1 /* 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. 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 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $ 36 */ 37 38 /* 39 * External virtual filesystem routines 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/buf.h> 45 #include <sys/conf.h> 46 #include <sys/dirent.h> 47 #include <sys/domain.h> 48 #include <sys/eventhandler.h> 49 #include <sys/fcntl.h> 50 #include <sys/kernel.h> 51 #include <sys/kthread.h> 52 #include <sys/malloc.h> 53 #include <sys/mbuf.h> 54 #include <sys/mount.h> 55 #include <sys/proc.h> 56 #include <sys/namei.h> 57 #include <sys/reboot.h> 58 #include <sys/socket.h> 59 #include <sys/stat.h> 60 #include <sys/sysctl.h> 61 #include <sys/syslog.h> 62 #include <sys/vmmeter.h> 63 #include <sys/vnode.h> 64 65 #include <machine/limits.h> 66 67 #include <vm/vm.h> 68 #include <vm/vm_object.h> 69 #include <vm/vm_extern.h> 70 #include <vm/vm_kern.h> 71 #include <vm/pmap.h> 72 #include <vm/vm_map.h> 73 #include <vm/vm_page.h> 74 #include <vm/vm_pager.h> 75 #include <vm/vnode_pager.h> 76 77 #include <sys/buf2.h> 78 #include <sys/thread2.h> 79 80 /* 81 * The workitem queue. 82 */ 83 #define SYNCER_MAXDELAY 32 84 static int sysctl_kern_syncdelay(SYSCTL_HANDLER_ARGS); 85 time_t syncdelay = 30; /* max time to delay syncing data */ 86 SYSCTL_PROC(_kern, OID_AUTO, syncdelay, CTLTYPE_INT | CTLFLAG_RW, 0, 0, 87 sysctl_kern_syncdelay, "I", "VFS data synchronization delay"); 88 time_t filedelay = 30; /* time to delay syncing files */ 89 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, 90 &filedelay, 0, "File synchronization delay"); 91 time_t dirdelay = 29; /* time to delay syncing directories */ 92 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, 93 &dirdelay, 0, "Directory synchronization delay"); 94 time_t metadelay = 28; /* time to delay syncing metadata */ 95 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, 96 &metadelay, 0, "VFS metadata synchronization delay"); 97 static int rushjob; /* number of slots to run ASAP */ 98 static int stat_rush_requests; /* number of times I/O speeded up */ 99 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, 100 &stat_rush_requests, 0, ""); 101 102 LIST_HEAD(synclist, vnode); 103 104 #define SC_FLAG_EXIT (0x1) /* request syncer exit */ 105 #define SC_FLAG_DONE (0x2) /* syncer confirm exit */ 106 107 struct syncer_ctx { 108 struct mount *sc_mp; 109 struct lwkt_token sc_token; 110 struct thread *sc_thread; 111 int sc_flags; 112 struct synclist *syncer_workitem_pending; 113 long syncer_mask; 114 int syncer_delayno; 115 int syncer_forced; 116 int syncer_rushjob; 117 }; 118 119 static void syncer_thread(void *); 120 121 static int 122 sysctl_kern_syncdelay(SYSCTL_HANDLER_ARGS) 123 { 124 int error; 125 int v = syncdelay; 126 127 error = sysctl_handle_int(oidp, &v, 0, req); 128 if (error || !req->newptr) 129 return (error); 130 if (v < 1) 131 v = 1; 132 if (v > SYNCER_MAXDELAY) 133 v = SYNCER_MAXDELAY; 134 syncdelay = v; 135 136 return(0); 137 } 138 139 /* 140 * The workitem queue. 