xref: /openbmc/linux/drivers/md/dm-delay.c (revision 9a29f5fc)
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 		dm_submit_bio_remap(bio, NULL);
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 	if (dc->kdelayd_wq)
125 		destroy_workqueue(dc->kdelayd_wq);
126 
127 	if (dc->read.dev)
128 		dm_put_device(ti, dc->read.dev);
129 	if (dc->write.dev)
130 		dm_put_device(ti, dc->write.dev);
131 	if (dc->flush.dev)
132 		dm_put_device(ti, dc->flush.dev);
133 
134 	mutex_destroy(&dc->timer_lock);
135 
136 	kfree(dc);
137 }
138 
139 static int delay_class_ctr(struct dm_target *ti, struct delay_class *c, char **argv)
140 {
141 	int ret;
142 	unsigned long long tmpll;
143 	char dummy;
144 
145 	if (sscanf(argv[1], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
146 		ti->error = "Invalid device sector";
147 		return -EINVAL;
148 	}
149 	c->start = tmpll;
150 
151 	if (sscanf(argv[2], "%u%c", &c->delay, &dummy) != 1) {
152 		ti->error = "Invalid delay";
153 		return -EINVAL;
154 	}
155 
156 	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &c->dev);
157 	if (ret) {
158 		ti->error = "Device lookup failed";
159 		return ret;
160 	}
161 
162 	return 0;
163 }
164 
165 /*
166  * Mapping parameters:
167  *    <device> <offset> <delay> [<write_device> <write_offset> <write_delay>]
168  *
169  * With separate write parameters, the first set is only used for reads.
170  * Offsets are specified in sectors.
171  * Delays are specified in milliseconds.
172  */
173 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
174 {
175 	struct delay_c *dc;
176 	int ret;
177 
178 	if (argc != 3 && argc != 6 && argc != 9) {
179 		ti->error = "Requires exactly 3, 6 or 9 arguments";
180 		return -EINVAL;
181 	}
182 
183 	dc = kzalloc(sizeof(*dc), GFP_KERNEL);
184 	if (!dc) {
185 		ti->error = "Cannot allocate context";
186 		return -ENOMEM;
187 	}
188 
189 	ti->private = dc;
190 	timer_setup(&dc->delay_timer, handle_delayed_timer, 0);
191 	INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
192 	INIT_LIST_HEAD(&dc->delayed_bios);
193 	mutex_init(&dc->timer_lock);
194 	atomic_set(&dc->may_delay, 1);
195 	dc->argc = argc;
196 
197 	ret = delay_class_ctr(ti, &dc->read, argv);
198 	if (ret)
199 		goto bad;
200 
201 	if (argc == 3) {
202 		ret = delay_class_ctr(ti, &dc->write, argv);
203 		if (ret)
204 			goto bad;
205 		ret = delay_class_ctr(ti, &dc->flush, argv);
206 		if (ret)
207 			goto bad;
208 		goto out;
209 	}
210 
211 	ret = delay_class_ctr(ti, &dc->write, argv + 3);
212 	if (ret)
213 		goto bad;
214 	if (argc == 6) {
215 		ret = delay_class_ctr(ti, &dc->flush, argv + 3);
216 		if (ret)
217 			goto bad;
218 		goto out;
219 	}
220 
221 	ret = delay_class_ctr(ti, &dc->flush, argv + 6);
222 	if (ret)
223 		goto bad;
224 
225 out:
226 	dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
227 	if (!dc->kdelayd_wq) {
228 		ret = -EINVAL;
229 		DMERR("Couldn't start kdelayd");
230 		goto bad;
231 	}
232 
233 	ti->num_flush_bios = 1;
234 	ti->num_discard_bios = 1;
235 	ti->accounts_remapped_io = true;
236 	ti->per_io_data_size = sizeof(struct dm_delay_info);
237 	return 0;
238 
239 bad:
240 	delay_dtr(ti);
241 	return ret;
242 }
243 
244 static int delay_bio(struct delay_c *dc, struct delay_class *c, struct bio *bio)
245 {
246 	struct dm_delay_info *delayed;
247 	unsigned long expires = 0;
248 
249 	if (!c->delay || !atomic_read(&dc->may_delay))
250 		return DM_MAPIO_REMAPPED;
251 
252 	delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
253 
254 	delayed->context = dc;
255 	delayed->expires = expires = jiffies + msecs_to_jiffies(c->delay);
