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("$FreeBSD$"); 43 44 #include "opt_ddb.h" 45 #include "opt_mac.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/event.h> 54 #include <sys/eventhandler.h> 55 #include <sys/extattr.h> 56 #include <sys/file.h> 57 #include <sys/fcntl.h> 58 #include <sys/kdb.h> 59 #include <sys/kernel.h> 60 #include <sys/kthread.h> 61 #include <sys/malloc.h> 62 #include <sys/mount.h> 63 #include <sys/namei.h> 64 #include <sys/priv.h> 65 #include <sys/reboot.h> 66 #include <sys/sleepqueue.h> 67 #include <sys/stat.h> 68 #include <sys/sysctl.h> 69 #include <sys/syslog.h> 70 #include <sys/vmmeter.h> 71 #include <sys/vnode.h> 72 73 #include <machine/stdarg.h> 74 75 #include <security/mac/mac_framework.h> 76 77 #include <vm/vm.h> 78 #include <vm/vm_object.h> 79 #include <vm/vm_extern.h> 80 #include <vm/pmap.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_page.h> 83 #include <vm/vm_kern.h> 84 #include <vm/uma.h> 85 86 #ifdef DDB 87 #include <ddb/ddb.h> 88 #endif 89 90 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure"); 91 92 static void delmntque(struct vnode *vp); 93 static void insmntque(struct vnode *vp, struct mount *mp); 94 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 95 int slpflag, int slptimeo); 96 static void syncer_shutdown(void *arg, int howto); 97 static int vtryrecycle(struct vnode *vp); 98 static void vbusy(struct vnode *vp); 99 static void vdropl(struct vnode *vp); 100 static void vinactive(struct vnode *, struct thread *); 101 static void v_incr_usecount(struct vnode *); 102 static void v_decr_usecount(struct vnode *); 103 static void v_decr_useonly(struct vnode *); 104 static void v_upgrade_usecount(struct vnode *); 105 static void vfree(struct vnode *); 106 static void vnlru_free(int); 107 static void vdestroy(struct vnode *); 108 static void vgonel(struct vnode *); 109 static void vfs_knllock(void *arg); 110 static void vfs_knlunlock(void *arg); 111 static int vfs_knllocked(void *arg); 112 113 114 /* 115 * Enable Giant pushdown based on whether or not the vm is mpsafe in this 116 * build. Without mpsafevm the buffer cache can not run Giant free. 117 */ 118 #if !defined(__powerpc__) 119 int mpsafe_vfs = 1; 120 #else 121 int mpsafe_vfs; 122 #endif 123 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs); 124 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0, 125 "MPSAFE VFS"); 126 127 /* 128 * Number of vnodes in existence. Increased whenever getnewvnode() 129 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed 130 * vnode. 131 */ 132 static unsigned long numvnodes; 133 134 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, ""); 135 136 /* 137 * Conversion tables for conversion from vnode types to inode formats 138 * and back. 139 */ 140 enum vtype iftovt_tab[16] = { 141 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 142 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 143 }; 144 int vttoif_tab[10] = { 145 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 146 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 147 }; 148 149 /* 150 * List of vnodes that are ready for recycling. 151 */ 152 static TAILQ_HEAD(freelst, vnode) vnode_free_list; 153 154 /* 155 * Free vnode target. Free vnodes may simply be files which have been stat'd 156 * but not read. This is somewhat common, and a small cache of such files 157 * should be kept to avoid recreation costs. 158 */ 159 static u_long wantfreevnodes; 160 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 161 /* Number of vnodes in the free list. */ 162 static u_long freevnodes; 163 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, ""); 164 165 /* 166 * Various variables used for debugging the new implementation of 167 * reassignbuf(). 168 * XXX these are probably of (very) limited utility now. 169 */ 170 static int reassignbufcalls; 171 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, ""); 172 173 /* 174 * Cache for the mount type id assigned to NFS. This is used for 175 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 176 */ 177 int nfs_mount_type = -1; 178 179 /* To keep more than one thread at a time from running vfs_getnewfsid */ 180 static struct mtx mntid_mtx; 181 182 /* 183 * Lock for any access to the following: 184 * vnode_free_list 185 * numvnodes 186 * freevnodes 187 */ 188 static struct mtx vnode_free_list_mtx; 189 190 /* Publicly exported FS */ 191 struct nfs_public nfs_pub; 192 193 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 194 static uma_zone_t vnode_zone; 195 static uma_zone_t vnodepoll_zone; 196 197 /* Set to 1 to print out reclaim of active vnodes */ 198 int prtactive; 199 200 /* 201 * The workitem queue. 202 * 203 * It is useful to delay writes of file data and filesystem metadata 204 * for tens of seconds so that quickly created and deleted files need 205 * not waste disk bandwidth being created and removed. To realize this, 206 * we append vnodes to a "workitem" queue. When running with a soft 207 * updates implementation, most pending metadata dependencies should 208 * not wait for more than a few seconds. Thus, mounted on block devices 209 * are delayed only about a half the time that file data is delayed. 210 * Similarly, directory updates are more critical, so are only delayed 211 * about a third the time that file data is delayed. Thus, there are 212 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 213 * one each second (driven off the filesystem syncer process). The 214 * syncer_delayno variable indicates the next queue that is to be processed. 215 * Items that need to be processed soon are placed in this queue: 216 * 217 * syncer_workitem_pending[syncer_delayno] 218 * 219 * A delay of fifteen seconds is done by placing the request fifteen 220 * entries later in the queue: 221 * 222 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 223 * 224 */ 225 static int syncer_delayno; 226 static long syncer_mask; 227 LIST_HEAD(synclist, bufobj); 228 static struct synclist *syncer_workitem_pending; 229 /* 230 * The sync_mtx protects: 231 * bo->bo_synclist 232 * sync_vnode_count 233 * syncer_delayno 234 * syncer_state 235 * syncer_workitem_pending 236 * syncer_worklist_len 237 * rushjob 238 */ 239 static struct mtx sync_mtx; 240 241 #define SYNCER_MAXDELAY 32 242 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 243 static int syncdelay = 30; /* max time to delay syncing data */ 244 static int filedelay = 30; /* time to delay syncing files */ 245 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, ""); 246 static int dirdelay = 29; /* time to delay syncing directories */ 247 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, ""); 248 static int metadelay = 28; /* time to delay syncing metadata */ 249 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, ""); 250 static int rushjob; /* number of slots to run ASAP */ 251 static int stat_rush_requests; /* number of times I/O speeded up */ 252 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, ""); 253 254 /* 255 * When shutting down the syncer, run it at four times normal speed. 256 */ 257 #define SYNCER_SHUTDOWN_SPEEDUP 4 258 static int sync_vnode_count; 259 static int syncer_worklist_len; 260 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 261 syncer_state; 262 263 /* 264 * Number of vnodes we want to exist at any one time. This is mostly used 265 * to size hash tables in vnode-related code. It is normally not used in 266 * getnewvnode(), as wantfreevnodes is normally nonzero.) 267 * 268 * XXX desiredvnodes is historical cruft and should not exist. 269 */ 270 int desiredvnodes; 271 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 272 &desiredvnodes, 0, "Maximum number of vnodes"); 273 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 274 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 275 static int vnlru_nowhere; 276 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 277 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 278 279 /* 280 * Macros to control when a vnode is freed and recycled. All require 281 * the vnode interlock. 282 */ 283 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 284 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 285 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 286 287 288 /* 289 * Initialize the vnode management data structures. 290 */ 291 #ifndef MAXVNODES_MAX 292 #define MAXVNODES_MAX 100000 293 #endif 294 static void 295 vntblinit(void *dummy __unused) 296 { 297 298 /* 299 * Desiredvnodes is a function of the physical memory size and 300 * the kernel's heap size. Specifically, desiredvnodes scales 301 * in proportion to the physical memory size until two fifths 302 * of the kernel's heap size is consumed by vnodes and vm 303 * objects. 304 */ 305 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size / 306 (5 * (sizeof(struct vm_object) + sizeof(struct vnode)))); 307 if (desiredvnodes > MAXVNODES_MAX) { 308 if (bootverbose) 309 printf("Reducing kern.maxvnodes %d -> %d\n", 310 desiredvnodes, MAXVNODES_MAX); 311 desiredvnodes = MAXVNODES_MAX; 312 } 313 wantfreevnodes = desiredvnodes / 4; 314 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 315 TAILQ_INIT(&vnode_free_list); 316 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 317 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 318 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 319 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 320 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 321 /* 322 * Initialize the filesystem syncer. 323 */ 324 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, 325 &syncer_mask); 326 syncer_maxdelay = syncer_mask + 1; 327 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 328 } 329 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL) 330 331 332 /* 333 * Mark a mount point as busy. Used to synchronize access and to delay 334 * unmounting. Interlock is not released on failure. 335 */ 336 int 337 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp, 338 struct thread *td) 339 { 340 int lkflags; 341 342 MNT_ILOCK(mp); 343 MNT_REF(mp); 344 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 345 if (flags & LK_NOWAIT) { 346 MNT_REL(mp); 347 MNT_IUNLOCK(mp); 348 return (ENOENT); 349 } 350 if (interlkp) 351 mtx_unlock(interlkp); 352 mp->mnt_kern_flag |= MNTK_MWAIT; 353 /* 354 * Since all busy locks are shared except the exclusive 355 * lock granted when unmounting, the only place that a 356 * wakeup needs to be done is at the release of the 357 * exclusive lock at the end of dounmount. 358 */ 359 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0); 360 MNT_REL(mp); 361 MNT_IUNLOCK(mp); 362 if (interlkp) 363 mtx_lock(interlkp); 364 return (ENOENT); 365 } 366 if (interlkp) 367 mtx_unlock(interlkp); 368 lkflags = LK_SHARED | LK_INTERLOCK; 369 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td)) 370 panic("vfs_busy: unexpected lock failure"); 371 return (0); 372 } 373 374 /* 375 * Free a busy filesystem. 376 */ 377 void 378 vfs_unbusy(struct mount *mp, struct thread *td) 379 { 380 381 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td); 382 vfs_rel(mp); 383 } 384 385 /* 386 * Lookup a mount point by filesystem identifier. 387 */ 388 struct mount * 389 vfs_getvfs(fsid_t *fsid) 390 { 391 struct mount *mp; 392 393 mtx_lock(&mountlist_mtx); 394 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 395 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 396 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 397 vfs_ref(mp); 398 mtx_unlock(&mountlist_mtx); 399 return (mp); 400 } 401 } 402 mtx_unlock(&mountlist_mtx); 403 return ((struct mount *) 0); 404 } 405 406 /* 407 * Check if a user can access priveledged mount options. 408 */ 409 int 410 vfs_suser(struct mount *mp, struct thread *td) 411 { 412 int error; 413 414 if ((mp->mnt_flag & MNT_USER) == 0 || 415 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 416 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 417 return (error); 418 } 419 return (0); 420 } 421 422 /* 423 * Get a new unique fsid. Try to make its val[0] unique, since this value 424 * will be used to create fake device numbers for stat(). Also try (but 425 * not so hard) make its val[0] unique mod 2^16, since some emulators only 426 * support 16-bit device numbers. We end up with unique val[0]'s for the 427 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 428 * 429 * Keep in mind that several mounts may be running in parallel. Starting 430 * the search one past where the previous search terminated is both a 431 * micro-optimization and a defense against returning the same fsid to 432 * different mounts. 433 */ 434 void 435 vfs_getnewfsid(struct mount *mp) 436 { 437 static u_int16_t mntid_base; 438 struct mount *nmp; 439 fsid_t tfsid; 440 int mtype; 441 442 mtx_lock(&mntid_mtx); 443 mtype = mp->mnt_vfc->vfc_typenum; 444 tfsid.val[1] = mtype; 445 mtype = (mtype & 0xFF) << 24; 446 for (;;) { 447 tfsid.val[0] = makedev(255, 448 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 449 mntid_base++; 450 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 451 break; 452 vfs_rel(nmp); 453 } 454 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 455 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 456 mtx_unlock(&mntid_mtx); 457 } 458 459 /* 460 * Knob to control the precision of file timestamps: 461 * 462 * 0 = seconds only; nanoseconds zeroed. 463 * 1 = seconds and nanoseconds, accurate within 1/HZ. 464 * 2 = seconds and nanoseconds, truncated to microseconds. 