1 /*
2 * Copyright (C) 2010-2012 by Dell Inc. All rights reserved.
3 * Copyright (C) 2011-2013 Red Hat, Inc.
4 *
5 * This file is released under the GPL.
6 *
7 * dm-switch is a device-mapper target that maps IO to underlying block
8 * devices efficiently when there are a large number of fixed-sized
9 * address regions but there is no simple pattern to allow for a compact
10 * mapping representation such as dm-stripe.
11 */
12
13 #include <linux/device-mapper.h>
14
15 #include <linux/module.h>
16 #include <linux/init.h>
17 #include <linux/vmalloc.h>
18
19 #define DM_MSG_PREFIX "switch"
20
21 /*
22 * One region_table_slot_t holds <region_entries_per_slot> region table
23 * entries each of which is <region_table_entry_bits> in size.
24 */
25 typedef unsigned long region_table_slot_t;
26
27 /*
28 * A device with the offset to its start sector.
29 */
30 struct switch_path {
31 struct dm_dev *dmdev;
32 sector_t start;
33 };
34
35 /*
36 * Context block for a dm switch device.
37 */
38 struct switch_ctx {
39 struct dm_target *ti;
40
41 unsigned nr_paths; /* Number of paths in path_list. */
42
43 unsigned region_size; /* Region size in 512-byte sectors */
44 unsigned long nr_regions; /* Number of regions making up the device */
45 signed char region_size_bits; /* log2 of region_size or -1 */
46
47 unsigned char region_table_entry_bits; /* Number of bits in one region table entry */
48 unsigned char region_entries_per_slot; /* Number of entries in one region table slot */
49 signed char region_entries_per_slot_bits; /* log2 of region_entries_per_slot or -1 */
50
51 region_table_slot_t *region_table; /* Region table */
52
53 /*
54 * Array of dm devices to switch between.
55 */
56 struct switch_path path_list[0];
57 };
58
alloc_switch_ctx(struct dm_target * ti,unsigned nr_paths,unsigned region_size)59 static struct switch_ctx *alloc_switch_ctx(struct dm_target *ti, unsigned nr_paths,
60 unsigned region_size)
61 {
62 struct switch_ctx *sctx;
63
64 sctx = kzalloc(sizeof(struct switch_ctx) + nr_paths * sizeof(struct switch_path),
65 GFP_KERNEL);
66 if (!sctx)
67 return NULL;
68
69 sctx->ti = ti;
70 sctx->region_size = region_size;
71
72 ti->private = sctx;
73
74 return sctx;
75 }
76
alloc_region_table(struct dm_target * ti,unsigned nr_paths)77 static int alloc_region_table(struct dm_target *ti, unsigned nr_paths)
78 {
79 struct switch_ctx *sctx = ti->private;
80 sector_t nr_regions = ti->len;
81 sector_t nr_slots;
82
83 if (!(sctx->region_size & (sctx->region_size - 1)))
84 sctx->region_size_bits = __ffs(sctx->region_size);
85 else
86 sctx->region_size_bits = -1;
87
88 sctx->region_table_entry_bits = 1;
89 while (sctx->region_table_entry_bits < sizeof(region_table_slot_t) * 8 &&
90 (region_table_slot_t)1 << sctx->region_table_entry_bits < nr_paths)
91 sctx->region_table_entry_bits++;
92
93 sctx->region_entries_per_slot = (sizeof(region_table_slot_t) * 8) / sctx->region_table_entry_bits;
94 if (!(sctx->region_entries_per_slot & (sctx->region_entries_per_slot - 1)))
95 sctx->region_entries_per_slot_bits = __ffs(sctx->region_entries_per_slot);
96 else
97 sctx->region_entries_per_slot_bits = -1;
98
99 if (sector_div(nr_regions, sctx->region_size))
100 nr_regions++;
101
102 if (nr_regions >= ULONG_MAX) {
103 ti->error = "Region table too large";
104 return -EINVAL;
105 }
106 sctx->nr_regions = nr_regions;
107
108 nr_slots = nr_regions;
109 if (sector_div(nr_slots, sctx->region_entries_per_slot))
110 nr_slots++;
111
112 if (nr_slots > ULONG_MAX / sizeof(region_table_slot_t)) {
113 ti->error = "Region table too large";
114 return -EINVAL;
115 }
116
117 sctx->region_table = vmalloc(array_size(nr_slots,
118 sizeof(region_table_slot_t)));
