xref: /openbmc/linux/drivers/md/dm-delay.c (revision 39b6f3aa)
1 /*
2  * Copyright (C) 2005-2007 Red Hat GmbH
3  *
4  * A target that delays reads and/or writes and can send
5  * them to different devices.
6  *
7  * This file is released under the GPL.
8  */
9 
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/blkdev.h>
13 #include <linux/bio.h>
14 #include <linux/slab.h>
15 
16 #include <linux/device-mapper.h>
17 
18 #define DM_MSG_PREFIX "delay"
19 
20 struct delay_c {
21 	struct timer_list delay_timer;
22 	struct mutex timer_lock;
23 	struct work_struct flush_expired_bios;
24 	struct list_head delayed_bios;
25 	atomic_t may_delay;
26 	mempool_t *delayed_pool;
27 
28 	struct dm_dev *dev_read;
29 	sector_t start_read;
30 	unsigned read_delay;
31 	unsigned reads;
32 
33 	struct dm_dev *dev_write;
34 	sector_t start_write;
35 	unsigned write_delay;
36 	unsigned writes;
37 };
38 
39 struct dm_delay_info {
40 	struct delay_c *context;
41 	struct list_head list;
42 	struct bio *bio;
43 	unsigned long expires;
44 };
45 
46 static DEFINE_MUTEX(delayed_bios_lock);
47 
48 static struct workqueue_struct *kdelayd_wq;
49 static struct kmem_cache *delayed_cache;
50 
51 static void handle_delayed_timer(unsigned long data)
52 {
53 	struct delay_c *dc = (struct delay_c *)data;
54 
55 	queue_work(kdelayd_wq, &dc->flush_expired_bios);
56 }
57 
58 static void queue_timeout(struct delay_c *dc, unsigned long expires)
59 {
60 	mutex_lock(&dc->timer_lock);
61 
62 	if (!timer_pending(&dc->delay_timer) || expires < dc->delay_timer.expires)
63 		mod_timer(&dc->delay_timer, expires);
64 
65 	mutex_unlock(&dc->timer_lock);
66 }
67 
68 static void flush_bios(struct bio *bio)
69 {
70 	struct bio *n;
71 
72 	while (bio) {
73 		n = bio->bi_next;
74 		bio->bi_next = NULL;
75 		generic_make_request(bio);
76 		bio = n;
77 	}
78 }
79 
80 static struct bio *flush_delayed_bios(struct delay_c *dc, int flush_all)
81 {
82 	struct dm_delay_info *delayed, *next;
83 	unsigned long next_expires = 0;
84 	int start_timer = 0;
85 	struct bio_list flush_bios = { };
86 
87 	mutex_lock(&delayed_bios_lock);
88 	list_for_each_entry_safe(delayed, next, &dc->delayed_bios, list) {
89 		if (flush_all || time_after_eq(jiffies, delayed->expires)) {
90 			list_del(&delayed->list);
91 			bio_list_add(&flush_bios, delayed->bio);
92 			if ((bio_data_dir(delayed->bio) == WRITE))
93 				delayed->context->writes--;
94 			else
95 				delayed->context->reads--;
96 			mempool_free(delayed, dc->delayed_pool);
97 			continue;
98 		}
99 
100 		if (!start_timer) {
101 			start_timer = 1;
102 			next_expires = delayed->expires;
103 		} else
104 			next_expires = min(next_expires, delayed->expires);
105 	}
106 
107 	mutex_unlock(&delayed_bios_lock);
108 
109 	if (start_timer)
110 		queue_timeout(dc, next_expires);
111 
112 	return bio_list_get(&flush_bios);
113 }
114 
115 static void flush_expired_bios(struct work_struct *work)
116 {
117 	struct delay_c *dc;
118 
119 	dc = container_of(work, struct delay_c, flush_expired_bios);
120 	flush_bios(flush_delayed_bios(dc, 0));
121 }
122 
123 /*
124  * Mapping parameters:
125  *    <device> <offset> <delay> [<write_device> <write_offset> <write_delay>]
126  *
127  * With separate write parameters, the first set is only used for reads.
128  * Delays are specified in milliseconds.