465 * >=3 = seconds and nanoseconds, maximum precision. 466 */ 467 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 468 469 static int timestamp_precision = TSP_SEC; 470 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 471 ×tamp_precision, 0, ""); 472 473 /* 474 * Get a current timestamp. 475 */ 476 void 477 vfs_timestamp(struct timespec *tsp) 478 { 479 struct timeval tv; 480 481 switch (timestamp_precision) { 482 case TSP_SEC: 483 tsp->tv_sec = time_second; 484 tsp->tv_nsec = 0; 485 break; 486 case TSP_HZ: 487 getnanotime(tsp); 488 break; 489 case TSP_USEC: 490 microtime(&tv); 491 TIMEVAL_TO_TIMESPEC(&tv, tsp); 492 break; 493 case TSP_NSEC: 494 default: 495 nanotime(tsp); 496 break; 497 } 498 } 499 500 /* 501 * Set vnode attributes to VNOVAL 502 */ 503 void 504 vattr_null(struct vattr *vap) 505 { 506 507 vap->va_type = VNON; 508 vap->va_size = VNOVAL; 509 vap->va_bytes = VNOVAL; 510 vap->va_mode = VNOVAL; 511 vap->va_nlink = VNOVAL; 512 vap->va_uid = VNOVAL; 513 vap->va_gid = VNOVAL; 514 vap->va_fsid = VNOVAL; 515 vap->va_fileid = VNOVAL; 516 vap->va_blocksize = VNOVAL; 517 vap->va_rdev = VNOVAL; 518 vap->va_atime.tv_sec = VNOVAL; 519 vap->va_atime.tv_nsec = VNOVAL; 520 vap->va_mtime.tv_sec = VNOVAL; 521 vap->va_mtime.tv_nsec = VNOVAL; 522 vap->va_ctime.tv_sec = VNOVAL; 523 vap->va_ctime.tv_nsec = VNOVAL; 524 vap->va_birthtime.tv_sec = VNOVAL; 525 vap->va_birthtime.tv_nsec = VNOVAL; 526 vap->va_flags = VNOVAL; 527 vap->va_gen = VNOVAL; 528 vap->va_vaflags = 0; 529 } 530 531 /* 532 * This routine is called when we have too many vnodes. It attempts 533 * to free <count> vnodes and will potentially free vnodes that still 534 * have VM backing store (VM backing store is typically the cause 535 * of a vnode blowout so we want to do this). Therefore, this operation 536 * is not considered cheap. 537 * 538 * A number of conditions may prevent a vnode from being reclaimed. 539 * the buffer cache may have references on the vnode, a directory 540 * vnode may still have references due to the namei cache representing 541 * underlying files, or the vnode may be in active use. It is not 542 * desireable to reuse such vnodes. These conditions may cause the 543 * number of vnodes to reach some minimum value regardless of what 544 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 545 */ 546 static int 547 vlrureclaim(struct mount *mp) 548 { 549 struct thread *td; 550 struct vnode *vp; 551 int done; 552 int trigger; 553 int usevnodes; 554 int count; 555 556 /* 557 * Calculate the trigger point, don't allow user 558 * screwups to blow us up. This prevents us from 559 * recycling vnodes with lots of resident pages. We 560 * aren't trying to free memory, we are trying to 561 * free vnodes. 562 */ 563 usevnodes = desiredvnodes; 564 if (usevnodes <= 0) 565 usevnodes = 1; 566 trigger = cnt.v_page_count * 2 / usevnodes; 567 done = 0; 568 td = curthread; 569 vn_start_write(NULL, &mp, V_WAIT); 570 MNT_ILOCK(mp); 571 count = mp->mnt_nvnodelistsize / 10 + 1; 572 while (count != 0) { 573 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 574 while (vp != NULL && vp->v_type == VMARKER) 575 vp = TAILQ_NEXT(vp, v_nmntvnodes); 576 if (vp == NULL) 577 break; 578 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 579 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 580 --count; 581 if (!VI_TRYLOCK(vp)) 582 goto next_iter; 583 /* 584 * If it's been deconstructed already, it's still 585 * referenced, or it exceeds the trigger, skip it. 586 */ 587 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) || 588 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 589 vp->v_object->resident_page_count > trigger)) { 590 VI_UNLOCK(vp); 591 goto next_iter; 592 } 593 MNT_IUNLOCK(mp); 594 vholdl(vp); 595 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) { 596 vdrop(vp); 597 goto next_iter_mntunlocked; 598 } 599 VI_LOCK(vp); 600 /* 601 * v_usecount may have been bumped after VOP_LOCK() dropped 602 * the vnode interlock and before it was locked again. 603 * 604 * It is not necessary to recheck VI_DOOMED because it can 605 * only be set by another thread that holds both the vnode 606 * lock and vnode interlock. If another thread has the 607 * vnode lock before we get to VOP_LOCK() and obtains the 608 * vnode interlock after VOP_LOCK() drops the vnode 609 * interlock, the other thread will be unable to drop the 610 * vnode lock before our VOP_LOCK() call fails. 611 */ 612 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) || 613 (vp->v_object != NULL && 614 vp->v_object->resident_page_count > trigger)) { 615 VOP_UNLOCK(vp, LK_INTERLOCK, td); 616 goto next_iter_mntunlocked; 617 } 618 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 619 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 620 vgonel(vp); 621 VOP_UNLOCK(vp, 0, td); 622 vdropl(vp); 623 done++; 624 next_iter_mntunlocked: 625 if ((count % 256) != 0) 626 goto relock_mnt; 627 goto yield; 628 next_iter: 629 if ((count % 256) != 0) 630 continue; 631 MNT_IUNLOCK(mp); 632 yield: 633 uio_yield(); 634 relock_mnt: 635 MNT_ILOCK(mp); 636 } 637 MNT_IUNLOCK(mp); 638 vn_finished_write(mp); 639 return done; 640 } 641 642 /* 643 * Attempt to keep the free list at wantfreevnodes length. 644 */ 645 static void 646 vnlru_free(int count) 647 { 648 struct vnode *vp; 649 int vfslocked; 650 651 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 652 for (; count > 0; count--) { 653 vp = TAILQ_FIRST(&vnode_free_list); 654 /* 655 * The list can be modified while the free_list_mtx 656 * has been dropped and vp could be NULL here. 657 */ 658 if (!vp) 659 break; 660 VNASSERT(vp->v_op != NULL, vp, 661 ("vnlru_free: vnode already reclaimed.")); 662 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 663 /* 664 * Don't recycle if we can't get the interlock. 665 */ 666 if (!VI_TRYLOCK(vp)) { 667 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 668 continue; 669 } 670 VNASSERT(VCANRECYCLE(vp), vp, 671 ("vp inconsistent on freelist")); 672 freevnodes--; 673 vp->v_iflag &= ~VI_FREE; 674 vholdl(vp); 675 mtx_unlock(&vnode_free_list_mtx); 676 VI_UNLOCK(vp); 677 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 678 vtryrecycle(vp); 679 VFS_UNLOCK_GIANT(vfslocked); 680 /* 681 * If the recycled succeeded this vdrop will actually free 682 * the vnode. If not it will simply place it back on 683 * the free list. 684 */ 685 vdrop(vp); 686 mtx_lock(&vnode_free_list_mtx); 687 } 688 } 689 /* 690 * Attempt to recycle vnodes in a context that is always safe to block. 691 * Calling vlrurecycle() from the bowels of filesystem code has some 692 * interesting deadlock problems. 693 */ 694 static struct proc *vnlruproc; 695 static int vnlruproc_sig; 696 697 static void 698 vnlru_proc(void) 699 { 700 struct mount *mp, *nmp; 701 int done; 702 struct proc *p = vnlruproc; 703 struct thread *td = FIRST_THREAD_IN_PROC(p); 704 705 mtx_lock(&Giant); 706 707 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 708 SHUTDOWN_PRI_FIRST); 709 710 for (;;) { 711 kthread_suspend_check(p); 712 mtx_lock(&vnode_free_list_mtx); 713 if (freevnodes > wantfreevnodes) 714 vnlru_free(freevnodes - wantfreevnodes); 715 if (numvnodes <= desiredvnodes * 9 / 10) { 716 vnlruproc_sig = 0; 717 wakeup(&vnlruproc_sig); 718 msleep(vnlruproc, &vnode_free_list_mtx, 719 PVFS|PDROP, "vlruwt", hz); 720 continue; 721 } 722 mtx_unlock(&vnode_free_list_mtx); 723 done = 0; 724 mtx_lock(&mountlist_mtx); 725 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 726 int vfsunlocked; 727 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) { 728 nmp = TAILQ_NEXT(mp, mnt_list); 729 continue; 730 } 731 if (!VFS_NEEDSGIANT(mp)) { 732 mtx_unlock(&Giant); 733 vfsunlocked = 1; 734 } else 735 vfsunlocked = 0; 736 done += vlrureclaim(mp); 737 if (vfsunlocked) 738 mtx_lock(&Giant); 739 mtx_lock(&mountlist_mtx); 740 nmp = TAILQ_NEXT(mp, mnt_list); 741 vfs_unbusy(mp, td); 742 } 743 mtx_unlock(&mountlist_mtx); 744 if (done == 0) { 745 #if 0 746 /* These messages are temporary debugging aids */ 747 if (vnlru_nowhere < 5) 748 printf("vnlru process getting nowhere..\n"); 749 else if (vnlru_nowhere == 5) 750 printf("vnlru process messages stopped.\n"); 751 #endif 752 vnlru_nowhere++; 753 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 754 } else 755 uio_yield(); 756 } 757 } 758 759 static struct kproc_desc vnlru_kp = { 760 "vnlru", 761 vnlru_proc, 762 &vnlruproc 763 }; 764 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 765 766 /* 767 * Routines having to do with the management of the vnode table. 768 */ 769 770 static void 771 vdestroy(struct vnode *vp) 772 { 773 struct bufobj *bo; 774 775 CTR1(KTR_VFS, "vdestroy vp %p", vp); 776 mtx_lock(&vnode_free_list_mtx); 777 numvnodes--; 778 mtx_unlock(&vnode_free_list_mtx); 779 bo = &vp->v_bufobj; 780 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 781 ("cleaned vnode still on the free list.")); 782 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 783 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 784 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 785 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 786 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 787 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 788 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 789 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 790 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 791 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 792 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 793 VI_UNLOCK(vp); 794 #ifdef MAC 795 mac_destroy_vnode(vp); 796 #endif 797 if (vp->v_pollinfo != NULL) { 798 knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note); 799 mtx_destroy(&vp->v_pollinfo->vpi_lock); 800 uma_zfree(vnodepoll_zone, vp->v_pollinfo); 801 } 802 #ifdef INVARIANTS 803 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */ 804 vp->v_op = NULL; 805 #endif 806 lockdestroy(vp->v_vnlock); 807 mtx_destroy(&vp->v_interlock); 808 uma_zfree(vnode_zone, vp); 809 } 810 811 /* 812 * Try to recycle a freed vnode. We abort if anyone picks up a reference 813 * before we actually vgone(). This function must be called with the vnode 814 * held to prevent the vnode from being returned to the free list midway 815 * through vgone(). 816 */ 817 static int 818 vtryrecycle(struct vnode *vp) 819 { 820 struct thread *td = curthread; 821 struct mount *vnmp; 822 823 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp); 824 VNASSERT(vp->v_holdcnt, vp, 825 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 826 /* 827 * This vnode may found and locked via some other list, if so we 828 * can't recycle it yet. 829 */ 830 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0) 831 return (EWOULDBLOCK); 832 /* 833 * Don't recycle if its filesystem is being suspended. 834 */ 835 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 836 VOP_UNLOCK(vp, 0, td); 837 return (EBUSY); 838 } 839 /* 840 * If we got this far, we need to acquire the interlock and see if 841 * anyone picked up this vnode from another list. If not, we will 842 * mark it with DOOMED via vgonel() so that anyone who does find it 843 * will skip over it. 844 */ 845 VI_LOCK(vp); 846 if (vp->v_usecount) { 847 VOP_UNLOCK(vp, LK_INTERLOCK, td); 848 vn_finished_write(vnmp); 849 return (EBUSY); 850 } 851 if ((vp->v_iflag & VI_DOOMED) == 0) 852 vgonel(vp); 853 VOP_UNLOCK(vp, LK_INTERLOCK, td); 854 vn_finished_write(vnmp); 855 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp); 856 return (0); 857 } 858 859 /* 860 * Return the next vnode from the free list. 861 */ 862 int 863 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 864 struct vnode **vpp) 865 { 866 struct vnode *vp = NULL; 867 struct bufobj *bo; 868 869 mtx_lock(&vnode_free_list_mtx); 870 /* 871 * Lend our context to reclaim vnodes if they've exceeded the max. 872 */ 873 if (freevnodes > wantfreevnodes) 874 vnlru_free(1); 875 /* 876 * Wait for available vnodes. 877 */ 878 if (numvnodes > desiredvnodes) { 879 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 880 /* 881 * File system is beeing suspended, we cannot risk a 882 * deadlock here, so allocate new vnode anyway. 883 */ 884 if (freevnodes > wantfreevnodes) 885 vnlru_free(freevnodes - wantfreevnodes); 886 goto alloc; 887 } 888 if (vnlruproc_sig == 0) { 889 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 890 wakeup(vnlruproc); 891 } 892 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 893 "vlruwk", hz); 894 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 895 if (numvnodes > desiredvnodes) { 896 mtx_unlock(&vnode_free_list_mtx); 897 return (ENFILE); 898 } 899 #endif 900 } 901 alloc: 902 numvnodes++; 903 mtx_unlock(&vnode_free_list_mtx); 904 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 905 /* 906 * Setup locks. 907 */ 908 vp->v_vnlock = &vp->v_lock; 909 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 910 /* 911 * By default, don't allow shared locks unless filesystems 912 * opt-in. 913 */ 914 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 915 /* 916 * Initialize bufobj. 917 */ 918 bo = &vp->v_bufobj; 919 bo->__bo_vnode = vp; 920 bo->bo_mtx = &vp->v_interlock; 921 bo->bo_ops = &buf_ops_bio; 922 bo->bo_private = vp; 923 TAILQ_INIT(&bo->bo_clean.bv_hd); 924 TAILQ_INIT(&bo->bo_dirty.bv_hd); 925 /* 926 * Initialize namecache. 