summaryrefslogtreecommitdiff
path: root/sys/miscfs/syncfs/sync_subr.c
blob: 5bdbe93dc06e27ce36ea371880bbbcf5e8dc75fc (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
/*	$NetBSD: sync_subr.c,v 1.41 2009/09/16 15:23:04 pooka Exp $	*/

/*-
 * Copyright (c) 2009 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Andrew Doran.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * Copyright 1997 Marshall Kirk McKusick. All Rights Reserved.
 *
 * This code is derived from work done by Greg Ganger at the
 * University of Michigan.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. None of the names of McKusick, Ganger, or the University of Michigan
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: sync_subr.c,v 1.41 2009/09/16 15:23:04 pooka Exp $");

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysctl.h>
#include <sys/kernel.h>
#include <sys/proc.h>
#include <sys/mount.h>
#include <sys/time.h>
#include <sys/vnode.h>
#include <sys/buf.h>
#include <sys/errno.h>
#include <sys/kmem.h>

#include <miscfs/genfs/genfs.h>
#include <miscfs/syncfs/syncfs.h>

static void	vn_syncer_add1(struct vnode *, int);

/*
 * Defines and variables for the syncer process.
 */
int syncer_maxdelay = SYNCER_MAXDELAY;	/* maximum delay time */
time_t syncdelay = 30;			/* max time to delay syncing data */
time_t filedelay = 30;			/* time to delay syncing files */
time_t dirdelay  = 15;			/* time to delay syncing directories */
time_t metadelay = 10;			/* time to delay syncing metadata */
time_t lockdelay = 1;			/* time to delay if locking fails */

kmutex_t syncer_mutex;			/* used to freeze syncer, long term */
static kmutex_t syncer_data_lock;	/* short term lock on data structures */

static int rushjob;			/* number of slots to run ASAP */
static kcondvar_t syncer_cv;		/* cv for rushjob */
static int stat_rush_requests;		/* number of times I/O speeded up */

static int syncer_delayno = 0;
static long syncer_last;
static struct synclist *syncer_workitem_pending;
struct lwp *updateproc = NULL;

static void sysctl_vfs_syncfs_setup(struct sysctllog **);

void
vn_initialize_syncerd(void)
{
	int i;

	syncer_last = SYNCER_MAXDELAY + 2;

	sysctl_vfs_syncfs_setup(NULL);

	syncer_workitem_pending =
	    kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP);

	for (i = 0; i < syncer_last; i++)
		TAILQ_INIT(&syncer_workitem_pending[i]);

	mutex_init(&syncer_mutex, MUTEX_DEFAULT, IPL_NONE);
	mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE);
	cv_init(&syncer_cv, "syncer");
}

/*
 * The workitem queue.
 *
 * It is useful to delay writes of file data and filesystem metadata
 * for tens of seconds so that quickly created and deleted files need
 * not waste disk bandwidth being created and removed. To realize this,
 * we append vnodes to a "workitem" queue. When running with a soft
 * updates implementation, most pending metadata dependencies should
 * not wait for more than a few seconds. Thus, mounted on block devices
 * are delayed only about a half the time that file data is delayed.
 * Similarly, directory updates are more critical, so are only delayed
 * about a third the time that file data is delayed. Thus, there are
 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
 * one each second (driven off the filesystem syner process). The
 * syncer_delayno variable indicates the next queue that is to be processed.
 * Items that need to be processed soon are placed in this queue:
 *
 *	syncer_workitem_pending[syncer_delayno]
 *
 * A delay of fifteen seconds is done by placing the request fifteen
 * entries later in the queue:
 *
 *	syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
 *
 */

/*
 * Add an item to the syncer work queue.
 */
static void
vn_syncer_add1(struct vnode *vp, int delayx)
{
	struct synclist *slp;

	KASSERT(mutex_owned(&syncer_data_lock));

	if (vp->v_iflag & VI_ONWORKLST) {
		slp = &syncer_workitem_pending[vp->v_synclist_slot];
		TAILQ_REMOVE(slp, vp, v_synclist);
	} else {
		/*
		 * We must not modify v_iflag if the vnode
		 * is already on a synclist: sched_sync()
		 * calls this routine while holding only
		 * syncer_data_lock in order to adjust the
		 * position of the vnode.  syncer_data_lock
		 * does not protect v_iflag.
		 */
		KASSERT(mutex_owned(&vp->v_interlock));
		vp->v_iflag |= VI_ONWORKLST;
	}

	if (delayx > syncer_maxdelay - 2)
		delayx = syncer_maxdelay - 2;
	vp->v_synclist_slot = (syncer_delayno + delayx) % syncer_last;

	slp = &syncer_workitem_pending[vp->v_synclist_slot];
	TAILQ_INSERT_TAIL(slp, vp, v_synclist);
}

void
vn_syncer_add_to_worklist(struct vnode *vp, int delayx)
{

	KASSERT(mutex_owned(&vp->v_interlock));

	mutex_enter(&syncer_data_lock);
	vn_syncer_add1(vp, delayx);
	mutex_exit(&syncer_data_lock);
}

/*
 * Remove an item from the syncer work queue.
 */
void
vn_syncer_remove_from_worklist(struct vnode *vp)
{
	struct synclist *slp;

