vfs_subr.c revision 184554
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 184554 2008-11-02 10:15:42Z attilio $"); 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/condvar.h> 52#include <sys/conf.h> 53#include <sys/dirent.h> 54#include <sys/event.h> 55#include <sys/eventhandler.h> 56#include <sys/extattr.h> 57#include <sys/file.h> 58#include <sys/fcntl.h> 59#include <sys/jail.h> 60#include <sys/kdb.h> 61#include <sys/kernel.h> 62#include <sys/kthread.h> 63#include <sys/lockf.h> 64#include <sys/malloc.h> 65#include <sys/mount.h> 66#include <sys/namei.h> 67#include <sys/priv.h> 68#include <sys/reboot.h> 69#include <sys/sleepqueue.h> 70#include <sys/stat.h> 71#include <sys/sysctl.h> 72#include <sys/syslog.h> 73#include <sys/vmmeter.h> 74#include <sys/vnode.h> 75 76#include <machine/stdarg.h> 77 78#include <security/mac/mac_framework.h> 79 80#include <vm/vm.h> 81#include <vm/vm_object.h> 82#include <vm/vm_extern.h> 83#include <vm/pmap.h> 84#include <vm/vm_map.h> 85#include <vm/vm_page.h> 86#include <vm/vm_kern.h> 87#include <vm/uma.h> 88 89#ifdef DDB 90#include <ddb/ddb.h> 91#endif 92 93#define WI_MPSAFEQ 0 94#define WI_GIANTQ 1 95 96static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure"); 97 98static void delmntque(struct vnode *vp); 99static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 100 int slpflag, int slptimeo); 101static void syncer_shutdown(void *arg, int howto); 102static int vtryrecycle(struct vnode *vp); 103static void vbusy(struct vnode *vp); 104static void vinactive(struct vnode *, struct thread *); 105static void v_incr_usecount(struct vnode *); 106static void v_decr_usecount(struct vnode *); 107static void v_decr_useonly(struct vnode *); 108static void v_upgrade_usecount(struct vnode *); 109static void vfree(struct vnode *); 110static void vnlru_free(int); 111static void vdestroy(struct vnode *); 112static void vgonel(struct vnode *); 113static void vfs_knllock(void *arg); 114static void vfs_knlunlock(void *arg); 115static int vfs_knllocked(void *arg); 116static void destroy_vpollinfo(struct vpollinfo *vi); 117 118/* 119 * Enable Giant pushdown based on whether or not the vm is mpsafe in this 120 * build. Without mpsafevm the buffer cache can not run Giant free. 121 */ 122int mpsafe_vfs = 1; 123TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs); 124SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0, 125 "MPSAFE VFS"); 126 127/* 128 * Number of vnodes in existence. Increased whenever getnewvnode() 129 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed 130 * vnode. 131 */ 132static unsigned long numvnodes; 133 134SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 135 136/* 137 * Conversion tables for conversion from vnode types to inode formats 138 * and back. 139 */ 140enum vtype iftovt_tab[16] = { 141 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 142 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 143}; 144int vttoif_tab[10] = { 145 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 146 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 147}; 148 149/* 150 * List of vnodes that are ready for recycling. 151 */ 152static TAILQ_HEAD(freelst, vnode) vnode_free_list; 153 154/* 155 * Free vnode target. Free vnodes may simply be files which have been stat'd 156 * but not read. This is somewhat common, and a small cache of such files 157 * should be kept to avoid recreation costs. 158 */ 159static u_long wantfreevnodes; 160SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 161/* Number of vnodes in the free list. */ 162static u_long freevnodes; 163SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 164 165/* 166 * Various variables used for debugging the new implementation of 167 * reassignbuf(). 168 * XXX these are probably of (very) limited utility now. 169 */ 170static int reassignbufcalls; 171SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 172 173/* 174 * Cache for the mount type id assigned to NFS. This is used for 175 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 176 */ 177int nfs_mount_type = -1; 178 179/* To keep more than one thread at a time from running vfs_getnewfsid */ 180static struct mtx mntid_mtx; 181 182/* 183 * Lock for any access to the following: 184 * vnode_free_list 185 * numvnodes 186 * freevnodes 187 */ 188static struct mtx vnode_free_list_mtx; 189 190/* Publicly exported FS */ 191struct nfs_public nfs_pub; 192 193/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 194static uma_zone_t vnode_zone; 195static uma_zone_t vnodepoll_zone; 196 197/* Set to 1 to print out reclaim of active vnodes */ 198int prtactive; 199 200/* 201 * The workitem queue. 202 * 203 * It is useful to delay writes of file data and filesystem metadata 204 * for tens of seconds so that quickly created and deleted files need 205 * not waste disk bandwidth being created and removed. To realize this, 206 * we append vnodes to a "workitem" queue. When running with a soft 207 * updates implementation, most pending metadata dependencies should 208 * not wait for more than a few seconds. Thus, mounted on block devices 209 * are delayed only about a half the time that file data is delayed. 210 * Similarly, directory updates are more critical, so are only delayed 211 * about a third the time that file data is delayed. Thus, there are 212 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 213 * one each second (driven off the filesystem syncer process). The 214 * syncer_delayno variable indicates the next queue that is to be processed. 215 * Items that need to be processed soon are placed in this queue: 216 * 217 * syncer_workitem_pending[syncer_delayno] 218 * 219 * A delay of fifteen seconds is done by placing the request fifteen 220 * entries later in the queue: 221 * 222 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 223 * 224 */ 225static int syncer_delayno; 226static long syncer_mask; 227LIST_HEAD(synclist, bufobj); 228static struct synclist *syncer_workitem_pending[2]; 229/* 230 * The sync_mtx protects: 231 * bo->bo_synclist 232 * sync_vnode_count 233 * syncer_delayno 234 * syncer_state 235 * syncer_workitem_pending 236 * syncer_worklist_len 237 * rushjob 238 */ 239static struct mtx sync_mtx; 240static struct cv sync_wakeup; 241 242#define SYNCER_MAXDELAY 32 243static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 244static int syncdelay = 30; /* max time to delay syncing data */ 245static int filedelay = 30; /* time to delay syncing files */ 246SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 247static int dirdelay = 29; /* time to delay syncing directories */ 248SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 249static int metadelay = 28; /* time to delay syncing metadata */ 250SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 251static int rushjob; /* number of slots to run ASAP */ 252static int stat_rush_requests; /* number of times I/O speeded up */ 253SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 254 255/* 256 * When shutting down the syncer, run it at four times normal speed. 257 */ 258#define SYNCER_SHUTDOWN_SPEEDUP 4 259static int sync_vnode_count; 260static int syncer_worklist_len; 261static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 262 syncer_state; 263 264/* 265 * Number of vnodes we want to exist at any one time. This is mostly used 266 * to size hash tables in vnode-related code. It is normally not used in 267 * getnewvnode(), as wantfreevnodes is normally nonzero.) 268 * 269 * XXX desiredvnodes is historical cruft and should not exist. 270 */ 271int desiredvnodes; 272SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 273 &desiredvnodes, 0, "Maximum number of vnodes"); 274SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 275 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 276static int vnlru_nowhere; 277SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 278 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 279 280/* 281 * Macros to control when a vnode is freed and recycled. All require 282 * the vnode interlock. 283 */ 284#define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 285#define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 286#define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 287 288 289/* 290 * Initialize the vnode management data structures. 291 */ 292#ifndef MAXVNODES_MAX 293#define MAXVNODES_MAX 100000 294#endif 295static void 296vntblinit(void *dummy __unused) 297{ 298 299 /* 300 * Desiredvnodes is a function of the physical memory size and 301 * the kernel's heap size. Specifically, desiredvnodes scales 302 * in proportion to the physical memory size until two fifths 303 * of the kernel's heap size is consumed by vnodes and vm 304 * objects. 305 */ 306 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size / 307 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 308 if (desiredvnodes > MAXVNODES_MAX) { 309 if (bootverbose) 310 printf("Reducing kern.maxvnodes %d -> %d\n", 311 desiredvnodes, MAXVNODES_MAX); 312 desiredvnodes = MAXVNODES_MAX; 313 } 314 wantfreevnodes = desiredvnodes / 4; 315 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 316 TAILQ_INIT(&vnode_free_list); 317 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 318 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 319 NULL, NULL, UMA_ALIGN_PTR, 0); 320 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 321 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 322 /* 323 * Initialize the filesystem syncer. 324 */ 325 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE, 326 &syncer_mask); 327 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE, 328 &syncer_mask); 329 syncer_maxdelay = syncer_mask + 1; 330 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 331 cv_init(&sync_wakeup, "syncer"); 332} 333SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 334 335 336/* 337 * Mark a mount point as busy. Used to synchronize access and to delay 338 * unmounting. Eventually, mountlist_mtx is not released on failure. 339 */ 340int 341vfs_busy(struct mount *mp, int flags) 342{ 343 344 MPASS((flags & ~MBF_MASK) == 0); 345 346 MNT_ILOCK(mp); 347 MNT_REF(mp); 348 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 349 if (flags & MBF_NOWAIT) { 350 MNT_REL(mp); 351 MNT_IUNLOCK(mp); 352 return (ENOENT); 353 } 354 if (flags & MBF_MNTLSTLOCK) 355 mtx_unlock(&mountlist_mtx); 356 mp->mnt_kern_flag |= MNTK_MWAIT; 357 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0); 358 MNT_REL(mp); 359 MNT_IUNLOCK(mp); 360 if (flags & MBF_MNTLSTLOCK) 361 mtx_lock(&mountlist_mtx); 362 return (ENOENT); 363 } 364 if (flags & MBF_MNTLSTLOCK) 365 mtx_unlock(&mountlist_mtx); 366 mp->mnt_lockref++; 367 MNT_IUNLOCK(mp); 368 return (0); 369} 370 371/* 372 * Free a busy filesystem. 373 */ 374void 375vfs_unbusy(struct mount *mp) 376{ 377 378 MNT_ILOCK(mp); 379 MNT_REL(mp); 380 mp->mnt_lockref--; 381 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 382 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 383 mp->mnt_kern_flag &= ~MNTK_DRAINING; 384 wakeup(&mp->mnt_lockref); 385 } 386 MNT_IUNLOCK(mp); 387} 388 389/* 390 * Lookup a mount point by filesystem identifier. 391 */ 392struct mount * 393vfs_getvfs(fsid_t *fsid) 394{ 395 struct mount *mp; 396 397 mtx_lock(&mountlist_mtx); 398 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 399 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 400 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 401 vfs_ref(mp); 402 mtx_unlock(&mountlist_mtx); 403 return (mp); 404 } 405 } 406 mtx_unlock(&mountlist_mtx); 407 return ((struct mount *) 0); 408} 409 410/* 411 * Check if a user can access privileged mount options. 412 */ 413int 414vfs_suser(struct mount *mp, struct thread *td) 415{ 416 int error; 417 418 /* 419 * If the thread is jailed, but this is not a jail-friendly file 420 * system, deny immediately. 421 */ 422 if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL)) 423 return (EPERM); 424 425 /* 426 * If the file system was mounted outside a jail and a jailed thread 427 * tries to access it, deny immediately. 428 */ 429 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred)) 430 return (EPERM); 431 432 /* 433 * If the file system was mounted inside different jail that the jail of 434 * the calling thread, deny immediately. 435 */ 436 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) && 437 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) { 438 return (EPERM); 439 } 440 441 if ((mp->mnt_flag & MNT_USER) == 0 || 442 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 443 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 444 return (error); 445 } 446 return (0); 447} 448 449/* 450 * Get a new unique fsid. Try to make its val[0] unique, since this value 451 * will be used to create fake device numbers for stat(). Also try (but 452 * not so hard) make its val[0] unique mod 2^16, since some emulators only 453 * support 16-bit device numbers. We end up with unique val[0]'s for the 454 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 455 * 456 * Keep in mind that several mounts may be running in parallel. Starting 457 * the search one past where the previous search terminated is both a 458 * micro-optimization and a defense against returning the same fsid to 459 * different mounts. 460 */ 461void 462vfs_getnewfsid(struct mount *mp) 463{ 464 static u_int16_t mntid_base; 465 struct mount *nmp; 466 fsid_t tfsid; 467 int mtype; 468 469 mtx_lock(&mntid_mtx); 470 mtype = mp->mnt_vfc->vfc_typenum; 471 tfsid.val[1] = mtype; 472 mtype = (mtype & 0xFF) << 24; 473 for (;;) { 474 tfsid.val[0] = makedev(255, 475 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 476 mntid_base++; 477 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 478 break; 479 vfs_rel(nmp); 480 } 481 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 482 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 483 mtx_unlock(&mntid_mtx); 484} 485 486/* 487 * Knob to control the precision of file timestamps: 488 * 489 * 0 = seconds only; nanoseconds zeroed. 490 * 1 = seconds and nanoseconds, accurate within 1/HZ. 491 * 2 = seconds and nanoseconds, truncated to microseconds. 492 * >=3 = seconds and nanoseconds, maximum precision. 493 */ 494enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 495 496static int timestamp_precision = TSP_SEC; 497SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 498 ×tamp_precision, 0, ""); 499 500/* 501 * Get a current timestamp. 