULL) { 1938 addalias(nvp, dev); 1939 return (nvp); 1940 } 1941 /* 1942 * Discard unneeded vnode, but save its node specific data. 1943 * Note that if there is a lock, it is carried over in the 1944 * node specific data to the replacement vnode. 1945 */ 1946 vref(ovp); 1947 ovp->v_data = nvp->v_data; 1948 ovp->v_tag = nvp->v_tag; 1949 nvp->v_data = NULL; 1950 lockdestroy(ovp->v_vnlock); 1951 lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg, 1952 nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK); 1953 ops = ovp->v_op; 1954 ovp->v_op = nvp->v_op; 1955 if (VOP_ISLOCKED(nvp, curthread)) { 1956 VOP_UNLOCK(nvp, 0, curthread); 1957 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread); 1958 } 1959 nvp->v_op = ops; 1960 delmntque(ovp); 1961 insmntque(ovp, nvp->v_mount); 1962 vrele(nvp); 1963 vgone(nvp); 1964 return (ovp); 1965} 1966 1967/* This is a local helper function that do the same as addaliasu, but for a 1968 * struct cdev *instead of an dev_t. */ 1969static void 1970addalias(nvp, dev) 1971 struct vnode *nvp; 1972 struct cdev *dev; 1973{ 1974 1975 KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode")); 1976 dev_ref(dev); 1977 nvp->v_rdev = dev; 1978 VI_LOCK(nvp); 1979 mtx_lock(&spechash_mtx); 1980 SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext); 1981 dev->si_usecount += nvp->v_usecount; 1982 mtx_unlock(&spechash_mtx); 1983 VI_UNLOCK(nvp); 1984} 1985 1986/* 1987 * Grab a particular vnode from the free list, increment its 1988 * reference count and lock it. The vnode lock bit is set if the 1989 * vnode is being eliminated in vgone. The process is awakened 1990 * when the transition is completed, and an error returned to 1991 * indicate that the vnode is no longer usable (possibly having 1992 * been changed to a new filesystem type). 1993 */ 1994int 1995vget(vp, flags, td) 1996 register struct vnode *vp; 1997 int flags; 1998 struct thread *td; 1999{ 2000 int error; 2001 2002 /* 2003 * If the vnode is in the process of being cleaned out for 2004 * another use, we wait for the cleaning to finish and then 2005 * return failure. Cleaning is determined by checking that 2006 * the VI_XLOCK flag is set. 2007 */ 2008 if ((flags & LK_INTERLOCK) == 0) 2009 VI_LOCK(vp); 2010 if (vp->v_iflag & VI_XLOCK && vp->v_vxthread != curthread) { 2011 if ((flags & LK_NOWAIT) == 0) { 2012 vp->v_iflag |= VI_XWANT; 2013 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0); 2014 return (ENOENT); 2015 } 2016 VI_UNLOCK(vp); 2017 return (EBUSY); 2018 } 2019 2020 v_incr_usecount(vp, 1); 2021 2022 if (VSHOULDBUSY(vp)) 2023 vbusy(vp); 2024 if (flags & LK_TYPE_MASK) { 2025 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) { 2026 /* 2027 * must expand vrele here because we do not want 2028 * to call VOP_INACTIVE if the reference count 2029 * drops back to zero since it was never really 2030 * active. We must remove it from the free list 2031 * before sleeping so that multiple processes do 2032 * not try to recycle it. 2033 */ 2034 VI_LOCK(vp); 2035 v_incr_usecount(vp, -1); 2036 if (VSHOULDFREE(vp)) 2037 vfree(vp); 2038 else 2039 vlruvp(vp); 2040 VI_UNLOCK(vp); 2041 } 2042 return (error); 2043 } 2044 VI_UNLOCK(vp); 2045 return (0); 2046} 2047 2048/* 2049 * Increase the reference count of a vnode. 2050 */ 2051void 2052vref(struct vnode *vp) 2053{ 2054 2055 VI_LOCK(vp); 2056 v_incr_usecount(vp, 1); 2057 VI_UNLOCK(vp); 2058} 2059 2060/* 2061 * Return reference count of a vnode. 2062 * 2063 * The results of this call are only guaranteed when some mechanism other 2064 * than the VI lock is used to stop other processes from gaining references 2065 * to the vnode. This may be the case if the caller holds the only reference. 2066 * This is also useful when stale data is acceptable as race conditions may 2067 * be accounted for by some other means. 2068 */ 2069int 2070vrefcnt(struct vnode *vp) 2071{ 2072 int usecnt; 2073 2074 VI_LOCK(vp); 2075 usecnt = vp->v_usecount; 2076 VI_UNLOCK(vp); 2077 2078 return (usecnt); 2079} 2080 2081 2082/* 2083 * Vnode put/release. 2084 * If count drops to zero, call inactive routine and return to freelist. 2085 */ 2086void 2087vrele(vp) 2088 struct vnode *vp; 2089{ 2090 struct thread *td = curthread; /* XXX */ 2091 2092 GIANT_REQUIRED; 2093 2094 KASSERT(vp != NULL, ("vrele: null vp")); 2095 2096 VI_LOCK(vp); 2097 2098 /* Skip this v_writecount check if we're going to panic below. */ 2099 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, 2100 ("vrele: missed vn_close")); 2101 2102 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2103 vp->v_usecount == 1)) { 2104 v_incr_usecount(vp, -1); 2105 VI_UNLOCK(vp); 2106 2107 return; 2108 } 2109 2110 if (vp->v_usecount == 1) { 2111 v_incr_usecount(vp, -1); 2112 /* 2113 * We must call VOP_INACTIVE with the node locked. Mark 2114 * as VI_DOINGINACT to avoid recursion. 2115 */ 2116 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) { 2117 VI_LOCK(vp); 2118 vp->v_iflag |= VI_DOINGINACT; 2119 VI_UNLOCK(vp); 2120 VOP_INACTIVE(vp, td); 2121 VI_LOCK(vp); 2122 KASSERT(vp->v_iflag & VI_DOINGINACT, 2123 ("vrele: lost VI_DOINGINACT")); 2124 vp->v_iflag &= ~VI_DOINGINACT; 2125 } else 2126 VI_LOCK(vp); 2127 if (VSHOULDFREE(vp)) 2128 vfree(vp); 2129 else 2130 vlruvp(vp); 2131 VI_UNLOCK(vp); 2132 2133 } else { 2134#ifdef DIAGNOSTIC 2135 vprint("vrele: negative ref count", vp); 2136#endif 2137 VI_UNLOCK(vp); 2138 panic("vrele: negative ref cnt"); 2139 } 2140} 2141 2142/* 2143 * Release an already locked vnode. This give the same effects as 2144 * unlock+vrele(), but takes less time and avoids releasing and 2145 * re-aquiring the lock (as vrele() aquires the lock internally.) 2146 */ 2147void 2148vput(vp) 2149 struct vnode *vp; 2150{ 2151 struct thread *td = curthread; /* XXX */ 2152 2153 GIANT_REQUIRED; 2154 2155 KASSERT(vp != NULL, ("vput: null vp")); 2156 VI_LOCK(vp); 2157 /* Skip this v_writecount check if we're going to panic below. */ 2158 KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, 2159 ("vput: missed vn_close")); 2160 2161 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2162 vp->v_usecount == 1)) { 2163 v_incr_usecount(vp, -1); 2164 VOP_UNLOCK(vp, LK_INTERLOCK, td); 2165 return; 2166 } 2167 2168 if (vp->v_usecount == 1) { 2169 v_incr_usecount(vp, -1); 2170 /* 2171 * We must call VOP_INACTIVE with the node locked, so 2172 * we just need to release the vnode mutex. Mark as 2173 * as VI_DOINGINACT to avoid recursion. 2174 */ 2175 vp->v_iflag |= VI_DOINGINACT; 2176 VI_UNLOCK(vp); 2177 VOP_INACTIVE(vp, td); 2178 VI_LOCK(vp); 2179 KASSERT(vp->v_iflag & VI_DOINGINACT, 2180 ("vput: lost VI_DOINGINACT")); 2181 vp->v_iflag &= ~VI_DOINGINACT; 2182 if (VSHOULDFREE(vp)) 2183 vfree(vp); 2184 else 2185 vlruvp(vp); 2186 VI_UNLOCK(vp); 2187 2188 } else { 2189#ifdef DIAGNOSTIC 2190 vprint("vput: negative ref count", vp); 2191#endif 2192 panic("vput: negative ref cnt"); 2193 } 2194} 2195 2196/* 2197 * Somebody doesn't want the vnode recycled. 