Commit | Line | Data |
---|---|---|
26b9f228 HM |
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 | ||
586e80e6 MP |
16 | #include <linux/device-mapper.h> |
17 | ||
26b9f228 HM |
18 | #define DM_MSG_PREFIX "delay" |
19 | ||
20 | struct delay_c { | |
21 | struct timer_list delay_timer; | |
ac818646 | 22 | struct mutex timer_lock; |
26b9f228 HM |
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 | ||
028867ac | 39 | struct dm_delay_info { |
26b9f228 HM |
40 | struct delay_c *context; |
41 | struct list_head list; | |
42 | struct bio *bio; | |
43 | unsigned long expires; | |
44 | }; | |
45 | ||
46 | static DEFINE_MUTEX(delayed_bios_lock); | |
47 | ||
48 | static struct workqueue_struct *kdelayd_wq; | |
49 | static struct kmem_cache *delayed_cache; | |
50 | ||
51 | static 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 | ||
58 | static void queue_timeout(struct delay_c *dc, unsigned long expires) | |
59 | { | |
ac818646 | 60 | mutex_lock(&dc->timer_lock); |
26b9f228 HM |
61 | |
62 | if (!timer_pending(&dc->delay_timer) || expires < dc->delay_timer.expires) | |
63 | mod_timer(&dc->delay_timer, expires); | |
64 | ||
ac818646 | 65 | mutex_unlock(&dc->timer_lock); |
26b9f228 HM |
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 | { | |
028867ac | 82 | struct dm_delay_info *delayed, *next; |
26b9f228 HM |
83 | unsigned long next_expires = 0; |
84 | int start_timer = 0; | |
051814c6 | 85 | struct bio_list flush_bios = { }; |
26b9f228 HM |
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 | ||
115 | static 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 | */ | |
130 | static 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], dc->start_read, ti->len, | |
160 | dm_table_get_mode(ti->table), &dc->dev_read)) { | |
161 | ti->error = "Device lookup failed"; | |
162 | goto bad; | |
163 | } | |
164 | ||
2e64a0f9 DM |
165 | dc->dev_write = NULL; |
166 | if (argc == 3) | |
26b9f228 | 167 | goto out; |
26b9f228 HM |
168 | |
169 | if (sscanf(argv[4], "%llu", &tmpll) != 1) { | |
170 | ti->error = "Invalid write device sector"; | |
2e64a0f9 | 171 | goto bad_dev_read; |
26b9f228 HM |
172 | } |
173 | dc->start_write = tmpll; | |
174 | ||
175 | if (sscanf(argv[5], "%u", &dc->write_delay) != 1) { | |
176 | ti->error = "Invalid write delay"; | |
2e64a0f9 | 177 | goto bad_dev_read; |
26b9f228 HM |
178 | } |
179 | ||
180 | if (dm_get_device(ti, argv[3], dc->start_write, ti->len, | |
181 | dm_table_get_mode(ti->table), &dc->dev_write)) { | |
182 | ti->error = "Write device lookup failed"; | |
2e64a0f9 | 183 | goto bad_dev_read; |
26b9f228 HM |
184 | } |
185 | ||
186 | out: | |
187 | dc->delayed_pool = mempool_create_slab_pool(128, delayed_cache); | |
188 | if (!dc->delayed_pool) { | |
189 | DMERR("Couldn't create delayed bio pool."); | |
2e64a0f9 | 190 | goto bad_dev_write; |
26b9f228 HM |
191 | } |
192 | ||
ac818646 | 193 | setup_timer(&dc->delay_timer, handle_delayed_timer, (unsigned long)dc); |
26b9f228 HM |
194 | |
195 | INIT_WORK(&dc->flush_expired_bios, flush_expired_bios); | |
196 | INIT_LIST_HEAD(&dc->delayed_bios); | |
ac818646 | 197 | mutex_init(&dc->timer_lock); |
26b9f228 HM |
198 | atomic_set(&dc->may_delay, 1); |
199 | ||
200 | ti->private = dc; | |
201 | return 0; | |
202 | ||
2e64a0f9 DM |
203 | bad_dev_write: |
204 | if (dc->dev_write) | |
205 | dm_put_device(ti, dc->dev_write); | |
206 | bad_dev_read: | |
207 | dm_put_device(ti, dc->dev_read); | |
26b9f228 HM |
208 | bad: |
209 | kfree(dc); | |
210 | return -EINVAL; | |
211 | } | |
212 | ||
213 | static void delay_dtr(struct dm_target *ti) | |
214 | { | |
215 | struct delay_c *dc = ti->private; | |
216 | ||
217 | flush_workqueue(kdelayd_wq); | |
218 | ||
219 | dm_put_device(ti, dc->dev_read); | |
220 | ||
221 | if (dc->dev_write) | |
222 | dm_put_device(ti, dc->dev_write); | |
223 | ||
224 | mempool_destroy(dc->delayed_pool); | |
225 | kfree(dc); | |
226 | } | |
227 | ||
228 | static int delay_bio(struct delay_c *dc, int delay, struct bio *bio) | |
229 | { | |
028867ac | 230 | struct dm_delay_info *delayed; |
26b9f228 HM |
231 | unsigned long expires = 0; |
232 | ||
233 | if (!delay || !atomic_read(&dc->may_delay)) | |
234 | return 1; | |
235 | ||
