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 | ||
586e80e6 | 7 | #include <linux/device-mapper.h> |
1da177e4 LT |
8 | |
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> | |
6f3c3f0a | 14 | #include <linux/log2.h> |
1da177e4 | 15 | |
72d94861 | 16 | #define DM_MSG_PREFIX "striped" |
a25eb944 | 17 | #define DM_IO_ERROR_THRESHOLD 15 |
72d94861 | 18 | |
1da177e4 LT |
19 | struct stripe { |
20 | struct dm_dev *dev; | |
21 | sector_t physical_start; | |
a25eb944 BW |
22 | |
23 | atomic_t error_count; | |
1da177e4 LT |
24 | }; |
25 | ||
26 | struct stripe_c { | |
27 | uint32_t stripes; | |
28 | ||
29 | /* The size of this target / num. stripes */ | |
30 | sector_t stripe_width; | |
31 | ||
32 | /* stripe chunk size */ | |
33 | uint32_t chunk_shift; | |
34 | sector_t chunk_mask; | |
35 | ||
a25eb944 BW |
36 | /* Needed for handling events */ |
37 | struct dm_target *ti; | |
38 | ||
39 | /* Work struct used for triggering events*/ | |
40 | struct work_struct kstriped_ws; | |
41 | ||
1da177e4 LT |
42 | struct stripe stripe[0]; |
43 | }; | |
44 | ||
a25eb944 BW |
45 | static struct workqueue_struct *kstriped; |
46 | ||
47 | /* | |
48 | * An event is triggered whenever a drive | |
49 | * drops out of a stripe volume. | |
50 | */ | |
51 | static void trigger_event(struct work_struct *work) | |
52 | { | |
53 | struct stripe_c *sc = container_of(work, struct stripe_c, kstriped_ws); | |
54 | ||
55 | dm_table_event(sc->ti->table); | |
56 | ||
57 | } | |
58 | ||
1da177e4 LT |
59 | static inline struct stripe_c *alloc_context(unsigned int stripes) |
60 | { | |
61 | size_t len; | |
62 | ||
d63a5ce3 MP |
63 | if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe), |
64 | stripes)) | |
1da177e4 LT |
65 | return NULL; |
66 | ||
67 | len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes); | |
68 | ||
69 | return kmalloc(len, GFP_KERNEL); | |
70 | } | |
71 | ||
72 | /* | |
73 | * Parse a single <dev> <sector> pair | |
74 | */ | |
75 | static int get_stripe(struct dm_target *ti, struct stripe_c *sc, | |
76 | unsigned int stripe, char **argv) | |
77 | { | |
4ee218cd | 78 | unsigned long long start; |
1da177e4 | 79 | |
4ee218cd | 80 | if (sscanf(argv[1], "%llu", &start) != 1) |
1da177e4 LT |
81 | return -EINVAL; |
82 | ||
8215d6ec | 83 | if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), |
1da177e4 LT |
84 | &sc->stripe[stripe].dev)) |
85 | return -ENXIO; | |
86 | ||
87 | sc->stripe[stripe].physical_start = start; | |
a25eb944 | 88 | |
1da177e4 LT |
89 | return 0; |
90 | } | |
91 | ||
92 | /* | |
93 | * Construct a striped mapping. | |
94 | * <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+ | |
95 | */ | |
96 | static 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) { | |
72d94861 | 107 | ti->error = "Not enough arguments"; |
1da177e4 LT |
108 | return -EINVAL; |
109 | } | |
110 | ||
111 | stripes = simple_strtoul(argv[0], &end, 10); | |
781248c1 | 112 | if (!stripes || *end) { |
72d94861 | 113 | ti->error = "Invalid stripe count"; |