2198 */ 2199void 2200vhold(struct vnode *vp) 2201{ 2202 2203 VI_LOCK(vp); 2204 vholdl(vp); 2205 VI_UNLOCK(vp); 2206} 2207 2208void 2209vholdl(vp) 2210 register struct vnode *vp; 2211{ 2212 2213 vp->v_holdcnt++; 2214 if (VSHOULDBUSY(vp)) 2215 vbusy(vp); 2216} 2217 2218/* 2219 * Note that there is one less who cares about this vnode. vdrop() is the 2220 * opposite of vhold(). 2221 */ 2222void 2223vdrop(struct vnode *vp) 2224{ 2225 2226 VI_LOCK(vp); 2227 vdropl(vp); 2228 VI_UNLOCK(vp); 2229} 2230 2231void 2232vdropl(vp) 2233 register struct vnode *vp; 2234{ 2235 2236 if (vp->v_holdcnt <= 0) 2237 panic("vdrop: holdcnt"); 2238 vp->v_holdcnt--; 2239 if (VSHOULDFREE(vp)) 2240 vfree(vp); 2241 else 2242 vlruvp(vp); 2243} 2244 2245/* 2246 * Remove any vnodes in the vnode table belonging to mount point mp. 2247 * 2248 * If FORCECLOSE is not specified, there should not be any active ones, 2249 * return error if any are found (nb: this is a user error, not a 2250 * system error). If FORCECLOSE is specified, detach any active vnodes 2251 * that are found. 2252 * 2253 * If WRITECLOSE is set, only flush out regular file vnodes open for 2254 * writing. 2255 * 2256 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2257 * 2258 * `rootrefs' specifies the base reference count for the root vnode 2259 * of this filesystem. The root vnode is considered busy if its 2260 * v_usecount exceeds this value. On a successful return, vflush() 2261 * will call vrele() on the root vnode exactly rootrefs times. 2262 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2263 * be zero. 2264 */ 2265#ifdef DIAGNOSTIC 2266static int busyprt = 0; /* print out busy vnodes */ 2267SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 2268#endif 2269 2270int 2271vflush(mp, rootrefs, flags) 2272 struct mount *mp; 2273 int rootrefs; 2274 int flags; 2275{ 2276 struct thread *td = curthread; /* XXX */ 2277 struct vnode *vp, *nvp, *rootvp = NULL; 2278 struct vattr vattr; 2279 int busy = 0, error; 2280 2281 if (rootrefs > 0) { 2282 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2283 ("vflush: bad args")); 2284 /* 2285 * Get the filesystem root vnode. We can vput() it 2286 * immediately, since with rootrefs > 0, it won't go away. 2287 */ 2288 if ((error = VFS_ROOT(mp, &rootvp)) != 0) 2289 return (error); 2290 vput(rootvp); 2291 2292 } 2293 MNT_ILOCK(mp); 2294loop: 2295 MNT_VNODE_FOREACH(vp, mp, nvp) { 2296 2297 VI_LOCK(vp); 2298 MNT_IUNLOCK(mp); 2299 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td); 2300 if (error) { 2301 MNT_ILOCK(mp); 2302 goto loop; 2303 } 2304 /* 2305 * Skip over a vnodes marked VV_SYSTEM. 2306 */ 2307 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2308 VOP_UNLOCK(vp, 0, td); 2309 MNT_ILOCK(mp); 2310 continue; 2311 } 2312 /* 2313 * If WRITECLOSE is set, flush out unlinked but still open 2314 * files (even if open only for reading) and regular file 2315 * vnodes open for writing. 2316 */ 2317 if (flags & WRITECLOSE) { 2318 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td); 2319 VI_LOCK(vp); 2320 2321 if ((vp->v_type == VNON || 2322 (error == 0 && vattr.va_nlink > 0)) && 2323 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2324 VOP_UNLOCK(vp, LK_INTERLOCK, td); 2325 MNT_ILOCK(mp); 2326 continue; 2327 } 2328 } else 2329 VI_LOCK(vp); 2330 2331 VOP_UNLOCK(vp, 0, td); 2332 2333 /* 2334 * With v_usecount == 0, all we need to do is clear out the 2335 * vnode data structures and we are done. 2336 */ 2337 if (vp->v_usecount == 0) { 2338 vgonel(vp, td); 2339 MNT_ILOCK(mp); 2340 continue; 2341 } 2342 2343 /* 2344 * If FORCECLOSE is set, forcibly close the vnode. For block 2345 * or character devices, revert to an anonymous device. For 2346 * all other files, just kill them. 2347 */ 2348 if (flags & FORCECLOSE) { 2349 if (vp->v_type != VCHR) 2350 vgonel(vp, td); 2351 else 2352 vgonechrl(vp, td); 2353 MNT_ILOCK(mp); 2354 continue; 2355 } 2356#ifdef DIAGNOSTIC 2357 if (busyprt) 2358 vprint("vflush: busy vnode", vp); 2359#endif 2360 VI_UNLOCK(vp); 2361 MNT_ILOCK(mp); 2362 busy++; 2363 } 2364 MNT_IUNLOCK(mp); 2365 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2366 /* 2367 * If just the root vnode is busy, and if its refcount 2368 * is equal to `rootrefs', then go ahead and kill it. 2369 */ 2370 VI_LOCK(rootvp); 2371 KASSERT(busy > 0, ("vflush: not busy")); 2372 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs")); 2373 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2374 vgonel(rootvp, td); 2375 busy = 0; 2376 } else 2377 VI_UNLOCK(rootvp); 2378 } 2379 if (busy) 2380 return (EBUSY); 2381 for (; rootrefs > 0; rootrefs--) 2382 vrele(rootvp); 2383 return (0); 2384} 2385 2386/* 2387 * This moves a now (likely recyclable) vnode to the end of the 2388 * mountlist. XXX However, it is temporarily disabled until we 2389 * can clean up ffs_sync() and friends, which have loop restart 2390 * conditions which this code causes to operate O(N^2). 2391 */ 2392static void 2393vlruvp(struct vnode *vp) 2394{ 2395#if 0 2396 struct mount *mp; 2397 2398 if ((mp = vp->v_mount) != NULL) { 2399 MNT_ILOCK(mp); 2400 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 2401 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 2402 MNT_IUNLOCK(mp); 2403 } 2404#endif 2405} 2406 2407static void 2408vx_lock(struct vnode *vp) 2409{ 2410 2411 ASSERT_VI_LOCKED(vp, "vx_lock"); 2412 2413 /* 2414 * Prevent the vnode from being recycled or brought into use while we 2415 * clean it out. 2416 */ 2417 if (vp->v_iflag & VI_XLOCK) 2418 panic("vclean: deadlock"); 2419 vp->v_iflag |= VI_XLOCK; 2420 vp->v_vxthread = curthread; 2421} 2422 2423static void 2424vx_unlock(struct vnode *vp) 2425{ 2426 ASSERT_VI_LOCKED(vp, "vx_unlock"); 2427 vp->v_iflag &= ~VI_XLOCK; 2428 vp->v_vxthread = NULL; 2429 if (vp->v_iflag & VI_XWANT) { 2430 vp->v_iflag &= ~VI_XWANT; 2431 wakeup(vp); 2432 } 2433} 2434 2435/* 2436 * Disassociate the underlying filesystem from a vnode. 2437 */ 2438static void 2439vclean(vp, flags, td) 2440 struct vnode *vp; 2441 int flags; 2442 struct thread *td; 2443{ 2444 int active; 2445 2446 ASSERT_VI_LOCKED(vp, "vclean"); 2447 /* 2448 * Check to see if the vnode is in use. If so we have to reference it 2449 * before we clean it out so that its count cannot fall to zero and 2450 * generate a race against ourselves to recycle it. 2451 */ 2452 if ((active = vp->v_usecount)) 2453 v_incr_usecount(vp, 1); 2454 2455 /* 2456 * Even if the count is zero, the VOP_INACTIVE routine may still 2457 * have the object locked while it cleans it out. The VOP_LOCK 2458 * ensures that the VOP_INACTIVE routine is done with its work. 2459 * For active vnodes, it ensures that no other activity can 2460 * occur while the underlying object is being cleaned out. 2461 */ 2462 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td); 2463 2464 /* 2465 * Clean out any buffers associated with the vnode. 2466 * If the flush fails, just toss the buffers. 