vfs_subr.c revision 284203
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: stable/10/sys/kern/vfs_subr.c 284203 2015-06-10 02:27:00Z kib $"); 43 44#include "opt_compat.h" 45#include "opt_ddb.h" 46#include "opt_watchdog.h" 47 48#include <sys/param.h> 49#include <sys/systm.h> 50#include <sys/bio.h> 51#include <sys/buf.h> 52#include <sys/condvar.h> 53#include <sys/conf.h> 54#include <sys/dirent.h> 55#include <sys/event.h> 56#include <sys/eventhandler.h> 57#include <sys/extattr.h> 58#include <sys/file.h> 59#include <sys/fcntl.h> 60#include <sys/jail.h> 61#include <sys/kdb.h> 62#include <sys/kernel.h> 63#include <sys/kthread.h> 64#include <sys/lockf.h> 65#include <sys/malloc.h> 66#include <sys/mount.h> 67#include <sys/namei.h> 68#include <sys/pctrie.h> 69#include <sys/priv.h> 70#include <sys/reboot.h> 71#include <sys/rwlock.h> 72#include <sys/sched.h> 73#include <sys/sleepqueue.h> 74#include <sys/smp.h> 75#include <sys/stat.h> 76#include <sys/sysctl.h> 77#include <sys/syslog.h> 78#include <sys/vmmeter.h> 79#include <sys/vnode.h> 80#include <sys/watchdog.h> 81 82#include <machine/stdarg.h> 83 84#include <security/mac/mac_framework.h> 85 86#include <vm/vm.h> 87#include <vm/vm_object.h> 88#include <vm/vm_extern.h> 89#include <vm/pmap.h> 90#include <vm/vm_map.h> 91#include <vm/vm_page.h> 92#include <vm/vm_kern.h> 93#include <vm/uma.h> 94 95#ifdef DDB 96#include <ddb/ddb.h> 97#endif 98 99static void delmntque(struct vnode *vp); 100static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 101 int slpflag, int slptimeo); 102static void syncer_shutdown(void *arg, int howto); 103static int vtryrecycle(struct vnode *vp); 104static void v_incr_usecount(struct vnode *); 105static void v_decr_usecount(struct vnode *); 106static void v_decr_useonly(struct vnode *); 107static void v_upgrade_usecount(struct vnode *); 108static void vnlru_free(int); 109static void vgonel(struct vnode *); 110static void vfs_knllock(void *arg); 111static void vfs_knlunlock(void *arg); 112static void vfs_knl_assert_locked(void *arg); 113static void vfs_knl_assert_unlocked(void *arg); 114static void destroy_vpollinfo(struct vpollinfo *vi); 115 116/* 117 * Number of vnodes in existence. Increased whenever getnewvnode() 118 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode. 119 */ 120static unsigned long numvnodes; 121 122SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 123 "Number of vnodes in existence"); 124 125static u_long vnodes_created; 126SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, 127 0, "Number of vnodes created by getnewvnode"); 128 129/* 130 * Conversion tables for conversion from vnode types to inode formats 131 * and back. 132 */ 133enum vtype iftovt_tab[16] = { 134 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 135 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 136}; 137int vttoif_tab[10] = { 138 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 139 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 140}; 141 142/* 143 * List of vnodes that are ready for recycling. 144 */ 145static TAILQ_HEAD(freelst, vnode) vnode_free_list; 146 147/* 148 * Free vnode target. Free vnodes may simply be files which have been stat'd 149 * but not read. This is somewhat common, and a small cache of such files 150 * should be kept to avoid recreation costs. 151 */ 152static u_long wantfreevnodes; 153SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 154/* Number of vnodes in the free list. */ 155static u_long freevnodes; 156SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, 157 "Number of vnodes in the free list"); 158 159static int vlru_allow_cache_src; 160SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW, 161 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode"); 162 163static u_long recycles_count; 164SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0, 165 "Number of vnodes recycled to avoid exceding kern.maxvnodes"); 166 167/* 168 * Various variables used for debugging the new implementation of 169 * reassignbuf(). 170 * XXX these are probably of (very) limited utility now. 171 */ 172static int reassignbufcalls; 173SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, 174 "Number of calls to reassignbuf"); 175 176/* 177 * Cache for the mount type id assigned to NFS. This is used for 178 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 179 */ 180int nfs_mount_type = -1; 181 182/* To keep more than one thread at a time from running vfs_getnewfsid */ 183static struct mtx mntid_mtx; 184 185/* 186 * Lock for any access to the following: 187 * vnode_free_list 188 * numvnodes 189 * freevnodes 190 */ 191static struct mtx vnode_free_list_mtx; 192 193/* Publicly exported FS */ 194struct nfs_public nfs_pub; 195 196static uma_zone_t buf_trie_zone; 197 198/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 199static uma_zone_t vnode_zone; 200static uma_zone_t vnodepoll_zone; 201 202/* 203 * The workitem queue. 204 * 205 * It is useful to delay writes of file data and filesystem metadata 206 * for tens of seconds so that quickly created and deleted files need 207 * not waste disk bandwidth being created and removed. To realize this, 208 * we append vnodes to a "workitem" queue. When running with a soft 209 * updates implementation, most pending metadata dependencies should 210 * not wait for more than a few seconds. Thus, mounted on block devices 211 * are delayed only about a half the time that file data is delayed. 212 * Similarly, directory updates are more critical, so are only delayed 213 * about a third the time that file data is delayed. Thus, there are 214 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 215 * one each second (driven off the filesystem syncer process). The 216 * syncer_delayno variable indicates the next queue that is to be processed. 217 * Items that need to be processed soon are placed in this queue: 218 * 219 * syncer_workitem_pending[syncer_delayno] 220 * 221 * A delay of fifteen seconds is done by placing the request fifteen 222 * entries later in the queue: 223 * 224 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 225 * 226 */ 227static int syncer_delayno; 228static long syncer_mask; 229LIST_HEAD(synclist, bufobj); 230static struct synclist *syncer_workitem_pending; 231/* 232 * The sync_mtx protects: 233 * bo->bo_synclist 234 * sync_vnode_count 235 * syncer_delayno 236 * syncer_state 237 * syncer_workitem_pending 238 * syncer_worklist_len 239 * rushjob 240 */ 241static struct mtx sync_mtx; 242static struct cv sync_wakeup; 243 244#define SYNCER_MAXDELAY 32 245static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 246static int syncdelay = 30; /* max time to delay syncing data */ 247static int filedelay = 30; /* time to delay syncing files */ 248SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, 249 "Time to delay syncing files (in seconds)"); 250static int dirdelay = 29; /* time to delay syncing directories */ 251SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, 252 "Time to delay syncing directories (in seconds)"); 253static int metadelay = 28; /* time to delay syncing metadata */ 254SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, 255 "Time to delay syncing metadata (in seconds)"); 256static int rushjob; /* number of slots to run ASAP */ 257static int stat_rush_requests; /* number of times I/O speeded up */ 258SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, 259 "Number of times I/O speeded up (rush requests)"); 260 261/* 262 * When shutting down the syncer, run it at four times normal speed. 263 */ 264#define SYNCER_SHUTDOWN_SPEEDUP 4 265static int sync_vnode_count; 266static int syncer_worklist_len; 267static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 268 syncer_state; 269 270/* 271 * Number of vnodes we want to exist at any one time. This is mostly used 272 * to size hash tables in vnode-related code. It is normally not used in 273 * getnewvnode(), as wantfreevnodes is normally nonzero.) 274 * 275 * XXX desiredvnodes is historical cruft and should not exist. 276 */ 277int desiredvnodes; 278SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 279 &desiredvnodes, 0, "Maximum number of vnodes"); 280SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 281 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 282static int vnlru_nowhere; 283SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 284 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 285 286/* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ 287static int vnsz2log; 288 289/* 290 * Support for the bufobj clean & dirty pctrie. 291 */ 292static void * 293buf_trie_alloc(struct pctrie *ptree) 294{ 295 296 return uma_zalloc(buf_trie_zone, M_NOWAIT); 297} 298 299static void 300buf_trie_free(struct pctrie *ptree, void *node) 301{ 302 303 uma_zfree(buf_trie_zone, node); 304} 305PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free); 306 307/* 308 * Initialize the vnode management data structures. 309 * 310 * Reevaluate the following cap on the number of vnodes after the physical 311 * memory size exceeds 512GB. In the limit, as the physical memory size 312 * grows, the ratio of physical pages to vnodes approaches sixteen to one. 313 */ 314#ifndef MAXVNODES_MAX 315#define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16)) 316#endif 317static void 318vntblinit(void *dummy __unused) 319{ 320 u_int i; 321 int physvnodes, virtvnodes; 322 323 /* 324 * Desiredvnodes is a function of the physical memory size and the 325 * kernel's heap size. Generally speaking, it scales with the 326 * physical memory size. The ratio of desiredvnodes to physical pages 327 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the 328 * marginal ratio of desiredvnodes to physical pages is one to 329 * sixteen. However, desiredvnodes is limited by the kernel's heap 330 * size. The memory required by desiredvnodes vnodes and vm objects 331 * may not exceed one seventh of the kernel's heap size. 332 */ 333 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4, 334 cnt.v_page_count) / 16; 335 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) + 336 sizeof(struct vnode))); 337 desiredvnodes = min(physvnodes, virtvnodes); 338 if (desiredvnodes > MAXVNODES_MAX) { 339 if (bootverbose) 340 printf("Reducing kern.maxvnodes %d -> %d\n", 341 desiredvnodes, MAXVNODES_MAX); 342 desiredvnodes = MAXVNODES_MAX; 343 } 344 wantfreevnodes = desiredvnodes / 4; 345 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 346 TAILQ_INIT(&vnode_free_list); 347 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 348 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 349 NULL, NULL, UMA_ALIGN_PTR, 0); 350 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 351 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 352 /* 353 * Preallocate enough nodes to support one-per buf so that 354 * we can not fail an insert. reassignbuf() callers can not 355 * tolerate the insertion failure. 356 */ 357 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), 358 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, 359 UMA_ZONE_NOFREE | UMA_ZONE_VM); 360 uma_prealloc(buf_trie_zone, nbuf); 361 /* 362 * Initialize the filesystem syncer. 363 */ 364 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 365 &syncer_mask); 366 syncer_maxdelay = syncer_mask + 1; 367 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 368 cv_init(&sync_wakeup, "syncer"); 369 for (i = 1; i <= sizeof(struct vnode); i <<= 1) 370 vnsz2log++; 371 vnsz2log--; 372} 373SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 374 375 376/* 377 * Mark a mount point as busy. Used to synchronize access and to delay 378 * unmounting. Eventually, mountlist_mtx is not released on failure. 379 * 380 * vfs_busy() is a custom lock, it can block the caller. 381 * vfs_busy() only sleeps if the unmount is active on the mount point. 382 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any 383 * vnode belonging to mp. 384 * 385 * Lookup uses vfs_busy() to traverse mount points. 386 * root fs var fs 387 * / vnode lock A / vnode lock (/var) D 388 * /var vnode lock B /log vnode lock(/var/log) E 389 * vfs_busy lock C vfs_busy lock F 390 * 391 * Within each file system, the lock order is C->A->B and F->D->E. 392 * 393 * When traversing across mounts, the system follows that lock order: 394 * 395 * C->A->B 396 * | 397 * +->F->D->E 398 * 399 * The lookup() process for namei("/var") illustrates the process: 400 * VOP_LOOKUP() obtains B while A is held 401 * vfs_busy() obtains a shared lock on F while A and B are held 402 * vput() releases lock on B 403 * vput() releases lock on A 404 * VFS_ROOT() obtains lock on D while shared lock on F is held 405 * vfs_unbusy() releases shared lock on F 406 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. 407 * Attempt to lock A (instead of vp_crossmp) while D is held would 408 * violate the global order, causing deadlocks. 409 * 410 * dounmount() locks B while F is drained. 411 */ 412int 413vfs_busy(struct mount *mp, int flags) 414{ 415 416 MPASS((flags & ~MBF_MASK) == 0); 417 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); 418 419 MNT_ILOCK(mp); 420 MNT_REF(mp); 421 /* 422 * If mount point is currenly being unmounted, sleep until the 423 * mount point fate is decided. If thread doing the unmounting fails, 424 * it will clear MNTK_UNMOUNT flag before waking us up, indicating 425 * that this mount point has survived the unmount attempt and vfs_busy 426 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE 427 * flag in addition to MNTK_UNMOUNT, indicating that mount point is 428 * about to be really destroyed. vfs_busy needs to release its 429 * reference on the mount point in this case and return with ENOENT, 430 * telling the caller that mount mount it tried to busy is no longer 431 * valid. 432 */ 433 while (mp->mnt_kern_flag & MNTK_UNMOUNT) { 434 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { 435 MNT_REL(mp); 436 MNT_IUNLOCK(mp); 437 CTR1(KTR_VFS, "%s: failed busying before sleeping", 438 __func__); 439 return (ENOENT); 440 } 441 if (flags & MBF_MNTLSTLOCK) 442 mtx_unlock(&mountlist_mtx); 443 mp->mnt_kern_flag |= MNTK_MWAIT; 444 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); 445 if (flags & MBF_MNTLSTLOCK) 446 mtx_lock(&mountlist_mtx); 447 MNT_ILOCK(mp); 448 } 449 if (flags & MBF_MNTLSTLOCK) 450 mtx_unlock(&mountlist_mtx); 451 mp->mnt_lockref++; 452 MNT_IUNLOCK(mp); 453 return (0); 454} 455 456/* 457 * Free a busy filesystem. 458 */ 459void 460vfs_unbusy(struct mount *mp) 461{ 462 463 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 464 MNT_ILOCK(mp); 465 MNT_REL(mp); 466 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); 467 mp->mnt_lockref--; 468 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 469 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 470 CTR1(KTR_VFS, "%s: waking up waiters", __func__); 471 mp->mnt_kern_flag &= ~MNTK_DRAINING; 472 wakeup(&mp->mnt_lockref); 473 } 474 MNT_IUNLOCK(mp); 475} 476 477/* 478 * Lookup a mount point by filesystem identifier. 479 */ 480struct mount * 481vfs_getvfs(fsid_t *fsid) 482{ 483 struct mount *mp; 484 485 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 486 mtx_lock(&mountlist_mtx); 487 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 488 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 489 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 490 vfs_ref(mp); 491 mtx_unlock(&mountlist_mtx); 492 return (mp); 493 } 494 } 495 mtx_unlock(&mountlist_mtx); 496 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 497 return ((struct mount *) 0); 498} 499 500/* 501 * Lookup a mount point by filesystem identifier, busying it before 502 * returning. 503 * 504 * To avoid congestion on mountlist_mtx, implement simple direct-mapped 505 * cache for popular filesystem identifiers. The cache is lockess, using 506 * the fact that struct mount's are never freed. In worst case we may 507 * get pointer to unmounted or even different filesystem, so we have to 508 * check what we got, and go slow way if so. 509 */ 510struct mount * 511vfs_busyfs(fsid_t *fsid) 512{ 513#define FSID_CACHE_SIZE 256 514 typedef struct mount * volatile vmp_t; 515 static vmp_t cache[FSID_CACHE_SIZE]; 516 struct mount *mp; 517 int error; 518 uint32_t hash; 519 520 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 521 hash = fsid->val[0] ^ fsid->val[1]; 522 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); 523 mp = cache[hash]; 524 if (mp == NULL || 525 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] || 526 mp->mnt_stat.f_fsid.val[1] != fsid->val[1]) 527 goto slow; 528 if (vfs_busy(mp, 0) != 0) { 529 cache[hash] = NULL; 530 goto slow; 531 } 532 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 533 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) 534 return (mp); 535 else 536 vfs_unbusy(mp); 537 538slow: 539 mtx_lock(&mountlist_mtx); 540 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 541 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 542 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 543 error = vfs_busy(mp, MBF_MNTLSTLOCK); 544 if (error) { 545 cache[hash] = NULL; 546 mtx_unlock(&mountlist_mtx); 547 return (NULL); 548 } 549 cache[hash] = mp; 550 return (mp); 551 } 552 } 553 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 554 mtx_unlock(&mountlist_mtx); 555 return ((struct mount *) 0); 556} 557 558/* 559 * Check if a user can access privileged mount options. 