xref: /openbmc/linux/drivers/md/dm-delay.c (revision 2fa49589)
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_class {
21 	struct dm_dev *dev;
22 	sector_t start;
23 	unsigned delay;
24 	unsigned ops;
25 };
26 
27 struct delay_c {
28 	struct timer_list delay_timer;
29 	struct mutex timer_lock;
30 	struct workqueue_struct *kdelayd_wq;
31 	struct work_struct flush_expired_bios;
32 	struct list_head delayed_bios;
33 	atomic_t may_delay;
34 
35 	struct delay_class read;
36 	struct delay_class write;
37 	struct delay_class flush;
38 
39 	int argc;
40 };
41 
42 struct dm_delay_info {
43 	struct delay_c *context;
44 	struct delay_class *class;
45 	struct list_head list;
46 	unsigned long expires;
47 };
48 
49 static DEFINE_MUTEX(delayed_bios_lock);
50 
51 static void handle_delayed_timer(struct timer_list *t)
52 {
53 	struct delay_c *dc = from_timer(dc, t, delay_timer);
54 
55 	queue_work(dc->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 	unsigned long 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 			struct bio *bio = dm_bio_from_per_bio_data(delayed,
91 						sizeof(struct dm_delay_info));
92 			list_del(&delayed->list);
93 			bio_list_add(&flush_bios, bio);
94 			delayed->class->ops--;
95 			continue;
96 		}
97 
98 		if (!start_timer) {
99 			start_timer = 1;
100 			next_expires = delayed->expires;
101 		} else
102 			next_expires = min(next_expires, delayed->expires);
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 static void delay_dtr(struct dm_target *ti)
121 {
122 	struct delay_c *dc = ti->private;
123 
124 	destroy_workqueue(dc->kdelayd_wq);
125 
126 	if (dc->read.dev)
127 		dm_put_device(ti, dc->read.dev);
128 	if (dc->write.dev)
129 		dm_put_device(ti, dc->write.dev);
130 	if (dc->flush.dev)
131 		dm_put_device(ti, dc->flush.dev);
132 
133 	mutex_destroy(&dc->timer_lock);
134 
135 	kfree(dc);
136 }
137 
138 static int delay_class_ctr(struct dm_target *ti, struct delay_class *c, char **argv)
139 {
140 	int ret;
141 	unsigned long long tmpll;
142 	char dummy;
143 
144 	if (sscanf(argv[1], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
145 		ti->error = "Invalid device sector";
146 		return -EINVAL;
147 	}
148 	c->start = tmpll;
149 
150 	if (sscanf(argv[2], "%u%c", &c->delay, &dummy) != 1) {
151 		ti->error = "Invalid delay";
152 		return -EINVAL;
153 	}
154 
155 	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &c->dev);
156 	if (ret) {
157 		ti->error = "Device lookup failed";
158 		return ret;
159 	}
160 
161 	return 0;
162 }
163 
164 /*
165  * Mapping parameters:
166  *    <device> <offset> <delay> [<write_device> <write_offset> <write_delay>]
167  *
168  * With separate write parameters, the first set is only used for reads.
169  * Offsets are specified in sectors.
170  * Delays are specified in milliseconds.
171  */
172 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
173 {
174 	struct delay_c *dc;
175 	int ret;
176 
177 	if (argc != 3 && argc != 6 && argc != 9) {
178 		ti->error = "Requires exactly 3, 6 or 9 arguments";
179 		return -EINVAL;
180 	}
181 
182 	dc = kzalloc(sizeof(*dc), GFP_KERNEL);
183 	if (!dc) {
184 		ti->error = "Cannot allocate context";
185 		return -ENOMEM;
186 	}
187 
188 	ti->private = dc;
189 	timer_setup(&dc->delay_timer, handle_delayed_timer, 0);
190 	INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
191 	INIT_LIST_HEAD(&dc->delayed_bios);
192 	mutex_init(&dc->timer_lock);
193 	atomic_set(&dc->may_delay, 1);
194 	dc->argc = argc;
195 
196 	ret = delay_class_ctr(ti, &dc->read, argv);
197 	if (ret)
198 		goto bad;
199 
200 	if (argc == 3) {
201 		ret = delay_class_ctr(ti, &dc->write, argv);
202 		if (ret)
203 			goto bad;
204 		ret = delay_class_ctr(ti, &dc->flush, argv);
205 		if (ret)
206 			goto bad;
207 		goto out;
208 	}
209 
210 	ret = delay_class_ctr(ti, &dc->write, argv + 3);
211 	if (ret)
212 		goto bad;
213 	if (argc == 6) {
214 		ret = delay_class_ctr(ti, &dc->flush, argv + 3);
215 		if (ret)
216 			goto bad;
217 		goto out;
218 	}
219 
220 	ret = delay_class_ctr(ti, &dc->flush, argv + 6);
221 	if (ret)
222 		goto bad;
223 
224 out:
225 	dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
226 	if (!dc->kdelayd_wq) {
227 		ret = -EINVAL;
