dm-stripe.c revision d87f4c14f27dc82d215108d8392a7d26687148a1
1/*
2 * Copyright (C) 2001-2003 Sistina Software (UK) Limited.
3 *
4 * This file is released under the GPL.
5 */
6
7#include <linux/device-mapper.h>
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>
14#include <linux/log2.h>
15
16#define DM_MSG_PREFIX "striped"
17#define DM_IO_ERROR_THRESHOLD 15
18
19struct stripe {
20	struct dm_dev *dev;
21	sector_t physical_start;
22
23	atomic_t error_count;
24};
25
26struct stripe_c {
27	uint32_t stripes;
28	int stripes_shift;
29	sector_t stripes_mask;
30
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
38	/* Needed for handling events */
39	struct dm_target *ti;
40
41	/* Work struct used for triggering events*/
42	struct work_struct kstriped_ws;
43
44	struct stripe stripe[0];
45};
46
47static struct workqueue_struct *kstriped;
48
49/*
50 * An event is triggered whenever a drive
51 * drops out of a stripe volume.
52 */
53static void trigger_event(struct work_struct *work)
54{
55	struct stripe_c *sc = container_of(work, struct stripe_c, kstriped_ws);
56
57	dm_table_event(sc->ti->table);
58
59}
60
61static inline struct stripe_c *alloc_context(unsigned int stripes)
62{
63	size_t len;
64
65	if (dm_array_too_big(sizeof(struct stripe_c), sizeof(struct stripe),
66			     stripes))
67		return NULL;
68
69	len = sizeof(struct stripe_c) + (sizeof(struct stripe) * stripes);
70
71	return kmalloc(len, GFP_KERNEL);
72}
73
74/*
75 * Parse a single <dev> <sector> pair
76 */
77static int get_stripe(struct dm_target *ti, struct stripe_c *sc,
78		      unsigned int stripe, char **argv)
79{
80	unsigned long long start;
81
82	if (sscanf(argv[1], "%llu", &start) != 1)
83		return -EINVAL;
84
85	if (dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
86			  &sc->stripe[stripe].dev))
87		return -ENXIO;
88
89	sc->stripe[stripe].physical_start = start;
90
91	return 0;
92}
93
94/*
95 * Construct a striped mapping.
96 * <number of stripes> <chunk size (2^^n)> [<dev_path> <offset>]+
97 */
98static int stripe_ctr(struct dm_target *ti, unsigned int argc, char **argv)
99{
100	struct stripe_c *sc;
101	sector_t width;
102	uint32_t stripes;
103	uint32_t chunk_size;
104	char *end;
105	int r;
106	unsigned int i;
107
108	if (argc < 2) {
109		ti->error = "Not enough arguments";
110		return -EINVAL;
111	}
112
113	stripes = simple_strtoul(argv[0], &end, 10);
114	if (!stripes || *end) {
115		ti->error = "Invalid stripe count";
116		return -EINVAL;
117	}
118
119	chunk_size = simple_strtoul(argv[1], &end, 10);
120	if (*end) {
121		ti->error = "Invalid chunk_size";
122		return -EINVAL;
123	}
124
125	/*
126	 * chunk_size is a power of two
127	 */
128	if (!is_power_of_2(chunk_size) ||
129	    (chunk_size < (PAGE_SIZE >> SECTOR_SHIFT))) {
130		ti->error = "Invalid chunk size";
131		return -EINVAL;
132	}
133
134	if (ti->len & (chunk_size - 1)) {
135		ti->error = "Target length not divisible by "
136		    "chunk size";
137		return -EINVAL;
138	}
139
140	width = ti->len;
141	if (sector_div(width, stripes)) {
142		ti->error = "Target length not divisible by "
143		    "number of stripes";
144		return -EINVAL;
145	}
146
147	/*
148	 * Do we have enough arguments for that many stripes ?
149	 */
150	if (argc != (2 + 2 * stripes)) {
151		ti->error = "Not enough destinations "
152			"specified";
153		return -EINVAL;
154	}
155
156	sc = alloc_context(stripes);
157	if (!sc) {
158		ti->error = "Memory allocation for striped context "
159		    "failed";
160		return -ENOMEM;
161	}
162
163	INIT_WORK(&sc->kstriped_ws, trigger_event);
164
165	/* Set pointer to dm target; used in trigger_event */
166	sc->ti = ti;
167	sc->stripes = stripes;
168	sc->stripe_width = width;
169
170	if (stripes & (stripes - 1))
171		sc->stripes_shift = -1;
172	else {
173		sc->stripes_shift = ffs(stripes) - 1;
174		sc->stripes_mask = ((sector_t) stripes) - 1;
175	}
176
177	ti->split_io = chunk_size;
178	ti->num_flush_requests = stripes;
179	ti->num_discard_requests = stripes;
180
181	sc->chunk_shift = ffs(chunk_size) - 1;
182	sc->chunk_mask = ((sector_t) chunk_size) - 1;
183
184	/*
185	 * Get the stripe destinations.