502 */ 503void 504vfs_timestamp(struct timespec *tsp) 505{ 506 struct timeval tv; 507 508 switch (timestamp_precision) { 509 case TSP_SEC: 510 tsp->tv_sec = time_second; 511 tsp->tv_nsec = 0; 512 break; 513 case TSP_HZ: 514 getnanotime(tsp); 515 break; 516 case TSP_USEC: 517 microtime(&tv); 518 TIMEVAL_TO_TIMESPEC(&tv, tsp); 519 break; 520 case TSP_NSEC: 521 default: 522 nanotime(tsp); 523 break; 524 } 525} 526 527/* 528 * Set vnode attributes to VNOVAL 529 */ 530void 531vattr_null(struct vattr *vap) 532{ 533 534 vap->va_type = VNON; 535 vap->va_size = VNOVAL; 536 vap->va_bytes = VNOVAL; 537 vap->va_mode = VNOVAL; 538 vap->va_nlink = VNOVAL; 539 vap->va_uid = VNOVAL; 540 vap->va_gid = VNOVAL; 541 vap->va_fsid = VNOVAL; 542 vap->va_fileid = VNOVAL; 543 vap->va_blocksize = VNOVAL; 544 vap->va_rdev = VNOVAL; 545 vap->va_atime.tv_sec = VNOVAL; 546 vap->va_atime.tv_nsec = VNOVAL; 547 vap->va_mtime.tv_sec = VNOVAL; 548 vap->va_mtime.tv_nsec = VNOVAL; 549 vap->va_ctime.tv_sec = VNOVAL; 550 vap->va_ctime.tv_nsec = VNOVAL; 551 vap->va_birthtime.tv_sec = VNOVAL; 552 vap->va_birthtime.tv_nsec = VNOVAL; 553 vap->va_flags = VNOVAL; 554 vap->va_gen = VNOVAL; 555 vap->va_vaflags = 0; 556} 557 558/* 559 * This routine is called when we have too many vnodes. It attempts 560 * to free <count> vnodes and will potentially free vnodes that still 561 * have VM backing store (VM backing store is typically the cause 562 * of a vnode blowout so we want to do this). Therefore, this operation 563 * is not considered cheap. 564 * 565 * A number of conditions may prevent a vnode from being reclaimed. 566 * the buffer cache may have references on the vnode, a directory 567 * vnode may still have references due to the namei cache representing 568 * underlying files, or the vnode may be in active use. It is not 569 * desireable to reuse such vnodes. These conditions may cause the 570 * number of vnodes to reach some minimum value regardless of what 571 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 572 */ 573static int 574vlrureclaim(struct mount *mp) 575{ 576 struct vnode *vp; 577 int done; 578 int trigger; 579 int usevnodes; 580 int count; 581 582 /* 583 * Calculate the trigger point, don't allow user 584 * screwups to blow us up. This prevents us from 585 * recycling vnodes with lots of resident pages. We 586 * aren't trying to free memory, we are trying to 587 * free vnodes. 588 */ 589 usevnodes = desiredvnodes; 590 if (usevnodes <= 0) 591 usevnodes = 1; 592 trigger = cnt.v_page_count * 2 / usevnodes; 593 done = 0; 594 vn_start_write(NULL, &mp, V_WAIT); 595 MNT_ILOCK(mp); 596 count = mp->mnt_nvnodelistsize / 10 + 1; 597 while (count != 0) { 598 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 599 while (vp != NULL && vp->v_type == VMARKER) 600 vp = TAILQ_NEXT(vp, v_nmntvnodes); 601 if (vp == NULL) 602 break; 603 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 604 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 605 --count; 606 if (!VI_TRYLOCK(vp)) 607 goto next_iter; 608 /* 609 * If it's been deconstructed already, it's still 610 * referenced, or it exceeds the trigger, skip it. 611 */ 612 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) || 613 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 614 vp->v_object->resident_page_count > trigger)) { 615 VI_UNLOCK(vp); 616 goto next_iter; 617 } 618 MNT_IUNLOCK(mp); 619 vholdl(vp); 620 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 621 vdrop(vp); 622 goto next_iter_mntunlocked; 623 } 624 VI_LOCK(vp); 625 /* 626 * v_usecount may have been bumped after VOP_LOCK() dropped 627 * the vnode interlock and before it was locked again. 628 * 629 * It is not necessary to recheck VI_DOOMED because it can 630 * only be set by another thread that holds both the vnode 631 * lock and vnode interlock. If another thread has the 632 * vnode lock before we get to VOP_LOCK() and obtains the 633 * vnode interlock after VOP_LOCK() drops the vnode 634 * interlock, the other thread will be unable to drop the 635 * vnode lock before our VOP_LOCK() call fails. 636 */ 637 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) || 638 (vp->v_object != NULL && 639 vp->v_object->resident_page_count > trigger)) { 640 VOP_UNLOCK(vp, LK_INTERLOCK); 641 goto next_iter_mntunlocked; 642 } 643 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 644 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 645 vgonel(vp); 646 VOP_UNLOCK(vp, 0); 647 vdropl(vp); 648 done++; 649next_iter_mntunlocked: 650 if ((count % 256) != 0) 651 goto relock_mnt; 652 goto yield; 653next_iter: 654 if ((count % 256) != 0) 655 continue; 656 MNT_IUNLOCK(mp); 657yield: 658 uio_yield(); 659relock_mnt: 660 MNT_ILOCK(mp); 661 } 662 MNT_IUNLOCK(mp); 663 vn_finished_write(mp); 664 return done; 665} 666 667/* 668 * Attempt to keep the free list at wantfreevnodes length. 669 */ 670static void 671vnlru_free(int count) 672{ 673 struct vnode *vp; 674 int vfslocked; 675 676 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 677 for (; count > 0; count--) { 678 vp = TAILQ_FIRST(&vnode_free_list); 679 /* 680 * The list can be modified while the free_list_mtx 681 * has been dropped and vp could be NULL here. 682 */ 683 if (!vp) 684 break; 685 VNASSERT(vp->v_op != NULL, vp, 686 ("vnlru_free: vnode already reclaimed.")); 687 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 688 /* 689 * Don't recycle if we can't get the interlock. 690 */ 691 if (!VI_TRYLOCK(vp)) { 692 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 693 continue; 694 } 695 VNASSERT(VCANRECYCLE(vp), vp, 696 ("vp inconsistent on freelist")); 697 freevnodes--; 698 vp->v_iflag &= ~VI_FREE; 699 vholdl(vp); 700 mtx_unlock(&vnode_free_list_mtx); 701 VI_UNLOCK(vp); 702 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 703 vtryrecycle(vp); 704 VFS_UNLOCK_GIANT(vfslocked); 705 /* 706 * If the recycled succeeded this vdrop will actually free 707 * the vnode. If not it will simply place it back on 708 * the free list. 709 */ 710 vdrop(vp); 711 mtx_lock(&vnode_free_list_mtx); 712 } 713} 714/* 715 * Attempt to recycle vnodes in a context that is always safe to block. 716 * Calling vlrurecycle() from the bowels of filesystem code has some 717 * interesting deadlock problems. 718 */ 719static struct proc *vnlruproc; 720static int vnlruproc_sig; 721 722static void 723vnlru_proc(void) 724{ 725 struct mount *mp, *nmp; 726 int done; 727 struct proc *p = vnlruproc; 728 729 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 730 SHUTDOWN_PRI_FIRST); 731 732 mtx_lock(&Giant); 733 734 for (;;) { 735 kproc_suspend_check(p); 736 mtx_lock(&vnode_free_list_mtx); 737 if (freevnodes > wantfreevnodes) 738 vnlru_free(freevnodes - wantfreevnodes); 739 if (numvnodes <= desiredvnodes * 9 / 10) { 740 vnlruproc_sig = 0; 741 wakeup(&vnlruproc_sig); 742 msleep(vnlruproc, &vnode_free_list_mtx, 743 PVFS|PDROP, "vlruwt", hz); 744 continue; 745 } 746 mtx_unlock(&vnode_free_list_mtx); 747 done = 0; 748 mtx_lock(&mountlist_mtx); 749 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 750 int vfsunlocked; 751 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 752 nmp = TAILQ_NEXT(mp, mnt_list); 753 continue; 754 } 755 if (!VFS_NEEDSGIANT(mp)) { 756 mtx_unlock(&Giant); 757 vfsunlocked = 1; 758 } else 759 vfsunlocked = 0; 760 done += vlrureclaim(mp); 761 if (vfsunlocked) 762 mtx_lock(&Giant); 763 mtx_lock(&mountlist_mtx); 764 nmp = TAILQ_NEXT(mp, mnt_list); 765 vfs_unbusy(mp); 766 } 767 mtx_unlock(&mountlist_mtx); 768 if (done == 0) { 769 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10); 770#if 0 771 /* These messages are temporary debugging aids */ 772 if (vnlru_nowhere < 5) 773 printf("vnlru process getting nowhere..\n"); 774 else if (vnlru_nowhere == 5) 775 printf("vnlru process messages stopped.\n"); 776#endif 777 vnlru_nowhere++; 778 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 779 } else 780 uio_yield(); 781 } 782} 783 784static struct kproc_desc vnlru_kp = { 785 "vnlru", 786 vnlru_proc, 787 &vnlruproc 788}; 789SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 790 &vnlru_kp); 791 792/* 793 * Routines having to do with the management of the vnode table. 794 */ 795 796static void 797vdestroy(struct vnode *vp) 798{ 799 struct bufobj *bo; 800 801 CTR1(KTR_VFS, "vdestroy vp %p", vp); 802 mtx_lock(&vnode_free_list_mtx); 803 numvnodes--; 804 mtx_unlock(&vnode_free_list_mtx); 805 bo = &vp->v_bufobj; 806 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 807 ("cleaned vnode still on the free list.")); 808 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 809 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 810 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 811 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 812 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 813 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 814 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 815 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 816 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 817 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 818 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 819 VI_UNLOCK(vp); 820#ifdef MAC 821 mac_vnode_destroy(vp); 822#endif 823 if (vp->v_pollinfo != NULL) 824 destroy_vpollinfo(vp->v_pollinfo); 825#ifdef INVARIANTS 826 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */ 827 vp->v_op = NULL; 828#endif 829 lockdestroy(vp->v_vnlock); 830 mtx_destroy(&vp->v_interlock); 831 mtx_destroy(BO_MTX(bo)); 832 uma_zfree(vnode_zone, vp); 833} 834 835/* 836 * Try to recycle a freed vnode. We abort if anyone picks up a reference 837 * before we actually vgone(). This function must be called with the vnode 838 * held to prevent the vnode from being returned to the free list midway 839 * through vgone(). 840 */ 841static int 842vtryrecycle(struct vnode *vp) 843{ 844 struct mount *vnmp; 845 846 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp); 847 VNASSERT(vp->v_holdcnt, vp, 848 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 849 /* 850 * This vnode may found and locked via some other list, if so we 851 * can't recycle it yet. 852 */ 853 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) 854 return (EWOULDBLOCK); 855 /* 856 * Don't recycle if its filesystem is being suspended. 857 */ 858 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 859 VOP_UNLOCK(vp, 0); 860 return (EBUSY); 861 } 862 /* 863 * If we got this far, we need to acquire the interlock and see if 864 * anyone picked up this vnode from another list. If not, we will 865 * mark it with DOOMED via vgonel() so that anyone who does find it 866 * will skip over it. 867 */ 868 VI_LOCK(vp); 869 if (vp->v_usecount) { 870 VOP_UNLOCK(vp, LK_INTERLOCK); 871 vn_finished_write(vnmp); 872 return (EBUSY); 873 } 874 if ((vp->v_iflag & VI_DOOMED) == 0) 875 vgonel(vp); 876 VOP_UNLOCK(vp, LK_INTERLOCK); 877 vn_finished_write(vnmp); 878 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp); 879 return (0); 880} 881 882/* 883 * Return the next vnode from the free list. 884 */ 885int 886getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 887 struct vnode **vpp) 888{ 889 struct vnode *vp = NULL; 890 struct bufobj *bo; 891 892 mtx_lock(&vnode_free_list_mtx); 893 /* 894 * Lend our context to reclaim vnodes if they've exceeded the max. 895 */ 896 if (freevnodes > wantfreevnodes) 897 vnlru_free(1); 898 /* 899 * Wait for available vnodes. 900 */ 901 if (numvnodes > desiredvnodes) { 902 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 903 /* 904 * File system is beeing suspended, we cannot risk a 905 * deadlock here, so allocate new vnode anyway. 906 */ 907 if (freevnodes > wantfreevnodes) 908 vnlru_free(freevnodes - wantfreevnodes); 909 goto alloc; 910 } 911 if (vnlruproc_sig == 0) { 912 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 913 wakeup(vnlruproc); 914 } 915 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 916 "vlruwk", hz); 917#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 918 if (numvnodes > desiredvnodes) { 919 mtx_unlock(&vnode_free_list_mtx); 920 return (ENFILE); 921 } 922#endif 923 } 924alloc: 925 numvnodes++; 926 mtx_unlock(&vnode_free_list_mtx); 927 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 928 /* 929 * Setup locks. 930 */ 931 vp->v_vnlock = &vp->v_lock; 932 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 933 /* 934 * By default, don't allow shared locks unless filesystems 935 * opt-in. 936 */ 937 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 938 /* 939 * Initialize bufobj. 940 */ 941 bo = &vp->v_bufobj; 942 bo->__bo_vnode = vp; 943 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF); 944 bo->bo_ops = &buf_ops_bio; 945 bo->bo_private = vp; 946 TAILQ_INIT(&bo->bo_clean.bv_hd); 947 TAILQ_INIT(&bo->bo_dirty.bv_hd); 948 /* 949 * Initialize namecache. 950 */ 951 LIST_INIT(&vp->v_cache_src); 952 TAILQ_INIT(&vp->v_cache_dst); 953 /* 954 * Finalize various vnode identity bits. 955 */ 956 vp->v_type = VNON; 957 vp->v_tag = tag; 958 vp->v_op = vops; 959 v_incr_usecount(vp); 960 vp->v_data = 0; 961#ifdef MAC 962 mac_vnode_init(vp); 963 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 964 mac_vnode_associate_singlelabel(mp, vp); 965 else if (mp == NULL && vops != &dead_vnodeops) 966 printf("NULL mp in getnewvnode()\n"); 967#endif 968 if (mp != NULL) { 969 bo->bo_bsize = mp->mnt_stat.f_iosize; 970 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 971 vp->v_vflag |= VV_NOKNOTE; 972 } 973 974 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp); 975 *vpp = vp; 976 return (0); 977} 978 979/* 980 * Delete from old mount point vnode list, if on one. 981 */ 982static void 983delmntque(struct vnode *vp) 984{ 985 struct mount *mp; 986 987 mp = vp->v_mount; 988 if (mp == NULL) 989 return; 990 MNT_ILOCK(mp); 991 vp->v_mount = NULL; 992 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 993 ("bad mount point vnode list size")); 994 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 995 mp->mnt_nvnodelistsize--; 996 MNT_REL(mp); 997 MNT_IUNLOCK(mp); 998} 999 1000static void 1001insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1002{ 1003 1004 vp->v_data = NULL; 1005 vp->v_op = &dead_vnodeops; 1006 /* XXX non mp-safe fs may still call insmntque with vnode 1007 unlocked */ 1008 if (!VOP_ISLOCKED(vp)) 1009 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1010 vgone(vp); 1011 vput(vp); 1012} 1013 1014/* 1015 * Insert into list of vnodes for the new mount point, if available. 