560 */ 561int 562vfs_suser(struct mount *mp, struct thread *td) 563{ 564 int error; 565 566 /* 567 * If the thread is jailed, but this is not a jail-friendly file 568 * system, deny immediately. 569 */ 570 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) 571 return (EPERM); 572 573 /* 574 * If the file system was mounted outside the jail of the calling 575 * thread, deny immediately. 576 */ 577 if (prison_check(td->td_ucred, mp->mnt_cred) != 0) 578 return (EPERM); 579 580 /* 581 * If file system supports delegated administration, we don't check 582 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified 583 * by the file system itself. 584 * If this is not the user that did original mount, we check for 585 * the PRIV_VFS_MOUNT_OWNER privilege. 586 */ 587 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && 588 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 589 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 590 return (error); 591 } 592 return (0); 593} 594 595/* 596 * Get a new unique fsid. Try to make its val[0] unique, since this value 597 * will be used to create fake device numbers for stat(). Also try (but 598 * not so hard) make its val[0] unique mod 2^16, since some emulators only 599 * support 16-bit device numbers. We end up with unique val[0]'s for the 600 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 601 * 602 * Keep in mind that several mounts may be running in parallel. Starting 603 * the search one past where the previous search terminated is both a 604 * micro-optimization and a defense against returning the same fsid to 605 * different mounts. 606 */ 607void 608vfs_getnewfsid(struct mount *mp) 609{ 610 static uint16_t mntid_base; 611 struct mount *nmp; 612 fsid_t tfsid; 613 int mtype; 614 615 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 616 mtx_lock(&mntid_mtx); 617 mtype = mp->mnt_vfc->vfc_typenum; 618 tfsid.val[1] = mtype; 619 mtype = (mtype & 0xFF) << 24; 620 for (;;) { 621 tfsid.val[0] = makedev(255, 622 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 623 mntid_base++; 624 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 625 break; 626 vfs_rel(nmp); 627 } 628 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 629 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 630 mtx_unlock(&mntid_mtx); 631} 632 633/* 634 * Knob to control the precision of file timestamps: 635 * 636 * 0 = seconds only; nanoseconds zeroed. 637 * 1 = seconds and nanoseconds, accurate within 1/HZ. 638 * 2 = seconds and nanoseconds, truncated to microseconds. 639 * >=3 = seconds and nanoseconds, maximum precision. 640 */ 641enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 642 643static int timestamp_precision = TSP_USEC; 644SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 645 ×tamp_precision, 0, "File timestamp precision (0: seconds, " 646 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, " 647 "3+: sec + ns (max. precision))"); 648 649/* 650 * Get a current timestamp. 651 */ 652void 653vfs_timestamp(struct timespec *tsp) 654{ 655 struct timeval tv; 656 657 switch (timestamp_precision) { 658 case TSP_SEC: 659 tsp->tv_sec = time_second; 660 tsp->tv_nsec = 0; 661 break; 662 case TSP_HZ: 663 getnanotime(tsp); 664 break; 665 case TSP_USEC: 666 microtime(&tv); 667 TIMEVAL_TO_TIMESPEC(&tv, tsp); 668 break; 669 case TSP_NSEC: 670 default: 671 nanotime(tsp); 672 break; 673 } 674} 675 676/* 677 * Set vnode attributes to VNOVAL 678 */ 679void 680vattr_null(struct vattr *vap) 681{ 682 683 vap->va_type = VNON; 684 vap->va_size = VNOVAL; 685 vap->va_bytes = VNOVAL; 686 vap->va_mode = VNOVAL; 687 vap->va_nlink = VNOVAL; 688 vap->va_uid = VNOVAL; 689 vap->va_gid = VNOVAL; 690 vap->va_fsid = VNOVAL; 691 vap->va_fileid = VNOVAL; 692 vap->va_blocksize = VNOVAL; 693 vap->va_rdev = VNOVAL; 694 vap->va_atime.tv_sec = VNOVAL; 695 vap->va_atime.tv_nsec = VNOVAL; 696 vap->va_mtime.tv_sec = VNOVAL; 697 vap->va_mtime.tv_nsec = VNOVAL; 698 vap->va_ctime.tv_sec = VNOVAL; 699 vap->va_ctime.tv_nsec = VNOVAL; 700 vap->va_birthtime.tv_sec = VNOVAL; 701 vap->va_birthtime.tv_nsec = VNOVAL; 702 vap->va_flags = VNOVAL; 703 vap->va_gen = VNOVAL; 704 vap->va_vaflags = 0; 705} 706 707/* 708 * This routine is called when we have too many vnodes. It attempts 709 * to free <count> vnodes and will potentially free vnodes that still 710 * have VM backing store (VM backing store is typically the cause 711 * of a vnode blowout so we want to do this). Therefore, this operation 712 * is not considered cheap. 713 * 714 * A number of conditions may prevent a vnode from being reclaimed. 715 * the buffer cache may have references on the vnode, a directory 716 * vnode may still have references due to the namei cache representing 717 * underlying files, or the vnode may be in active use. It is not 718 * desireable to reuse such vnodes. These conditions may cause the 719 * number of vnodes to reach some minimum value regardless of what 720 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 721 */ 722static int 723vlrureclaim(struct mount *mp) 724{ 725 struct vnode *vp; 726 int done; 727 int trigger; 728 int usevnodes; 729 int count; 730 731 /* 732 * Calculate the trigger point, don't allow user 733 * screwups to blow us up. This prevents us from 734 * recycling vnodes with lots of resident pages. We 735 * aren't trying to free memory, we are trying to 736 * free vnodes. 737 */ 738 usevnodes = desiredvnodes; 739 if (usevnodes <= 0) 740 usevnodes = 1; 741 trigger = cnt.v_page_count * 2 / usevnodes; 742 done = 0; 743 vn_start_write(NULL, &mp, V_WAIT); 744 MNT_ILOCK(mp); 745 count = mp->mnt_nvnodelistsize / 10 + 1; 746 while (count != 0) { 747 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 748 while (vp != NULL && vp->v_type == VMARKER) 749 vp = TAILQ_NEXT(vp, v_nmntvnodes); 750 if (vp == NULL) 751 break; 752 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 753 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 754 --count; 755 if (!VI_TRYLOCK(vp)) 756 goto next_iter; 757 /* 758 * If it's been deconstructed already, it's still 759 * referenced, or it exceeds the trigger, skip it. 760 */ 761 if (vp->v_usecount || 762 (!vlru_allow_cache_src && 763 !LIST_EMPTY(&(vp)->v_cache_src)) || 764 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 765 vp->v_object->resident_page_count > trigger)) { 766 VI_UNLOCK(vp); 767 goto next_iter; 768 } 769 MNT_IUNLOCK(mp); 770 vholdl(vp); 771 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 772 vdrop(vp); 773 goto next_iter_mntunlocked; 774 } 775 VI_LOCK(vp); 776 /* 777 * v_usecount may have been bumped after VOP_LOCK() dropped 778 * the vnode interlock and before it was locked again. 779 * 780 * It is not necessary to recheck VI_DOOMED because it can 781 * only be set by another thread that holds both the vnode 782 * lock and vnode interlock. If another thread has the 783 * vnode lock before we get to VOP_LOCK() and obtains the 784 * vnode interlock after VOP_LOCK() drops the vnode 785 * interlock, the other thread will be unable to drop the 786 * vnode lock before our VOP_LOCK() call fails. 787 */ 788 if (vp->v_usecount || 789 (!vlru_allow_cache_src && 790 !LIST_EMPTY(&(vp)->v_cache_src)) || 791 (vp->v_object != NULL && 792 vp->v_object->resident_page_count > trigger)) { 793 VOP_UNLOCK(vp, LK_INTERLOCK); 794 vdrop(vp); 795 goto next_iter_mntunlocked; 796 } 797 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 798 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 799 atomic_add_long(&recycles_count, 1); 800 vgonel(vp); 801 VOP_UNLOCK(vp, 0); 802 vdropl(vp); 803 done++; 804next_iter_mntunlocked: 805 if (!should_yield()) 806 goto relock_mnt; 807 goto yield; 808next_iter: 809 if (!should_yield()) 810 continue; 811 MNT_IUNLOCK(mp); 812yield: 813 kern_yield(PRI_USER); 814relock_mnt: 815 MNT_ILOCK(mp); 816 } 817 MNT_IUNLOCK(mp); 818 vn_finished_write(mp); 819 return done; 820} 821 822/* 823 * Attempt to keep the free list at wantfreevnodes length. 824 */ 825static void 826vnlru_free(int count) 827{ 828 struct vnode *vp; 829 830 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 831 for (; count > 0; count--) { 832 vp = TAILQ_FIRST(&vnode_free_list); 833 /* 834 * The list can be modified while the free_list_mtx 835 * has been dropped and vp could be NULL here. 836 */ 837 if (!vp) 838 break; 839 VNASSERT(vp->v_op != NULL, vp, 840 ("vnlru_free: vnode already reclaimed.")); 841 KASSERT((vp->v_iflag & VI_FREE) != 0, 842 ("Removing vnode not on freelist")); 843 KASSERT((vp->v_iflag & VI_ACTIVE) == 0, 844 ("Mangling active vnode")); 845 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); 846 /* 847 * Don't recycle if we can't get the interlock. 848 */ 849 if (!VI_TRYLOCK(vp)) { 850 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); 851 continue; 852 } 853 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0, 854 vp, ("vp inconsistent on freelist")); 855 856 /* 857 * The clear of VI_FREE prevents activation of the 858 * vnode. There is no sense in putting the vnode on 859 * the mount point active list, only to remove it 860 * later during recycling. Inline the relevant part 861 * of vholdl(), to avoid triggering assertions or 862 * activating. 863 */ 864 freevnodes--; 865 vp->v_iflag &= ~VI_FREE; 866 vp->v_holdcnt++; 867 868 mtx_unlock(&vnode_free_list_mtx); 869 VI_UNLOCK(vp); 870 vtryrecycle(vp); 871 /* 872 * If the recycled succeeded this vdrop will actually free 873 * the vnode. If not it will simply place it back on 874 * the free list. 875 */ 876 vdrop(vp); 877 mtx_lock(&vnode_free_list_mtx); 878 } 879} 880/* 881 * Attempt to recycle vnodes in a context that is always safe to block. 882 * Calling vlrurecycle() from the bowels of filesystem code has some 883 * interesting deadlock problems. 884 */ 885static struct proc *vnlruproc; 886static int vnlruproc_sig; 887 888static void 889vnlru_proc(void) 890{ 891 struct mount *mp, *nmp; 892 int done; 893 struct proc *p = vnlruproc; 894 895 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 896 SHUTDOWN_PRI_FIRST); 897 898 for (;;) { 899 kproc_suspend_check(p); 900 mtx_lock(&vnode_free_list_mtx); 901 if (freevnodes > wantfreevnodes) 902 vnlru_free(freevnodes - wantfreevnodes); 903 if (numvnodes <= desiredvnodes * 9 / 10) { 904 vnlruproc_sig = 0; 905 wakeup(&vnlruproc_sig); 906 msleep(vnlruproc, &vnode_free_list_mtx, 907 PVFS|PDROP, "vlruwt", hz); 908 continue; 909 } 910 mtx_unlock(&vnode_free_list_mtx); 911 done = 0; 912 mtx_lock(&mountlist_mtx); 913 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 914 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 915 nmp = TAILQ_NEXT(mp, mnt_list); 916 continue; 917 } 918 done += vlrureclaim(mp); 919 mtx_lock(&mountlist_mtx); 920 nmp = TAILQ_NEXT(mp, mnt_list); 921 vfs_unbusy(mp); 922 } 923 mtx_unlock(&mountlist_mtx); 924 if (done == 0) { 925#if 0 926 /* These messages are temporary debugging aids */ 927 if (vnlru_nowhere < 5) 928 printf("vnlru process getting nowhere..\n"); 929 else if (vnlru_nowhere == 5) 930 printf("vnlru process messages stopped.\n"); 931#endif 932 vnlru_nowhere++; 933 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 934 } else 935 kern_yield(PRI_USER); 936 } 937} 938 939static struct kproc_desc vnlru_kp = { 940 "vnlru", 941 vnlru_proc, 942 &vnlruproc 943}; 944SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 945 &vnlru_kp); 946 947/* 948 * Routines having to do with the management of the vnode table. 949 */ 950 951/* 952 * Try to recycle a freed vnode. We abort if anyone picks up a reference 953 * before we actually vgone(). This function must be called with the vnode 954 * held to prevent the vnode from being returned to the free list midway 955 * through vgone(). 956 */ 957static int 958vtryrecycle(struct vnode *vp) 959{ 960 struct mount *vnmp; 961 962 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 963 VNASSERT(vp->v_holdcnt, vp, 964 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 965 /* 966 * This vnode may found and locked via some other list, if so we 967 * can't recycle it yet. 968 */ 969 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 970 CTR2(KTR_VFS, 971 "%s: impossible to recycle, vp %p lock is already held", 972 __func__, vp); 973 return (EWOULDBLOCK); 974 } 975 /* 976 * Don't recycle if its filesystem is being suspended. 977 */ 978 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 979 VOP_UNLOCK(vp, 0); 980 CTR2(KTR_VFS, 981 "%s: impossible to recycle, cannot start the write for %p", 982 __func__, vp); 983 return (EBUSY); 984 } 985 /* 986 * If we got this far, we need to acquire the interlock and see if 987 * anyone picked up this vnode from another list. If not, we will 988 * mark it with DOOMED via vgonel() so that anyone who does find it 989 * will skip over it. 990 */ 991 VI_LOCK(vp); 992 if (vp->v_usecount) { 993 VOP_UNLOCK(vp, LK_INTERLOCK); 994 vn_finished_write(vnmp); 995 CTR2(KTR_VFS, 996 "%s: impossible to recycle, %p is already referenced", 997 __func__, vp); 998 return (EBUSY); 999 } 1000 if ((vp->v_iflag & VI_DOOMED) == 0) { 1001 atomic_add_long(&recycles_count, 1); 1002 vgonel(vp); 1003 } 1004 VOP_UNLOCK(vp, LK_INTERLOCK); 1005 vn_finished_write(vnmp); 1006 return (0); 1007} 1008 1009/* 1010 * Wait for available vnodes. 1011 */ 1012static int 1013getnewvnode_wait(int suspended) 1014{ 1015 1016 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 1017 if (numvnodes > desiredvnodes) { 1018 if (suspended) { 1019 /* 1020 * File system is beeing suspended, we cannot risk a 1021 * deadlock here, so allocate new vnode anyway. 1022 */ 1023 if (freevnodes > wantfreevnodes) 1024 vnlru_free(freevnodes - wantfreevnodes); 1025 return (0); 1026 } 1027 if (vnlruproc_sig == 0) { 1028 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 1029 wakeup(vnlruproc); 1030 } 1031 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 1032 "vlruwk", hz); 1033 } 1034 return (numvnodes > desiredvnodes ? ENFILE : 0); 1035} 1036 1037void 1038getnewvnode_reserve(u_int count) 1039{ 1040 struct thread *td; 1041 1042 td = curthread; 1043 /* First try to be quick and racy. */ 1044 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) { 1045 td->td_vp_reserv += count; 1046 return; 1047 } else 1048 atomic_subtract_long(&numvnodes, count); 1049 1050 mtx_lock(&vnode_free_list_mtx); 1051 while (count > 0) { 1052 if (getnewvnode_wait(0) == 0) { 1053 count--; 1054 td->td_vp_reserv++; 1055 atomic_add_long(&numvnodes, 1); 1056 } 1057 } 1058 mtx_unlock(&vnode_free_list_mtx); 1059} 1060 1061void 1062getnewvnode_drop_reserve(void) 1063{ 1064 struct thread *td; 1065 1066 td = curthread; 1067 atomic_subtract_long(&numvnodes, td->td_vp_reserv); 1068 td->td_vp_reserv = 0; 1069} 1070 1071/* 1072 * Return the next vnode from the free list. 1073 */ 1074int 1075getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 1076 struct vnode **vpp) 1077{ 1078 struct vnode *vp; 1079 struct bufobj *bo; 1080 struct thread *td; 1081 int error; 1082 1083 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); 1084 vp = NULL; 1085 td = curthread; 1086 if (td->td_vp_reserv > 0) { 1087 td->td_vp_reserv -= 1; 1088 goto alloc; 1089 } 1090 mtx_lock(&vnode_free_list_mtx); 1091 /* 1092 * Lend our context to reclaim vnodes if they've exceeded the max. 1093 */ 1094 if (freevnodes > wantfreevnodes) 1095 vnlru_free(1); 1096 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag & 1097 MNTK_SUSPEND)); 1098#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 1099 if (error != 0) { 1100 mtx_unlock(&vnode_free_list_mtx); 1101 return (error); 1102 } 1103#endif 1104 atomic_add_long(&numvnodes, 1); 1105 mtx_unlock(&vnode_free_list_mtx); 1106alloc: 1107 atomic_add_long(&vnodes_created, 1); 1108 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 1109 /* 1110 * Setup locks. 1111 */ 1112 vp->v_vnlock = &vp->v_lock; 1113 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 1114 /* 1115 * By default, don't allow shared locks unless filesystems 1116 * opt-in. 1117 */ 1118 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); 1119 /* 1120 * Initialize bufobj. 1121 */ 1122 bo = &vp->v_bufobj; 1123 bo->__bo_vnode = vp; 1124 rw_init(BO_LOCKPTR(bo), "bufobj interlock"); 1125 bo->bo_ops = &buf_ops_bio; 1126 bo->bo_private = vp; 1127 TAILQ_INIT(&bo->bo_clean.bv_hd); 1128 TAILQ_INIT(&bo->bo_dirty.bv_hd); 1129 /* 1130 * Initialize namecache. 1131 */ 1132 LIST_INIT(&vp->v_cache_src); 1133 TAILQ_INIT(&vp->v_cache_dst); 1134 /* 1135 * Finalize various vnode identity bits. 1136 */ 1137 vp->v_type = VNON; 1138 vp->v_tag = tag; 1139 vp->v_op = vops; 1140 v_incr_usecount(vp); 1141 vp->v_data = NULL; 1142#ifdef MAC 1143 mac_vnode_init(vp); 1144 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 1145 mac_vnode_associate_singlelabel(mp, vp); 1146 else if (mp == NULL && vops != &dead_vnodeops) 1147 printf("NULL mp in getnewvnode()\n"); 1148#endif 1149 if (mp != NULL) { 1150 bo->bo_bsize = mp->mnt_stat.f_iosize; 1151 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 1152 vp->v_vflag |= VV_NOKNOTE; 1153 } 1154 rangelock_init(&vp->v_rl); 1155 1156 /* 1157 * For the filesystems which do not use vfs_hash_insert(), 1158 * still initialize v_hash to have vfs_hash_index() useful. 