blob: 35c94ff24ad5867917ed715a8ce387fbb8d3b150 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001-2003 Sistina Software (UK) Limited.
3 *
4 * This file is released under the GPL.
5 */
6
Mikulas Patocka586e80e2008-10-21 17:44:59 +01007#include <linux/device-mapper.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07008
9#include <linux/module.h>
10#include <linux/init.h>
11#include <linux/blkdev.h>
12#include <linux/bio.h>
13#include <linux/slab.h>
vignesh babu6f3c3f02007-10-19 22:38:44 +010014#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015
Alasdair G Kergon72d94862006-06-26 00:27:35 -070016#define DM_MSG_PREFIX "striped"
Brian Wooda25eb942008-02-08 02:11:22 +000017#define DM_IO_ERROR_THRESHOLD 15
Alasdair G Kergon72d94862006-06-26 00:27:35 -070018
Linus Torvalds1da177e2005-04-16 15:20:36 -070019struct stripe {
20 struct dm_dev *dev;
21 sector_t physical_start;
Brian Wooda25eb942008-02-08 02:11:22 +000022
23 atomic_t error_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -070024};
25
26struct stripe_c {
27 uint32_t stripes;
Mikulas Patockac96053b2010-08-12 04:14:21 +010028 int stripes_shift;
29 sector_t stripes_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31 /* The size of this target / num. stripes */
32 sector_t stripe_width;
33
34 /* stripe chunk size */
35 uint32_t chunk_shift;
36 sector_t chunk_mask;
37
Brian Wooda25eb942008-02-08 02:11:22 +000038 /* Needed for handling events */
39 struct dm_target *ti;
40
41 /* Work struct used for triggering events*/
Tejun Heof521f072011-01-13 19:59:57 +000042 struct work_struct trigger_event;
Brian Wooda25eb942008-02-08 02:11:22 +000043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 struct stripe stripe[0];
45};
46
Brian Wooda25eb942008-02-08 02:11:22 +000047/*
48 * An event is triggered whenever a drive
49 * drops out of a stripe volume.
50 */
51static void trigger_event(struct work_struct *work)
52{
Tejun Heof521f072011-01-13 19:59:57 +000053 struct stripe_c *sc = container_of(work, struct stripe_c,
54 trigger_event);
Brian Wooda25eb942008-02-08 02:11:22 +000055 dm_table_event(sc->ti->table);
Brian Wooda25eb942008-02-08 02:11:22 +000056}
57
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static inline struct stripe_c *alloc_context(unsigned int stripes)
59{
60 size_t len;
61
Mikulas Patockad63a5ce2008-10-21 17:44:57 +010062 if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe),
63 stripes))
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 return NULL;
65
66 len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes);
67
68 return kmalloc(len, GFP_KERNEL);
69}
70
71/*
72 * Parse a single <dev> <sector> pair
73 */
74static int get_stripe(struct dm_target *ti, struct stripe_c *sc,
75 unsigned int stripe, char **argv)
76{
Andrew Morton4ee218c2006-03-27 01:17:48 -080077 unsigned long long start;
Mikulas Patocka31998ef2012-03-28 18:41:26 +010078 char dummy;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Mikulas Patocka31998ef2012-03-28 18:41:26 +010080 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 return -EINVAL;
82
Nikanth Karthikesan8215d6e2010-03-06 02:32:27 +000083 if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 &sc->stripe[stripe].dev))
85 return -ENXIO;
86
87 sc->stripe[stripe].physical_start = start;
Brian Wooda25eb942008-02-08 02:11:22 +000088
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 return 0;
90}
91
92/*
93 * Construct a striped mapping.
94 * <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+
95 */
96static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
97{
98 struct stripe_c *sc;
99 sector_t width;
100 uint32_t stripes;
101 uint32_t chunk_size;
102 char *end;
103 int r;
104 unsigned int i;
105
106 if (argc < 2) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700107 ti->error = "Not enough arguments";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 return -EINVAL;
109 }
110
111 stripes = simple_strtoul(argv[0], &end, 10);
Nikanth Karthikesan781248c2010-02-16 18:42:47 +0000112 if (!stripes || *end) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700113 ti->error = "Invalid stripe count";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 return -EINVAL;
115 }
116
117 chunk_size = simple_strtoul(argv[1], &end, 10);
118 if (*end) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700119 ti->error = "Invalid chunk_size";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return -EINVAL;
121 }
122
123 /*
124 * chunk_size is a power of two
125 */
vignesh babu6f3c3f02007-10-19 22:38:44 +0100126 if (!is_power_of_2(chunk_size) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 (chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700128 ti->error = "Invalid chunk size";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 return -EINVAL;
130 }
131
Kevin Corrya22c96c2006-03-27 01:18:01 -0800132 if (ti->len & (chunk_size - 1)) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700133 ti->error = "Target length not divisible by "
Kevin Corry8ba32fd2006-03-16 23:04:03 -0800134 "chunk size";
135 return -EINVAL;
136 }
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 width = ti->len;
139 if (sector_div(width, stripes)) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700140 ti->error = "Target length not divisible by "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 "number of stripes";
142 return -EINVAL;
143 }
144
145 /*
146 * Do we have enough arguments for that many stripes ?