927 */ 928 LIST_INIT(&vp->v_cache_src); 929 TAILQ_INIT(&vp->v_cache_dst); 930 /* 931 * Finalize various vnode identity bits. 932 */ 933 vp->v_type = VNON; 934 vp->v_tag = tag; 935 vp->v_op = vops; 936 v_incr_usecount(vp); 937 vp->v_data = 0; 938 #ifdef MAC 939 mac_init_vnode(vp); 940 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 941 mac_associate_vnode_singlelabel(mp, vp); 942 else if (mp == NULL) 943 printf("NULL mp in getnewvnode()\n"); 944 #endif 945 if (mp != NULL) { 946 insmntque(vp, mp); 947 bo->bo_bsize = mp->mnt_stat.f_iosize; 948 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 949 vp->v_vflag |= VV_NOKNOTE; 950 } 951 952 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp); 953 *vpp = vp; 954 return (0); 955 } 956 957 /* 958 * Delete from old mount point vnode list, if on one. 959 */ 960 static void 961 delmntque(struct vnode *vp) 962 { 963 struct mount *mp; 964 965 mp = vp->v_mount; 966 if (mp == NULL) 967 return; 968 MNT_ILOCK(mp); 969 vp->v_mount = NULL; 970 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 971 ("bad mount point vnode list size")); 972 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 973 mp->mnt_nvnodelistsize--; 974 MNT_REL(mp); 975 MNT_IUNLOCK(mp); 976 } 977 978 /* 979 * Insert into list of vnodes for the new mount point, if available. 980 */ 981 static void 982 insmntque(struct vnode *vp, struct mount *mp) 983 { 984 985 vp->v_mount = mp; 986 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 987 MNT_ILOCK(mp); 988 MNT_REF(mp); 989 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 990 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 991 ("neg mount point vnode list size")); 992 mp->mnt_nvnodelistsize++; 993 MNT_IUNLOCK(mp); 994 } 995 996 /* 997 * Flush out and invalidate all buffers associated with a bufobj 998 * Called with the underlying object locked. 999 */ 1000 int 1001 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag, 1002 int slptimeo) 1003 { 1004 int error; 1005 1006 BO_LOCK(bo); 1007 if (flags & V_SAVE) { 1008 error = bufobj_wwait(bo, slpflag, slptimeo); 1009 if (error) { 1010 BO_UNLOCK(bo); 1011 return (error); 1012 } 1013 if (bo->bo_dirty.bv_cnt > 0) { 1014 BO_UNLOCK(bo); 1015 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0) 1016 return (error); 1017 /* 1018 * XXX We could save a lock/unlock if this was only 1019 * enabled under INVARIANTS 1020 */ 1021 BO_LOCK(bo); 1022 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1023 panic("vinvalbuf: dirty bufs"); 1024 } 1025 } 1026 /* 1027 * If you alter this loop please notice that interlock is dropped and 1028 * reacquired in flushbuflist. Special care is needed to ensure that 1029 * no race conditions occur from this. 1030 */ 1031 do { 1032 error = flushbuflist(&bo->bo_clean, 1033 flags, bo, slpflag, slptimeo); 1034 if (error == 0) 1035 error = flushbuflist(&bo->bo_dirty, 1036 flags, bo, slpflag, slptimeo); 1037 if (error != 0 && error != EAGAIN) { 1038 BO_UNLOCK(bo); 1039 return (error); 1040 } 1041 } while (error != 0); 1042 1043 /* 1044 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1045 * have write I/O in-progress but if there is a VM object then the 1046 * VM object can also have read-I/O in-progress. 1047 */ 1048 do { 1049 bufobj_wwait(bo, 0, 0); 1050 BO_UNLOCK(bo); 1051 if (bo->bo_object != NULL) { 1052 VM_OBJECT_LOCK(bo->bo_object); 1053 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1054 VM_OBJECT_UNLOCK(bo->bo_object); 1055 } 1056 BO_LOCK(bo); 1057 } while (bo->bo_numoutput > 0); 1058 BO_UNLOCK(bo); 1059 1060 /* 1061 * Destroy the copy in the VM cache, too. 1062 */ 1063 if (bo->bo_object != NULL) { 1064 VM_OBJECT_LOCK(bo->bo_object); 1065 vm_object_page_remove(bo->bo_object, 0, 0, 1066 (flags & V_SAVE) ? TRUE : FALSE); 1067 VM_OBJECT_UNLOCK(bo->bo_object); 1068 } 1069 1070 #ifdef INVARIANTS 1071 BO_LOCK(bo); 1072 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1073 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1074 panic("vinvalbuf: flush failed"); 1075 BO_UNLOCK(bo); 1076 #endif 1077 return (0); 1078 } 1079 1080 /* 1081 * Flush out and invalidate all buffers associated with a vnode. 1082 * Called with the underlying object locked. 1083 */ 1084 int 1085 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag, 1086 int slptimeo) 1087 { 1088 1089 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags); 1090 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1091 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo)); 1092 } 1093 1094 /* 1095 * Flush out buffers on the specified list. 1096 * 1097 */ 1098 static int 1099 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1100 int slptimeo) 1101 { 1102 struct buf *bp, *nbp; 1103 int retval, error; 1104 daddr_t lblkno; 1105 b_xflags_t xflags; 1106 1107 ASSERT_BO_LOCKED(bo); 1108 1109 retval = 0; 1110 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1111 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1112 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1113 continue; 1114 } 1115 lblkno = 0; 1116 xflags = 0; 1117 if (nbp != NULL) { 1118 lblkno = nbp->b_lblkno; 1119 xflags = nbp->b_xflags & 1120 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1121 } 1122 retval = EAGAIN; 1123 error = BUF_TIMELOCK(bp, 1124 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1125 "flushbuf", slpflag, slptimeo); 1126 if (error) { 1127 BO_LOCK(bo); 1128 return (error != ENOLCK ? error : EAGAIN); 1129 } 1130 KASSERT(bp->b_bufobj == bo, 1131 ("bp %p wrong b_bufobj %p should be %p", 1132 bp, bp->b_bufobj, bo)); 1133 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1134 BUF_UNLOCK(bp); 1135 BO_LOCK(bo); 1136 return (EAGAIN); 1137 } 1138 /* 1139 * XXX Since there are no node locks for NFS, I 1140 * believe there is a slight chance that a delayed 1141 * write will occur while sleeping just above, so 1142 * check for it. 1143 */ 1144 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1145 (flags & V_SAVE)) { 1146 bremfree(bp); 1147 bp->b_flags |= B_ASYNC; 1148 bwrite(bp); 1149 BO_LOCK(bo); 1150 return (EAGAIN); /* XXX: why not loop ? */ 1151 } 1152 bremfree(bp); 1153 bp->b_flags |= (B_INVAL | B_RELBUF); 1154 bp->b_flags &= ~B_ASYNC; 1155 brelse(bp); 1156 BO_LOCK(bo); 1157 if (nbp != NULL && 1158 (nbp->b_bufobj != bo || 1159 nbp->b_lblkno != lblkno || 1160 (nbp->b_xflags & 1161 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1162 break; /* nbp invalid */ 1163 } 1164 return (retval); 1165 } 1166 1167 /* 1168 * Truncate a file's buffer and pages to a specified length. This 1169 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1170 * sync activity. 1171 */ 1172 int 1173 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, 1174 off_t length, int blksize) 1175 { 1176 struct buf *bp, *nbp; 1177 int anyfreed; 1178 int trunclbn; 1179 struct bufobj *bo; 1180 1181 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length); 1182 /* 1183 * Round up to the *next* lbn. 1184 */ 1185 trunclbn = (length + blksize - 1) / blksize; 1186 1187 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1188 restart: 1189 VI_LOCK(vp); 1190 bo = &vp->v_bufobj; 1191 anyfreed = 1; 1192 for (;anyfreed;) { 1193 anyfreed = 0; 1194 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1195 if (bp->b_lblkno < trunclbn) 1196 continue; 1197 if (BUF_LOCK(bp, 1198 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1199 VI_MTX(vp)) == ENOLCK) 1200 goto restart; 1201 1202 bremfree(bp); 1203 bp->b_flags |= (B_INVAL | B_RELBUF); 1204 bp->b_flags &= ~B_ASYNC; 1205 brelse(bp); 1206 anyfreed = 1; 1207 1208 if (nbp != NULL && 1209 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1210 (nbp->b_vp != vp) || 1211 (nbp->b_flags & B_DELWRI))) { 1212 goto restart; 1213 } 1214 VI_LOCK(vp); 1215 } 1216 1217 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1218 if (bp->b_lblkno < trunclbn) 1219 continue; 1220 if (BUF_LOCK(bp, 1221 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1222 VI_MTX(vp)) == ENOLCK) 1223 goto restart; 1224 bremfree(bp); 1225 bp->b_flags |= (B_INVAL | B_RELBUF); 1226 bp->b_flags &= ~B_ASYNC; 1227 brelse(bp); 1228 anyfreed = 1; 1229 if (nbp != NULL && 1230 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1231 (nbp->b_vp != vp) || 1232 (nbp->b_flags & B_DELWRI) == 0)) { 1233 goto restart; 1234 } 1235 VI_LOCK(vp); 1236 } 1237 } 1238 1239 if (length > 0) { 1240 restartsync: 1241 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1242 if (bp->b_lblkno > 0) 1243 continue; 1244 /* 1245 * Since we hold the vnode lock this should only 1246 * fail if we're racing with the buf daemon. 1247 */ 1248 if (BUF_LOCK(bp, 1249 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1250 VI_MTX(vp)) == ENOLCK) { 1251 goto restart; 1252 } 1253 VNASSERT((bp->b_flags & B_DELWRI), vp, 1254 ("buf(%p) on dirty queue without DELWRI", bp)); 1255 1256 bremfree(bp); 1257 bawrite(bp); 1258 VI_LOCK(vp); 1259 goto restartsync; 1260 } 1261 } 1262 1263 bufobj_wwait(bo, 0, 0); 1264 VI_UNLOCK(vp); 1265 vnode_pager_setsize(vp, length); 1266 1267 return (0); 1268 } 1269 1270 /* 1271 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1272 * a vnode. 1273 * 1274 * NOTE: We have to deal with the special case of a background bitmap 1275 * buffer, a situation where two buffers will have the same logical 1276 * block offset. We want (1) only the foreground buffer to be accessed 1277 * in a lookup and (2) must differentiate between the foreground and 1278 * background buffer in the splay tree algorithm because the splay 1279 * tree cannot normally handle multiple entities with the same 'index'. 1280 * We accomplish this by adding differentiating flags to the splay tree's 1281 * numerical domain. 1282 */ 1283 static 1284 struct buf * 1285 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1286 { 1287 struct buf dummy; 1288 struct buf *lefttreemax, *righttreemin, *y; 1289 1290 if (root == NULL) 1291 return (NULL); 1292 lefttreemax = righttreemin = &dummy; 1293 for (;;) { 1294 if (lblkno < root->b_lblkno || 1295 (lblkno == root->b_lblkno && 1296 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1297 if ((y = root->b_left) == NULL) 1298 break; 1299 if (lblkno < y->b_lblkno) { 1300 /* Rotate right. */ 1301 root->b_left = y->b_right; 1302 y->b_right = root; 1303 root = y; 1304 if ((y = root->b_left) == NULL) 1305 break; 1306 } 1307 /* Link into the new root's right tree. */ 1308 righttreemin->b_left = root; 1309 righttreemin = root; 1310 } else if (lblkno > root->b_lblkno || 1311 (lblkno == root->b_lblkno && 1312 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1313 if ((y = root->b_right) == NULL) 1314 break; 1315 if (lblkno > y->b_lblkno) { 1316 /* Rotate left. */ 1317 root->b_right = y->b_left; 1318 y->b_left = root; 1319 root = y; 1320 if ((y = root->b_right) == NULL) 1321 break; 1322 } 1323 /* Link into the new root's left tree. */ 1324 lefttreemax->b_right = root; 1325 lefttreemax = root; 1326 } else { 1327 break; 1328 } 1329 root = y; 1330 } 1331 /* Assemble the new root. */ 1332 lefttreemax->b_right = root->b_left; 1333 righttreemin->b_left = root->b_right; 1334 root->b_left = dummy.b_right; 1335 root->b_right = dummy.b_left; 1336 return (root); 1337 } 1338 1339 static void 1340 buf_vlist_remove(struct buf *bp) 1341 { 1342 struct buf *root; 1343 struct bufv *bv; 1344 1345 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1346 ASSERT_BO_LOCKED(bp->b_bufobj); 1347 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1348 (BX_VNDIRTY|BX_VNCLEAN), 1349 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1350 if (bp->b_xflags & BX_VNDIRTY) 1351 bv = &bp->b_bufobj->bo_dirty; 1352 else 1353 bv = &bp->b_bufobj->bo_clean; 1354 if (bp != bv->bv_root) { 1355 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1356 KASSERT(root == bp, ("splay lookup failed in remove")); 1357 } 1358 if (bp->b_left == NULL) { 1359 root = bp->b_right; 1360 } else { 1361 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1362 root->b_right = bp->b_right; 1363 } 1364 bv->bv_root = root; 1365 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1366 bv->bv_cnt--; 1367 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1368 } 1369 1370 /* 1371 * Add the buffer to the sorted clean or dirty block list using a 1372 * splay tree algorithm. 1373 * 1374 * NOTE: xflags is passed as a constant, optimizing this inline function! 1375 */ 1376 static void 1377 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1378 { 1379 struct buf *root; 1380 struct bufv *bv; 1381 1382 ASSERT_BO_LOCKED(bo); 1383 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1384 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1385 bp->b_xflags |= xflags; 1386 if (xflags & BX_VNDIRTY) 1387 bv = &bo->bo_dirty; 1388 else 1389 bv = &bo->bo_clean; 1390 1391 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1392 if (root == NULL) { 1393 bp->b_left = NULL; 1394 bp->b_right = NULL; 1395 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1396 } else if (bp->b_lblkno < root->b_lblkno || 1397 (bp->b_lblkno == root->b_lblkno && 1398 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1399 bp->b_left = root->b_left; 1400 bp->b_right = root; 1401 root->b_left = NULL; 1402 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1403 } else { 1404 bp->b_right = root->b_right; 1405 bp->b_left = root; 1406 root->b_right = NULL; 1407 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1408 } 1409 bv->bv_cnt++; 1410 bv->bv_root = bp; 1411 } 1412 1413 /* 1414 * Lookup a buffer using the splay tree. Note that we specifically avoid 1415 * shadow buffers used in background bitmap writes. 