1159 * E.g., nullfs uses vfs_hash_index() on the lower vnode for 1160 * its own hashing. 1161 */ 1162 vp->v_hash = (uintptr_t)vp >> vnsz2log; 1163 1164 *vpp = vp; 1165 return (0); 1166} 1167 1168/* 1169 * Delete from old mount point vnode list, if on one. 1170 */ 1171static void 1172delmntque(struct vnode *vp) 1173{ 1174 struct mount *mp; 1175 int active; 1176 1177 mp = vp->v_mount; 1178 if (mp == NULL) 1179 return; 1180 MNT_ILOCK(mp); 1181 VI_LOCK(vp); 1182 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize, 1183 ("Active vnode list size %d > Vnode list size %d", 1184 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize)); 1185 active = vp->v_iflag & VI_ACTIVE; 1186 vp->v_iflag &= ~VI_ACTIVE; 1187 if (active) { 1188 mtx_lock(&vnode_free_list_mtx); 1189 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); 1190 mp->mnt_activevnodelistsize--; 1191 mtx_unlock(&vnode_free_list_mtx); 1192 } 1193 vp->v_mount = NULL; 1194 VI_UNLOCK(vp); 1195 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 1196 ("bad mount point vnode list size")); 1197 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1198 mp->mnt_nvnodelistsize--; 1199 MNT_REL(mp); 1200 MNT_IUNLOCK(mp); 1201} 1202 1203static void 1204insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1205{ 1206 1207 vp->v_data = NULL; 1208 vp->v_op = &dead_vnodeops; 1209 vgone(vp); 1210 vput(vp); 1211} 1212 1213/* 1214 * Insert into list of vnodes for the new mount point, if available. 1215 */ 1216int 1217insmntque1(struct vnode *vp, struct mount *mp, 1218 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1219{ 1220 1221 KASSERT(vp->v_mount == NULL, 1222 ("insmntque: vnode already on per mount vnode list")); 1223 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1224 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); 1225 1226 /* 1227 * We acquire the vnode interlock early to ensure that the 1228 * vnode cannot be recycled by another process releasing a 1229 * holdcnt on it before we get it on both the vnode list 1230 * and the active vnode list. The mount mutex protects only 1231 * manipulation of the vnode list and the vnode freelist 1232 * mutex protects only manipulation of the active vnode list. 1233 * Hence the need to hold the vnode interlock throughout. 1234 */ 1235 MNT_ILOCK(mp); 1236 VI_LOCK(vp); 1237 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1238 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1239 mp->mnt_nvnodelistsize == 0)) && 1240 (vp->v_vflag & VV_FORCEINSMQ) == 0) { 1241 VI_UNLOCK(vp); 1242 MNT_IUNLOCK(mp); 1243 if (dtr != NULL) 1244 dtr(vp, dtr_arg); 1245 return (EBUSY); 1246 } 1247 vp->v_mount = mp; 1248 MNT_REF(mp); 1249 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1250 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1251 ("neg mount point vnode list size")); 1252 mp->mnt_nvnodelistsize++; 1253 KASSERT((vp->v_iflag & VI_ACTIVE) == 0, 1254 ("Activating already active vnode")); 1255 vp->v_iflag |= VI_ACTIVE; 1256 mtx_lock(&vnode_free_list_mtx); 1257 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); 1258 mp->mnt_activevnodelistsize++; 1259 mtx_unlock(&vnode_free_list_mtx); 1260 VI_UNLOCK(vp); 1261 MNT_IUNLOCK(mp); 1262 return (0); 1263} 1264 1265int 1266insmntque(struct vnode *vp, struct mount *mp) 1267{ 1268 1269 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1270} 1271 1272/* 1273 * Flush out and invalidate all buffers associated with a bufobj 1274 * Called with the underlying object locked. 1275 */ 1276int 1277bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1278{ 1279 int error; 1280 1281 BO_LOCK(bo); 1282 if (flags & V_SAVE) { 1283 error = bufobj_wwait(bo, slpflag, slptimeo); 1284 if (error) { 1285 BO_UNLOCK(bo); 1286 return (error); 1287 } 1288 if (bo->bo_dirty.bv_cnt > 0) { 1289 BO_UNLOCK(bo); 1290 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1291 return (error); 1292 /* 1293 * XXX We could save a lock/unlock if this was only 1294 * enabled under INVARIANTS 1295 */ 1296 BO_LOCK(bo); 1297 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1298 panic("vinvalbuf: dirty bufs"); 1299 } 1300 } 1301 /* 1302 * If you alter this loop please notice that interlock is dropped and 1303 * reacquired in flushbuflist. Special care is needed to ensure that 1304 * no race conditions occur from this. 1305 */ 1306 do { 1307 error = flushbuflist(&bo->bo_clean, 1308 flags, bo, slpflag, slptimeo); 1309 if (error == 0 && !(flags & V_CLEANONLY)) 1310 error = flushbuflist(&bo->bo_dirty, 1311 flags, bo, slpflag, slptimeo); 1312 if (error != 0 && error != EAGAIN) { 1313 BO_UNLOCK(bo); 1314 return (error); 1315 } 1316 } while (error != 0); 1317 1318 /* 1319 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1320 * have write I/O in-progress but if there is a VM object then the 1321 * VM object can also have read-I/O in-progress. 1322 */ 1323 do { 1324 bufobj_wwait(bo, 0, 0); 1325 BO_UNLOCK(bo); 1326 if (bo->bo_object != NULL) { 1327 VM_OBJECT_WLOCK(bo->bo_object); 1328 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1329 VM_OBJECT_WUNLOCK(bo->bo_object); 1330 } 1331 BO_LOCK(bo); 1332 } while (bo->bo_numoutput > 0); 1333 BO_UNLOCK(bo); 1334 1335 /* 1336 * Destroy the copy in the VM cache, too. 1337 */ 1338 if (bo->bo_object != NULL && 1339 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) { 1340 VM_OBJECT_WLOCK(bo->bo_object); 1341 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? 1342 OBJPR_CLEANONLY : 0); 1343 VM_OBJECT_WUNLOCK(bo->bo_object); 1344 } 1345 1346#ifdef INVARIANTS 1347 BO_LOCK(bo); 1348 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 && 1349 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1350 panic("vinvalbuf: flush failed"); 1351 BO_UNLOCK(bo); 1352#endif 1353 return (0); 1354} 1355 1356/* 1357 * Flush out and invalidate all buffers associated with a vnode. 1358 * Called with the underlying object locked. 1359 */ 1360int 1361vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1362{ 1363 1364 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 1365 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1366 if (vp->v_object != NULL && vp->v_object->handle != vp) 1367 return (0); 1368 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1369} 1370 1371/* 1372 * Flush out buffers on the specified list. 1373 * 1374 */ 1375static int 1376flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1377 int slptimeo) 1378{ 1379 struct buf *bp, *nbp; 1380 int retval, error; 1381 daddr_t lblkno; 1382 b_xflags_t xflags; 1383 1384 ASSERT_BO_WLOCKED(bo); 1385 1386 retval = 0; 1387 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1388 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1389 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1390 continue; 1391 } 1392 lblkno = 0; 1393 xflags = 0; 1394 if (nbp != NULL) { 1395 lblkno = nbp->b_lblkno; 1396 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); 1397 } 1398 retval = EAGAIN; 1399 error = BUF_TIMELOCK(bp, 1400 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), 1401 "flushbuf", slpflag, slptimeo); 1402 if (error) { 1403 BO_LOCK(bo); 1404 return (error != ENOLCK ? error : EAGAIN); 1405 } 1406 KASSERT(bp->b_bufobj == bo, 1407 ("bp %p wrong b_bufobj %p should be %p", 1408 bp, bp->b_bufobj, bo)); 1409 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1410 BUF_UNLOCK(bp); 1411 BO_LOCK(bo); 1412 return (EAGAIN); 1413 } 1414 /* 1415 * XXX Since there are no node locks for NFS, I 1416 * believe there is a slight chance that a delayed 1417 * write will occur while sleeping just above, so 1418 * check for it. 1419 */ 1420 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1421 (flags & V_SAVE)) { 1422 bremfree(bp); 1423 bp->b_flags |= B_ASYNC; 1424 bwrite(bp); 1425 BO_LOCK(bo); 1426 return (EAGAIN); /* XXX: why not loop ? */ 1427 } 1428 bremfree(bp); 1429 bp->b_flags |= (B_INVAL | B_RELBUF); 1430 bp->b_flags &= ~B_ASYNC; 1431 brelse(bp); 1432 BO_LOCK(bo); 1433 if (nbp != NULL && 1434 (nbp->b_bufobj != bo || 1435 nbp->b_lblkno != lblkno || 1436 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1437 break; /* nbp invalid */ 1438 } 1439 return (retval); 1440} 1441 1442/* 1443 * Truncate a file's buffer and pages to a specified length. This 1444 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1445 * sync activity. 1446 */ 1447int 1448vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize) 1449{ 1450 struct buf *bp, *nbp; 1451 int anyfreed; 1452 int trunclbn; 1453 struct bufobj *bo; 1454 1455 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, 1456 vp, cred, blksize, (uintmax_t)length); 1457 1458 /* 1459 * Round up to the *next* lbn. 1460 */ 1461 trunclbn = (length + blksize - 1) / blksize; 1462 1463 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1464restart: 1465 bo = &vp->v_bufobj; 1466 BO_LOCK(bo); 1467 anyfreed = 1; 1468 for (;anyfreed;) { 1469 anyfreed = 0; 1470 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1471 if (bp->b_lblkno < trunclbn) 1472 continue; 1473 if (BUF_LOCK(bp, 1474 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1475 BO_LOCKPTR(bo)) == ENOLCK) 1476 goto restart; 1477 1478 bremfree(bp); 1479 bp->b_flags |= (B_INVAL | B_RELBUF); 1480 bp->b_flags &= ~B_ASYNC; 1481 brelse(bp); 1482 anyfreed = 1; 1483 1484 BO_LOCK(bo); 1485 if (nbp != NULL && 1486 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1487 (nbp->b_vp != vp) || 1488 (nbp->b_flags & B_DELWRI))) { 1489 BO_UNLOCK(bo); 1490 goto restart; 1491 } 1492 } 1493 1494 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1495 if (bp->b_lblkno < trunclbn) 1496 continue; 1497 if (BUF_LOCK(bp, 1498 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1499 BO_LOCKPTR(bo)) == ENOLCK) 1500 goto restart; 1501 bremfree(bp); 1502 bp->b_flags |= (B_INVAL | B_RELBUF); 1503 bp->b_flags &= ~B_ASYNC; 1504 brelse(bp); 1505 anyfreed = 1; 1506 1507 BO_LOCK(bo); 1508 if (nbp != NULL && 1509 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1510 (nbp->b_vp != vp) || 1511 (nbp->b_flags & B_DELWRI) == 0)) { 1512 BO_UNLOCK(bo); 1513 goto restart; 1514 } 1515 } 1516 } 1517 1518 if (length > 0) { 1519restartsync: 1520 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1521 if (bp->b_lblkno > 0) 1522 continue; 1523 /* 1524 * Since we hold the vnode lock this should only 1525 * fail if we're racing with the buf daemon. 1526 */ 1527 if (BUF_LOCK(bp, 1528 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1529 BO_LOCKPTR(bo)) == ENOLCK) { 1530 goto restart; 1531 } 1532 VNASSERT((bp->b_flags & B_DELWRI), vp, 1533 ("buf(%p) on dirty queue without DELWRI", bp)); 1534 1535 bremfree(bp); 1536 bawrite(bp); 1537 BO_LOCK(bo); 1538 goto restartsync; 1539 } 1540 } 1541 1542 bufobj_wwait(bo, 0, 0); 1543 BO_UNLOCK(bo); 1544 vnode_pager_setsize(vp, length); 1545 1546 return (0); 1547} 1548 1549static void 1550buf_vlist_remove(struct buf *bp) 1551{ 1552 struct bufv *bv; 1553 1554 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1555 ASSERT_BO_WLOCKED(bp->b_bufobj); 1556 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1557 (BX_VNDIRTY|BX_VNCLEAN), 1558 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1559 if (bp->b_xflags & BX_VNDIRTY) 1560 bv = &bp->b_bufobj->bo_dirty; 1561 else 1562 bv = &bp->b_bufobj->bo_clean; 1563 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); 1564 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1565 bv->bv_cnt--; 1566 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1567} 1568 1569/* 1570 * Add the buffer to the sorted clean or dirty block list. 1571 * 1572 * NOTE: xflags is passed as a constant, optimizing this inline function! 1573 */ 1574static void 1575buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1576{ 1577 struct bufv *bv; 1578 struct buf *n; 1579 int error; 1580 1581 ASSERT_BO_WLOCKED(bo); 1582 KASSERT((bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); 1583 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1584 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1585 bp->b_xflags |= xflags; 1586 if (xflags & BX_VNDIRTY) 1587 bv = &bo->bo_dirty; 1588 else 1589 bv = &bo->bo_clean; 1590 1591 /* 1592 * Keep the list ordered. Optimize empty list insertion. Assume 1593 * we tend to grow at the tail so lookup_le should usually be cheaper 1594 * than _ge. 1595 */ 1596 if (bv->bv_cnt == 0 || 1597 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) 1598 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1599 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) 1600 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); 1601 else 1602 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); 1603 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); 1604 if (error) 1605 panic("buf_vlist_add: Preallocated nodes insufficient."); 1606 bv->bv_cnt++; 1607} 1608 1609/* 1610 * Lookup a buffer using the splay tree. Note that we specifically avoid 1611 * shadow buffers used in background bitmap writes. 1612 * 1613 * This code isn't quite efficient as it could be because we are maintaining 1614 * two sorted lists and do not know which list the block resides in. 1615 * 1616 * During a "make buildworld" the desired buffer is found at one of 1617 * the roots more than 60% of the time. Thus, checking both roots 1618 * before performing either splay eliminates unnecessary splays on the 1619 * first tree splayed. 1620 */ 1621struct buf * 1622gbincore(struct bufobj *bo, daddr_t lblkno) 1623{ 1624 struct buf *bp; 1625 1626 ASSERT_BO_LOCKED(bo); 1627 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); 1628 if (bp != NULL) 1629 return (bp); 1630 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno); 1631} 1632 1633/* 1634 * Associate a buffer with a vnode. 1635 */ 1636void 1637bgetvp(struct vnode *vp, struct buf *bp) 1638{ 1639 struct bufobj *bo; 1640 1641 bo = &vp->v_bufobj; 1642 ASSERT_BO_WLOCKED(bo); 1643 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1644 1645 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1646 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1647 ("bgetvp: bp already attached! %p", bp)); 1648 1649 vhold(vp); 1650 bp->b_vp = vp; 1651 bp->b_bufobj = bo; 1652 /* 1653 * Insert onto list for new vnode. 1654 */ 1655 buf_vlist_add(bp, bo, BX_VNCLEAN); 1656} 1657 1658/* 1659 * Disassociate a buffer from a vnode. 1660 */ 1661void 1662brelvp(struct buf *bp) 1663{ 1664 struct bufobj *bo; 1665 struct vnode *vp; 1666 1667 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1668 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1669 1670 /* 1671 * Delete from old vnode list, if on one. 1672 */ 1673 vp = bp->b_vp; /* XXX */ 1674 bo = bp->b_bufobj; 1675 BO_LOCK(bo); 1676 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1677 buf_vlist_remove(bp); 1678 else 1679 panic("brelvp: Buffer %p not on queue.", bp); 1680 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1681 bo->bo_flag &= ~BO_ONWORKLST; 1682 mtx_lock(&sync_mtx); 1683 LIST_REMOVE(bo, bo_synclist); 1684 syncer_worklist_len--; 1685 mtx_unlock(&sync_mtx); 1686 } 1687 bp->b_vp = NULL; 1688 bp->b_bufobj = NULL; 1689 BO_UNLOCK(bo); 1690 vdrop(vp); 1691} 1692 1693/* 1694 * Add an item to the syncer work queue. 1695 */ 1696static void 1697vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1698{ 1699 int slot; 1700 1701 ASSERT_BO_WLOCKED(bo); 1702 1703 mtx_lock(&sync_mtx); 1704 if (bo->bo_flag & BO_ONWORKLST) 1705 LIST_REMOVE(bo, bo_synclist); 1706 else { 1707 bo->bo_flag |= BO_ONWORKLST; 1708 syncer_worklist_len++; 1709 } 1710 1711 if (delay > syncer_maxdelay - 2) 1712 delay = syncer_maxdelay - 2; 1713 slot = (syncer_delayno + delay) & syncer_mask; 1714 1715 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); 1716 mtx_unlock(&sync_mtx); 1717} 1718 1719static int 1720sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1721{ 1722 int error, len; 1723 1724 mtx_lock(&sync_mtx); 1725 len = syncer_worklist_len - sync_vnode_count; 1726 mtx_unlock(&sync_mtx); 1727 error = SYSCTL_OUT(req, &len, sizeof(len)); 1728 return (error); 1729} 1730 1731SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1732 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1733 1734static struct proc *updateproc; 1735static void sched_sync(void); 1736static struct kproc_desc up_kp = { 1737 "syncer", 1738 sched_sync, 1739 &updateproc 1740}; 1741SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1742 1743static int 1744sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1745{ 1746 struct vnode *vp; 1747 struct mount *mp; 1748 1749 *bo = LIST_FIRST(slp); 1750 if (*bo == NULL) 1751 return (0); 1752 vp = (*bo)->__bo_vnode; /* XXX */ 1753 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1754 return (1); 1755 /* 1756 * We use vhold in case the vnode does not 1757 * successfully sync. vhold prevents the vnode from 1758 * going away when we unlock the sync_mtx so that 1759 * we can acquire the vnode interlock. 1760 */ 1761 vholdl(vp); 1762 mtx_unlock(&sync_mtx); 1763 VI_UNLOCK(vp); 1764 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1765 vdrop(vp); 1766 mtx_lock(&sync_mtx); 1767 return (*bo == LIST_FIRST(slp)); 1768 } 1769 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1770 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1771 VOP_UNLOCK(vp, 0); 1772 vn_finished_write(mp); 1773 BO_LOCK(*bo); 1774 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1775 /* 1776 * Put us back on the worklist. The worklist 1777 * routine will remove us from our current 1778 * position and then add us back in at a later 1779 * position. 