147 */
148 if (argc != (2 + 2 * stripes)) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700149 ti->error = "Not enough destinations "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 "specified";
151 return -EINVAL;
152 }
153
154 sc = alloc_context(stripes);
155 if (!sc) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700156 ti->error = "Memory allocation for striped context "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 "failed";
158 return -ENOMEM;
159 }
160
Tejun Heof521f072011-01-13 19:59:57 +0000161 INIT_WORK(&sc->trigger_event, trigger_event);
Brian Wooda25eb942008-02-08 02:11:22 +0000162
163 /* Set pointer to dm target; used in trigger_event */
164 sc->ti = ti;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 sc->stripes = stripes;
166 sc->stripe_width = width;
Mikulas Patockac96053b2010-08-12 04:14:21 +0100167
168 if (stripes & (stripes - 1))
169 sc->stripes_shift = -1;
170 else {
171 sc->stripes_shift = ffs(stripes) - 1;
172 sc->stripes_mask = ((sector_t) stripes) - 1;
173 }
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 ti->split_io = chunk_size;
Mikulas Patocka374bf7e2009-06-22 10:12:22 +0100176 ti->num_flush_requests = stripes;
Mikulas Patocka7b76ec12010-08-12 04:14:26 +0100177 ti->num_discard_requests = stripes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Mikulas Patockac96053b2010-08-12 04:14:21 +0100179 sc->chunk_shift = ffs(chunk_size) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 sc->chunk_mask = ((sector_t) chunk_size) - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
182 /*
183 * Get the stripe destinations.
184 */
185 for (i = 0; i < stripes; i++) {
186 argv += 2;
187
188 r = get_stripe(ti, sc, i, argv);
189 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700190 ti->error = "Couldn't parse stripe destination";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 while (i--)
192 dm_put_device(ti, sc->stripe[i].dev);
193 kfree(sc);
194 return r;
195 }
Brian Wooda25eb942008-02-08 02:11:22 +0000196 atomic_set(&(sc->stripe[i].error_count), 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 }
198
199 ti->private = sc;
Brian Wooda25eb942008-02-08 02:11:22 +0000200
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 return 0;
202}
203
204static void stripe_dtr(struct dm_target *ti)
205{
206 unsigned int i;
207 struct stripe_c *sc = (struct stripe_c *) ti->private;
208
209 for (i = 0; i < sc->stripes; i++)
210 dm_put_device(ti, sc->stripe[i].dev);
211
Tejun Heof521f072011-01-13 19:59:57 +0000212 flush_work_sync(&sc->trigger_event);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 kfree(sc);
214}
215
Mikulas Patocka65988522010-08-12 04:14:14 +0100216static void stripe_map_sector(struct stripe_c *sc, sector_t sector,
217 uint32_t *stripe, sector_t *result)
218{
219 sector_t offset = dm_target_offset(sc->ti, sector);
220 sector_t chunk = offset >> sc->chunk_shift;
221
Mikulas Patockac96053b2010-08-12 04:14:21 +0100222 if (sc->stripes_shift < 0)
223 *stripe = sector_div(chunk, sc->stripes);
224 else {
225 *stripe = chunk & sc->stripes_mask;
226 chunk >>= sc->stripes_shift;
227 }
228
Mikulas Patocka65988522010-08-12 04:14:14 +0100229 *result = (chunk << sc->chunk_shift) | (offset & sc->chunk_mask);
230}
231
Mikulas Patocka7b76ec12010-08-12 04:14:26 +0100232static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector,
233 uint32_t target_stripe, sector_t *result)
234{
235 uint32_t stripe;
236
237 stripe_map_sector(sc, sector, &stripe, result);
238 if (stripe == target_stripe)
239 return;
240 *result &= ~sc->chunk_mask; /* round down */
241 if (target_stripe < stripe)
242 *result += sc->chunk_mask + 1; /* next chunk */
243}
244
245static int stripe_map_discard(struct stripe_c *sc, struct bio *bio,
246 uint32_t target_stripe)
247{
248 sector_t begin, end;
249
250 stripe_map_range_sector(sc, bio->bi_sector, target_stripe, &begin);
251 stripe_map_range_sector(sc, bio->bi_sector + bio_sectors(bio),
252 target_stripe, &end);
253 if (begin < end) {
254 bio->bi_bdev = sc->stripe[target_stripe].dev->bdev;
255 bio->bi_sector = begin + sc->stripe[target_stripe].physical_start;
256 bio->bi_size = to_bytes(end - begin);
257 return DM_MAPIO_REMAPPED;
258 } else {
259 /* The range doesn't map to the target stripe */