1016 */ 1017int 1018insmntque1(struct vnode *vp, struct mount *mp, 1019 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1020{ 1021 int locked; 1022 1023 KASSERT(vp->v_mount == NULL, 1024 ("insmntque: vnode already on per mount vnode list")); 1025 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1026#ifdef DEBUG_VFS_LOCKS 1027 if (!VFS_NEEDSGIANT(mp)) 1028 ASSERT_VOP_ELOCKED(vp, 1029 "insmntque: mp-safe fs and non-locked vp"); 1030#endif 1031 MNT_ILOCK(mp); 1032 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1033 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1034 mp->mnt_nvnodelistsize == 0)) { 1035 locked = VOP_ISLOCKED(vp); 1036 if (!locked || (locked == LK_EXCLUSIVE && 1037 (vp->v_vflag & VV_FORCEINSMQ) == 0)) { 1038 MNT_IUNLOCK(mp); 1039 if (dtr != NULL) 1040 dtr(vp, dtr_arg); 1041 return (EBUSY); 1042 } 1043 } 1044 vp->v_mount = mp; 1045 MNT_REF(mp); 1046 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1047 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1048 ("neg mount point vnode list size")); 1049 mp->mnt_nvnodelistsize++; 1050 MNT_IUNLOCK(mp); 1051 return (0); 1052} 1053 1054int 1055insmntque(struct vnode *vp, struct mount *mp) 1056{ 1057 1058 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1059} 1060 1061/* 1062 * Flush out and invalidate all buffers associated with a bufobj 1063 * Called with the underlying object locked. 1064 */ 1065int 1066bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1067{ 1068 int error; 1069 1070 BO_LOCK(bo); 1071 if (flags & V_SAVE) { 1072 error = bufobj_wwait(bo, slpflag, slptimeo); 1073 if (error) { 1074 BO_UNLOCK(bo); 1075 return (error); 1076 } 1077 if (bo->bo_dirty.bv_cnt > 0) { 1078 BO_UNLOCK(bo); 1079 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1080 return (error); 1081 /* 1082 * XXX We could save a lock/unlock if this was only 1083 * enabled under INVARIANTS 1084 */ 1085 BO_LOCK(bo); 1086 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1087 panic("vinvalbuf: dirty bufs"); 1088 } 1089 } 1090 /* 1091 * If you alter this loop please notice that interlock is dropped and 1092 * reacquired in flushbuflist. Special care is needed to ensure that 1093 * no race conditions occur from this. 1094 */ 1095 do { 1096 error = flushbuflist(&bo->bo_clean, 1097 flags, bo, slpflag, slptimeo); 1098 if (error == 0) 1099 error = flushbuflist(&bo->bo_dirty, 1100 flags, bo, slpflag, slptimeo); 1101 if (error != 0 && error != EAGAIN) { 1102 BO_UNLOCK(bo); 1103 return (error); 1104 } 1105 } while (error != 0); 1106 1107 /* 1108 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1109 * have write I/O in-progress but if there is a VM object then the 1110 * VM object can also have read-I/O in-progress. 1111 */ 1112 do { 1113 bufobj_wwait(bo, 0, 0); 1114 BO_UNLOCK(bo); 1115 if (bo->bo_object != NULL) { 1116 VM_OBJECT_LOCK(bo->bo_object); 1117 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1118 VM_OBJECT_UNLOCK(bo->bo_object); 1119 } 1120 BO_LOCK(bo); 1121 } while (bo->bo_numoutput > 0); 1122 BO_UNLOCK(bo); 1123 1124 /* 1125 * Destroy the copy in the VM cache, too. 1126 */ 1127 if (bo->bo_object != NULL) { 1128 VM_OBJECT_LOCK(bo->bo_object); 1129 vm_object_page_remove(bo->bo_object, 0, 0, 1130 (flags & V_SAVE) ? TRUE : FALSE); 1131 VM_OBJECT_UNLOCK(bo->bo_object); 1132 } 1133 1134#ifdef INVARIANTS 1135 BO_LOCK(bo); 1136 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1137 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1138 panic("vinvalbuf: flush failed"); 1139 BO_UNLOCK(bo); 1140#endif 1141 return (0); 1142} 1143 1144/* 1145 * Flush out and invalidate all buffers associated with a vnode. 1146 * Called with the underlying object locked. 1147 */ 1148int 1149vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1150{ 1151 1152 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags); 1153 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1154 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1155} 1156 1157/* 1158 * Flush out buffers on the specified list. 1159 * 1160 */ 1161static int 1162flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1163 int slptimeo) 1164{ 1165 struct buf *bp, *nbp; 1166 int retval, error; 1167 daddr_t lblkno; 1168 b_xflags_t xflags; 1169 1170 ASSERT_BO_LOCKED(bo); 1171 1172 retval = 0; 1173 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1174 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1175 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1176 continue; 1177 } 1178 lblkno = 0; 1179 xflags = 0; 1180 if (nbp != NULL) { 1181 lblkno = nbp->b_lblkno; 1182 xflags = nbp->b_xflags & 1183 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1184 } 1185 retval = EAGAIN; 1186 error = BUF_TIMELOCK(bp, 1187 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1188 "flushbuf", slpflag, slptimeo); 1189 if (error) { 1190 BO_LOCK(bo); 1191 return (error != ENOLCK ? error : EAGAIN); 1192 } 1193 KASSERT(bp->b_bufobj == bo, 1194 ("bp %p wrong b_bufobj %p should be %p", 1195 bp, bp->b_bufobj, bo)); 1196 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1197 BUF_UNLOCK(bp); 1198 BO_LOCK(bo); 1199 return (EAGAIN); 1200 } 1201 /* 1202 * XXX Since there are no node locks for NFS, I 1203 * believe there is a slight chance that a delayed 1204 * write will occur while sleeping just above, so 1205 * check for it. 1206 */ 1207 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1208 (flags & V_SAVE)) { 1209 bremfree(bp); 1210 bp->b_flags |= B_ASYNC; 1211 bwrite(bp); 1212 BO_LOCK(bo); 1213 return (EAGAIN); /* XXX: why not loop ? */ 1214 } 1215 bremfree(bp); 1216 bp->b_flags |= (B_INVAL | B_RELBUF); 1217 bp->b_flags &= ~B_ASYNC; 1218 brelse(bp); 1219 BO_LOCK(bo); 1220 if (nbp != NULL && 1221 (nbp->b_bufobj != bo || 1222 nbp->b_lblkno != lblkno || 1223 (nbp->b_xflags & 1224 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1225 break; /* nbp invalid */ 1226 } 1227 return (retval); 1228} 1229 1230/* 1231 * Truncate a file's buffer and pages to a specified length. This 1232 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1233 * sync activity. 1234 */ 1235int 1236vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, 1237 off_t length, int blksize) 1238{ 1239 struct buf *bp, *nbp; 1240 int anyfreed; 1241 int trunclbn; 1242 struct bufobj *bo; 1243 1244 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length); 1245 /* 1246 * Round up to the *next* lbn. 1247 */ 1248 trunclbn = (length + blksize - 1) / blksize; 1249 1250 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1251restart: 1252 bo = &vp->v_bufobj; 1253 BO_LOCK(bo); 1254 anyfreed = 1; 1255 for (;anyfreed;) { 1256 anyfreed = 0; 1257 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1258 if (bp->b_lblkno < trunclbn) 1259 continue; 1260 if (BUF_LOCK(bp, 1261 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1262 BO_MTX(bo)) == ENOLCK) 1263 goto restart; 1264 1265 bremfree(bp); 1266 bp->b_flags |= (B_INVAL | B_RELBUF); 1267 bp->b_flags &= ~B_ASYNC; 1268 brelse(bp); 1269 anyfreed = 1; 1270 1271 if (nbp != NULL && 1272 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1273 (nbp->b_vp != vp) || 1274 (nbp->b_flags & B_DELWRI))) { 1275 goto restart; 1276 } 1277 BO_LOCK(bo); 1278 } 1279 1280 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1281 if (bp->b_lblkno < trunclbn) 1282 continue; 1283 if (BUF_LOCK(bp, 1284 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1285 BO_MTX(bo)) == ENOLCK) 1286 goto restart; 1287 bremfree(bp); 1288 bp->b_flags |= (B_INVAL | B_RELBUF); 1289 bp->b_flags &= ~B_ASYNC; 1290 brelse(bp); 1291 anyfreed = 1; 1292 if (nbp != NULL && 1293 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1294 (nbp->b_vp != vp) || 1295 (nbp->b_flags & B_DELWRI) == 0)) { 1296 goto restart; 1297 } 1298 BO_LOCK(bo); 1299 } 1300 } 1301 1302 if (length > 0) { 1303restartsync: 1304 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1305 if (bp->b_lblkno > 0) 1306 continue; 1307 /* 1308 * Since we hold the vnode lock this should only 1309 * fail if we're racing with the buf daemon. 1310 */ 1311 if (BUF_LOCK(bp, 1312 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1313 BO_MTX(bo)) == ENOLCK) { 1314 goto restart; 1315 } 1316 VNASSERT((bp->b_flags & B_DELWRI), vp, 1317 ("buf(%p) on dirty queue without DELWRI", bp)); 1318 1319 bremfree(bp); 1320 bawrite(bp); 1321 BO_LOCK(bo); 1322 goto restartsync; 1323 } 1324 } 1325 1326 bufobj_wwait(bo, 0, 0); 1327 BO_UNLOCK(bo); 1328 vnode_pager_setsize(vp, length); 1329 1330 return (0); 1331} 1332 1333/* 1334 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1335 * a vnode. 1336 * 1337 * NOTE: We have to deal with the special case of a background bitmap 1338 * buffer, a situation where two buffers will have the same logical 1339 * block offset. We want (1) only the foreground buffer to be accessed 1340 * in a lookup and (2) must differentiate between the foreground and 1341 * background buffer in the splay tree algorithm because the splay 1342 * tree cannot normally handle multiple entities with the same 'index'. 1343 * We accomplish this by adding differentiating flags to the splay tree's 1344 * numerical domain. 1345 */ 1346static 1347struct buf * 1348buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1349{ 1350 struct buf dummy; 1351 struct buf *lefttreemax, *righttreemin, *y; 1352 1353 if (root == NULL) 1354 return (NULL); 1355 lefttreemax = righttreemin = &dummy; 1356 for (;;) { 1357 if (lblkno < root->b_lblkno || 1358 (lblkno == root->b_lblkno && 1359 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1360 if ((y = root->b_left) == NULL) 1361 break; 1362 if (lblkno < y->b_lblkno) { 1363 /* Rotate right. */ 1364 root->b_left = y->b_right; 1365 y->b_right = root; 1366 root = y; 1367 if ((y = root->b_left) == NULL) 1368 break; 1369 } 1370 /* Link into the new root's right tree. */ 1371 righttreemin->b_left = root; 1372 righttreemin = root; 1373 } else if (lblkno > root->b_lblkno || 1374 (lblkno == root->b_lblkno && 1375 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1376 if ((y = root->b_right) == NULL) 1377 break; 1378 if (lblkno > y->b_lblkno) { 1379 /* Rotate left. */ 1380 root->b_right = y->b_left; 1381 y->b_left = root; 1382 root = y; 1383 if ((y = root->b_right) == NULL) 1384 break; 1385 } 1386 /* Link into the new root's left tree. */ 1387 lefttreemax->b_right = root; 1388 lefttreemax = root; 1389 } else { 1390 break; 1391 } 1392 root = y; 1393 } 1394 /* Assemble the new root. */ 1395 lefttreemax->b_right = root->b_left; 1396 righttreemin->b_left = root->b_right; 1397 root->b_left = dummy.b_right; 1398 root->b_right = dummy.b_left; 1399 return (root); 1400} 1401 1402static void 1403buf_vlist_remove(struct buf *bp) 1404{ 1405 struct buf *root; 1406 struct bufv *bv; 1407 1408 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1409 ASSERT_BO_LOCKED(bp->b_bufobj); 1410 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1411 (BX_VNDIRTY|BX_VNCLEAN), 1412 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1413 if (bp->b_xflags & BX_VNDIRTY) 1414 bv = &bp->b_bufobj->bo_dirty; 1415 else 1416 bv = &bp->b_bufobj->bo_clean; 1417 if (bp != bv->bv_root) { 1418 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1419 KASSERT(root == bp, ("splay lookup failed in remove")); 1420 } 1421 if (bp->b_left == NULL) { 1422 root = bp->b_right; 1423 } else { 1424 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1425 root->b_right = bp->b_right; 1426 } 1427 bv->bv_root = root; 1428 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1429 bv->bv_cnt--; 1430 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1431} 1432 1433/* 1434 * Add the buffer to the sorted clean or dirty block list using a 1435 * splay tree algorithm. 1436 * 1437 * NOTE: xflags is passed as a constant, optimizing this inline function! 1438 */ 1439static void 1440buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1441{ 1442 struct buf *root; 1443 struct bufv *bv; 1444 1445 ASSERT_BO_LOCKED(bo); 1446 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1447 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1448 bp->b_xflags |= xflags; 1449 if (xflags & BX_VNDIRTY) 1450 bv = &bo->bo_dirty; 1451 else 1452 bv = &bo->bo_clean; 1453 1454 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1455 if (root == NULL) { 1456 bp->b_left = NULL; 1457 bp->b_right = NULL; 1458 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1459 } else if (bp->b_lblkno < root->b_lblkno || 1460 (bp->b_lblkno == root->b_lblkno && 1461 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1462 bp->b_left = root->b_left; 1463 bp->b_right = root; 1464 root->b_left = NULL; 1465 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1466 } else { 1467 bp->b_right = root->b_right; 1468 bp->b_left = root; 1469 root->b_right = NULL; 1470 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1471 } 1472 bv->bv_cnt++; 1473 bv->bv_root = bp; 1474} 1475 1476/* 1477 * Lookup a buffer using the splay tree. Note that we specifically avoid 1478 * shadow buffers used in background bitmap writes. 1479 * 1480 * This code isn't quite efficient as it could be because we are maintaining 1481 * two sorted lists and do not know which list the block resides in. 1482 * 1483 * During a "make buildworld" the desired buffer is found at one of 1484 * the roots more than 60% of the time. Thus, checking both roots 1485 * before performing either splay eliminates unnecessary splays on the 1486 * first tree splayed. 1487 */ 1488struct buf * 1489gbincore(struct bufobj *bo, daddr_t lblkno) 1490{ 1491 struct buf *bp; 1492 1493 ASSERT_BO_LOCKED(bo); 1494 if ((bp = bo->bo_clean.bv_root) != NULL && 1495 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1496 return (bp); 1497 if ((bp = bo->bo_dirty.bv_root) != NULL && 1498 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1499 return (bp); 1500 if ((bp = bo->bo_clean.bv_root) != NULL) { 1501 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1502 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1503 return (bp); 1504 } 1505 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1506 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1507 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1508 return (bp); 1509 } 1510 return (NULL); 1511} 1512 1513/* 1514 * Associate a buffer with a vnode. 1515 */ 1516void 1517bgetvp(struct vnode *vp, struct buf *bp) 1518{ 1519 struct bufobj *bo; 1520 1521 bo = &vp->v_bufobj; 1522 ASSERT_BO_LOCKED(bo); 1523 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1524 1525 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1526 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1527 ("bgetvp: bp already attached! %p", bp)); 1528 1529 vhold(vp); 1530 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT) 1531 bp->b_flags |= B_NEEDSGIANT; 1532 bp->b_vp = vp; 1533 bp->b_bufobj = bo; 1534 /* 1535 * Insert onto list for new vnode. 