186	 */
187	for (i = 0; i < stripes; i++) {
188		argv += 2;
189
190		r = get_stripe(ti, sc, i, argv);
191		if (r < 0) {
192			ti->error = "Couldn't parse stripe destination";
193			while (i--)
194				dm_put_device(ti, sc->stripe[i].dev);
195			kfree(sc);
196			return r;
197		}
198		atomic_set(&(sc->stripe[i].error_count), 0);
199	}
200
201	ti->private = sc;
202
203	return 0;
204}
205
206static void stripe_dtr(struct dm_target *ti)
207{
208	unsigned int i;
209	struct stripe_c *sc = (struct stripe_c *) ti->private;
210
211	for (i = 0; i < sc->stripes; i++)
212		dm_put_device(ti, sc->stripe[i].dev);
213
214	flush_workqueue(kstriped);
215	kfree(sc);
216}
217
218static void stripe_map_sector(struct stripe_c *sc, sector_t sector,
219			      uint32_t *stripe, sector_t *result)
220{
221	sector_t offset = dm_target_offset(sc->ti, sector);
222	sector_t chunk = offset >> sc->chunk_shift;
223
224	if (sc->stripes_shift < 0)
225		*stripe = sector_div(chunk, sc->stripes);
226	else {
227		*stripe = chunk & sc->stripes_mask;
228		chunk >>= sc->stripes_shift;
229	}
230
231	*result = (chunk << sc->chunk_shift) | (offset & sc->chunk_mask);
232}
233
234static void stripe_map_range_sector(struct stripe_c *sc, sector_t sector,
235				    uint32_t target_stripe, sector_t *result)
236{
237	uint32_t stripe;
238
239	stripe_map_sector(sc, sector, &stripe, result);
240	if (stripe == target_stripe)
241		return;
242	*result &= ~sc->chunk_mask;			/* round down */
243	if (target_stripe < stripe)
244		*result += sc->chunk_mask + 1;		/* next chunk */
245}
246
247static int stripe_map_discard(struct stripe_c *sc, struct bio *bio,
248			      uint32_t target_stripe)
249{
250	sector_t begin, end;
251
252	stripe_map_range_sector(sc, bio->bi_sector, target_stripe, &begin);
253	stripe_map_range_sector(sc, bio->bi_sector + bio_sectors(bio),
254				target_stripe, &end);
255	if (begin < end) {
256		bio->bi_bdev = sc->stripe[target_stripe].dev->bdev;
257		bio->bi_sector = begin + sc->stripe[target_stripe].physical_start;
258		bio->bi_size = to_bytes(end - begin);
259		return DM_MAPIO_REMAPPED;
260	} else {
261		/* The range doesn't map to the target stripe */
262		bio_endio(bio, 0);
263		return DM_MAPIO_SUBMITTED;
264	}
265}
266
267static int stripe_map(struct dm_target *ti, struct bio *bio,
268		      union map_info *map_context)
269{
270	struct stripe_c *sc = ti->private;
271	uint32_t stripe;
272	unsigned target_request_nr;
273
274	if (bio->bi_rw & REQ_FLUSH) {
275		target_request_nr = map_context->target_request_nr;
276		BUG_ON(target_request_nr >= sc->stripes);
277		bio->bi_bdev = sc->stripe[target_request_nr].dev->bdev;
278		return DM_MAPIO_REMAPPED;
279	}
280	if (unlikely(bio->bi_rw & REQ_DISCARD)) {
281		target_request_nr = map_context->target_request_nr;
282		BUG_ON(target_request_nr >= sc->stripes);
283		return stripe_map_discard(sc, bio, target_request_nr);
284	}
285
286	stripe_map_sector(sc, bio->bi_sector, &stripe, &bio->bi_sector);
287
288	bio->bi_sector += sc->stripe[stripe].physical_start;
289	bio->bi_bdev = sc->stripe[stripe].dev->bdev;
290
291	return DM_MAPIO_REMAPPED;
292}
293
294/*
295 * Stripe status:
296 *
297 * INFO
298 * #stripes [stripe_name <stripe_name>] [group word count]
299 * [error count 'A|D' <error count 'A|D'>]
300 *
301 * TABLE
302 * #stripes [stripe chunk size]
303 * [stripe_name physical_start <stripe_name physical_start>]
304 *
305 */
306
307static int stripe_status(struct dm_target *ti,
308			 status_type_t type, char *result, unsigned int maxlen)