260 bio_endio(bio, 0);
261 return DM_MAPIO_SUBMITTED;
262 }
263}
264
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265static int stripe_map(struct dm_target *ti, struct bio *bio,
266 union map_info *map_context)
267{
Mikulas Patocka65988522010-08-12 04:14:14 +0100268 struct stripe_c *sc = ti->private;
Mikulas Patocka374bf7e2009-06-22 10:12:22 +0100269 uint32_t stripe;
Mike Snitzer57cba5d2010-08-12 04:14:04 +0100270 unsigned target_request_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271
Tejun Heod87f4c12010-09-03 11:56:19 +0200272 if (bio->bi_rw & REQ_FLUSH) {
Mike Snitzer57cba5d2010-08-12 04:14:04 +0100273 target_request_nr = map_context->target_request_nr;
274 BUG_ON(target_request_nr >= sc->stripes);
275 bio->bi_bdev = sc->stripe[target_request_nr].dev->bdev;
Mikulas Patocka374bf7e2009-06-22 10:12:22 +0100276 return DM_MAPIO_REMAPPED;
277 }
Mikulas Patocka7b76ec12010-08-12 04:14:26 +0100278 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
279 target_request_nr = map_context->target_request_nr;
280 BUG_ON(target_request_nr >= sc->stripes);
281 return stripe_map_discard(sc, bio, target_request_nr);
282 }
Mikulas Patocka374bf7e2009-06-22 10:12:22 +0100283
Mikulas Patocka65988522010-08-12 04:14:14 +0100284 stripe_map_sector(sc, bio->bi_sector, &stripe, &bio->bi_sector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Mikulas Patocka65988522010-08-12 04:14:14 +0100286 bio->bi_sector += sc->stripe[stripe].physical_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 bio->bi_bdev = sc->stripe[stripe].dev->bdev;
Mikulas Patocka65988522010-08-12 04:14:14 +0100288
Kiyoshi Uedad2a7ad22006-12-08 02:41:06 -0800289 return DM_MAPIO_REMAPPED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290}
291
Brian Wood4f7f5c62008-02-08 02:11:24 +0000292/*
293 * Stripe status:
294 *
295 * INFO
296 * #stripes [stripe_name <stripe_name>] [group word count]
297 * [error count 'A|D' <error count 'A|D'>]
298 *
299 * TABLE
300 * #stripes [stripe chunk size]
301 * [stripe_name physical_start <stripe_name physical_start>]
302 *
303 */
304
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305static int stripe_status(struct dm_target *ti,
306 status_type_t type, char *result, unsigned int maxlen)
307{
308 struct stripe_c *sc = (struct stripe_c *) ti->private;
Brian Wood4f7f5c62008-02-08 02:11:24 +0000309 char buffer[sc->stripes + 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 unsigned int sz = 0;
311 unsigned int i;
312
313 switch (type) {
314 case STATUSTYPE_INFO:
Brian Wood4f7f5c62008-02-08 02:11:24 +0000315 DMEMIT("%d ", sc->stripes);
316 for (i = 0; i < sc->stripes; i++) {
317 DMEMIT("%s ", sc->stripe[i].dev->name);
318 buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
319 'D' : 'A';
320 }
321 buffer[i] = '\0';
322 DMEMIT("1 %s", buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 break;
324
325 case STATUSTYPE_TABLE:
Andrew Morton4ee218c2006-03-27 01:17:48 -0800326 DMEMIT("%d %llu", sc->stripes,
327 (unsigned long long)sc->chunk_mask + 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 for (i = 0; i < sc->stripes; i++)
Andrew Morton4ee218c2006-03-27 01:17:48 -0800329 DMEMIT(" %s %llu", sc->stripe[i].dev->name,
330 (unsigned long long)sc->stripe[i].physical_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 break;
332 }
333 return 0;
334}
335
Brian Wooda25eb942008-02-08 02:11:22 +0000336static int stripe_end_io(struct dm_target *ti, struct bio *bio,
337 int error, union map_info *map_context)
338{
339 unsigned i;
340 char major_minor[16];
341 struct stripe_c *sc = ti->private;
342
343 if (!error)
344 return 0; /* I/O complete */
345
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200346 if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD))
Brian Wooda25eb942008-02-08 02:11:22 +0000347 return error;
348
349 if (error == -EOPNOTSUPP)
350 return error;
351
352 memset(major_minor, 0, sizeof(major_minor));
353 sprintf(major_minor, "%d:%d",
Tejun Heof331c022008-09-03 09:01:48 +0200354 MAJOR(disk_devt(bio->bi_bdev->bd_disk)),
355 MINOR(disk_devt(bio->bi_bdev->bd_disk)));
Brian Wooda25eb942008-02-08 02:11:22 +0000356
357 /*
358 * Test to see which stripe drive triggered the event
359 * and increment error count for all stripes on that device.