1536 */ 1537 buf_vlist_add(bp, bo, BX_VNCLEAN); 1538} 1539 1540/* 1541 * Disassociate a buffer from a vnode. 1542 */ 1543void 1544brelvp(struct buf *bp) 1545{ 1546 struct bufobj *bo; 1547 struct vnode *vp; 1548 1549 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1550 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1551 1552 /* 1553 * Delete from old vnode list, if on one. 1554 */ 1555 vp = bp->b_vp; /* XXX */ 1556 bo = bp->b_bufobj; 1557 BO_LOCK(bo); 1558 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1559 buf_vlist_remove(bp); 1560 else 1561 panic("brelvp: Buffer %p not on queue.", bp); 1562 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1563 bo->bo_flag &= ~BO_ONWORKLST; 1564 mtx_lock(&sync_mtx); 1565 LIST_REMOVE(bo, bo_synclist); 1566 syncer_worklist_len--; 1567 mtx_unlock(&sync_mtx); 1568 } 1569 bp->b_flags &= ~B_NEEDSGIANT; 1570 bp->b_vp = NULL; 1571 bp->b_bufobj = NULL; 1572 BO_UNLOCK(bo); 1573 vdrop(vp); 1574} 1575 1576/* 1577 * Add an item to the syncer work queue. 1578 */ 1579static void 1580vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1581{ 1582 int queue, slot; 1583 1584 ASSERT_BO_LOCKED(bo); 1585 1586 mtx_lock(&sync_mtx); 1587 if (bo->bo_flag & BO_ONWORKLST) 1588 LIST_REMOVE(bo, bo_synclist); 1589 else { 1590 bo->bo_flag |= BO_ONWORKLST; 1591 syncer_worklist_len++; 1592 } 1593 1594 if (delay > syncer_maxdelay - 2) 1595 delay = syncer_maxdelay - 2; 1596 slot = (syncer_delayno + delay) & syncer_mask; 1597 1598 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ : 1599 WI_MPSAFEQ; 1600 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo, 1601 bo_synclist); 1602 mtx_unlock(&sync_mtx); 1603} 1604 1605static int 1606sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1607{ 1608 int error, len; 1609 1610 mtx_lock(&sync_mtx); 1611 len = syncer_worklist_len - sync_vnode_count; 1612 mtx_unlock(&sync_mtx); 1613 error = SYSCTL_OUT(req, &len, sizeof(len)); 1614 return (error); 1615} 1616 1617SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1618 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1619 1620static struct proc *updateproc; 1621static void sched_sync(void); 1622static struct kproc_desc up_kp = { 1623 "syncer", 1624 sched_sync, 1625 &updateproc 1626}; 1627SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1628 1629static int 1630sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1631{ 1632 struct vnode *vp; 1633 struct mount *mp; 1634 1635 *bo = LIST_FIRST(slp); 1636 if (*bo == NULL) 1637 return (0); 1638 vp = (*bo)->__bo_vnode; /* XXX */ 1639 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1640 return (1); 1641 /* 1642 * We use vhold in case the vnode does not 1643 * successfully sync. vhold prevents the vnode from 1644 * going away when we unlock the sync_mtx so that 1645 * we can acquire the vnode interlock. 1646 */ 1647 vholdl(vp); 1648 mtx_unlock(&sync_mtx); 1649 VI_UNLOCK(vp); 1650 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1651 vdrop(vp); 1652 mtx_lock(&sync_mtx); 1653 return (*bo == LIST_FIRST(slp)); 1654 } 1655 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1656 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1657 VOP_UNLOCK(vp, 0); 1658 vn_finished_write(mp); 1659 BO_LOCK(*bo); 1660 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1661 /* 1662 * Put us back on the worklist. The worklist 1663 * routine will remove us from our current 1664 * position and then add us back in at a later 1665 * position. 1666 */ 1667 vn_syncer_add_to_worklist(*bo, syncdelay); 1668 } 1669 BO_UNLOCK(*bo); 1670 vdrop(vp); 1671 mtx_lock(&sync_mtx); 1672 return (0); 1673} 1674 1675/* 1676 * System filesystem synchronizer daemon. 1677 */ 1678static void 1679sched_sync(void) 1680{ 1681 struct synclist *gnext, *next; 1682 struct synclist *gslp, *slp; 1683 struct bufobj *bo; 1684 long starttime; 1685 struct thread *td = curthread; 1686 int last_work_seen; 1687 int net_worklist_len; 1688 int syncer_final_iter; 1689 int first_printf; 1690 int error; 1691 1692 last_work_seen = 0; 1693 syncer_final_iter = 0; 1694 first_printf = 1; 1695 syncer_state = SYNCER_RUNNING; 1696 starttime = time_uptime; 1697 td->td_pflags |= TDP_NORUNNINGBUF; 1698 1699 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1700 SHUTDOWN_PRI_LAST); 1701 1702 mtx_lock(&sync_mtx); 1703 for (;;) { 1704 if (syncer_state == SYNCER_FINAL_DELAY && 1705 syncer_final_iter == 0) { 1706 mtx_unlock(&sync_mtx); 1707 kproc_suspend_check(td->td_proc); 1708 mtx_lock(&sync_mtx); 1709 } 1710 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1711 if (syncer_state != SYNCER_RUNNING && 1712 starttime != time_uptime) { 1713 if (first_printf) { 1714 printf("\nSyncing disks, vnodes remaining..."); 1715 first_printf = 0; 1716 } 1717 printf("%d ", net_worklist_len); 1718 } 1719 starttime = time_uptime; 1720 1721 /* 1722 * Push files whose dirty time has expired. Be careful 1723 * of interrupt race on slp queue. 1724 * 1725 * Skip over empty worklist slots when shutting down. 1726 */ 1727 do { 1728 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1729 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1730 syncer_delayno += 1; 1731 if (syncer_delayno == syncer_maxdelay) 1732 syncer_delayno = 0; 1733 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1734 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1735 /* 1736 * If the worklist has wrapped since the 1737 * it was emptied of all but syncer vnodes, 1738 * switch to the FINAL_DELAY state and run 1739 * for one more second. 1740 */ 1741 if (syncer_state == SYNCER_SHUTTING_DOWN && 1742 net_worklist_len == 0 && 1743 last_work_seen == syncer_delayno) { 1744 syncer_state = SYNCER_FINAL_DELAY; 1745 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1746 } 1747 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1748 LIST_EMPTY(gslp) && syncer_worklist_len > 0); 1749 1750 /* 1751 * Keep track of the last time there was anything 1752 * on the worklist other than syncer vnodes. 1753 * Return to the SHUTTING_DOWN state if any 1754 * new work appears. 1755 */ 1756 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1757 last_work_seen = syncer_delayno; 1758 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1759 syncer_state = SYNCER_SHUTTING_DOWN; 1760 while (!LIST_EMPTY(slp)) { 1761 error = sync_vnode(slp, &bo, td); 1762 if (error == 1) { 1763 LIST_REMOVE(bo, bo_synclist); 1764 LIST_INSERT_HEAD(next, bo, bo_synclist); 1765 continue; 1766 } 1767 } 1768 if (!LIST_EMPTY(gslp)) { 1769 mtx_unlock(&sync_mtx); 1770 mtx_lock(&Giant); 1771 mtx_lock(&sync_mtx); 1772 while (!LIST_EMPTY(gslp)) { 1773 error = sync_vnode(gslp, &bo, td); 1774 if (error == 1) { 1775 LIST_REMOVE(bo, bo_synclist); 1776 LIST_INSERT_HEAD(gnext, bo, 1777 bo_synclist); 1778 continue; 1779 } 1780 } 1781 mtx_unlock(&Giant); 1782 } 1783 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1784 syncer_final_iter--; 1785 /* 1786 * The variable rushjob allows the kernel to speed up the 1787 * processing of the filesystem syncer process. A rushjob 1788 * value of N tells the filesystem syncer to process the next 1789 * N seconds worth of work on its queue ASAP. Currently rushjob 1790 * is used by the soft update code to speed up the filesystem 1791 * syncer process when the incore state is getting so far 1792 * ahead of the disk that the kernel memory pool is being 1793 * threatened with exhaustion. 1794 */ 1795 if (rushjob > 0) { 1796 rushjob -= 1; 1797 continue; 1798 } 1799 /* 1800 * Just sleep for a short period of time between 1801 * iterations when shutting down to allow some I/O 1802 * to happen. 1803 * 1804 * If it has taken us less than a second to process the 1805 * current work, then wait. Otherwise start right over 1806 * again. We can still lose time if any single round 1807 * takes more than two seconds, but it does not really 1808 * matter as we are just trying to generally pace the 1809 * filesystem activity. 1810 */ 1811 if (syncer_state != SYNCER_RUNNING) 1812 cv_timedwait(&sync_wakeup, &sync_mtx, 1813 hz / SYNCER_SHUTDOWN_SPEEDUP); 1814 else if (time_uptime == starttime) 1815 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1816 } 1817} 1818 1819/* 1820 * Request the syncer daemon to speed up its work. 1821 * We never push it to speed up more than half of its 1822 * normal turn time, otherwise it could take over the cpu. 1823 */ 1824int 1825speedup_syncer(void) 1826{ 1827 int ret = 0; 1828 1829 mtx_lock(&sync_mtx); 1830 if (rushjob < syncdelay / 2) { 1831 rushjob += 1; 1832 stat_rush_requests += 1; 1833 ret = 1; 1834 } 1835 mtx_unlock(&sync_mtx); 1836 cv_broadcast(&sync_wakeup); 1837 return (ret); 1838} 1839 1840/* 1841 * Tell the syncer to speed up its work and run though its work 1842 * list several times, then tell it to shut down. 1843 */ 1844static void 1845syncer_shutdown(void *arg, int howto) 1846{ 1847 1848 if (howto & RB_NOSYNC) 1849 return; 1850 mtx_lock(&sync_mtx); 1851 syncer_state = SYNCER_SHUTTING_DOWN; 1852 rushjob = 0; 1853 mtx_unlock(&sync_mtx); 1854 cv_broadcast(&sync_wakeup); 1855 kproc_shutdown(arg, howto); 1856} 1857 1858/* 1859 * Reassign a buffer from one vnode to another. 1860 * Used to assign file specific control information 1861 * (indirect blocks) to the vnode to which they belong. 1862 */ 1863void 1864reassignbuf(struct buf *bp) 1865{ 1866 struct vnode *vp; 1867 struct bufobj *bo; 1868 int delay; 1869#ifdef INVARIANTS 1870 struct bufv *bv; 1871#endif 1872 1873 vp = bp->b_vp; 1874 bo = bp->b_bufobj; 1875 ++reassignbufcalls; 1876 1877 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1878 bp, bp->b_vp, bp->b_flags); 1879 /* 1880 * B_PAGING flagged buffers cannot be reassigned because their vp 1881 * is not fully linked in. 1882 */ 1883 if (bp->b_flags & B_PAGING) 1884 panic("cannot reassign paging buffer"); 1885 1886 /* 1887 * Delete from old vnode list, if on one. 1888 */ 1889 BO_LOCK(bo); 1890 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1891 buf_vlist_remove(bp); 1892 else 1893 panic("reassignbuf: Buffer %p not on queue.", bp); 1894 /* 1895 * If dirty, put on list of dirty buffers; otherwise insert onto list 1896 * of clean buffers. 1897 */ 1898 if (bp->b_flags & B_DELWRI) { 1899 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 1900 switch (vp->v_type) { 1901 case VDIR: 1902 delay = dirdelay; 1903 break; 1904 case VCHR: 1905 delay = metadelay; 1906 break; 1907 default: 1908 delay = filedelay; 1909 } 1910 vn_syncer_add_to_worklist(bo, delay); 1911 } 1912 buf_vlist_add(bp, bo, BX_VNDIRTY); 1913 } else { 1914 buf_vlist_add(bp, bo, BX_VNCLEAN); 1915 1916 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1917 mtx_lock(&sync_mtx); 1918 LIST_REMOVE(bo, bo_synclist); 1919 syncer_worklist_len--; 1920 mtx_unlock(&sync_mtx); 1921 bo->bo_flag &= ~BO_ONWORKLST; 1922 } 1923 } 1924#ifdef INVARIANTS 1925 bv = &bo->bo_clean; 1926 bp = TAILQ_FIRST(&bv->bv_hd); 1927 KASSERT(bp == NULL || bp->b_bufobj == bo, 1928 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1929 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1930 KASSERT(bp == NULL || bp->b_bufobj == bo, 1931 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1932 bv = &bo->bo_dirty; 1933 bp = TAILQ_FIRST(&bv->bv_hd); 1934 KASSERT(bp == NULL || bp->b_bufobj == bo, 1935 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1936 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1937 KASSERT(bp == NULL || bp->b_bufobj == bo, 1938 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1939#endif 1940 BO_UNLOCK(bo); 1941} 1942 1943/* 1944 * Increment the use and hold counts on the vnode, taking care to reference 1945 * the driver's usecount if this is a chardev. The vholdl() will remove 1946 * the vnode from the free list if it is presently free. Requires the 1947 * vnode interlock and returns with it held. 1948 */ 1949static void 1950v_incr_usecount(struct vnode *vp) 1951{ 1952 1953 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n", 1954 vp, vp->v_holdcnt, vp->v_usecount); 1955 vp->v_usecount++; 1956 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1957 dev_lock(); 1958 vp->v_rdev->si_usecount++; 1959 dev_unlock(); 1960 } 1961 vholdl(vp); 1962} 1963 1964/* 1965 * Turn a holdcnt into a use+holdcnt such that only one call to 1966 * v_decr_usecount is needed. 1967 */ 1968static void 1969v_upgrade_usecount(struct vnode *vp) 1970{ 1971 1972 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n", 1973 vp, vp->v_holdcnt, vp->v_usecount); 1974 vp->v_usecount++; 1975 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1976 dev_lock(); 1977 vp->v_rdev->si_usecount++; 1978 dev_unlock(); 1979 } 1980} 1981 1982/* 1983 * Decrement the vnode use and hold count along with the driver's usecount 1984 * if this is a chardev. The vdropl() below releases the vnode interlock 1985 * as it may free the vnode. 1986 */ 1987static void 1988v_decr_usecount(struct vnode *vp) 1989{ 1990 1991 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n", 1992 vp, vp->v_holdcnt, vp->v_usecount); 1993 ASSERT_VI_LOCKED(vp, __FUNCTION__); 1994 VNASSERT(vp->v_usecount > 0, vp, 1995 ("v_decr_usecount: negative usecount")); 1996 vp->v_usecount--; 1997 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1998 dev_lock(); 1999 vp->v_rdev->si_usecount--; 2000 dev_unlock(); 2001 } 2002 vdropl(vp); 2003} 2004 2005/* 2006 * Decrement only the use count and driver use count. This is intended to 2007 * be paired with a follow on vdropl() to release the remaining hold count. 2008 * In this way we may vgone() a vnode with a 0 usecount without risk of 2009 * having it end up on a free list because the hold count is kept above 0. 2010 */ 2011static void 2012v_decr_useonly(struct vnode *vp) 2013{ 2014 2015 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n", 2016 vp, vp->v_holdcnt, vp->v_usecount); 2017 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2018 VNASSERT(vp->v_usecount > 0, vp, 2019 ("v_decr_useonly: negative usecount")); 2020 vp->v_usecount--; 2021 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2022 dev_lock(); 2023 vp->v_rdev->si_usecount--; 2024 dev_unlock(); 2025 } 2026} 2027 2028/* 2029 * Grab a particular vnode from the free list, increment its 2030 * reference count and lock it. VI_DOOMED is set if the vnode 2031 * is being destroyed. Only callers who specify LK_RETRY will 2032 * see doomed vnodes. If inactive processing was delayed in 2033 * vput try to do it here. 2034 */ 2035int 2036vget(struct vnode *vp, int flags, struct thread *td) 2037{ 2038 int error; 2039 2040 error = 0; 2041 VFS_ASSERT_GIANT(vp->v_mount); 2042 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2043 ("vget: invalid lock operation")); 2044 if ((flags & LK_INTERLOCK) == 0) 2045 VI_LOCK(vp); 2046 vholdl(vp); 2047 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2048 vdrop(vp); 2049 return (error); 2050 } 2051 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2052 panic("vget: vn_lock failed to return ENOENT\n"); 2053 VI_LOCK(vp); 2054 /* Upgrade our holdcnt to a usecount. */ 2055 v_upgrade_usecount(vp); 2056 /* 2057 * We don't guarantee that any particular close will 2058 * trigger inactive processing so just make a best effort 2059 * here at preventing a reference to a removed file. If 2060 * we don't succeed no harm is done. 2061 */ 2062 if (vp->v_iflag & VI_OWEINACT) { 2063 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2064 (flags & LK_NOWAIT) == 0) 2065 vinactive(vp, td); 2066 vp->v_iflag &= ~VI_OWEINACT; 2067 } 2068 VI_UNLOCK(vp); 2069 return (0); 2070} 2071 2072/* 2073 * Increase the reference count of a vnode. 