1416 * 1417 * This code isn't quite efficient as it could be because we are maintaining 1418 * two sorted lists and do not know which list the block resides in. 1419 * 1420 * During a "make buildworld" the desired buffer is found at one of 1421 * the roots more than 60% of the time. Thus, checking both roots 1422 * before performing either splay eliminates unnecessary splays on the 1423 * first tree splayed. 1424 */ 1425 struct buf * 1426 gbincore(struct bufobj *bo, daddr_t lblkno) 1427 { 1428 struct buf *bp; 1429 1430 ASSERT_BO_LOCKED(bo); 1431 if ((bp = bo->bo_clean.bv_root) != NULL && 1432 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1433 return (bp); 1434 if ((bp = bo->bo_dirty.bv_root) != NULL && 1435 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1436 return (bp); 1437 if ((bp = bo->bo_clean.bv_root) != NULL) { 1438 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1439 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1440 return (bp); 1441 } 1442 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1443 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1444 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1445 return (bp); 1446 } 1447 return (NULL); 1448 } 1449 1450 /* 1451 * Associate a buffer with a vnode. 1452 */ 1453 void 1454 bgetvp(struct vnode *vp, struct buf *bp) 1455 { 1456 1457 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1458 1459 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1460 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1461 ("bgetvp: bp already attached! %p", bp)); 1462 1463 ASSERT_VI_LOCKED(vp, "bgetvp"); 1464 vholdl(vp); 1465 if (VFS_NEEDSGIANT(vp->v_mount) || 1466 vp->v_bufobj.bo_flag & BO_NEEDSGIANT) 1467 bp->b_flags |= B_NEEDSGIANT; 1468 bp->b_vp = vp; 1469 bp->b_bufobj = &vp->v_bufobj; 1470 /* 1471 * Insert onto list for new vnode. 1472 */ 1473 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN); 1474 } 1475 1476 /* 1477 * Disassociate a buffer from a vnode. 1478 */ 1479 void 1480 brelvp(struct buf *bp) 1481 { 1482 struct bufobj *bo; 1483 struct vnode *vp; 1484 1485 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1486 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1487 1488 /* 1489 * Delete from old vnode list, if on one. 1490 */ 1491 vp = bp->b_vp; /* XXX */ 1492 bo = bp->b_bufobj; 1493 BO_LOCK(bo); 1494 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1495 buf_vlist_remove(bp); 1496 else 1497 panic("brelvp: Buffer %p not on queue.", bp); 1498 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1499 bo->bo_flag &= ~BO_ONWORKLST; 1500 mtx_lock(&sync_mtx); 1501 LIST_REMOVE(bo, bo_synclist); 1502 syncer_worklist_len--; 1503 mtx_unlock(&sync_mtx); 1504 } 1505 bp->b_flags &= ~B_NEEDSGIANT; 1506 bp->b_vp = NULL; 1507 bp->b_bufobj = NULL; 1508 vdropl(vp); 1509 } 1510 1511 /* 1512 * Add an item to the syncer work queue. 1513 */ 1514 static void 1515 vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1516 { 1517 int slot; 1518 1519 ASSERT_BO_LOCKED(bo); 1520 1521 mtx_lock(&sync_mtx); 1522 if (bo->bo_flag & BO_ONWORKLST) 1523 LIST_REMOVE(bo, bo_synclist); 1524 else { 1525 bo->bo_flag |= BO_ONWORKLST; 1526 syncer_worklist_len++; 1527 } 1528 1529 if (delay > syncer_maxdelay - 2) 1530 delay = syncer_maxdelay - 2; 1531 slot = (syncer_delayno + delay) & syncer_mask; 1532 1533 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); 1534 mtx_unlock(&sync_mtx); 1535 } 1536 1537 static int 1538 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1539 { 1540 int error, len; 1541 1542 mtx_lock(&sync_mtx); 1543 len = syncer_worklist_len - sync_vnode_count; 1544 mtx_unlock(&sync_mtx); 1545 error = SYSCTL_OUT(req, &len, sizeof(len)); 1546 return (error); 1547 } 1548 1549 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1550 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1551 1552 static struct proc *updateproc; 1553 static void sched_sync(void); 1554 static struct kproc_desc up_kp = { 1555 "syncer", 1556 sched_sync, 1557 &updateproc 1558 }; 1559 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 1560 1561 static int 1562 sync_vnode(struct bufobj *bo, struct thread *td) 1563 { 1564 struct vnode *vp; 1565 struct mount *mp; 1566 1567 vp = bo->__bo_vnode; /* XXX */ 1568 if (VOP_ISLOCKED(vp, NULL) != 0) 1569 return (1); 1570 if (VI_TRYLOCK(vp) == 0) 1571 return (1); 1572 /* 1573 * We use vhold in case the vnode does not 1574 * successfully sync. vhold prevents the vnode from 1575 * going away when we unlock the sync_mtx so that 1576 * we can acquire the vnode interlock. 1577 */ 1578 vholdl(vp); 1579 mtx_unlock(&sync_mtx); 1580 VI_UNLOCK(vp); 1581 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1582 vdrop(vp); 1583 mtx_lock(&sync_mtx); 1584 return (1); 1585 } 1586 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 1587 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1588 VOP_UNLOCK(vp, 0, td); 1589 vn_finished_write(mp); 1590 VI_LOCK(vp); 1591 if ((bo->bo_flag & BO_ONWORKLST) != 0) { 1592 /* 1593 * Put us back on the worklist. The worklist 1594 * routine will remove us from our current 1595 * position and then add us back in at a later 1596 * position. 1597 */ 1598 vn_syncer_add_to_worklist(bo, syncdelay); 1599 } 1600 vdropl(vp); 1601 mtx_lock(&sync_mtx); 1602 return (0); 1603 } 1604 1605 /* 1606 * System filesystem synchronizer daemon. 1607 */ 1608 static void 1609 sched_sync(void) 1610 { 1611 struct synclist *next; 1612 struct synclist *slp; 1613 struct bufobj *bo; 1614 long starttime; 1615 struct thread *td = FIRST_THREAD_IN_PROC(updateproc); 1616 static int dummychan; 1617 int last_work_seen; 1618 int net_worklist_len; 1619 int syncer_final_iter; 1620 int first_printf; 1621 int error; 1622 1623 mtx_lock(&Giant); 1624 last_work_seen = 0; 1625 syncer_final_iter = 0; 1626 first_printf = 1; 1627 syncer_state = SYNCER_RUNNING; 1628 starttime = time_uptime; 1629 td->td_pflags |= TDP_NORUNNINGBUF; 1630 1631 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1632 SHUTDOWN_PRI_LAST); 1633 1634 mtx_lock(&sync_mtx); 1635 for (;;) { 1636 if (syncer_state == SYNCER_FINAL_DELAY && 1637 syncer_final_iter == 0) { 1638 mtx_unlock(&sync_mtx); 1639 kthread_suspend_check(td->td_proc); 1640 mtx_lock(&sync_mtx); 1641 } 1642 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1643 if (syncer_state != SYNCER_RUNNING && 1644 starttime != time_uptime) { 1645 if (first_printf) { 1646 printf("\nSyncing disks, vnodes remaining..."); 1647 first_printf = 0; 1648 } 1649 printf("%d ", net_worklist_len); 1650 } 1651 starttime = time_uptime; 1652 1653 /* 1654 * Push files whose dirty time has expired. Be careful 1655 * of interrupt race on slp queue. 1656 * 1657 * Skip over empty worklist slots when shutting down. 1658 */ 1659 do { 1660 slp = &syncer_workitem_pending[syncer_delayno]; 1661 syncer_delayno += 1; 1662 if (syncer_delayno == syncer_maxdelay) 1663 syncer_delayno = 0; 1664 next = &syncer_workitem_pending[syncer_delayno]; 1665 /* 1666 * If the worklist has wrapped since the 1667 * it was emptied of all but syncer vnodes, 1668 * switch to the FINAL_DELAY state and run 1669 * for one more second. 1670 */ 1671 if (syncer_state == SYNCER_SHUTTING_DOWN && 1672 net_worklist_len == 0 && 1673 last_work_seen == syncer_delayno) { 1674 syncer_state = SYNCER_FINAL_DELAY; 1675 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1676 } 1677 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1678 syncer_worklist_len > 0); 1679 1680 /* 1681 * Keep track of the last time there was anything 1682 * on the worklist other than syncer vnodes. 1683 * Return to the SHUTTING_DOWN state if any 1684 * new work appears. 1685 */ 1686 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1687 last_work_seen = syncer_delayno; 1688 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1689 syncer_state = SYNCER_SHUTTING_DOWN; 1690 while ((bo = LIST_FIRST(slp)) != NULL) { 1691 error = sync_vnode(bo, td); 1692 if (error == 1) { 1693 LIST_REMOVE(bo, bo_synclist); 1694 LIST_INSERT_HEAD(next, bo, bo_synclist); 1695 continue; 1696 } 1697 } 1698 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1699 syncer_final_iter--; 1700 /* 1701 * The variable rushjob allows the kernel to speed up the 1702 * processing of the filesystem syncer process. A rushjob 1703 * value of N tells the filesystem syncer to process the next 1704 * N seconds worth of work on its queue ASAP. Currently rushjob 1705 * is used by the soft update code to speed up the filesystem 1706 * syncer process when the incore state is getting so far 1707 * ahead of the disk that the kernel memory pool is being 1708 * threatened with exhaustion. 1709 */ 1710 if (rushjob > 0) { 1711 rushjob -= 1; 1712 continue; 1713 } 1714 /* 1715 * Just sleep for a short period of time between 1716 * iterations when shutting down to allow some I/O 1717 * to happen. 1718 * 1719 * If it has taken us less than a second to process the 1720 * current work, then wait. Otherwise start right over 1721 * again. We can still lose time if any single round 1722 * takes more than two seconds, but it does not really 1723 * matter as we are just trying to generally pace the 1724 * filesystem activity. 1725 */ 1726 if (syncer_state != SYNCER_RUNNING) 1727 msleep(&dummychan, &sync_mtx, PPAUSE, "syncfnl", 1728 hz / SYNCER_SHUTDOWN_SPEEDUP); 1729 else if (time_uptime == starttime) 1730 msleep(&lbolt, &sync_mtx, PPAUSE, "syncer", 0); 1731 } 1732 } 1733 1734 /* 1735 * Request the syncer daemon to speed up its work. 1736 * We never push it to speed up more than half of its 1737 * normal turn time, otherwise it could take over the cpu. 1738 */ 1739 int 1740 speedup_syncer(void) 1741 { 1742 struct thread *td; 1743 int ret = 0; 1744 1745 td = FIRST_THREAD_IN_PROC(updateproc); 1746 sleepq_remove(td, &lbolt); 1747 mtx_lock(&sync_mtx); 1748 if (rushjob < syncdelay / 2) { 1749 rushjob += 1; 1750 stat_rush_requests += 1; 1751 ret = 1; 1752 } 1753 mtx_unlock(&sync_mtx); 1754 return (ret); 1755 } 1756 1757 /* 1758 * Tell the syncer to speed up its work and run though its work 1759 * list several times, then tell it to shut down. 1760 */ 1761 static void 1762 syncer_shutdown(void *arg, int howto) 1763 { 1764 struct thread *td; 1765 1766 if (howto & RB_NOSYNC) 1767 return; 1768 td = FIRST_THREAD_IN_PROC(updateproc); 1769 sleepq_remove(td, &lbolt); 1770 mtx_lock(&sync_mtx); 1771 syncer_state = SYNCER_SHUTTING_DOWN; 1772 rushjob = 0; 1773 mtx_unlock(&sync_mtx); 1774 kproc_shutdown(arg, howto); 1775 } 1776 1777 /* 1778 * Reassign a buffer from one vnode to another. 1779 * Used to assign file specific control information 1780 * (indirect blocks) to the vnode to which they belong. 1781 */ 1782 void 1783 reassignbuf(struct buf *bp) 1784 { 1785 struct vnode *vp; 1786 struct bufobj *bo; 1787 int delay; 1788 #ifdef INVARIANTS 1789 struct bufv *bv; 1790 #endif 1791 1792 vp = bp->b_vp; 1793 bo = bp->b_bufobj; 1794 ++reassignbufcalls; 1795 1796 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1797 bp, bp->b_vp, bp->b_flags); 1798 /* 1799 * B_PAGING flagged buffers cannot be reassigned because their vp 1800 * is not fully linked in. 1801 */ 1802 if (bp->b_flags & B_PAGING) 1803 panic("cannot reassign paging buffer"); 1804 1805 /* 1806 * Delete from old vnode list, if on one. 1807 */ 1808 VI_LOCK(vp); 1809 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1810 buf_vlist_remove(bp); 1811 else 1812 panic("reassignbuf: Buffer %p not on queue.", bp); 1813 /* 1814 * If dirty, put on list of dirty buffers; otherwise insert onto list 1815 * of clean buffers. 1816 */ 1817 if (bp->b_flags & B_DELWRI) { 1818 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 1819 switch (vp->v_type) { 1820 case VDIR: 1821 delay = dirdelay; 1822 break; 1823 case VCHR: 1824 delay = metadelay; 1825 break; 1826 default: 1827 delay = filedelay; 1828 } 1829 vn_syncer_add_to_worklist(bo, delay); 1830 } 1831 buf_vlist_add(bp, bo, BX_VNDIRTY); 1832 } else { 1833 buf_vlist_add(bp, bo, BX_VNCLEAN); 1834 1835 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1836 mtx_lock(&sync_mtx); 1837 LIST_REMOVE(bo, bo_synclist); 1838 syncer_worklist_len--; 1839 mtx_unlock(&sync_mtx); 1840 bo->bo_flag &= ~BO_ONWORKLST; 1841 } 1842 } 1843 #ifdef INVARIANTS 1844 bv = &bo->bo_clean; 1845 bp = TAILQ_FIRST(&bv->bv_hd); 1846 KASSERT(bp == NULL || bp->b_bufobj == bo, 1847 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1848 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1849 KASSERT(bp == NULL || bp->b_bufobj == bo, 1850 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1851 bv = &bo->bo_dirty; 1852 bp = TAILQ_FIRST(&bv->bv_hd); 1853 KASSERT(bp == NULL || bp->b_bufobj == bo, 1854 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1855 bp = TAILQ_LAST(&bv->bv_hd, buflists); 1856 KASSERT(bp == NULL || bp->b_bufobj == bo, 1857 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 1858 #endif 1859 VI_UNLOCK(vp); 1860 } 1861 1862 /* 1863 * Increment the use and hold counts on the vnode, taking care to reference 1864 * the driver's usecount if this is a chardev. The vholdl() will remove 1865 * the vnode from the free list if it is presently free. Requires the 1866 * vnode interlock and returns with it held. 1867 */ 1868 static void 1869 v_incr_usecount(struct vnode *vp) 1870 { 1871 1872 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n", 1873 vp, vp->v_holdcnt, vp->v_usecount); 1874 vp->v_usecount++; 1875 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1876 dev_lock(); 1877 vp->v_rdev->si_usecount++; 1878 dev_unlock(); 1879 } 1880 vholdl(vp); 1881 } 1882 1883 /* 1884 * Turn a holdcnt into a use+holdcnt such that only one call to 1885 * v_decr_usecount is needed. 