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.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
39 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
40 * $DragonFly: src/sys/kern/vfs_sync.c,v 1.7 2005/09/17 07:43:00 dillon Exp $
44 * External virtual filesystem routines
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/domain.h>
54 #include <sys/eventhandler.h>
55 #include <sys/fcntl.h>
56 #include <sys/kernel.h>
57 #include <sys/kthread.h>
58 #include <sys/malloc.h>
60 #include <sys/mount.h>
62 #include <sys/namei.h>
63 #include <sys/reboot.h>
64 #include <sys/socket.h>
66 #include <sys/sysctl.h>
67 #include <sys/syslog.h>
68 #include <sys/vmmeter.h>
69 #include <sys/vnode.h>
71 #include <machine/limits.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_kern.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_pager.h>
81 #include <vm/vnode_pager.h>
84 #include <sys/thread2.h>
89 #define SYNCER_MAXDELAY 32
90 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
91 time_t syncdelay = 30; /* max time to delay syncing data */
92 SYSCTL_INT(_kern, OID_AUTO, syncdelay, CTLFLAG_RW,
93 &syncdelay, 0, "VFS data synchronization delay");
94 time_t filedelay = 30; /* time to delay syncing files */
95 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW,
96 &filedelay, 0, "File synchronization delay");
97 time_t dirdelay = 29; /* time to delay syncing directories */
98 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW,
99 &dirdelay, 0, "Directory synchronization delay");
100 time_t metadelay = 28; /* time to delay syncing metadata */
101 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW,
102 &metadelay, 0, "VFS metadata synchronization delay");
103 static int rushjob; /* number of slots to run ASAP */
104 static int stat_rush_requests; /* number of times I/O speeded up */
105 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW,
106 &stat_rush_requests, 0, "");
108 static int syncer_delayno = 0;
109 static long syncer_mask;
110 LIST_HEAD(synclist, vnode);
111 static struct synclist *syncer_workitem_pending;
114 * Called from vfsinit()
119 syncer_workitem_pending = hashinit(syncer_maxdelay, M_DEVBUF,
121 syncer_maxdelay = syncer_mask + 1;
125 * The workitem queue.
127 * It is useful to delay writes of file data and filesystem metadata
128 * for tens of seconds so that quickly created and deleted files need
129 * not waste disk bandwidth being created and removed. To realize this,
130 * we append vnodes to a "workitem" queue. When running with a soft
131 * updates implementation, most pending metadata dependencies should
132 * not wait for more than a few seconds. Thus, mounted on block devices
133 * are delayed only about a half the time that file data is delayed.
134 * Similarly, directory updates are more critical, so are only delayed
135 * about a third the time that file data is delayed. Thus, there are
136 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
137 * one each second (driven off the filesystem syncer process). The
138 * syncer_delayno variable indicates the next queue that is to be processed.
139 * Items that need to be processed soon are placed in this queue:
141 * syncer_workitem_pending[syncer_delayno]
143 * A delay of fifteen seconds is done by placing the request fifteen
144 * entries later in the queue:
146 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
151 * Add an item to the syncer work queue.
154 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
160 if (vp->v_flag & VONWORKLST) {
161 LIST_REMOVE(vp, v_synclist);
164 if (delay > syncer_maxdelay - 2)
165 delay = syncer_maxdelay - 2;
166 slot = (syncer_delayno + delay) & syncer_mask;
168 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
169 vp->v_flag |= VONWORKLST;
173 struct thread *updatethread;
174 static void sched_sync (void);
175 static struct kproc_desc up_kp = {
180 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
183 * System filesystem synchronizer daemon.
188 struct synclist *slp;
191 struct thread *td = curthread;
193 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
197 kproc_suspend_loop();
199 starttime = time_second;
202 * Push files whose dirty time has expired. Be careful
203 * of interrupt race on slp queue.
206 slp = &syncer_workitem_pending[syncer_delayno];
208 if (syncer_delayno == syncer_maxdelay)
212 while ((vp = LIST_FIRST(slp)) != NULL) {
213 if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT, td) == 0) {
214 VOP_FSYNC(vp, MNT_LAZY, td);
220 * If the vnode is still at the head of the list
221 * we were not able to completely flush it. To
222 * give other vnodes a fair shake we move it to
225 * Note that v_tag VT_VFS vnodes can remain on the
226 * worklist with no dirty blocks, but sync_fsync()
227 * moves it to a later slot so we will never see it
230 if (LIST_FIRST(slp) == vp) {
231 if (RB_EMPTY(&vp->v_rbdirty_tree) &&
232 !vn_isdisk(vp, NULL)) {
233 panic("sched_sync: fsync failed vp %p tag %d", vp, vp->v_tag);
235 vn_syncer_add_to_worklist(vp, syncdelay);
241 * Do soft update processing.
244 (*bioops.io_sync)(NULL);
247 * The variable rushjob allows the kernel to speed up the
248 * processing of the filesystem syncer process. A rushjob
249 * value of N tells the filesystem syncer to process the next
250 * N seconds worth of work on its queue ASAP. Currently rushjob
251 * is used by the soft update code to speed up the filesystem
252 * syncer process when the incore state is getting so far
253 * ahead of the disk that the kernel memory pool is being
254 * threatened with exhaustion.
261 * If it has taken us less than a second to process the
262 * current work, then wait. Otherwise start right over
263 * again. We can still lose time if any single round
264 * takes more than two seconds, but it does not really
265 * matter as we are just trying to generally pace the
266 * filesystem activity.