141 * 142 * It is useful to delay writes of file data and filesystem metadata 143 * for tens of seconds so that quickly created and deleted files need 144 * not waste disk bandwidth being created and removed. To realize this, 145 * we append vnodes to a "workitem" queue. When running with a soft 146 * updates implementation, most pending metadata dependencies should 147 * not wait for more than a few seconds. Thus, mounted on block devices 148 * are delayed only about a half the time that file data is delayed. 149 * Similarly, directory updates are more critical, so are only delayed 150 * about a third the time that file data is delayed. Thus, there are 151 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 152 * one each second (driven off the filesystem syncer process). The 153 * syncer_delayno variable indicates the next queue that is to be processed. 154 * Items that need to be processed soon are placed in this queue: 155 * 156 * syncer_workitem_pending[syncer_delayno] 157 * 158 * A delay of fifteen seconds is done by placing the request fifteen 159 * entries later in the queue: 160 * 161 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 162 * 163 */ 164 165 /* 166 * Add an item to the syncer work queue. 167 * 168 * WARNING: Cannot get vp->v_token here if not already held, we must 169 * depend on the syncer_token (which might already be held by 170 * the caller) to protect v_synclist and VONWORKLST. 171 * 172 * MPSAFE 173 */ 174 void 175 vn_syncer_add(struct vnode *vp, int delay) 176 { 177 struct syncer_ctx *ctx; 178 int slot; 179 180 ctx = vp->v_mount->mnt_syncer_ctx; 181 lwkt_gettoken(&ctx->sc_token); 182 183 if (vp->v_flag & VONWORKLST) 184 LIST_REMOVE(vp, v_synclist); 185 if (delay <= 0) { 186 slot = -delay & ctx->syncer_mask; 187 } else { 188 if (delay > SYNCER_MAXDELAY - 2) 189 delay = SYNCER_MAXDELAY - 2; 190 slot = (ctx->syncer_delayno + delay) & ctx->syncer_mask; 191 } 192 193 LIST_INSERT_HEAD(&ctx->syncer_workitem_pending[slot], vp, v_synclist); 194 vsetflags(vp, VONWORKLST); 195 196 lwkt_reltoken(&ctx->sc_token); 197 } 198 199 /* 200 * Removes the vnode from the syncer list. Since we might block while 201 * acquiring the syncer_token we have to [re]check conditions to determine 202 * that it is ok to remove the vnode. 203 * 204 * vp->v_token held on call 205 */ 206 void 207 vn_syncer_remove(struct vnode *vp) 208 { 209 struct syncer_ctx *ctx; 210 211 ctx = vp->v_mount->mnt_syncer_ctx; 212 lwkt_gettoken(&ctx->sc_token); 213 214 if ((vp->v_flag & (VISDIRTY | VONWORKLST | VOBJDIRTY)) == VONWORKLST && 215 RB_EMPTY(&vp->v_rbdirty_tree)) { 216 vclrflags(vp, VONWORKLST); 217 LIST_REMOVE(vp, v_synclist); 218 } 219 220 lwkt_reltoken(&ctx->sc_token); 221 } 222 223 /* 224 * vnode must be locked 225 */ 226 void 227 vclrisdirty(struct vnode *vp) 228 { 229 vclrflags(vp, VISDIRTY); 230 if (vp->v_flag & VONWORKLST) 231 vn_syncer_remove(vp); 232 } 233 234 void 235 vclrobjdirty(struct vnode *vp) 236 { 237 vclrflags(vp, VOBJDIRTY); 238 if (vp->v_flag & VONWORKLST) 239 vn_syncer_remove(vp); 240 } 241 242 /* 243 * vnode must be stable 244 */ 245 void 246 vsetisdirty(struct vnode *vp) 247 { 248 struct syncer_ctx *ctx; 249 250 if ((vp->v_flag & VISDIRTY) == 0) { 251 ctx = vp->v_mount->mnt_syncer_ctx; 252 vsetflags(vp, VISDIRTY); 253 lwkt_gettoken(&ctx->sc_token); 254 if ((vp->v_flag & VONWORKLST) == 0) 255 vn_syncer_add(vp, syncdelay); 256 