2074 */ 2075void 2076vref(struct vnode *vp) 2077{ 2078 2079 VI_LOCK(vp); 2080 v_incr_usecount(vp); 2081 VI_UNLOCK(vp); 2082} 2083 2084/* 2085 * Return reference count of a vnode. 2086 * 2087 * The results of this call are only guaranteed when some mechanism other 2088 * than the VI lock is used to stop other processes from gaining references 2089 * to the vnode. This may be the case if the caller holds the only reference. 2090 * This is also useful when stale data is acceptable as race conditions may 2091 * be accounted for by some other means. 2092 */ 2093int 2094vrefcnt(struct vnode *vp) 2095{ 2096 int usecnt; 2097 2098 VI_LOCK(vp); 2099 usecnt = vp->v_usecount; 2100 VI_UNLOCK(vp); 2101 2102 return (usecnt); 2103} 2104 2105 2106/* 2107 * Vnode put/release. 2108 * If count drops to zero, call inactive routine and return to freelist. 2109 */ 2110void 2111vrele(struct vnode *vp) 2112{ 2113 struct thread *td = curthread; /* XXX */ 2114 2115 KASSERT(vp != NULL, ("vrele: null vp")); 2116 VFS_ASSERT_GIANT(vp->v_mount); 2117 2118 VI_LOCK(vp); 2119 2120 /* Skip this v_writecount check if we're going to panic below. */ 2121 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2122 ("vrele: missed vn_close")); 2123 2124 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2125 vp->v_usecount == 1)) { 2126 v_decr_usecount(vp); 2127 return; 2128 } 2129 if (vp->v_usecount != 1) { 2130#ifdef DIAGNOSTIC 2131 vprint("vrele: negative ref count", vp); 2132#endif 2133 VI_UNLOCK(vp); 2134 panic("vrele: negative ref cnt"); 2135 } 2136 /* 2137 * We want to hold the vnode until the inactive finishes to 2138 * prevent vgone() races. We drop the use count here and the 2139 * hold count below when we're done. 2140 */ 2141 v_decr_useonly(vp); 2142 /* 2143 * We must call VOP_INACTIVE with the node locked. Mark 2144 * as VI_DOINGINACT to avoid recursion. 2145 */ 2146 vp->v_iflag |= VI_OWEINACT; 2147 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) { 2148 VI_LOCK(vp); 2149 if (vp->v_usecount > 0) 2150 vp->v_iflag &= ~VI_OWEINACT; 2151 if (vp->v_iflag & VI_OWEINACT) 2152 vinactive(vp, td); 2153 VOP_UNLOCK(vp, 0); 2154 } else { 2155 VI_LOCK(vp); 2156 if (vp->v_usecount > 0) 2157 vp->v_iflag &= ~VI_OWEINACT; 2158 } 2159 vdropl(vp); 2160} 2161 2162/* 2163 * Release an already locked vnode. This give the same effects as 2164 * unlock+vrele(), but takes less time and avoids releasing and 2165 * re-aquiring the lock (as vrele() acquires the lock internally.) 2166 */ 2167void 2168vput(struct vnode *vp) 2169{ 2170 struct thread *td = curthread; /* XXX */ 2171 int error; 2172 2173 KASSERT(vp != NULL, ("vput: null vp")); 2174 ASSERT_VOP_LOCKED(vp, "vput"); 2175 VFS_ASSERT_GIANT(vp->v_mount); 2176 VI_LOCK(vp); 2177 /* Skip this v_writecount check if we're going to panic below. */ 2178 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2179 ("vput: missed vn_close")); 2180 error = 0; 2181 2182 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2183 vp->v_usecount == 1)) { 2184 VOP_UNLOCK(vp, 0); 2185 v_decr_usecount(vp); 2186 return; 2187 } 2188 2189 if (vp->v_usecount != 1) { 2190#ifdef DIAGNOSTIC 2191 vprint("vput: negative ref count", vp); 2192#endif 2193 panic("vput: negative ref cnt"); 2194 } 2195 /* 2196 * We want to hold the vnode until the inactive finishes to 2197 * prevent vgone() races. We drop the use count here and the 2198 * hold count below when we're done. 2199 */ 2200 v_decr_useonly(vp); 2201 vp->v_iflag |= VI_OWEINACT; 2202 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2203 error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT); 2204 VI_LOCK(vp); 2205 if (error) { 2206 if (vp->v_usecount > 0) 2207 vp->v_iflag &= ~VI_OWEINACT; 2208 goto done; 2209 } 2210 } 2211 if (vp->v_usecount > 0) 2212 vp->v_iflag &= ~VI_OWEINACT; 2213 if (vp->v_iflag & VI_OWEINACT) 2214 vinactive(vp, td); 2215 VOP_UNLOCK(vp, 0); 2216done: 2217 vdropl(vp); 2218} 2219 2220/* 2221 * Somebody doesn't want the vnode recycled. 2222 */ 2223void 2224vhold(struct vnode *vp) 2225{ 2226 2227 VI_LOCK(vp); 2228 vholdl(vp); 2229 VI_UNLOCK(vp); 2230} 2231 2232void 2233vholdl(struct vnode *vp) 2234{ 2235 2236 vp->v_holdcnt++; 2237 if (VSHOULDBUSY(vp)) 2238 vbusy(vp); 2239} 2240 2241/* 2242 * Note that there is one less who cares about this vnode. vdrop() is the 2243 * opposite of vhold(). 2244 */ 2245void 2246vdrop(struct vnode *vp) 2247{ 2248 2249 VI_LOCK(vp); 2250 vdropl(vp); 2251} 2252 2253/* 2254 * Drop the hold count of the vnode. If this is the last reference to 2255 * the vnode we will free it if it has been vgone'd otherwise it is 2256 * placed on the free list. 2257 */ 2258void 2259vdropl(struct vnode *vp) 2260{ 2261 2262 ASSERT_VI_LOCKED(vp, "vdropl"); 2263 if (vp->v_holdcnt <= 0) 2264 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2265 vp->v_holdcnt--; 2266 if (vp->v_holdcnt == 0) { 2267 if (vp->v_iflag & VI_DOOMED) { 2268 vdestroy(vp); 2269 return; 2270 } else 2271 vfree(vp); 2272 } 2273 VI_UNLOCK(vp); 2274} 2275 2276/* 2277 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2278 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2279 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2280 * failed lock upgrade. 2281 */ 2282static void 2283vinactive(struct vnode *vp, struct thread *td) 2284{ 2285 2286 ASSERT_VOP_ELOCKED(vp, "vinactive"); 2287 ASSERT_VI_LOCKED(vp, "vinactive"); 2288 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2289 ("vinactive: recursed on VI_DOINGINACT")); 2290 vp->v_iflag |= VI_DOINGINACT; 2291 vp->v_iflag &= ~VI_OWEINACT; 2292 VI_UNLOCK(vp); 2293 VOP_INACTIVE(vp, td); 2294 VI_LOCK(vp); 2295 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2296 ("vinactive: lost VI_DOINGINACT")); 2297 vp->v_iflag &= ~VI_DOINGINACT; 2298} 2299 2300/* 2301 * Remove any vnodes in the vnode table belonging to mount point mp. 2302 * 2303 * If FORCECLOSE is not specified, there should not be any active ones, 2304 * return error if any are found (nb: this is a user error, not a 2305 * system error). If FORCECLOSE is specified, detach any active vnodes 2306 * that are found. 2307 * 2308 * If WRITECLOSE is set, only flush out regular file vnodes open for 2309 * writing. 2310 * 2311 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2312 * 2313 * `rootrefs' specifies the base reference count for the root vnode 2314 * of this filesystem. The root vnode is considered busy if its 2315 * v_usecount exceeds this value. On a successful return, vflush(, td) 2316 * will call vrele() on the root vnode exactly rootrefs times. 2317 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2318 * be zero. 2319 */ 2320#ifdef DIAGNOSTIC 2321static int busyprt = 0; /* print out busy vnodes */ 2322SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 2323#endif 2324 2325int 2326vflush( struct mount *mp, int rootrefs, int flags, struct thread *td) 2327{ 2328 struct vnode *vp, *mvp, *rootvp = NULL; 2329 struct vattr vattr; 2330 int busy = 0, error; 2331 2332 CTR1(KTR_VFS, "vflush: mp %p", mp); 2333 if (rootrefs > 0) { 2334 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2335 ("vflush: bad args")); 2336 /* 2337 * Get the filesystem root vnode. We can vput() it 2338 * immediately, since with rootrefs > 0, it won't go away. 2339 */ 2340 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0) 2341 return (error); 2342 vput(rootvp); 2343 2344 } 2345 MNT_ILOCK(mp); 2346loop: 2347 MNT_VNODE_FOREACH(vp, mp, mvp) { 2348 2349 VI_LOCK(vp); 2350 vholdl(vp); 2351 MNT_IUNLOCK(mp); 2352 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2353 if (error) { 2354 vdrop(vp); 2355 MNT_ILOCK(mp); 2356 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2357 goto loop; 2358 } 2359 /* 2360 * Skip over a vnodes marked VV_SYSTEM. 2361 */ 2362 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2363 VOP_UNLOCK(vp, 0); 2364 vdrop(vp); 2365 MNT_ILOCK(mp); 2366 continue; 2367 } 2368 /* 2369 * If WRITECLOSE is set, flush out unlinked but still open 2370 * files (even if open only for reading) and regular file 2371 * vnodes open for writing. 2372 */ 2373 if (flags & WRITECLOSE) { 2374 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2375 VI_LOCK(vp); 2376 2377 if ((vp->v_type == VNON || 2378 (error == 0 && vattr.va_nlink > 0)) && 2379 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2380 VOP_UNLOCK(vp, 0); 2381 vdropl(vp); 2382 MNT_ILOCK(mp); 2383 continue; 2384 } 2385 } else 2386 VI_LOCK(vp); 2387 /* 2388 * With v_usecount == 0, all we need to do is clear out the 2389 * vnode data structures and we are done. 2390 * 2391 * If FORCECLOSE is set, forcibly close the vnode. 2392 */ 2393 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2394 VNASSERT(vp->v_usecount == 0 || 2395 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2396 ("device VNODE %p is FORCECLOSED", vp)); 2397 vgonel(vp); 2398 } else { 2399 busy++; 2400#ifdef DIAGNOSTIC 2401 if (busyprt) 2402 vprint("vflush: busy vnode", vp); 2403#endif 2404 } 2405 VOP_UNLOCK(vp, 0); 2406 vdropl(vp); 2407 MNT_ILOCK(mp); 2408 } 2409 MNT_IUNLOCK(mp); 2410 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2411 /* 2412 * If just the root vnode is busy, and if its refcount 2413 * is equal to `rootrefs', then go ahead and kill it. 2414 */ 2415 VI_LOCK(rootvp); 2416 KASSERT(busy > 0, ("vflush: not busy")); 2417 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2418 ("vflush: usecount %d < rootrefs %d", 2419 rootvp->v_usecount, rootrefs)); 2420 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2421 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2422 vgone(rootvp); 2423 VOP_UNLOCK(rootvp, 0); 2424 busy = 0; 2425 } else 2426 VI_UNLOCK(rootvp); 2427 } 2428 if (busy) 2429 return (EBUSY); 2430 for (; rootrefs > 0; rootrefs--) 2431 vrele(rootvp); 2432 return (0); 2433} 2434 2435/* 2436 * Recycle an unused vnode to the front of the free list. 2437 */ 2438int 2439vrecycle(struct vnode *vp, struct thread *td) 2440{ 2441 int recycled; 2442 2443 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2444 recycled = 0; 2445 VI_LOCK(vp); 2446 if (vp->v_usecount == 0) { 2447 recycled = 1; 2448 vgonel(vp); 2449 } 2450 VI_UNLOCK(vp); 2451 return (recycled); 2452} 2453 2454/* 2455 * Eliminate all activity associated with a vnode 2456 * in preparation for reuse. 2457 */ 2458void 2459vgone(struct vnode *vp) 2460{ 2461 VI_LOCK(vp); 2462 vgonel(vp); 2463 VI_UNLOCK(vp); 2464} 2465 2466/* 2467 * vgone, with the vp interlock held. 2468 */ 2469void 2470vgonel(struct vnode *vp) 2471{ 2472 struct thread *td; 2473 int oweinact; 2474 int active; 2475 struct mount *mp; 2476 2477 CTR1(KTR_VFS, "vgonel: vp %p", vp); 2478 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2479 ASSERT_VI_LOCKED(vp, "vgonel"); 2480 VNASSERT(vp->v_holdcnt, vp, 2481 ("vgonel: vp %p has no reference.", vp)); 2482 td = curthread; 2483 2484 /* 2485 * Don't vgonel if we're already doomed. 2486 */ 2487 if (vp->v_iflag & VI_DOOMED) 2488 return; 2489 vp->v_iflag |= VI_DOOMED; 2490 /* 2491 * Check to see if the vnode is in use. If so, we have to call 2492 * VOP_CLOSE() and VOP_INACTIVE(). 2493 */ 2494 active = vp->v_usecount; 2495 oweinact = (vp->v_iflag & VI_OWEINACT); 2496 VI_UNLOCK(vp); 2497 /* 2498 * Clean out any buffers associated with the vnode. 2499 * If the flush fails, just toss the buffers. 2500 */ 2501 mp = NULL; 2502 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2503 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2504 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) 2505 vinvalbuf(vp, 0, 0, 0); 2506 2507 /* 2508 * If purging an active vnode, it must be closed and 2509 * deactivated before being reclaimed. 2510 */ 2511 if (active) 2512 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2513 if (oweinact || active) { 2514 VI_LOCK(vp); 2515 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2516 vinactive(vp, td); 2517 VI_UNLOCK(vp); 2518 } 2519 /* 2520 * Reclaim the vnode. 2521 */ 2522 if (VOP_RECLAIM(vp, td)) 2523 panic("vgone: cannot reclaim"); 2524 if (mp != NULL) 2525 vn_finished_secondary_write(mp); 2526 VNASSERT(vp->v_object == NULL, vp, 2527 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2528 /* 2529 * Clear the advisory locks and wake up waiting threads. 2530 */ 2531 lf_purgelocks(vp, &(vp->v_lockf)); 2532 /* 2533 * Delete from old mount point vnode list. 2534 */ 2535 delmntque(vp); 2536 cache_purge(vp); 2537 /* 2538 * Done with purge, reset to the standard lock and invalidate 2539 * the vnode. 2540 */ 2541 VI_LOCK(vp); 2542 vp->v_vnlock = &vp->v_lock; 2543 vp->v_op = &dead_vnodeops; 2544 vp->v_tag = "none"; 2545 vp->v_type = VBAD; 2546} 2547 2548/* 2549 * Calculate the total number of references to a special device. 2550 */ 2551int 2552vcount(struct vnode *vp) 2553{ 2554 int count; 2555 2556 dev_lock(); 2557 count = vp->v_rdev->si_usecount; 2558 dev_unlock(); 2559 return (count); 2560} 2561 2562/* 2563 * Same as above, but using the struct cdev *as argument 2564 */ 2565int 2566count_dev(struct cdev *dev) 2567{ 2568 int count; 2569 2570 dev_lock(); 2571 count = dev->si_usecount; 2572 dev_unlock(); 2573 return(count); 2574} 2575 2576/* 2577 * Print out a description of a vnode. 2578 */ 2579static char *typename[] = 2580{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2581 "VMARKER"}; 2582 2583void 2584vn_printf(struct vnode *vp, const char *fmt, ...) 