1780 */ 1781 vn_syncer_add_to_worklist(*bo, syncdelay); 1782 } 1783 BO_UNLOCK(*bo); 1784 vdrop(vp); 1785 mtx_lock(&sync_mtx); 1786 return (0); 1787} 1788 1789static int first_printf = 1; 1790 1791/* 1792 * System filesystem synchronizer daemon. 1793 */ 1794static void 1795sched_sync(void) 1796{ 1797 struct synclist *next, *slp; 1798 struct bufobj *bo; 1799 long starttime; 1800 struct thread *td = curthread; 1801 int last_work_seen; 1802 int net_worklist_len; 1803 int syncer_final_iter; 1804 int error; 1805 1806 last_work_seen = 0; 1807 syncer_final_iter = 0; 1808 syncer_state = SYNCER_RUNNING; 1809 starttime = time_uptime; 1810 td->td_pflags |= TDP_NORUNNINGBUF; 1811 1812 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1813 SHUTDOWN_PRI_LAST); 1814 1815 mtx_lock(&sync_mtx); 1816 for (;;) { 1817 if (syncer_state == SYNCER_FINAL_DELAY && 1818 syncer_final_iter == 0) { 1819 mtx_unlock(&sync_mtx); 1820 kproc_suspend_check(td->td_proc); 1821 mtx_lock(&sync_mtx); 1822 } 1823 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1824 if (syncer_state != SYNCER_RUNNING && 1825 starttime != time_uptime) { 1826 if (first_printf) { 1827 printf("\nSyncing disks, vnodes remaining..."); 1828 first_printf = 0; 1829 } 1830 printf("%d ", net_worklist_len); 1831 } 1832 starttime = time_uptime; 1833 1834 /* 1835 * Push files whose dirty time has expired. Be careful 1836 * of interrupt race on slp queue. 1837 * 1838 * Skip over empty worklist slots when shutting down. 1839 */ 1840 do { 1841 slp = &syncer_workitem_pending[syncer_delayno]; 1842 syncer_delayno += 1; 1843 if (syncer_delayno == syncer_maxdelay) 1844 syncer_delayno = 0; 1845 next = &syncer_workitem_pending[syncer_delayno]; 1846 /* 1847 * If the worklist has wrapped since the 1848 * it was emptied of all but syncer vnodes, 1849 * switch to the FINAL_DELAY state and run 1850 * for one more second. 1851 */ 1852 if (syncer_state == SYNCER_SHUTTING_DOWN && 1853 net_worklist_len == 0 && 1854 last_work_seen == syncer_delayno) { 1855 syncer_state = SYNCER_FINAL_DELAY; 1856 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1857 } 1858 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1859 syncer_worklist_len > 0); 1860 1861 /* 1862 * Keep track of the last time there was anything 1863 * on the worklist other than syncer vnodes. 1864 * Return to the SHUTTING_DOWN state if any 1865 * new work appears. 1866 */ 1867 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1868 last_work_seen = syncer_delayno; 1869 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1870 syncer_state = SYNCER_SHUTTING_DOWN; 1871 while (!LIST_EMPTY(slp)) { 1872 error = sync_vnode(slp, &bo, td); 1873 if (error == 1) { 1874 LIST_REMOVE(bo, bo_synclist); 1875 LIST_INSERT_HEAD(next, bo, bo_synclist); 1876 continue; 1877 } 1878 1879 if (first_printf == 0) 1880 wdog_kern_pat(WD_LASTVAL); 1881 1882 } 1883 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1884 syncer_final_iter--; 1885 /* 1886 * The variable rushjob allows the kernel to speed up the 1887 * processing of the filesystem syncer process. A rushjob 1888 * value of N tells the filesystem syncer to process the next 1889 * N seconds worth of work on its queue ASAP. Currently rushjob 1890 * is used by the soft update code to speed up the filesystem 1891 * syncer process when the incore state is getting so far 1892 * ahead of the disk that the kernel memory pool is being 1893 * threatened with exhaustion. 1894 */ 1895 if (rushjob > 0) { 1896 rushjob -= 1; 1897 continue; 1898 } 1899 /* 1900 * Just sleep for a short period of time between 1901 * iterations when shutting down to allow some I/O 1902 * to happen. 1903 * 1904 * If it has taken us less than a second to process the 1905 * current work, then wait. Otherwise start right over 1906 * again. We can still lose time if any single round 1907 * takes more than two seconds, but it does not really 1908 * matter as we are just trying to generally pace the 1909 * filesystem activity. 1910 */ 1911 if (syncer_state != SYNCER_RUNNING || 1912 time_uptime == starttime) { 1913 thread_lock(td); 1914 sched_prio(td, PPAUSE); 1915 thread_unlock(td); 1916 } 1917 if (syncer_state != SYNCER_RUNNING) 1918 cv_timedwait(&sync_wakeup, &sync_mtx, 1919 hz / SYNCER_SHUTDOWN_SPEEDUP); 1920 else if (time_uptime == starttime) 1921 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1922 } 1923} 1924 1925/* 1926 * Request the syncer daemon to speed up its work. 1927 * We never push it to speed up more than half of its 1928 * normal turn time, otherwise it could take over the cpu. 1929 */ 1930int 1931speedup_syncer(void) 1932{ 1933 int ret = 0; 1934 1935 mtx_lock(&sync_mtx); 1936 if (rushjob < syncdelay / 2) { 1937 rushjob += 1; 1938 stat_rush_requests += 1; 1939 ret = 1; 1940 } 1941 mtx_unlock(&sync_mtx); 1942 cv_broadcast(&sync_wakeup); 1943 return (ret); 1944} 1945 1946/* 1947 * Tell the syncer to speed up its work and run though its work 1948 * list several times, then tell it to shut down. 1949 */ 1950static void 1951syncer_shutdown(void *arg, int howto) 1952{ 1953 1954 if (howto & RB_NOSYNC) 1955 return; 1956 mtx_lock(&sync_mtx); 1957 syncer_state = SYNCER_SHUTTING_DOWN; 1958 rushjob = 0; 1959 mtx_unlock(&sync_mtx); 1960 cv_broadcast(&sync_wakeup); 1961 kproc_shutdown(arg, howto); 1962} 1963 1964void 1965syncer_suspend(void) 1966{ 1967 1968 syncer_shutdown(updateproc, 0); 1969} 1970 1971void 1972syncer_resume(void) 1973{ 1974 1975 mtx_lock(&sync_mtx); 1976 first_printf = 1; 1977 syncer_state = SYNCER_RUNNING; 1978 mtx_unlock(&sync_mtx); 1979 cv_broadcast(&sync_wakeup); 1980 kproc_resume(updateproc); 1981} 1982 1983/* 1984 * Reassign a buffer from one vnode to another. 1985 * Used to assign file specific control information 1986 * (indirect blocks) to the vnode to which they belong. 1987 */ 1988void 1989reassignbuf(struct buf *bp) 1990{ 1991 struct vnode *vp; 1992 struct bufobj *bo; 1993 int delay; 1994#ifdef INVARIANTS 1995 struct bufv *bv; 1996#endif 1997 1998 vp = bp->b_vp; 1999 bo = bp->b_bufobj; 2000 ++reassignbufcalls; 2001 2002 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 2003 bp, bp->b_vp, bp->b_flags); 2004 /* 2005 * B_PAGING flagged buffers cannot be reassigned because their vp 2006 * is not fully linked in. 2007 */ 2008 if (bp->b_flags & B_PAGING) 2009 panic("cannot reassign paging buffer"); 2010 2011 /* 2012 * Delete from old vnode list, if on one. 2013 */ 2014 BO_LOCK(bo); 2015 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 2016 buf_vlist_remove(bp); 2017 else 2018 panic("reassignbuf: Buffer %p not on queue.", bp); 2019 /* 2020 * If dirty, put on list of dirty buffers; otherwise insert onto list 2021 * of clean buffers. 2022 */ 2023 if (bp->b_flags & B_DELWRI) { 2024 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 2025 switch (vp->v_type) { 2026 case VDIR: 2027 delay = dirdelay; 2028 break; 2029 case VCHR: 2030 delay = metadelay; 2031 break; 2032 default: 2033 delay = filedelay; 2034 } 2035 vn_syncer_add_to_worklist(bo, delay); 2036 } 2037 buf_vlist_add(bp, bo, BX_VNDIRTY); 2038 } else { 2039 buf_vlist_add(bp, bo, BX_VNCLEAN); 2040 2041 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 2042 mtx_lock(&sync_mtx); 2043 LIST_REMOVE(bo, bo_synclist); 2044 syncer_worklist_len--; 2045 mtx_unlock(&sync_mtx); 2046 bo->bo_flag &= ~BO_ONWORKLST; 2047 } 2048 } 2049#ifdef INVARIANTS 2050 bv = &bo->bo_clean; 2051 bp = TAILQ_FIRST(&bv->bv_hd); 2052 KASSERT(bp == NULL || bp->b_bufobj == bo, 2053 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2054 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2055 KASSERT(bp == NULL || bp->b_bufobj == bo, 2056 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2057 bv = &bo->bo_dirty; 2058 bp = TAILQ_FIRST(&bv->bv_hd); 2059 KASSERT(bp == NULL || bp->b_bufobj == bo, 2060 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2061 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2062 KASSERT(bp == NULL || bp->b_bufobj == bo, 2063 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2064#endif 2065 BO_UNLOCK(bo); 2066} 2067 2068/* 2069 * Increment the use and hold counts on the vnode, taking care to reference 2070 * the driver's usecount if this is a chardev. The vholdl() will remove 2071 * the vnode from the free list if it is presently free. Requires the 2072 * vnode interlock and returns with it held. 2073 */ 2074static void 2075v_incr_usecount(struct vnode *vp) 2076{ 2077 2078 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2079 vholdl(vp); 2080 vp->v_usecount++; 2081 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2082 dev_lock(); 2083 vp->v_rdev->si_usecount++; 2084 dev_unlock(); 2085 } 2086} 2087 2088/* 2089 * Turn a holdcnt into a use+holdcnt such that only one call to 2090 * v_decr_usecount is needed. 2091 */ 2092static void 2093v_upgrade_usecount(struct vnode *vp) 2094{ 2095 2096 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2097 vp->v_usecount++; 2098 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2099 dev_lock(); 2100 vp->v_rdev->si_usecount++; 2101 dev_unlock(); 2102 } 2103} 2104 2105/* 2106 * Decrement the vnode use and hold count along with the driver's usecount 2107 * if this is a chardev. The vdropl() below releases the vnode interlock 2108 * as it may free the vnode. 2109 */ 2110static void 2111v_decr_usecount(struct vnode *vp) 2112{ 2113 2114 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2115 VNASSERT(vp->v_usecount > 0, vp, 2116 ("v_decr_usecount: negative usecount")); 2117 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2118 vp->v_usecount--; 2119 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2120 dev_lock(); 2121 vp->v_rdev->si_usecount--; 2122 dev_unlock(); 2123 } 2124 vdropl(vp); 2125} 2126 2127/* 2128 * Decrement only the use count and driver use count. This is intended to 2129 * be paired with a follow on vdropl() to release the remaining hold count. 2130 * In this way we may vgone() a vnode with a 0 usecount without risk of 2131 * having it end up on a free list because the hold count is kept above 0. 2132 */ 2133static void 2134v_decr_useonly(struct vnode *vp) 2135{ 2136 2137 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2138 VNASSERT(vp->v_usecount > 0, vp, 2139 ("v_decr_useonly: negative usecount")); 2140 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2141 vp->v_usecount--; 2142 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2143 dev_lock(); 2144 vp->v_rdev->si_usecount--; 2145 dev_unlock(); 2146 } 2147} 2148 2149/* 2150 * Grab a particular vnode from the free list, increment its 2151 * reference count and lock it. VI_DOOMED is set if the vnode 2152 * is being destroyed. Only callers who specify LK_RETRY will 2153 * see doomed vnodes. If inactive processing was delayed in 2154 * vput try to do it here. 2155 */ 2156int 2157vget(struct vnode *vp, int flags, struct thread *td) 2158{ 2159 int error; 2160 2161 error = 0; 2162 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2163 ("vget: invalid lock operation")); 2164 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 2165 2166 if ((flags & LK_INTERLOCK) == 0) 2167 VI_LOCK(vp); 2168 vholdl(vp); 2169 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2170 vdrop(vp); 2171 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, 2172 vp); 2173 return (error); 2174 } 2175 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2176 panic("vget: vn_lock failed to return ENOENT\n"); 2177 VI_LOCK(vp); 2178 /* Upgrade our holdcnt to a usecount. */ 2179 v_upgrade_usecount(vp); 2180 /* 2181 * We don't guarantee that any particular close will 2182 * trigger inactive processing so just make a best effort 2183 * here at preventing a reference to a removed file. If 2184 * we don't succeed no harm is done. 2185 */ 2186 if (vp->v_iflag & VI_OWEINACT) { 2187 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2188 (flags & LK_NOWAIT) == 0) 2189 vinactive(vp, td); 2190 vp->v_iflag &= ~VI_OWEINACT; 2191 } 2192 VI_UNLOCK(vp); 2193 return (0); 2194} 2195 2196/* 2197 * Increase the reference count of a vnode. 2198 */ 2199void 2200vref(struct vnode *vp) 2201{ 2202 2203 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2204 VI_LOCK(vp); 2205 v_incr_usecount(vp); 2206 VI_UNLOCK(vp); 2207} 2208 2209/* 2210 * Return reference count of a vnode. 2211 * 2212 * The results of this call are only guaranteed when some mechanism other 2213 * than the VI lock is used to stop other processes from gaining references 2214 * to the vnode. This may be the case if the caller holds the only reference. 2215 * This is also useful when stale data is acceptable as race conditions may 2216 * be accounted for by some other means. 2217 */ 2218int 2219vrefcnt(struct vnode *vp) 2220{ 2221 int usecnt; 2222 2223 VI_LOCK(vp); 2224 usecnt = vp->v_usecount; 2225 VI_UNLOCK(vp); 2226 2227 return (usecnt); 2228} 2229 2230#define VPUTX_VRELE 1 2231#define VPUTX_VPUT 2 2232#define VPUTX_VUNREF 3 2233 2234static void 2235vputx(struct vnode *vp, int func) 2236{ 2237 int error; 2238 2239 KASSERT(vp != NULL, ("vputx: null vp")); 2240 if (func == VPUTX_VUNREF) 2241 ASSERT_VOP_LOCKED(vp, "vunref"); 2242 else if (func == VPUTX_VPUT) 2243 ASSERT_VOP_LOCKED(vp, "vput"); 2244 else 2245 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); 2246 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2247 VI_LOCK(vp); 2248 2249 /* Skip this v_writecount check if we're going to panic below. */ 2250 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2251 ("vputx: missed vn_close")); 2252 error = 0; 2253 2254 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2255 vp->v_usecount == 1)) { 2256 if (func == VPUTX_VPUT) 2257 VOP_UNLOCK(vp, 0); 2258 v_decr_usecount(vp); 2259 return; 2260 } 2261 2262 if (vp->v_usecount != 1) { 2263 vprint("vputx: negative ref count", vp); 2264 panic("vputx: negative ref cnt"); 2265 } 2266 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); 2267 /* 2268 * We want to hold the vnode until the inactive finishes to 2269 * prevent vgone() races. We drop the use count here and the 2270 * hold count below when we're done. 2271 */ 2272 v_decr_useonly(vp); 2273 /* 2274 * We must call VOP_INACTIVE with the node locked. Mark 2275 * as VI_DOINGINACT to avoid recursion. 2276 */ 2277 vp->v_iflag |= VI_OWEINACT; 2278 switch (func) { 2279 case VPUTX_VRELE: 2280 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2281 VI_LOCK(vp); 2282 break; 2283 case VPUTX_VPUT: 2284 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2285 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | 2286 LK_NOWAIT); 2287 VI_LOCK(vp); 2288 } 2289 break; 2290 case VPUTX_VUNREF: 2291 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2292 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); 2293 VI_LOCK(vp); 2294 } 2295 break; 2296 } 2297 if (vp->v_usecount > 0) 2298 vp->v_iflag &= ~VI_OWEINACT; 2299 if (error == 0) { 2300 if (vp->v_iflag & VI_OWEINACT) 2301 vinactive(vp, curthread); 2302 if (func != VPUTX_VUNREF) 2303 VOP_UNLOCK(vp, 0); 2304 } 2305 vdropl(vp); 2306} 2307 2308/* 2309 * Vnode put/release. 2310 * If count drops to zero, call inactive routine and return to freelist. 2311 */ 2312void 2313vrele(struct vnode *vp) 2314{ 2315 2316 vputx(vp, VPUTX_VRELE); 2317} 2318 2319/* 2320 * Release an already locked vnode. This give the same effects as 2321 * unlock+vrele(), but takes less time and avoids releasing and 2322 * re-aquiring the lock (as vrele() acquires the lock internally.) 2323 */ 2324void 2325vput(struct vnode *vp) 2326{ 2327 2328 vputx(vp, VPUTX_VPUT); 2329} 2330 2331/* 2332 * Release an exclusively locked vnode. Do not unlock the vnode lock. 2333 */ 2334void 2335vunref(struct vnode *vp) 2336{ 2337 2338 vputx(vp, VPUTX_VUNREF); 2339} 2340 2341/* 2342 * Somebody doesn't want the vnode recycled. 2343 */ 2344void 2345vhold(struct vnode *vp) 2346{ 2347 2348 VI_LOCK(vp); 2349 vholdl(vp); 2350 VI_UNLOCK(vp); 2351} 2352 2353/* 2354 * Increase the hold count and activate if this is the first reference. 2355 */ 2356void 2357vholdl(struct vnode *vp) 2358{ 2359 struct mount *mp; 2360 2361 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2362#ifdef INVARIANTS 2363 /* getnewvnode() calls v_incr_usecount() without holding interlock. */ 2364 if (vp->v_type != VNON || vp->v_data != NULL) { 2365 ASSERT_VI_LOCKED(vp, "vholdl"); 2366 VNASSERT(vp->v_holdcnt > 0 || (vp->v_iflag & VI_FREE) != 0, 2367 vp, ("vholdl: free vnode is held")); 2368 } 2369#endif 2370 vp->v_holdcnt++; 2371 if ((vp->v_iflag & VI_FREE) == 0) 2372 return; 2373 VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count")); 2374 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed.")); 2375 /* 2376 * Remove a vnode from the free list, mark it as in use, 2377 * and put it on the active list. 2378 */ 2379 mtx_lock(&vnode_free_list_mtx); 2380 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); 2381 freevnodes--; 2382 vp->v_iflag &= ~(VI_FREE|VI_AGE); 2383 KASSERT((vp->v_iflag & VI_ACTIVE) == 0, 2384 ("Activating already active vnode")); 2385 vp->v_iflag |= VI_ACTIVE; 2386 mp = vp->v_mount; 2387 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); 2388 mp->mnt_activevnodelistsize++; 2389 mtx_unlock(&vnode_free_list_mtx); 2390} 2391 2392/* 2393 * Note that there is one less who cares about this vnode. 