1886 */ 1887 static void 1888 v_upgrade_usecount(struct vnode *vp) 1889 { 1890 1891 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n", 1892 vp, vp->v_holdcnt, vp->v_usecount); 1893 vp->v_usecount++; 1894 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1895 dev_lock(); 1896 vp->v_rdev->si_usecount++; 1897 dev_unlock(); 1898 } 1899 } 1900 1901 /* 1902 * Decrement the vnode use and hold count along with the driver's usecount 1903 * if this is a chardev. The vdropl() below releases the vnode interlock 1904 * as it may free the vnode. 1905 */ 1906 static void 1907 v_decr_usecount(struct vnode *vp) 1908 { 1909 1910 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n", 1911 vp, vp->v_holdcnt, vp->v_usecount); 1912 ASSERT_VI_LOCKED(vp, __FUNCTION__); 1913 VNASSERT(vp->v_usecount > 0, vp, 1914 ("v_decr_usecount: negative usecount")); 1915 vp->v_usecount--; 1916 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1917 dev_lock(); 1918 vp->v_rdev->si_usecount--; 1919 dev_unlock(); 1920 } 1921 vdropl(vp); 1922 } 1923 1924 /* 1925 * Decrement only the use count and driver use count. This is intended to 1926 * be paired with a follow on vdropl() to release the remaining hold count. 1927 * In this way we may vgone() a vnode with a 0 usecount without risk of 1928 * having it end up on a free list because the hold count is kept above 0. 1929 */ 1930 static void 1931 v_decr_useonly(struct vnode *vp) 1932 { 1933 1934 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n", 1935 vp, vp->v_holdcnt, vp->v_usecount); 1936 ASSERT_VI_LOCKED(vp, __FUNCTION__); 1937 VNASSERT(vp->v_usecount > 0, vp, 1938 ("v_decr_useonly: negative usecount")); 1939 vp->v_usecount--; 1940 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 1941 dev_lock(); 1942 vp->v_rdev->si_usecount--; 1943 dev_unlock(); 1944 } 1945 } 1946 1947 /* 1948 * Grab a particular vnode from the free list, increment its 1949 * reference count and lock it. The vnode lock bit is set if the 1950 * vnode is being eliminated in vgone. The process is awakened 1951 * when the transition is completed, and an error returned to 1952 * indicate that the vnode is no longer usable (possibly having 1953 * been changed to a new filesystem type). 1954 */ 1955 int 1956 vget(struct vnode *vp, int flags, struct thread *td) 1957 { 1958 int oweinact; 1959 int oldflags; 1960 int error; 1961 1962 error = 0; 1963 oldflags = flags; 1964 oweinact = 0; 1965 VFS_ASSERT_GIANT(vp->v_mount); 1966 if ((flags & LK_INTERLOCK) == 0) 1967 VI_LOCK(vp); 1968 /* 1969 * If the inactive call was deferred because vput() was called 1970 * with a shared lock, we have to do it here before another thread 1971 * gets a reference to data that should be dead. 1972 */ 1973 if (vp->v_iflag & VI_OWEINACT) { 1974 if (flags & LK_NOWAIT) { 1975 VI_UNLOCK(vp); 1976 return (EBUSY); 1977 } 1978 flags &= ~LK_TYPE_MASK; 1979 flags |= LK_EXCLUSIVE; 1980 oweinact = 1; 1981 } 1982 vholdl(vp); 1983 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) { 1984 vdrop(vp); 1985 return (error); 1986 } 1987 VI_LOCK(vp); 1988 /* Upgrade our holdcnt to a usecount. */ 1989 v_upgrade_usecount(vp); 1990 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 1991 panic("vget: vn_lock failed to return ENOENT\n"); 1992 if (oweinact) { 1993 if (vp->v_iflag & VI_OWEINACT) 1994 vinactive(vp, td); 1995 VI_UNLOCK(vp); 1996 if ((oldflags & LK_TYPE_MASK) == 0) 1997 VOP_UNLOCK(vp, 0, td); 1998 } else 1999 VI_UNLOCK(vp); 2000 return (0); 2001 } 2002 2003 /* 2004 * Increase the reference count of a vnode. 2005 */ 2006 void 2007 vref(struct vnode *vp) 2008 { 2009 2010 VI_LOCK(vp); 2011 v_incr_usecount(vp); 2012 VI_UNLOCK(vp); 2013 } 2014 2015 /* 2016 * Return reference count of a vnode. 2017 * 2018 * The results of this call are only guaranteed when some mechanism other 2019 * than the VI lock is used to stop other processes from gaining references 2020 * to the vnode. This may be the case if the caller holds the only reference. 2021 * This is also useful when stale data is acceptable as race conditions may 2022 * be accounted for by some other means. 2023 */ 2024 int 2025 vrefcnt(struct vnode *vp) 2026 { 2027 int usecnt; 2028 2029 VI_LOCK(vp); 2030 usecnt = vp->v_usecount; 2031 VI_UNLOCK(vp); 2032 2033 return (usecnt); 2034 } 2035 2036 2037 /* 2038 * Vnode put/release. 2039 * If count drops to zero, call inactive routine and return to freelist. 2040 */ 2041 void 2042 vrele(struct vnode *vp) 2043 { 2044 struct thread *td = curthread; /* XXX */ 2045 2046 KASSERT(vp != NULL, ("vrele: null vp")); 2047 VFS_ASSERT_GIANT(vp->v_mount); 2048 2049 VI_LOCK(vp); 2050 2051 /* Skip this v_writecount check if we're going to panic below. */ 2052 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2053 ("vrele: missed vn_close")); 2054 2055 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2056 vp->v_usecount == 1)) { 2057 v_decr_usecount(vp); 2058 return; 2059 } 2060 if (vp->v_usecount != 1) { 2061 #ifdef DIAGNOSTIC 2062 vprint("vrele: negative ref count", vp); 2063 #endif 2064 VI_UNLOCK(vp); 2065 panic("vrele: negative ref cnt"); 2066 } 2067 /* 2068 * We want to hold the vnode until the inactive finishes to 2069 * prevent vgone() races. We drop the use count here and the 2070 * hold count below when we're done. 2071 */ 2072 v_decr_useonly(vp); 2073 /* 2074 * We must call VOP_INACTIVE with the node locked. Mark 2075 * as VI_DOINGINACT to avoid recursion. 2076 */ 2077 vp->v_iflag |= VI_OWEINACT; 2078 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) { 2079 VI_LOCK(vp); 2080 if (vp->v_usecount > 0) 2081 vp->v_iflag &= ~VI_OWEINACT; 2082 if (vp->v_iflag & VI_OWEINACT) 2083 vinactive(vp, td); 2084 VOP_UNLOCK(vp, 0, td); 2085 } else { 2086 VI_LOCK(vp); 2087 if (vp->v_usecount > 0) 2088 vp->v_iflag &= ~VI_OWEINACT; 2089 } 2090 vdropl(vp); 2091 } 2092 2093 /* 2094 * Release an already locked vnode. This give the same effects as 2095 * unlock+vrele(), but takes less time and avoids releasing and 2096 * re-aquiring the lock (as vrele() aquires the lock internally.) 2097 */ 2098 void 2099 vput(struct vnode *vp) 2100 { 2101 struct thread *td = curthread; /* XXX */ 2102 int error; 2103 2104 KASSERT(vp != NULL, ("vput: null vp")); 2105 ASSERT_VOP_LOCKED(vp, "vput"); 2106 VFS_ASSERT_GIANT(vp->v_mount); 2107 VI_LOCK(vp); 2108 /* Skip this v_writecount check if we're going to panic below. */ 2109 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2110 ("vput: missed vn_close")); 2111 error = 0; 2112 2113 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2114 vp->v_usecount == 1)) { 2115 VOP_UNLOCK(vp, 0, td); 2116 v_decr_usecount(vp); 2117 return; 2118 } 2119 2120 if (vp->v_usecount != 1) { 2121 #ifdef DIAGNOSTIC 2122 vprint("vput: negative ref count", vp); 2123 #endif 2124 panic("vput: negative ref cnt"); 2125 } 2126 /* 2127 * We want to hold the vnode until the inactive finishes to 2128 * prevent vgone() races. We drop the use count here and the 2129 * hold count below when we're done. 2130 */ 2131 v_decr_useonly(vp); 2132 vp->v_iflag |= VI_OWEINACT; 2133 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) { 2134 error = VOP_LOCK(vp, LK_EXCLUPGRADE|LK_INTERLOCK|LK_NOWAIT, td); 2135 VI_LOCK(vp); 2136 if (error) { 2137 if (vp->v_usecount > 0) 2138 vp->v_iflag &= ~VI_OWEINACT; 2139 goto done; 2140 } 2141 } 2142 if (vp->v_usecount > 0) 2143 vp->v_iflag &= ~VI_OWEINACT; 2144 if (vp->v_iflag & VI_OWEINACT) 2145 vinactive(vp, td); 2146 VOP_UNLOCK(vp, 0, td); 2147 done: 2148 vdropl(vp); 2149 } 2150 2151 /* 2152 * Somebody doesn't want the vnode recycled. 2153 */ 2154 void 2155 vhold(struct vnode *vp) 2156 { 2157 2158 VI_LOCK(vp); 2159 vholdl(vp); 2160 VI_UNLOCK(vp); 2161 } 2162 2163 void 2164 vholdl(struct vnode *vp) 2165 { 2166 2167 vp->v_holdcnt++; 2168 if (VSHOULDBUSY(vp)) 2169 vbusy(vp); 2170 } 2171 2172 /* 2173 * Note that there is one less who cares about this vnode. vdrop() is the 2174 * opposite of vhold(). 2175 */ 2176 void 2177 vdrop(struct vnode *vp) 2178 { 2179 2180 VI_LOCK(vp); 2181 vdropl(vp); 2182 } 2183 2184 /* 2185 * Drop the hold count of the vnode. If this is the last reference to 2186 * the vnode we will free it if it has been vgone'd otherwise it is 2187 * placed on the free list. 2188 */ 2189 static void 2190 vdropl(struct vnode *vp) 2191 { 2192 2193 if (vp->v_holdcnt <= 0) 2194 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2195 vp->v_holdcnt--; 2196 if (vp->v_holdcnt == 0) { 2197 if (vp->v_iflag & VI_DOOMED) { 2198 vdestroy(vp); 2199 return; 2200 } else 2201 vfree(vp); 2202 } 2203 VI_UNLOCK(vp); 2204 } 2205 2206 /* 2207 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2208 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2209 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2210 * failed lock upgrade. 2211 */ 2212 static void 2213 vinactive(struct vnode *vp, struct thread *td) 2214 { 2215 2216 ASSERT_VOP_LOCKED(vp, "vinactive"); 2217 ASSERT_VI_LOCKED(vp, "vinactive"); 2218 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2219 ("vinactive: recursed on VI_DOINGINACT")); 2220 vp->v_iflag |= VI_DOINGINACT; 2221 vp->v_iflag &= ~VI_OWEINACT; 2222 VI_UNLOCK(vp); 2223 VOP_INACTIVE(vp, td); 2224 VI_LOCK(vp); 2225 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2226 ("vinactive: lost VI_DOINGINACT")); 2227 vp->v_iflag &= ~VI_DOINGINACT; 2228 } 2229 2230 /* 2231 * Remove any vnodes in the vnode table belonging to mount point mp. 2232 * 2233 * If FORCECLOSE is not specified, there should not be any active ones, 2234 * return error if any are found (nb: this is a user error, not a 2235 * system error). If FORCECLOSE is specified, detach any active vnodes 2236 * that are found. 2237 * 2238 * If WRITECLOSE is set, only flush out regular file vnodes open for 2239 * writing. 2240 * 2241 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2242 * 2243 * `rootrefs' specifies the base reference count for the root vnode 2244 * of this filesystem. The root vnode is considered busy if its 2245 * v_usecount exceeds this value. On a successful return, vflush(, td) 2246 * will call vrele() on the root vnode exactly rootrefs times. 2247 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2248 * be zero. 2249 */ 2250 #ifdef DIAGNOSTIC 2251 static int busyprt = 0; /* print out busy vnodes */ 2252 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 2253 #endif 2254 2255 int 2256 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td) 2257 { 2258 struct vnode *vp, *mvp, *rootvp = NULL; 2259 struct vattr vattr; 2260 int busy = 0, error; 2261 2262 CTR1(KTR_VFS, "vflush: mp %p", mp); 2263 if (rootrefs > 0) { 2264 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2265 ("vflush: bad args")); 2266 /* 2267 * Get the filesystem root vnode. We can vput() it 2268 * immediately, since with rootrefs > 0, it won't go away. 2269 */ 2270 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0) 2271 return (error); 2272 vput(rootvp); 2273 2274 } 2275 MNT_ILOCK(mp); 2276 loop: 2277 MNT_VNODE_FOREACH(vp, mp, mvp) { 2278 2279 VI_LOCK(vp); 2280 vholdl(vp); 2281 MNT_IUNLOCK(mp); 2282 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td); 2283 if (error) { 2284 vdrop(vp); 2285 MNT_ILOCK(mp); 2286 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2287 goto loop; 2288 } 2289 /* 2290 * Skip over a vnodes marked VV_SYSTEM. 2291 */ 2292 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2293 VOP_UNLOCK(vp, 0, td); 2294 vdrop(vp); 2295 MNT_ILOCK(mp); 2296 continue; 2297 } 2298 /* 2299 * If WRITECLOSE is set, flush out unlinked but still open 2300 * files (even if open only for reading) and regular file 2301 * vnodes open for writing. 2302 */ 2303 if (flags & WRITECLOSE) { 2304 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td); 2305 VI_LOCK(vp); 2306 2307 if ((vp->v_type == VNON || 2308 (error == 0 && vattr.va_nlink > 0)) && 2309 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2310 VOP_UNLOCK(vp, 0, td); 2311 vdropl(vp); 2312 MNT_ILOCK(mp); 2313 continue; 2314 } 2315 } else 2316 VI_LOCK(vp); 2317 /* 2318 * With v_usecount == 0, all we need to do is clear out the 2319 * vnode data structures and we are done. 2320 * 2321 * If FORCECLOSE is set, forcibly close the vnode. 2322 */ 2323 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2324 VNASSERT(vp->v_usecount == 0 || 2325 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2326 ("device VNODE %p is FORCECLOSED", vp)); 2327 vgonel(vp); 2328 } else { 2329 busy++; 2330 #ifdef DIAGNOSTIC 2331 if (busyprt) 2332 vprint("vflush: busy vnode", vp); 2333 #endif 2334 } 2335 VOP_UNLOCK(vp, 0, td); 2336 vdropl(vp); 2337 MNT_ILOCK(mp); 2338 } 2339 MNT_IUNLOCK(mp); 2340 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2341 /* 2342 * If just the root vnode is busy, and if its refcount 2343 * is equal to `rootrefs', then go ahead and kill it. 2344 */ 2345 VI_LOCK(rootvp); 2346 KASSERT(busy > 0, ("vflush: not busy")); 2347 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2348 ("vflush: usecount %d < rootrefs %d", 2349 rootvp->v_usecount, rootrefs)); 2350 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2351 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td); 2352 vgone(rootvp); 2353 VOP_UNLOCK(rootvp, 0, td); 2354 busy = 0; 2355 } else 2356 VI_UNLOCK(rootvp); 2357 } 2358 if (busy) 2359 return (EBUSY); 2360 for (; rootrefs > 0; rootrefs--) 2361 vrele(rootvp); 2362 return (0); 2363 } 2364 2365 /* 2366 * Recycle an unused vnode to the front of the free list. 2367 */ 2368 int 2369 vrecycle(struct vnode *vp, struct thread *td) 2370 { 2371 int recycled; 2372 2373 ASSERT_VOP_LOCKED(vp, "vrecycle"); 2374 recycled = 0; 2375 VI_LOCK(vp); 2376 if (vp->v_usecount == 0) { 2377 recycled = 1; 2378 vgonel(vp); 2379 } 2380 VI_UNLOCK(vp); 2381 return (recycled); 2382 } 2383 2384 /* 2385 * Eliminate all activity associated with a vnode 2386 * in preparation for reuse. 