2585{ 2586 va_list ap; 2587 char buf[256], buf2[16]; 2588 u_long flags; 2589 2590 va_start(ap, fmt); 2591 vprintf(fmt, ap); 2592 va_end(ap); 2593 printf("%p: ", (void *)vp); 2594 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2595 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2596 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2597 buf[0] = '\0'; 2598 buf[1] = '\0'; 2599 if (vp->v_vflag & VV_ROOT) 2600 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2601 if (vp->v_vflag & VV_ISTTY) 2602 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2603 if (vp->v_vflag & VV_NOSYNC) 2604 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2605 if (vp->v_vflag & VV_CACHEDLABEL) 2606 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2607 if (vp->v_vflag & VV_TEXT) 2608 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2609 if (vp->v_vflag & VV_COPYONWRITE) 2610 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2611 if (vp->v_vflag & VV_SYSTEM) 2612 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2613 if (vp->v_vflag & VV_PROCDEP) 2614 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2615 if (vp->v_vflag & VV_NOKNOTE) 2616 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2617 if (vp->v_vflag & VV_DELETED) 2618 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2619 if (vp->v_vflag & VV_MD) 2620 strlcat(buf, "|VV_MD", sizeof(buf)); 2621 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | 2622 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2623 VV_NOKNOTE | VV_DELETED | VV_MD); 2624 if (flags != 0) { 2625 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2626 strlcat(buf, buf2, sizeof(buf)); 2627 } 2628 if (vp->v_iflag & VI_MOUNT) 2629 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2630 if (vp->v_iflag & VI_AGE) 2631 strlcat(buf, "|VI_AGE", sizeof(buf)); 2632 if (vp->v_iflag & VI_DOOMED) 2633 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2634 if (vp->v_iflag & VI_FREE) 2635 strlcat(buf, "|VI_FREE", sizeof(buf)); 2636 if (vp->v_iflag & VI_OBJDIRTY) 2637 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf)); 2638 if (vp->v_iflag & VI_DOINGINACT) 2639 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2640 if (vp->v_iflag & VI_OWEINACT) 2641 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2642 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2643 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT); 2644 if (flags != 0) { 2645 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2646 strlcat(buf, buf2, sizeof(buf)); 2647 } 2648 printf(" flags (%s)\n", buf + 1); 2649 if (mtx_owned(VI_MTX(vp))) 2650 printf(" VI_LOCKed"); 2651 if (vp->v_object != NULL) 2652 printf(" v_object %p ref %d pages %d\n", 2653 vp->v_object, vp->v_object->ref_count, 2654 vp->v_object->resident_page_count); 2655 printf(" "); 2656 lockmgr_printinfo(vp->v_vnlock); 2657 printf("\n"); 2658 if (vp->v_data != NULL) 2659 VOP_PRINT(vp); 2660} 2661 2662#ifdef DDB 2663/* 2664 * List all of the locked vnodes in the system. 2665 * Called when debugging the kernel. 2666 */ 2667DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2668{ 2669 struct mount *mp, *nmp; 2670 struct vnode *vp; 2671 2672 /* 2673 * Note: because this is DDB, we can't obey the locking semantics 2674 * for these structures, which means we could catch an inconsistent 2675 * state and dereference a nasty pointer. Not much to be done 2676 * about that. 2677 */ 2678 db_printf("Locked vnodes\n"); 2679 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2680 nmp = TAILQ_NEXT(mp, mnt_list); 2681 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2682 if (vp->v_type != VMARKER && 2683 VOP_ISLOCKED(vp)) 2684 vprint("", vp); 2685 } 2686 nmp = TAILQ_NEXT(mp, mnt_list); 2687 } 2688} 2689 2690/* 2691 * Show details about the given vnode. 2692 */ 2693DB_SHOW_COMMAND(vnode, db_show_vnode) 2694{ 2695 struct vnode *vp; 2696 2697 if (!have_addr) 2698 return; 2699 vp = (struct vnode *)addr; 2700 vn_printf(vp, "vnode "); 2701} 2702 2703/* 2704 * Show details about the given mount point. 2705 */ 2706DB_SHOW_COMMAND(mount, db_show_mount) 2707{ 2708 struct mount *mp; 2709 struct statfs *sp; 2710 struct vnode *vp; 2711 char buf[512]; 2712 u_int flags; 2713 2714 if (!have_addr) { 2715 /* No address given, print short info about all mount points. */ 2716 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2717 db_printf("%p %s on %s (%s)\n", mp, 2718 mp->mnt_stat.f_mntfromname, 2719 mp->mnt_stat.f_mntonname, 2720 mp->mnt_stat.f_fstypename); 2721 if (db_pager_quit) 2722 break; 2723 } 2724 db_printf("\nMore info: show mount <addr>\n"); 2725 return; 2726 } 2727 2728 mp = (struct mount *)addr; 2729 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 2730 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 2731 2732 buf[0] = '\0'; 2733 flags = mp->mnt_flag; 2734#define MNT_FLAG(flag) do { \ 2735 if (flags & (flag)) { \ 2736 if (buf[0] != '\0') \ 2737 strlcat(buf, ", ", sizeof(buf)); \ 2738 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 2739 flags &= ~(flag); \ 2740 } \ 2741} while (0) 2742 MNT_FLAG(MNT_RDONLY); 2743 MNT_FLAG(MNT_SYNCHRONOUS); 2744 MNT_FLAG(MNT_NOEXEC); 2745 MNT_FLAG(MNT_NOSUID); 2746 MNT_FLAG(MNT_UNION); 2747 MNT_FLAG(MNT_ASYNC); 2748 MNT_FLAG(MNT_SUIDDIR); 2749 MNT_FLAG(MNT_SOFTDEP); 2750 MNT_FLAG(MNT_NOSYMFOLLOW); 2751 MNT_FLAG(MNT_GJOURNAL); 2752 MNT_FLAG(MNT_MULTILABEL); 2753 MNT_FLAG(MNT_ACLS); 2754 MNT_FLAG(MNT_NOATIME); 2755 MNT_FLAG(MNT_NOCLUSTERR); 2756 MNT_FLAG(MNT_NOCLUSTERW); 2757 MNT_FLAG(MNT_EXRDONLY); 2758 MNT_FLAG(MNT_EXPORTED); 2759 MNT_FLAG(MNT_DEFEXPORTED); 2760 MNT_FLAG(MNT_EXPORTANON); 2761 MNT_FLAG(MNT_EXKERB); 2762 MNT_FLAG(MNT_EXPUBLIC); 2763 MNT_FLAG(MNT_LOCAL); 2764 MNT_FLAG(MNT_QUOTA); 2765 MNT_FLAG(MNT_ROOTFS); 2766 MNT_FLAG(MNT_USER); 2767 MNT_FLAG(MNT_IGNORE); 2768 MNT_FLAG(MNT_UPDATE); 2769 MNT_FLAG(MNT_DELEXPORT); 2770 MNT_FLAG(MNT_RELOAD); 2771 MNT_FLAG(MNT_FORCE); 2772 MNT_FLAG(MNT_SNAPSHOT); 2773 MNT_FLAG(MNT_BYFSID); 2774#undef MNT_FLAG 2775 if (flags != 0) { 2776 if (buf[0] != '\0') 2777 strlcat(buf, ", ", sizeof(buf)); 2778 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2779 "0x%08x", flags); 2780 } 2781 db_printf(" mnt_flag = %s\n", buf); 2782 2783 buf[0] = '\0'; 2784 flags = mp->mnt_kern_flag; 2785#define MNT_KERN_FLAG(flag) do { \ 2786 if (flags & (flag)) { \ 2787 if (buf[0] != '\0') \ 2788 strlcat(buf, ", ", sizeof(buf)); \ 2789 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 2790 flags &= ~(flag); \ 2791 } \ 2792} while (0) 2793 MNT_KERN_FLAG(MNTK_UNMOUNTF); 2794 MNT_KERN_FLAG(MNTK_ASYNC); 2795 MNT_KERN_FLAG(MNTK_SOFTDEP); 2796 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 2797 MNT_KERN_FLAG(MNTK_UNMOUNT); 2798 MNT_KERN_FLAG(MNTK_MWAIT); 2799 MNT_KERN_FLAG(MNTK_SUSPEND); 2800 MNT_KERN_FLAG(MNTK_SUSPEND2); 2801 MNT_KERN_FLAG(MNTK_SUSPENDED); 2802 MNT_KERN_FLAG(MNTK_MPSAFE); 2803 MNT_KERN_FLAG(MNTK_NOKNOTE); 2804 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 2805#undef MNT_KERN_FLAG 2806 if (flags != 0) { 2807 if (buf[0] != '\0') 2808 strlcat(buf, ", ", sizeof(buf)); 2809 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2810 "0x%08x", flags); 2811 } 2812 db_printf(" mnt_kern_flag = %s\n", buf); 2813 2814 sp = &mp->mnt_stat; 2815 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 2816 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 2817 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 2818 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 2819 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 2820 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 2821 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 2822 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 2823 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 2824 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 2825 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 2826 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 2827 2828 db_printf(" mnt_cred = { uid=%u ruid=%u", 2829 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 2830 if (mp->mnt_cred->cr_prison != NULL) 2831 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 2832 db_printf(" }\n"); 2833 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 2834 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 2835 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 2836 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 2837 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync); 2838 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 2839 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 2840 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 2841 db_printf(" mnt_holdcnt = %d\n", mp->mnt_holdcnt); 2842 db_printf(" mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters); 2843 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 2844 db_printf(" mnt_secondary_accwrites = %d\n", 2845 mp->mnt_secondary_accwrites); 2846 db_printf(" mnt_gjprovider = %s\n", 2847 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 2848 db_printf("\n"); 2849 2850 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2851 if (vp->v_type != VMARKER) { 2852 vn_printf(vp, "vnode "); 2853 if (db_pager_quit) 2854 break; 2855 } 2856 } 2857} 2858#endif /* DDB */ 2859 2860/* 2861 * Fill in a struct xvfsconf based on a struct vfsconf. 2862 */ 2863static void 2864vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2865{ 2866 2867 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2868 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2869 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2870 xvfsp->vfc_flags = vfsp->vfc_flags; 2871 /* 2872 * These are unused in userland, we keep them 2873 * to not break binary compatibility. 2874 */ 2875 xvfsp->vfc_vfsops = NULL; 2876 xvfsp->vfc_next = NULL; 2877} 2878 2879/* 2880 * Top level filesystem related information gathering. 2881 */ 2882static int 2883sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2884{ 2885 struct vfsconf *vfsp; 2886 struct xvfsconf xvfsp; 2887 int error; 2888 2889 error = 0; 2890 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2891 bzero(&xvfsp, sizeof(xvfsp)); 2892 vfsconf2x(vfsp, &xvfsp); 2893 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 2894 if (error) 2895 break; 2896 } 2897 return (error); 2898} 2899 2900SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist, 2901 "S,xvfsconf", "List of all configured filesystems"); 2902 2903#ifndef BURN_BRIDGES 2904static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 2905 2906static int 2907vfs_sysctl(SYSCTL_HANDLER_ARGS) 2908{ 2909 int *name = (int *)arg1 - 1; /* XXX */ 2910 u_int namelen = arg2 + 1; /* XXX */ 2911 struct vfsconf *vfsp; 2912 struct xvfsconf xvfsp; 2913 2914 printf("WARNING: userland calling deprecated sysctl, " 2915 "please rebuild world\n"); 2916 2917#if 1 || defined(COMPAT_PRELITE2) 2918 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2919 if (namelen == 1) 2920 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2921#endif 2922 2923 switch (name[1]) { 2924 case VFS_MAXTYPENUM: 2925 if (namelen != 2) 2926 return (ENOTDIR); 2927 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2928 case VFS_CONF: 2929 if (namelen != 3) 2930 return (ENOTDIR); /* overloaded */ 2931 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 2932 if (vfsp->vfc_typenum == name[2]) 2933 break; 2934 if (vfsp == NULL) 2935 return (EOPNOTSUPP); 2936 bzero(&xvfsp, sizeof(xvfsp)); 2937 vfsconf2x(vfsp, &xvfsp); 2938 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 2939 } 2940 return (EOPNOTSUPP); 2941} 2942 2943static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 2944 vfs_sysctl, "Generic filesystem"); 2945 2946#if 1 || defined(COMPAT_PRELITE2) 2947 2948static int 2949sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2950{ 2951 int error; 2952 struct vfsconf *vfsp; 2953 struct ovfsconf ovfs; 2954 2955 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2956 bzero(&ovfs, sizeof(ovfs)); 2957 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2958 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2959 ovfs.vfc_index = vfsp->vfc_typenum; 2960 ovfs.vfc_refcount = vfsp->vfc_refcount; 2961 ovfs.vfc_flags = vfsp->vfc_flags; 2962 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2963 if (error) 2964 return error; 2965 } 2966 return 0; 2967} 2968 2969#endif /* 1 || COMPAT_PRELITE2 */ 2970#endif /* !BURN_BRIDGES */ 2971 2972#define KINFO_VNODESLOP 10 2973#ifdef notyet 2974/* 2975 * Dump vnode list (via sysctl). 2976 */ 2977/* ARGSUSED */ 2978static int 2979sysctl_vnode(SYSCTL_HANDLER_ARGS) 2980{ 2981 struct xvnode *xvn; 2982 struct mount *mp; 2983 struct vnode *vp; 2984 int error, len, n; 2985 2986 /* 2987 * Stale numvnodes access is not fatal here. 2988 */ 2989 req->lock = 0; 2990 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 2991 if (!req->oldptr) 2992 /* Make an estimate */ 2993 return (SYSCTL_OUT(req, 0, len)); 2994 2995 error = sysctl_wire_old_buffer(req, 0); 2996 if (error != 0) 2997 return (error); 2998 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 2999 n = 0; 3000 mtx_lock(&mountlist_mtx); 3001 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3002 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3003 continue; 3004 MNT_ILOCK(mp); 3005 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3006 if (n == len) 3007 break; 3008 vref(vp); 3009 xvn[n].xv_size = sizeof *xvn; 3010 xvn[n].xv_vnode = vp; 3011 xvn[n].xv_id = 0; /* XXX compat */ 3012#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3013 XV_COPY(usecount); 3014 XV_COPY(writecount); 3015 XV_COPY(holdcnt); 3016 XV_COPY(mount); 3017 XV_COPY(numoutput); 3018 XV_COPY(type); 3019#undef XV_COPY 3020 xvn[n].xv_flag = vp->v_vflag; 3021 3022 switch (vp->v_type) { 3023 case VREG: 3024 case VDIR: 3025 case VLNK: 3026 break; 3027 case VBLK: 3028 case VCHR: 3029 if (vp->v_rdev == NULL) { 3030 vrele(vp); 3031 continue; 3032 } 3033 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3034 break; 3035 case VSOCK: 3036 xvn[n].xv_socket = vp->v_socket; 3037 break; 3038 case VFIFO: 3039 xvn[n].xv_fifo = vp->v_fifoinfo; 3040 break; 3041 case VNON: 3042 case VBAD: 3043 default: 3044 /* shouldn't happen? */ 3045 vrele(vp); 3046 continue; 3047 } 3048 vrele(vp); 3049 ++n; 3050 } 3051 MNT_IUNLOCK(mp); 3052 mtx_lock(&mountlist_mtx); 3053 vfs_unbusy(mp); 3054 if (n == len) 3055 break; 3056 } 3057 mtx_unlock(&mountlist_mtx); 3058 3059 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3060 free(xvn, M_TEMP); 3061 return (error); 3062} 3063 3064SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3065 0, 0, sysctl_vnode, "S,xvnode", ""); 3066#endif 3067 3068/* 3069 * Unmount all filesystems. The list is traversed in reverse order 3070 * of mounting to avoid dependencies. 3071 */ 3072void 3073vfs_unmountall(void) 3074{ 3075 struct mount *mp; 3076 struct thread *td; 3077 int error; 3078 3079 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 3080 td = curthread; 3081 /* 3082 * Since this only runs when rebooting, it is not interlocked. 