2467 */ 2468 if (flags & DOCLOSE) { 2469 struct buf *bp; 2470 bp = TAILQ_FIRST(&vp->v_dirtyblkhd); 2471 if (bp != NULL) 2472 (void) vn_write_suspend_wait(vp, NULL, V_WAIT); 2473 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0) 2474 vinvalbuf(vp, 0, NOCRED, td, 0, 0); 2475 } 2476 2477 VOP_DESTROYVOBJECT(vp); 2478 2479 /* 2480 * Any other processes trying to obtain this lock must first 2481 * wait for VXLOCK to clear, then call the new lock operation. 2482 */ 2483 VOP_UNLOCK(vp, 0, td); 2484 2485 /* 2486 * If purging an active vnode, it must be closed and 2487 * deactivated before being reclaimed. Note that the 2488 * VOP_INACTIVE will unlock the vnode. 2489 */ 2490 if (active) { 2491 if (flags & DOCLOSE) 2492 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2493 VI_LOCK(vp); 2494 if ((vp->v_iflag & VI_DOINGINACT) == 0) { 2495 vp->v_iflag |= VI_DOINGINACT; 2496 VI_UNLOCK(vp); 2497 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0) 2498 panic("vclean: cannot relock."); 2499 VOP_INACTIVE(vp, td); 2500 VI_LOCK(vp); 2501 KASSERT(vp->v_iflag & VI_DOINGINACT, 2502 ("vclean: lost VI_DOINGINACT")); 2503 vp->v_iflag &= ~VI_DOINGINACT; 2504 } 2505 VI_UNLOCK(vp); 2506 } 2507 /* 2508 * Reclaim the vnode. 2509 */ 2510 if (VOP_RECLAIM(vp, td)) 2511 panic("vclean: cannot reclaim"); 2512 2513 if (active) { 2514 /* 2515 * Inline copy of vrele() since VOP_INACTIVE 2516 * has already been called. 2517 */ 2518 VI_LOCK(vp); 2519 v_incr_usecount(vp, -1); 2520 if (vp->v_usecount <= 0) { 2521#ifdef INVARIANTS 2522 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 2523 vprint("vclean: bad ref count", vp); 2524 panic("vclean: ref cnt"); 2525 } 2526#endif 2527 if (VSHOULDFREE(vp)) 2528 vfree(vp); 2529 } 2530 VI_UNLOCK(vp); 2531 } 2532 /* 2533 * Delete from old mount point vnode list. 2534 */ 2535 delmntque(vp); 2536 cache_purge(vp); 2537 VI_LOCK(vp); 2538 if (VSHOULDFREE(vp)) 2539 vfree(vp); 2540 2541 /* 2542 * Done with purge, reset to the standard lock and 2543 * notify sleepers of the grim news. 2544 */ 2545 vp->v_vnlock = &vp->v_lock; 2546 vp->v_op = dead_vnodeop_p; 2547 if (vp->v_pollinfo != NULL) 2548 vn_pollgone(vp); 2549 vp->v_tag = "none"; 2550} 2551 2552/* 2553 * Eliminate all activity associated with the requested vnode 2554 * and with all vnodes aliased to the requested vnode. 2555 */ 2556int 2557vop_revoke(ap) 2558 struct vop_revoke_args /* { 2559 struct vnode *a_vp; 2560 int a_flags; 2561 } */ *ap; 2562{ 2563 struct vnode *vp, *vq; 2564 struct cdev *dev; 2565 2566 KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke")); 2567 vp = ap->a_vp; 2568 KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR")); 2569 2570 VI_LOCK(vp); 2571 /* 2572 * If a vgone (or vclean) is already in progress, 2573 * wait until it is done and return. 2574 */ 2575 if (vp->v_iflag & VI_XLOCK) { 2576 vp->v_iflag |= VI_XWANT; 2577 msleep(vp, VI_MTX(vp), PINOD | PDROP, 2578 "vop_revokeall", 0); 2579 return (0); 2580 } 2581 VI_UNLOCK(vp); 2582 dev = vp->v_rdev; 2583 for (;;) { 2584 mtx_lock(&spechash_mtx); 2585 vq = SLIST_FIRST(&dev->si_hlist); 2586 mtx_unlock(&spechash_mtx); 2587 if (vq == NULL) 2588 break; 2589 vgone(vq); 2590 } 2591 return (0); 2592} 2593 2594/* 2595 * Recycle an unused vnode to the front of the free list. 2596 * Release the passed interlock if the vnode will be recycled. 2597 */ 2598int 2599vrecycle(vp, inter_lkp, td) 2600 struct vnode *vp; 2601 struct mtx *inter_lkp; 2602 struct thread *td; 2603{ 2604 2605 VI_LOCK(vp); 2606 if (vp->v_usecount == 0) { 2607 if (inter_lkp) { 2608 mtx_unlock(inter_lkp); 2609 } 2610 vgonel(vp, td); 2611 return (1); 2612 } 2613 VI_UNLOCK(vp); 2614 return (0); 2615} 2616 2617/* 2618 * Eliminate all activity associated with a vnode 2619 * in preparation for reuse. 2620 */ 2621void 2622vgone(vp) 2623 register struct vnode *vp; 2624{ 2625 struct thread *td = curthread; /* XXX */ 2626 2627 VI_LOCK(vp); 2628 vgonel(vp, td); 2629} 2630 2631/* 2632 * Disassociate a character device from the its underlying filesystem and 2633 * attach it to spec. This is for use when the chr device is still active 2634 * and the filesystem is going away. 2635 */ 2636static void 2637vgonechrl(struct vnode *vp, struct thread *td) 2638{ 2639 ASSERT_VI_LOCKED(vp, "vgonechrl"); 2640 vx_lock(vp); 2641 /* 2642 * This is a custom version of vclean() which does not tearm down 2643 * the bufs or vm objects held by this vnode. This allows filesystems 2644 * to continue using devices which were discovered via another 2645 * filesystem that has been unmounted. 2646 */ 2647 if (vp->v_usecount != 0) { 2648 v_incr_usecount(vp, 1); 2649 /* 2650 * Ensure that no other activity can occur while the 2651 * underlying object is being cleaned out. 2652 */ 2653 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td); 2654 /* 2655 * Any other processes trying to obtain this lock must first 2656 * wait for VXLOCK to clear, then call the new lock operation. 2657 */ 2658 VOP_UNLOCK(vp, 0, td); 2659 vp->v_vnlock = &vp->v_lock; 2660 vp->v_tag = "orphanchr"; 2661 vp->v_op = spec_vnodeop_p; 2662 delmntque(vp); 2663 cache_purge(vp); 2664 vrele(vp); 2665 VI_LOCK(vp); 2666 } else 2667 vclean(vp, 0, td); 2668 vp->v_op = spec_vnodeop_p; 2669 vx_unlock(vp); 2670 VI_UNLOCK(vp); 2671} 2672 2673/* 2674 * vgone, with the vp interlock held. 2675 */ 2676void 2677vgonel(vp, td) 2678 struct vnode *vp; 2679 struct thread *td; 2680{ 2681 /* 2682 * If a vgone (or vclean) is already in progress, 2683 * wait until it is done and return. 2684 */ 2685 ASSERT_VI_LOCKED(vp, "vgonel"); 2686 if (vp->v_iflag & VI_XLOCK) { 2687 vp->v_iflag |= VI_XWANT; 2688 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0); 2689 return; 2690 } 2691 vx_lock(vp); 2692 2693 /* 2694 * Clean out the filesystem specific data. 2695 */ 2696 vclean(vp, DOCLOSE, td); 2697 VI_UNLOCK(vp); 2698 2699 /* 2700 * If special device, remove it from special device alias list 2701 * if it is on one. 2702 */ 2703 VI_LOCK(vp); 2704 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2705 mtx_lock(&spechash_mtx); 2706 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext); 2707 vp->v_rdev->si_usecount -= vp->v_usecount; 2708 mtx_unlock(&spechash_mtx); 2709 dev_rel(vp->v_rdev); 2710 vp->v_rdev = NULL; 2711 } 2712 2713 /* 2714 * If it is on the freelist and not already at the head, 2715 * move it to the head of the list. The test of the 2716 * VDOOMED flag and the reference count of zero is because 2717 * it will be removed from the free list by getnewvnode, 2718 * but will not have its reference count incremented until 2719 * after calling vgone. If the reference count were 2720 * incremented first, vgone would (incorrectly) try to 2721 * close the previous instance of the underlying object. 2722 */ 2723 if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) { 2724 mtx_lock(&vnode_free_list_mtx); 2725 if (vp->v_iflag & VI_FREE) { 2726 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2727 } else { 2728 vp->v_iflag |= VI_FREE; 2729 freevnodes++; 2730 } 2731 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2732 mtx_unlock(&vnode_free_list_mtx); 2733 } 2734 2735 vp->v_type = VBAD; 2736 vx_unlock(vp); 2737 VI_UNLOCK(vp); 2738} 2739 2740/* 2741 * Lookup a vnode by device number. 