Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001-2003 Sistina Software (UK) Limited. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7fff5e8f | 7 | #include "dm.h" |
586e80e6 | 8 | #include <linux/device-mapper.h> |
1da177e4 LT |
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> | |
6f3c3f0a | 15 | #include <linux/log2.h> |
1da177e4 | 16 | |
72d94861 | 17 | #define DM_MSG_PREFIX "striped" |
a25eb944 | 18 | #define DM_IO_ERROR_THRESHOLD 15 |
72d94861 | 19 | |
1da177e4 LT |
20 | struct stripe { |
21 | struct dm_dev *dev; | |
22 | sector_t physical_start; | |
a25eb944 BW |
23 | |
24 | atomic_t error_count; | |
1da177e4 LT |
25 | }; |
26 | ||
27 | struct stripe_c { | |
28 | uint32_t stripes; | |
c96053b7 | 29 | int stripes_shift; |
1da177e4 LT |
30 | |
31 | /* The size of this target / num. stripes */ | |
32 | sector_t stripe_width; | |
33 | ||
eb850de6 | 34 | uint32_t chunk_size; |
33d07c0d | 35 | int chunk_size_shift; |
1da177e4 | 36 | |
a25eb944 BW |
37 | /* Needed for handling events */ |
38 | struct dm_target *ti; | |
39 | ||
40 | /* Work struct used for triggering events*/ | |
f521f074 | 41 | struct work_struct trigger_event; |
a25eb944 | 42 | |
1da177e4 LT |
43 | struct stripe stripe[0]; |
44 | }; | |
45 | ||
a25eb944 BW |
46 | /* |
47 | * An event is triggered whenever a drive | |
48 | * drops out of a stripe volume. | |
49 | */ | |
50 | static void trigger_event(struct work_struct *work) | |
51 | { | |
f521f074 TH |
52 | struct stripe_c *sc = container_of(work, struct stripe_c, |
53 | trigger_event); | |
a25eb944 | 54 | dm_table_event(sc->ti->table); |
a25eb944 BW |
55 | } |
56 | ||
1da177e4 LT |
57 | static inline struct stripe_c *alloc_context(unsigned int stripes) |
58 | { | |
59 | size_t len; | |
60 | ||
d63a5ce3 MP |
61 | if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe), |
62 | stripes)) | |
1da177e4 LT |
63 | return NULL; |
64 | ||
65 | len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes); | |
66 | ||
67 | return kmalloc(len, GFP_KERNEL); | |
68 | } | |
69 | ||
70 | /* | |
71 | * Parse a single <dev> <sector> pair | |
72 | */ | |
73 | static int get_stripe(struct dm_target *ti, struct stripe_c *sc, | |
74 | unsigned int stripe, char **argv) | |
75 | { | |
4ee218cd | 76 | unsigned long long start; |
31998ef1 | 77 | char dummy; |
1da177e4 | 78 | |
31998ef1 | 79 | if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1) |
1da177e4 LT |
80 | return -EINVAL; |
81 | ||
8215d6ec | 82 | if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), |
1da177e4 LT |
83 | &sc->stripe[stripe].dev)) |
84 | return -ENXIO; | |
85 | ||
86 | sc->stripe[stripe].physical_start = start; | |
a25eb944 | 87 | |
1da177e4 LT |
88 | return 0; |
89 | } | |
90 | ||
91 | /* | |
92 | * Construct a striped mapping. | |
eb850de6 | 93 | * <number of stripes> <chunk size> [<dev_path> <offset>]+ |
1da177e4 LT |
94 | */ |
95 | static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
96 | { | |
97 | struct stripe_c *sc; | |
d793e684 | 98 | sector_t width, tmp_len; |
1da177e4 LT |
99 | uint32_t stripes; |
100 | uint32_t chunk_size; | |
1da177e4 LT |
101 | int r; |
102 | unsigned int i; | |
103 | ||
104 | if (argc < 2) { | |