119 if (!sctx->region_table) {
120 ti->error = "Cannot allocate region table";
121 return -ENOMEM;
122 }
123
124 return 0;
125 }
126
switch_get_position(struct switch_ctx * sctx,unsigned long region_nr,unsigned long * region_index,unsigned * bit)127 static void switch_get_position(struct switch_ctx *sctx, unsigned long region_nr,
128 unsigned long *region_index, unsigned *bit)
129 {
130 if (sctx->region_entries_per_slot_bits >= 0) {
131 *region_index = region_nr >> sctx->region_entries_per_slot_bits;
132 *bit = region_nr & (sctx->region_entries_per_slot - 1);
133 } else {
134 *region_index = region_nr / sctx->region_entries_per_slot;
135 *bit = region_nr % sctx->region_entries_per_slot;
136 }
137
138 *bit *= sctx->region_table_entry_bits;
139 }
140
switch_region_table_read(struct switch_ctx * sctx,unsigned long region_nr)141 static unsigned switch_region_table_read(struct switch_ctx *sctx, unsigned long region_nr)
142 {
143 unsigned long region_index;
144 unsigned bit;
145
146 switch_get_position(sctx, region_nr, ®ion_index, &bit);
147
148 return (READ_ONCE(sctx->region_table[region_index]) >> bit) &
149 ((1 << sctx->region_table_entry_bits) - 1);
150 }
151
152 /*
153 * Find which path to use at given offset.
154 */
switch_get_path_nr(struct switch_ctx * sctx,sector_t offset)155 static unsigned switch_get_path_nr(struct switch_ctx *sctx, sector_t offset)
156 {
157 unsigned path_nr;
158 sector_t p;
159
160 p = offset;
161 if (sctx->region_size_bits >= 0)
162 p >>= sctx->region_size_bits;
163 else
164 sector_div(p, sctx->region_size);
165
166 path_nr = switch_region_table_read(sctx, p);
167
168 /* This can only happen if the processor uses non-atomic stores. */
169 if (unlikely(path_nr >= sctx->nr_paths))
170 path_nr = 0;
171
172 return path_nr;
173 }
174
switch_region_table_write(struct switch_ctx * sctx,unsigned long region_nr,unsigned value)175 static void switch_region_table_write(struct switch_ctx *sctx, unsigned long region_nr,
176 unsigned value)
177 {
178 unsigned long region_index;
179 unsigned bit;
180 region_table_slot_t pte;
181
182 switch_get_position(sctx, region_nr, ®ion_index, &bit);
183
184 pte = sctx->region_table[region_index];
185 pte &= ~((((region_table_slot_t)1 << sctx->region_table_entry_bits) - 1) << bit);
186 pte |= (region_table_slot_t)value << bit;
187 sctx->region_table[region_index] = pte;
188 }
189
190 /*
191 * Fill the region table with an initial round robin pattern.
192 */
initialise_region_table(struct switch_ctx * sctx)193 static void initialise_region_table(struct switch_ctx *sctx)
194 {
195 unsigned path_nr = 0;
196 unsigned long region_nr;
197
198 for (region_nr = 0; region_nr < sctx->nr_regions; region_nr++) {
199 switch_region_table_write(sctx, region_nr, path_nr);
200 if (++path_nr >= sctx->nr_paths)
201 path_nr = 0;
202 }
203 }
204
parse_path(struct dm_arg_set * as,struct dm_target * ti)205 static int parse_path(struct dm_arg_set *as, struct dm_target *ti)
206 {
207 struct switch_ctx *sctx = ti->private;
208 unsigned long long start;
209 int r;
210
211 r = dm_get_device(ti, dm_shift_arg(as), dm_table_get_mode(ti->table),
212 &sctx->path_list[sctx->nr_paths].dmdev);
213 if (r) {
214 ti->error = "Device lookup failed";
215 return r;
216 }
217
218 if (kstrtoull(dm_shift_arg(as), 10, &start) || start != (sector_t)start) {
219 ti->error = "Invalid device starting offset";
220 dm_put_device(ti, sctx->path_list[sctx->nr_paths].dmdev);
221 return -EINVAL;
222 }
223
224 sctx->path_list[sctx->nr_paths].start = start;
225
226 sctx->nr_paths++;
227
228 return 0;
229 }
230
231 /*
232 * Destructor: Don't free the dm_target, just the ti->private data (if any).