lwkt_reltoken(&ctx->sc_token); 257 } 258 } 259 260 void 261 vsetobjdirty(struct vnode *vp) 262 { 263 struct syncer_ctx *ctx; 264 265 if ((vp->v_flag & VOBJDIRTY) == 0) { 266 ctx = vp->v_mount->mnt_syncer_ctx; 267 vsetflags(vp, VOBJDIRTY); 268 lwkt_gettoken(&ctx->sc_token); 269 if ((vp->v_flag & VONWORKLST) == 0) 270 vn_syncer_add(vp, syncdelay); 271 lwkt_reltoken(&ctx->sc_token); 272 } 273 } 274 275 /* 276 * Create per-filesystem syncer process 277 */ 278 void 279 vn_syncer_thr_create(struct mount *mp) 280 { 281 struct syncer_ctx *ctx; 282 static int syncalloc = 0; 283 284 ctx = kmalloc(sizeof(struct syncer_ctx), M_TEMP, M_WAITOK | M_ZERO); 285 ctx->sc_mp = mp; 286 ctx->sc_flags = 0; 287 ctx->syncer_workitem_pending = hashinit(SYNCER_MAXDELAY, M_DEVBUF, 288 &ctx->syncer_mask); 289 ctx->syncer_delayno = 0; 290 lwkt_token_init(&ctx->sc_token, "syncer"); 291 mp->mnt_syncer_ctx = ctx; 292 kthread_create(syncer_thread, ctx, &ctx->sc_thread, 293 "syncer%d", ++syncalloc & 0x7FFFFFFF); 294 } 295 296 /* 297 * Stop per-filesystem syncer process 298 */ 299 void 300 vn_syncer_thr_stop(struct mount *mp) 301 { 302 struct syncer_ctx *ctx; 303 304 ctx = mp->mnt_syncer_ctx; 305 if (ctx == NULL) 306 return; 307 308 lwkt_gettoken(&ctx->sc_token); 309 310 /* Signal the syncer process to exit */ 311 ctx->sc_flags |= SC_FLAG_EXIT; 312 wakeup(ctx); 313 314 /* Wait till syncer process exits */ 315 while ((ctx->sc_flags & SC_FLAG_DONE) == 0) 316 tsleep(&ctx->sc_flags, 0, "syncexit", hz); 317 318 mp->mnt_syncer_ctx = NULL; 319 lwkt_reltoken(&ctx->sc_token); 320 321 hashdestroy(ctx->syncer_workitem_pending, M_DEVBUF, ctx->syncer_mask); 322 kfree(ctx, M_TEMP); 323 } 324 325 struct thread *updatethread; 326 327 /* 328 * System filesystem synchronizer daemon. 329 */ 330 static void 331 syncer_thread(void *_ctx) 332 { 333 struct syncer_ctx *ctx = _ctx; 334 struct synclist *slp; 335 struct vnode *vp; 336 long starttime; 337 int *sc_flagsp; 338 int sc_flags; 339 int vnodes_synced = 0; 340 int delta; 341 int dummy = 0; 342 343 for (;;) { 344 kproc_suspend_loop(); 345 346 starttime = time_uptime; 347 lwkt_gettoken(&ctx->sc_token); 348 349 /* 350 * Push files whose dirty time has expired. Be careful 351 * of interrupt race on slp queue. 352 */ 353 slp = &ctx->syncer_workitem_pending[ctx->syncer_delayno]; 354 ctx->syncer_delayno = (ctx->syncer_delayno + 1) & 355 ctx->syncer_mask; 356 357 while ((vp = LIST_FIRST(slp)) != NULL) { 358 if (ctx->syncer_forced) { 359 if (vget(vp, LK_EXCLUSIVE) == 0) { 360 VOP_FSYNC(vp, MNT_NOWAIT, 0); 361 vput(vp); 362 vnodes_synced++; 363 } 364 } else { 365 if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 366 VOP_FSYNC(vp, MNT_LAZY, 0); 367 vput(vp); 368 vnodes_synced++; 369 } 370 } 371 372 /* 373 * vp is stale but can still be used if we can 374 * verify that it remains at the head of the list. 375 * Be careful not to try to get vp->v_token as 376 * vp can become stale if this blocks. 