3083 */ 3084 while(!TAILQ_EMPTY(&mountlist)) { 3085 mp = TAILQ_LAST(&mountlist, mntlist); 3086 error = dounmount(mp, MNT_FORCE, td); 3087 if (error) { 3088 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3089 /* 3090 * XXX: Due to the way in which we mount the root 3091 * file system off of devfs, devfs will generate a 3092 * "busy" warning when we try to unmount it before 3093 * the root. Don't print a warning as a result in 3094 * order to avoid false positive errors that may 3095 * cause needless upset. 3096 */ 3097 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3098 printf("unmount of %s failed (", 3099 mp->mnt_stat.f_mntonname); 3100 if (error == EBUSY) 3101 printf("BUSY)\n"); 3102 else 3103 printf("%d)\n", error); 3104 } 3105 } else { 3106 /* The unmount has removed mp from the mountlist */ 3107 } 3108 } 3109} 3110 3111/* 3112 * perform msync on all vnodes under a mount point 3113 * the mount point must be locked. 3114 */ 3115void 3116vfs_msync(struct mount *mp, int flags) 3117{ 3118 struct vnode *vp, *mvp; 3119 struct vm_object *obj; 3120 3121 MNT_ILOCK(mp); 3122 MNT_VNODE_FOREACH(vp, mp, mvp) { 3123 VI_LOCK(vp); 3124 if ((vp->v_iflag & VI_OBJDIRTY) && 3125 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3126 MNT_IUNLOCK(mp); 3127 if (!vget(vp, 3128 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3129 curthread)) { 3130 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3131 vput(vp); 3132 MNT_ILOCK(mp); 3133 continue; 3134 } 3135 3136 obj = vp->v_object; 3137 if (obj != NULL) { 3138 VM_OBJECT_LOCK(obj); 3139 vm_object_page_clean(obj, 0, 0, 3140 flags == MNT_WAIT ? 3141 OBJPC_SYNC : OBJPC_NOSYNC); 3142 VM_OBJECT_UNLOCK(obj); 3143 } 3144 vput(vp); 3145 } 3146 MNT_ILOCK(mp); 3147 } else 3148 VI_UNLOCK(vp); 3149 } 3150 MNT_IUNLOCK(mp); 3151} 3152 3153/* 3154 * Mark a vnode as free, putting it up for recycling. 3155 */ 3156static void 3157vfree(struct vnode *vp) 3158{ 3159 3160 CTR1(KTR_VFS, "vfree vp %p", vp); 3161 ASSERT_VI_LOCKED(vp, "vfree"); 3162 mtx_lock(&vnode_free_list_mtx); 3163 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed.")); 3164 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); 3165 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't")); 3166 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp, 3167 ("vfree: Freeing doomed vnode")); 3168 if (vp->v_iflag & VI_AGE) { 3169 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 3170 } else { 3171 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 3172 } 3173 freevnodes++; 3174 vp->v_iflag &= ~VI_AGE; 3175 vp->v_iflag |= VI_FREE; 3176 mtx_unlock(&vnode_free_list_mtx); 3177} 3178 3179/* 3180 * Opposite of vfree() - mark a vnode as in use. 3181 */ 3182static void 3183vbusy(struct vnode *vp) 3184{ 3185 CTR1(KTR_VFS, "vbusy vp %p", vp); 3186 ASSERT_VI_LOCKED(vp, "vbusy"); 3187 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 3188 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed.")); 3189 3190 mtx_lock(&vnode_free_list_mtx); 3191 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 3192 freevnodes--; 3193 vp->v_iflag &= ~(VI_FREE|VI_AGE); 3194 mtx_unlock(&vnode_free_list_mtx); 3195} 3196 3197static void 3198destroy_vpollinfo(struct vpollinfo *vi) 3199{ 3200 knlist_destroy(&vi->vpi_selinfo.si_note); 3201 mtx_destroy(&vi->vpi_lock); 3202 uma_zfree(vnodepoll_zone, vi); 3203} 3204 3205/* 3206 * Initalize per-vnode helper structure to hold poll-related state. 3207 */ 3208void 3209v_addpollinfo(struct vnode *vp) 3210{ 3211 struct vpollinfo *vi; 3212 3213 if (vp->v_pollinfo != NULL) 3214 return; 3215 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3216 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3217 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3218 vfs_knlunlock, vfs_knllocked); 3219 VI_LOCK(vp); 3220 if (vp->v_pollinfo != NULL) { 3221 VI_UNLOCK(vp); 3222 destroy_vpollinfo(vi); 3223 return; 3224 } 3225 vp->v_pollinfo = vi; 3226 VI_UNLOCK(vp); 3227} 3228 3229/* 3230 * Record a process's interest in events which might happen to 3231 * a vnode. Because poll uses the historic select-style interface 3232 * internally, this routine serves as both the ``check for any 3233 * pending events'' and the ``record my interest in future events'' 3234 * functions. (These are done together, while the lock is held, 3235 * to avoid race conditions.) 3236 */ 3237int 3238vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3239{ 3240 3241 v_addpollinfo(vp); 3242 mtx_lock(&vp->v_pollinfo->vpi_lock); 3243 if (vp->v_pollinfo->vpi_revents & events) { 3244 /* 3245 * This leaves events we are not interested 3246 * in available for the other process which 3247 * which presumably had requested them 3248 * (otherwise they would never have been 3249 * recorded). 3250 */ 3251 events &= vp->v_pollinfo->vpi_revents; 3252 vp->v_pollinfo->vpi_revents &= ~events; 3253 3254 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3255 return (events); 3256 } 3257 vp->v_pollinfo->vpi_events |= events; 3258 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3259 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3260 return (0); 3261} 3262 3263/* 3264 * Routine to create and manage a filesystem syncer vnode. 3265 */ 3266#define sync_close ((int (*)(struct vop_close_args *))nullop) 3267static int sync_fsync(struct vop_fsync_args *); 3268static int sync_inactive(struct vop_inactive_args *); 3269static int sync_reclaim(struct vop_reclaim_args *); 3270 3271static struct vop_vector sync_vnodeops = { 3272 .vop_bypass = VOP_EOPNOTSUPP, 3273 .vop_close = sync_close, /* close */ 3274 .vop_fsync = sync_fsync, /* fsync */ 3275 .vop_inactive = sync_inactive, /* inactive */ 3276 .vop_reclaim = sync_reclaim, /* reclaim */ 3277 .vop_lock1 = vop_stdlock, /* lock */ 3278 .vop_unlock = vop_stdunlock, /* unlock */ 3279 .vop_islocked = vop_stdislocked, /* islocked */ 3280}; 3281 3282/* 3283 * Create a new filesystem syncer vnode for the specified mount point. 3284 */ 3285int 3286vfs_allocate_syncvnode(struct mount *mp) 3287{ 3288 struct vnode *vp; 3289 struct bufobj *bo; 3290 static long start, incr, next; 3291 int error; 3292 3293 /* Allocate a new vnode */ 3294 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) { 3295 mp->mnt_syncer = NULL; 3296 return (error); 3297 } 3298 vp->v_type = VNON; 3299 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3300 vp->v_vflag |= VV_FORCEINSMQ; 3301 error = insmntque(vp, mp); 3302 if (error != 0) 3303 panic("vfs_allocate_syncvnode: insmntque failed"); 3304 vp->v_vflag &= ~VV_FORCEINSMQ; 3305 VOP_UNLOCK(vp, 0); 3306 /* 3307 * Place the vnode onto the syncer worklist. We attempt to 3308 * scatter them about on the list so that they will go off 3309 * at evenly distributed times even if all the filesystems 3310 * are mounted at once. 3311 */ 3312 next += incr; 3313 if (next == 0 || next > syncer_maxdelay) { 3314 start /= 2; 3315 incr /= 2; 3316 if (start == 0) { 3317 start = syncer_maxdelay / 2; 3318 incr = syncer_maxdelay; 3319 } 3320 next = start; 3321 } 3322 bo = &vp->v_bufobj; 3323 BO_LOCK(bo); 3324 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3325 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3326 mtx_lock(&sync_mtx); 3327 sync_vnode_count++; 3328 mtx_unlock(&sync_mtx); 3329 BO_UNLOCK(bo); 3330 mp->mnt_syncer = vp; 3331 return (0); 3332} 3333 3334/* 3335 * Do a lazy sync of the filesystem. 3336 */ 3337static int 3338sync_fsync(struct vop_fsync_args *ap) 3339{ 3340 struct vnode *syncvp = ap->a_vp; 3341 struct mount *mp = syncvp->v_mount; 3342 int error; 3343 struct bufobj *bo; 3344 3345 /* 3346 * We only need to do something if this is a lazy evaluation. 3347 */ 3348 if (ap->a_waitfor != MNT_LAZY) 3349 return (0); 3350 3351 /* 3352 * Move ourselves to the back of the sync list. 3353 */ 3354 bo = &syncvp->v_bufobj; 3355 BO_LOCK(bo); 3356 vn_syncer_add_to_worklist(bo, syncdelay); 3357 BO_UNLOCK(bo); 3358 3359 /* 3360 * Walk the list of vnodes pushing all that are dirty and 3361 * not already on the sync list. 3362 */ 3363 mtx_lock(&mountlist_mtx); 3364 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { 3365 mtx_unlock(&mountlist_mtx); 3366 return (0); 3367 } 3368 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3369 vfs_unbusy(mp); 3370 return (0); 3371 } 3372 MNT_ILOCK(mp); 3373 mp->mnt_noasync++; 3374 mp->mnt_kern_flag &= ~MNTK_ASYNC; 3375 MNT_IUNLOCK(mp); 3376 vfs_msync(mp, MNT_NOWAIT); 3377 error = VFS_SYNC(mp, MNT_LAZY, ap->a_td); 3378 MNT_ILOCK(mp); 3379 mp->mnt_noasync--; 3380 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 3381 mp->mnt_kern_flag |= MNTK_ASYNC; 3382 MNT_IUNLOCK(mp); 3383 vn_finished_write(mp); 3384 vfs_unbusy(mp); 3385 return (error); 3386} 3387 3388/* 3389 * The syncer vnode is no referenced. 3390 */ 3391static int 3392sync_inactive(struct vop_inactive_args *ap) 3393{ 3394 3395 vgone(ap->a_vp); 3396 return (0); 3397} 3398 3399/* 3400 * The syncer vnode is no longer needed and is being decommissioned. 3401 * 3402 * Modifications to the worklist must be protected by sync_mtx. 3403 */ 3404static int 3405sync_reclaim(struct vop_reclaim_args *ap) 3406{ 3407 struct vnode *vp = ap->a_vp; 3408 struct bufobj *bo; 3409 3410 bo = &vp->v_bufobj; 3411 BO_LOCK(bo); 3412 vp->v_mount->mnt_syncer = NULL; 3413 if (bo->bo_flag & BO_ONWORKLST) { 3414 mtx_lock(&sync_mtx); 3415 LIST_REMOVE(bo, bo_synclist); 3416 syncer_worklist_len--; 3417 sync_vnode_count--; 3418 mtx_unlock(&sync_mtx); 3419 bo->bo_flag &= ~BO_ONWORKLST; 3420 } 3421 BO_UNLOCK(bo); 3422 3423 return (0); 3424} 3425 3426/* 3427 * Check if vnode represents a disk device 3428 */ 3429int 3430vn_isdisk(struct vnode *vp, int *errp) 3431{ 3432 int error; 3433 3434 error = 0; 3435 dev_lock(); 3436 if (vp->v_type != VCHR) 3437 error = ENOTBLK; 3438 else if (vp->v_rdev == NULL) 3439 error = ENXIO; 3440 else if (vp->v_rdev->si_devsw == NULL) 3441 error = ENXIO; 3442 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3443 error = ENOTBLK; 3444 dev_unlock(); 3445 if (errp != NULL) 3446 *errp = error; 3447 return (error == 0); 3448} 3449 3450/* 3451 * Common filesystem object access control check routine. Accepts a 3452 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3453 * and optional call-by-reference privused argument allowing vaccess() 3454 * to indicate to the caller whether privilege was used to satisfy the 3455 * request (obsoleted). Returns 0 on success, or an errno on failure. 3456 * 3457 * The ifdef'd CAPABILITIES version is here for reference, but is not 3458 * actually used. 3459 */ 3460int 3461vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3462 accmode_t accmode, struct ucred *cred, int *privused) 3463{ 3464 accmode_t dac_granted; 3465 accmode_t priv_granted; 3466 3467 /* 3468 * Look for a normal, non-privileged way to access the file/directory 3469 * as requested. If it exists, go with that. 3470 */ 3471 3472 if (privused != NULL) 3473 *privused = 0; 3474 3475 dac_granted = 0; 3476 3477 /* Check the owner. */ 3478 if (cred->cr_uid == file_uid) { 3479 dac_granted |= VADMIN; 3480 if (file_mode & S_IXUSR) 3481 dac_granted |= VEXEC; 3482 if (file_mode & S_IRUSR) 3483 dac_granted |= VREAD; 3484 if (file_mode & S_IWUSR) 3485 dac_granted |= (VWRITE | VAPPEND); 3486 3487 if ((accmode & dac_granted) == accmode) 3488 return (0); 3489 3490 goto privcheck; 3491 } 3492 3493 /* Otherwise, check the groups (first match) */ 3494 if (groupmember(file_gid, cred)) { 3495 if (file_mode & S_IXGRP) 3496 dac_granted |= VEXEC; 3497 if (file_mode & S_IRGRP) 3498 dac_granted |= VREAD; 3499 if (file_mode & S_IWGRP) 3500 dac_granted |= (VWRITE | VAPPEND); 3501 3502 if ((accmode & dac_granted) == accmode) 3503 return (0); 3504 3505 goto privcheck; 3506 } 3507 3508 /* Otherwise, check everyone else. */ 3509 if (file_mode & S_IXOTH) 3510 dac_granted |= VEXEC; 3511 if (file_mode & S_IROTH) 3512 dac_granted |= VREAD; 3513 if (file_mode & S_IWOTH) 3514 dac_granted |= (VWRITE | VAPPEND); 3515 if ((accmode & dac_granted) == accmode) 3516 return (0); 3517 3518privcheck: 3519 /* 3520 * Build a privilege mask to determine if the set of privileges 3521 * satisfies the requirements when combined with the granted mask 3522 * from above. For each privilege, if the privilege is required, 3523 * bitwise or the request type onto the priv_granted mask. 3524 */ 3525 priv_granted = 0; 3526 3527 if (type == VDIR) { 3528 /* 3529 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3530 * requests, instead of PRIV_VFS_EXEC. 3531 */ 3532 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3533 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3534 priv_granted |= VEXEC; 3535 } else { 3536 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3537 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3538 priv_granted |= VEXEC; 3539 } 3540 3541 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3542 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3543 priv_granted |= VREAD; 3544 3545 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3546 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3547 priv_granted |= (VWRITE | VAPPEND); 3548 3549 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3550 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3551 priv_granted |= VADMIN; 3552 3553 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3554 /* XXX audit: privilege used */ 3555 if (privused != NULL) 3556 *privused = 1; 3557 return (0); 3558 } 3559 3560 return ((accmode & VADMIN) ? EPERM : EACCES); 3561} 3562 3563/* 3564 * Credential check based on process requesting service, and per-attribute 3565 * permissions. 3566 */ 3567int 3568extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3569 struct thread *td, accmode_t accmode) 3570{ 3571 3572 /* 3573 * Kernel-invoked always succeeds. 3574 */ 3575 if (cred == NOCRED) 3576 return (0); 3577 3578 /* 3579 * Do not allow privileged processes in jail to directly manipulate 3580 * system attributes. 