2394 * vdrop() is the opposite of vhold(). 2395 */ 2396void 2397vdrop(struct vnode *vp) 2398{ 2399 2400 VI_LOCK(vp); 2401 vdropl(vp); 2402} 2403 2404/* 2405 * Drop the hold count of the vnode. If this is the last reference to 2406 * the vnode we place it on the free list unless it has been vgone'd 2407 * (marked VI_DOOMED) in which case we will free it. 2408 */ 2409void 2410vdropl(struct vnode *vp) 2411{ 2412 struct bufobj *bo; 2413 struct mount *mp; 2414 int active; 2415 2416 ASSERT_VI_LOCKED(vp, "vdropl"); 2417 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2418 if (vp->v_holdcnt <= 0) 2419 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2420 vp->v_holdcnt--; 2421 if (vp->v_holdcnt > 0) { 2422 VI_UNLOCK(vp); 2423 return; 2424 } 2425 if ((vp->v_iflag & VI_DOOMED) == 0) { 2426 /* 2427 * Mark a vnode as free: remove it from its active list 2428 * and put it up for recycling on the freelist. 2429 */ 2430 VNASSERT(vp->v_op != NULL, vp, 2431 ("vdropl: vnode already reclaimed.")); 2432 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 2433 ("vnode already free")); 2434 VNASSERT(vp->v_holdcnt == 0, vp, 2435 ("vdropl: freeing when we shouldn't")); 2436 active = vp->v_iflag & VI_ACTIVE; 2437 vp->v_iflag &= ~VI_ACTIVE; 2438 mp = vp->v_mount; 2439 mtx_lock(&vnode_free_list_mtx); 2440 if (active) { 2441 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, 2442 v_actfreelist); 2443 mp->mnt_activevnodelistsize--; 2444 } 2445 if (vp->v_iflag & VI_AGE) { 2446 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist); 2447 } else { 2448 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); 2449 } 2450 freevnodes++; 2451 vp->v_iflag &= ~VI_AGE; 2452 vp->v_iflag |= VI_FREE; 2453 mtx_unlock(&vnode_free_list_mtx); 2454 VI_UNLOCK(vp); 2455 return; 2456 } 2457 /* 2458 * The vnode has been marked for destruction, so free it. 2459 */ 2460 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); 2461 atomic_subtract_long(&numvnodes, 1); 2462 bo = &vp->v_bufobj; 2463 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 2464 ("cleaned vnode still on the free list.")); 2465 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 2466 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 2467 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 2468 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 2469 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 2470 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 2471 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, 2472 ("clean blk trie not empty")); 2473 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 2474 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, 2475 ("dirty blk trie not empty")); 2476 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 2477 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 2478 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); 2479 VI_UNLOCK(vp); 2480#ifdef MAC 2481 mac_vnode_destroy(vp); 2482#endif 2483 if (vp->v_pollinfo != NULL) 2484 destroy_vpollinfo(vp->v_pollinfo); 2485#ifdef INVARIANTS 2486 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ 2487 vp->v_op = NULL; 2488#endif 2489 rangelock_destroy(&vp->v_rl); 2490 lockdestroy(vp->v_vnlock); 2491 mtx_destroy(&vp->v_interlock); 2492 rw_destroy(BO_LOCKPTR(bo)); 2493 uma_zfree(vnode_zone, vp); 2494} 2495 2496/* 2497 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2498 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2499 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2500 * failed lock upgrade. 2501 */ 2502void 2503vinactive(struct vnode *vp, struct thread *td) 2504{ 2505 struct vm_object *obj; 2506 2507 ASSERT_VOP_ELOCKED(vp, "vinactive"); 2508 ASSERT_VI_LOCKED(vp, "vinactive"); 2509 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2510 ("vinactive: recursed on VI_DOINGINACT")); 2511 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2512 vp->v_iflag |= VI_DOINGINACT; 2513 vp->v_iflag &= ~VI_OWEINACT; 2514 VI_UNLOCK(vp); 2515 /* 2516 * Before moving off the active list, we must be sure that any 2517 * modified pages are on the vnode's dirty list since these will 2518 * no longer be checked once the vnode is on the inactive list. 2519 * Because the vnode vm object keeps a hold reference on the vnode 2520 * if there is at least one resident non-cached page, the vnode 2521 * cannot leave the active list without the page cleanup done. 2522 */ 2523 obj = vp->v_object; 2524 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { 2525 VM_OBJECT_WLOCK(obj); 2526 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC); 2527 VM_OBJECT_WUNLOCK(obj); 2528 } 2529 VOP_INACTIVE(vp, td); 2530 VI_LOCK(vp); 2531 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2532 ("vinactive: lost VI_DOINGINACT")); 2533 vp->v_iflag &= ~VI_DOINGINACT; 2534} 2535 2536/* 2537 * Remove any vnodes in the vnode table belonging to mount point mp. 2538 * 2539 * If FORCECLOSE is not specified, there should not be any active ones, 2540 * return error if any are found (nb: this is a user error, not a 2541 * system error). If FORCECLOSE is specified, detach any active vnodes 2542 * that are found. 2543 * 2544 * If WRITECLOSE is set, only flush out regular file vnodes open for 2545 * writing. 2546 * 2547 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2548 * 2549 * `rootrefs' specifies the base reference count for the root vnode 2550 * of this filesystem. The root vnode is considered busy if its 2551 * v_usecount exceeds this value. On a successful return, vflush(, td) 2552 * will call vrele() on the root vnode exactly rootrefs times. 2553 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2554 * be zero. 2555 */ 2556#ifdef DIAGNOSTIC 2557static int busyprt = 0; /* print out busy vnodes */ 2558SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); 2559#endif 2560 2561int 2562vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) 2563{ 2564 struct vnode *vp, *mvp, *rootvp = NULL; 2565 struct vattr vattr; 2566 int busy = 0, error; 2567 2568 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, 2569 rootrefs, flags); 2570 if (rootrefs > 0) { 2571 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2572 ("vflush: bad args")); 2573 /* 2574 * Get the filesystem root vnode. We can vput() it 2575 * immediately, since with rootrefs > 0, it won't go away. 2576 */ 2577 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { 2578 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", 2579 __func__, error); 2580 return (error); 2581 } 2582 vput(rootvp); 2583 } 2584loop: 2585 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 2586 vholdl(vp); 2587 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2588 if (error) { 2589 vdrop(vp); 2590 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 2591 goto loop; 2592 } 2593 /* 2594 * Skip over a vnodes marked VV_SYSTEM. 2595 */ 2596 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2597 VOP_UNLOCK(vp, 0); 2598 vdrop(vp); 2599 continue; 2600 } 2601 /* 2602 * If WRITECLOSE is set, flush out unlinked but still open 2603 * files (even if open only for reading) and regular file 2604 * vnodes open for writing. 2605 */ 2606 if (flags & WRITECLOSE) { 2607 if (vp->v_object != NULL) { 2608 VM_OBJECT_WLOCK(vp->v_object); 2609 vm_object_page_clean(vp->v_object, 0, 0, 0); 2610 VM_OBJECT_WUNLOCK(vp->v_object); 2611 } 2612 error = VOP_FSYNC(vp, MNT_WAIT, td); 2613 if (error != 0) { 2614 VOP_UNLOCK(vp, 0); 2615 vdrop(vp); 2616 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 2617 return (error); 2618 } 2619 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2620 VI_LOCK(vp); 2621 2622 if ((vp->v_type == VNON || 2623 (error == 0 && vattr.va_nlink > 0)) && 2624 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2625 VOP_UNLOCK(vp, 0); 2626 vdropl(vp); 2627 continue; 2628 } 2629 } else 2630 VI_LOCK(vp); 2631 /* 2632 * With v_usecount == 0, all we need to do is clear out the 2633 * vnode data structures and we are done. 2634 * 2635 * If FORCECLOSE is set, forcibly close the vnode. 2636 */ 2637 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2638 VNASSERT(vp->v_usecount == 0 || 2639 vp->v_op != &devfs_specops || 2640 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2641 ("device VNODE %p is FORCECLOSED", vp)); 2642 vgonel(vp); 2643 } else { 2644 busy++; 2645#ifdef DIAGNOSTIC 2646 if (busyprt) 2647 vprint("vflush: busy vnode", vp); 2648#endif 2649 } 2650 VOP_UNLOCK(vp, 0); 2651 vdropl(vp); 2652 } 2653 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2654 /* 2655 * If just the root vnode is busy, and if its refcount 2656 * is equal to `rootrefs', then go ahead and kill it. 2657 */ 2658 VI_LOCK(rootvp); 2659 KASSERT(busy > 0, ("vflush: not busy")); 2660 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2661 ("vflush: usecount %d < rootrefs %d", 2662 rootvp->v_usecount, rootrefs)); 2663 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2664 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2665 vgone(rootvp); 2666 VOP_UNLOCK(rootvp, 0); 2667 busy = 0; 2668 } else 2669 VI_UNLOCK(rootvp); 2670 } 2671 if (busy) { 2672 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, 2673 busy); 2674 return (EBUSY); 2675 } 2676 for (; rootrefs > 0; rootrefs--) 2677 vrele(rootvp); 2678 return (0); 2679} 2680 2681/* 2682 * Recycle an unused vnode to the front of the free list. 2683 */ 2684int 2685vrecycle(struct vnode *vp) 2686{ 2687 int recycled; 2688 2689 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2690 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2691 recycled = 0; 2692 VI_LOCK(vp); 2693 if (vp->v_usecount == 0) { 2694 recycled = 1; 2695 vgonel(vp); 2696 } 2697 VI_UNLOCK(vp); 2698 return (recycled); 2699} 2700 2701/* 2702 * Eliminate all activity associated with a vnode 2703 * in preparation for reuse. 2704 */ 2705void 2706vgone(struct vnode *vp) 2707{ 2708 VI_LOCK(vp); 2709 vgonel(vp); 2710 VI_UNLOCK(vp); 2711} 2712 2713static void 2714notify_lowervp_vfs_dummy(struct mount *mp __unused, 2715 struct vnode *lowervp __unused) 2716{ 2717} 2718 2719/* 2720 * Notify upper mounts about reclaimed or unlinked vnode. 2721 */ 2722void 2723vfs_notify_upper(struct vnode *vp, int event) 2724{ 2725 static struct vfsops vgonel_vfsops = { 2726 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, 2727 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, 2728 }; 2729 struct mount *mp, *ump, *mmp; 2730 2731 mp = vp->v_mount; 2732 if (mp == NULL) 2733 return; 2734 2735 MNT_ILOCK(mp); 2736 if (TAILQ_EMPTY(&mp->mnt_uppers)) 2737 goto unlock; 2738 MNT_IUNLOCK(mp); 2739 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); 2740 mmp->mnt_op = &vgonel_vfsops; 2741 mmp->mnt_kern_flag |= MNTK_MARKER; 2742 MNT_ILOCK(mp); 2743 mp->mnt_kern_flag |= MNTK_VGONE_UPPER; 2744 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { 2745 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { 2746 ump = TAILQ_NEXT(ump, mnt_upper_link); 2747 continue; 2748 } 2749 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); 2750 MNT_IUNLOCK(mp); 2751 switch (event) { 2752 case VFS_NOTIFY_UPPER_RECLAIM: 2753 VFS_RECLAIM_LOWERVP(ump, vp); 2754 break; 2755 case VFS_NOTIFY_UPPER_UNLINK: 2756 VFS_UNLINK_LOWERVP(ump, vp); 2757 break; 2758 default: 2759 KASSERT(0, ("invalid event %d", event)); 2760 break; 2761 } 2762 MNT_ILOCK(mp); 2763 ump = TAILQ_NEXT(mmp, mnt_upper_link); 2764 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); 2765 } 2766 free(mmp, M_TEMP); 2767 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; 2768 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { 2769 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; 2770 wakeup(&mp->mnt_uppers); 2771 } 2772unlock: 2773 MNT_IUNLOCK(mp); 2774} 2775 2776/* 2777 * vgone, with the vp interlock held. 2778 */ 2779void 2780vgonel(struct vnode *vp) 2781{ 2782 struct thread *td; 2783 int oweinact; 2784 int active; 2785 struct mount *mp; 2786 2787 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2788 ASSERT_VI_LOCKED(vp, "vgonel"); 2789 VNASSERT(vp->v_holdcnt, vp, 2790 ("vgonel: vp %p has no reference.", vp)); 2791 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2792 td = curthread; 2793 2794 /* 2795 * Don't vgonel if we're already doomed. 2796 */ 2797 if (vp->v_iflag & VI_DOOMED) 2798 return; 2799 vp->v_iflag |= VI_DOOMED; 2800 2801 /* 2802 * Check to see if the vnode is in use. If so, we have to call 2803 * VOP_CLOSE() and VOP_INACTIVE(). 2804 */ 2805 active = vp->v_usecount; 2806 oweinact = (vp->v_iflag & VI_OWEINACT); 2807 VI_UNLOCK(vp); 2808 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); 2809 2810 /* 2811 * If purging an active vnode, it must be closed and 2812 * deactivated before being reclaimed. 2813 */ 2814 if (active) 2815 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2816 if (oweinact || active) { 2817 VI_LOCK(vp); 2818 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2819 vinactive(vp, td); 2820 VI_UNLOCK(vp); 2821 } 2822 if (vp->v_type == VSOCK) 2823 vfs_unp_reclaim(vp); 2824 2825 /* 2826 * Clean out any buffers associated with the vnode. 2827 * If the flush fails, just toss the buffers. 2828 */ 2829 mp = NULL; 2830 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2831 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2832 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { 2833 while (vinvalbuf(vp, 0, 0, 0) != 0) 2834 ; 2835 } 2836#ifdef INVARIANTS 2837 BO_LOCK(&vp->v_bufobj); 2838 KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && 2839 vp->v_bufobj.bo_dirty.bv_cnt == 0 && 2840 TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && 2841 vp->v_bufobj.bo_clean.bv_cnt == 0, 2842 ("vp %p bufobj not invalidated", vp)); 2843 vp->v_bufobj.bo_flag |= BO_DEAD; 2844 BO_UNLOCK(&vp->v_bufobj); 2845#endif 2846 2847 /* 2848 * Reclaim the vnode. 2849 */ 2850 if (VOP_RECLAIM(vp, td)) 2851 panic("vgone: cannot reclaim"); 2852 if (mp != NULL) 2853 vn_finished_secondary_write(mp); 2854 VNASSERT(vp->v_object == NULL, vp, 2855 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2856 /* 2857 * Clear the advisory locks and wake up waiting threads. 2858 */ 2859 (void)VOP_ADVLOCKPURGE(vp); 2860 /* 2861 * Delete from old mount point vnode list. 2862 */ 2863 delmntque(vp); 2864 cache_purge(vp); 2865 /* 2866 * Done with purge, reset to the standard lock and invalidate 2867 * the vnode. 2868 */ 2869 VI_LOCK(vp); 2870 vp->v_vnlock = &vp->v_lock; 2871 vp->v_op = &dead_vnodeops; 2872 vp->v_tag = "none"; 2873 vp->v_type = VBAD; 2874} 2875 2876/* 2877 * Calculate the total number of references to a special device. 2878 */ 2879int 2880vcount(struct vnode *vp) 2881{ 2882 int count; 2883 2884 dev_lock(); 2885 count = vp->v_rdev->si_usecount; 2886 dev_unlock(); 2887 return (count); 2888} 2889 2890/* 2891 * Same as above, but using the struct cdev *as argument 2892 */ 2893int 2894count_dev(struct cdev *dev) 2895{ 2896 int count; 2897 2898 dev_lock(); 2899 count = dev->si_usecount; 2900 dev_unlock(); 2901 return(count); 2902} 2903 2904/* 2905 * Print out a description of a vnode. 2906 */ 2907static char *typename[] = 2908{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2909 "VMARKER"}; 2910 2911void 2912vn_printf(struct vnode *vp, const char *fmt, ...) 2913{ 2914 va_list ap; 2915 char buf[256], buf2[16]; 2916 u_long flags; 2917 2918 va_start(ap, fmt); 2919 vprintf(fmt, ap); 2920 va_end(ap); 2921 printf("%p: ", (void *)vp); 2922 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2923 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2924 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2925 buf[0] = '\0'; 2926 buf[1] = '\0'; 2927 if (vp->v_vflag & VV_ROOT) 2928 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2929 if (vp->v_vflag & VV_ISTTY) 2930 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2931 if (vp->v_vflag & VV_NOSYNC) 2932 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2933 if (vp->v_vflag & VV_ETERNALDEV) 2934 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); 2935 if (vp->v_vflag & VV_CACHEDLABEL) 2936 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2937 if (vp->v_vflag & VV_TEXT) 2938 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2939 if (vp->v_vflag & VV_COPYONWRITE) 2940 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2941 if (vp->v_vflag & VV_SYSTEM) 2942 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2943 if (vp->v_vflag & VV_PROCDEP) 2944 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2945 if (vp->v_vflag & VV_NOKNOTE) 2946 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2947 if (vp->v_vflag & VV_DELETED) 2948 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2949 if (vp->v_vflag & VV_MD) 2950 strlcat(buf, "|VV_MD", sizeof(buf)); 2951 if (vp->v_vflag & VV_FORCEINSMQ) 2952 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); 2953 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | 2954 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2955 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); 2956 if (flags != 0) { 2957 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2958 strlcat(buf, buf2, sizeof(buf)); 2959 } 2960 if (vp->v_iflag & VI_MOUNT) 2961 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2962 if (vp->v_iflag & VI_AGE) 2963 strlcat(buf, "|VI_AGE", sizeof(buf)); 2964 if (vp->v_iflag & VI_DOOMED) 2965 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2966 if (vp->v_iflag & VI_FREE) 2967 strlcat(buf, "|VI_FREE", sizeof(buf)); 2968 if (vp->v_iflag & VI_ACTIVE) 2969 strlcat(buf, "|VI_ACTIVE", sizeof(buf)); 2970 if (vp->v_iflag & VI_DOINGINACT) 2971 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2972 if (vp->v_iflag & VI_OWEINACT) 2973 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2974 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2975 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT); 2976 if (flags != 0) { 2977 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2978 strlcat(buf, buf2, sizeof(buf)); 2979 } 2980 printf(" flags (%s)\n", buf + 1); 2981 if (mtx_owned(VI_MTX(vp))) 2982 printf(" VI_LOCKed"); 2983 if (vp->v_object != NULL) 2984 printf(" v_object %p ref %d pages %d " 2985 "cleanbuf %d dirtybuf %d\n", 2986 vp->v_object, vp->v_object->ref_count, 2987 vp->v_object->resident_page_count, 2988 vp->v_bufobj.bo_dirty.bv_cnt, 2989 vp->v_bufobj.bo_clean.bv_cnt); 2990 printf(" "); 2991 lockmgr_printinfo(vp->v_vnlock); 2992 if (vp->v_data != NULL) 2993 VOP_PRINT(vp); 2994} 2995 2996#ifdef DDB 2997/* 2998 * List all of the locked vnodes in the system. 