2742 */ 2743int 2744vfinddev(dev, vpp) 2745 struct cdev *dev; 2746 struct vnode **vpp; 2747{ 2748 struct vnode *vp; 2749 2750 mtx_lock(&spechash_mtx); 2751 SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) { 2752 *vpp = vp; 2753 mtx_unlock(&spechash_mtx); 2754 return (1); 2755 } 2756 mtx_unlock(&spechash_mtx); 2757 return (0); 2758} 2759 2760/* 2761 * Calculate the total number of references to a special device. 2762 */ 2763int 2764vcount(vp) 2765 struct vnode *vp; 2766{ 2767 int count; 2768 2769 mtx_lock(&spechash_mtx); 2770 count = vp->v_rdev->si_usecount; 2771 mtx_unlock(&spechash_mtx); 2772 return (count); 2773} 2774 2775/* 2776 * Same as above, but using the struct cdev *as argument 2777 */ 2778int 2779count_dev(dev) 2780 struct cdev *dev; 2781{ 2782 int count; 2783 2784 mtx_lock(&spechash_mtx); 2785 count = dev->si_usecount; 2786 mtx_unlock(&spechash_mtx); 2787 return(count); 2788} 2789 2790/* 2791 * Print out a description of a vnode. 2792 */ 2793static char *typename[] = 2794{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"}; 2795 2796void 2797vprint(label, vp) 2798 char *label; 2799 struct vnode *vp; 2800{ 2801 char buf[96]; 2802 2803 if (label != NULL) 2804 printf("%s: %p: ", label, (void *)vp); 2805 else 2806 printf("%p: ", (void *)vp); 2807 printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,", 2808 vp->v_tag, typename[vp->v_type], vp->v_usecount, 2809 vp->v_writecount, vp->v_holdcnt); 2810 buf[0] = '\0'; 2811 if (vp->v_vflag & VV_ROOT) 2812 strcat(buf, "|VV_ROOT"); 2813 if (vp->v_vflag & VV_TEXT) 2814 strcat(buf, "|VV_TEXT"); 2815 if (vp->v_vflag & VV_SYSTEM) 2816 strcat(buf, "|VV_SYSTEM"); 2817 if (vp->v_iflag & VI_XLOCK) 2818 strcat(buf, "|VI_XLOCK"); 2819 if (vp->v_iflag & VI_XWANT) 2820 strcat(buf, "|VI_XWANT"); 2821 if (vp->v_iflag & VI_BWAIT) 2822 strcat(buf, "|VI_BWAIT"); 2823 if (vp->v_iflag & VI_DOOMED) 2824 strcat(buf, "|VI_DOOMED"); 2825 if (vp->v_iflag & VI_FREE) 2826 strcat(buf, "|VI_FREE"); 2827 if (vp->v_vflag & VV_OBJBUF) 2828 strcat(buf, "|VV_OBJBUF"); 2829 if (buf[0] != '\0') 2830 printf(" flags (%s),", &buf[1]); 2831 lockmgr_printinfo(vp->v_vnlock); 2832 printf("\n"); 2833 if (vp->v_data != NULL) 2834 VOP_PRINT(vp); 2835} 2836 2837#ifdef DDB 2838#include <ddb/ddb.h> 2839/* 2840 * List all of the locked vnodes in the system. 2841 * Called when debugging the kernel. 2842 */ 2843DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2844{ 2845 struct mount *mp, *nmp; 2846 struct vnode *vp; 2847 2848 /* 2849 * Note: because this is DDB, we can't obey the locking semantics 2850 * for these structures, which means we could catch an inconsistent 2851 * state and dereference a nasty pointer. Not much to be done 2852 * about that. 2853 */ 2854 printf("Locked vnodes\n"); 2855 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2856 nmp = TAILQ_NEXT(mp, mnt_list); 2857 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2858 if (VOP_ISLOCKED(vp, NULL)) 2859 vprint(NULL, vp); 2860 } 2861 nmp = TAILQ_NEXT(mp, mnt_list); 2862 } 2863} 2864#endif 2865 2866/* 2867 * Fill in a struct xvfsconf based on a struct vfsconf. 2868 */ 2869static void 2870vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2871{ 2872 2873 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2874 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2875 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2876 xvfsp->vfc_flags = vfsp->vfc_flags; 2877 /* 2878 * These are unused in userland, we keep them 2879 * to not break binary compatibility. 2880 */ 2881 xvfsp->vfc_vfsops = NULL; 2882 xvfsp->vfc_next = NULL; 2883} 2884 2885/* 2886 * Top level filesystem related information gathering. 2887 */ 2888static int 2889sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2890{ 2891 struct vfsconf *vfsp; 2892 struct xvfsconf *xvfsp; 2893 int cnt, error, i; 2894 2895 cnt = 0; 2896 for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next) 2897 cnt++; 2898 xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK); 2899 /* 2900 * Handle the race that we will have here when struct vfsconf 2901 * will be locked down by using both cnt and checking vfc_next 2902 * against NULL to determine the end of the loop. The race will 2903 * happen because we will have to unlock before calling malloc(). 2904 * We are protected by Giant for now. 2905 */ 2906 i = 0; 2907 for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) { 2908 vfsconf2x(vfsp, xvfsp + i); 2909 i++; 2910 } 2911 error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i); 2912 free(xvfsp, M_TEMP); 2913 return (error); 2914} 2915 2916SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist, 2917 "S,xvfsconf", "List of all configured filesystems"); 2918 2919#ifndef BURN_BRIDGES 2920static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 2921 2922static int 2923vfs_sysctl(SYSCTL_HANDLER_ARGS) 2924{ 2925 int *name = (int *)arg1 - 1; /* XXX */ 2926 u_int namelen = arg2 + 1; /* XXX */ 2927 struct vfsconf *vfsp; 2928 struct xvfsconf xvfsp; 2929 2930 printf("WARNING: userland calling deprecated sysctl, " 2931 "please rebuild world\n"); 2932 2933#if 1 || defined(COMPAT_PRELITE2) 2934 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2935 if (namelen == 1) 2936 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2937#endif 2938 2939 switch (name[1]) { 2940 case VFS_MAXTYPENUM: 2941 if (namelen != 2) 2942 return (ENOTDIR); 2943 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2944 case VFS_CONF: 2945 if (namelen != 3) 2946 return (ENOTDIR); /* overloaded */ 2947 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) 2948 if (vfsp->vfc_typenum == name[2]) 2949 break; 2950 if (vfsp == NULL) 2951 return (EOPNOTSUPP); 2952 vfsconf2x(vfsp, &xvfsp); 2953 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 2954 } 2955 return (EOPNOTSUPP); 2956} 2957 2958SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl, 2959 "Generic filesystem"); 2960 2961#if 1 || defined(COMPAT_PRELITE2) 2962 2963static int 2964sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2965{ 2966 int error; 2967 struct vfsconf *vfsp; 2968 struct ovfsconf ovfs; 2969 2970 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) { 2971 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2972 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2973 ovfs.vfc_index = vfsp->vfc_typenum; 2974 ovfs.vfc_refcount = vfsp->vfc_refcount; 2975 ovfs.vfc_flags = vfsp->vfc_flags; 2976 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2977 if (error) 2978 return error; 2979 } 2980 return 0; 2981} 2982 2983#endif /* 1 || COMPAT_PRELITE2 */ 2984#endif /* !BURN_BRIDGES */ 2985 2986#define KINFO_VNODESLOP 10 2987#ifdef notyet 2988/* 2989 * Dump vnode list (via sysctl). 