236 | delayed = mempool_alloc(dc->delayed_pool, GFP_NOIO); | |
237 | ||
238 | delayed->context = dc; | |
239 | delayed->bio = bio; | |
240 | delayed->expires = expires = jiffies + (delay * HZ / 1000); | |
241 | ||
242 | mutex_lock(&delayed_bios_lock); | |
243 | ||
244 | if (bio_data_dir(bio) == WRITE) | |
245 | dc->writes++; | |
246 | else | |
247 | dc->reads++; | |
248 | ||
249 | list_add_tail(&delayed->list, &dc->delayed_bios); | |
250 | ||
251 | mutex_unlock(&delayed_bios_lock); | |
252 | ||
253 | queue_timeout(dc, expires); | |
254 | ||
255 | return 0; | |
256 | } | |
257 | ||
258 | static void delay_presuspend(struct dm_target *ti) | |
259 | { | |
260 | struct delay_c *dc = ti->private; | |
261 | ||
262 | atomic_set(&dc->may_delay, 0); | |
263 | del_timer_sync(&dc->delay_timer); | |
264 | flush_bios(flush_delayed_bios(dc, 1)); | |
265 | } | |
266 | ||
267 | static void delay_resume(struct dm_target *ti) | |
268 | { | |
269 | struct delay_c *dc = ti->private; | |
270 | ||
271 | atomic_set(&dc->may_delay, 1); | |
272 | } | |
273 | ||
274 | static int delay_map(struct dm_target *ti, struct bio *bio, | |
275 | union map_info *map_context) | |
276 | { | |
277 | struct delay_c *dc = ti->private; | |
278 | ||
279 | if ((bio_data_dir(bio) == WRITE) && (dc->dev_write)) { | |
280 | bio->bi_bdev = dc->dev_write->bdev; | |
281 | bio->bi_sector = dc->start_write + | |
282 | (bio->bi_sector - ti->begin); | |
283 | ||
284 | return delay_bio(dc, dc->write_delay, bio); | |
285 | } | |
286 | ||
287 | bio->bi_bdev = dc->dev_read->bdev; | |
288 | bio->bi_sector = dc->start_read + | |
289 | (bio->bi_sector - ti->begin); | |
290 | ||
291 | return delay_bio(dc, dc->read_delay, bio); | |
292 | } | |
293 | ||
294 | static int delay_status(struct dm_target *ti, status_type_t type, | |
295 | char *result, unsigned maxlen) | |
296 | { | |
297 | struct delay_c *dc = ti->private; | |
298 | int sz = 0; | |
299 | ||
300 | switch (type) { | |
301 | case STATUSTYPE_INFO: | |
302 | DMEMIT("%u %u", dc->reads, dc->writes); | |
303 | break; | |
304 | ||
305 | case STATUSTYPE_TABLE: | |
306 | DMEMIT("%s %llu %u", dc->dev_read->name, | |
307 | (unsigned long long) dc->start_read, | |
308 | dc->read_delay); | |
309 | if (dc->dev_write) | |
79662d1e | 310 | DMEMIT(" %s %llu %u", dc->dev_write->name, |
26b9f228 HM |
311 | (unsigned long long) dc->start_write, |
312 | dc->write_delay); | |
313 | break; | |
314 | } | |
315 | ||
316 | return 0; | |
317 | } | |
318 | ||
319 | static struct target_type delay_target = { | |
320 | .name = "delay", | |
321 | .version = {1, 0, 2}, | |
322 | .module = THIS_MODULE, | |
323 | .ctr = delay_ctr, | |
324 | .dtr = delay_dtr, | |
325 | .map = delay_map, | |
326 | .presuspend = delay_presuspend, | |
327 | .resume = delay_resume, | |
328 | .status = delay_status, | |
329 | }; | |
330 | ||
331 | static int __init dm_delay_init(void) | |
332 | { | |
333 | int r = -ENOMEM; | |
334 | ||
335 | kdelayd_wq = create_workqueue("kdelayd"); | |
336 | if (!kdelayd_wq) { | |
337 | DMERR("Couldn't start kdelayd"); | |
338 | goto bad_queue; | |
339 | } | |
340 | ||
028867ac | 341 | delayed_cache = KMEM_CACHE(dm_delay_info, 0); |
26b9f228 HM |
342 | if (!delayed_cache) { |
343 | DMERR("Couldn't create delayed bio cache."); | |
344 | goto bad_memcache; | |
345 | } | |
346 | ||
347 | r = dm_register_target(&delay_target); | |
348 | if (r < 0) { | |
349 | DMERR("register failed %d", r); | |
350 | goto bad_register; | |
351 | } | |
352 | ||
353 | return 0; | |
354 | ||
355 | bad_register: | |
356 | kmem_cache_destroy(delayed_cache); | |
357 | bad_memcache: | |
358 | destroy_workqueue(kdelayd_wq); | |
359 | bad_queue: | |
360 | return r; | |
361 | } | |
362 | ||
363 | static void __exit dm_delay_exit(void) | |
364 | { | |
10d3bd09 | 365 | dm_unregister_target(&delay_target); |
26b9f228 HM |
366 | kmem_cache_destroy(delayed_cache); |
367 | destroy_workqueue(kdelayd_wq); | |
368 | } | |
369 | ||
370 | /* Module hooks */ | |
371 | module_init(dm_delay_init); | |
372 | module_exit(dm_delay_exit); | |
373 | ||
374 | MODULE_DESCRIPTION(DM_NAME " delay target"); | |
375 | MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>"); | |
376 | MODULE_LICENSE("GPL"); |