2999 * Called when debugging the kernel. 3000 */ 3001DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 3002{ 3003 struct mount *mp; 3004 struct vnode *vp; 3005 3006 /* 3007 * Note: because this is DDB, we can't obey the locking semantics 3008 * for these structures, which means we could catch an inconsistent 3009 * state and dereference a nasty pointer. Not much to be done 3010 * about that. 3011 */ 3012 db_printf("Locked vnodes\n"); 3013 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3014 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3015 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) 3016 vprint("", vp); 3017 } 3018 } 3019} 3020 3021/* 3022 * Show details about the given vnode. 3023 */ 3024DB_SHOW_COMMAND(vnode, db_show_vnode) 3025{ 3026 struct vnode *vp; 3027 3028 if (!have_addr) 3029 return; 3030 vp = (struct vnode *)addr; 3031 vn_printf(vp, "vnode "); 3032} 3033 3034/* 3035 * Show details about the given mount point. 3036 */ 3037DB_SHOW_COMMAND(mount, db_show_mount) 3038{ 3039 struct mount *mp; 3040 struct vfsopt *opt; 3041 struct statfs *sp; 3042 struct vnode *vp; 3043 char buf[512]; 3044 uint64_t mflags; 3045 u_int flags; 3046 3047 if (!have_addr) { 3048 /* No address given, print short info about all mount points. */ 3049 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3050 db_printf("%p %s on %s (%s)\n", mp, 3051 mp->mnt_stat.f_mntfromname, 3052 mp->mnt_stat.f_mntonname, 3053 mp->mnt_stat.f_fstypename); 3054 if (db_pager_quit) 3055 break; 3056 } 3057 db_printf("\nMore info: show mount <addr>\n"); 3058 return; 3059 } 3060 3061 mp = (struct mount *)addr; 3062 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 3063 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 3064 3065 buf[0] = '\0'; 3066 mflags = mp->mnt_flag; 3067#define MNT_FLAG(flag) do { \ 3068 if (mflags & (flag)) { \ 3069 if (buf[0] != '\0') \ 3070 strlcat(buf, ", ", sizeof(buf)); \ 3071 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 3072 mflags &= ~(flag); \ 3073 } \ 3074} while (0) 3075 MNT_FLAG(MNT_RDONLY); 3076 MNT_FLAG(MNT_SYNCHRONOUS); 3077 MNT_FLAG(MNT_NOEXEC); 3078 MNT_FLAG(MNT_NOSUID); 3079 MNT_FLAG(MNT_NFS4ACLS); 3080 MNT_FLAG(MNT_UNION); 3081 MNT_FLAG(MNT_ASYNC); 3082 MNT_FLAG(MNT_SUIDDIR); 3083 MNT_FLAG(MNT_SOFTDEP); 3084 MNT_FLAG(MNT_NOSYMFOLLOW); 3085 MNT_FLAG(MNT_GJOURNAL); 3086 MNT_FLAG(MNT_MULTILABEL); 3087 MNT_FLAG(MNT_ACLS); 3088 MNT_FLAG(MNT_NOATIME); 3089 MNT_FLAG(MNT_NOCLUSTERR); 3090 MNT_FLAG(MNT_NOCLUSTERW); 3091 MNT_FLAG(MNT_SUJ); 3092 MNT_FLAG(MNT_EXRDONLY); 3093 MNT_FLAG(MNT_EXPORTED); 3094 MNT_FLAG(MNT_DEFEXPORTED); 3095 MNT_FLAG(MNT_EXPORTANON); 3096 MNT_FLAG(MNT_EXKERB); 3097 MNT_FLAG(MNT_EXPUBLIC); 3098 MNT_FLAG(MNT_LOCAL); 3099 MNT_FLAG(MNT_QUOTA); 3100 MNT_FLAG(MNT_ROOTFS); 3101 MNT_FLAG(MNT_USER); 3102 MNT_FLAG(MNT_IGNORE); 3103 MNT_FLAG(MNT_UPDATE); 3104 MNT_FLAG(MNT_DELEXPORT); 3105 MNT_FLAG(MNT_RELOAD); 3106 MNT_FLAG(MNT_FORCE); 3107 MNT_FLAG(MNT_SNAPSHOT); 3108 MNT_FLAG(MNT_BYFSID); 3109#undef MNT_FLAG 3110 if (mflags != 0) { 3111 if (buf[0] != '\0') 3112 strlcat(buf, ", ", sizeof(buf)); 3113 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 3114 "0x%016jx", mflags); 3115 } 3116 db_printf(" mnt_flag = %s\n", buf); 3117 3118 buf[0] = '\0'; 3119 flags = mp->mnt_kern_flag; 3120#define MNT_KERN_FLAG(flag) do { \ 3121 if (flags & (flag)) { \ 3122 if (buf[0] != '\0') \ 3123 strlcat(buf, ", ", sizeof(buf)); \ 3124 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 3125 flags &= ~(flag); \ 3126 } \ 3127} while (0) 3128 MNT_KERN_FLAG(MNTK_UNMOUNTF); 3129 MNT_KERN_FLAG(MNTK_ASYNC); 3130 MNT_KERN_FLAG(MNTK_SOFTDEP); 3131 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 3132 MNT_KERN_FLAG(MNTK_DRAINING); 3133 MNT_KERN_FLAG(MNTK_REFEXPIRE); 3134 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); 3135 MNT_KERN_FLAG(MNTK_SHARED_WRITES); 3136 MNT_KERN_FLAG(MNTK_NO_IOPF); 3137 MNT_KERN_FLAG(MNTK_VGONE_UPPER); 3138 MNT_KERN_FLAG(MNTK_VGONE_WAITER); 3139 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); 3140 MNT_KERN_FLAG(MNTK_MARKER); 3141 MNT_KERN_FLAG(MNTK_USES_BCACHE); 3142 MNT_KERN_FLAG(MNTK_NOASYNC); 3143 MNT_KERN_FLAG(MNTK_UNMOUNT); 3144 MNT_KERN_FLAG(MNTK_MWAIT); 3145 MNT_KERN_FLAG(MNTK_SUSPEND); 3146 MNT_KERN_FLAG(MNTK_SUSPEND2); 3147 MNT_KERN_FLAG(MNTK_SUSPENDED); 3148 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 3149 MNT_KERN_FLAG(MNTK_NOKNOTE); 3150#undef MNT_KERN_FLAG 3151 if (flags != 0) { 3152 if (buf[0] != '\0') 3153 strlcat(buf, ", ", sizeof(buf)); 3154 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 3155 "0x%08x", flags); 3156 } 3157 db_printf(" mnt_kern_flag = %s\n", buf); 3158 3159 db_printf(" mnt_opt = "); 3160 opt = TAILQ_FIRST(mp->mnt_opt); 3161 if (opt != NULL) { 3162 db_printf("%s", opt->name); 3163 opt = TAILQ_NEXT(opt, link); 3164 while (opt != NULL) { 3165 db_printf(", %s", opt->name); 3166 opt = TAILQ_NEXT(opt, link); 3167 } 3168 } 3169 db_printf("\n"); 3170 3171 sp = &mp->mnt_stat; 3172 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 3173 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 3174 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 3175 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 3176 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 3177 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 3178 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 3179 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 3180 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 3181 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 3182 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 3183 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 3184 3185 db_printf(" mnt_cred = { uid=%u ruid=%u", 3186 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 3187 if (jailed(mp->mnt_cred)) 3188 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 3189 db_printf(" }\n"); 3190 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 3191 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 3192 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 3193 db_printf(" mnt_activevnodelistsize = %d\n", 3194 mp->mnt_activevnodelistsize); 3195 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 3196 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 3197 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 3198 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 3199 db_printf(" mnt_lockref = %d\n", mp->mnt_lockref); 3200 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 3201 db_printf(" mnt_secondary_accwrites = %d\n", 3202 mp->mnt_secondary_accwrites); 3203 db_printf(" mnt_gjprovider = %s\n", 3204 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 3205 3206 db_printf("\n\nList of active vnodes\n"); 3207 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) { 3208 if (vp->v_type != VMARKER) { 3209 vn_printf(vp, "vnode "); 3210 if (db_pager_quit) 3211 break; 3212 } 3213 } 3214 db_printf("\n\nList of inactive vnodes\n"); 3215 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3216 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) { 3217 vn_printf(vp, "vnode "); 3218 if (db_pager_quit) 3219 break; 3220 } 3221 } 3222} 3223#endif /* DDB */ 3224 3225/* 3226 * Fill in a struct xvfsconf based on a struct vfsconf. 3227 */ 3228static int 3229vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) 3230{ 3231 struct xvfsconf xvfsp; 3232 3233 bzero(&xvfsp, sizeof(xvfsp)); 3234 strcpy(xvfsp.vfc_name, vfsp->vfc_name); 3235 xvfsp.vfc_typenum = vfsp->vfc_typenum; 3236 xvfsp.vfc_refcount = vfsp->vfc_refcount; 3237 xvfsp.vfc_flags = vfsp->vfc_flags; 3238 /* 3239 * These are unused in userland, we keep them 3240 * to not break binary compatibility. 3241 */ 3242 xvfsp.vfc_vfsops = NULL; 3243 xvfsp.vfc_next = NULL; 3244 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3245} 3246 3247#ifdef COMPAT_FREEBSD32 3248struct xvfsconf32 { 3249 uint32_t vfc_vfsops; 3250 char vfc_name[MFSNAMELEN]; 3251 int32_t vfc_typenum; 3252 int32_t vfc_refcount; 3253 int32_t vfc_flags; 3254 uint32_t vfc_next; 3255}; 3256 3257static int 3258vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) 3259{ 3260 struct xvfsconf32 xvfsp; 3261 3262 strcpy(xvfsp.vfc_name, vfsp->vfc_name); 3263 xvfsp.vfc_typenum = vfsp->vfc_typenum; 3264 xvfsp.vfc_refcount = vfsp->vfc_refcount; 3265 xvfsp.vfc_flags = vfsp->vfc_flags; 3266 xvfsp.vfc_vfsops = 0; 3267 xvfsp.vfc_next = 0; 3268 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3269} 3270#endif 3271 3272/* 3273 * Top level filesystem related information gathering. 3274 */ 3275static int 3276sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 3277{ 3278 struct vfsconf *vfsp; 3279 int error; 3280 3281 error = 0; 3282 vfsconf_slock(); 3283 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3284#ifdef COMPAT_FREEBSD32 3285 if (req->flags & SCTL_MASK32) 3286 error = vfsconf2x32(req, vfsp); 3287 else 3288#endif 3289 error = vfsconf2x(req, vfsp); 3290 if (error) 3291 break; 3292 } 3293 vfsconf_sunlock(); 3294 return (error); 3295} 3296 3297SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | 3298 CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, 3299 "S,xvfsconf", "List of all configured filesystems"); 3300 3301#ifndef BURN_BRIDGES 3302static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 3303 3304static int 3305vfs_sysctl(SYSCTL_HANDLER_ARGS) 3306{ 3307 int *name = (int *)arg1 - 1; /* XXX */ 3308 u_int namelen = arg2 + 1; /* XXX */ 3309 struct vfsconf *vfsp; 3310 3311 log(LOG_WARNING, "userland calling deprecated sysctl, " 3312 "please rebuild world\n"); 3313 3314#if 1 || defined(COMPAT_PRELITE2) 3315 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 3316 if (namelen == 1) 3317 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 3318#endif 3319 3320 switch (name[1]) { 3321 case VFS_MAXTYPENUM: 3322 if (namelen != 2) 3323 return (ENOTDIR); 3324 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 3325 case VFS_CONF: 3326 if (namelen != 3) 3327 return (ENOTDIR); /* overloaded */ 3328 vfsconf_slock(); 3329 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3330 if (vfsp->vfc_typenum == name[2]) 3331 break; 3332 } 3333 vfsconf_sunlock(); 3334 if (vfsp == NULL) 3335 return (EOPNOTSUPP); 3336#ifdef COMPAT_FREEBSD32 3337 if (req->flags & SCTL_MASK32) 3338 return (vfsconf2x32(req, vfsp)); 3339 else 3340#endif 3341 return (vfsconf2x(req, vfsp)); 3342 } 3343 return (EOPNOTSUPP); 3344} 3345 3346static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | 3347 CTLFLAG_MPSAFE, vfs_sysctl, 3348 "Generic filesystem"); 3349 3350#if 1 || defined(COMPAT_PRELITE2) 3351 3352static int 3353sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 3354{ 3355 int error; 3356 struct vfsconf *vfsp; 3357 struct ovfsconf ovfs; 3358 3359 vfsconf_slock(); 3360 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3361 bzero(&ovfs, sizeof(ovfs)); 3362 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 3363 strcpy(ovfs.vfc_name, vfsp->vfc_name); 3364 ovfs.vfc_index = vfsp->vfc_typenum; 3365 ovfs.vfc_refcount = vfsp->vfc_refcount; 3366 ovfs.vfc_flags = vfsp->vfc_flags; 3367 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 3368 if (error != 0) { 3369 vfsconf_sunlock(); 3370 return (error); 3371 } 3372 } 3373 vfsconf_sunlock(); 3374 return (0); 3375} 3376 3377#endif /* 1 || COMPAT_PRELITE2 */ 3378#endif /* !BURN_BRIDGES */ 3379 3380#define KINFO_VNODESLOP 10 3381#ifdef notyet 3382/* 3383 * Dump vnode list (via sysctl). 3384 */ 3385/* ARGSUSED */ 3386static int 3387sysctl_vnode(SYSCTL_HANDLER_ARGS) 3388{ 3389 struct xvnode *xvn; 3390 struct mount *mp; 3391 struct vnode *vp; 3392 int error, len, n; 3393 3394 /* 3395 * Stale numvnodes access is not fatal here. 3396 */ 3397 req->lock = 0; 3398 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3399 if (!req->oldptr) 3400 /* Make an estimate */ 3401 return (SYSCTL_OUT(req, 0, len)); 3402 3403 error = sysctl_wire_old_buffer(req, 0); 3404 if (error != 0) 3405 return (error); 3406 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3407 n = 0; 3408 mtx_lock(&mountlist_mtx); 3409 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3410 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3411 continue; 3412 MNT_ILOCK(mp); 3413 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3414 if (n == len) 3415 break; 3416 vref(vp); 3417 xvn[n].xv_size = sizeof *xvn; 3418 xvn[n].xv_vnode = vp; 3419 xvn[n].xv_id = 0; /* XXX compat */ 3420#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3421 XV_COPY(usecount); 3422 XV_COPY(writecount); 3423 XV_COPY(holdcnt); 3424 XV_COPY(mount); 3425 XV_COPY(numoutput); 3426 XV_COPY(type); 3427#undef XV_COPY 3428 xvn[n].xv_flag = vp->v_vflag; 3429 3430 switch (vp->v_type) { 3431 case VREG: 3432 case VDIR: 3433 case VLNK: 3434 break; 3435 case VBLK: 3436 case VCHR: 3437 if (vp->v_rdev == NULL) { 3438 vrele(vp); 3439 continue; 3440 } 3441 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3442 break; 3443 case VSOCK: 3444 xvn[n].xv_socket = vp->v_socket; 3445 break; 3446 case VFIFO: 3447 xvn[n].xv_fifo = vp->v_fifoinfo; 3448 break; 3449 case VNON: 3450 case VBAD: 3451 default: 3452 /* shouldn't happen? */ 3453 vrele(vp); 3454 continue; 3455 } 3456 vrele(vp); 3457 ++n; 3458 } 3459 MNT_IUNLOCK(mp); 3460 mtx_lock(&mountlist_mtx); 3461 vfs_unbusy(mp); 3462 if (n == len) 3463 break; 3464 } 3465 mtx_unlock(&mountlist_mtx); 3466 3467 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3468 free(xvn, M_TEMP); 3469 return (error); 3470} 3471 3472SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | 3473 CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", 3474 ""); 3475#endif 3476 3477/* 3478 * Unmount all filesystems. The list is traversed in reverse order 3479 * of mounting to avoid dependencies. 3480 */ 3481void 3482vfs_unmountall(void) 3483{ 3484 struct mount *mp; 3485 struct thread *td; 3486 int error; 3487 3488 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); 3489 td = curthread; 3490 3491 /* 3492 * Since this only runs when rebooting, it is not interlocked. 3493 */ 3494 while(!TAILQ_EMPTY(&mountlist)) { 3495 mp = TAILQ_LAST(&mountlist, mntlist); 3496 vfs_ref(mp); 3497 error = dounmount(mp, MNT_FORCE, td); 3498 if (error != 0) { 3499 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3500 /* 3501 * XXX: Due to the way in which we mount the root 3502 * file system off of devfs, devfs will generate a 3503 * "busy" warning when we try to unmount it before 3504 * the root. Don't print a warning as a result in 3505 * order to avoid false positive errors that may 3506 * cause needless upset. 3507 */ 3508 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3509 printf("unmount of %s failed (", 3510 mp->mnt_stat.f_mntonname); 3511 if (error == EBUSY) 3512 printf("BUSY)\n"); 3513 else 3514 printf("%d)\n", error); 3515 } 3516 } else { 3517 /* The unmount has removed mp from the mountlist */ 3518 } 3519 } 3520} 3521 3522/* 3523 * perform msync on all vnodes under a mount point 3524 * the mount point must be locked. 3525 */ 3526void 3527vfs_msync(struct mount *mp, int flags) 3528{ 3529 struct vnode *vp, *mvp; 3530 struct vm_object *obj; 3531 3532 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 3533 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { 3534 obj = vp->v_object; 3535 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && 3536 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3537 if (!vget(vp, 3538 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3539 curthread)) { 3540 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3541 vput(vp); 3542 continue; 3543 } 3544 3545 obj = vp->v_object; 3546 if (obj != NULL) { 3547 VM_OBJECT_WLOCK(obj); 3548 vm_object_page_clean(obj, 0, 0, 3549 flags == MNT_WAIT ? 