377 * 378 * If the vp is still at the head of the list were 379 * unable to completely flush it and move it to 380 * a later slot to give other vnodes a fair shot. 381 * 382 * Note that v_tag VT_VFS vnodes can remain on the 383 * worklist with no dirty blocks, but sync_fsync() 384 * moves it to a later slot so we will never see it 385 * here. 386 * 387 * It is possible to race a vnode with no dirty 388 * buffers being removed from the list. If this 389 * occurs we will move the vnode in the synclist 390 * and then the other thread will remove it. Do 391 * not try to remove it here. 392 */ 393 if (LIST_FIRST(slp) == vp) 394 vn_syncer_add(vp, syncdelay); 395 } 396 397 sc_flags = ctx->sc_flags; 398 399 /* Exit on unmount */ 400 if (sc_flags & SC_FLAG_EXIT) 401 break; 402 403 lwkt_reltoken(&ctx->sc_token); 404 405 /* 406 * Do sync processing for each mount. 407 */ 408 if (ctx->sc_mp) 409 bio_ops_sync(ctx->sc_mp); 410 411 /* 412 * The variable rushjob allows the kernel to speed up the 413 * processing of the filesystem syncer process. A rushjob 414 * value of N tells the filesystem syncer to process the next 415 * N seconds worth of work on its queue ASAP. Currently rushjob 416 * is used by the soft update code to speed up the filesystem 417 * syncer process when the incore state is getting so far 418 * ahead of the disk that the kernel memory pool is being 419 * threatened with exhaustion. 420 */ 421 delta = rushjob - ctx->syncer_rushjob; 422 if ((u_int)delta > syncdelay / 2) { 423 ctx->syncer_rushjob = rushjob - syncdelay / 2; 424 tsleep(&dummy, 0, "rush", 1); 425 continue; 426 } 427 if (delta) { 428 ++ctx->syncer_rushjob; 429 tsleep(&dummy, 0, "rush", 1); 430 continue; 431 } 432 433 /* 434 * If it has taken us less than a second to process the 435 * current work, then wait. Otherwise start right over 436 * again. We can still lose time if any single round 437 * takes more than two seconds, but it does not really 438 * matter as we are just trying to generally pace the 439 * filesystem activity. 440 */ 441 if (time_uptime == starttime) 442 tsleep(ctx, 0, "syncer", hz); 443 } 444 445 /* 446 * Unmount/exit path for per-filesystem syncers; sc_token held 447 */ 448 ctx->sc_flags |= SC_FLAG_DONE; 449 sc_flagsp = &ctx->sc_flags; 450 lwkt_reltoken(&ctx->sc_token); 451 wakeup(sc_flagsp); 452 453 kthread_exit(); 454 } 455 456 /* 457 * Request that the syncer daemon for a specific mount speed up its work. 458 * If mp is NULL the caller generally wants to speed up all syncers. 459 */ 460 void 461 speedup_syncer(struct mount *mp) 462 { 463 /* 464 * Don't bother protecting the test. unsleep_and_wakeup_thread() 465 * will only do something real if the thread is in the right state. 466 */ 467 atomic_add_int(&rushjob, 1); 468 ++stat_rush_requests; 469 if (mp) 470 wakeup(mp->mnt_syncer_ctx); 471 } 472 473 /* 474 * Routine to create and manage a filesystem syncer vnode. 475 */ 476 static int sync_close(struct vop_close_args *); 477 static int sync_fsync(struct vop_fsync_args *); 478 static int sync_inactive(struct vop_inactive_args *); 479 static int sync_reclaim (struct vop_reclaim_args *); 480 static int sync_print(struct vop_print_args *); 481 482 static struct vop_ops sync_vnode_vops = { 483 .vop_default = vop_eopnotsupp, 484 .vop_close = sync_close, 485 .vop_fsync = sync_fsync, 486 .vop_inactive = sync_inactive, 487 .vop_reclaim = sync_reclaim, 488 .vop_print = sync_print, 489 }; 490 491 static struct vop_ops *sync_vnode_vops_p = &sync_vnode_vops; 492 493 VNODEOP_SET(sync_vnode_vops); 494 495 /* 496 * Create a new filesystem syncer vnode for the specified mount point. 