Loading...
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
20struct 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
39struct dm_delay_info {
40 struct delay_c *context;
41 struct list_head list;
42 struct bio *bio;
43 unsigned long expires;
44};
45
46static DEFINE_MUTEX(delayed_bios_lock);
47
48static struct workqueue_struct *kdelayd_wq;
49static struct kmem_cache *delayed_cache;
50
51static 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
58static 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
68static 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
80static 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
115static 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 */
130static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
131{
132 struct delay_c *dc;
133 unsigned long long tmpll;
134
135 if (argc != 3 && argc != 6) {
136 ti->error = "requires exactly 3 or 6 arguments";
137 return -EINVAL;
138 }
139
140 dc = kmalloc(sizeof(*dc), GFP_KERNEL);
141 if (!dc) {
142 ti->error = "Cannot allocate context";
143 return -ENOMEM;
144 }
145
146 dc->reads = dc->writes = 0;
147
148 if (sscanf(argv[1], "%llu", &tmpll) != 1) {
149 ti->error = "Invalid device sector";
150 goto bad;
151 }
152 dc->start_read = tmpll;
153
154 if (sscanf(argv[2], "%u", &dc->read_delay) != 1) {
155 ti->error = "Invalid delay";
156 goto bad;
157 }
158
159 if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
160 &dc->dev_read)) {
161 ti->error = "Device lookup failed";
162 goto bad;
163 }
164
165 dc->dev_write = NULL;
166 if (argc == 3)
167 goto out;
168
169 if (sscanf(argv[4], "%llu", &tmpll) != 1) {
170 ti->error = "Invalid write device sector";
171 goto bad_dev_read;
172 }
173 dc->start_write = tmpll;
174
175 if (sscanf(argv[5], "%u", &dc->write_delay) != 1) {
176 ti->error = "Invalid write delay";
177 goto bad_dev_read;
178 }
179
180 if (dm_get_device(ti, argv[3], dm_table_get_mode(ti->table),
181 &dc->dev_write)) {
182 ti->error = "Write device lookup failed";
183 goto bad_dev_read;
184 }
185
186out:
187 dc->delayed_pool = mempool_create_slab_pool(128, delayed_cache);
188 if (!dc->delayed_pool) {
189 DMERR("Couldn't create delayed bio pool.");
190 goto bad_dev_write;
191 }
192
193 setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc);
194
195 INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
196 INIT_LIST_HEAD(&dc->delayed_bios);
197 mutex_init(&dc->timer_lock);
198 atomic_set(&dc->may_delay, 1);
199
200 ti->num_flush_requests = 1;
201 ti->num_discard_requests = 1;
202 ti->private = dc;
203 return 0;
204
205bad_dev_write:
206 if (dc->dev_write)
207 dm_put_device(ti, dc->dev_write);
208bad_dev_read:
209 dm_put_device(ti, dc->dev_read);
210bad:
211 kfree(dc);
212 return -EINVAL;
213}
214
215static void delay_dtr(struct dm_target *ti)
216{
217 struct delay_c *dc = ti->private;
218
219 flush_workqueue(kdelayd_wq);
220
221 dm_put_device(ti, dc->dev_read);
222
223 if (dc->dev_write)
224 dm_put_device(ti, dc->dev_write);
225
226 mempool_destroy(dc->delayed_pool);
227 kfree(dc);
228}
229
230static int delay_bio(struct delay_c *dc, int delay, struct bio *bio)
231{
232 struct dm_delay_info *delayed;
233 unsigned long expires = 0;
234
235 if (!delay || !atomic_read(&dc->may_delay))
236 return 1;
237
238 delayed = mempool_alloc(dc->delayed_pool, GFP_NOIO);
239
240 delayed->context = dc;
241 delayed->bio = bio;
242 delayed->expires = expires = jiffies + (delay * HZ / 1000);
243
244 mutex_lock(&delayed_bios_lock);
245
246 if (bio_data_dir(bio) == WRITE)
247 dc->writes++;
248 else
249 dc->reads++;
250
251 list_add_tail(&delayed->list, &dc->delayed_bios);
252
253 mutex_unlock(&delayed_bios_lock);
254
255 queue_timeout(dc, expires);
256
257 return 0;
258}
259
260static void delay_presuspend(struct dm_target *ti)
261{
262 struct delay_c *dc = ti->private;
263
264 atomic_set(&dc->may_delay, 0);
265 del_timer_sync(&dc->delay_timer);
266 flush_bios(flush_delayed_bios(dc, 1));
267}
268
269static void delay_resume(struct dm_target *ti)
270{
271 struct delay_c *dc = ti->private;
272
273 atomic_set(&dc->may_delay, 1);
274}
275
276static int delay_map(struct dm_target *ti, struct bio *bio,
277 union map_info *map_context)
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