309{
310	struct stripe_c *sc = (struct stripe_c *) ti->private;
311	char buffer[sc->stripes + 1];
312	unsigned int sz = 0;
313	unsigned int i;
314
315	switch (type) {
316	case STATUSTYPE_INFO:
317		DMEMIT("%d ", sc->stripes);
318		for (i = 0; i < sc->stripes; i++)  {
319			DMEMIT("%s ", sc->stripe[i].dev->name);
320			buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
321				'D' : 'A';
322		}
323		buffer[i] = '\0';
324		DMEMIT("1 %s", buffer);
325		break;
326
327	case STATUSTYPE_TABLE:
328		DMEMIT("%d %llu", sc->stripes,
329			(unsigned long long)sc->chunk_mask + 1);
330		for (i = 0; i < sc->stripes; i++)
331			DMEMIT(" %s %llu", sc->stripe[i].dev->name,
332			    (unsigned long long)sc->stripe[i].physical_start);
333		break;
334	}
335	return 0;
336}
337
338static int stripe_end_io(struct dm_target *ti, struct bio *bio,
339			 int error, union map_info *map_context)
340{
341	unsigned i;
342	char major_minor[16];
343	struct stripe_c *sc = ti->private;
344
345	if (!error)
346		return 0; /* I/O complete */
347
348	if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD))
349		return error;
350
351	if (error == -EOPNOTSUPP)
352		return error;
353
354	memset(major_minor, 0, sizeof(major_minor));
355	sprintf(major_minor, "%d:%d",
356		MAJOR(disk_devt(bio->bi_bdev->bd_disk)),
357		MINOR(disk_devt(bio->bi_bdev->bd_disk)));
358
359	/*
360	 * Test to see which stripe drive triggered the event
361	 * and increment error count for all stripes on that device.
362	 * If the error count for a given device exceeds the threshold
363	 * value we will no longer trigger any further events.
364	 */
365	for (i = 0; i < sc->stripes; i++)
366		if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
367			atomic_inc(&(sc->stripe[i].error_count));
368			if (atomic_read(&(sc->stripe[i].error_count)) <
369			    DM_IO_ERROR_THRESHOLD)
370				queue_work(kstriped, &sc->kstriped_ws);
371		}
372
373	return error;
374}
375
376static int stripe_iterate_devices(struct dm_target *ti,
377				  iterate_devices_callout_fn fn, void *data)
378{
379	struct stripe_c *sc = ti->private;
380	int ret = 0;
381	unsigned i = 0;
382
383	do {
384		ret = fn(ti, sc->stripe[i].dev,
385			 sc->stripe[i].physical_start,
386			 sc->stripe_width, data);
387	} while (!ret && ++i < sc->stripes);
388
389	return ret;
390}
391
392static void stripe_io_hints(struct dm_target *ti,
393			    struct queue_limits *limits)
394{
395	struct stripe_c *sc = ti->private;
396	unsigned chunk_size = (sc->chunk_mask + 1) << 9;
397
398	blk_limits_io_min(limits, chunk_size);
399	blk_limits_io_opt(limits, chunk_size * sc->stripes);
400}
401
402static struct target_type stripe_target = {
403	.name   = "striped",
404	.version = {1, 3, 0},
405	.module = THIS_MODULE,
406	.ctr    = stripe_ctr,
407	.dtr    = stripe_dtr,
408	.map    = stripe_map,
409	.end_io = stripe_end_io,
410	.status = stripe_status,
411	.iterate_devices = stripe_iterate_devices,
412	.io_hints = stripe_io_hints,
413};
414
415int __init dm_stripe_init(void)
416{
417	int r;
418
419	r = dm_register_target(&stripe_target);
420	if (r < 0) {
421		DMWARN("target registration failed");
422		return r;
423	}
424
425	kstriped = create_singlethread_workqueue("kstriped");
426	if (!kstriped) {
427		DMERR("failed to create workqueue kstriped");
428		dm_unregister_target(&stripe_target);
429		return -ENOMEM;
430	}
431
432	return r;
433}
434
435void dm_stripe_exit(void)
436{
437	dm_unregister_target(&stripe_target);
438	destroy_workqueue(kstriped);
439
440	return;
441}
442