2990 */ 2991/* ARGSUSED */ 2992static int 2993sysctl_vnode(SYSCTL_HANDLER_ARGS) 2994{ 2995 struct xvnode *xvn; 2996 struct thread *td = req->td; 2997 struct mount *mp; 2998 struct vnode *vp; 2999 int error, len, n; 3000 3001 /* 3002 * Stale numvnodes access is not fatal here. 3003 */ 3004 req->lock = 0; 3005 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3006 if (!req->oldptr) 3007 /* Make an estimate */ 3008 return (SYSCTL_OUT(req, 0, len)); 3009 3010 error = sysctl_wire_old_buffer(req, 0); 3011 if (error != 0) 3012 return (error); 3013 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3014 n = 0; 3015 mtx_lock(&mountlist_mtx); 3016 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3017 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) 3018 continue; 3019 MNT_ILOCK(mp); 3020 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3021 if (n == len) 3022 break; 3023 vref(vp); 3024 xvn[n].xv_size = sizeof *xvn; 3025 xvn[n].xv_vnode = vp; 3026#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3027 XV_COPY(usecount); 3028 XV_COPY(writecount); 3029 XV_COPY(holdcnt); 3030 XV_COPY(id); 3031 XV_COPY(mount); 3032 XV_COPY(numoutput); 3033 XV_COPY(type); 3034#undef XV_COPY 3035 xvn[n].xv_flag = vp->v_vflag; 3036 3037 switch (vp->v_type) { 3038 case VREG: 3039 case VDIR: 3040 case VLNK: 3041 xvn[n].xv_dev = vp->v_cachedfs; 3042 xvn[n].xv_ino = vp->v_cachedid; 3043 break; 3044 case VBLK: 3045 case VCHR: 3046 if (vp->v_rdev == NULL) { 3047 vrele(vp); 3048 continue; 3049 } 3050 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3051 break; 3052 case VSOCK: 3053 xvn[n].xv_socket = vp->v_socket; 3054 break; 3055 case VFIFO: 3056 xvn[n].xv_fifo = vp->v_fifoinfo; 3057 break; 3058 case VNON: 3059 case VBAD: 3060 default: 3061 /* shouldn't happen? */ 3062 vrele(vp); 3063 continue; 3064 } 3065 vrele(vp); 3066 ++n; 3067 } 3068 MNT_IUNLOCK(mp); 3069 mtx_lock(&mountlist_mtx); 3070 vfs_unbusy(mp, td); 3071 if (n == len) 3072 break; 3073 } 3074 mtx_unlock(&mountlist_mtx); 3075 3076 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3077 free(xvn, M_TEMP); 3078 return (error); 3079} 3080 3081SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3082 0, 0, sysctl_vnode, "S,xvnode", ""); 3083#endif 3084 3085/* 3086 * Check to see if a filesystem is mounted on a block device. 3087 */ 3088int 3089vfs_mountedon(vp) 3090 struct vnode *vp; 3091{ 3092 3093 if (vp->v_rdev->si_mountpoint != NULL) 3094 return (EBUSY); 3095 return (0); 3096} 3097 3098/* 3099 * Unmount all filesystems. The list is traversed in reverse order 3100 * of mounting to avoid dependencies. 3101 */ 3102void 3103vfs_unmountall() 3104{ 3105 struct mount *mp; 3106 struct thread *td; 3107 int error; 3108 3109 if (curthread != NULL) 3110 td = curthread; 3111 else 3112 td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */ 3113 /* 3114 * Since this only runs when rebooting, it is not interlocked. 3115 */ 3116 while(!TAILQ_EMPTY(&mountlist)) { 3117 mp = TAILQ_LAST(&mountlist, mntlist); 3118 error = dounmount(mp, MNT_FORCE, td); 3119 if (error) { 3120 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3121 printf("unmount of %s failed (", 3122 mp->mnt_stat.f_mntonname); 3123 if (error == EBUSY) 3124 printf("BUSY)\n"); 3125 else 3126 printf("%d)\n", error); 3127 } else { 3128 /* The unmount has removed mp from the mountlist */ 3129 } 3130 } 3131} 3132 3133/* 3134 * perform msync on all vnodes under a mount point 3135 * the mount point must be locked. 3136 */ 3137void 3138vfs_msync(struct mount *mp, int flags) 3139{ 3140 struct vnode *vp, *nvp; 3141 struct vm_object *obj; 3142 int tries; 3143 3144 GIANT_REQUIRED; 3145 3146 tries = 5; 3147 MNT_ILOCK(mp); 3148loop: 3149 TAILQ_FOREACH_SAFE(vp, &mp->mnt_nvnodelist, v_nmntvnodes, nvp) { 3150 if (vp->v_mount != mp) { 3151 if (--tries > 0) 3152 goto loop; 3153 break; 3154 } 3155 3156 VI_LOCK(vp); 3157 if (vp->v_iflag & VI_XLOCK) { 3158 VI_UNLOCK(vp); 3159 continue; 3160 } 3161 3162 if ((vp->v_iflag & VI_OBJDIRTY) && 3163 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 3164 MNT_IUNLOCK(mp); 3165 if (!vget(vp, 3166 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3167 curthread)) { 3168 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3169 vput(vp); 3170 MNT_ILOCK(mp); 3171 continue; 3172 } 3173 3174 if (VOP_GETVOBJECT(vp, &obj) == 0) { 3175 VM_OBJECT_LOCK(obj); 3176 vm_object_page_clean(obj, 0, 0, 3177 flags == MNT_WAIT ? 3178 OBJPC_SYNC : OBJPC_NOSYNC); 3179 VM_OBJECT_UNLOCK(obj); 3180 } 3181 vput(vp); 3182 } 3183 MNT_ILOCK(mp); 3184 if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) { 3185 if (--tries > 0) 3186 goto loop; 3187 break; 3188 } 3189 } else 3190 VI_UNLOCK(vp); 3191 } 3192 MNT_IUNLOCK(mp); 3193} 3194 3195/* 3196 * Create the VM object needed for VMIO and mmap support. This 3197 * is done for all VREG files in the system. Some filesystems might 3198 * afford the additional metadata buffering capability of the 3199 * VMIO code by making the device node be VMIO mode also. 3200 * 3201 * vp must be locked when vfs_object_create is called. 3202 */ 3203int 3204vfs_object_create(vp, td, cred) 3205 struct vnode *vp; 3206 struct thread *td; 3207 struct ucred *cred; 3208{ 3209 3210 GIANT_REQUIRED; 3211 return (VOP_CREATEVOBJECT(vp, cred, td)); 3212} 3213 3214/* 3215 * Mark a vnode as free, putting it up for recycling. 3216 */ 3217void 3218vfree(vp) 3219 struct vnode *vp; 3220{ 3221 3222 ASSERT_VI_LOCKED(vp, "vfree"); 3223 mtx_lock(&vnode_free_list_mtx); 3224 KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free")); 3225 if (vp->v_iflag & VI_AGE) { 3226 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 3227 } else { 3228 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 3229 } 3230 freevnodes++; 3231 mtx_unlock(&vnode_free_list_mtx); 3232 vp->v_iflag &= ~VI_AGE; 3233 vp->v_iflag |= VI_FREE; 3234} 3235 3236/* 3237 * Opposite of vfree() - mark a vnode as in use. 3238 */ 3239void 3240vbusy(vp) 3241 struct vnode *vp; 3242{ 3243 3244 ASSERT_VI_LOCKED(vp, "vbusy"); 3245 KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free")); 3246 3247 mtx_lock(&vnode_free_list_mtx); 3248 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 3249 freevnodes--; 3250 mtx_unlock(&vnode_free_list_mtx); 3251 3252 vp->v_iflag &= ~(VI_FREE|VI_AGE); 3253} 3254 3255/* 3256 * Initalize per-vnode helper structure to hold poll-related state. 3257 */ 3258void 3259v_addpollinfo(struct vnode *vp) 3260{ 3261 3262 vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK); 3263 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3264} 3265 3266/* 3267 * Record a process's interest in events which might happen to 3268 * a vnode. Because poll uses the historic select-style interface 3269 * internally, this routine serves as both the ``check for any 3270 * pending events'' and the ``record my interest in future events'' 3271 * functions. (These are done together, while the lock is held, 3272 * to avoid race conditions.) 