233 */
switch_dtr(struct dm_target * ti)234 static void switch_dtr(struct dm_target *ti)
235 {
236 struct switch_ctx *sctx = ti->private;
237
238 while (sctx->nr_paths--)
239 dm_put_device(ti, sctx->path_list[sctx->nr_paths].dmdev);
240
241 vfree(sctx->region_table);
242 kfree(sctx);
243 }
244
245 /*
246 * Constructor arguments:
247 * <num_paths> <region_size> <num_optional_args> [<optional_args>...]
248 * [<dev_path> <offset>]+
249 *
250 * Optional args are to allow for future extension: currently this
251 * parameter must be 0.
252 */
switch_ctr(struct dm_target * ti,unsigned argc,char ** argv)253 static int switch_ctr(struct dm_target *ti, unsigned argc, char **argv)
254 {
255 static const struct dm_arg _args[] = {
256 {1, (KMALLOC_MAX_SIZE - sizeof(struct switch_ctx)) / sizeof(struct switch_path), "Invalid number of paths"},
257 {1, UINT_MAX, "Invalid region size"},
258 {0, 0, "Invalid number of optional args"},
259 };
260
261 struct switch_ctx *sctx;
262 struct dm_arg_set as;
263 unsigned nr_paths, region_size, nr_optional_args;
264 int r;
265
266 as.argc = argc;
267 as.argv = argv;
268
269 r = dm_read_arg(_args, &as, &nr_paths, &ti->error);
270 if (r)
271 return -EINVAL;
272
273 r = dm_read_arg(_args + 1, &as, ®ion_size, &ti->error);
274 if (r)
275 return r;
276
277 r = dm_read_arg_group(_args + 2, &as, &nr_optional_args, &ti->error);
278 if (r)
279 return r;
280 /* parse optional arguments here, if we add any */
281
282 if (as.argc != nr_paths * 2) {
283 ti->error = "Incorrect number of path arguments";
284 return -EINVAL;
285 }
286
287 sctx = alloc_switch_ctx(ti, nr_paths, region_size);
288 if (!sctx) {
289 ti->error = "Cannot allocate redirection context";
290 return -ENOMEM;
291 }
292
293 r = dm_set_target_max_io_len(ti, region_size);
294 if (r)
295 goto error;
296
297 while (as.argc) {
298 r = parse_path(&as, ti);
299 if (r)
300 goto error;
301 }
302
303 r = alloc_region_table(ti, nr_paths);
304 if (r)
305 goto error;
306
307 initialise_region_table(sctx);
308
309 /* For UNMAP, sending the request down any path is sufficient */
310 ti->num_discard_bios = 1;
311
312 return 0;
313
314 error:
315 switch_dtr(ti);
316
317 return r;
318 }
319
switch_map(struct dm_target * ti,struct bio * bio)320 static int switch_map(struct dm_target *ti, struct bio *bio)
321 {
322 struct switch_ctx *sctx = ti->private;
323 sector_t offset = dm_target_offset(ti, bio->bi_iter.bi_sector);
324 unsigned path_nr = switch_get_path_nr(sctx, offset);
325
326 bio_set_dev(bio, sctx->path_list[path_nr].dmdev->bdev);
327 bio->bi_iter.bi_sector = sctx->path_list[path_nr].start + offset;
328
329 return DM_MAPIO_REMAPPED;
330 }
331
332 /*
333 * We need to parse hex numbers in the message as quickly as possible.
334 *
335 * This table-based hex parser improves performance.
336 * It improves a time to load 1000000 entries compared to the condition-based
337 * parser.