296static int delay_status(struct dm_target *ti, status_type_t type,
297 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 return 0;
319}
320
321static int delay_iterate_devices(struct dm_target *ti,
322 iterate_devices_callout_fn fn, void *data)
323{
324 struct delay_c *dc = ti->private;
325 int ret = 0;
326
327 ret = fn(ti, dc->dev_read, dc->start_read, ti->len, data);
328 if (ret)
329 goto out;
330
331 if (dc->dev_write)
332 ret = fn(ti, dc->dev_write, dc->start_write, ti->len, data);
333
334out:
335 return ret;
336}
337
338static struct target_type delay_target = {
339 .name = "delay",
340 .version = {1, 1, 0},
341 .module = THIS_MODULE,
342 .ctr = delay_ctr,
343 .dtr = delay_dtr,
344 .map = delay_map,
345 .presuspend = delay_presuspend,
346 .resume = delay_resume,
347 .status = delay_status,
348 .iterate_devices = delay_iterate_devices,
349};
350
351static int __init dm_delay_init(void)
352{
353 int r = -ENOMEM;
354
355 kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
356 if (!kdelayd_wq) {
357 DMERR("Couldn't start kdelayd");
358 goto bad_queue;
359 }
360
361 delayed_cache = KMEM_CACHE(dm_delay_info, 0);
362 if (!delayed_cache) {
363 DMERR("Couldn't create delayed bio cache.");
364 goto bad_memcache;
365 }
366
367 r = dm_register_target(&delay_target);
368 if (r < 0) {
369 DMERR("register failed %d", r);
370 goto bad_register;
371 }
372
373 return 0;
374
375bad_register:
376 kmem_cache_destroy(delayed_cache);
377bad_memcache:
378 destroy_workqueue(kdelayd_wq);
379bad_queue:
380 return r;
381}
382
383static void __exit dm_delay_exit(void)
384{
385 dm_unregister_target(&delay_target);
386 kmem_cache_destroy(delayed_cache);
387 destroy_workqueue(kdelayd_wq);
388}
389
390/* Module hooks */
391module_init(dm_delay_init);
392module_exit(dm_delay_exit);
393
394MODULE_DESCRIPTION(DM_NAME " delay target");
395MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
396MODULE_LICENSE("GPL");
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
20struct 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
39struct dm_delay_info {
40 struct delay_c *context;
41 struct list_head list;
42 unsigned long expires;
43};
44
45static DEFINE_MUTEX(delayed_bios_lock);
46
47static void handle_delayed_timer(struct timer_list *t)
48{
49 struct delay_c *dc = from_timer(dc, t, delay_timer);
50
51 queue_work(dc->kdelayd_wq, &dc->flush_expired_bios);
52}
53
54static 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
64static 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
76static 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
112static 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 * Offsets are specified in sectors.
126 * Delays are specified in milliseconds.
127 */
128static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
129{
130 struct delay_c *dc;
131 unsigned long long tmpll;
132 char dummy;
133 int ret;
134
135 if (argc != 3 && argc != 6) {
136 ti->error = "Requires exactly 3 or 6 arguments";
137 return -EINVAL;
138 }
139
140 dc = kmalloc(sizeof(*dc), GFP_KERNEL);
141 if (!dc) {
142 ti->error = "Cannot allocate context";
143 return -ENOMEM;
144 }
145
146 dc->reads = dc->writes = 0;
147
148 ret = -EINVAL;
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 ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
161 &dc->dev_read);
162 if (ret) {
163 ti->error = "Device lookup failed";
164 goto bad;
165 }
166
167 ret = -EINVAL;
168 dc->dev_write = NULL;
169 if (argc == 3)
170 goto out;
171
172 if (sscanf(argv[4], "%llu%c", &tmpll, &dummy) != 1) {
173 ti->error = "Invalid write device sector";
174 goto bad_dev_read;
175 }
176 dc->start_write = tmpll;
177
178 if (sscanf(argv[5], "%u%c", &dc->write_delay, &dummy) != 1) {
179 ti->error = "Invalid write delay";
180 goto bad_dev_read;
181 }
182
183 ret = dm_get_device(ti, argv[3], dm_table_get_mode(ti->table),
184 &dc->dev_write);
185 if (ret) {
186 ti->error = "Write device lookup failed";
187 goto bad_dev_read;
188 }
189
190out:
191 ret = -EINVAL;
192 dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
193 if (!dc->kdelayd_wq) {
194 DMERR("Couldn't start kdelayd");
195 goto bad_queue;
196 }
197
198 timer_setup(&dc->delay_timer, handle_delayed_timer, 0);
199
200 INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
201 INIT_LIST_HEAD(&dc->delayed_bios);