3273 */ 3274int 3275vn_pollrecord(vp, td, events) 3276 struct vnode *vp; 3277 struct thread *td; 3278 short events; 3279{ 3280 3281 if (vp->v_pollinfo == NULL) 3282 v_addpollinfo(vp); 3283 mtx_lock(&vp->v_pollinfo->vpi_lock); 3284 if (vp->v_pollinfo->vpi_revents & events) { 3285 /* 3286 * This leaves events we are not interested 3287 * in available for the other process which 3288 * which presumably had requested them 3289 * (otherwise they would never have been 3290 * recorded). 3291 */ 3292 events &= vp->v_pollinfo->vpi_revents; 3293 vp->v_pollinfo->vpi_revents &= ~events; 3294 3295 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3296 return events; 3297 } 3298 vp->v_pollinfo->vpi_events |= events; 3299 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3300 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3301 return 0; 3302} 3303 3304/* 3305 * Note the occurrence of an event. If the VN_POLLEVENT macro is used, 3306 * it is possible for us to miss an event due to race conditions, but 3307 * that condition is expected to be rare, so for the moment it is the 3308 * preferred interface. 3309 */ 3310void 3311vn_pollevent(vp, events) 3312 struct vnode *vp; 3313 short events; 3314{ 3315 3316 if (vp->v_pollinfo == NULL) 3317 v_addpollinfo(vp); 3318 mtx_lock(&vp->v_pollinfo->vpi_lock); 3319 if (vp->v_pollinfo->vpi_events & events) { 3320 /* 3321 * We clear vpi_events so that we don't 3322 * call selwakeup() twice if two events are 3323 * posted before the polling process(es) is 3324 * awakened. This also ensures that we take at 3325 * most one selwakeup() if the polling process 3326 * is no longer interested. However, it does 3327 * mean that only one event can be noticed at 3328 * a time. (Perhaps we should only clear those 3329 * event bits which we note?) XXX 3330 */ 3331 vp->v_pollinfo->vpi_events = 0; /* &= ~events ??? */ 3332 vp->v_pollinfo->vpi_revents |= events; 3333 selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO); 3334 } 3335 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3336} 3337 3338/* 3339 * Wake up anyone polling on vp because it is being revoked. 3340 * This depends on dead_poll() returning POLLHUP for correct 3341 * behavior. 3342 */ 3343void 3344vn_pollgone(vp) 3345 struct vnode *vp; 3346{ 3347 3348 mtx_lock(&vp->v_pollinfo->vpi_lock); 3349 VN_KNOTE(vp, NOTE_REVOKE); 3350 if (vp->v_pollinfo->vpi_events) { 3351 vp->v_pollinfo->vpi_events = 0; 3352 selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO); 3353 } 3354 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3355} 3356 3357 3358 3359/* 3360 * Routine to create and manage a filesystem syncer vnode. 3361 */ 3362#define sync_close ((int (*)(struct vop_close_args *))nullop) 3363static int sync_fsync(struct vop_fsync_args *); 3364static int sync_inactive(struct vop_inactive_args *); 3365static int sync_reclaim(struct vop_reclaim_args *); 3366 3367static vop_t **sync_vnodeop_p; 3368static struct vnodeopv_entry_desc sync_vnodeop_entries[] = { 3369 { &vop_default_desc, (vop_t *) vop_eopnotsupp }, 3370 { &vop_close_desc, (vop_t *) sync_close }, /* close */ 3371 { &vop_fsync_desc, (vop_t *) sync_fsync }, /* fsync */ 3372 { &vop_inactive_desc, (vop_t *) sync_inactive }, /* inactive */ 3373 { &vop_reclaim_desc, (vop_t *) sync_reclaim }, /* reclaim */ 3374 { &vop_lock_desc, (vop_t *) vop_stdlock }, /* lock */ 3375 { &vop_unlock_desc, (vop_t *) vop_stdunlock }, /* unlock */ 3376 { &vop_islocked_desc, (vop_t *) vop_stdislocked }, /* islocked */ 3377 { NULL, NULL } 3378}; 3379static struct vnodeopv_desc sync_vnodeop_opv_desc = 3380 { &sync_vnodeop_p, sync_vnodeop_entries }; 3381 3382VNODEOP_SET(sync_vnodeop_opv_desc); 3383 3384/* 3385 * Create a new filesystem syncer vnode for the specified mount point. 3386 */ 3387int 3388vfs_allocate_syncvnode(mp) 3389 struct mount *mp; 3390{ 3391 struct vnode *vp; 3392 static long start, incr, next; 3393 int error; 3394 3395 /* Allocate a new vnode */ 3396 if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) { 3397 mp->mnt_syncer = NULL; 3398 return (error); 3399 } 3400 vp->v_type = VNON; 3401 /* 3402 * Place the vnode onto the syncer worklist. We attempt to 3403 * scatter them about on the list so that they will go off 3404 * at evenly distributed times even if all the filesystems 3405 * are mounted at once. 3406 */ 3407 next += incr; 3408 if (next == 0 || next > syncer_maxdelay) { 3409 start /= 2; 3410 incr /= 2; 3411 if (start == 0) { 3412 start = syncer_maxdelay / 2; 3413 incr = syncer_maxdelay; 3414 } 3415 next = start; 3416 } 3417 VI_LOCK(vp); 3418 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0); 3419 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3420 mtx_lock(&sync_mtx); 3421 sync_vnode_count++; 3422 mtx_unlock(&sync_mtx); 3423 VI_UNLOCK(vp); 3424 mp->mnt_syncer = vp; 3425 return (0); 3426} 3427 3428/* 3429 * Do a lazy sync of the filesystem. 3430 */ 3431static int 3432sync_fsync(ap) 3433 struct vop_fsync_args /* { 3434 struct vnode *a_vp; 3435 struct ucred *a_cred; 3436 int a_waitfor; 3437 struct thread *a_td; 3438 } */ *ap; 3439{ 3440 struct vnode *syncvp = ap->a_vp; 3441 struct mount *mp = syncvp->v_mount; 3442 struct thread *td = ap->a_td; 3443 int error, asyncflag; 3444 3445 /* 3446 * We only need to do something if this is a lazy evaluation. 3447 */ 3448 if (ap->a_waitfor != MNT_LAZY) 3449 return (0); 3450 3451 /* 3452 * Move ourselves to the back of the sync list. 3453 */ 3454 VI_LOCK(syncvp); 3455 vn_syncer_add_to_worklist(syncvp, syncdelay); 3456 VI_UNLOCK(syncvp); 3457 3458 /* 3459 * Walk the list of vnodes pushing all that are dirty and 3460 * not already on the sync list. 3461 */ 3462 mtx_lock(&mountlist_mtx); 3463 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) { 3464 mtx_unlock(&mountlist_mtx); 3465 return (0); 3466 } 3467 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3468 vfs_unbusy(mp, td); 3469 return (0); 3470 } 3471 asyncflag = mp->mnt_flag & MNT_ASYNC; 3472 mp->mnt_flag &= ~MNT_ASYNC; 3473 vfs_msync(mp, MNT_NOWAIT); 3474 error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td); 3475 if (asyncflag) 3476 mp->mnt_flag |= MNT_ASYNC; 3477 vn_finished_write(mp); 3478 vfs_unbusy(mp, td); 3479 return (error); 3480} 3481 3482/* 3483 * The syncer vnode is no referenced. 3484 */ 3485static int 3486sync_inactive(ap) 3487 struct vop_inactive_args /* { 3488 struct vnode *a_vp; 3489 struct thread *a_td; 3490 } */ *ap; 3491{ 3492 3493 VOP_UNLOCK(ap->a_vp, 0, ap->a_td); 3494 vgone(ap->a_vp); 3495 return (0); 3496} 3497 3498/* 3499 * The syncer vnode is no longer needed and is being decommissioned. 