2387 */ 2388 void 2389 vgone(struct vnode *vp) 2390 { 2391 VI_LOCK(vp); 2392 vgonel(vp); 2393 VI_UNLOCK(vp); 2394 } 2395 2396 /* 2397 * vgone, with the vp interlock held. 2398 */ 2399 void 2400 vgonel(struct vnode *vp) 2401 { 2402 struct thread *td; 2403 int oweinact; 2404 int active; 2405 struct mount *mp; 2406 2407 CTR1(KTR_VFS, "vgonel: vp %p", vp); 2408 ASSERT_VOP_LOCKED(vp, "vgonel"); 2409 ASSERT_VI_LOCKED(vp, "vgonel"); 2410 VNASSERT(vp->v_holdcnt, vp, 2411 ("vgonel: vp %p has no reference.", vp)); 2412 td = curthread; 2413 2414 /* 2415 * Don't vgonel if we're already doomed. 2416 */ 2417 if (vp->v_iflag & VI_DOOMED) 2418 return; 2419 vp->v_iflag |= VI_DOOMED; 2420 /* 2421 * Check to see if the vnode is in use. If so, we have to call 2422 * VOP_CLOSE() and VOP_INACTIVE(). 2423 */ 2424 active = vp->v_usecount; 2425 oweinact = (vp->v_iflag & VI_OWEINACT); 2426 VI_UNLOCK(vp); 2427 /* 2428 * Clean out any buffers associated with the vnode. 2429 * If the flush fails, just toss the buffers. 2430 */ 2431 mp = NULL; 2432 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2433 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2434 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0) 2435 vinvalbuf(vp, 0, td, 0, 0); 2436 2437 /* 2438 * If purging an active vnode, it must be closed and 2439 * deactivated before being reclaimed. 2440 */ 2441 if (active) 2442 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2443 if (oweinact || active) { 2444 VI_LOCK(vp); 2445 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2446 vinactive(vp, td); 2447 VI_UNLOCK(vp); 2448 } 2449 /* 2450 * Reclaim the vnode. 2451 */ 2452 if (VOP_RECLAIM(vp, td)) 2453 panic("vgone: cannot reclaim"); 2454 if (mp != NULL) 2455 vn_finished_secondary_write(mp); 2456 VNASSERT(vp->v_object == NULL, vp, 2457 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2458 /* 2459 * Delete from old mount point vnode list. 2460 */ 2461 delmntque(vp); 2462 cache_purge(vp); 2463 /* 2464 * Done with purge, reset to the standard lock and invalidate 2465 * the vnode. 2466 */ 2467 VI_LOCK(vp); 2468 vp->v_vnlock = &vp->v_lock; 2469 vp->v_op = &dead_vnodeops; 2470 vp->v_tag = "none"; 2471 vp->v_type = VBAD; 2472 } 2473 2474 /* 2475 * Calculate the total number of references to a special device. 2476 */ 2477 int 2478 vcount(struct vnode *vp) 2479 { 2480 int count; 2481 2482 dev_lock(); 2483 count = vp->v_rdev->si_usecount; 2484 dev_unlock(); 2485 return (count); 2486 } 2487 2488 /* 2489 * Same as above, but using the struct cdev *as argument 2490 */ 2491 int 2492 count_dev(struct cdev *dev) 2493 { 2494 int count; 2495 2496 dev_lock(); 2497 count = dev->si_usecount; 2498 dev_unlock(); 2499 return(count); 2500 } 2501 2502 /* 2503 * Print out a description of a vnode. 2504 */ 2505 static char *typename[] = 2506 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2507 "VMARKER"}; 2508 2509 void 2510 vn_printf(struct vnode *vp, const char *fmt, ...) 2511 { 2512 va_list ap; 2513 char buf[96]; 2514 2515 va_start(ap, fmt); 2516 vprintf(fmt, ap); 2517 va_end(ap); 2518 printf("%p: ", (void *)vp); 2519 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2520 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2521 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2522 buf[0] = '\0'; 2523 buf[1] = '\0'; 2524 if (vp->v_vflag & VV_ROOT) 2525 strcat(buf, "|VV_ROOT"); 2526 if (vp->v_vflag & VV_TEXT) 2527 strcat(buf, "|VV_TEXT"); 2528 if (vp->v_vflag & VV_SYSTEM) 2529 strcat(buf, "|VV_SYSTEM"); 2530 if (vp->v_vflag & VV_DELETED) 2531 strcat(buf, "|VV_DELETED"); 2532 if (vp->v_iflag & VI_DOOMED) 2533 strcat(buf, "|VI_DOOMED"); 2534 if (vp->v_iflag & VI_FREE) 2535 strcat(buf, "|VI_FREE"); 2536 printf(" flags (%s)\n", buf + 1); 2537 if (mtx_owned(VI_MTX(vp))) 2538 printf(" VI_LOCKed"); 2539 if (vp->v_object != NULL) 2540 printf(" v_object %p ref %d pages %d\n", 2541 vp->v_object, vp->v_object->ref_count, 2542 vp->v_object->resident_page_count); 2543 printf(" "); 2544 lockmgr_printinfo(vp->v_vnlock); 2545 printf("\n"); 2546 if (vp->v_data != NULL) 2547 VOP_PRINT(vp); 2548 } 2549 2550 #ifdef DDB 2551 /* 2552 * List all of the locked vnodes in the system. 2553 * Called when debugging the kernel. 2554 */ 2555 DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2556 { 2557 struct mount *mp, *nmp; 2558 struct vnode *vp; 2559 2560 /* 2561 * Note: because this is DDB, we can't obey the locking semantics 2562 * for these structures, which means we could catch an inconsistent 2563 * state and dereference a nasty pointer. Not much to be done 2564 * about that. 2565 */ 2566 printf("Locked vnodes\n"); 2567 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2568 nmp = TAILQ_NEXT(mp, mnt_list); 2569 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2570 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL)) 2571 vprint("", vp); 2572 } 2573 nmp = TAILQ_NEXT(mp, mnt_list); 2574 } 2575 } 2576 2577 /* 2578 * Show details about the given vnode. 2579 */ 2580 DB_SHOW_COMMAND(vnode, db_show_vnode) 2581 { 2582 struct vnode *vp; 2583 2584 if (!have_addr) 2585 return; 2586 vp = (struct vnode *)addr; 2587 vn_printf(vp, "vnode "); 2588 } 2589 #endif /* DDB */ 2590 2591 /* 2592 * Fill in a struct xvfsconf based on a struct vfsconf. 2593 */ 2594 static void 2595 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2596 { 2597 2598 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2599 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2600 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2601 xvfsp->vfc_flags = vfsp->vfc_flags; 2602 /* 2603 * These are unused in userland, we keep them 2604 * to not break binary compatibility. 2605 */ 2606 xvfsp->vfc_vfsops = NULL; 2607 xvfsp->vfc_next = NULL; 2608 } 2609 2610 /* 2611 * Top level filesystem related information gathering. 2612 */ 2613 static int 2614 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2615 { 2616 struct vfsconf *vfsp; 2617 struct xvfsconf xvfsp; 2618 int error; 2619 2620 error = 0; 2621 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2622 bzero(&xvfsp, sizeof(xvfsp)); 2623 vfsconf2x(vfsp, &xvfsp); 2624 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 2625 if (error) 2626 break; 2627 } 2628 return (error); 2629 } 2630 2631 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist, 2632 "S,xvfsconf", "List of all configured filesystems"); 2633 2634 #ifndef BURN_BRIDGES 2635 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 2636 2637 static int 2638 vfs_sysctl(SYSCTL_HANDLER_ARGS) 2639 { 2640 int *name = (int *)arg1 - 1; /* XXX */ 2641 u_int namelen = arg2 + 1; /* XXX */ 2642 struct vfsconf *vfsp; 2643 struct xvfsconf xvfsp; 2644 2645 printf("WARNING: userland calling deprecated sysctl, " 2646 "please rebuild world\n"); 2647 2648 #if 1 || defined(COMPAT_PRELITE2) 2649 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 2650 if (namelen == 1) 2651 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 2652 #endif 2653 2654 switch (name[1]) { 2655 case VFS_MAXTYPENUM: 2656 if (namelen != 2) 2657 return (ENOTDIR); 2658 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 2659 case VFS_CONF: 2660 if (namelen != 3) 2661 return (ENOTDIR); /* overloaded */ 2662 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 2663 if (vfsp->vfc_typenum == name[2]) 2664 break; 2665 if (vfsp == NULL) 2666 return (EOPNOTSUPP); 2667 bzero(&xvfsp, sizeof(xvfsp)); 2668 vfsconf2x(vfsp, &xvfsp); 2669 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 2670 } 2671 return (EOPNOTSUPP); 2672 } 2673 2674 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 2675 vfs_sysctl, "Generic filesystem"); 2676 2677 #if 1 || defined(COMPAT_PRELITE2) 2678 2679 static int 2680 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 2681 { 2682 int error; 2683 struct vfsconf *vfsp; 2684 struct ovfsconf ovfs; 2685 2686 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2687 bzero(&ovfs, sizeof(ovfs)); 2688 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 2689 strcpy(ovfs.vfc_name, vfsp->vfc_name); 2690 ovfs.vfc_index = vfsp->vfc_typenum; 2691 ovfs.vfc_refcount = vfsp->vfc_refcount; 2692 ovfs.vfc_flags = vfsp->vfc_flags; 2693 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 2694 if (error) 2695 return error; 2696 } 2697 return 0; 2698 } 2699 2700 #endif /* 1 || COMPAT_PRELITE2 */ 2701 #endif /* !BURN_BRIDGES */ 2702 2703 #define KINFO_VNODESLOP 10 2704 #ifdef notyet 2705 /* 2706 * Dump vnode list (via sysctl). 2707 */ 2708 /* ARGSUSED */ 2709 static int 2710 sysctl_vnode(SYSCTL_HANDLER_ARGS) 2711 { 2712 struct xvnode *xvn; 2713 struct thread *td = req->td; 2714 struct mount *mp; 2715 struct vnode *vp; 2716 int error, len, n; 2717 2718 /* 2719 * Stale numvnodes access is not fatal here. 2720 */ 2721 req->lock = 0; 2722 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 2723 if (!req->oldptr) 2724 /* Make an estimate */ 2725 return (SYSCTL_OUT(req, 0, len)); 2726 2727 error = sysctl_wire_old_buffer(req, 0); 2728 if (error != 0) 2729 return (error); 2730 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 2731 n = 0; 2732 mtx_lock(&mountlist_mtx); 2733 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2734 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) 2735 continue; 2736 MNT_ILOCK(mp); 2737 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2738 if (n == len) 2739 break; 2740 vref(vp); 2741 xvn[n].xv_size = sizeof *xvn; 2742 xvn[n].xv_vnode = vp; 2743 xvn[n].xv_id = 0; /* XXX compat */ 2744 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 2745 XV_COPY(usecount); 2746 XV_COPY(writecount); 2747 XV_COPY(holdcnt); 2748 XV_COPY(mount); 2749 XV_COPY(numoutput); 2750 XV_COPY(type); 2751 #undef XV_COPY 2752 xvn[n].xv_flag = vp->v_vflag; 2753 2754 switch (vp->v_type) { 2755 case VREG: 2756 case VDIR: 2757 case VLNK: 2758 break; 2759 case VBLK: 2760 case VCHR: 2761 if (vp->v_rdev == NULL) { 2762 vrele(vp); 2763 continue; 2764 } 2765 xvn[n].xv_dev = dev2udev(vp->v_rdev); 2766 break; 2767 case VSOCK: 2768 xvn[n].xv_socket = vp->v_socket; 2769 break; 2770 case VFIFO: 2771 xvn[n].xv_fifo = vp->v_fifoinfo; 2772 break; 2773 case VNON: 2774 case VBAD: 2775 default: 2776 /* shouldn't happen? */ 2777 vrele(vp); 2778 continue; 2779 } 2780 vrele(vp); 2781 ++n; 2782 } 2783 MNT_IUNLOCK(mp); 2784 mtx_lock(&mountlist_mtx); 2785 vfs_unbusy(mp, td); 2786 if (n == len) 2787 break; 2788 } 2789 mtx_unlock(&mountlist_mtx); 2790 2791 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 2792 free(xvn, M_TEMP); 2793 return (error); 2794 } 2795 2796 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 2797 0, 0, sysctl_vnode, "S,xvnode", ""); 2798 #endif 2799 2800 /* 2801 * Unmount all filesystems. The list is traversed in reverse order 2802 * of mounting to avoid dependencies. 2803 */ 2804 void 2805 vfs_unmountall(void) 2806 { 2807 struct mount *mp; 2808 struct thread *td; 2809 int error; 2810 2811 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 2812 td = curthread; 2813 /* 2814 * Since this only runs when rebooting, it is not interlocked. 2815 */ 2816 while(!TAILQ_EMPTY(&mountlist)) { 2817 mp = TAILQ_LAST(&mountlist, mntlist); 2818 error = dounmount(mp, MNT_FORCE, td); 2819 if (error) { 2820 TAILQ_REMOVE(&mountlist, mp, mnt_list); 2821 /* 2822 * XXX: Due to the way in which we mount the root 2823 * file system off of devfs, devfs will generate a 2824 * "busy" warning when we try to unmount it before 2825 * the root. Don't print a warning as a result in 2826 * order to avoid false positive errors that may 2827 * cause needless upset. 2828 */ 2829 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 2830 printf("unmount of %s failed (", 2831 mp->mnt_stat.f_mntonname); 2832 if (error == EBUSY) 2833 printf("BUSY)\n"); 2834 else 2835 printf("%d)\n", error); 2836 } 2837 } else { 2838 /* The unmount has removed mp from the mountlist */ 2839 } 2840 } 2841 } 2842 2843 /* 2844 * perform msync on all vnodes under a mount point 2845 * the mount point must be locked. 2846 */ 2847 void 2848 vfs_msync(struct mount *mp, int flags) 2849 { 2850 struct vnode *vp, *mvp; 2851 struct vm_object *obj; 2852 2853 MNT_ILOCK(mp); 2854 MNT_VNODE_FOREACH(vp, mp, mvp) { 2855 VI_LOCK(vp); 2856 if ((vp->v_iflag & VI_OBJDIRTY) && 2857 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) { 2858 MNT_IUNLOCK(mp); 2859 if (!vget(vp, 2860 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 2861 curthread)) { 2862 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 2863 vput(vp); 2864 MNT_ILOCK(mp); 2865 continue; 2866 } 2867 2868 obj = vp->v_object; 2869 if (obj != NULL) { 2870 VM_OBJECT_LOCK(obj); 2871 vm_object_page_clean(obj, 0, 0, 2872 flags == MNT_WAIT ? 2873 OBJPC_SYNC : OBJPC_NOSYNC); 2874 VM_OBJECT_UNLOCK(obj); 2875 } 2876 vput(vp); 2877 } 2878 MNT_ILOCK(mp); 2879 } else 2880 VI_UNLOCK(vp); 2881 } 2882 MNT_IUNLOCK(mp); 2883 } 2884 2885 /* 2886 * Mark a vnode as free, putting it up for recycling. 