	KASSERT(mutex_owned(&vp->v_interlock));

	mutex_enter(&syncer_data_lock);

	if (vp->v_iflag & VI_ONWORKLST) {
		vp->v_iflag &= ~VI_ONWORKLST;
		slp = &syncer_workitem_pending[vp->v_synclist_slot];
		TAILQ_REMOVE(slp, vp, v_synclist);
	}

	mutex_exit(&syncer_data_lock);
}

/*
 * System filesystem synchronizer daemon.
 */
void
sched_sync(void *v)
{
	struct synclist *slp;
	struct vnode *vp;
	long starttime;
	bool synced;

	updateproc = curlwp;

	for (;;) {
		mutex_enter(&syncer_mutex);
		mutex_enter(&syncer_data_lock);

		starttime = time_second;

		/*
		 * Push files whose dirty time has expired.
		 */
		slp = &syncer_workitem_pending[syncer_delayno];
		syncer_delayno += 1;
		if (syncer_delayno >= syncer_last)
			syncer_delayno = 0;

		while ((vp = TAILQ_FIRST(slp)) != NULL) {
			/* We are locking in the wrong direction. */
			synced = false;
			if (mutex_tryenter(&vp->v_interlock)) {
				mutex_exit(&syncer_data_lock);
				if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT |
				    LK_INTERLOCK) == 0) {
					synced = true;
					(void) VOP_FSYNC(vp, curlwp->l_cred,
					    FSYNC_LAZY, 0, 0);
					vput(vp);
				}
				mutex_enter(&syncer_data_lock);
			}

			/*
			 * XXX The vnode may have been recycled, in which
			 * case it may have a new identity.
			 */
			if (TAILQ_FIRST(slp) == vp) {
				/*
				 * Put us back on the worklist.  The worklist
				 * routine will remove us from our current
				 * position and then add us back in at a later
				 * position.
				 *
				 * Try again sooner rather than later if
				 * we were unable to lock the vnode.  Lock
				 * failure should not prevent us from doing
				 * the sync "soon".
				 *
				 * If we locked it yet arrive here, it's
				 * likely that lazy sync is in progress and
				 * so the vnode still has dirty metadata. 
				 * syncdelay is mainly to get this vnode out
				 * of the way so we do not consider it again
				 * "soon" in this loop, so the delay time is
				 * not critical as long as it is not "soon". 
				 * While write-back strategy is the file
				 * system's domain, we expect write-back to
				 * occur no later than syncdelay seconds
				 * into the future.
				 */
				vn_syncer_add1(vp,
				    synced ? syncdelay : lockdelay);
			}
		}
		mutex_exit(&syncer_mutex);

		/*
		 * Wait until there are more workitems to process.
		 */
		if (rushjob > 0) {
			/*
			 * The variable rushjob allows the kernel to speed
			 * up the processing of the filesystem syncer
			 * process. A rushjob value of N tells the
			 * filesystem syncer to process the next N seconds
			 * worth of work on its queue ASAP. Currently
			 * rushjob is used by the soft update code to
			 * speed up the filesystem syncer process when the
			 * incore state is getting so far ahead of the
			 * disk that the kernel memory pool is being
			 * threatened with exhaustion.
			 */
			rushjob--;
		} else {
			/*
			 * If it has taken us less than a second to
			 * process the current work, then wait. Otherwise
			 * start right over again. We can still lose time
			 * if any single round takes more than two
			 * seconds, but it does not really matter as we
			 * are just trying to generally pace the
			 * filesystem activity.
			 */
			if (time_second == starttime)
				cv_timedwait(&syncer_cv, &syncer_data_lock, hz);
		}
		mutex_exit(&syncer_data_lock);
	}
}

/*
 * Request the syncer daemon to speed up its work.
 * We never push it to speed up more than half of its
 * normal turn time, otherwise it could take over the CPU.
 */
int
speedup_syncer(void)
{

	mutex_enter(&syncer_data_lock);
	if (rushjob >= syncdelay / 2) {
		mutex_exit(&syncer_data_lock);
		return (0);
	}
	rushjob++;
	cv_signal(&syncer_cv);
	stat_rush_requests += 1;
	mutex_exit(&syncer_data_lock);

	return (1);
}

static void
sysctl_vfs_syncfs_setup(struct sysctllog **clog)
{
	const struct sysctlnode *rnode, *cnode;
        
	sysctl_createv(clog, 0, NULL, &rnode,
			CTLFLAG_PERMANENT,
			CTLTYPE_NODE, "vfs", NULL,
			NULL, 0, NULL, 0,
			CTL_VFS, CTL_EOL);

	sysctl_createv(clog, 0, &rnode, &rnode,
			CTLFLAG_PERMANENT,
			CTLTYPE_NODE, "sync",
			SYSCTL_DESCR("syncer options"),
			NULL, 0, NULL, 0,
			CTL_CREATE, CTL_EOL);

	sysctl_createv(clog, 0, &rnode, &cnode,
			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
			CTLTYPE_INT, "delay",
			SYSCTL_DESCR("max time to delay syncing data"),
			NULL, 0, &syncdelay, 0,
			CTL_CREATE, CTL_EOL);

	sysctl_createv(clog, 0, &rnode, &cnode,
			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
			CTLTYPE_INT, "filedelay",
			SYSCTL_DESCR("time to delay syncing files"),
			NULL, 0, &filedelay, 0,
			CTL_CREATE, CTL_EOL);

	sysctl_createv(clog, 0, &rnode, &cnode,
			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
			CTLTYPE_INT, "dirdelay",
			SYSCTL_DESCR("time to delay syncing directories"),
			NULL, 0, &dirdelay, 0,
			CTL_CREATE, CTL_EOL);

	sysctl_createv(clog, 0, &rnode, &cnode,
			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
			CTLTYPE_INT, "metadelay",
			SYSCTL_DESCR("time to delay syncing metadata"),
			NULL, 0, &metadelay, 0,
			CTL_CREATE, CTL_EOL);
}