256 
257 	mutex_lock(&delayed_bios_lock);
258 	c->ops++;
259 	list_add_tail(&delayed->list, &dc->delayed_bios);
260 	mutex_unlock(&delayed_bios_lock);
261 
262 	queue_timeout(dc, expires);
263 
264 	return DM_MAPIO_SUBMITTED;
265 }
266 
267 static void delay_presuspend(struct dm_target *ti)
268 {
269 	struct delay_c *dc = ti->private;
270 
271 	atomic_set(&dc->may_delay, 0);
272 	del_timer_sync(&dc->delay_timer);
273 	flush_bios(flush_delayed_bios(dc, 1));
274 }
275 
276 static void delay_resume(struct dm_target *ti)
277 {
278 	struct delay_c *dc = ti->private;
279 
280 	atomic_set(&dc->may_delay, 1);
281 }
282 
283 static int delay_map(struct dm_target *ti, struct bio *bio)
284 {
285 	struct delay_c *dc = ti->private;
286 	struct delay_class *c;
287 	struct dm_delay_info *delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
288 
289 	if (bio_data_dir(bio) == WRITE) {
290 		if (unlikely(bio->bi_opf & REQ_PREFLUSH))
291 			c = &dc->flush;
292 		else
293 			c = &dc->write;
294 	} else {
295 		c = &dc->read;
296 	}
297 	delayed->class = c;
298 	bio_set_dev(bio, c->dev->bdev);
299 	bio->bi_iter.bi_sector = c->start + dm_target_offset(ti, bio->bi_iter.bi_sector);
300 
301 	return delay_bio(dc, c, bio);
302 }
303 
304 #define DMEMIT_DELAY_CLASS(c) \
305 	DMEMIT("%s %llu %u", (c)->dev->name, (unsigned long long)(c)->start, (c)->delay)
306 
307 static void delay_status(struct dm_target *ti, status_type_t type,
308 			 unsigned status_flags, char *result, unsigned maxlen)
309 {
310 	struct delay_c *dc = ti->private;
311 	int sz = 0;
312 
313 	switch (type) {
314 	case STATUSTYPE_INFO:
315 		DMEMIT("%u %u %u", dc->read.ops, dc->write.ops, dc->flush.ops);
316 		break;
317 
318 	case STATUSTYPE_TABLE:
319 		DMEMIT_DELAY_CLASS(&dc->read);
320 		if (dc->argc >= 6) {
321 			DMEMIT(" ");
322 			DMEMIT_DELAY_CLASS(&dc->write);
323 		}
324 		if (dc->argc >= 9) {
325 			DMEMIT(" ");
326 			DMEMIT_DELAY_CLASS(&dc->flush);
327 		}
328 		break;
329 
330 	case STATUSTYPE_IMA:
331 		*result = '\0';
332 		break;
333 	}
334 }
335 
336 static int delay_iterate_devices(struct dm_target *ti,
337 				 iterate_devices_callout_fn fn, void *data)
338 {
339 	struct delay_c *dc = ti->private;
340 	int ret = 0;
341 
342 	ret = fn(ti, dc->read.dev, dc->read.start, ti->len, data);
343 	if (ret)
344 		goto out;
345 	ret = fn(ti, dc->write.dev, dc->write.start, ti->len, data);
346 	if (ret)
347 		goto out;
348 	ret = fn(ti, dc->flush.dev, dc->flush.start, ti->len, data);
349 	if (ret)
350 		goto out;
351 
352 out:
353 	return ret;
354 }
355 
356 static struct target_type delay_target = {
357 	.name	     = "delay",
358 	.version     = {1, 3, 0},
359 	.features    = DM_TARGET_PASSES_INTEGRITY,
360 	.module      = THIS_MODULE,
361 	.ctr	     = delay_ctr,
362 	.dtr	     = delay_dtr,
363 	.map	     = delay_map,
364 	.presuspend  = delay_presuspend,
365 	.resume	     = delay_resume,
366 	.status	     = delay_status,
367 	.iterate_devices = delay_iterate_devices,
368 };
369 
370 static int __init dm_delay_init(void)
371 {
372 	int r;
373 
374 	r = dm_register_target(&delay_target);
375 	if (r < 0) {
376 		DMERR("register failed %d", r);
377 		goto bad_register;
378 	}
379 
380 	return 0;
381 
382 bad_register:
383 	return r;
384 }
385 
386 static void __exit dm_delay_exit(void)
387 {
388 	dm_unregister_target(&delay_target);
389 }
390 
391 /* Module hooks */
392 module_init(dm_delay_init);
393 module_exit(dm_delay_exit);
394 
395 MODULE_DESCRIPTION(DM_NAME " delay target");
396 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
397 MODULE_LICENSE("GPL");
398