drivers/md/dm-delay.c

0001 /*
0002  * Copyright (C) 2005-2007 Red Hat GmbH
0003  *
0004  * A target that delays reads and/or writes and can send
0005  * them to different devices.
0006  *
0007  * This file is released under the GPL.
0008  */
0009
0010 #include <linux/module.h>
0011 #include <linux/init.h>
0012 #include <linux/blkdev.h>
0013 #include <linux/bio.h>
0014 #include <linux/slab.h>
0015
0016 #include <linux/device-mapper.h>
0017
0018 #define DM_MSG_PREFIX "delay"
0019
0020 struct delay_class {
0021     struct dm_dev *dev;
0022     sector_t start;
0023     unsigned delay;
0024     unsigned ops;
0025 };
0026
0027 struct delay_c {
0028     struct timer_list delay_timer;
0029     struct mutex timer_lock;
0030     struct workqueue_struct *kdelayd_wq;
0031     struct work_struct flush_expired_bios;
0032     struct list_head delayed_bios;
0033     atomic_t may_delay;
0034
0035     struct delay_class read;
0036     struct delay_class write;
0037     struct delay_class flush;
0038
0039     int argc;
0040 };
0041
0042 struct dm_delay_info {
0043     struct delay_c *context;
0044     struct delay_class *class;
0045     struct list_head list;
0046     unsigned long expires;
0047 };
0048
0049 static DEFINE_MUTEX(delayed_bios_lock);
0050
0051 static void handle_delayed_timer(struct timer_list *t)
0052 {
0053     struct delay_c *dc = from_timer(dc, t, delay_timer);
0054
0055     queue_work(dc->kdelayd_wq, &dc->flush_expired_bios);
0056 }
0057
0058 static void queue_timeout(struct delay_c *dc, unsigned long expires)
0059 {
0060     mutex_lock(&dc->timer_lock);
0061
0062     if (!timer_pending(&dc->delay_timer) || expires < dc->delay_timer.expires)
0063         mod_timer(&dc->delay_timer, expires);
0064
0065     mutex_unlock(&dc->timer_lock);
0066 }
0067
0068 static void flush_bios(struct bio *bio)
0069 {
0070     struct bio *n;
0071
0072     while (bio) {
0073         n = bio->bi_next;
0074         bio->bi_next = NULL;
0075         dm_submit_bio_remap(bio, NULL);
0076         bio = n;
0077     }
0078 }
0079
0080 static struct bio *flush_delayed_bios(struct delay_c *dc, int flush_all)
0081 {
0082     struct dm_delay_info *delayed, *next;
0083     unsigned long next_expires = 0;
0084     unsigned long start_timer = 0;
0085     struct bio_list flush_bios = { };
0086
0087     mutex_lock(&delayed_bios_lock);
0088     list_for_each_entry_safe(delayed, next, &dc->delayed_bios, list) {
0089         if (flush_all || time_after_eq(jiffies, delayed->expires)) {
0090             struct bio *bio = dm_bio_from_per_bio_data(delayed,
0091                         sizeof(struct dm_delay_info));
0092             list_del(&delayed->list);
0093             bio_list_add(&flush_bios, bio);
0094             delayed->class->ops--;
0095             continue;
0096         }
0097
0098         if (!start_timer) {
0099             start_timer = 1;
0100             next_expires = delayed->expires;
0101         } else
0102             next_expires = min(next_expires, delayed->expires);
0103     }
0104     mutex_unlock(&delayed_bios_lock);
0105
0106     if (start_timer)
0107         queue_timeout(dc, next_expires);
0108
0109     return bio_list_get(&flush_bios);
0110 }
0111
0112 static void flush_expired_bios(struct work_struct *work)
0113 {
0114     struct delay_c *dc;
0115
0116     dc = container_of(work, struct delay_c, flush_expired_bios);
0117     flush_bios(flush_delayed_bios(dc, 0));
0118 }
0119
0120 static void delay_dtr(struct dm_target *ti)
0121 {
0122     struct delay_c *dc = ti->private;
0123
0124     if (dc->kdelayd_wq)
0125         destroy_workqueue(dc->kdelayd_wq);
0126
0127     if (dc->read.dev)
0128         dm_put_device(ti, dc->read.dev);
0129     if (dc->write.dev)
0130         dm_put_device(ti, dc->write.dev);
0131     if (dc->flush.dev)
0132         dm_put_device(ti, dc->flush.dev);
0133
0134     mutex_destroy(&dc->timer_lock);
0135
0136     kfree(dc);
0137 }
0138
0139 static int delay_class_ctr(struct dm_target *ti, struct delay_class *c, char **argv)
0140 {
0141     int ret;
0142     unsigned long long tmpll;
0143     char dummy;
0144
0145     if (sscanf(argv[1], "%llu%c", &tmpll, &dummy) != 1 || tmpll != (sector_t)tmpll) {
0146         ti->error = "Invalid device sector";
0147         return -EINVAL;
0148     }
0149     c->start = tmpll;
0150
0151     if (sscanf(argv[2], "%u%c", &c->delay, &dummy) != 1) {
0152         ti->error = "Invalid delay";
0153         return -EINVAL;
0154     }
0155
0156     ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &c->dev);
0157     if (ret) {
0158         ti->error = "Device lookup failed";
0159         return ret;
0160     }
0161
0162     return 0;
0163 }
0164
0165 /*
0166  * Mapping parameters:
0167  *    <device> <offset> <delay> [<write_device> <write_offset> <write_delay>]
0168  *
0169  * With separate write parameters, the first set is only used for reads.