497 * This vnode is placed on the worklist and is responsible for sync'ing 498 * the filesystem. 499 * 500 * NOTE: read-only mounts are also placed on the worklist. The filesystem 501 * sync code is also responsible for cleaning up vnodes. 502 */ 503 int 504 vfs_allocate_syncvnode(struct mount *mp) 505 { 506 struct vnode *vp; 507 static long start, incr, next; 508 int error; 509 510 /* Allocate a new vnode */ 511 error = getspecialvnode(VT_VFS, mp, &sync_vnode_vops_p, &vp, 0, 0); 512 if (error) { 513 mp->mnt_syncer = NULL; 514 return (error); 515 } 516 vp->v_type = VNON; 517 /* 518 * Place the vnode onto the syncer worklist. We attempt to 519 * scatter them about on the list so that they will go off 520 * at evenly distributed times even if all the filesystems 521 * are mounted at once. 522 */ 523 next += incr; 524 if (next == 0 || next > SYNCER_MAXDELAY) { 525 start /= 2; 526 incr /= 2; 527 if (start == 0) { 528 start = SYNCER_MAXDELAY / 2; 529 incr = SYNCER_MAXDELAY; 530 } 531 next = start; 532 } 533 534 /* 535 * Only put the syncer vnode onto the syncer list if we have a 536 * syncer thread. Some VFS's (aka NULLFS) don't need a syncer 537 * thread. 538 */ 539 if (mp->mnt_syncer_ctx) 540 vn_syncer_add(vp, syncdelay > 0 ? next % syncdelay : 0); 541 542 /* 543 * The mnt_syncer field inherits the vnode reference, which is 544 * held until later decomissioning. 545 */ 546 mp->mnt_syncer = vp; 547 vx_unlock(vp); 548 return (0); 549 } 550 551 static int 552 sync_close(struct vop_close_args *ap) 553 { 554 return (0); 555 } 556 557 /* 558 * Do a lazy sync of the filesystem. 559 * 560 * sync_fsync { struct vnode *a_vp, int a_waitfor } 561 */ 562 static int 563 sync_fsync(struct vop_fsync_args *ap) 564 { 565 struct vnode *syncvp = ap->a_vp; 566 struct mount *mp = syncvp->v_mount; 567 int asyncflag; 568 569 /* 570 * We only need to do something if this is a lazy evaluation. 571 */ 572 if ((ap->a_waitfor & MNT_LAZY) == 0) 573 return (0); 574 575 /* 576 * Move ourselves to the back of the sync list. 577 */ 578 vn_syncer_add(syncvp, syncdelay); 579 580 /* 581 * Walk the list of vnodes pushing all that are dirty and 582 * not already on the sync list, and freeing vnodes which have 583 * no refs and whos VM objects are empty. vfs_msync() handles 584 * the VM issues and must be called whether the mount is readonly 585 * or not. 586 */ 587 if (vfs_busy(mp, LK_NOWAIT) != 0) 588 return (0); 589 if (mp->mnt_flag & MNT_RDONLY) { 590 vfs_msync(mp, MNT_NOWAIT); 591 } else { 592 asyncflag = mp->mnt_flag & MNT_ASYNC; 593 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */ 594 vfs_msync(mp, MNT_NOWAIT); 595 VFS_SYNC(mp, MNT_NOWAIT | MNT_LAZY); 596 if (asyncflag) 597 mp->mnt_flag |= MNT_ASYNC; 598 } 599 vfs_unbusy(mp); 600 return (0); 601 } 602 603 /* 604 * The syncer vnode is no longer referenced. 605 * 606 * sync_inactive { struct vnode *a_vp, struct proc *a_p } 607 */ 608 static int 609 sync_inactive(struct vop_inactive_args *ap) 610 { 611 vgone_vxlocked(ap->a_vp); 612 return (0); 613 } 614 615 /* 616 * The syncer vnode is no longer needed and is being decommissioned. 