338 * table-based parser condition-based parser
339 * PA-RISC 0.29s 0.31s
340 * Opteron 0.0495s 0.0498s
341 */
342 static const unsigned char hex_table[256] = {
343 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
344 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
345 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
346 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 255, 255, 255, 255, 255, 255,
347 255, 10, 11, 12, 13, 14, 15, 255, 255, 255, 255, 255, 255, 255, 255, 255,
348 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
349 255, 10, 11, 12, 13, 14, 15, 255, 255, 255, 255, 255, 255, 255, 255, 255,
350 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
351 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
352 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
353 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
354 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
355 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
356 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
357 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255,
358 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255
359 };
360
parse_hex(const char ** string)361 static __always_inline unsigned long parse_hex(const char **string)
362 {
363 unsigned char d;
364 unsigned long r = 0;
365
366 while ((d = hex_table[(unsigned char)**string]) < 16) {
367 r = (r << 4) | d;
368 (*string)++;
369 }
370
371 return r;
372 }
373
process_set_region_mappings(struct switch_ctx * sctx,unsigned argc,char ** argv)374 static int process_set_region_mappings(struct switch_ctx *sctx,
375 unsigned argc, char **argv)
376 {
377 unsigned i;
378 unsigned long region_index = 0;
379
380 for (i = 1; i < argc; i++) {
381 unsigned long path_nr;
382 const char *string = argv[i];
383
384 if ((*string & 0xdf) == 'R') {
385 unsigned long cycle_length, num_write;
386
387 string++;
388 if (unlikely(*string == ',')) {
389 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
390 return -EINVAL;
391 }
392 cycle_length = parse_hex(&string);
393 if (unlikely(*string != ',')) {
394 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
395 return -EINVAL;
396 }
397 string++;
398 if (unlikely(!*string)) {
399 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
400 return -EINVAL;
401 }
402 num_write = parse_hex(&string);
403 if (unlikely(*string)) {
404 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
405 return -EINVAL;
406 }
407
408 if (unlikely(!cycle_length) || unlikely(cycle_length - 1 > region_index)) {
409 DMWARN("invalid set_region_mappings cycle length: %lu > %lu",
410 cycle_length - 1, region_index);
411 return -EINVAL;
412 }
413 if (unlikely(region_index + num_write < region_index) ||
414 unlikely(region_index + num_write >= sctx->nr_regions)) {
415 DMWARN("invalid set_region_mappings region number: %lu + %lu >= %lu",
416 region_index, num_write, sctx->nr_regions);
417 return -EINVAL;
418 }
419
420 while (num_write--) {
421 region_index++;
422 path_nr = switch_region_table_read(sctx, region_index - cycle_length);
423 switch_region_table_write(sctx, region_index, path_nr);
424 }
425
426 continue;
427 }
428
429 if (*string == ':')
430 region_index++;
431 else {
432 region_index = parse_hex(&string);
433 if (unlikely(*string != ':')) {
434 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
435 return -EINVAL;
436 }
437 }
438
439 string++;
440 if (unlikely(!*string)) {
441 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
442 return -EINVAL;
443 }
444
445 path_nr = parse_hex(&string);
446 if (unlikely(*string)) {
447 DMWARN("invalid set_region_mappings argument: '%s'", argv[i]);
448 return -EINVAL;
449 }
450 if (unlikely(region_index >= sctx->nr_regions)) {
451 DMWARN("invalid set_region_mappings region number: %lu >= %lu", region_index, sctx->nr_regions);
452 return -EINVAL;
453 }
454 if (unlikely(path_nr >= sctx->nr_paths)) {
455 DMWARN("invalid set_region_mappings device: %lu >= %u", path_nr, sctx->nr_paths);
456 return -EINVAL;
457 }
458
459 switch_region_table_write(sctx, region_index, path_nr);
460 }
461
462 return 0;
463 }
464
465 /*
466 * Messages are processed one-at-a-time.
467 *
468 * Only set_region_mappings is supported.