3500 * 3501 * Modifications to the worklist must be protected by sync_mtx. 3502 */ 3503static int 3504sync_reclaim(ap) 3505 struct vop_reclaim_args /* { 3506 struct vnode *a_vp; 3507 } */ *ap; 3508{ 3509 struct vnode *vp = ap->a_vp; 3510 3511 VI_LOCK(vp); 3512 vp->v_mount->mnt_syncer = NULL; 3513 if (vp->v_iflag & VI_ONWORKLST) { 3514 mtx_lock(&sync_mtx); 3515 LIST_REMOVE(vp, v_synclist); 3516 syncer_worklist_len--; 3517 sync_vnode_count--; 3518 mtx_unlock(&sync_mtx); 3519 vp->v_iflag &= ~VI_ONWORKLST; 3520 } 3521 VI_UNLOCK(vp); 3522 3523 return (0); 3524} 3525 3526/* 3527 * extract the struct cdev *from a VCHR 3528 */ 3529struct cdev * 3530vn_todev(vp) 3531 struct vnode *vp; 3532{ 3533 3534 if (vp->v_type != VCHR) 3535 return (NULL); 3536 return (vp->v_rdev); 3537} 3538 3539/* 3540 * Check if vnode represents a disk device 3541 */ 3542int 3543vn_isdisk(vp, errp) 3544 struct vnode *vp; 3545 int *errp; 3546{ 3547 int error; 3548 3549 error = 0; 3550 if (vp->v_type != VCHR) 3551 error = ENOTBLK; 3552 else if (vp->v_rdev == NULL) 3553 error = ENXIO; 3554 else if (!(devsw(vp->v_rdev)->d_flags & D_DISK)) 3555 error = ENOTBLK; 3556 if (errp != NULL) 3557 *errp = error; 3558 return (error == 0); 3559} 3560 3561/* 3562 * Free data allocated by namei(); see namei(9) for details. 3563 */ 3564void 3565NDFREE(ndp, flags) 3566 struct nameidata *ndp; 3567 const u_int flags; 3568{ 3569 3570 if (!(flags & NDF_NO_FREE_PNBUF) && 3571 (ndp->ni_cnd.cn_flags & HASBUF)) { 3572 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf); 3573 ndp->ni_cnd.cn_flags &= ~HASBUF; 3574 } 3575 if (!(flags & NDF_NO_DVP_UNLOCK) && 3576 (ndp->ni_cnd.cn_flags & LOCKPARENT) && 3577 ndp->ni_dvp != ndp->ni_vp) 3578 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread); 3579 if (!(flags & NDF_NO_DVP_RELE) && 3580 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) { 3581 vrele(ndp->ni_dvp); 3582 ndp->ni_dvp = NULL; 3583 } 3584 if (!(flags & NDF_NO_VP_UNLOCK) && 3585 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp) 3586 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread); 3587 if (!(flags & NDF_NO_VP_RELE) && 3588 ndp->ni_vp) { 3589 vrele(ndp->ni_vp); 3590 ndp->ni_vp = NULL; 3591 } 3592 if (!(flags & NDF_NO_STARTDIR_RELE) && 3593 (ndp->ni_cnd.cn_flags & SAVESTART)) { 3594 vrele(ndp->ni_startdir); 3595 ndp->ni_startdir = NULL; 3596 } 3597} 3598 3599/* 3600 * Common filesystem object access control check routine. Accepts a 3601 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3602 * and optional call-by-reference privused argument allowing vaccess() 3603 * to indicate to the caller whether privilege was used to satisfy the 3604 * request (obsoleted). Returns 0 on success, or an errno on failure. 3605 */ 3606int 3607vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused) 3608 enum vtype type; 3609 mode_t file_mode; 3610 uid_t file_uid; 3611 gid_t file_gid; 3612 mode_t acc_mode; 3613 struct ucred *cred; 3614 int *privused; 3615{ 3616 mode_t dac_granted; 3617#ifdef CAPABILITIES 3618 mode_t cap_granted; 3619#endif 3620 3621 /* 3622 * Look for a normal, non-privileged way to access the file/directory 3623 * as requested. If it exists, go with that. 3624 */ 3625 3626 if (privused != NULL) 3627 *privused = 0; 3628 3629 dac_granted = 0; 3630 3631 /* Check the owner. */ 3632 if (cred->cr_uid == file_uid) { 3633 dac_granted |= VADMIN; 3634 if (file_mode & S_IXUSR) 3635 dac_granted |= VEXEC; 3636 if (file_mode & S_IRUSR) 3637 dac_granted |= VREAD; 3638 if (file_mode & S_IWUSR) 3639 dac_granted |= (VWRITE | VAPPEND); 3640 3641 if ((acc_mode & dac_granted) == acc_mode) 3642 return (0); 3643 3644 goto privcheck; 3645 } 3646 3647 /* Otherwise, check the groups (first match) */ 3648 if (groupmember(file_gid, cred)) { 3649 if (file_mode & S_IXGRP) 3650 dac_granted |= VEXEC; 3651 if (file_mode & S_IRGRP) 3652 dac_granted |= VREAD; 3653 if (file_mode & S_IWGRP) 3654 dac_granted |= (VWRITE | VAPPEND); 3655 3656 if ((acc_mode & dac_granted) == acc_mode) 3657 return (0); 3658 3659 goto privcheck; 3660 } 3661 3662 /* Otherwise, check everyone else. */ 3663 if (file_mode & S_IXOTH) 3664 dac_granted |= VEXEC; 3665 if (file_mode & S_IROTH) 3666 dac_granted |= VREAD; 3667 if (file_mode & S_IWOTH) 3668 dac_granted |= (VWRITE | VAPPEND); 3669 if ((acc_mode & dac_granted) == acc_mode) 3670 return (0); 3671 3672privcheck: 3673 if (!suser_cred(cred, PRISON_ROOT)) { 3674 /* XXX audit: privilege used */ 3675 if (privused != NULL) 3676 *privused = 1; 3677 return (0); 3678 } 3679 3680#ifdef CAPABILITIES 3681 /* 3682 * Build a capability mask to determine if the set of capabilities 3683 * satisfies the requirements when combined with the granted mask 3684 * from above. 3685 * For each capability, if the capability is required, bitwise 3686 * or the request type onto the cap_granted mask. 3687 */ 3688 cap_granted = 0; 3689 3690 if (type == VDIR) { 3691 /* 3692 * For directories, use CAP_DAC_READ_SEARCH to satisfy 3693 * VEXEC requests, instead of CAP_DAC_EXECUTE. 3694 */ 3695 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3696 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT)) 3697 cap_granted |= VEXEC; 3698 } else { 3699 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3700 !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT)) 3701 cap_granted |= VEXEC; 3702 } 3703 3704 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) && 3705 !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT)) 3706 cap_granted |= VREAD; 3707 3708 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3709 !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT)) 3710 cap_granted |= (VWRITE | VAPPEND); 3711 3712 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3713 !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT)) 3714 cap_granted |= VADMIN; 3715 3716 if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) { 3717 /* XXX audit: privilege used */ 3718 if (privused != NULL) 3719 *privused = 1; 3720 return (0); 3721 } 3722#endif 3723 3724 return ((acc_mode & VADMIN) ? EPERM : EACCES); 3725} 3726 3727/* 3728 * Credential check based on process requesting service, and per-attribute 3729 * permissions. 3730 */ 3731int 3732extattr_check_cred(struct vnode *vp, int attrnamespace, 3733 struct ucred *cred, struct thread *td, int access) 3734{ 3735 3736 /* 3737 * Kernel-invoked always succeeds. 3738 */ 3739 if (cred == NOCRED) 3740 return (0); 3741 3742 /* 3743 * Do not allow privileged processes in jail to directly 3744 * manipulate system attributes. 3745 * 3746 * XXX What capability should apply here? 3747 * Probably CAP_SYS_SETFFLAG. 