1da177e4 LT |
114 | return -EINVAL; |
115 | } | |
116 | ||
117 | chunk_size = simple_strtoul(argv[1], &end, 10); | |
118 | if (*end) { | |
72d94861 | 119 | ti->error = "Invalid chunk_size"; |
1da177e4 LT |
120 | return -EINVAL; |
121 | } | |
122 | ||
123 | /* | |
124 | * chunk_size is a power of two | |
125 | */ | |
6f3c3f0a | 126 | if (!is_power_of_2(chunk_size) || |
1da177e4 | 127 | (chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) { |
72d94861 | 128 | ti->error = "Invalid chunk size"; |
1da177e4 LT |
129 | return -EINVAL; |
130 | } | |
131 | ||
a22c96c7 | 132 | if (ti->len & (chunk_size - 1)) { |
72d94861 | 133 | ti->error = "Target length not divisible by " |
8ba32fde KC |
134 | "chunk size"; |
135 | return -EINVAL; | |
136 | } | |
137 | ||
1da177e4 LT |
138 | width = ti->len; |
139 | if (sector_div(width, stripes)) { | |
72d94861 | 140 | ti->error = "Target length not divisible by " |
1da177e4 LT |
141 | "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)) { | |
72d94861 | 149 | ti->error = "Not enough destinations " |
1da177e4 LT |
150 | "specified"; |
151 | return -EINVAL; | |
152 | } | |
153 | ||
154 | sc = alloc_context(stripes); | |
155 | if (!sc) { | |
72d94861 | 156 | ti->error = "Memory allocation for striped context " |
1da177e4 LT |
157 | "failed"; |
158 | return -ENOMEM; | |
159 | } | |
160 | ||
a25eb944 BW |
161 | INIT_WORK(&sc->kstriped_ws, trigger_event); |
162 | ||
163 | /* Set pointer to dm target; used in trigger_event */ | |
164 | sc->ti = ti; | |
165 | ||
1da177e4 LT |
166 | sc->stripes = stripes; |
167 | sc->stripe_width = width; | |
168 | ti->split_io = chunk_size; | |
374bf7e7 | 169 | ti->num_flush_requests = stripes; |
1da177e4 LT |
170 | |
171 | sc->chunk_mask = ((sector_t) chunk_size) - 1; | |
172 | for (sc->chunk_shift = 0; chunk_size; sc->chunk_shift++) | |
173 | chunk_size >>= 1; | |
174 | sc->chunk_shift--; | |
175 | ||
176 | /* | |
177 | * Get the stripe destinations. | |
178 | */ | |
179 | for (i = 0; i < stripes; i++) { | |
180 | argv += 2; | |
181 | ||
182 | r = get_stripe(ti, sc, i, argv); | |
183 | if (r < 0) { | |
72d94861 | 184 | ti->error = "Couldn't parse stripe destination"; |
1da177e4 LT |
185 | while (i--) |
186 | dm_put_device(ti, sc->stripe[i].dev); | |
187 | kfree(sc); | |
188 | return r; | |
189 | } | |
a25eb944 | 190 | atomic_set(&(sc->stripe[i].error_count), 0); |
1da177e4 LT |
191 | } |
192 | ||
193 | ti->private = sc; | |
a25eb944 | 194 | |
1da177e4 LT |
195 | return 0; |
196 | } | |
197 | ||
198 | static void stripe_dtr(struct dm_target *ti) | |
199 | { | |
200 | unsigned int i; | |
201 | struct stripe_c *sc = (struct stripe_c *) ti->private; | |
202 | ||
203 | for (i = 0; i < sc->stripes; i++) | |
204 | dm_put_device(ti, sc->stripe[i].dev); | |
205 | ||
a25eb944 | 206 | flush_workqueue(kstriped); |
1da177e4 LT |
207 | kfree(sc); |
208 | } | |
209 | ||
210 | static int stripe_map(struct dm_target *ti, struct bio *bio, | |
211 | union map_info *map_context) | |
212 | { | |