268 if (time_second == starttime)
269 tsleep(&lbolt, 0, "syncer", 0);
274 * Request the syncer daemon to speed up its work.
275 * We never push it to speed up more than half of its
276 * normal turn time, otherwise it could take over the cpu.
278 * YYY wchan field protected by the BGL.
284 if (updatethread->td_wchan == &lbolt) { /* YYY */
285 unsleep(updatethread);
286 lwkt_schedule(updatethread);
289 if (rushjob < syncdelay / 2) {
291 stat_rush_requests += 1;
298 * Routine to create and manage a filesystem syncer vnode.
300 #define sync_close ((int (*) (struct vop_close_args *))nullop)
301 static int sync_fsync (struct vop_fsync_args *);
302 static int sync_inactive (struct vop_inactive_args *);
303 static int sync_reclaim (struct vop_reclaim_args *);
304 #define sync_lock ((int (*) (struct vop_lock_args *))vop_stdlock)
305 #define sync_unlock ((int (*) (struct vop_unlock_args *))vop_stdunlock)
306 static int sync_print (struct vop_print_args *);
307 #define sync_islocked ((int(*) (struct vop_islocked_args *))vop_stdislocked)
309 static struct vop_ops *sync_vnode_vops;
310 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
311 { &vop_default_desc, vop_eopnotsupp },
312 { &vop_close_desc, (void *) sync_close }, /* close */
313 { &vop_fsync_desc, (void *) sync_fsync }, /* fsync */
314 { &vop_inactive_desc, (void *) sync_inactive }, /* inactive */
315 { &vop_reclaim_desc, (void *) sync_reclaim }, /* reclaim */
316 { &vop_lock_desc, (void *) sync_lock }, /* lock */
317 { &vop_unlock_desc, (void *) sync_unlock }, /* unlock */
318 { &vop_print_desc, (void *) sync_print }, /* print */
319 { &vop_islocked_desc, (void *) sync_islocked }, /* islocked */
323 static struct vnodeopv_desc sync_vnodeop_opv_desc =
324 { &sync_vnode_vops, sync_vnodeop_entries, 0 };
326 VNODEOP_SET(sync_vnodeop_opv_desc);
329 * Create a new filesystem syncer vnode for the specified mount point.
330 * This vnode is placed on the worklist and is responsible for sync'ing
333 * NOTE: read-only mounts are also placed on the worklist. The filesystem
334 * sync code is also responsible for cleaning up vnodes.
337 vfs_allocate_syncvnode(struct mount *mp)
340 static long start, incr, next;
343 /* Allocate a new vnode */
344 error = getspecialvnode(VT_VFS, mp, &sync_vnode_vops, &vp, 0, 0);
346 mp->mnt_syncer = NULL;
351 * Place the vnode onto the syncer worklist. We attempt to
352 * scatter them about on the list so that they will go off
353 * at evenly distributed times even if all the filesystems
354 * are mounted at once.
357 if (next == 0 || next > syncer_maxdelay) {
361 start = syncer_maxdelay / 2;
362 incr = syncer_maxdelay;
366 vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
373 * Do a lazy sync of the filesystem.
375 * sync_fsync { struct vnode *a_vp, struct ucred *a_cred, int a_waitfor,
376 * struct thread *a_td }
379 sync_fsync(struct vop_fsync_args *ap)
381 struct vnode *syncvp = ap->a_vp;
382 struct mount *mp = syncvp->v_mount;
383 struct thread *td = ap->a_td;
387 * We only need to do something if this is a lazy evaluation.
389 if (ap->a_waitfor != MNT_LAZY)
393 * Move ourselves to the back of the sync list.
395 vn_syncer_add_to_worklist(syncvp, syncdelay);
398 * Walk the list of vnodes pushing all that are dirty and
399 * not already on the sync list, and freeing vnodes which have
400 * no refs and whos VM objects are empty. vfs_msync() handles
401 * the VM issues and must be called whether the mount is readonly
404 if (vfs_busy(mp, LK_NOWAIT, td) != 0)
406 if (mp->mnt_flag & MNT_RDONLY) {
407 vfs_msync(mp, MNT_NOWAIT);
409 asyncflag = mp->mnt_flag & MNT_ASYNC;
410 mp->mnt_flag &= ~MNT_ASYNC; /* ZZZ hack */
411 vfs_msync(mp, MNT_NOWAIT);
412 VFS_SYNC(mp, MNT_LAZY, td);
414 mp->mnt_flag |= MNT_ASYNC;
421 * The syncer vnode is no referenced.
423 * sync_inactive { struct vnode *a_vp, struct proc *a_p }
426 sync_inactive(struct vop_inactive_args *ap)
433 * The syncer vnode is no longer needed and is being decommissioned.
435 * Modifications to the worklist must be protected with a critical
438 * sync_reclaim { struct vnode *a_vp }
441 sync_reclaim(struct vop_reclaim_args *ap)
443 struct vnode *vp = ap->a_vp;
446 vp->v_mount->mnt_syncer = NULL;
447 if (vp->v_flag & VONWORKLST) {
448 LIST_REMOVE(vp, v_synclist);
449 vp->v_flag &= ~VONWORKLST;
457 * Print out a syncer vnode.
459 * sync_print { struct vnode *a_vp }
462 sync_print(struct vop_print_args *ap)
464 struct vnode *vp = ap->a_vp;
466 printf("syncer vnode");
467 lockmgr_printinfo(&vp->v_lock);