3550 OBJPC_SYNC : OBJPC_NOSYNC); 3551 VM_OBJECT_WUNLOCK(obj); 3552 } 3553 vput(vp); 3554 } 3555 } else 3556 VI_UNLOCK(vp); 3557 } 3558} 3559 3560static void 3561destroy_vpollinfo_free(struct vpollinfo *vi) 3562{ 3563 3564 knlist_destroy(&vi->vpi_selinfo.si_note); 3565 mtx_destroy(&vi->vpi_lock); 3566 uma_zfree(vnodepoll_zone, vi); 3567} 3568 3569static void 3570destroy_vpollinfo(struct vpollinfo *vi) 3571{ 3572 3573 knlist_clear(&vi->vpi_selinfo.si_note, 1); 3574 seldrain(&vi->vpi_selinfo); 3575 destroy_vpollinfo_free(vi); 3576} 3577 3578/* 3579 * Initalize per-vnode helper structure to hold poll-related state. 3580 */ 3581void 3582v_addpollinfo(struct vnode *vp) 3583{ 3584 struct vpollinfo *vi; 3585 3586 if (vp->v_pollinfo != NULL) 3587 return; 3588 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3589 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3590 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3591 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); 3592 VI_LOCK(vp); 3593 if (vp->v_pollinfo != NULL) { 3594 VI_UNLOCK(vp); 3595 destroy_vpollinfo_free(vi); 3596 return; 3597 } 3598 vp->v_pollinfo = vi; 3599 VI_UNLOCK(vp); 3600} 3601 3602/* 3603 * Record a process's interest in events which might happen to 3604 * a vnode. Because poll uses the historic select-style interface 3605 * internally, this routine serves as both the ``check for any 3606 * pending events'' and the ``record my interest in future events'' 3607 * functions. (These are done together, while the lock is held, 3608 * to avoid race conditions.) 3609 */ 3610int 3611vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3612{ 3613 3614 v_addpollinfo(vp); 3615 mtx_lock(&vp->v_pollinfo->vpi_lock); 3616 if (vp->v_pollinfo->vpi_revents & events) { 3617 /* 3618 * This leaves events we are not interested 3619 * in available for the other process which 3620 * which presumably had requested them 3621 * (otherwise they would never have been 3622 * recorded). 3623 */ 3624 events &= vp->v_pollinfo->vpi_revents; 3625 vp->v_pollinfo->vpi_revents &= ~events; 3626 3627 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3628 return (events); 3629 } 3630 vp->v_pollinfo->vpi_events |= events; 3631 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3632 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3633 return (0); 3634} 3635 3636/* 3637 * Routine to create and manage a filesystem syncer vnode. 3638 */ 3639#define sync_close ((int (*)(struct vop_close_args *))nullop) 3640static int sync_fsync(struct vop_fsync_args *); 3641static int sync_inactive(struct vop_inactive_args *); 3642static int sync_reclaim(struct vop_reclaim_args *); 3643 3644static struct vop_vector sync_vnodeops = { 3645 .vop_bypass = VOP_EOPNOTSUPP, 3646 .vop_close = sync_close, /* close */ 3647 .vop_fsync = sync_fsync, /* fsync */ 3648 .vop_inactive = sync_inactive, /* inactive */ 3649 .vop_reclaim = sync_reclaim, /* reclaim */ 3650 .vop_lock1 = vop_stdlock, /* lock */ 3651 .vop_unlock = vop_stdunlock, /* unlock */ 3652 .vop_islocked = vop_stdislocked, /* islocked */ 3653}; 3654 3655/* 3656 * Create a new filesystem syncer vnode for the specified mount point. 3657 */ 3658void 3659vfs_allocate_syncvnode(struct mount *mp) 3660{ 3661 struct vnode *vp; 3662 struct bufobj *bo; 3663 static long start, incr, next; 3664 int error; 3665 3666 /* Allocate a new vnode */ 3667 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); 3668 if (error != 0) 3669 panic("vfs_allocate_syncvnode: getnewvnode() failed"); 3670 vp->v_type = VNON; 3671 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3672 vp->v_vflag |= VV_FORCEINSMQ; 3673 error = insmntque(vp, mp); 3674 if (error != 0) 3675 panic("vfs_allocate_syncvnode: insmntque() failed"); 3676 vp->v_vflag &= ~VV_FORCEINSMQ; 3677 VOP_UNLOCK(vp, 0); 3678 /* 3679 * Place the vnode onto the syncer worklist. We attempt to 3680 * scatter them about on the list so that they will go off 3681 * at evenly distributed times even if all the filesystems 3682 * are mounted at once. 3683 */ 3684 next += incr; 3685 if (next == 0 || next > syncer_maxdelay) { 3686 start /= 2; 3687 incr /= 2; 3688 if (start == 0) { 3689 start = syncer_maxdelay / 2; 3690 incr = syncer_maxdelay; 3691 } 3692 next = start; 3693 } 3694 bo = &vp->v_bufobj; 3695 BO_LOCK(bo); 3696 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3697 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3698 mtx_lock(&sync_mtx); 3699 sync_vnode_count++; 3700 if (mp->mnt_syncer == NULL) { 3701 mp->mnt_syncer = vp; 3702 vp = NULL; 3703 } 3704 mtx_unlock(&sync_mtx); 3705 BO_UNLOCK(bo); 3706 if (vp != NULL) { 3707 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3708 vgone(vp); 3709 vput(vp); 3710 } 3711} 3712 3713void 3714vfs_deallocate_syncvnode(struct mount *mp) 3715{ 3716 struct vnode *vp; 3717 3718 mtx_lock(&sync_mtx); 3719 vp = mp->mnt_syncer; 3720 if (vp != NULL) 3721 mp->mnt_syncer = NULL; 3722 mtx_unlock(&sync_mtx); 3723 if (vp != NULL) 3724 vrele(vp); 3725} 3726 3727/* 3728 * Do a lazy sync of the filesystem. 3729 */ 3730static int 3731sync_fsync(struct vop_fsync_args *ap) 3732{ 3733 struct vnode *syncvp = ap->a_vp; 3734 struct mount *mp = syncvp->v_mount; 3735 int error, save; 3736 struct bufobj *bo; 3737 3738 /* 3739 * We only need to do something if this is a lazy evaluation. 3740 */ 3741 if (ap->a_waitfor != MNT_LAZY) 3742 return (0); 3743 3744 /* 3745 * Move ourselves to the back of the sync list. 3746 */ 3747 bo = &syncvp->v_bufobj; 3748 BO_LOCK(bo); 3749 vn_syncer_add_to_worklist(bo, syncdelay); 3750 BO_UNLOCK(bo); 3751 3752 /* 3753 * Walk the list of vnodes pushing all that are dirty and 3754 * not already on the sync list. 3755 */ 3756 if (vfs_busy(mp, MBF_NOWAIT) != 0) 3757 return (0); 3758 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3759 vfs_unbusy(mp); 3760 return (0); 3761 } 3762 save = curthread_pflags_set(TDP_SYNCIO); 3763 vfs_msync(mp, MNT_NOWAIT); 3764 error = VFS_SYNC(mp, MNT_LAZY); 3765 curthread_pflags_restore(save); 3766 vn_finished_write(mp); 3767 vfs_unbusy(mp); 3768 return (error); 3769} 3770 3771/* 3772 * The syncer vnode is no referenced. 3773 */ 3774static int 3775sync_inactive(struct vop_inactive_args *ap) 3776{ 3777 3778 vgone(ap->a_vp); 3779 return (0); 3780} 3781 3782/* 3783 * The syncer vnode is no longer needed and is being decommissioned. 3784 * 3785 * Modifications to the worklist must be protected by sync_mtx. 3786 */ 3787static int 3788sync_reclaim(struct vop_reclaim_args *ap) 3789{ 3790 struct vnode *vp = ap->a_vp; 3791 struct bufobj *bo; 3792 3793 bo = &vp->v_bufobj; 3794 BO_LOCK(bo); 3795 mtx_lock(&sync_mtx); 3796 if (vp->v_mount->mnt_syncer == vp) 3797 vp->v_mount->mnt_syncer = NULL; 3798 if (bo->bo_flag & BO_ONWORKLST) { 3799 LIST_REMOVE(bo, bo_synclist); 3800 syncer_worklist_len--; 3801 sync_vnode_count--; 3802 bo->bo_flag &= ~BO_ONWORKLST; 3803 } 3804 mtx_unlock(&sync_mtx); 3805 BO_UNLOCK(bo); 3806 3807 return (0); 3808} 3809 3810/* 3811 * Check if vnode represents a disk device 3812 */ 3813int 3814vn_isdisk(struct vnode *vp, int *errp) 3815{ 3816 int error; 3817 3818 error = 0; 3819 dev_lock(); 3820 if (vp->v_type != VCHR) 3821 error = ENOTBLK; 3822 else if (vp->v_rdev == NULL) 3823 error = ENXIO; 3824 else if (vp->v_rdev->si_devsw == NULL) 3825 error = ENXIO; 3826 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3827 error = ENOTBLK; 3828 dev_unlock(); 3829 if (errp != NULL) 3830 *errp = error; 3831 return (error == 0); 3832} 3833 3834/* 3835 * Common filesystem object access control check routine. Accepts a 3836 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3837 * and optional call-by-reference privused argument allowing vaccess() 3838 * to indicate to the caller whether privilege was used to satisfy the 3839 * request (obsoleted). Returns 0 on success, or an errno on failure. 3840 */ 3841int 3842vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3843 accmode_t accmode, struct ucred *cred, int *privused) 3844{ 3845 accmode_t dac_granted; 3846 accmode_t priv_granted; 3847 3848 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, 3849 ("invalid bit in accmode")); 3850 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), 3851 ("VAPPEND without VWRITE")); 3852 3853 /* 3854 * Look for a normal, non-privileged way to access the file/directory 3855 * as requested. If it exists, go with that. 3856 */ 3857 3858 if (privused != NULL) 3859 *privused = 0; 3860 3861 dac_granted = 0; 3862 3863 /* Check the owner. */ 3864 if (cred->cr_uid == file_uid) { 3865 dac_granted |= VADMIN; 3866 if (file_mode & S_IXUSR) 3867 dac_granted |= VEXEC; 3868 if (file_mode & S_IRUSR) 3869 dac_granted |= VREAD; 3870 if (file_mode & S_IWUSR) 3871 dac_granted |= (VWRITE | VAPPEND); 3872 3873 if ((accmode & dac_granted) == accmode) 3874 return (0); 3875 3876 goto privcheck; 3877 } 3878 3879 /* Otherwise, check the groups (first match) */ 3880 if (groupmember(file_gid, cred)) { 3881 if (file_mode & S_IXGRP) 3882 dac_granted |= VEXEC; 3883 if (file_mode & S_IRGRP) 3884 dac_granted |= VREAD; 3885 if (file_mode & S_IWGRP) 3886 dac_granted |= (VWRITE | VAPPEND); 3887 3888 if ((accmode & dac_granted) == accmode) 3889 return (0); 3890 3891 goto privcheck; 3892 } 3893 3894 /* Otherwise, check everyone else. */ 3895 if (file_mode & S_IXOTH) 3896 dac_granted |= VEXEC; 3897 if (file_mode & S_IROTH) 3898 dac_granted |= VREAD; 3899 if (file_mode & S_IWOTH) 3900 dac_granted |= (VWRITE | VAPPEND); 3901 if ((accmode & dac_granted) == accmode) 3902 return (0); 3903 3904privcheck: 3905 /* 3906 * Build a privilege mask to determine if the set of privileges 3907 * satisfies the requirements when combined with the granted mask 3908 * from above. For each privilege, if the privilege is required, 3909 * bitwise or the request type onto the priv_granted mask. 3910 */ 3911 priv_granted = 0; 3912 3913 if (type == VDIR) { 3914 /* 3915 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3916 * requests, instead of PRIV_VFS_EXEC. 3917 */ 3918 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3919 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3920 priv_granted |= VEXEC; 3921 } else { 3922 /* 3923 * Ensure that at least one execute bit is on. Otherwise, 3924 * a privileged user will always succeed, and we don't want 3925 * this to happen unless the file really is executable. 3926 */ 3927 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3928 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && 3929 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3930 priv_granted |= VEXEC; 3931 } 3932 3933 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3934 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3935 priv_granted |= VREAD; 3936 3937 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3938 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3939 priv_granted |= (VWRITE | VAPPEND); 3940 3941 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3942 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3943 priv_granted |= VADMIN; 3944 3945 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3946 /* XXX audit: privilege used */ 3947 if (privused != NULL) 3948 *privused = 1; 3949 return (0); 3950 } 3951 3952 return ((accmode & VADMIN) ? EPERM : EACCES); 3953} 3954 3955/* 3956 * Credential check based on process requesting service, and per-attribute 3957 * permissions. 3958 */ 3959int 3960extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3961 struct thread *td, accmode_t accmode) 3962{ 3963 3964 /* 3965 * Kernel-invoked always succeeds. 3966 */ 3967 if (cred == NOCRED) 3968 return (0); 3969 3970 /* 3971 * Do not allow privileged processes in jail to directly manipulate 3972 * system attributes. 3973 */ 3974 switch (attrnamespace) { 3975 case EXTATTR_NAMESPACE_SYSTEM: 3976 /* Potentially should be: return (EPERM); */ 3977 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3978 case EXTATTR_NAMESPACE_USER: 3979 return (VOP_ACCESS(vp, accmode, cred, td)); 3980 default: 3981 return (EPERM); 3982 } 3983} 3984 3985#ifdef DEBUG_VFS_LOCKS 3986/* 3987 * This only exists to supress warnings from unlocked specfs accesses. It is 3988 * no longer ok to have an unlocked VFS. 3989 */ 3990#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3991 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3992 3993int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3994SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, 3995 "Drop into debugger on lock violation"); 3996 3997int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3998SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 3999 0, "Check for interlock across VOPs"); 4000 4001int vfs_badlock_print = 1; /* Print lock violations. */ 4002SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 4003 0, "Print lock violations"); 4004 4005#ifdef KDB 4006int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 4007SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, 4008 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); 4009#endif 4010 4011static void 4012vfs_badlock(const char *msg, const char *str, struct vnode *vp) 4013{ 4014 4015#ifdef KDB 4016 if (vfs_badlock_backtrace) 4017 kdb_backtrace(); 4018#endif 4019 if (vfs_badlock_print) 4020 printf("%s: %p %s\n", str, (void *)vp, msg); 4021 if (vfs_badlock_ddb) 4022 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 4023} 4024 4025void 4026assert_vi_locked(struct vnode *vp, const char *str) 4027{ 4028 4029 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 4030 vfs_badlock("interlock is not locked but should be", str, vp); 4031} 4032 4033void 4034assert_vi_unlocked(struct vnode *vp, const char *str) 4035{ 4036 4037 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 4038 vfs_badlock("interlock is locked but should not be", str, vp); 4039} 4040 4041void 4042assert_vop_locked(struct vnode *vp, const char *str) 4043{ 4044 int locked; 4045 4046 if (!IGNORE_LOCK(vp)) { 4047 locked = VOP_ISLOCKED(vp); 4048 if (locked == 0 || locked == LK_EXCLOTHER) 4049 vfs_badlock("is not locked but should be", str, vp); 4050 } 4051} 4052 4053void 4054assert_vop_unlocked(struct vnode *vp, const char *str) 4055{ 4056 4057 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 4058 vfs_badlock("is locked but should not be", str, vp); 4059} 4060 4061void 4062assert_vop_elocked(struct vnode *vp, const char *str) 4063{ 4064 4065 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 4066 vfs_badlock("is not exclusive locked but should be", str, vp); 4067} 4068 4069#if 0 4070void 4071assert_vop_elocked_other(struct vnode *vp, const char *str) 4072{ 4073 4074 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 4075 vfs_badlock("is not exclusive locked by another thread", 4076 str, vp); 4077} 4078 4079void 4080assert_vop_slocked(struct vnode *vp, const char *str) 4081{ 4082 4083 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 4084 vfs_badlock("is not locked shared but should be", str, vp); 4085} 4086#endif /* 0 */ 4087#endif /* DEBUG_VFS_LOCKS */ 4088 4089void 4090vop_rename_fail(struct vop_rename_args *ap) 4091{ 4092 4093 if (ap->a_tvp != NULL) 4094 vput(ap->a_tvp); 4095 if (ap->a_tdvp == ap->a_tvp) 4096 vrele(ap->a_tdvp); 4097 else 4098 vput(ap->a_tdvp); 4099 vrele(ap->a_fdvp); 4100 vrele(ap->a_fvp); 4101} 4102 4103void 4104vop_rename_pre(void *ap) 4105{ 4106 struct vop_rename_args *a = ap; 4107 4108#ifdef DEBUG_VFS_LOCKS 4109 if (a->a_tvp) 4110 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 4111 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 4112 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 4113 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 4114 4115 /* Check the source (from). */ 4116 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && 4117 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) 4118 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 4119 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) 4120 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 4121 4122 /* Check the target. */ 4123 if (a->a_tvp) 4124 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 4125 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 4126#endif 4127 if (a->a_tdvp != a->a_fdvp) 4128 vhold(a->a_fdvp); 4129 if (a->a_tvp != a->a_fvp) 4130 vhold(a->a_fvp); 4131 vhold(a->a_tdvp); 4132 if (a->a_tvp) 4133 vhold(a->a_tvp); 4134} 4135 4136void 4137vop_strategy_pre(void *ap) 4138{ 4139#ifdef DEBUG_VFS_LOCKS 4140 struct vop_strategy_args *a; 4141 struct buf *bp; 4142 4143 a = ap; 4144 bp = a->a_bp; 4145 4146 /* 4147 * Cluster ops lock their component buffers but not the IO container. 