228 		DMERR("Couldn't start kdelayd");
229 		goto bad;
230 	}
231 
232 	ti->num_flush_bios = 1;
233 	ti->num_discard_bios = 1;
234 	ti->per_io_data_size = sizeof(struct dm_delay_info);
235 	return 0;
236 
237 bad:
238 	delay_dtr(ti);
239 	return ret;
240 }
241 
242 static int delay_bio(struct delay_c *dc, struct delay_class *c, struct bio *bio)
243 {
244 	struct dm_delay_info *delayed;
245 	unsigned long expires = 0;
246 
247 	if (!c->delay || !atomic_read(&dc->may_delay))
248 		return DM_MAPIO_REMAPPED;
249 
250 	delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
251 
252 	delayed->context = dc;
253 	delayed->expires = expires = jiffies + msecs_to_jiffies(c->delay);
254 
255 	mutex_lock(&delayed_bios_lock);
256 	c->ops++;
257 	list_add_tail(&delayed->list, &dc->delayed_bios);
258 	mutex_unlock(&delayed_bios_lock);
259 
260 	queue_timeout(dc, expires);
261 
262 	return DM_MAPIO_SUBMITTED;
263 }
264 
265 static void delay_presuspend(struct dm_target *ti)
266 {
267 	struct delay_c *dc = ti->private;
268 
269 	atomic_set(&dc->may_delay, 0);
270 	del_timer_sync(&dc->delay_timer);
271 	flush_bios(flush_delayed_bios(dc, 1));
272 }
273 
274 static void delay_resume(struct dm_target *ti)
275 {
276 	struct delay_c *dc = ti->private;
277 
278 	atomic_set(&dc->may_delay, 1);
279 }
280 
281 static int delay_map(struct dm_target *ti, struct bio *bio)
282 {
283 	struct delay_c *dc = ti->private;
284 	struct delay_class *c;
285 	struct dm_delay_info *delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
286 
287 	if (bio_data_dir(bio) == WRITE) {
288 		if (unlikely(bio->bi_opf & REQ_PREFLUSH))
289 			c = &dc->flush;
290 		else
291 			c = &dc->write;
292 	} else {
293 		c = &dc->read;
294 	}
295 	delayed->class = c;
296 	bio_set_dev(bio, c->dev->bdev);
297 	if (bio_sectors(bio))
298 		bio->bi_iter.bi_sector = c->start + dm_target_offset(ti, bio->bi_iter.bi_sector);
299 
300 	return delay_bio(dc, c, bio);
301 }
302 
303 #define DMEMIT_DELAY_CLASS(c) \
304 	DMEMIT("%s %llu %u", (c)->dev->name, (unsigned long long)(c)->start, (c)->delay)
305 
306 static void delay_status(struct dm_target *ti, status_type_t type,
307 			 unsigned status_flags, char *result, unsigned maxlen)
308 {
309 	struct delay_c *dc = ti->private;
310 	int sz = 0;
311 
312 	switch (type) {
313 	case STATUSTYPE_INFO:
314 		DMEMIT("%u %u %u", dc->read.ops, dc->write.ops, dc->flush.ops);
315 		break;
316 
317 	case STATUSTYPE_TABLE:
318 		DMEMIT_DELAY_CLASS(&dc->read);
319 		if (dc->argc >= 6) {
320 			DMEMIT(" ");
321 			DMEMIT_DELAY_CLASS(&dc->write);
322 		}
323 		if (dc->argc >= 9) {
324 			DMEMIT(" ");
325 			DMEMIT_DELAY_CLASS(&dc->flush);
326 		}
327 		break;
328 	}
329 }
330 
331 static int delay_iterate_devices(struct dm_target *ti,
332 				 iterate_devices_callout_fn fn, void *data)
333 {
334 	struct delay_c *dc = ti->private;
335 	int ret = 0;
336 
337 	ret = fn(ti, dc->read.dev, dc->read.start, ti->len, data);
338 	if (ret)
339 		goto out;
340 	ret = fn(ti, dc->write.dev, dc->write.start, ti->len, data);
341 	if (ret)
342 		goto out;
343 	ret = fn(ti, dc->flush.dev, dc->flush.start, ti->len, data);
344 	if (ret)
345 		goto out;
346 
347 out:
348 	return ret;
349 }
350 
351 static struct target_type delay_target = {
352 	.name	     = "delay",
353 	.version     = {1, 2, 1},
354 	.features    = DM_TARGET_PASSES_INTEGRITY,
355 	.module      = THIS_MODULE,
356 	.ctr	     = delay_ctr,
357 	.dtr	     = delay_dtr,
358 	.map	     = delay_map,
359 	.presuspend  = delay_presuspend,
360 	.resume	     = delay_resume,
361 	.status	     = delay_status,
362 	.iterate_devices = delay_iterate_devices,
363 };
364 
365 static int __init dm_delay_init(void)
366 {
367 	int r;
368 
369 	r = dm_register_target(&delay_target);
370 	if (r < 0) {
371 		DMERR("register failed %d", r);
372 		goto bad_register;
373 	}
374 
375 	return 0;
376 
377 bad_register:
378 	return r;
379 }
380 
381 static void __exit dm_delay_exit(void)
382 {
383 	dm_unregister_target(&delay_target);
384 }
385 
386 /* Module hooks */
387 module_init(dm_delay_init);
388 module_exit(dm_delay_exit);
389 
390 MODULE_DESCRIPTION(DM_NAME " delay target");
391 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
392 MODULE_LICENSE("GPL");
393