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 #define SC_FLAG_BIOOPS_ALL (0x4) /* do bufops_sync(NULL) */ 107 108 struct syncer_ctx { 109 struct mount *sc_mp; 110 struct lwkt_token sc_token; 111 struct thread *sc_thread; 112 int sc_flags; 113 114 struct synclist *syncer_workitem_pending; 115 long syncer_mask; 116 int syncer_delayno; 117 int syncer_forced; 118 }; 119 120 static struct syncer_ctx syncer_ctx0; 121 122 static void syncer_thread(void *); 123 124 static void 125 syncer_ctx_init(struct syncer_ctx *ctx, struct mount *mp) 126 { 127 ctx->sc_mp = mp; 128 ctx->sc_flags = 0; 129 ctx->syncer_workitem_pending = hashinit(SYNCER_MAXDELAY, M_DEVBUF, 130 &ctx->syncer_mask); 131 ctx->syncer_delayno = 0; 132 lwkt_token_init(&ctx->sc_token, "syncer"); 133 } 134 135 /* 136 * Called from vfsinit() 137 */ 138 void 139 vfs_sync_init(void) 140 { 141 syncer_ctx_init(&syncer_ctx0, NULL); 142 syncer_ctx0.sc_flags |= SC_FLAG_BIOOPS_ALL; 143 144 /* Support schedcpu wakeup of syncer0 */ 145 lbolt_syncer = &syncer_ctx0; 146 } 147 148 static int 149 sysctl_kern_syncdelay(SYSCTL_HANDLER_ARGS) 150 { 151 int error; 152 int v = syncdelay; 153 154 error = sysctl_handle_int(oidp, &v, 0, req); 155 if (error || !req->newptr) 156 return (error); 157 if (v < 1) 158 v = 1; 159 if (v > SYNCER_MAXDELAY) 160 v = SYNCER_MAXDELAY; 161 syncdelay = v; 162 163 return(0); 164 } 165 166 static struct syncer_ctx * 167 vn_get_syncer(struct vnode *vp) 168 { 169 struct mount *mp; 170 struct syncer_ctx *ctx; 171 172 if ((mp = vp->v_mount) != NULL) 173 ctx = mp->mnt_syncer_ctx; 174 else 175 ctx = &syncer_ctx0; 176 return (ctx); 177 } 178 179 /* 180 * The workitem queue. 181 * 182 * It is useful to delay writes of file data and filesystem metadata 183 * for tens of seconds so that quickly created and deleted files need 184 * not waste disk bandwidth being created and removed. To realize this, 185 * we append vnodes to a "workitem" queue. When running with a soft 186 * updates implementation, most pending metadata dependencies should 187 * not wait for more than a few seconds. Thus, mounted on block devices 188 * are delayed only about a half the time that file data is delayed. 189 * Similarly, directory updates are more critical, so are only delayed 190 * about a third the time that file data is delayed. Thus, there are 191 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 192 * one each second (driven off the filesystem syncer process). The 193 * syncer_delayno variable indicates the next queue that is to be processed. 194 * Items that need to be processed soon are placed in this queue: 195 * 196 * syncer_workitem_pending[syncer_delayno] 197 * 198 * A delay of fifteen seconds is done by placing the request fifteen 199 * entries later in the queue: 200 * 201 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 202 * 203 */ 204 205 /* 206 * Add an item to the syncer work queue. 207 * 208 * WARNING: Cannot get vp->v_token here if not already held, we must 209 * depend on the syncer_token (which might already be held by 210 * the caller) to protect v_synclist and VONWORKLST. 