129  */
130 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
131 {
132 	struct delay_c *dc;
133 	unsigned long long tmpll;
134 	char dummy;
135 
136 	if (argc != 3 && argc != 6) {
137 		ti->error = "requires exactly 3 or 6 arguments";
138 		return -EINVAL;
139 	}
140 
141 	dc = kmalloc(sizeof(*dc), GFP_KERNEL);
142 	if (!dc) {
143 		ti->error = "Cannot allocate context";
144 		return -ENOMEM;
145 	}
146 
147 	dc->reads = dc->writes = 0;
148 
149 	if (sscanf(argv[1], "%llu%c", &tmpll, &dummy) != 1) {
150 		ti->error = "Invalid device sector";
151 		goto bad;
152 	}
153 	dc->start_read = tmpll;
154 
155 	if (sscanf(argv[2], "%u%c", &dc->read_delay, &dummy) != 1) {
156 		ti->error = "Invalid delay";
157 		goto bad;
158 	}
159 
160 	if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
161 			  &dc->dev_read)) {
162 		ti->error = "Device lookup failed";
163 		goto bad;
164 	}
165 
166 	dc->dev_write = NULL;
167 	if (argc == 3)
168 		goto out;
169 
170 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1) {
171 		ti->error = "Invalid write device sector";
172 		goto bad_dev_read;
173 	}
174 	dc->start_write = tmpll;
175 
176 	if (sscanf(argv[5], "%u%c", &dc->write_delay, &dummy) != 1) {
177 		ti->error = "Invalid write delay";
178 		goto bad_dev_read;
179 	}
180 
181 	if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table),
182 			  &dc->dev_write)) {
183 		ti->error = "Write device lookup failed";
184 		goto bad_dev_read;
185 	}
186 
187 out:
188 	dc->delayed_pool = mempool_create_slab_pool(128, delayed_cache);
189 	if (!dc->delayed_pool) {
190 		DMERR("Couldn't create delayed bio pool.");
191 		goto bad_dev_write;
192 	}
193 
194 	setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc);
195 
196 	INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
197 	INIT_LIST_HEAD(&dc->delayed_bios);
198 	mutex_init(&dc->timer_lock);
199 	atomic_set(&dc->may_delay, 1);
200 
201 	ti->num_flush_bios = 1;
202 	ti->num_discard_bios = 1;
203 	ti->private = dc;
204 	return 0;
205 
206 bad_dev_write:
207 	if (dc->dev_write)
208 		dm_put_device(ti, dc->dev_write);
209 bad_dev_read:
210 	dm_put_device(ti, dc->dev_read);
211 bad:
212 	kfree(dc);
213 	return -EINVAL;
214 }
215 
216 static void delay_dtr(struct dm_target *ti)
217 {
218 	struct delay_c *dc = ti->private;
219 
220 	flush_workqueue(kdelayd_wq);
221 
222 	dm_put_device(ti, dc->dev_read);
223 
224 	if (dc->dev_write)
225 		dm_put_device(ti, dc->dev_write);
226 
227 	mempool_destroy(dc->delayed_pool);
228 	kfree(dc);
229 }
230 
231 static int delay_bio(struct delay_c *dc, int delay, struct bio *bio)
232 {
233 	struct dm_delay_info *delayed;
234 	unsigned long expires = 0;
235 
236 	if (!delay || !atomic_read(&dc->may_delay))
237 		return 1;
238 
239 	delayed = mempool_alloc(dc->delayed_pool, GFP_NOIO);
240 
241 	delayed->context = dc;
242 	delayed->bio = bio;
243 	delayed->expires = expires = jiffies + (delay * HZ / 1000);
244 
245 	mutex_lock(&delayed_bios_lock);
246 
247 	if (bio_data_dir(bio) == WRITE)
248 		dc->writes++;
249 	else
250 		dc->reads++;
251 
252 	list_add_tail(&delayed->list, &dc->delayed_bios);
253 
254 	mutex_unlock(&delayed_bios_lock);
255 
256 	queue_timeout(dc, expires);
257 
258 	return 0;
259 }
260 
261 static void delay_presuspend(struct dm_target *ti)
262 {
263 	struct delay_c *dc = ti->private;
264 
265 	atomic_set(&dc->may_delay, 0);