0170  * Offsets are specified in sectors.
0171  * Delays are specified in milliseconds.
0172  */
0173 static int delay_ctr(struct dm_target *ti, unsigned int argc, char **argv)
0174 {
0175     struct delay_c *dc;
0176     int ret;
0177
0178     if (argc != 3 && argc != 6 && argc != 9) {
0179         ti->error = "Requires exactly 3, 6 or 9 arguments";
0180         return -EINVAL;
0181     }
0182
0183     dc = kzalloc(sizeof(*dc), GFP_KERNEL);
0184     if (!dc) {
0185         ti->error = "Cannot allocate context";
0186         return -ENOMEM;
0187     }
0188
0189     ti->private = dc;
0190     timer_setup(&dc->delay_timer, handle_delayed_timer, 0);
0191     INIT_WORK(&dc->flush_expired_bios, flush_expired_bios);
0192     INIT_LIST_HEAD(&dc->delayed_bios);
0193     mutex_init(&dc->timer_lock);
0194     atomic_set(&dc->may_delay, 1);
0195     dc->argc = argc;
0196
0197     ret = delay_class_ctr(ti, &dc->read, argv);
0198     if (ret)
0199         goto bad;
0200
0201     if (argc == 3) {
0202         ret = delay_class_ctr(ti, &dc->write, argv);
0203         if (ret)
0204             goto bad;
0205         ret = delay_class_ctr(ti, &dc->flush, argv);
0206         if (ret)
0207             goto bad;
0208         goto out;
0209     }
0210
0211     ret = delay_class_ctr(ti, &dc->write, argv + 3);
0212     if (ret)
0213         goto bad;
0214     if (argc == 6) {
0215         ret = delay_class_ctr(ti, &dc->flush, argv + 3);
0216         if (ret)
0217             goto bad;
0218         goto out;
0219     }
0220
0221     ret = delay_class_ctr(ti, &dc->flush, argv + 6);
0222     if (ret)
0223         goto bad;
0224
0225 out:
0226     dc->kdelayd_wq = alloc_workqueue("kdelayd", WQ_MEM_RECLAIM, 0);
0227     if (!dc->kdelayd_wq) {
0228         ret = -EINVAL;
0229         DMERR("Couldn't start kdelayd");
0230         goto bad;
0231     }
0232
0233     ti->num_flush_bios = 1;
0234     ti->num_discard_bios = 1;
0235     ti->accounts_remapped_io = true;
0236     ti->per_io_data_size = sizeof(struct dm_delay_info);
0237     return 0;
0238
0239 bad:
0240     delay_dtr(ti);
0241     return ret;
0242 }
0243
0244 static int delay_bio(struct delay_c *dc, struct delay_class *c, struct bio *bio)
0245 {
0246     struct dm_delay_info *delayed;
0247     unsigned long expires = 0;
0248
0249     if (!c->delay || !atomic_read(&dc->may_delay))
0250         return DM_MAPIO_REMAPPED;
0251
0252     delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
0253
0254     delayed->context = dc;
0255     delayed->expires = expires = jiffies + msecs_to_jiffies(c->delay);
0256
0257     mutex_lock(&delayed_bios_lock);
0258     c->ops++;
0259     list_add_tail(&delayed->list, &dc->delayed_bios);
0260     mutex_unlock(&delayed_bios_lock);
0261
0262     queue_timeout(dc, expires);
0263
0264     return DM_MAPIO_SUBMITTED;
0265 }
0266
0267 static void delay_presuspend(struct dm_target *ti)
0268 {
0269     struct delay_c *dc = ti->private;
0270
0271     atomic_set(&dc->may_delay, 0);
0272     del_timer_sync(&dc->delay_timer);
0273     flush_bios(flush_delayed_bios(dc, 1));
0274 }
0275
0276 static void delay_resume(struct dm_target *ti)
0277 {
0278     struct delay_c *dc = ti->private;
0279
0280     atomic_set(&dc->may_delay, 1);
0281 }
0282
0283 static int delay_map(struct dm_target *ti, struct bio *bio)
0284 {
