xref: /linux/drivers/md/dm-delay.c (revision 93d90ad708b8da6efc0e487b66111aa9db7f70c7)
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 workqueue_struct *kdelayd_wq;
24 	struct work_struct flush_expired_bios;
25 	struct list_head delayed_bios;
26 	atomic_t may_delay;
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 	unsigned long expires;
43 };
44 
45 static DEFINE_MUTEX(delayed_bios_lock);
46 
47 static void handle_delayed_timer(unsigned long data)
48 {
49 	struct delay_c *dc = (struct delay_c *)data;
50 
51 	queue_work(dc->kdelayd_wq, &dc->flush_expired_bios);
52 }
53 
54 static void queue_timeout(struct delay_c *dc, unsigned long expires)
55 {
56 	mutex_lock(&dc->timer_lock);
57 
58 	if (!timer_pending(&dc->delay_timer) || expires < dc->delay_timer.expires)
59 		mod_timer(&dc->delay_timer, expires);
60 
61 	mutex_unlock(&dc->timer_lock);
62 }
63 
64 static void flush_bios(struct bio *bio)
65 {
66 	struct bio *n;
67 
68 	while (bio) {
69 		n = bio->bi_next;
70 		bio->bi_next = NULL;
71 		generic_make_request(bio);
72 		bio = n;
73 	}
74 }
75 
76 static struct bio *flush_delayed_bios(struct delay_c *dc, int flush_all)
77 {
78 	struct dm_delay_info *delayed, *next;
79 	unsigned long next_expires = 0;
80 	int start_timer = 0;
81 	struct bio_list flush_bios = { };
82 
83 	mutex_lock(&delayed_bios_lock);
84 	list_for_each_entry_safe(delayed, next, &dc->delayed_bios, list) {
85 		if (flush_all || time_after_eq(jiffies, delayed->expires)) {
86 			struct bio *bio = dm_bio_from_per_bio_data(delayed,
87 						sizeof(struct dm_delay_info));
88 			list_del(&delayed->list);
89 			bio_list_add(&flush_bios, bio);
90 			if ((bio_data_dir(bio) == WRITE))
91 				delayed->context->writes--;
92 			else
93 				delayed->context->reads--;
94 			continue;
95 		}
96 
97 		if (!start_timer) {
98 			start_timer = 1;
99 			next_expires = delayed->expires;
100 		} else
101 			next_expires = min(next_expires, delayed->expires);
102 	}
103 
104 	mutex_unlock(&delayed_bios_lock);
105 
106 	if (start_timer)
107 		queue_timeout(dc, next_expires);
108 
109 	return bio_list_get(&flush_bios);
110 }
111 
112 static void flush_expired_bios(struct work_struct *work)
113 {
114 	struct delay_c *dc;
115 
116 	dc = container_of(work, struct delay_c, flush_expired_bios);
117 	flush_bios(flush_delayed_bios(dc, 0));
118 }
119 
120 /*
121  * Mapping parameters:
122  *    <device> <offset> <delay> [<write_device> <write_offset> <write_delay>]
123  *
124  * With separate write parameters, the first set is only used for reads.
125  * Delays are specified in milliseconds.
126  */
127 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
128 {
129 	struct delay_c *dc;
130 	unsigned long long tmpll;
131 	char dummy;
132 
133 	if (argc != 3 && argc != 6) {
134 		ti->error = "requires exactly 3 or 6 arguments";
135 		return -EINVAL;
136 	}
137 
138 	dc = kmalloc(sizeof(*dc), GFP_KERNEL);
139 	if (!dc) {
140 		ti->error = "Cannot allocate context";
141 		return -ENOMEM;
142 	}
143 
144 	dc->reads = dc->writes = 0;
145 
146 	if (sscanf(argv[1], "%llu%c", &tmpll, &dummy) != 1) {
147 		ti->error = "Invalid device sector";
148 		goto bad;
149 	}
150 	dc->start_read = tmpll;
151 
152 	if (sscanf(argv[2], "%u%c", &dc->read_delay, &dummy) != 1) {
153 		ti->error = "Invalid delay";
154 		goto bad;
155 	}
156 
157 	if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
158 			  &dc->dev_read)) {
159 		ti->error = "Device lookup failed";
160 		goto bad;
161 	}
162 
163 	dc->dev_write = NULL;
164 	if (argc == 3)
165 		goto out;
166 
167 	if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1) {
168 		ti->error = "Invalid write device sector";
169 		goto bad_dev_read;
170 	}
171 	dc->start_write = tmpll;
172 
173 	if (sscanf(argv[5], "%u%c", &dc->write_delay, &dummy) != 1) {
174 		ti->error = "Invalid write delay";
175 		goto bad_dev_read;
176 	}
177 
178 	if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table),
179 			  &dc->dev_write)) {
180 		ti->error = "Write device lookup failed";
181 		goto bad_dev_read;
182 	}
183 
184 out:
185 	dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
186 	if (!dc->kdelayd_wq) {
187 		DMERR("Couldn't start kdelayd");
188 		goto bad_queue;
189 	}
190 
191 	setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc);
192 
193 	INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
194 	INIT_LIST_HEAD(&dc->delayed_bios);
195 	mutex_init(&dc->timer_lock);