2887 */ 2888 static void 2889 vfree(struct vnode *vp) 2890 { 2891 2892 CTR1(KTR_VFS, "vfree vp %p", vp); 2893 ASSERT_VI_LOCKED(vp, "vfree"); 2894 mtx_lock(&vnode_free_list_mtx); 2895 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed.")); 2896 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); 2897 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't")); 2898 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp, 2899 ("vfree: Freeing doomed vnode")); 2900 if (vp->v_iflag & VI_AGE) { 2901 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 2902 } else { 2903 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 2904 } 2905 freevnodes++; 2906 vp->v_iflag &= ~VI_AGE; 2907 vp->v_iflag |= VI_FREE; 2908 mtx_unlock(&vnode_free_list_mtx); 2909 } 2910 2911 /* 2912 * Opposite of vfree() - mark a vnode as in use. 2913 */ 2914 static void 2915 vbusy(struct vnode *vp) 2916 { 2917 CTR1(KTR_VFS, "vbusy vp %p", vp); 2918 ASSERT_VI_LOCKED(vp, "vbusy"); 2919 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 2920 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed.")); 2921 2922 mtx_lock(&vnode_free_list_mtx); 2923 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 2924 freevnodes--; 2925 vp->v_iflag &= ~(VI_FREE|VI_AGE); 2926 mtx_unlock(&vnode_free_list_mtx); 2927 } 2928 2929 /* 2930 * Initalize per-vnode helper structure to hold poll-related state. 2931 */ 2932 void 2933 v_addpollinfo(struct vnode *vp) 2934 { 2935 struct vpollinfo *vi; 2936 2937 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 2938 if (vp->v_pollinfo != NULL) { 2939 uma_zfree(vnodepoll_zone, vi); 2940 return; 2941 } 2942 vp->v_pollinfo = vi; 2943 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 2944 knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock, 2945 vfs_knlunlock, vfs_knllocked); 2946 } 2947 2948 /* 2949 * Record a process's interest in events which might happen to 2950 * a vnode. Because poll uses the historic select-style interface 2951 * internally, this routine serves as both the ``check for any 2952 * pending events'' and the ``record my interest in future events'' 2953 * functions. (These are done together, while the lock is held, 2954 * to avoid race conditions.) 2955 */ 2956 int 2957 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 2958 { 2959 2960 if (vp->v_pollinfo == NULL) 2961 v_addpollinfo(vp); 2962 mtx_lock(&vp->v_pollinfo->vpi_lock); 2963 if (vp->v_pollinfo->vpi_revents & events) { 2964 /* 2965 * This leaves events we are not interested 2966 * in available for the other process which 2967 * which presumably had requested them 2968 * (otherwise they would never have been 2969 * recorded). 2970 */ 2971 events &= vp->v_pollinfo->vpi_revents; 2972 vp->v_pollinfo->vpi_revents &= ~events; 2973 2974 mtx_unlock(&vp->v_pollinfo->vpi_lock); 2975 return events; 2976 } 2977 vp->v_pollinfo->vpi_events |= events; 2978 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 2979 mtx_unlock(&vp->v_pollinfo->vpi_lock); 2980 return 0; 2981 } 2982 2983 /* 2984 * Routine to create and manage a filesystem syncer vnode. 2985 */ 2986 #define sync_close ((int (*)(struct vop_close_args *))nullop) 2987 static int sync_fsync(struct vop_fsync_args *); 2988 static int sync_inactive(struct vop_inactive_args *); 2989 static int sync_reclaim(struct vop_reclaim_args *); 2990 2991 static struct vop_vector sync_vnodeops = { 2992 .vop_bypass = VOP_EOPNOTSUPP, 2993 .vop_close = sync_close, /* close */ 2994 .vop_fsync = sync_fsync, /* fsync */ 2995 .vop_inactive = sync_inactive, /* inactive */ 2996 .vop_reclaim = sync_reclaim, /* reclaim */ 2997 ._vop_lock = vop_stdlock, /* lock */ 2998 .vop_unlock = vop_stdunlock, /* unlock */ 2999 .vop_islocked = vop_stdislocked, /* islocked */ 3000 }; 3001 3002 /* 3003 * Create a new filesystem syncer vnode for the specified mount point. 3004 */ 3005 int 3006 vfs_allocate_syncvnode(struct mount *mp) 3007 { 3008 struct vnode *vp; 3009 static long start, incr, next; 3010 int error; 3011 3012 /* Allocate a new vnode */ 3013 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) { 3014 mp->mnt_syncer = NULL; 3015 return (error); 3016 } 3017 vp->v_type = VNON; 3018 /* 3019 * Place the vnode onto the syncer worklist. We attempt to 3020 * scatter them about on the list so that they will go off 3021 * at evenly distributed times even if all the filesystems 3022 * are mounted at once. 3023 */ 3024 next += incr; 3025 if (next == 0 || next > syncer_maxdelay) { 3026 start /= 2; 3027 incr /= 2; 3028 if (start == 0) { 3029 start = syncer_maxdelay / 2; 3030 incr = syncer_maxdelay; 3031 } 3032 next = start; 3033 } 3034 VI_LOCK(vp); 3035 vn_syncer_add_to_worklist(&vp->v_bufobj, 3036 syncdelay > 0 ? next % syncdelay : 0); 3037 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3038 mtx_lock(&sync_mtx); 3039 sync_vnode_count++; 3040 mtx_unlock(&sync_mtx); 3041 VI_UNLOCK(vp); 3042 mp->mnt_syncer = vp; 3043 return (0); 3044 } 3045 3046 /* 3047 * Do a lazy sync of the filesystem. 3048 */ 3049 static int 3050 sync_fsync(struct vop_fsync_args *ap) 3051 { 3052 struct vnode *syncvp = ap->a_vp; 3053 struct mount *mp = syncvp->v_mount; 3054 struct thread *td = ap->a_td; 3055 int error; 3056 struct bufobj *bo; 3057 3058 /* 3059 * We only need to do something if this is a lazy evaluation. 3060 */ 3061 if (ap->a_waitfor != MNT_LAZY) 3062 return (0); 3063 3064 /* 3065 * Move ourselves to the back of the sync list. 3066 */ 3067 bo = &syncvp->v_bufobj; 3068 BO_LOCK(bo); 3069 vn_syncer_add_to_worklist(bo, syncdelay); 3070 BO_UNLOCK(bo); 3071 3072 /* 3073 * Walk the list of vnodes pushing all that are dirty and 3074 * not already on the sync list. 3075 */ 3076 mtx_lock(&mountlist_mtx); 3077 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) { 3078 mtx_unlock(&mountlist_mtx); 3079 return (0); 3080 } 3081 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3082 vfs_unbusy(mp, td); 3083 return (0); 3084 } 3085 MNT_ILOCK(mp); 3086 mp->mnt_noasync++; 3087 mp->mnt_kern_flag &= ~MNTK_ASYNC; 3088 MNT_IUNLOCK(mp); 3089 vfs_msync(mp, MNT_NOWAIT); 3090 error = VFS_SYNC(mp, MNT_LAZY, td); 3091 MNT_ILOCK(mp); 3092 mp->mnt_noasync--; 3093 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 3094 mp->mnt_kern_flag |= MNTK_ASYNC; 3095 MNT_IUNLOCK(mp); 3096 vn_finished_write(mp); 3097 vfs_unbusy(mp, td); 3098 return (error); 3099 } 3100 3101 /* 3102 * The syncer vnode is no referenced. 3103 */ 3104 static int 3105 sync_inactive(struct vop_inactive_args *ap) 3106 { 3107 3108 vgone(ap->a_vp); 3109 return (0); 3110 } 3111 3112 /* 3113 * The syncer vnode is no longer needed and is being decommissioned. 3114 * 3115 * Modifications to the worklist must be protected by sync_mtx. 3116 */ 3117 static int 3118 sync_reclaim(struct vop_reclaim_args *ap) 3119 { 3120 struct vnode *vp = ap->a_vp; 3121 struct bufobj *bo; 3122 3123 VI_LOCK(vp); 3124 bo = &vp->v_bufobj; 3125 vp->v_mount->mnt_syncer = NULL; 3126 if (bo->bo_flag & BO_ONWORKLST) { 3127 mtx_lock(&sync_mtx); 3128 LIST_REMOVE(bo, bo_synclist); 3129 syncer_worklist_len--; 3130 sync_vnode_count--; 3131 mtx_unlock(&sync_mtx); 3132 bo->bo_flag &= ~BO_ONWORKLST; 3133 } 3134 VI_UNLOCK(vp); 3135 3136 return (0); 3137 } 3138 3139 /* 3140 * Check if vnode represents a disk device 3141 */ 3142 int 3143 vn_isdisk(struct vnode *vp, int *errp) 3144 { 3145 int error; 3146 3147 error = 0; 3148 dev_lock(); 3149 if (vp->v_type != VCHR) 3150 error = ENOTBLK; 3151 else if (vp->v_rdev == NULL) 3152 error = ENXIO; 3153 else if (vp->v_rdev->si_devsw == NULL) 3154 error = ENXIO; 3155 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3156 error = ENOTBLK; 3157 dev_unlock(); 3158 if (errp != NULL) 3159 *errp = error; 3160 return (error == 0); 3161 } 3162 3163 /* 3164 * Common filesystem object access control check routine. Accepts a 3165 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3166 * and optional call-by-reference privused argument allowing vaccess() 3167 * to indicate to the caller whether privilege was used to satisfy the 3168 * request (obsoleted). Returns 0 on success, or an errno on failure. 3169 * 3170 * The ifdef'd CAPABILITIES version is here for reference, but is not 3171 * actually used. 3172 */ 3173 int 3174 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3175 mode_t acc_mode, struct ucred *cred, int *privused) 3176 { 3177 mode_t dac_granted; 3178 mode_t priv_granted; 3179 3180 /* 3181 * Look for a normal, non-privileged way to access the file/directory 3182 * as requested. If it exists, go with that. 3183 */ 3184 3185 if (privused != NULL) 3186 *privused = 0; 3187 3188 dac_granted = 0; 3189 3190 /* Check the owner. */ 3191 if (cred->cr_uid == file_uid) { 3192 dac_granted |= VADMIN; 3193 if (file_mode & S_IXUSR) 3194 dac_granted |= VEXEC; 3195 if (file_mode & S_IRUSR) 3196 dac_granted |= VREAD; 3197 if (file_mode & S_IWUSR) 3198 dac_granted |= (VWRITE | VAPPEND); 3199 3200 if ((acc_mode & dac_granted) == acc_mode) 3201 return (0); 3202 3203 goto privcheck; 3204 } 3205 3206 /* Otherwise, check the groups (first match) */ 3207 if (groupmember(file_gid, cred)) { 3208 if (file_mode & S_IXGRP) 3209 dac_granted |= VEXEC; 3210 if (file_mode & S_IRGRP) 3211 dac_granted |= VREAD; 3212 if (file_mode & S_IWGRP) 3213 dac_granted |= (VWRITE | VAPPEND); 3214 3215 if ((acc_mode & dac_granted) == acc_mode) 3216 return (0); 3217 3218 goto privcheck; 3219 } 3220 3221 /* Otherwise, check everyone else. */ 3222 if (file_mode & S_IXOTH) 3223 dac_granted |= VEXEC; 3224 if (file_mode & S_IROTH) 3225 dac_granted |= VREAD; 3226 if (file_mode & S_IWOTH) 3227 dac_granted |= (VWRITE | VAPPEND); 3228 if ((acc_mode & dac_granted) == acc_mode) 3229 return (0); 3230 3231 privcheck: 3232 /* 3233 * Build a privilege mask to determine if the set of privileges 3234 * satisfies the requirements when combined with the granted mask 3235 * from above. For each privilege, if the privilege is required, 3236 * bitwise or the request type onto the priv_granted mask. 3237 */ 3238 priv_granted = 0; 3239 3240 if (type == VDIR) { 3241 /* 3242 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3243 * requests, instead of PRIV_VFS_EXEC. 3244 */ 3245 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3246 !priv_check_cred(cred, PRIV_VFS_LOOKUP, SUSER_ALLOWJAIL)) 3247 priv_granted |= VEXEC; 3248 } else { 3249 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3250 !priv_check_cred(cred, PRIV_VFS_EXEC, SUSER_ALLOWJAIL)) 3251 priv_granted |= VEXEC; 3252 } 3253 3254 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) && 3255 !priv_check_cred(cred, PRIV_VFS_READ, SUSER_ALLOWJAIL)) 3256 priv_granted |= VREAD; 3257 3258 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3259 !priv_check_cred(cred, PRIV_VFS_WRITE, SUSER_ALLOWJAIL)) 3260 priv_granted |= (VWRITE | VAPPEND); 3261 3262 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3263 !priv_check_cred(cred, PRIV_VFS_ADMIN, SUSER_ALLOWJAIL)) 3264 priv_granted |= VADMIN; 3265 3266 if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) { 3267 /* XXX audit: privilege used */ 3268 if (privused != NULL) 3269 *privused = 1; 3270 return (0); 3271 } 3272 3273 return ((acc_mode & VADMIN) ? EPERM : EACCES); 3274 } 3275 3276 /* 3277 * Credential check based on process requesting service, and per-attribute 3278 * permissions. 3279 */ 3280 int 3281 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3282 struct thread *td, int access) 3283 { 3284 3285 /* 3286 * Kernel-invoked always succeeds. 3287 */ 3288 if (cred == NOCRED) 3289 return (0); 3290 3291 /* 3292 * Do not allow privileged processes in jail to directly manipulate 3293 * system attributes. 3294 */ 3295 switch (attrnamespace) { 3296 case EXTATTR_NAMESPACE_SYSTEM: 3297 /* Potentially should be: return (EPERM); */ 3298 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3299 case EXTATTR_NAMESPACE_USER: 3300 return (VOP_ACCESS(vp, access, cred, td)); 3301 default: 3302 return (EPERM); 3303 } 3304 } 3305 3306 #ifdef DEBUG_VFS_LOCKS 3307 /* 3308 * This only exists to supress warnings from unlocked specfs accesses. It is 3309 * no longer ok to have an unlocked VFS. 3310 */ 3311 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD) 3312 3313 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3314 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, ""); 3315 3316 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3317 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, ""); 3318 3319 int vfs_badlock_print = 1; /* Print lock violations. */ 3320 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, ""); 3321 3322 #ifdef KDB 3323 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3324 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, ""); 3325 #endif 3326 3327 static void 3328 vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3329 { 3330 3331 #ifdef KDB 3332 if (vfs_badlock_backtrace) 3333 kdb_backtrace(); 3334 #endif 3335 if (vfs_badlock_print) 3336 printf("%s: %p %s\n", str, (void *)vp, msg); 3337 if (vfs_badlock_ddb) 3338 kdb_enter("lock violation"); 3339 } 3340 3341 void 3342 assert_vi_locked(struct vnode *vp, const char *str) 3343 { 3344 3345 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3346 vfs_badlock("interlock is not locked but should be", str, vp); 3347 } 3348 3349 void 3350 assert_vi_unlocked(struct vnode *vp, const char *str) 3351 { 3352 3353 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3354 vfs_badlock("interlock is locked but should not be", str, vp); 3355 } 3356 3357 void 3358 assert_vop_locked(struct vnode *vp, const char *str) 3359 { 3360 3361 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0) 3362 vfs_badlock("is not locked but should be", str, vp); 3363 } 3364 3365 void 3366 