72d94861 | 105 | ti->error = "Not enough arguments"; |
1da177e4 LT |
106 | return -EINVAL; |
107 | } | |
108 | ||
1a66a08a | 109 | if (kstrtouint(argv[0], 10, &stripes) || !stripes) { |
72d94861 | 110 | ti->error = "Invalid stripe count"; |
1da177e4 LT |
111 | return -EINVAL; |
112 | } | |
113 | ||
8f069b41 | 114 | if (kstrtouint(argv[1], 10, &chunk_size) || !chunk_size) { |
eb850de6 | 115 | ti->error = "Invalid chunk_size"; |
1da177e4 LT |
116 | return -EINVAL; |
117 | } | |
118 | ||
eb850de6 | 119 | width = ti->len; |
d793e684 | 120 | if (sector_div(width, stripes)) { |
72d94861 | 121 | ti->error = "Target length not divisible by " |
d793e684 | 122 | "number of stripes"; |
8ba32fde KC |
123 | return -EINVAL; |
124 | } | |
125 | ||
d793e684 MS |
126 | tmp_len = width; |
127 | if (sector_div(tmp_len, chunk_size)) { | |
72d94861 | 128 | ti->error = "Target length not divisible by " |
d793e684 | 129 | "chunk size"; |
1da177e4 LT |
130 | return -EINVAL; |
131 | } | |
132 | ||
133 | /* | |
134 | * Do we have enough arguments for that many stripes ? | |
135 | */ | |
136 | if (argc != (2 + 2 * stripes)) { | |
72d94861 | 137 | ti->error = "Not enough destinations " |
1da177e4 LT |
138 | "specified"; |
139 | return -EINVAL; | |
140 | } | |
141 | ||
142 | sc = alloc_context(stripes); | |
143 | if (!sc) { | |
72d94861 | 144 | ti->error = "Memory allocation for striped context " |
1da177e4 LT |
145 | "failed"; |
146 | return -ENOMEM; | |
147 | } | |
148 | ||
f521f074 | 149 | INIT_WORK(&sc->trigger_event, trigger_event); |
a25eb944 BW |
150 | |
151 | /* Set pointer to dm target; used in trigger_event */ | |
152 | sc->ti = ti; | |
1da177e4 LT |
153 | sc->stripes = stripes; |
154 | sc->stripe_width = width; | |
c96053b7 MP |
155 | |
156 | if (stripes & (stripes - 1)) | |
157 | sc->stripes_shift = -1; | |
1df05483 MP |
158 | else |
159 | sc->stripes_shift = __ffs(stripes); | |
c96053b7 | 160 | |
542f9038 MS |
161 | r = dm_set_target_max_io_len(ti, chunk_size); |
162 | if (r) | |
163 | return r; | |
164 | ||
55a62eef AK |
165 | ti->num_flush_bios = stripes; |
166 | ti->num_discard_bios = stripes; | |
167 | ti->num_write_same_bios = stripes; | |
1da177e4 | 168 | |
eb850de6 | 169 | sc->chunk_size = chunk_size; |
33d07c0d MP |
170 | if (chunk_size & (chunk_size - 1)) |
171 | sc->chunk_size_shift = -1; | |
172 | else | |
173 | sc->chunk_size_shift = __ffs(chunk_size); | |
1da177e4 LT |
174 | |
175 | /* | |
176 | * Get the stripe destinations. | |
177 | */ | |
178 | for (i = 0; i < stripes; i++) { | |
179 | argv += 2; | |
180 | ||
181 | r = get_stripe(ti, sc, i, argv); | |
182 | if (r < 0) { | |
72d94861 | 183 | ti->error = "Couldn't parse stripe destination"; |
1da177e4 LT |
184 | while (i--) |
185 | dm_put_device(ti, sc->stripe[i].dev); | |
186 | kfree(sc); | |
187 | return r; | |
188 | } | |
a25eb944 | 189 | atomic_set(&(sc->stripe[i].error_count), 0); |
1da177e4 LT |
190 | } |
191 | ||
192 | ti->private = sc; | |
a25eb944 | 193 | |
1da177e4 LT |
194 | return 0; |
195 | } | |
196 | ||
197 | static void stripe_dtr(struct dm_target *ti) | |
198 | { | |
199 | unsigned int i; | |
200 | struct stripe_c *sc = (struct stripe_c *) ti->private; | |
201 | ||