211 * 212 * MPSAFE 213 */ 214 void 215 vn_syncer_add(struct vnode *vp, int delay) 216 { 217 struct syncer_ctx *ctx; 218 int slot; 219 220 ctx = vn_get_syncer(vp); 221 222 lwkt_gettoken(&ctx->sc_token); 223 224 if (vp->v_flag & VONWORKLST) 225 LIST_REMOVE(vp, v_synclist); 226 if (delay <= 0) { 227 slot = -delay & ctx->syncer_mask; 228 } else { 229 if (delay > SYNCER_MAXDELAY - 2) 230 delay = SYNCER_MAXDELAY - 2; 231 slot = (ctx->syncer_delayno + delay) & ctx->syncer_mask; 232 } 233 234 LIST_INSERT_HEAD(&ctx->syncer_workitem_pending[slot], vp, v_synclist); 235 vsetflags(vp, VONWORKLST); 236 237 lwkt_reltoken(&ctx->sc_token); 238 } 239 240 /* 241 * Removes the vnode from the syncer list. Since we might block while 242 * acquiring the syncer_token we have to recheck conditions. 243 * 244 * vp->v_token held on call 245 */ 246 void 247 vn_syncer_remove(struct vnode *vp) 248 { 249 struct syncer_ctx *ctx; 250 251 ctx = vn_get_syncer(vp); 252 253 lwkt_gettoken(&ctx->sc_token); 254 255 if ((vp->v_flag & (VISDIRTY | VONWORKLST)) == VONWORKLST && 256 RB_EMPTY(&vp->v_rbdirty_tree)) { 257 vclrflags(vp, VONWORKLST); 258 LIST_REMOVE(vp, v_synclist); 259 } 260 261 lwkt_reltoken(&ctx->sc_token); 262 } 263 264 /* 265 * vnode must be locked 266 */ 267 void 268 vclrisdirty(struct vnode *vp) 269 { 270 vclrflags(vp, VISDIRTY); 271 if (vp->v_flag & VONWORKLST) 272 vn_syncer_remove(vp); 273 } 274 275 /* 276 * vnode must be stable 277 */ 278 void 279 vsetisdirty(struct vnode *vp) 280 { 281 if ((vp->v_flag & VISDIRTY) == 0) { 282 vsetflags(vp, VISDIRTY); 283 vn_syncer_add(vp, syncdelay); 284 } 285 } 286 287 /* 288 * Create per-filesystem syncer process 289 */ 290 void 291 vn_syncer_thr_create(struct mount *mp) 292 { 293 struct syncer_ctx *ctx; 294 static int syncalloc = 0; 295 int rc; 296 297 if (mp->mnt_kern_flag & MNTK_THR_SYNC) { 298 ctx = kmalloc(sizeof(struct syncer_ctx), M_TEMP, 299 M_WAITOK | M_ZERO); 300 syncer_ctx_init(ctx, mp); 301 mp->mnt_syncer_ctx = ctx; 302 rc = kthread_create(syncer_thread, ctx, &ctx->sc_thread, 303 "syncer%d", ++syncalloc); 304 } else { 305 mp->mnt_syncer_ctx = &syncer_ctx0; 306 } 307 } 308 309 /* 310 * Stop per-filesystem syncer process 311 */ 312 void 313 vn_syncer_thr_stop(struct mount *mp) 314 { 315 struct syncer_ctx *ctx; 316 317 ctx = mp->mnt_syncer_ctx; 318 if (ctx == NULL || ctx == &syncer_ctx0) 319 return; 320 KKASSERT(mp->mnt_kern_flag & MNTK_THR_SYNC); 321 322 lwkt_gettoken(&ctx->sc_token); 323 324 /* Signal the syncer process to exit */ 325 ctx->sc_flags |= SC_FLAG_EXIT; 326 wakeup(ctx); 327 328 /* Wait till syncer process exits */ 329 while ((ctx->sc_flags & SC_FLAG_DONE) == 0) 330 tsleep(&ctx->sc_flags, 0, "syncexit", hz); 331 332 mp->mnt_syncer_ctx = NULL; 333 lwkt_reltoken(&ctx->sc_token); 334 335 hashdestroy(ctx->syncer_workitem_pending, M_DEVBUF, ctx->syncer_mask); 336 kfree(ctx, M_TEMP); 337 } 338 339 struct thread *updatethread; 340 341 /* 342 * System filesystem synchronizer daemon. 343 */ 344 static void 345 syncer_thread(void *_ctx) 346 { 347 struct thread *td = curthread; 348 struct syncer_ctx *ctx = _ctx; 349 struct synclist *slp; 350 struct vnode *vp; 351 long starttime; 352 int *sc_flagsp; 353 int sc_flags; 354 int vnodes_synced = 0; 355 356 /* 357 * syncer0 runs till system shutdown; per-filesystem syncers are 358 * terminated on filesystem unmount 359 */ 360 if (ctx == &syncer_ctx0) 361 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 362 SHUTDOWN_PRI_LAST); 363 for (;;) { 364 kproc_suspend_loop(); 365 366 starttime = time_uptime; 367 lwkt_gettoken(&ctx->sc_token); 368 369 /* 370 * Push files whose dirty time has expired. Be careful 371 * of interrupt race on slp queue. 