4148 */ 4149 if ((bp->b_flags & B_CLUSTER) != 0) 4150 return; 4151 4152 if (panicstr == NULL && !BUF_ISLOCKED(bp)) { 4153 if (vfs_badlock_print) 4154 printf( 4155 "VOP_STRATEGY: bp is not locked but should be\n"); 4156 if (vfs_badlock_ddb) 4157 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 4158 } 4159#endif 4160} 4161 4162void 4163vop_lock_pre(void *ap) 4164{ 4165#ifdef DEBUG_VFS_LOCKS 4166 struct vop_lock1_args *a = ap; 4167 4168 if ((a->a_flags & LK_INTERLOCK) == 0) 4169 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 4170 else 4171 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 4172#endif 4173} 4174 4175void 4176vop_lock_post(void *ap, int rc) 4177{ 4178#ifdef DEBUG_VFS_LOCKS 4179 struct vop_lock1_args *a = ap; 4180 4181 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 4182 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) 4183 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 4184#endif 4185} 4186 4187void 4188vop_unlock_pre(void *ap) 4189{ 4190#ifdef DEBUG_VFS_LOCKS 4191 struct vop_unlock_args *a = ap; 4192 4193 if (a->a_flags & LK_INTERLOCK) 4194 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 4195 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 4196#endif 4197} 4198 4199void 4200vop_unlock_post(void *ap, int rc) 4201{ 4202#ifdef DEBUG_VFS_LOCKS 4203 struct vop_unlock_args *a = ap; 4204 4205 if (a->a_flags & LK_INTERLOCK) 4206 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 4207#endif 4208} 4209 4210void 4211vop_create_post(void *ap, int rc) 4212{ 4213 struct vop_create_args *a = ap; 4214 4215 if (!rc) 4216 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4217} 4218 4219void 4220vop_deleteextattr_post(void *ap, int rc) 4221{ 4222 struct vop_deleteextattr_args *a = ap; 4223 4224 if (!rc) 4225 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4226} 4227 4228void 4229vop_link_post(void *ap, int rc) 4230{ 4231 struct vop_link_args *a = ap; 4232 4233 if (!rc) { 4234 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 4235 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 4236 } 4237} 4238 4239void 4240vop_mkdir_post(void *ap, int rc) 4241{ 4242 struct vop_mkdir_args *a = ap; 4243 4244 if (!rc) 4245 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4246} 4247 4248void 4249vop_mknod_post(void *ap, int rc) 4250{ 4251 struct vop_mknod_args *a = ap; 4252 4253 if (!rc) 4254 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4255} 4256 4257void 4258vop_remove_post(void *ap, int rc) 4259{ 4260 struct vop_remove_args *a = ap; 4261 4262 if (!rc) { 4263 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4264 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4265 } 4266} 4267 4268void 4269vop_rename_post(void *ap, int rc) 4270{ 4271 struct vop_rename_args *a = ap; 4272 4273 if (!rc) { 4274 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 4275 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 4276 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 4277 if (a->a_tvp) 4278 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 4279 } 4280 if (a->a_tdvp != a->a_fdvp) 4281 vdrop(a->a_fdvp); 4282 if (a->a_tvp != a->a_fvp) 4283 vdrop(a->a_fvp); 4284 vdrop(a->a_tdvp); 4285 if (a->a_tvp) 4286 vdrop(a->a_tvp); 4287} 4288 4289void 4290vop_rmdir_post(void *ap, int rc) 4291{ 4292 struct vop_rmdir_args *a = ap; 4293 4294 if (!rc) { 4295 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4296 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4297 } 4298} 4299 4300void 4301vop_setattr_post(void *ap, int rc) 4302{ 4303 struct vop_setattr_args *a = ap; 4304 4305 if (!rc) 4306 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4307} 4308 4309void 4310vop_setextattr_post(void *ap, int rc) 4311{ 4312 struct vop_setextattr_args *a = ap; 4313 4314 if (!rc) 4315 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4316} 4317 4318void 4319vop_symlink_post(void *ap, int rc) 4320{ 4321 struct vop_symlink_args *a = ap; 4322 4323 if (!rc) 4324 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4325} 4326 4327static struct knlist fs_knlist; 4328 4329static void 4330vfs_event_init(void *arg) 4331{ 4332 knlist_init_mtx(&fs_knlist, NULL); 4333} 4334/* XXX - correct order? */ 4335SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 4336 4337void 4338vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) 4339{ 4340 4341 KNOTE_UNLOCKED(&fs_knlist, event); 4342} 4343 4344static int filt_fsattach(struct knote *kn); 4345static void filt_fsdetach(struct knote *kn); 4346static int filt_fsevent(struct knote *kn, long hint); 4347 4348struct filterops fs_filtops = { 4349 .f_isfd = 0, 4350 .f_attach = filt_fsattach, 4351 .f_detach = filt_fsdetach, 4352 .f_event = filt_fsevent 4353}; 4354 4355static int 4356filt_fsattach(struct knote *kn) 4357{ 4358 4359 kn->kn_flags |= EV_CLEAR; 4360 knlist_add(&fs_knlist, kn, 0); 4361 return (0); 4362} 4363 4364static void 4365filt_fsdetach(struct knote *kn) 4366{ 4367 4368 knlist_remove(&fs_knlist, kn, 0); 4369} 4370 4371static int 4372filt_fsevent(struct knote *kn, long hint) 4373{ 4374 4375 kn->kn_fflags |= hint; 4376 return (kn->kn_fflags != 0); 4377} 4378 4379static int 4380sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4381{ 4382 struct vfsidctl vc; 4383 int error; 4384 struct mount *mp; 4385 4386 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4387 if (error) 4388 return (error); 4389 if (vc.vc_vers != VFS_CTL_VERS1) 4390 return (EINVAL); 4391 mp = vfs_getvfs(&vc.vc_fsid); 4392 if (mp == NULL) 4393 return (ENOENT); 4394 /* ensure that a specific sysctl goes to the right filesystem. */ 4395 if (strcmp(vc.vc_fstypename, "*") != 0 && 4396 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4397 vfs_rel(mp); 4398 return (EINVAL); 4399 } 4400 VCTLTOREQ(&vc, req); 4401 error = VFS_SYSCTL(mp, vc.vc_op, req); 4402 vfs_rel(mp); 4403 return (error); 4404} 4405 4406SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, 4407 NULL, 0, sysctl_vfs_ctl, "", 4408 "Sysctl by fsid"); 4409 4410/* 4411 * Function to initialize a va_filerev field sensibly. 4412 * XXX: Wouldn't a random number make a lot more sense ?? 4413 */ 4414u_quad_t 4415init_va_filerev(void) 4416{ 4417 struct bintime bt; 4418 4419 getbinuptime(&bt); 4420 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4421} 4422 4423static int filt_vfsread(struct knote *kn, long hint); 4424static int filt_vfswrite(struct knote *kn, long hint); 4425static int filt_vfsvnode(struct knote *kn, long hint); 4426static void filt_vfsdetach(struct knote *kn); 4427static struct filterops vfsread_filtops = { 4428 .f_isfd = 1, 4429 .f_detach = filt_vfsdetach, 4430 .f_event = filt_vfsread 4431}; 4432static struct filterops vfswrite_filtops = { 4433 .f_isfd = 1, 4434 .f_detach = filt_vfsdetach, 4435 .f_event = filt_vfswrite 4436}; 4437static struct filterops vfsvnode_filtops = { 4438 .f_isfd = 1, 4439 .f_detach = filt_vfsdetach, 4440 .f_event = filt_vfsvnode 4441}; 4442 4443static void 4444vfs_knllock(void *arg) 4445{ 4446 struct vnode *vp = arg; 4447 4448 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4449} 4450 4451static void 4452vfs_knlunlock(void *arg) 4453{ 4454 struct vnode *vp = arg; 4455 4456 VOP_UNLOCK(vp, 0); 4457} 4458 4459static void 4460vfs_knl_assert_locked(void *arg) 4461{ 4462#ifdef DEBUG_VFS_LOCKS 4463 struct vnode *vp = arg; 4464 4465 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); 4466#endif 4467} 4468 4469static void 4470vfs_knl_assert_unlocked(void *arg) 4471{ 4472#ifdef DEBUG_VFS_LOCKS 4473 struct vnode *vp = arg; 4474 4475 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); 4476#endif 4477} 4478 4479int 4480vfs_kqfilter(struct vop_kqfilter_args *ap) 4481{ 4482 struct vnode *vp = ap->a_vp; 4483 struct knote *kn = ap->a_kn; 4484 struct knlist *knl; 4485 4486 switch (kn->kn_filter) { 4487 case EVFILT_READ: 4488 kn->kn_fop = &vfsread_filtops; 4489 break; 4490 case EVFILT_WRITE: 4491 kn->kn_fop = &vfswrite_filtops; 4492 break; 4493 case EVFILT_VNODE: 4494 kn->kn_fop = &vfsvnode_filtops; 4495 break; 4496 default: 4497 return (EINVAL); 4498 } 4499 4500 kn->kn_hook = (caddr_t)vp; 4501 4502 v_addpollinfo(vp); 4503 if (vp->v_pollinfo == NULL) 4504 return (ENOMEM); 4505 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4506 vhold(vp); 4507 knlist_add(knl, kn, 0); 4508 4509 return (0); 4510} 4511 4512/* 4513 * Detach knote from vnode 4514 */ 4515static void 4516filt_vfsdetach(struct knote *kn) 4517{ 4518 struct vnode *vp = (struct vnode *)kn->kn_hook; 4519 4520 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4521 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4522 vdrop(vp); 4523} 4524 4525/*ARGSUSED*/ 4526static int 4527filt_vfsread(struct knote *kn, long hint) 4528{ 4529 struct vnode *vp = (struct vnode *)kn->kn_hook; 4530 struct vattr va; 4531 int res; 4532 4533 /* 4534 * filesystem is gone, so set the EOF flag and schedule 4535 * the knote for deletion. 4536 */ 4537 if (hint == NOTE_REVOKE) { 4538 VI_LOCK(vp); 4539 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4540 VI_UNLOCK(vp); 4541 return (1); 4542 } 4543 4544 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4545 return (0); 4546 4547 VI_LOCK(vp); 4548 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4549 res = (kn->kn_data != 0); 4550 VI_UNLOCK(vp); 4551 return (res); 4552} 4553 4554/*ARGSUSED*/ 4555static int 4556filt_vfswrite(struct knote *kn, long hint) 4557{ 4558 struct vnode *vp = (struct vnode *)kn->kn_hook; 4559 4560 VI_LOCK(vp); 4561 4562 /* 4563 * filesystem is gone, so set the EOF flag and schedule 4564 * the knote for deletion. 4565 */ 4566 if (hint == NOTE_REVOKE) 4567 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4568 4569 kn->kn_data = 0; 4570 VI_UNLOCK(vp); 4571 return (1); 4572} 4573 4574static int 4575filt_vfsvnode(struct knote *kn, long hint) 4576{ 4577 struct vnode *vp = (struct vnode *)kn->kn_hook; 4578 int res; 4579 4580 VI_LOCK(vp); 4581 if (kn->kn_sfflags & hint) 4582 kn->kn_fflags |= hint; 4583 if (hint == NOTE_REVOKE) { 4584 kn->kn_flags |= EV_EOF; 4585 VI_UNLOCK(vp); 4586 return (1); 4587 } 4588 res = (kn->kn_fflags != 0); 4589 VI_UNLOCK(vp); 4590 return (res); 4591} 4592 4593int 4594vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4595{ 4596 int error; 4597 4598 if (dp->d_reclen > ap->a_uio->uio_resid) 4599 return (ENAMETOOLONG); 4600 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4601 if (error) { 4602 if (ap->a_ncookies != NULL) { 4603 if (ap->a_cookies != NULL) 4604 free(ap->a_cookies, M_TEMP); 4605 ap->a_cookies = NULL; 4606 *ap->a_ncookies = 0; 4607 } 4608 return (error); 4609 } 4610 if (ap->a_ncookies == NULL) 4611 return (0); 4612 4613 KASSERT(ap->a_cookies, 4614 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4615 4616 *ap->a_cookies = realloc(*ap->a_cookies, 4617 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4618 (*ap->a_cookies)[*ap->a_ncookies] = off; 4619 return (0); 4620} 4621 4622/* 4623 * Mark for update the access time of the file if the filesystem 4624 * supports VOP_MARKATIME. This functionality is used by execve and 4625 * mmap, so we want to avoid the I/O implied by directly setting 4626 * va_atime for the sake of efficiency. 4627 */ 4628void 4629vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4630{ 4631 struct mount *mp; 4632 4633 mp = vp->v_mount; 4634 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); 4635 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 4636 (void)VOP_MARKATIME(vp); 4637} 4638 4639/* 4640 * The purpose of this routine is to remove granularity from accmode_t, 4641 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, 4642 * VADMIN and VAPPEND. 4643 * 4644 * If it returns 0, the caller is supposed to continue with the usual 4645 * access checks using 'accmode' as modified by this routine. If it 4646 * returns nonzero value, the caller is supposed to return that value 4647 * as errno. 4648 * 4649 * Note that after this routine runs, accmode may be zero. 4650 */ 4651int 4652vfs_unixify_accmode(accmode_t *accmode) 4653{ 4654 /* 4655 * There is no way to specify explicit "deny" rule using 4656 * file mode or POSIX.1e ACLs. 4657 */ 4658 if (*accmode & VEXPLICIT_DENY) { 4659 *accmode = 0; 4660 return (0); 4661 } 4662 4663 /* 4664 * None of these can be translated into usual access bits. 4665 * Also, the common case for NFSv4 ACLs is to not contain 4666 * either of these bits. Caller should check for VWRITE 4667 * on the containing directory instead. 4668 */ 4669 if (*accmode & (VDELETE_CHILD | VDELETE)) 4670 return (EPERM); 4671 4672 if (*accmode & VADMIN_PERMS) { 4673 *accmode &= ~VADMIN_PERMS; 4674 *accmode |= VADMIN; 4675 } 4676 4677 /* 4678 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL 4679 * or VSYNCHRONIZE using file mode or POSIX.1e ACL. 4680 */ 4681 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); 4682 4683 return (0); 4684} 4685 4686/* 4687 * These are helper functions for filesystems to traverse all 4688 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. 4689 * 4690 * This interface replaces MNT_VNODE_FOREACH. 4691 */ 4692 4693MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 4694 4695struct vnode * 4696__mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) 4697{ 4698 struct vnode *vp; 4699 4700 if (should_yield()) 4701 kern_yield(PRI_USER); 4702 MNT_ILOCK(mp); 4703 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4704 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 4705 while (vp != NULL && (vp->v_type == VMARKER || 4706 (vp->v_iflag & VI_DOOMED) != 0)) 4707 vp = TAILQ_NEXT(vp, v_nmntvnodes); 4708 4709 /* Check if we are done */ 4710 if (vp == NULL) { 4711 __mnt_vnode_markerfree_all(mvp, mp); 4712 /* MNT_IUNLOCK(mp); -- done in above function */ 4713 mtx_assert(MNT_MTX(mp), MA_NOTOWNED); 4714 return (NULL); 4715 } 4716 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 4717 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 4718 VI_LOCK(vp); 4719 MNT_IUNLOCK(mp); 4720 return (vp); 4721} 4722 4723struct vnode * 4724__mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) 4725{ 4726 struct vnode *vp; 4727 4728 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); 4729 MNT_ILOCK(mp); 4730 MNT_REF(mp); 4731 (*mvp)->v_type = VMARKER; 4732 4733 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 4734 while (vp != NULL && (vp->v_type == VMARKER || 4735 (vp->v_iflag & VI_DOOMED) != 0)) 4736 vp = TAILQ_NEXT(vp, v_nmntvnodes); 4737 4738 /* Check if we are done */ 4739 if (vp == NULL) { 4740 MNT_REL(mp); 4741 MNT_IUNLOCK(mp); 4742 free(*mvp, M_VNODE_MARKER); 4743 *mvp = NULL; 4744 return (NULL); 4745 } 4746 (*mvp)->v_mount = mp; 4747 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 4748 VI_LOCK(vp); 4749 MNT_IUNLOCK(mp); 4750 return (vp); 4751} 4752 4753 4754void 4755__mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) 4756{ 4757 4758 if (*mvp == NULL) { 4759 MNT_IUNLOCK(mp); 4760 return; 4761 } 4762 4763 mtx_assert(MNT_MTX(mp), MA_OWNED); 4764 4765 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4766 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 4767 MNT_REL(mp); 4768 MNT_IUNLOCK(mp); 4769 free(*mvp, M_VNODE_MARKER); 4770 *mvp = NULL; 4771} 4772 4773/* 4774 * These are helper functions for filesystems to traverse their 4775 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h 4776 */ 4777static void 4778mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) 4779{ 4780 4781 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4782 4783 MNT_ILOCK(mp); 4784 MNT_REL(mp); 4785 MNT_IUNLOCK(mp); 4786 free(*mvp, M_VNODE_MARKER); 4787 *mvp = NULL; 4788} 4789 4790static struct vnode * 4791mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) 4792{ 4793 struct vnode *vp, *nvp; 4794 4795 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 4796 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4797restart: 4798 vp = TAILQ_NEXT(*mvp, v_actfreelist); 4799 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); 4800 while (vp != NULL) { 4801 if (vp->v_type == VMARKER) { 4802 vp = TAILQ_NEXT(vp, v_actfreelist); 4803 continue; 4804 } 4805 if (!VI_TRYLOCK(vp)) { 4806 if (mp_ncpus == 1 || should_yield()) { 4807 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist); 4808 mtx_unlock(&vnode_free_list_mtx); 4809 pause("vnacti", 1); 4810 mtx_lock(&vnode_free_list_mtx); 4811 goto restart; 4812 } 4813 continue; 4814 } 4815 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); 4816 KASSERT(vp->v_mount == mp || vp->v_mount == NULL, 4817 ("alien vnode on the active list %p %p", vp, mp)); 4818 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0) 4819 break; 4820 nvp = TAILQ_NEXT(vp, v_actfreelist); 4821 VI_UNLOCK(vp); 4822 vp = nvp; 4823 } 4824 4825 /* Check if we are done */ 4826 if (vp == NULL) { 4827 mtx_unlock(&vnode_free_list_mtx); 4828 mnt_vnode_markerfree_active(mvp, mp); 4829 return (NULL); 4830 } 4831 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist); 4832 mtx_unlock(&vnode_free_list_mtx); 4833 ASSERT_VI_LOCKED(vp, "active iter"); 4834 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp)); 4835 return (vp); 4836} 4837 4838struct vnode * 4839__mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) 4840{ 4841 4842 if (should_yield()) 4843 kern_yield(PRI_USER); 4844 mtx_lock(&vnode_free_list_mtx); 4845 return (mnt_vnode_next_active(mvp, mp)); 4846} 4847 4848struct vnode * 4849__mnt_vnode_first_active(struct vnode **mvp, struct mount *mp) 4850{ 4851 struct vnode *vp; 4852 4853 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); 4854 MNT_ILOCK(mp); 4855 MNT_REF(mp); 4856 MNT_IUNLOCK(mp); 4857 (*mvp)->v_type = VMARKER; 4858 (*mvp)->v_mount = mp; 4859 4860 mtx_lock(&vnode_free_list_mtx); 4861 vp = TAILQ_FIRST(&mp->mnt_activevnodelist); 4862 if (vp == NULL) { 4863 mtx_unlock(&vnode_free_list_mtx); 4864 mnt_vnode_markerfree_active(mvp, mp); 4865 return (NULL); 4866 } 4867 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist); 4868 return (mnt_vnode_next_active(mvp, mp)); 4869} 4870 4871void 4872__mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) 4873{ 4874 4875 if (*mvp == NULL) 4876 return; 4877 4878 mtx_lock(&vnode_free_list_mtx); 4879 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); 4880 mtx_unlock(&vnode_free_list_mtx); 4881 mnt_vnode_markerfree_active(mvp, mp); 4882} 4883