360 * If the error count for a given device exceeds the threshold
361 * value we will no longer trigger any further events.
362 */
363 for (i = 0; i < sc->stripes; i++)
364 if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
365 atomic_inc(&(sc->stripe[i].error_count));
366 if (atomic_read(&(sc->stripe[i].error_count)) <
367 DM_IO_ERROR_THRESHOLD)
Tejun Heof521f072011-01-13 19:59:57 +0000368 schedule_work(&sc->trigger_event);
Brian Wooda25eb942008-02-08 02:11:22 +0000369 }
370
371 return error;
372}
373
Mike Snitzeraf4874e2009-06-22 10:12:33 +0100374static int stripe_iterate_devices(struct dm_target *ti,
375 iterate_devices_callout_fn fn, void *data)
376{
377 struct stripe_c *sc = ti->private;
378 int ret = 0;
379 unsigned i = 0;
380
Mike Snitzer5dea2712009-07-23 20:30:42 +0100381 do {
Mike Snitzeraf4874e2009-06-22 10:12:33 +0100382 ret = fn(ti, sc->stripe[i].dev,
Mike Snitzer5dea2712009-07-23 20:30:42 +0100383 sc->stripe[i].physical_start,
384 sc->stripe_width, data);
385 } while (!ret && ++i < sc->stripes);
Mike Snitzeraf4874e2009-06-22 10:12:33 +0100386
387 return ret;
388}
389
Mike Snitzer40bea432009-09-04 20:40:25 +0100390static void stripe_io_hints(struct dm_target *ti,
391 struct queue_limits *limits)
392{
393 struct stripe_c *sc = ti->private;
394 unsigned chunk_size = (sc->chunk_mask + 1) << 9;
395
396 blk_limits_io_min(limits, chunk_size);
Martin K. Petersen3c5820c2009-09-11 21:54:52 +0200397 blk_limits_io_opt(limits, chunk_size * sc->stripes);
Mike Snitzer40bea432009-09-04 20:40:25 +0100398}
399
Mustafa Mesanovic29915202011-03-24 13:54:35 +0000400static int stripe_merge(struct dm_target *ti, struct bvec_merge_data *bvm,
401 struct bio_vec *biovec, int max_size)
402{
403 struct stripe_c *sc = ti->private;
404 sector_t bvm_sector = bvm->bi_sector;
405 uint32_t stripe;
406 struct request_queue *q;
407
408 stripe_map_sector(sc, bvm_sector, &stripe, &bvm_sector);
409
410 q = bdev_get_queue(sc->stripe[stripe].dev->bdev);
411 if (!q->merge_bvec_fn)
412 return max_size;
413
414 bvm->bi_bdev = sc->stripe[stripe].dev->bdev;
415 bvm->bi_sector = sc->stripe[stripe].physical_start + bvm_sector;
416
417 return min(max_size, q->merge_bvec_fn(q, bvm, biovec));
418}
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420static struct target_type stripe_target = {
421 .name = "striped",
Mustafa Mesanovic29915202011-03-24 13:54:35 +0000422 .version = {1, 4, 0},
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 .module = THIS_MODULE,
424 .ctr = stripe_ctr,
425 .dtr = stripe_dtr,
426 .map = stripe_map,
Brian Wooda25eb942008-02-08 02:11:22 +0000427 .end_io = stripe_end_io,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 .status = stripe_status,
Mike Snitzeraf4874e2009-06-22 10:12:33 +0100429 .iterate_devices = stripe_iterate_devices,
Mike Snitzer40bea432009-09-04 20:40:25 +0100430 .io_hints = stripe_io_hints,
Mustafa Mesanovic29915202011-03-24 13:54:35 +0000431 .merge = stripe_merge,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432};
433
434int __init dm_stripe_init(void)
435{
436 int r;
437
438 r = dm_register_target(&stripe_target);
Heinz Mauelshagen6edebde2008-11-13 23:38:56 +0000439 if (r < 0) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700440 DMWARN("target registration failed");
Heinz Mauelshagen6edebde2008-11-13 23:38:56 +0000441 return r;
442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443
444 return r;
445}
446
447void dm_stripe_exit(void)
448{
Mikulas Patocka10d3bd02009-01-06 03:04:58 +0000449 dm_unregister_target(&stripe_target);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450}