372 */ 373 slp = &ctx->syncer_workitem_pending[ctx->syncer_delayno]; 374 ctx->syncer_delayno = (ctx->syncer_delayno + 1) & 375 ctx->syncer_mask; 376 377 while ((vp = LIST_FIRST(slp)) != NULL) { 378 if (ctx->syncer_forced) { 379 if (vget(vp, LK_EXCLUSIVE) == 0) { 380 VOP_FSYNC(vp, MNT_NOWAIT, 0); 381 vput(vp); 382 vnodes_synced++; 383 } 384 } else { 385 if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 386 VOP_FSYNC(vp, MNT_LAZY, 0); 387 vput(vp); 388 vnodes_synced++; 389 } 390 } 391 392 /* 393 * vp is stale but can still be used if we can 394 * verify that it remains at the head of the list. 395 * Be careful not to try to get vp->v_token as 396 * vp can become stale if this blocks. 397 * 398 * If the vp is still at the head of the list were 399 * unable to completely flush it and move it to 400 * a later slot to give other vnodes a fair shot. 401 * 402 * Note that v_tag VT_VFS vnodes can remain on the 403 * worklist with no dirty blocks, but sync_fsync() 404 * moves it to a later slot so we will never see it 405 * here. 406 * 407 * It is possible to race a vnode with no dirty 408 * buffers being removed from the list. If this 409 * occurs we will move the vnode in the synclist 410 * and then the other thread will remove it. Do 411 * not try to remove it here. 412 */ 413 if (LIST_FIRST(slp) == vp) 414 vn_syncer_add(vp, syncdelay); 415 } 416 417 sc_flags = ctx->sc_flags; 418 419 /* Exit on unmount */ 420 if (sc_flags & SC_FLAG_EXIT) 421 break; 422 423 lwkt_reltoken(&ctx->sc_token); 424 425 /* 426 * Do sync processing for each mount. 427 */ 428 if (ctx->sc_mp || sc_flags & SC_FLAG_BIOOPS_ALL) 429 bio_ops_sync(ctx->sc_mp); 430 431 /* 432 * The variable rushjob allows the kernel to speed up the 433 * processing of the filesystem syncer process. A rushjob 434 * value of N tells the filesystem syncer to process the next 435 * N seconds worth of work on its queue ASAP. Currently rushjob 436 * is used by the soft update code to speed up the filesystem 437 * syncer process when the incore state is getting so far 438 * ahead of the disk that the kernel memory pool is being 439 * threatened with exhaustion. 440 */ 441 if (ctx == &syncer_ctx0 && rushjob > 0) { 442 atomic_subtract_int(&rushjob, 1); 443 continue; 444 } 445 /* 446 * If it has taken us less than a second to process the 447 * current work, then wait. Otherwise start right over 448 * again. We can still lose time if any single round 449 * takes more than two seconds, but it does not really 450 * matter as we are just trying to generally pace the 451 * filesystem activity. 452 */ 453 if (time_uptime == starttime) 454 tsleep(ctx, 0, "syncer", hz); 455 } 456 457 /* 458 * Unmount/exit path for per-filesystem syncers; sc_token held 459 */ 460 ctx->sc_flags |= SC_FLAG_DONE; 461 sc_flagsp = &ctx->sc_flags; 462 lwkt_reltoken(&ctx->sc_token); 463 wakeup(sc_flagsp); 464 465 kthread_exit(); 466 } 467 468 static void 469 syncer_thread_start(void) 470 { 471 syncer_thread(&syncer_ctx0); 472 } 473 474 static struct kproc_desc up_kp = { 475 "syncer0", 476 syncer_thread_start, 477 &updatethread 478 }; 479 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp) 480 481 /* 482 * Request the syncer daemon to speed up its work. 483 * We never push it to speed up more than half of its 484 * normal turn time, otherwise it could take over the cpu. 485 */ 486 int 487 speedup_syncer(void) 488 { 489 /* 490 * Don't bother protecting the test. unsleep_and_wakeup_thread() 491 * will only do something real if the thread is in the right state. 