202 | for (i = 0; i < sc->stripes; i++) | |
203 | dm_put_device(ti, sc->stripe[i].dev); | |
204 | ||
43829731 | 205 | flush_work(&sc->trigger_event); |
1da177e4 LT |
206 | kfree(sc); |
207 | } | |
208 | ||
65988525 MP |
209 | static void stripe_map_sector(struct stripe_c *sc, sector_t sector, |
210 | uint32_t *stripe, sector_t *result) | |
211 | { | |
eb850de6 | 212 | sector_t chunk = dm_target_offset(sc->ti, sector); |
33d07c0d MP |
213 | sector_t chunk_offset; |
214 | ||
215 | if (sc->chunk_size_shift < 0) | |
216 | chunk_offset = sector_div(chunk, sc->chunk_size); | |
217 | else { | |
218 | chunk_offset = chunk & (sc->chunk_size - 1); | |
219 | chunk >>= sc->chunk_size_shift; | |
220 | } | |
65988525 | 221 | |
c96053b7 MP |
222 | if (sc->stripes_shift < 0) |
223 | *stripe = sector_div(chunk, sc->stripes); | |
224 | else { | |
1df05483 | 225 | *stripe = chunk & (sc->stripes - 1); |
c96053b7 MP |
226 | chunk >>= sc->stripes_shift; |
227 | } | |
228 | ||
33d07c0d MP |
229 | if (sc->chunk_size_shift < 0) |
230 | chunk *= sc->chunk_size; | |
231 | else | |
232 | chunk <<= sc->chunk_size_shift; | |
233 | ||
234 | *result = chunk + chunk_offset; | |
65988525 MP |
235 | } |
236 | ||
7b76ec11 MP |
237 | static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector, |
238 | uint32_t target_stripe, sector_t *result) | |
239 | { | |
240 | uint32_t stripe; | |
241 | ||
242 | stripe_map_sector(sc, sector, &stripe, result); | |
243 | if (stripe == target_stripe) | |
244 | return; | |
eb850de6 MS |
245 | |
246 | /* round down */ | |
247 | sector = *result; | |
33d07c0d MP |
248 | if (sc->chunk_size_shift < 0) |
249 | *result -= sector_div(sector, sc->chunk_size); | |
250 | else | |
251 | *result = sector & ~(sector_t)(sc->chunk_size - 1); | |
eb850de6 | 252 | |
7b76ec11 | 253 | if (target_stripe < stripe) |
eb850de6 | 254 | *result += sc->chunk_size; /* next chunk */ |
7b76ec11 MP |
255 | } |
256 | ||
45e621d4 MS |
257 | static int stripe_map_range(struct stripe_c *sc, struct bio *bio, |
258 | uint32_t target_stripe) | |
7b76ec11 MP |
259 | { |
260 | sector_t begin, end; | |
261 | ||
262 | stripe_map_range_sector(sc, bio->bi_sector, target_stripe, &begin); | |
f73a1c7d | 263 | stripe_map_range_sector(sc, bio_end_sector(bio), |
7b76ec11 MP |
264 | target_stripe, &end); |
265 | if (begin < end) { | |
266 | bio->bi_bdev = sc->stripe[target_stripe].dev->bdev; | |
267 | bio->bi_sector = begin + sc->stripe[target_stripe].physical_start; | |
268 | bio->bi_size = to_bytes(end - begin); | |
269 | return DM_MAPIO_REMAPPED; | |
270 | } else { | |
271 | /* The range doesn't map to the target stripe */ | |
272 | bio_endio(bio, 0); | |
273 | return DM_MAPIO_SUBMITTED; | |
274 | } | |
275 | } | |
276 | ||
7de3ee57 | 277 | static int stripe_map(struct dm_target *ti, struct bio *bio) |
1da177e4 | 278 | { |
65988525 | 279 | struct stripe_c *sc = ti->private; |
374bf7e7 | 280 | uint32_t stripe; |
55a62eef | 281 | unsigned target_bio_nr; |
1da177e4 | 282 | |
d87f4c14 | 283 | if (bio->bi_rw & REQ_FLUSH) { |
55a62eef AK |
284 | target_bio_nr = dm_bio_get_target_bio_nr(bio); |
285 | BUG_ON(target_bio_nr >= sc->stripes); | |
286 | bio->bi_bdev = sc->stripe[target_bio_nr].dev->bdev; | |