202 mutex_init(&dc->timer_lock);
203 atomic_set(&dc->may_delay, 1);
204
205 ti->num_flush_bios = 1;
206 ti->num_discard_bios = 1;
207 ti->per_io_data_size = sizeof(struct dm_delay_info);
208 ti->private = dc;
209 return 0;
210
211bad_queue:
212 if (dc->dev_write)
213 dm_put_device(ti, dc->dev_write);
214bad_dev_read:
215 dm_put_device(ti, dc->dev_read);
216bad:
217 kfree(dc);
218 return ret;
219}
220
221static void delay_dtr(struct dm_target *ti)
222{
223 struct delay_c *dc = ti->private;
224
225 destroy_workqueue(dc->kdelayd_wq);
226
227 dm_put_device(ti, dc->dev_read);
228
229 if (dc->dev_write)
230 dm_put_device(ti, dc->dev_write);
231
232 mutex_destroy(&dc->timer_lock);
233
234 kfree(dc);
235}
236
237static int delay_bio(struct delay_c *dc, int delay, struct bio *bio)
238{
239 struct dm_delay_info *delayed;
240 unsigned long expires = 0;
241
242 if (!delay || !atomic_read(&dc->may_delay))
243 return DM_MAPIO_REMAPPED;
244
245 delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
246
247 delayed->context = dc;
248 delayed->expires = expires = jiffies + msecs_to_jiffies(delay);
249
250 mutex_lock(&delayed_bios_lock);
251
252 if (bio_data_dir(bio) == WRITE)
253 dc->writes++;
254 else
255 dc->reads++;
256
257 list_add_tail(&delayed->list, &dc->delayed_bios);
258
259 mutex_unlock(&delayed_bios_lock);
260
261 queue_timeout(dc, expires);
262
263 return DM_MAPIO_SUBMITTED;
264}
265
266static void delay_presuspend(struct dm_target *ti)
267{
268 struct delay_c *dc = ti->private;
269
270 atomic_set(&dc->may_delay, 0);
271 del_timer_sync(&dc->delay_timer);
272 flush_bios(flush_delayed_bios(dc, 1));
273}
274
275static void delay_resume(struct dm_target *ti)
276{
277 struct delay_c *dc = ti->private;
278
279 atomic_set(&dc->may_delay, 1);
280}
281
282static int delay_map(struct dm_target *ti, struct bio *bio)
283{
284 struct delay_c *dc = ti->private;
285
286 if ((bio_data_dir(bio) == WRITE) && (dc->dev_write)) {
287 bio_set_dev(bio, dc->dev_write->bdev);
288 if (bio_sectors(bio))
289 bio->bi_iter.bi_sector = dc->start_write +
290 dm_target_offset(ti, bio->bi_iter.bi_sector);
291
292 return delay_bio(dc, dc->write_delay, bio);
293 }
294
295 bio_set_dev(bio, dc->dev_read->bdev);
296 bio->bi_iter.bi_sector = dc->start_read +
297 dm_target_offset(ti, bio->bi_iter.bi_sector);
298
299 return delay_bio(dc, dc->read_delay, bio);
300}
301
302static void delay_status(struct dm_target *ti, status_type_t type,
303 unsigned status_flags, char *result, unsigned maxlen)
304{
305 struct delay_c *dc = ti->private;
306 int sz = 0;
307
308 switch (type) {
309 case STATUSTYPE_INFO:
310 DMEMIT("%u %u", dc->reads, dc->writes);
311 break;
312
313 case STATUSTYPE_TABLE:
314 DMEMIT("%s %llu %u", dc->dev_read->name,
315 (unsigned long long) dc->start_read,
316 dc->read_delay);
317 if (dc->dev_write)
318 DMEMIT(" %s %llu %u", dc->dev_write->name,
319 (unsigned long long) dc->start_write,
320 dc->write_delay);
321 break;
322 }
323}
324
325static int delay_iterate_devices(struct dm_target *ti,
326 iterate_devices_callout_fn fn, void *data)
327{
328 struct delay_c *dc = ti->private;
329 int ret = 0;
330
331 ret = fn(ti, dc->dev_read, dc->start_read, ti->len, data);
332 if (ret)
333 goto out;
334
335 if (dc->dev_write)
336 ret = fn(ti, dc->dev_write, dc->start_write, ti->len, data);
337
338out:
339 return ret;
340}
341
342static struct target_type delay_target = {
343 .name = "delay",
344 .version = {1, 2, 1},
345 .features = DM_TARGET_PASSES_INTEGRITY,
346 .module = THIS_MODULE,
347 .ctr = delay_ctr,
348 .dtr = delay_dtr,
349 .map = delay_map,
350 .presuspend = delay_presuspend,
351 .resume = delay_resume,
352 .status = delay_status,
353 .iterate_devices = delay_iterate_devices,
354};
355
356static int __init dm_delay_init(void)
357{
358 int r;
359
360 r = dm_register_target(&delay_target);
361 if (r < 0) {
362 DMERR("register failed %d", r);
363 goto bad_register;
364 }
365
366 return 0;
367
368bad_register:
369 return r;
370}
371
372static void __exit dm_delay_exit(void)
373{
374 dm_unregister_target(&delay_target);
375}
376
377/* Module hooks */
378module_init(dm_delay_init);
379module_exit(dm_delay_exit);
380
381MODULE_DESCRIPTION(DM_NAME " delay target");
382MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
383MODULE_LICENSE("GPL");