617 * This can only occur when the last reference has been released on 618 * mp->mnt_syncer, so mp->mnt_syncer had better be NULL. 619 * 620 * Modifications to the worklist must be protected with a critical 621 * section. 622 * 623 * sync_reclaim { struct vnode *a_vp } 624 */ 625 static int 626 sync_reclaim(struct vop_reclaim_args *ap) 627 { 628 struct vnode *vp = ap->a_vp; 629 struct syncer_ctx *ctx; 630 631 ctx = vp->v_mount->mnt_syncer_ctx; 632 if (ctx) { 633 lwkt_gettoken(&ctx->sc_token); 634 KKASSERT(vp->v_mount->mnt_syncer != vp); 635 if (vp->v_flag & VONWORKLST) { 636 LIST_REMOVE(vp, v_synclist); 637 vclrflags(vp, VONWORKLST); 638 } 639 lwkt_reltoken(&ctx->sc_token); 640 } else { 641 KKASSERT((vp->v_flag & VONWORKLST) == 0); 642 } 643 644 return (0); 645 } 646 647 /* 648 * This is very similar to vmntvnodescan() but it only scans the 649 * vnodes on the syncer list. VFS's which support faster VFS_SYNC 650 * operations use the VISDIRTY flag on the vnode to ensure that vnodes 651 * with dirty inodes are added to the syncer in addition to vnodes 652 * with dirty buffers, and can use this function instead of nmntvnodescan(). 653 * 654 * This is important when a system has millions of vnodes. 655 */ 656 int 657 vsyncscan( 658 struct mount *mp, 659 int vmsc_flags, 660 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 661 void *data 662 ) { 663 struct syncer_ctx *ctx; 664 struct synclist *slp; 665 struct vnode *vp; 666 int i; 667 int count; 668 int lkflags; 669 670 if (vmsc_flags & VMSC_NOWAIT) 671 lkflags = LK_NOWAIT; 672 else 673 lkflags = 0; 674 675 /* 676 * Syncer list context. This API requires a dedicated syncer thread. 677 * (MNTK_THR_SYNC). 678 */ 679 KKASSERT(mp->mnt_kern_flag & MNTK_THR_SYNC); 680 ctx = mp->mnt_syncer_ctx; 681 lwkt_gettoken(&ctx->sc_token); 682 683 /* 684 * Setup for loop. Allow races against the syncer thread but 685 * require that the syncer thread no be lazy if we were told 686 * not to be lazy. 687 */ 688 i = ctx->syncer_delayno & ctx->syncer_mask; 689 if ((vmsc_flags & VMSC_NOWAIT) == 0) 690 ++ctx->syncer_forced; 691 for (count = 0; count <= ctx->syncer_mask; ++count) { 692 slp = &ctx->syncer_workitem_pending[i]; 693 694 while ((vp = LIST_FIRST(slp)) != NULL) { 695 KKASSERT(vp->v_mount == mp); 696 if (vmsc_flags & VMSC_GETVP) { 697 if (vget(vp, LK_EXCLUSIVE | lkflags) == 0) { 698 slowfunc(mp, vp, data); 699 vput(vp); 700 } 701 } else if (vmsc_flags & VMSC_GETVX) { 702 vx_get(vp); 703 slowfunc(mp, vp, data); 704 vx_put(vp); 705 } else { 706 vhold(vp); 707 slowfunc(mp, vp, data); 708 vdrop(vp); 709 } 710 711 /* 712 * vp could be invalid. However, if vp is still at 713 * the head of the list it is clearly valid and we 714 * can safely move it. 715 */ 716 if (LIST_FIRST(slp) == vp) 717 vn_syncer_add(vp, -(i + syncdelay)); 718 } 719 i = (i + 1) & ctx->syncer_mask; 720 } 721 722 if ((vmsc_flags & VMSC_NOWAIT) == 0) 723 --ctx->syncer_forced; 724 lwkt_reltoken(&ctx->sc_token); 725 return(0); 726 } 727 728 /* 729 * Print out a syncer vnode. 730 * 731 * sync_print { struct vnode *a_vp } 732 */ 733 static int 734 sync_print(struct vop_print_args *ap) 735 { 736 struct vnode *vp = ap->a_vp; 737 738 kprintf("syncer vnode"); 739 lockmgr_printinfo(&vp->v_lock); 740 kprintf("\n"); 741 return (0); 742 } 743 744