266 	del_timer_sync(&dc->delay_timer);
267 	flush_bios(flush_delayed_bios(dc, 1));
268 }
269 
270 static void delay_resume(struct dm_target *ti)
271 {
272 	struct delay_c *dc = ti->private;
273 
274 	atomic_set(&dc->may_delay, 1);
275 }
276 
277 static int delay_map(struct dm_target *ti, struct bio *bio)
278 {
279 	struct delay_c *dc = ti->private;
280 
281 	if ((bio_data_dir(bio) == WRITE) && (dc->dev_write)) {
282 		bio->bi_bdev = dc->dev_write->bdev;
283 		if (bio_sectors(bio))
284 			bio->bi_sector = dc->start_write +
285 					 dm_target_offset(ti, bio->bi_sector);
286 
287 		return delay_bio(dc, dc->write_delay, bio);
288 	}
289 
290 	bio->bi_bdev = dc->dev_read->bdev;
291 	bio->bi_sector = dc->start_read + dm_target_offset(ti, bio->bi_sector);
292 
293 	return delay_bio(dc, dc->read_delay, bio);
294 }
295 
296 static void delay_status(struct dm_target *ti, status_type_t type,
297 			 unsigned status_flags, char *result, unsigned maxlen)
298 {
299 	struct delay_c *dc = ti->private;
300 	int sz = 0;
301 
302 	switch (type) {
303 	case STATUSTYPE_INFO:
304 		DMEMIT("%u %u", dc->reads, dc->writes);
305 		break;
306 
307 	case STATUSTYPE_TABLE:
308 		DMEMIT("%s %llu %u", dc->dev_read->name,
309 		       (unsigned long long) dc->start_read,
310 		       dc->read_delay);
311 		if (dc->dev_write)
312 			DMEMIT(" %s %llu %u", dc->dev_write->name,
313 			       (unsigned long long) dc->start_write,
314 			       dc->write_delay);
315 		break;
316 	}
317 }
318 
319 static int delay_iterate_devices(struct dm_target *ti,
320 				 iterate_devices_callout_fn fn, void *data)
321 {
322 	struct delay_c *dc = ti->private;
323 	int ret = 0;
324 
325 	ret = fn(ti, dc->dev_read, dc->start_read, ti->len, data);
326 	if (ret)
327 		goto out;
328 
329 	if (dc->dev_write)
330 		ret = fn(ti, dc->dev_write, dc->start_write, ti->len, data);
331 
332 out:
333 	return ret;
334 }
335 
336 static struct target_type delay_target = {
337 	.name	     = "delay",
338 	.version     = {1, 2, 1},
339 	.module      = THIS_MODULE,
340 	.ctr	     = delay_ctr,
341 	.dtr	     = delay_dtr,
342 	.map	     = delay_map,
343 	.presuspend  = delay_presuspend,
344 	.resume	     = delay_resume,
345 	.status	     = delay_status,
346 	.iterate_devices = delay_iterate_devices,
347 };
348 
349 static int __init dm_delay_init(void)
350 {
351 	int r = -ENOMEM;
352 
353 	kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
354 	if (!kdelayd_wq) {
355 		DMERR("Couldn't start kdelayd");
356 		goto bad_queue;
357 	}
358 
359 	delayed_cache = KMEM_CACHE(dm_delay_info, 0);
360 	if (!delayed_cache) {
361 		DMERR("Couldn't create delayed bio cache.");
362 		goto bad_memcache;
363 	}
364 
365 	r = dm_register_target(&delay_target);
366 	if (r < 0) {
367 		DMERR("register failed %d", r);
368 		goto bad_register;
369 	}
370 
371 	return 0;
372 
373 bad_register:
374 	kmem_cache_destroy(delayed_cache);
375 bad_memcache:
376 	destroy_workqueue(kdelayd_wq);
377 bad_queue:
378 	return r;
379 }
380 
381 static void __exit dm_delay_exit(void)
382 {
383 	dm_unregister_target(&delay_target);
384 	kmem_cache_destroy(delayed_cache);
385 	destroy_workqueue(kdelayd_wq);
386 }
387 
388 /* Module hooks */
389 module_init(dm_delay_init);
390 module_exit(dm_delay_exit);
391 
392 MODULE_DESCRIPTION(DM_NAME " delay target");
393 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
394 MODULE_LICENSE("GPL");
395