469 */
switch_message(struct dm_target * ti,unsigned argc,char ** argv,char * result,unsigned maxlen)470 static int switch_message(struct dm_target *ti, unsigned argc, char **argv,
471 char *result, unsigned maxlen)
472 {
473 static DEFINE_MUTEX(message_mutex);
474
475 struct switch_ctx *sctx = ti->private;
476 int r = -EINVAL;
477
478 mutex_lock(&message_mutex);
479
480 if (!strcasecmp(argv[0], "set_region_mappings"))
481 r = process_set_region_mappings(sctx, argc, argv);
482 else
483 DMWARN("Unrecognised message received.");
484
485 mutex_unlock(&message_mutex);
486
487 return r;
488 }
489
switch_status(struct dm_target * ti,status_type_t type,unsigned status_flags,char * result,unsigned maxlen)490 static void switch_status(struct dm_target *ti, status_type_t type,
491 unsigned status_flags, char *result, unsigned maxlen)
492 {
493 struct switch_ctx *sctx = ti->private;
494 unsigned sz = 0;
495 int path_nr;
496
497 switch (type) {
498 case STATUSTYPE_INFO:
499 result[0] = '\0';
500 break;
501
502 case STATUSTYPE_TABLE:
503 DMEMIT("%u %u 0", sctx->nr_paths, sctx->region_size);
504 for (path_nr = 0; path_nr < sctx->nr_paths; path_nr++)
505 DMEMIT(" %s %llu", sctx->path_list[path_nr].dmdev->name,
506 (unsigned long long)sctx->path_list[path_nr].start);
507 break;
508 }
509 }
510
511 /*
512 * Switch ioctl:
513 *
514 * Passthrough all ioctls to the path for sector 0
515 */
switch_prepare_ioctl(struct dm_target * ti,struct block_device ** bdev)516 static int switch_prepare_ioctl(struct dm_target *ti, struct block_device **bdev)
517 {
518 struct switch_ctx *sctx = ti->private;
519 unsigned path_nr;
520
521 path_nr = switch_get_path_nr(sctx, 0);
522
523 *bdev = sctx->path_list[path_nr].dmdev->bdev;
524
525 /*
526 * Only pass ioctls through if the device sizes match exactly.
527 */
528 if (ti->len + sctx->path_list[path_nr].start !=
529 i_size_read((*bdev)->bd_inode) >> SECTOR_SHIFT)
530 return 1;
531 return 0;
532 }
533
switch_iterate_devices(struct dm_target * ti,iterate_devices_callout_fn fn,void * data)534 static int switch_iterate_devices(struct dm_target *ti,
535 iterate_devices_callout_fn fn, void *data)
536 {
537 struct switch_ctx *sctx = ti->private;
538 int path_nr;
539 int r;
540
541 for (path_nr = 0; path_nr < sctx->nr_paths; path_nr++) {
542 r = fn(ti, sctx->path_list[path_nr].dmdev,
543 sctx->path_list[path_nr].start, ti->len, data);
544 if (r)
545 return r;
546 }
547
548 return 0;
549 }
550
551 static struct target_type switch_target = {
552 .name = "switch",
553 .version = {1, 1, 0},
554 .module = THIS_MODULE,
555 .ctr = switch_ctr,
556 .dtr = switch_dtr,
557 .map = switch_map,
558 .message = switch_message,
559 .status = switch_status,
560 .prepare_ioctl = switch_prepare_ioctl,
561 .iterate_devices = switch_iterate_devices,
562 };
563
dm_switch_init(void)564 static int __init dm_switch_init(void)
565 {
566 int r;
567
568 r = dm_register_target(&switch_target);
569 if (r < 0)
570 DMERR("dm_register_target() failed %d", r);
571
572 return r;
573 }
574
dm_switch_exit(void)575 static void __exit dm_switch_exit(void)
576 {
577 dm_unregister_target(&switch_target);
578 }
579
580 module_init(dm_switch_init);
581 module_exit(dm_switch_exit);
582
583 MODULE_DESCRIPTION(DM_NAME " dynamic path switching target");
584 MODULE_AUTHOR("Kevin D. O'Kelley <Kevin_OKelley@dell.com>");
585 MODULE_AUTHOR("Narendran Ganapathy <Narendran_Ganapathy@dell.com>");
586 MODULE_AUTHOR("Jim Ramsay <Jim_Ramsay@dell.com>");
587 MODULE_AUTHOR("Mikulas Patocka <mpatocka@redhat.com>");
588 MODULE_LICENSE("GPL");
589