3581 */ 3582 switch (attrnamespace) { 3583 case EXTATTR_NAMESPACE_SYSTEM: 3584 /* Potentially should be: return (EPERM); */ 3585 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3586 case EXTATTR_NAMESPACE_USER: 3587 return (VOP_ACCESS(vp, accmode, cred, td)); 3588 default: 3589 return (EPERM); 3590 } 3591} 3592 3593#ifdef DEBUG_VFS_LOCKS 3594/* 3595 * This only exists to supress warnings from unlocked specfs accesses. It is 3596 * no longer ok to have an unlocked VFS. 3597 */ 3598#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3599 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3600 3601int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3602SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, ""); 3603 3604int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3605SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, ""); 3606 3607int vfs_badlock_print = 1; /* Print lock violations. */ 3608SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, ""); 3609 3610#ifdef KDB 3611int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3612SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, ""); 3613#endif 3614 3615static void 3616vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3617{ 3618 3619#ifdef KDB 3620 if (vfs_badlock_backtrace) 3621 kdb_backtrace(); 3622#endif 3623 if (vfs_badlock_print) 3624 printf("%s: %p %s\n", str, (void *)vp, msg); 3625 if (vfs_badlock_ddb) 3626 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3627} 3628 3629void 3630assert_vi_locked(struct vnode *vp, const char *str) 3631{ 3632 3633 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3634 vfs_badlock("interlock is not locked but should be", str, vp); 3635} 3636 3637void 3638assert_vi_unlocked(struct vnode *vp, const char *str) 3639{ 3640 3641 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3642 vfs_badlock("interlock is locked but should not be", str, vp); 3643} 3644 3645void 3646assert_vop_locked(struct vnode *vp, const char *str) 3647{ 3648 3649 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0) 3650 vfs_badlock("is not locked but should be", str, vp); 3651} 3652 3653void 3654assert_vop_unlocked(struct vnode *vp, const char *str) 3655{ 3656 3657 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 3658 vfs_badlock("is locked but should not be", str, vp); 3659} 3660 3661void 3662assert_vop_elocked(struct vnode *vp, const char *str) 3663{ 3664 3665 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 3666 vfs_badlock("is not exclusive locked but should be", str, vp); 3667} 3668 3669#if 0 3670void 3671assert_vop_elocked_other(struct vnode *vp, const char *str) 3672{ 3673 3674 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 3675 vfs_badlock("is not exclusive locked by another thread", 3676 str, vp); 3677} 3678 3679void 3680assert_vop_slocked(struct vnode *vp, const char *str) 3681{ 3682 3683 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 3684 vfs_badlock("is not locked shared but should be", str, vp); 3685} 3686#endif /* 0 */ 3687#endif /* DEBUG_VFS_LOCKS */ 3688 3689void 3690vop_rename_pre(void *ap) 3691{ 3692 struct vop_rename_args *a = ap; 3693 3694#ifdef DEBUG_VFS_LOCKS 3695 if (a->a_tvp) 3696 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3697 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3698 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3699 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3700 3701 /* Check the source (from). */ 3702 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp) 3703 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3704 if (a->a_tvp != a->a_fvp) 3705 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3706 3707 /* Check the target. */ 3708 if (a->a_tvp) 3709 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3710 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3711#endif 3712 if (a->a_tdvp != a->a_fdvp) 3713 vhold(a->a_fdvp); 3714 if (a->a_tvp != a->a_fvp) 3715 vhold(a->a_fvp); 3716 vhold(a->a_tdvp); 3717 if (a->a_tvp) 3718 vhold(a->a_tvp); 3719} 3720 3721void 3722vop_strategy_pre(void *ap) 3723{ 3724#ifdef DEBUG_VFS_LOCKS 3725 struct vop_strategy_args *a; 3726 struct buf *bp; 3727 3728 a = ap; 3729 bp = a->a_bp; 3730 3731 /* 3732 * Cluster ops lock their component buffers but not the IO container. 3733 */ 3734 if ((bp->b_flags & B_CLUSTER) != 0) 3735 return; 3736 3737 if (!BUF_ISLOCKED(bp)) { 3738 if (vfs_badlock_print) 3739 printf( 3740 "VOP_STRATEGY: bp is not locked but should be\n"); 3741 if (vfs_badlock_ddb) 3742 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3743 } 3744#endif 3745} 3746 3747void 3748vop_lookup_pre(void *ap) 3749{ 3750#ifdef DEBUG_VFS_LOCKS 3751 struct vop_lookup_args *a; 3752 struct vnode *dvp; 3753 3754 a = ap; 3755 dvp = a->a_dvp; 3756 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3757 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3758#endif 3759} 3760 3761void 3762vop_lookup_post(void *ap, int rc) 3763{ 3764#ifdef DEBUG_VFS_LOCKS 3765 struct vop_lookup_args *a; 3766 struct vnode *dvp; 3767 struct vnode *vp; 3768 3769 a = ap; 3770 dvp = a->a_dvp; 3771 vp = *(a->a_vpp); 3772 3773 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3774 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3775 3776 if (!rc) 3777 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 3778#endif 3779} 3780 3781void 3782vop_lock_pre(void *ap) 3783{ 3784#ifdef DEBUG_VFS_LOCKS 3785 struct vop_lock1_args *a = ap; 3786 3787 if ((a->a_flags & LK_INTERLOCK) == 0) 3788 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3789 else 3790 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3791#endif 3792} 3793 3794void 3795vop_lock_post(void *ap, int rc) 3796{ 3797#ifdef DEBUG_VFS_LOCKS 3798 struct vop_lock1_args *a = ap; 3799 3800 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3801 if (rc == 0) 3802 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3803#endif 3804} 3805 3806void 3807vop_unlock_pre(void *ap) 3808{ 3809#ifdef DEBUG_VFS_LOCKS 3810 struct vop_unlock_args *a = ap; 3811 3812 if (a->a_flags & LK_INTERLOCK) 3813 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3814 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3815#endif 3816} 3817 3818void 3819vop_unlock_post(void *ap, int rc) 3820{ 3821#ifdef DEBUG_VFS_LOCKS 3822 struct vop_unlock_args *a = ap; 3823 3824 if (a->a_flags & LK_INTERLOCK) 3825 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3826#endif 3827} 3828 3829void 3830vop_create_post(void *ap, int rc) 3831{ 3832 struct vop_create_args *a = ap; 3833 3834 if (!rc) 3835 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3836} 3837 3838void 3839vop_link_post(void *ap, int rc) 3840{ 3841 struct vop_link_args *a = ap; 3842 3843 if (!rc) { 3844 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 3845 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 3846 } 3847} 3848 3849void 3850vop_mkdir_post(void *ap, int rc) 3851{ 3852 struct vop_mkdir_args *a = ap; 3853 3854 if (!rc) 3855 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3856} 3857 3858void 3859vop_mknod_post(void *ap, int rc) 3860{ 3861 struct vop_mknod_args *a = ap; 3862 3863 if (!rc) 3864 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3865} 3866 3867void 3868vop_remove_post(void *ap, int rc) 3869{ 3870 struct vop_remove_args *a = ap; 3871 3872 if (!rc) { 3873 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3874 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 3875 } 3876} 3877 3878void 3879vop_rename_post(void *ap, int rc) 3880{ 3881 struct vop_rename_args *a = ap; 3882 3883 if (!rc) { 3884 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 3885 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 3886 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 3887 if (a->a_tvp) 3888 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 3889 } 3890 if (a->a_tdvp != a->a_fdvp) 3891 vdrop(a->a_fdvp); 3892 if (a->a_tvp != a->a_fvp) 3893 vdrop(a->a_fvp); 3894 vdrop(a->a_tdvp); 3895 if (a->a_tvp) 3896 vdrop(a->a_tvp); 3897} 3898 3899void 3900vop_rmdir_post(void *ap, int rc) 3901{ 3902 struct vop_rmdir_args *a = ap; 3903 3904 if (!rc) { 3905 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3906 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 3907 } 3908} 3909 3910void 3911vop_setattr_post(void *ap, int rc) 3912{ 3913 struct vop_setattr_args *a = ap; 3914 3915 if (!rc) 3916 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 3917} 3918 3919void 3920vop_symlink_post(void *ap, int rc) 3921{ 3922 struct vop_symlink_args *a = ap; 3923 3924 if (!rc) 3925 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3926} 3927 3928static struct knlist fs_knlist; 3929 3930static void 3931vfs_event_init(void *arg) 3932{ 3933 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL); 3934} 3935/* XXX - correct order? */ 3936SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 3937 3938void 3939vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused) 3940{ 3941 3942 KNOTE_UNLOCKED(&fs_knlist, event); 3943} 3944 3945static int filt_fsattach(struct knote *kn); 3946static void filt_fsdetach(struct knote *kn); 3947static int filt_fsevent(struct knote *kn, long hint); 3948 3949struct filterops fs_filtops = 3950 { 0, filt_fsattach, filt_fsdetach, filt_fsevent }; 3951 3952static int 3953filt_fsattach(struct knote *kn) 3954{ 3955 3956 kn->kn_flags |= EV_CLEAR; 3957 knlist_add(&fs_knlist, kn, 0); 3958 return (0); 3959} 3960 3961static void 3962filt_fsdetach(struct knote *kn) 3963{ 3964 3965 knlist_remove(&fs_knlist, kn, 0); 3966} 3967 3968static int 3969filt_fsevent(struct knote *kn, long hint) 3970{ 3971 3972 kn->kn_fflags |= hint; 3973 return (kn->kn_fflags != 0); 3974} 3975 3976static int 3977sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 3978{ 3979 struct vfsidctl vc; 3980 int error; 3981 struct mount *mp; 3982 3983 error = SYSCTL_IN(req, &vc, sizeof(vc)); 3984 if (error) 3985 return (error); 3986 if (vc.vc_vers != VFS_CTL_VERS1) 3987 return (EINVAL); 3988 mp = vfs_getvfs(&vc.vc_fsid); 3989 if (mp == NULL) 3990 return (ENOENT); 3991 /* ensure that a specific sysctl goes to the right filesystem. */ 3992 if (strcmp(vc.vc_fstypename, "*") != 0 && 3993 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 3994 vfs_rel(mp); 3995 return (EINVAL); 3996 } 3997 VCTLTOREQ(&vc, req); 3998 error = VFS_SYSCTL(mp, vc.vc_op, req); 3999 vfs_rel(mp); 4000 return (error); 4001} 4002 4003SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", 4004 "Sysctl by fsid"); 4005 4006/* 4007 * Function to initialize a va_filerev field sensibly. 4008 * XXX: Wouldn't a random number make a lot more sense ?? 4009 */ 4010u_quad_t 4011init_va_filerev(void) 4012{ 4013 struct bintime bt; 4014 4015 getbinuptime(&bt); 4016 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4017} 4018 4019static int filt_vfsread(struct knote *kn, long hint); 4020static int filt_vfswrite(struct knote *kn, long hint); 4021static int filt_vfsvnode(struct knote *kn, long hint); 4022static void filt_vfsdetach(struct knote *kn); 4023static struct filterops vfsread_filtops = 4024 { 1, NULL, filt_vfsdetach, filt_vfsread }; 4025static struct filterops vfswrite_filtops = 4026 { 1, NULL, filt_vfsdetach, filt_vfswrite }; 4027static struct filterops vfsvnode_filtops = 4028 { 1, NULL, filt_vfsdetach, filt_vfsvnode }; 4029 4030static void 4031vfs_knllock(void *arg) 4032{ 4033 struct vnode *vp = arg; 4034 4035 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4036} 4037 4038static void 4039vfs_knlunlock(void *arg) 4040{ 4041 struct vnode *vp = arg; 4042 4043 VOP_UNLOCK(vp, 0); 4044} 4045 4046static int 4047vfs_knllocked(void *arg) 4048{ 4049 struct vnode *vp = arg; 4050 4051 return (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 4052} 4053 4054int 4055vfs_kqfilter(struct vop_kqfilter_args *ap) 4056{ 4057 struct vnode *vp = ap->a_vp; 4058 struct knote *kn = ap->a_kn; 4059 struct knlist *knl; 4060 4061 switch (kn->kn_filter) { 4062 case EVFILT_READ: 4063 kn->kn_fop = &vfsread_filtops; 4064 break; 4065 case EVFILT_WRITE: 4066 kn->kn_fop = &vfswrite_filtops; 4067 break; 4068 case EVFILT_VNODE: 4069 kn->kn_fop = &vfsvnode_filtops; 4070 break; 4071 default: 4072 return (EINVAL); 4073 } 4074 4075 kn->kn_hook = (caddr_t)vp; 4076 4077 v_addpollinfo(vp); 4078 if (vp->v_pollinfo == NULL) 4079 return (ENOMEM); 4080 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4081 knlist_add(knl, kn, 0); 4082 4083 return (0); 4084} 4085 4086/* 4087 * Detach knote from vnode 4088 */ 4089static void 4090filt_vfsdetach(struct knote *kn) 4091{ 4092 struct vnode *vp = (struct vnode *)kn->kn_hook; 4093 4094 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4095 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4096} 4097 4098/*ARGSUSED*/ 4099static int 4100filt_vfsread(struct knote *kn, long hint) 4101{ 4102 struct vnode *vp = (struct vnode *)kn->kn_hook; 4103 struct vattr va; 4104 4105 /* 4106 * filesystem is gone, so set the EOF flag and schedule 4107 * the knote for deletion. 4108 */ 4109 if (hint == NOTE_REVOKE) { 4110 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4111 return (1); 4112 } 4113 4114 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4115 return (0); 4116 4117 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4118 return (kn->kn_data != 0); 4119} 4120 4121/*ARGSUSED*/ 4122static int 4123filt_vfswrite(struct knote *kn, long hint) 4124{ 4125 /* 4126 * filesystem is gone, so set the EOF flag and schedule 4127 * the knote for deletion. 4128 */ 4129 if (hint == NOTE_REVOKE) 4130 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4131 4132 kn->kn_data = 0; 4133 return (1); 4134} 4135 4136static int 4137filt_vfsvnode(struct knote *kn, long hint) 4138{ 4139 if (kn->kn_sfflags & hint) 4140 kn->kn_fflags |= hint; 4141 if (hint == NOTE_REVOKE) { 4142 kn->kn_flags |= EV_EOF; 4143 return (1); 4144 } 4145 return (kn->kn_fflags != 0); 4146} 4147 4148int 4149vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4150{ 4151 int error; 4152 4153 if (dp->d_reclen > ap->a_uio->uio_resid) 4154 return (ENAMETOOLONG); 4155 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4156 if (error) { 4157 if (ap->a_ncookies != NULL) { 4158 if (ap->a_cookies != NULL) 4159 free(ap->a_cookies, M_TEMP); 4160 ap->a_cookies = NULL; 4161 *ap->a_ncookies = 0; 4162 } 4163 return (error); 4164 } 4165 if (ap->a_ncookies == NULL) 4166 return (0); 4167 4168 KASSERT(ap->a_cookies, 4169 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4170 4171 *ap->a_cookies = realloc(*ap->a_cookies, 4172 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4173 (*ap->a_cookies)[*ap->a_ncookies] = off; 4174 return (0); 4175} 4176 4177/* 4178 * Mark for update the access time of the file if the filesystem 4179 * supports VA_MARK_ATIME. This functionality is used by execve 4180 * and mmap, so we want to avoid the synchronous I/O implied by 4181 * directly setting va_atime for the sake of efficiency. 4182 */ 4183void 4184vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4185{ 4186 struct vattr atimeattr; 4187 4188 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 4189 VATTR_NULL(&atimeattr); 4190 atimeattr.va_vaflags |= VA_MARK_ATIME; 4191 (void)VOP_SETATTR(vp, &atimeattr, cred); 4192 } 4193} 4194