492 */ 493 wakeup(lbolt_syncer); 494 if (rushjob < syncdelay / 2) { 495 atomic_add_int(&rushjob, 1); 496 stat_rush_requests += 1; 497 return (1); 498 } 499 return(0); 500 } 501 502 /* 503 * Routine to create and manage a filesystem syncer vnode. 504 */ 505 static int sync_close(struct vop_close_args *); 506 static int sync_fsync(struct vop_fsync_args *); 507 static int sync_inactive(struct vop_inactive_args *); 508 static int sync_reclaim (struct vop_reclaim_args *); 509 static int sync_print(struct vop_print_args *); 510 511 static struct vop_ops sync_vnode_vops = { 512 .vop_default = vop_eopnotsupp, 513 .vop_close = sync_close, 514 .vop_fsync = sync_fsync, 515 .vop_inactive = sync_inactive, 516 .vop_reclaim = sync_reclaim, 517 .vop_print = sync_print, 518 }; 519 520 static struct vop_ops *sync_vnode_vops_p = &sync_vnode_vops; 521 522 VNODEOP_SET(sync_vnode_vops); 523 524 /* 525 * Create a new filesystem syncer vnode for the specified mount point. 526 * This vnode is placed on the worklist and is responsible for sync'ing 527 * the filesystem. 528 * 529 * NOTE: read-only mounts are also placed on the worklist. The filesystem 530 * sync code is also responsible for cleaning up vnodes. 531 */ 532 int 533 vfs_allocate_syncvnode(struct mount *mp) 534 { 535 struct vnode *vp; 536 static long start, incr, next; 537 int error; 538 539 /* Allocate a new vnode */ 540 error = getspecialvnode(VT_VFS, mp, &sync_vnode_vops_p, &vp, 0, 0); 541 if (error) { 542 mp->mnt_syncer = NULL; 543 return (error); 544 } 545 vp->v_type = VNON; 546 /* 547 * Place the vnode onto the syncer worklist. We attempt to 548 * scatter them about on the list so that they will go off 549 * at evenly distributed times even if all the filesystems 550 * are mounted at once. 551 */ 552 next += incr; 553 if (next == 0 || next > SYNCER_MAXDELAY) { 554 start /= 2; 555 incr /= 2; 556 if (start == 0) { 557 start = SYNCER_MAXDELAY / 2; 558 incr = SYNCER_MAXDELAY; 559 } 560 next = start; 561 } 562 vn_syncer_add(vp, syncdelay > 0 ? next % syncdelay : 0); 563 564 /* 565 * The mnt_syncer field inherits the vnode reference, which is 566 * held until later decomissioning. 567 */ 568 mp->mnt_syncer = vp; 569 vx_unlock(vp); 570 return (0); 571 } 572 573 static int 574 sync_close(struct vop_close_args *ap) 575 { 576 return (0); 577 } 578 579 /* 580 * Do a lazy sync of the filesystem. 581 * 582 * sync_fsync { struct vnode *a_vp, int a_waitfor } 583 */ 584 static int 585 sync_fsync(struct vop_fsync_args *ap) 586 { 587 struct vnode *syncvp = ap->a_vp; 588 struct mount *mp = syncvp->v_mount; 589 int asyncflag; 590 591 /* 592 * We only need to do something if this is a lazy evaluation. 593 */ 594 if ((ap->a_waitfor & MNT_LAZY) == 0) 595 return (0); 596 597 /* 598 * Move ourselves to the back of the sync list. 599 */ 600 vn_syncer_add(syncvp, syncdelay); 601 602 /* 603 * Walk the list of vnodes pushing all that are dirty and 604 * not already on the sync list, and freeing vnodes which have 605 * no refs and whos VM objects are empty. vfs_msync() handles 606 * the VM issues and must be called whether the mount is readonly 607 * or not. 608 */ 609 if (vfs_busy(mp, LK_NOWAIT) != 0) 610 return (0); 611 if (mp->mnt_flag & MNT_RDONLY) { 612 vfs_msync(mp, MNT_NOWAIT); 613 } else { 614 asyncflag = mp->mnt_flag & MNT_ASYNC; 615 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */ 616 vfs_msync(mp, MNT_NOWAIT); 617 VFS_SYNC(mp, MNT_NOWAIT | MNT_LAZY); 618 if (asyncflag) 619 mp->mnt_flag |= MNT_ASYNC; 620 } 621 vfs_unbusy(mp); 622 return (0); 623 } 624 625 /* 626 * The syncer vnode is no longer referenced. 