3748 */ 3749 switch (attrnamespace) { 3750 case EXTATTR_NAMESPACE_SYSTEM: 3751 /* Potentially should be: return (EPERM); */ 3752 return (suser_cred(cred, 0)); 3753 case EXTATTR_NAMESPACE_USER: 3754 return (VOP_ACCESS(vp, access, cred, td)); 3755 default: 3756 return (EPERM); 3757 } 3758} 3759 3760#ifdef DEBUG_VFS_LOCKS 3761/* 3762 * This only exists to supress warnings from unlocked specfs accesses. It is 3763 * no longer ok to have an unlocked VFS. 3764 */ 3765#define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD) 3766 3767int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3768int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3769int vfs_badlock_print = 1; /* Print lock violations. */ 3770 3771static void 3772vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3773{ 3774 3775 if (vfs_badlock_print) 3776 printf("%s: %p %s\n", str, (void *)vp, msg); 3777 if (vfs_badlock_ddb) 3778 Debugger("lock violation"); 3779} 3780 3781void 3782assert_vi_locked(struct vnode *vp, const char *str) 3783{ 3784 3785 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3786 vfs_badlock("interlock is not locked but should be", str, vp); 3787} 3788 3789void 3790assert_vi_unlocked(struct vnode *vp, const char *str) 3791{ 3792 3793 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3794 vfs_badlock("interlock is locked but should not be", str, vp); 3795} 3796 3797void 3798assert_vop_locked(struct vnode *vp, const char *str) 3799{ 3800 3801 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0) 3802 vfs_badlock("is not locked but should be", str, vp); 3803} 3804 3805void 3806assert_vop_unlocked(struct vnode *vp, const char *str) 3807{ 3808 3809 if (vp && !IGNORE_LOCK(vp) && 3810 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) 3811 vfs_badlock("is locked but should not be", str, vp); 3812} 3813 3814#if 0 3815void 3816assert_vop_elocked(struct vnode *vp, const char *str) 3817{ 3818 3819 if (vp && !IGNORE_LOCK(vp) && 3820 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE) 3821 vfs_badlock("is not exclusive locked but should be", str, vp); 3822} 3823 3824void 3825assert_vop_elocked_other(struct vnode *vp, const char *str) 3826{ 3827 3828 if (vp && !IGNORE_LOCK(vp) && 3829 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER) 3830 vfs_badlock("is not exclusive locked by another thread", 3831 str, vp); 3832} 3833 3834void 3835assert_vop_slocked(struct vnode *vp, const char *str) 3836{ 3837 3838 if (vp && !IGNORE_LOCK(vp) && 3839 VOP_ISLOCKED(vp, curthread) != LK_SHARED) 3840 vfs_badlock("is not locked shared but should be", str, vp); 3841} 3842#endif /* 0 */ 3843 3844void 3845vop_rename_pre(void *ap) 3846{ 3847 struct vop_rename_args *a = ap; 3848 3849 if (a->a_tvp) 3850 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3851 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3852 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3853 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3854 3855 /* Check the source (from). */ 3856 if (a->a_tdvp != a->a_fdvp) 3857 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3858 if (a->a_tvp != a->a_fvp) 3859 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked"); 3860 3861 /* Check the target. */ 3862 if (a->a_tvp) 3863 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3864 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3865} 3866 3867void 3868vop_strategy_pre(void *ap) 3869{ 3870 struct vop_strategy_args *a; 3871 struct buf *bp; 3872 3873 a = ap; 3874 bp = a->a_bp; 3875 3876 /* 3877 * Cluster ops lock their component buffers but not the IO container. 3878 */ 3879 if ((bp->b_flags & B_CLUSTER) != 0) 3880 return; 3881 3882 if (BUF_REFCNT(bp) < 1) { 3883 if (vfs_badlock_print) 3884 printf( 3885 "VOP_STRATEGY: bp is not locked but should be\n"); 3886 if (vfs_badlock_ddb) 3887 Debugger("lock violation"); 3888 } 3889} 3890 3891void 3892vop_lookup_pre(void *ap) 3893{ 3894 struct vop_lookup_args *a; 3895 struct vnode *dvp; 3896 3897 a = ap; 3898 dvp = a->a_dvp; 3899 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3900 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3901} 3902 3903void 3904vop_lookup_post(void *ap, int rc) 3905{ 3906 struct vop_lookup_args *a; 3907 struct componentname *cnp; 3908 struct vnode *dvp; 3909 struct vnode *vp; 3910 int flags; 3911 3912 a = ap; 3913 dvp = a->a_dvp; 3914 cnp = a->a_cnp; 3915 vp = *(a->a_vpp); 3916 flags = cnp->cn_flags; 3917 3918 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3919 3920 /* 3921 * If this is the last path component for this lookup and LOCKPARENT 3922 * is set, OR if there is an error the directory has to be locked. 3923 */ 3924 if ((flags & LOCKPARENT) && (flags & ISLASTCN)) 3925 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)"); 3926 else if (rc != 0) 3927 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)"); 3928 else if (dvp != vp) 3929 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)"); 3930 if (flags & PDIRUNLOCK) 3931 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)"); 3932} 3933 3934void 3935vop_lock_pre(void *ap) 3936{ 3937 struct vop_lock_args *a = ap; 3938 3939 if ((a->a_flags & LK_INTERLOCK) == 0) 3940 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3941 else 3942 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3943} 3944 3945void 3946vop_lock_post(void *ap, int rc) 3947{ 3948 struct vop_lock_args *a = ap; 3949 3950 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3951 if (rc == 0) 3952 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3953} 3954 3955void 3956vop_unlock_pre(void *ap) 3957{ 3958 struct vop_unlock_args *a = ap; 3959 3960 if (a->a_flags & LK_INTERLOCK) 3961 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3962 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3963} 3964 3965void 3966vop_unlock_post(void *ap, int rc) 3967{ 3968 struct vop_unlock_args *a = ap; 3969 3970 if (a->a_flags & LK_INTERLOCK) 3971 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3972} 3973#endif /* DEBUG_VFS_LOCKS */ 3974 3975static struct klist fs_klist = SLIST_HEAD_INITIALIZER(&fs_klist); 3976 3977void 3978vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused) 3979{ 3980 3981 KNOTE(&fs_klist, event); 3982} 3983 3984static int filt_fsattach(struct knote *kn); 3985static void filt_fsdetach(struct knote *kn); 3986static int filt_fsevent(struct knote *kn, long hint); 3987 3988struct filterops fs_filtops = 3989 { 0, filt_fsattach, filt_fsdetach, filt_fsevent }; 3990 3991static int 3992filt_fsattach(struct knote *kn) 3993{ 3994 3995 kn->kn_flags |= EV_CLEAR; 3996 SLIST_INSERT_HEAD(&fs_klist, kn, kn_selnext); 3997 return (0); 3998} 3999 4000static void 4001filt_fsdetach(struct knote *kn) 4002{ 4003 4004 SLIST_REMOVE(&fs_klist, kn, knote, kn_selnext); 4005} 4006 4007static int 4008filt_fsevent(struct knote *kn, long hint) 4009{ 4010 4011 kn->kn_fflags |= hint; 4012 return (kn->kn_fflags != 0); 4013} 4014 4015static int 4016sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4017{ 4018 struct vfsidctl vc; 4019 int error; 4020 struct mount *mp; 4021 4022 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4023 if (error) 4024 return (error); 4025 if (vc.vc_vers != VFS_CTL_VERS1) 4026 return (EINVAL); 4027 mp = vfs_getvfs(&vc.vc_fsid); 4028 if (mp == NULL) 4029 return (ENOENT); 4030 /* ensure that a specific sysctl goes to the right filesystem. */ 4031 if (strcmp(vc.vc_fstypename, "*") != 0 && 4032 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4033 return (EINVAL); 4034 } 4035 VCTLTOREQ(&vc, req); 4036 return (VFS_SYSCTL(mp, vc.vc_op, req)); 4037} 4038 4039SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_RD, 4040 NULL, 0, sysctl_vfs_ctl, "", "Message queue IDs"); 4041