196 	atomic_set(&dc->may_delay, 1);
197 
198 	ti->num_flush_bios = 1;
199 	ti->num_discard_bios = 1;
200 	ti->per_bio_data_size = sizeof(struct dm_delay_info);
201 	ti->private = dc;
202 	return 0;
203 
204 bad_queue:
205 	if (dc->dev_write)
206 		dm_put_device(ti, dc->dev_write);
207 bad_dev_read:
208 	dm_put_device(ti, dc->dev_read);
209 bad:
210 	kfree(dc);
211 	return -EINVAL;
212 }
213 
214 static void delay_dtr(struct dm_target *ti)
215 {
216 	struct delay_c *dc = ti->private;
217 
218 	destroy_workqueue(dc->kdelayd_wq);
219 
220 	dm_put_device(ti, dc->dev_read);
221 
222 	if (dc->dev_write)
223 		dm_put_device(ti, dc->dev_write);
224 
225 	kfree(dc);
226 }
227 
228 static int delay_bio(struct delay_c *dc, int delay, struct bio *bio)
229 {
230 	struct dm_delay_info *delayed;
231 	unsigned long expires = 0;
232 
233 	if (!delay || !atomic_read(&dc->may_delay))
234 		return 1;
235 
236 	delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
237 
238 	delayed->context = dc;
239 	delayed->expires = expires = jiffies + (delay * HZ / 1000);
240 
241 	mutex_lock(&delayed_bios_lock);
242 
243 	if (bio_data_dir(bio) == WRITE)
244 		dc->writes++;
245 	else
246 		dc->reads++;
247 
248 	list_add_tail(&delayed->list, &dc->delayed_bios);
249 
250 	mutex_unlock(&delayed_bios_lock);
251 
252 	queue_timeout(dc, expires);
253 
254 	return 0;
255 }
256 
257 static void delay_presuspend(struct dm_target *ti)
258 {
259 	struct delay_c *dc = ti->private;
260 
261 	atomic_set(&dc->may_delay, 0);
262 	del_timer_sync(&dc->delay_timer);
263 	flush_bios(flush_delayed_bios(dc, 1));
264 }
265 
266 static void delay_resume(struct dm_target *ti)
267 {
268 	struct delay_c *dc = ti->private;
269 
270 	atomic_set(&dc->may_delay, 1);
271 }
272 
273 static int delay_map(struct dm_target *ti, struct bio *bio)
274 {
275 	struct delay_c *dc = ti->private;
276 
277 	if ((bio_data_dir(bio) == WRITE) && (dc->dev_write)) {
278 		bio->bi_bdev = dc->dev_write->bdev;
279 		if (bio_sectors(bio))
280 			bio->bi_iter.bi_sector = dc->start_write +
281 				dm_target_offset(ti, bio->bi_iter.bi_sector);
282 
283 		return delay_bio(dc, dc->write_delay, bio);
284 	}
285 
286 	bio->bi_bdev = dc->dev_read->bdev;
287 	bio->bi_iter.bi_sector = dc->start_read +
288 		dm_target_offset(ti, bio->bi_iter.bi_sector);
289 
290 	return delay_bio(dc, dc->read_delay, bio);
291 }
292 
293 static void delay_status(struct dm_target *ti, status_type_t type,
294 			 unsigned status_flags, char *result, unsigned maxlen)
295 {
296 	struct delay_c *dc = ti->private;
297 	int sz = 0;
298 
299 	switch (type) {
300 	case STATUSTYPE_INFO:
301 		DMEMIT("%u %u", dc->reads, dc->writes);
302 		break;
303 
304 	case STATUSTYPE_TABLE:
305 		DMEMIT("%s %llu %u", dc->dev_read->name,
306 		       (unsigned long long) dc->start_read,
307 		       dc->read_delay);
308 		if (dc->dev_write)
309 			DMEMIT(" %s %llu %u", dc->dev_write->name,
310 			       (unsigned long long) dc->start_write,
311 			       dc->write_delay);
312 		break;
313 	}
314 }
315 
316 static int delay_iterate_devices(struct dm_target *ti,
317 				 iterate_devices_callout_fn fn, void *data)
318 {
319 	struct delay_c *dc = ti->private;
320 	int ret = 0;
321 
322 	ret = fn(ti, dc->dev_read, dc->start_read, ti->len, data);
323 	if (ret)
324 		goto out;
325 
326 	if (dc->dev_write)
327 		ret = fn(ti, dc->dev_write, dc->start_write, ti->len, data);
328 
329 out:
330 	return ret;
331 }
332 
333 static struct target_type delay_target = {
334 	.name	     = "delay",
335 	.version     = {1, 2, 1},
336 	.module      = THIS_MODULE,
337 	.ctr	     = delay_ctr,
338 	.dtr	     = delay_dtr,
339 	.map	     = delay_map,
340 	.presuspend  = delay_presuspend,
341 	.resume	     = delay_resume,
342 	.status	     = delay_status,
343 	.iterate_devices = delay_iterate_devices,
344 };
345 
346 static int __init dm_delay_init(void)
347 {
348 	int r;
349 
350 	r = dm_register_target(&delay_target);
351 	if (r < 0) {
352 		DMERR("register failed %d", r);
353 		goto bad_register;
354 	}
355 
356 	return 0;
357 
358 bad_register:
359 	return r;
360 }
361 
362 static void __exit dm_delay_exit(void)
363 {
364 	dm_unregister_target(&delay_target);
365 }
366 
367 /* Module hooks */
368 module_init(dm_delay_init);
369 module_exit(dm_delay_exit);
370 
371 MODULE_DESCRIPTION(DM_NAME " delay target");
372 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
373 MODULE_LICENSE("GPL");
374