213 | struct stripe_c *sc = (struct stripe_c *) ti->private; | |
374bf7e7 MP |
214 | sector_t offset, chunk; |
215 | uint32_t stripe; | |
1da177e4 | 216 | |
374bf7e7 MP |
217 | if (unlikely(bio_empty_barrier(bio))) { |
218 | BUG_ON(map_context->flush_request >= sc->stripes); | |
219 | bio->bi_bdev = sc->stripe[map_context->flush_request].dev->bdev; | |
220 | return DM_MAPIO_REMAPPED; | |
221 | } | |
222 | ||
223 | offset = bio->bi_sector - ti->begin; | |
224 | chunk = offset >> sc->chunk_shift; | |
225 | stripe = sector_div(chunk, sc->stripes); | |
1da177e4 LT |
226 | |
227 | bio->bi_bdev = sc->stripe[stripe].dev->bdev; | |
228 | bio->bi_sector = sc->stripe[stripe].physical_start + | |
229 | (chunk << sc->chunk_shift) + (offset & sc->chunk_mask); | |
d2a7ad29 | 230 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
231 | } |
232 | ||
4f7f5c67 BW |
233 | /* |
234 | * Stripe status: | |
235 | * | |
236 | * INFO | |
237 | * #stripes [stripe_name <stripe_name>] [group word count] | |
238 | * [error count 'A|D' <error count 'A|D'>] | |
239 | * | |
240 | * TABLE | |
241 | * #stripes [stripe chunk size] | |
242 | * [stripe_name physical_start <stripe_name physical_start>] | |
243 | * | |
244 | */ | |
245 | ||
1da177e4 LT |
246 | static int stripe_status(struct dm_target *ti, |
247 | status_type_t type, char *result, unsigned int maxlen) | |
248 | { | |
249 | struct stripe_c *sc = (struct stripe_c *) ti->private; | |
4f7f5c67 | 250 | char buffer[sc->stripes + 1]; |
1da177e4 LT |
251 | unsigned int sz = 0; |
252 | unsigned int i; | |
253 | ||
254 | switch (type) { | |
255 | case STATUSTYPE_INFO: | |
4f7f5c67 BW |
256 | DMEMIT("%d ", sc->stripes); |
257 | for (i = 0; i < sc->stripes; i++) { | |
258 | DMEMIT("%s ", sc->stripe[i].dev->name); | |
259 | buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ? | |
260 | 'D' : 'A'; | |
261 | } | |
262 | buffer[i] = '\0'; | |
263 | DMEMIT("1 %s", buffer); | |
1da177e4 LT |
264 | break; |
265 | ||
266 | case STATUSTYPE_TABLE: | |
4ee218cd AM |
267 | DMEMIT("%d %llu", sc->stripes, |
268 | (unsigned long long)sc->chunk_mask + 1); | |
1da177e4 | 269 | for (i = 0; i < sc->stripes; i++) |
4ee218cd AM |
270 | DMEMIT(" %s %llu", sc->stripe[i].dev->name, |
271 | (unsigned long long)sc->stripe[i].physical_start); | |
1da177e4 LT |
272 | break; |
273 | } | |
274 | return 0; | |
275 | } | |
276 | ||
a25eb944 BW |
277 | static int stripe_end_io(struct dm_target *ti, struct bio *bio, |
278 | int error, union map_info *map_context) | |
279 | { | |
280 | unsigned i; | |
281 | char major_minor[16]; | |
282 | struct stripe_c *sc = ti->private; | |
283 | ||
284 | if (!error) | |
285 | return 0; /* I/O complete */ | |
286 | ||
1f98a13f | 287 | if ((error == -EWOULDBLOCK) && bio_rw_flagged(bio, BIO_RW_AHEAD)) |
a25eb944 BW |
288 | return error; |
289 | ||
290 | if (error == -EOPNOTSUPP) | |
291 | return error; | |
292 | ||
293 | memset(major_minor, 0, sizeof(major_minor)); | |
294 | sprintf(major_minor, "%d:%d", | |