627 * 628 * sync_inactive { struct vnode *a_vp, struct proc *a_p } 629 */ 630 static int 631 sync_inactive(struct vop_inactive_args *ap) 632 { 633 vgone_vxlocked(ap->a_vp); 634 return (0); 635 } 636 637 /* 638 * The syncer vnode is no longer needed and is being decommissioned. 639 * This can only occur when the last reference has been released on 640 * mp->mnt_syncer, so mp->mnt_syncer had better be NULL. 641 * 642 * Modifications to the worklist must be protected with a critical 643 * section. 644 * 645 * sync_reclaim { struct vnode *a_vp } 646 */ 647 static int 648 sync_reclaim(struct vop_reclaim_args *ap) 649 { 650 struct vnode *vp = ap->a_vp; 651 struct syncer_ctx *ctx; 652 653 ctx = vn_get_syncer(vp); 654 655 lwkt_gettoken(&ctx->sc_token); 656 KKASSERT(vp->v_mount->mnt_syncer != vp); 657 if (vp->v_flag & VONWORKLST) { 658 LIST_REMOVE(vp, v_synclist); 659 vclrflags(vp, VONWORKLST); 660 } 661 lwkt_reltoken(&ctx->sc_token); 662 663 return (0); 664 } 665 666 /* 667 * This is very similar to vmntvnodescan() but it only scans the 668 * vnodes on the syncer list. VFS's which support faster VFS_SYNC 669 * operations use the VISDIRTY flag on the vnode to ensure that vnodes 670 * with dirty inodes are added to the syncer in addition to vnodes 671 * with dirty buffers, and can use this function instead of nmntvnodescan(). 672 * 673 * This is important when a system has millions of vnodes. 674 */ 675 int 676 vsyncscan( 677 struct mount *mp, 678 int vmsc_flags, 679 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 680 void *data 681 ) { 682 struct syncer_ctx *ctx; 683 struct synclist *slp; 684 struct vnode *vp; 685 int b; 686 int i; 687 int lkflags; 688 689 if (vmsc_flags & VMSC_NOWAIT) 690 lkflags = LK_NOWAIT; 691 else 692 lkflags = 0; 693 694 /* 695 * Syncer list context. This API requires a dedicated syncer thread. 696 * (MNTK_THR_SYNC). 697 */ 698 KKASSERT(mp->mnt_kern_flag & MNTK_THR_SYNC); 699 ctx = mp->mnt_syncer_ctx; 700 KKASSERT(ctx != &syncer_ctx0); 701 702 lwkt_gettoken(&ctx->sc_token); 703 704 /* 705 * Setup for loop. Allow races against the syncer thread but 706 * require that the syncer thread no be lazy if we were told 707 * not to be lazy. 708 */ 709 b = ctx->syncer_delayno & ctx->syncer_mask; 710 i = b; 711 if ((vmsc_flags & VMSC_NOWAIT) == 0) 712 ++ctx->syncer_forced; 713 714 do { 715 slp = &ctx->syncer_workitem_pending[i]; 716 717 while ((vp = LIST_FIRST(slp)) != NULL) { 718 KKASSERT(vp->v_mount == mp); 719 if (vget(vp, LK_EXCLUSIVE | lkflags) == 0) { 720 slowfunc(mp, vp, data); 721 vput(vp); 722 } 723 if (LIST_FIRST(slp) == vp) 724 vn_syncer_add(vp, -(i + syncdelay)); 725 } 726 i = (i + 1) & ctx->syncer_mask; 727 } while (i != b); 728 729 if ((vmsc_flags & VMSC_NOWAIT) == 0) 730 --ctx->syncer_forced; 731 lwkt_reltoken(&ctx->sc_token); 732 return(0); 733 } 734 735 /* 736 * Print out a syncer vnode. 737 * 738 * sync_print { struct vnode *a_vp } 739 */ 740 static int 741 sync_print(struct vop_print_args *ap) 742 { 743 struct vnode *vp = ap->a_vp; 744 745 kprintf("syncer vnode"); 746 lockmgr_printinfo(&vp->v_lock); 747 kprintf("\n"); 748 return (0); 749 } 750 751