assert_vop_unlocked(struct vnode *vp, const char *str) 3367 { 3368 3369 if (vp && !IGNORE_LOCK(vp) && 3370 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE) 3371 vfs_badlock("is locked but should not be", str, vp); 3372 } 3373 3374 void 3375 assert_vop_elocked(struct vnode *vp, const char *str) 3376 { 3377 3378 if (vp && !IGNORE_LOCK(vp) && 3379 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE) 3380 vfs_badlock("is not exclusive locked but should be", str, vp); 3381 } 3382 3383 #if 0 3384 void 3385 assert_vop_elocked_other(struct vnode *vp, const char *str) 3386 { 3387 3388 if (vp && !IGNORE_LOCK(vp) && 3389 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER) 3390 vfs_badlock("is not exclusive locked by another thread", 3391 str, vp); 3392 } 3393 3394 void 3395 assert_vop_slocked(struct vnode *vp, const char *str) 3396 { 3397 3398 if (vp && !IGNORE_LOCK(vp) && 3399 VOP_ISLOCKED(vp, curthread) != LK_SHARED) 3400 vfs_badlock("is not locked shared but should be", str, vp); 3401 } 3402 #endif /* 0 */ 3403 #endif /* DEBUG_VFS_LOCKS */ 3404 3405 void 3406 vop_rename_pre(void *ap) 3407 { 3408 struct vop_rename_args *a = ap; 3409 3410 #ifdef DEBUG_VFS_LOCKS 3411 if (a->a_tvp) 3412 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3413 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3414 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3415 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3416 3417 /* Check the source (from). */ 3418 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp) 3419 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3420 if (a->a_tvp != a->a_fvp) 3421 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3422 3423 /* Check the target. */ 3424 if (a->a_tvp) 3425 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3426 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3427 #endif 3428 if (a->a_tdvp != a->a_fdvp) 3429 vhold(a->a_fdvp); 3430 if (a->a_tvp != a->a_fvp) 3431 vhold(a->a_fvp); 3432 vhold(a->a_tdvp); 3433 if (a->a_tvp) 3434 vhold(a->a_tvp); 3435 } 3436 3437 void 3438 vop_strategy_pre(void *ap) 3439 { 3440 #ifdef DEBUG_VFS_LOCKS 3441 struct vop_strategy_args *a; 3442 struct buf *bp; 3443 3444 a = ap; 3445 bp = a->a_bp; 3446 3447 /* 3448 * Cluster ops lock their component buffers but not the IO container. 3449 */ 3450 if ((bp->b_flags & B_CLUSTER) != 0) 3451 return; 3452 3453 if (BUF_REFCNT(bp) < 1) { 3454 if (vfs_badlock_print) 3455 printf( 3456 "VOP_STRATEGY: bp is not locked but should be\n"); 3457 if (vfs_badlock_ddb) 3458 kdb_enter("lock violation"); 3459 } 3460 #endif 3461 } 3462 3463 void 3464 vop_lookup_pre(void *ap) 3465 { 3466 #ifdef DEBUG_VFS_LOCKS 3467 struct vop_lookup_args *a; 3468 struct vnode *dvp; 3469 3470 a = ap; 3471 dvp = a->a_dvp; 3472 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3473 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3474 #endif 3475 } 3476 3477 void 3478 vop_lookup_post(void *ap, int rc) 3479 { 3480 #ifdef DEBUG_VFS_LOCKS 3481 struct vop_lookup_args *a; 3482 struct vnode *dvp; 3483 struct vnode *vp; 3484 3485 a = ap; 3486 dvp = a->a_dvp; 3487 vp = *(a->a_vpp); 3488 3489 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3490 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3491 3492 if (!rc) 3493 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 3494 #endif 3495 } 3496 3497 void 3498 vop_lock_pre(void *ap) 3499 { 3500 #ifdef DEBUG_VFS_LOCKS 3501 struct _vop_lock_args *a = ap; 3502 3503 if ((a->a_flags & LK_INTERLOCK) == 0) 3504 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3505 else 3506 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3507 #endif 3508 } 3509 3510 void 3511 vop_lock_post(void *ap, int rc) 3512 { 3513 #ifdef DEBUG_VFS_LOCKS 3514 struct _vop_lock_args *a = ap; 3515 3516 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3517 if (rc == 0) 3518 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3519 #endif 3520 } 3521 3522 void 3523 vop_unlock_pre(void *ap) 3524 { 3525 #ifdef DEBUG_VFS_LOCKS 3526 struct vop_unlock_args *a = ap; 3527 3528 if (a->a_flags & LK_INTERLOCK) 3529 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3530 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3531 #endif 3532 } 3533 3534 void 3535 vop_unlock_post(void *ap, int rc) 3536 { 3537 #ifdef DEBUG_VFS_LOCKS 3538 struct vop_unlock_args *a = ap; 3539 3540 if (a->a_flags & LK_INTERLOCK) 3541 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3542 #endif 3543 } 3544 3545 void 3546 vop_create_post(void *ap, int rc) 3547 { 3548 struct vop_create_args *a = ap; 3549 3550 if (!rc) 3551 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3552 } 3553 3554 void 3555 vop_link_post(void *ap, int rc) 3556 { 3557 struct vop_link_args *a = ap; 3558 3559 if (!rc) { 3560 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 3561 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 3562 } 3563 } 3564 3565 void 3566 vop_mkdir_post(void *ap, int rc) 3567 { 3568 struct vop_mkdir_args *a = ap; 3569 3570 if (!rc) 3571 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3572 } 3573 3574 void 3575 vop_mknod_post(void *ap, int rc) 3576 { 3577 struct vop_mknod_args *a = ap; 3578 3579 if (!rc) 3580 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3581 } 3582 3583 void 3584 vop_remove_post(void *ap, int rc) 3585 { 3586 struct vop_remove_args *a = ap; 3587 3588 if (!rc) { 3589 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3590 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 3591 } 3592 } 3593 3594 void 3595 vop_rename_post(void *ap, int rc) 3596 { 3597 struct vop_rename_args *a = ap; 3598 3599 if (!rc) { 3600 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 3601 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 3602 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 3603 if (a->a_tvp) 3604 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 3605 } 3606 if (a->a_tdvp != a->a_fdvp) 3607 vdrop(a->a_fdvp); 3608 if (a->a_tvp != a->a_fvp) 3609 vdrop(a->a_fvp); 3610 vdrop(a->a_tdvp); 3611 if (a->a_tvp) 3612 vdrop(a->a_tvp); 3613 } 3614 3615 void 3616 vop_rmdir_post(void *ap, int rc) 3617 { 3618 struct vop_rmdir_args *a = ap; 3619 3620 if (!rc) { 3621 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 3622 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 3623 } 3624 } 3625 3626 void 3627 vop_setattr_post(void *ap, int rc) 3628 { 3629 struct vop_setattr_args *a = ap; 3630 3631 if (!rc) 3632 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 3633 } 3634 3635 void 3636 vop_symlink_post(void *ap, int rc) 3637 { 3638 struct vop_symlink_args *a = ap; 3639 3640 if (!rc) 3641 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3642 } 3643 3644 static struct knlist fs_knlist; 3645 3646 static void 3647 vfs_event_init(void *arg) 3648 { 3649 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL); 3650 } 3651 /* XXX - correct order? */ 3652 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 3653 3654 void 3655 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused) 3656 { 3657 3658 KNOTE_UNLOCKED(&fs_knlist, event); 3659 } 3660 3661 static int filt_fsattach(struct knote *kn); 3662 static void filt_fsdetach(struct knote *kn); 3663 static int filt_fsevent(struct knote *kn, long hint); 3664 3665 struct filterops fs_filtops = 3666 { 0, filt_fsattach, filt_fsdetach, filt_fsevent }; 3667 3668 static int 3669 filt_fsattach(struct knote *kn) 3670 { 3671 3672 kn->kn_flags |= EV_CLEAR; 3673 knlist_add(&fs_knlist, kn, 0); 3674 return (0); 3675 } 3676 3677 static void 3678 filt_fsdetach(struct knote *kn) 3679 { 3680 3681 knlist_remove(&fs_knlist, kn, 0); 3682 } 3683 3684 static int 3685 filt_fsevent(struct knote *kn, long hint) 3686 { 3687 3688 kn->kn_fflags |= hint; 3689 return (kn->kn_fflags != 0); 3690 } 3691 3692 static int 3693 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 3694 { 3695 struct vfsidctl vc; 3696 int error; 3697 struct mount *mp; 3698 3699 error = SYSCTL_IN(req, &vc, sizeof(vc)); 3700 if (error) 3701 return (error); 3702 if (vc.vc_vers != VFS_CTL_VERS1) 3703 return (EINVAL); 3704 mp = vfs_getvfs(&vc.vc_fsid); 3705 if (mp == NULL) 3706 return (ENOENT); 3707 /* ensure that a specific sysctl goes to the right filesystem. */ 3708 if (strcmp(vc.vc_fstypename, "*") != 0 && 3709 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 3710 vfs_rel(mp); 3711 return (EINVAL); 3712 } 3713 VCTLTOREQ(&vc, req); 3714 error = VFS_SYSCTL(mp, vc.vc_op, req); 3715 vfs_rel(mp); 3716 return (error); 3717 } 3718 3719 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, 3720 NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); 3721 3722 /* 3723 * Function to initialize a va_filerev field sensibly. 3724 * XXX: Wouldn't a random number make a lot more sense ?? 3725 */ 3726 u_quad_t 3727 init_va_filerev(void) 3728 { 3729 struct bintime bt; 3730 3731 getbinuptime(&bt); 3732 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 3733 } 3734 3735 static int filt_vfsread(struct knote *kn, long hint); 3736 static int filt_vfswrite(struct knote *kn, long hint); 3737 static int filt_vfsvnode(struct knote *kn, long hint); 3738 static void filt_vfsdetach(struct knote *kn); 3739 static struct filterops vfsread_filtops = 3740 { 1, NULL, filt_vfsdetach, filt_vfsread }; 3741 static struct filterops vfswrite_filtops = 3742 { 1, NULL, filt_vfsdetach, filt_vfswrite }; 3743 static struct filterops vfsvnode_filtops = 3744 { 1, NULL, filt_vfsdetach, filt_vfsvnode }; 3745 3746 static void 3747 vfs_knllock(void *arg) 3748 { 3749 struct vnode *vp = arg; 3750 3751 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread); 3752 } 3753 3754 static void 3755 vfs_knlunlock(void *arg) 3756 { 3757 struct vnode *vp = arg; 3758 3759 VOP_UNLOCK(vp, 0, curthread); 3760 } 3761 3762 static int 3763 vfs_knllocked(void *arg) 3764 { 3765 struct vnode *vp = arg; 3766 3767 return (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE); 3768 } 3769 3770 int 3771 vfs_kqfilter(struct vop_kqfilter_args *ap) 3772 { 3773 struct vnode *vp = ap->a_vp; 3774 struct knote *kn = ap->a_kn; 3775 struct knlist *knl; 3776 3777 switch (kn->kn_filter) { 3778 case EVFILT_READ: 3779 kn->kn_fop = &vfsread_filtops; 3780 break; 3781 case EVFILT_WRITE: 3782 kn->kn_fop = &vfswrite_filtops; 3783 break; 3784 case EVFILT_VNODE: 3785 kn->kn_fop = &vfsvnode_filtops; 3786 break; 3787 default: 3788 return (EINVAL); 3789 } 3790 3791 kn->kn_hook = (caddr_t)vp; 3792 3793 if (vp->v_pollinfo == NULL) 3794 v_addpollinfo(vp); 3795 if (vp->v_pollinfo == NULL) 3796 return (ENOMEM); 3797 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 3798 knlist_add(knl, kn, 0); 3799 3800 return (0); 3801 } 3802 3803 /* 3804 * Detach knote from vnode 3805 */ 3806 static void 3807 filt_vfsdetach(struct knote *kn) 3808 { 3809 struct vnode *vp = (struct vnode *)kn->kn_hook; 3810 3811 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 3812 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 3813 } 3814 3815 /*ARGSUSED*/ 3816 static int 3817 filt_vfsread(struct knote *kn, long hint) 3818 { 3819 struct vnode *vp = (struct vnode *)kn->kn_hook; 3820 struct vattr va; 3821 3822 /* 3823 * filesystem is gone, so set the EOF flag and schedule 3824 * the knote for deletion. 3825 */ 3826 if (hint == NOTE_REVOKE) { 3827 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 3828 return (1); 3829 } 3830 3831 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread)) 3832 return (0); 3833 3834 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 3835 return (kn->kn_data != 0); 3836 } 3837 3838 /*ARGSUSED*/ 3839 static int 3840 filt_vfswrite(struct knote *kn, long hint) 3841 { 3842 /* 3843 * filesystem is gone, so set the EOF flag and schedule 3844 * the knote for deletion. 3845 */ 3846 if (hint == NOTE_REVOKE) 3847 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 3848 3849 kn->kn_data = 0; 3850 return (1); 3851 } 3852 3853 static int 3854 filt_vfsvnode(struct knote *kn, long hint) 3855 { 3856 if (kn->kn_sfflags & hint) 3857 kn->kn_fflags |= hint; 3858 if (hint == NOTE_REVOKE) { 3859 kn->kn_flags |= EV_EOF; 3860 return (1); 3861 } 3862 return (kn->kn_fflags != 0); 3863 } 3864 3865 int 3866 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 3867 { 3868 int error; 3869 3870 if (dp->d_reclen > ap->a_uio->uio_resid) 3871 return (ENAMETOOLONG); 3872 error = uiomove(dp, dp->d_reclen, ap->a_uio); 3873 if (error) { 3874 if (ap->a_ncookies != NULL) { 3875 if (ap->a_cookies != NULL) 3876 free(ap->a_cookies, M_TEMP); 3877 ap->a_cookies = NULL; 3878 *ap->a_ncookies = 0; 3879 } 3880 return (error); 3881 } 3882 if (ap->a_ncookies == NULL) 3883 return (0); 3884 3885 KASSERT(ap->a_cookies, 3886 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 3887 3888 *ap->a_cookies = realloc(*ap->a_cookies, 3889 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 3890 (*ap->a_cookies)[*ap->a_ncookies] = off; 3891 return (0); 3892 } 3893 3894 /* 3895 * Mark for update the access time of the file if the filesystem 3896 * supports VA_MARK_ATIME. This functionality is used by execve 3897 * and mmap, so we want to avoid the synchronous I/O implied by 3898 * directly setting va_atime for the sake of efficiency. 3899 */ 3900 void 3901 vfs_mark_atime(struct vnode *vp, struct thread *td) 3902 { 3903 struct vattr atimeattr; 3904 3905 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 3906 VATTR_NULL(&atimeattr); 3907 atimeattr.va_vaflags |= VA_MARK_ATIME; 3908 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td); 3909 } 3910 } 3911