374bf7e7 MP |
287 | return DM_MAPIO_REMAPPED; |
288 | } | |
45e621d4 MS |
289 | if (unlikely(bio->bi_rw & REQ_DISCARD) || |
290 | unlikely(bio->bi_rw & REQ_WRITE_SAME)) { | |
55a62eef AK |
291 | target_bio_nr = dm_bio_get_target_bio_nr(bio); |
292 | BUG_ON(target_bio_nr >= sc->stripes); | |
293 | return stripe_map_range(sc, bio, target_bio_nr); | |
7b76ec11 | 294 | } |
374bf7e7 | 295 | |
65988525 | 296 | stripe_map_sector(sc, bio->bi_sector, &stripe, &bio->bi_sector); |
1da177e4 | 297 | |
65988525 | 298 | bio->bi_sector += sc->stripe[stripe].physical_start; |
1da177e4 | 299 | bio->bi_bdev = sc->stripe[stripe].dev->bdev; |
65988525 | 300 | |
d2a7ad29 | 301 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
302 | } |
303 | ||
4f7f5c67 BW |
304 | /* |
305 | * Stripe status: | |
306 | * | |
307 | * INFO | |
308 | * #stripes [stripe_name <stripe_name>] [group word count] | |
309 | * [error count 'A|D' <error count 'A|D'>] | |
310 | * | |
311 | * TABLE | |
312 | * #stripes [stripe chunk size] | |
313 | * [stripe_name physical_start <stripe_name physical_start>] | |
314 | * | |
315 | */ | |
316 | ||
fd7c092e MP |
317 | static void stripe_status(struct dm_target *ti, status_type_t type, |
318 | unsigned status_flags, char *result, unsigned maxlen) | |
1da177e4 LT |
319 | { |
320 | struct stripe_c *sc = (struct stripe_c *) ti->private; | |
4f7f5c67 | 321 | char buffer[sc->stripes + 1]; |
1da177e4 LT |
322 | unsigned int sz = 0; |
323 | unsigned int i; | |
324 | ||
325 | switch (type) { | |
326 | case STATUSTYPE_INFO: | |
4f7f5c67 BW |
327 | DMEMIT("%d ", sc->stripes); |
328 | for (i = 0; i < sc->stripes; i++) { | |
329 | DMEMIT("%s ", sc->stripe[i].dev->name); | |
330 | buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? | |
331 | 'D' : 'A'; | |
332 | } | |
333 | buffer[i] = '\0'; | |
334 | DMEMIT("1 %s", buffer); | |
1da177e4 LT |
335 | break; |
336 | ||
337 | case STATUSTYPE_TABLE: | |
4ee218cd | 338 | DMEMIT("%d %llu", sc->stripes, |
eb850de6 | 339 | (unsigned long long)sc->chunk_size); |
1da177e4 | 340 | for (i = 0; i < sc->stripes; i++) |
4ee218cd AM |
341 | DMEMIT(" %s %llu", sc->stripe[i].dev->name, |
342 | (unsigned long long)sc->stripe[i].physical_start); | |
1da177e4 LT |
343 | break; |
344 | } | |
1da177e4 LT |
345 | } |
346 | ||
7de3ee57 | 347 | static int stripe_end_io(struct dm_target *ti, struct bio *bio, int error) |
a25eb944 BW |
348 | { |
349 | unsigned i; | |
350 | char major_minor[16]; | |
351 | struct stripe_c *sc = ti->private; | |
352 | ||
353 | if (!error) | |
354 | return 0; /* I/O complete */ | |
355 | ||
7b6d91da | 356 | if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD)) |
a25eb944 BW |
357 | return error; |
358 | ||
359 | if (error == -EOPNOTSUPP) | |
360 | return error; | |
361 | ||
362 | memset(major_minor, 0, sizeof(major_minor)); | |
363 | sprintf(major_minor, "%d:%d", | |
f331c029 TH |
364 | MAJOR(disk_devt(bio->bi_bdev->bd_disk)), |
365 | MINOR(disk_devt(bio->bi_bdev->bd_disk))); | |
a25eb944 BW |
366 | |
367 | /* | |
368 | * Test to see which stripe drive triggered the event | |
369 | * and increment error count for all stripes on that device. | |