f331c029 TH |
295 | MAJOR(disk_devt(bio->bi_bdev->bd_disk)), |
296 | MINOR(disk_devt(bio->bi_bdev->bd_disk))); | |
a25eb944 BW |
297 | |
298 | /* | |
299 | * Test to see which stripe drive triggered the event | |
300 | * and increment error count for all stripes on that device. | |
301 | * If the error count for a given device exceeds the threshold | |
302 | * value we will no longer trigger any further events. | |
303 | */ | |
304 | for (i = 0; i < sc->stripes; i++) | |
305 | if (!strcmp(sc->stripe[i].dev->name, major_minor)) { | |
306 | atomic_inc(&(sc->stripe[i].error_count)); | |
307 | if (atomic_read(&(sc->stripe[i].error_count)) < | |
308 | DM_IO_ERROR_THRESHOLD) | |
309 | queue_work(kstriped, &sc->kstriped_ws); | |
310 | } | |
311 | ||
312 | return error; | |
313 | } | |
314 | ||
af4874e0 MS |
315 | static int stripe_iterate_devices(struct dm_target *ti, |
316 | iterate_devices_callout_fn fn, void *data) | |
317 | { | |
318 | struct stripe_c *sc = ti->private; | |
319 | int ret = 0; | |
320 | unsigned i = 0; | |
321 | ||
5dea271b | 322 | do { |
af4874e0 | 323 | ret = fn(ti, sc->stripe[i].dev, |
5dea271b MS |
324 | sc->stripe[i].physical_start, |
325 | sc->stripe_width, data); | |
326 | } while (!ret && ++i < sc->stripes); | |
af4874e0 MS |
327 | |
328 | return ret; | |
329 | } | |
330 | ||
40bea431 MS |
331 | static void stripe_io_hints(struct dm_target *ti, |
332 | struct queue_limits *limits) | |
333 | { | |
334 | struct stripe_c *sc = ti->private; | |
335 | unsigned chunk_size = (sc->chunk_mask + 1) << 9; | |
336 | ||
337 | blk_limits_io_min(limits, chunk_size); | |
3c5820c7 | 338 | blk_limits_io_opt(limits, chunk_size * sc->stripes); |
40bea431 MS |
339 | } |
340 | ||
1da177e4 LT |
341 | static struct target_type stripe_target = { |
342 | .name = "striped", | |
40bea431 | 343 | .version = {1, 3, 0}, |
1da177e4 LT |
344 | .module = THIS_MODULE, |
345 | .ctr = stripe_ctr, | |
346 | .dtr = stripe_dtr, | |
347 | .map = stripe_map, | |
a25eb944 | 348 | .end_io = stripe_end_io, |
1da177e4 | 349 | .status = stripe_status, |
af4874e0 | 350 | .iterate_devices = stripe_iterate_devices, |
40bea431 | 351 | .io_hints = stripe_io_hints, |
1da177e4 LT |
352 | }; |
353 | ||
354 | int __init dm_stripe_init(void) | |
355 | { | |
356 | int r; | |
357 | ||
358 | r = dm_register_target(&stripe_target); | |
6edebdee | 359 | if (r < 0) { |
72d94861 | 360 | DMWARN("target registration failed"); |
6edebdee HM |
361 | return r; |
362 | } | |
1da177e4 | 363 | |
a25eb944 BW |
364 | kstriped = create_singlethread_workqueue("kstriped"); |
365 | if (!kstriped) { | |
366 | DMERR("failed to create workqueue kstriped"); | |
367 | dm_unregister_target(&stripe_target); | |
368 | return -ENOMEM; | |
369 | } | |
370 | ||
1da177e4 LT |
371 | return r; |
372 | } | |
373 | ||
374 | void dm_stripe_exit(void) | |
375 | { | |
10d3bd09 | 376 | dm_unregister_target(&stripe_target); |
a25eb944 BW |
377 | destroy_workqueue(kstriped); |
378 | ||
1da177e4 LT |
379 | return; |
380 | } |