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_class {
21 struct dm_dev *dev;
22 sector_t start;
23 unsigned delay;
24 unsigned ops;
25};
26
27struct 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
42struct dm_delay_info {
43 struct delay_c *context;
44 struct delay_class *class;
45 struct list_head list;
46 unsigned long expires;
47};
48
49static DEFINE_MUTEX(delayed_bios_lock);
50
51static 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
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 dm_submit_bio_remap(bio, NULL);
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 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
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
120static 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
139static 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 */
173static 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
225out:
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
239bad:
240 delay_dtr(ti);
241 return ret;
242}
243
244static 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
267static 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
276static 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
283static 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
307static 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
336static 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
352out:
353 return ret;
354}
355
356static 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
370static 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
382bad_register:
383 return r;
384}
385
386static void __exit dm_delay_exit(void)
387{
388 dm_unregister_target(&delay_target);
389}
390
391/* Module hooks */
392module_init(dm_delay_init);
393module_exit(dm_delay_exit);
394
395MODULE_DESCRIPTION(DM_NAME " delay target");
396MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
397MODULE_LICENSE("GPL");