1 // SPDX-License-Identifier: GPL-2.0-only
3  * Copyright (C) 2005-2007 Red Hat GmbH
17 #include <linux/delay.h>
19 #include <linux/device-mapper.h>
21 #define DM_MSG_PREFIX "delay"
28 	unsigned int delay;  member
43 	struct delay_class read;  member
61 	queue_work(dc->kdelayd_wq, &dc->flush_expired_bios);  in handle_delayed_timer()
66 	timer_reduce(&dc->delay_timer, expires);  in queue_timeout()
71 	return !!dc->worker;  in delay_is_fast()
79 		n = bio->bi_next;  in flush_bios()
80 		bio->bi_next = NULL;  in flush_bios()
95 	mutex_lock(&dc->process_bios_lock);  in flush_delayed_bios()
96 	spin_lock(&dc->delayed_bios_lock);  in flush_delayed_bios()
97 	list_replace_init(&dc->delayed_bios, &local_list);  in flush_delayed_bios()
98 	spin_unlock(&dc->delayed_bios_lock);  in flush_delayed_bios()
101 		if (flush_all || time_after_eq(jiffies, delayed->expires)) {  in flush_delayed_bios()
104 			list_del(&delayed->list);  in flush_delayed_bios()
106 			delayed->class->ops--;  in flush_delayed_bios()
113 				next_expires = delayed->expires;  in flush_delayed_bios()
115 				next_expires = min(next_expires, delayed->expires);  in flush_delayed_bios()
119 	spin_lock(&dc->delayed_bios_lock);  in flush_delayed_bios()
120 	list_splice(&local_list, &dc->delayed_bios);  in flush_delayed_bios()
121 	spin_unlock(&dc->delayed_bios_lock);  in flush_delayed_bios()
122 	mutex_unlock(&dc->process_bios_lock);  in flush_delayed_bios()
136 		spin_lock(&dc->delayed_bios_lock);  in flush_worker_fn()
137 		if (unlikely(list_empty(&dc->delayed_bios))) {  in flush_worker_fn()
139 			spin_unlock(&dc->delayed_bios_lock);  in flush_worker_fn()
142 			spin_unlock(&dc->delayed_bios_lock);  in flush_worker_fn()
143 			fsleep(dc->worker_sleep_us);  in flush_worker_fn()
161 	struct delay_c *dc = ti->private;  in delay_dtr()
163 	if (dc->kdelayd_wq) {  in delay_dtr()
164 		timer_shutdown_sync(&dc->delay_timer);  in delay_dtr()
165 		destroy_workqueue(dc->kdelayd_wq);  in delay_dtr()
168 	if (dc->read.dev)  in delay_dtr()
169 		dm_put_device(ti, dc->read.dev);  in delay_dtr()
170 	if (dc->write.dev)  in delay_dtr()
171 		dm_put_device(ti, dc->write.dev);  in delay_dtr()
172 	if (dc->flush.dev)  in delay_dtr()
173 		dm_put_device(ti, dc->flush.dev);  in delay_dtr()
174 	if (dc->worker)  in delay_dtr()
175 		kthread_stop(dc->worker);  in delay_dtr()
177 	mutex_destroy(&dc->process_bios_lock);  in delay_dtr()
189 		ti->error = "Invalid device sector";  in delay_class_ctr()
190 		return -EINVAL;  in delay_class_ctr()
192 	c->start = tmpll;  in delay_class_ctr()
194 	if (sscanf(argv[2], "%u%c", &c->delay, &dummy) != 1) {  in delay_class_ctr()
195 		ti->error = "Invalid delay";  in delay_class_ctr()
196 		return -EINVAL;  in delay_class_ctr()
199 	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &c->dev);  in delay_class_ctr()
201 		ti->error = "Device lookup failed";  in delay_class_ctr()
210  *    <device> <offset> <delay> [<write_device> <write_offset> <write_delay>]
223 		ti->error = "Requires exactly 3, 6 or 9 arguments";  in delay_ctr()
224 		return -EINVAL;  in delay_ctr()
229 		ti->error = "Cannot allocate context";  in delay_ctr()
230 		return -ENOMEM;  in delay_ctr()
233 	ti->private = dc;  in delay_ctr()
234 	INIT_LIST_HEAD(&dc->delayed_bios);  in delay_ctr()
235 	mutex_init(&dc->process_bios_lock);  in delay_ctr()
236 	spin_lock_init(&dc->delayed_bios_lock);  in delay_ctr()
237 	dc->may_delay = true;  in delay_ctr()
238 	dc->argc = argc;  in delay_ctr()
240 	ret = delay_class_ctr(ti, &dc->read, argv);  in delay_ctr()
243 	min_delay = max_delay = dc->read.delay;  in delay_ctr()
246 		ret = delay_class_ctr(ti, &dc->write, argv);  in delay_ctr()
249 		ret = delay_class_ctr(ti, &dc->flush, argv);  in delay_ctr()
255 	ret = delay_class_ctr(ti, &dc->write, argv + 3);  in delay_ctr()
258 	max_delay = max(max_delay, dc->write.delay);  in delay_ctr()