0285     struct delay_c *dc = ti->private;
0286     struct delay_class *c;
0287     struct dm_delay_info *delayed = dm_per_bio_data(bio, sizeof(struct dm_delay_info));
0288
0289     if (bio_data_dir(bio) == WRITE) {
0290         if (unlikely(bio->bi_opf & REQ_PREFLUSH))
0291             c = &dc->flush;
0292         else
0293             c = &dc->write;
0294     } else {
0295         c = &dc->read;
0296     }
0297     delayed->class = c;
0298     bio_set_dev(bio, c->dev->bdev);
0299     bio->bi_iter.bi_sector = c->start + dm_target_offset(ti, bio->bi_iter.bi_sector);
0300
0301     return delay_bio(dc, c, bio);
0302 }
0303
0304 #define DMEMIT_DELAY_CLASS(c) \
0305     DMEMIT("%s %llu %u", (c)->dev->name, (unsigned long long)(c)->start, (c)->delay)
0306
0307 static void delay_status(struct dm_target *ti, status_type_t type,
0308              unsigned status_flags, char *result, unsigned maxlen)
0309 {
0310     struct delay_c *dc = ti->private;
0311     int sz = 0;
0312
0313     switch (type) {
0314     case STATUSTYPE_INFO:
0315         DMEMIT("%u %u %u", dc->read.ops, dc->write.ops, dc->flush.ops);
0316         break;
0317
0318     case STATUSTYPE_TABLE:
0319         DMEMIT_DELAY_CLASS(&dc->read);
0320         if (dc->argc >= 6) {
0321             DMEMIT(" ");
0322             DMEMIT_DELAY_CLASS(&dc->write);
0323         }
0324         if (dc->argc >= 9) {
0325             DMEMIT(" ");
0326             DMEMIT_DELAY_CLASS(&dc->flush);
0327         }
0328         break;
0329
0330     case STATUSTYPE_IMA:
0331         *result = '\0';
0332         break;
0333     }
0334 }
0335
0336 static int delay_iterate_devices(struct dm_target *ti,
0337                  iterate_devices_callout_fn fn, void *data)
0338 {
0339     struct delay_c *dc = ti->private;
0340     int ret = 0;
0341
0342     ret = fn(ti, dc->read.dev, dc->read.start, ti->len, data);
0343     if (ret)
0344         goto out;
0345     ret = fn(ti, dc->write.dev, dc->write.start, ti->len, data);
0346     if (ret)
0347         goto out;
0348     ret = fn(ti, dc->flush.dev, dc->flush.start, ti->len, data);
0349     if (ret)
0350         goto out;
0351
0352 out:
0353     return ret;
0354 }
0355
0356 static struct target_type delay_target = {
0357     .name        = "delay",
0358     .version     = {1, 3, 0},
0359     .features    = DM_TARGET_PASSES_INTEGRITY,
0360     .module      = THIS_MODULE,
0361     .ctr         = delay_ctr,
0362     .dtr         = delay_dtr,
0363     .map         = delay_map,
0364     .presuspend  = delay_presuspend,
0365     .resume      = delay_resume,
0366     .status      = delay_status,
0367     .iterate_devices = delay_iterate_devices,
0368 };
0369
0370 static int __init dm_delay_init(void)
0371 {
0372     int r;
0373
0374     r = dm_register_target(&delay_target);
0375     if (r < 0) {
0376         DMERR("register failed %d", r);
0377         goto bad_register;
0378     }
0379
0380     return 0;
0381
0382 bad_register:
0383     return r;
0384 }
0385
0386 static void __exit dm_delay_exit(void)
0387 {
0388     dm_unregister_target(&delay_target);
0389 }
0390
0391 /* Module hooks */
0392 module_init(dm_delay_init);
0393 module_exit(dm_delay_exit);
0394
0395 MODULE_DESCRIPTION(DM_NAME " delay target");
0396 MODULE_AUTHOR("Heinz Mauelshagen <mauelshagen@redhat.com>");
0397 MODULE_LICENSE("GPL");