370 | * If the error count for a given device exceeds the threshold | |
371 | * value we will no longer trigger any further events. | |
372 | */ | |
373 | for (i = 0; i < sc->stripes; i++) | |
374 | if (!strcmp(sc->stripe[i].dev->name, major_minor)) { | |
375 | atomic_inc(&(sc->stripe[i].error_count)); | |
376 | if (atomic_read(&(sc->stripe[i].error_count)) < | |
377 | DM_IO_ERROR_THRESHOLD) | |
f521f074 | 378 | schedule_work(&sc->trigger_event); |
a25eb944 BW |
379 | } |
380 | ||
381 | return error; | |
382 | } | |
383 | ||
af4874e0 MS |
384 | static int stripe_iterate_devices(struct dm_target *ti, |
385 | iterate_devices_callout_fn fn, void *data) | |
386 | { | |
387 | struct stripe_c *sc = ti->private; | |
388 | int ret = 0; | |
389 | unsigned i = 0; | |
390 | ||
5dea271b | 391 | do { |
af4874e0 | 392 | ret = fn(ti, sc->stripe[i].dev, |
5dea271b MS |
393 | sc->stripe[i].physical_start, |
394 | sc->stripe_width, data); | |
395 | } while (!ret && ++i < sc->stripes); | |
af4874e0 MS |
396 | |
397 | return ret; | |
398 | } | |
399 | ||
40bea431 MS |
400 | static void stripe_io_hints(struct dm_target *ti, |
401 | struct queue_limits *limits) | |
402 | { | |
403 | struct stripe_c *sc = ti->private; | |
eb850de6 | 404 | unsigned chunk_size = sc->chunk_size << SECTOR_SHIFT; |
40bea431 MS |
405 | |
406 | blk_limits_io_min(limits, chunk_size); | |
3c5820c7 | 407 | blk_limits_io_opt(limits, chunk_size * sc->stripes); |
40bea431 MS |
408 | } |
409 | ||
29915202 MM |
410 | static int stripe_merge(struct dm_target *ti, struct bvec_merge_data *bvm, |
411 | struct bio_vec *biovec, int max_size) | |
412 | { | |
413 | struct stripe_c *sc = ti->private; | |
414 | sector_t bvm_sector = bvm->bi_sector; | |
415 | uint32_t stripe; | |
416 | struct request_queue *q; | |
417 | ||
418 | stripe_map_sector(sc, bvm_sector, &stripe, &bvm_sector); | |
419 | ||
420 | q = bdev_get_queue(sc->stripe[stripe].dev->bdev); | |
421 | if (!q->merge_bvec_fn) | |
422 | return max_size; | |
423 | ||
424 | bvm->bi_bdev = sc->stripe[stripe].dev->bdev; | |
425 | bvm->bi_sector = sc->stripe[stripe].physical_start + bvm_sector; | |
426 | ||
427 | return min(max_size, q->merge_bvec_fn(q, bvm, biovec)); | |
428 | } | |
429 | ||
1da177e4 LT |
430 | static struct target_type stripe_target = { |
431 | .name = "striped", | |
fd7c092e | 432 | .version = {1, 5, 1}, |
1da177e4 LT |
433 | .module = THIS_MODULE, |
434 | .ctr = stripe_ctr, | |
435 | .dtr = stripe_dtr, | |
436 | .map = stripe_map, | |
a25eb944 | 437 | .end_io = stripe_end_io, |
1da177e4 | 438 | .status = stripe_status, |
af4874e0 | 439 | .iterate_devices = stripe_iterate_devices, |
40bea431 | 440 | .io_hints = stripe_io_hints, |
29915202 | 441 | .merge = stripe_merge, |
1da177e4 LT |
442 | }; |
443 | ||
444 | int __init dm_stripe_init(void) | |
445 | { | |
446 | int r; | |
447 | ||
448 | r = dm_register_target(&stripe_target); | |
6edebdee | 449 | if (r < 0) { |
72d94861 | 450 | DMWARN("target registration failed"); |
6edebdee HM |
451 | return r; |
452 | } | |
1da177e4 LT |
453 | |
454 | return r; | |
455 | } | |
456 | ||
457 | void dm_stripe_exit(void) | |
458 | { | |
10d3bd09 | 459 | dm_unregister_target(&stripe_target); |
1da177e4 | 460 | } |