dm-stripe.c revision c96053b767d494d7c30e2be68097ac9defa9403f
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
180	sc->chunk_shift = ffs(chunk_size) - 1;
181	sc->chunk_mask = ((sector_t) chunk_size) - 1;
182
183	/*
184	 * Get the stripe destinations.
185	 */
186	for (i = 0; i < stripes; i++) {
187		argv += 2;
188
189		r = get_stripe(ti, sc, i, argv);
190		if (r < 0) {
191			ti->error = "Couldn't parse stripe destination";
192			while (i--)
193				dm_put_device(ti, sc->stripe[i].dev);
194			kfree(sc);
195			return r;
196		}
197		atomic_set(&(sc->stripe[i].error_count), 0);
198	}
199
200	ti->private = sc;
201
202	return 0;
203}
204
205static void stripe_dtr(struct dm_target *ti)
206{
207	unsigned int i;
208	struct stripe_c *sc = (struct stripe_c *) ti->private;
209
210	for (i = 0; i < sc->stripes; i++)
211		dm_put_device(ti, sc->stripe[i].dev);
212
213	flush_workqueue(kstriped);
214	kfree(sc);
215}
216
217static void stripe_map_sector(struct stripe_c *sc, sector_t sector,
218			      uint32_t *stripe, sector_t *result)
219{
220	sector_t offset = dm_target_offset(sc->ti, sector);
221	sector_t chunk = offset >> sc->chunk_shift;
222
223	if (sc->stripes_shift < 0)
224		*stripe = sector_div(chunk, sc->stripes);
225	else {
226		*stripe = chunk & sc->stripes_mask;
227		chunk >>= sc->stripes_shift;
228	}
229
230	*result = (chunk << sc->chunk_shift) | (offset & sc->chunk_mask);
231}
232
233static int stripe_map(struct dm_target *ti, struct bio *bio,
234		      union map_info *map_context)
235{
236	struct stripe_c *sc = ti->private;
237	uint32_t stripe;
238	unsigned target_request_nr;
239
240	if (unlikely(bio_empty_barrier(bio))) {
241		target_request_nr = map_context->target_request_nr;
242		BUG_ON(target_request_nr >= sc->stripes);
243		bio->bi_bdev = sc->stripe[target_request_nr].dev->bdev;
244		return DM_MAPIO_REMAPPED;
245	}
246
247	stripe_map_sector(sc, bio->bi_sector, &stripe, &bio->bi_sector);
248
249	bio->bi_sector += sc->stripe[stripe].physical_start;
250	bio->bi_bdev = sc->stripe[stripe].dev->bdev;
251
252	return DM_MAPIO_REMAPPED;
253}
254
255/*
256 * Stripe status:
257 *
258 * INFO
259 * #stripes [stripe_name <stripe_name>] [group word count]
260 * [error count 'A|D' <error count 'A|D'>]
261 *
262 * TABLE
263 * #stripes [stripe chunk size]
264 * [stripe_name physical_start <stripe_name physical_start>]
265 *
266 */
267
268static int stripe_status(struct dm_target *ti,
269			 status_type_t type, char *result, unsigned int maxlen)
270{
271	struct stripe_c *sc = (struct stripe_c *) ti->private;
272	char buffer[sc->stripes + 1];
273	unsigned int sz = 0;
274	unsigned int i;
275
276	switch (type) {
277	case STATUSTYPE_INFO:
278		DMEMIT("%d ", sc->stripes);
279		for (i = 0; i < sc->stripes; i++)  {
280			DMEMIT("%s ", sc->stripe[i].dev->name);
281			buffer[i] = atomic_read(&(sc->stripe[i].error_count)) ?
282				'D' : 'A';
283		}
284		buffer[i] = '\0';
285		DMEMIT("1 %s", buffer);
286		break;
287
288	case STATUSTYPE_TABLE:
289		DMEMIT("%d %llu", sc->stripes,
290			(unsigned long long)sc->chunk_mask + 1);
291		for (i = 0; i < sc->stripes; i++)
292			DMEMIT(" %s %llu", sc->stripe[i].dev->name,
293			    (unsigned long long)sc->stripe[i].physical_start);
294		break;
295	}
296	return 0;
297}
298
299static int stripe_end_io(struct dm_target *ti, struct bio *bio,
300			 int error, union map_info *map_context)
301{
302	unsigned i;
303	char major_minor[16];
304	struct stripe_c *sc = ti->private;
305
306	if (!error)
307		return 0; /* I/O complete */
308
309	if ((error == -EWOULDBLOCK) && (bio->bi_rw & REQ_RAHEAD))
310		return error;
311
312	if (error == -EOPNOTSUPP)
313		return error;
314
315	memset(major_minor, 0, sizeof(major_minor));
316	sprintf(major_minor, "%d:%d",
317		MAJOR(disk_devt(bio->bi_bdev->bd_disk)),
318		MINOR(disk_devt(bio->bi_bdev->bd_disk)));
319
320	/*
321	 * Test to see which stripe drive triggered the event
322	 * and increment error count for all stripes on that device.
323	 * If the error count for a given device exceeds the threshold
324	 * value we will no longer trigger any further events.
325	 */
326	for (i = 0; i < sc->stripes; i++)
327		if (!strcmp(sc->stripe[i].dev->name, major_minor)) {
328			atomic_inc(&(sc->stripe[i].error_count));
329			if (atomic_read(&(sc->stripe[i].error_count)) <
330			    DM_IO_ERROR_THRESHOLD)
331				queue_work(kstriped, &sc->kstriped_ws);
332		}
333
334	return error;
335}
336
337static int stripe_iterate_devices(struct dm_target *ti,
338				  iterate_devices_callout_fn fn, void *data)
339{
340	struct stripe_c *sc = ti->private;
341	int ret = 0;
342	unsigned i = 0;
343
344	do {
345		ret = fn(ti, sc->stripe[i].dev,
346			 sc->stripe[i].physical_start,
347			 sc->stripe_width, data);
348	} while (!ret && ++i < sc->stripes);
349
350	return ret;
351}
352
353static void stripe_io_hints(struct dm_target *ti,
354			    struct queue_limits *limits)
355{
356	struct stripe_c *sc = ti->private;
357	unsigned chunk_size = (sc->chunk_mask + 1) << 9;
358
359	blk_limits_io_min(limits, chunk_size);
360	blk_limits_io_opt(limits, chunk_size * sc->stripes);
361}
362
363static struct target_type stripe_target = {
364	.name   = "striped",
365	.version = {1, 3, 0},
366	.module = THIS_MODULE,
367	.ctr    = stripe_ctr,
368	.dtr    = stripe_dtr,
369	.map    = stripe_map,
370	.end_io = stripe_end_io,
371	.status = stripe_status,
372	.iterate_devices = stripe_iterate_devices,
373	.io_hints = stripe_io_hints,
374};
375
376int __init dm_stripe_init(void)
377{
378	int r;
379
380	r = dm_register_target(&stripe_target);
381	if (r < 0) {
382		DMWARN("target registration failed");
383		return r;
384	}
385
386	kstriped = create_singlethread_workqueue("kstriped");
387	if (!kstriped) {
388		DMERR("failed to create workqueue kstriped");
389		dm_unregister_target(&stripe_target);
390		return -ENOMEM;
391	}
392
393	return r;
394}
395
396void dm_stripe_exit(void)
397{
398	dm_unregister_target(&stripe_target);
399	destroy_workqueue(kstriped);
400
401	return;
402}
403