259 	min_delay = min_not_zero(min_delay, dc->write.delay);  in delay_ctr()
262 		ret = delay_class_ctr(ti, &dc->flush, argv + 3);  in delay_ctr()
268 	ret = delay_class_ctr(ti, &dc->flush, argv + 6);  in delay_ctr()
271 	max_delay = max(max_delay, dc->flush.delay);  in delay_ctr()
272 	min_delay = min_not_zero(min_delay, dc->flush.delay);  in delay_ctr()
277 			dc->worker_sleep_us = 1000;  in delay_ctr()
279 			dc->worker_sleep_us = (min_delay * 1000) >> SLEEP_SHIFT;  in delay_ctr()
284 		dc->worker = kthread_run(&flush_worker_fn, dc, "dm-delay-flush-worker");  in delay_ctr()
285 		if (IS_ERR(dc->worker)) {  in delay_ctr()
286 			ret = PTR_ERR(dc->worker);  in delay_ctr()
287 			dc->worker = NULL;  in delay_ctr()
291 		timer_setup(&dc->delay_timer, handle_delayed_timer, 0);  in delay_ctr()
292 		INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);  in delay_ctr()
293 		dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);  in delay_ctr()
294 		if (!dc->kdelayd_wq) {  in delay_ctr()
295 			ret = -EINVAL;  in delay_ctr()
301 	ti->num_flush_bios = 1;  in delay_ctr()
302 	ti->num_discard_bios = 1;  in delay_ctr()
303 	ti->accounts_remapped_io = true;  in delay_ctr()
304 	ti->per_io_data_size = sizeof(struct dm_delay_info);  in delay_ctr()
317 	if (!c->delay)  in delay_bio()
322 	delayed->context = dc;  in delay_bio()
323 	delayed->expires = expires = jiffies + msecs_to_jiffies(c->delay);  in delay_bio()
325 	spin_lock(&dc->delayed_bios_lock);  in delay_bio()
326 	if (unlikely(!dc->may_delay)) {  in delay_bio()
327 		spin_unlock(&dc->delayed_bios_lock);  in delay_bio()
330 	c->ops++;  in delay_bio()
331 	list_add_tail(&delayed->list, &dc->delayed_bios);  in delay_bio()
332 	spin_unlock(&dc->delayed_bios_lock);  in delay_bio()
335 		wake_up_process(dc->worker);  in delay_bio()
344 	struct delay_c *dc = ti->private;  in delay_presuspend()
346 	spin_lock(&dc->delayed_bios_lock);  in delay_presuspend()
347 	dc->may_delay = false;  in delay_presuspend()
348 	spin_unlock(&dc->delayed_bios_lock);  in delay_presuspend()
351 		timer_delete(&dc->delay_timer);  in delay_presuspend()
357 	struct delay_c *dc = ti->private;  in delay_resume()
359 	dc->may_delay = true;  in delay_resume()
364 	struct delay_c *dc = ti->private;  in delay_map()
369 		if (unlikely(bio->bi_opf & REQ_PREFLUSH))  in delay_map()
370 			c = &dc->flush;  in delay_map()
372 			c = &dc->write;  in delay_map()
374 		c = &dc->read;  in delay_map()
376 	delayed->class = c;  in delay_map()
377 	bio_set_dev(bio, c->dev->bdev);  in delay_map()
378 	bio->bi_iter.bi_sector = c->start + dm_target_offset(ti, bio->bi_iter.bi_sector);  in delay_map()
387 	struct delay_c *dc = ti->private;  in delay_report_zones()
388 	struct delay_class *c = &dc->read;  in delay_report_zones()
390 	return dm_report_zones(c->dev->bdev, c->start,  in delay_report_zones()
391 			c->start + dm_target_offset(ti, args->next_sector),  in delay_report_zones()
399 	DMEMIT("%s %llu %u", (c)->dev->name, (unsigned long long)(c)->start, (c)->delay)
404 	struct delay_c *dc = ti->private;  in delay_status()
409 		DMEMIT("%u %u %u", dc->read.ops, dc->write.ops, dc->flush.ops);  in delay_status()
413 		DMEMIT_DELAY_CLASS(&dc->read);  in delay_status()
414 		if (dc->argc >= 6) {  in delay_status()
416 			DMEMIT_DELAY_CLASS(&dc->write);  in delay_status()
418 		if (dc->argc >= 9) {  in delay_status()
420 			DMEMIT_DELAY_CLASS(&dc->flush);  in delay_status()
433 	struct delay_c *dc = ti->private;  in delay_iterate_devices()
436 	ret = fn(ti, dc->read.dev, dc->read.start, ti->len, data);  in delay_iterate_devices()
439 	ret = fn(ti, dc->write.dev, dc->write.start, ti->len, data);  in delay_iterate_devices()
442 	ret = fn(ti, dc->flush.dev, dc->flush.start, ti->len, data);  in delay_iterate_devices()
451 	.name	     = "delay",
464 module_dm(delay);
466 MODULE_DESCRIPTION(DM_NAME " delay target");