Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software | |
891ce207 | 3 | * Copyright (C) 2006 Red Hat GmbH |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
952b3557 MP |
8 | #include "dm.h" |
9 | ||
586e80e6 | 10 | #include <linux/device-mapper.h> |
1da177e4 LT |
11 | |
12 | #include <linux/bio.h> | |
10f1d5d1 | 13 | #include <linux/completion.h> |
1da177e4 LT |
14 | #include <linux/mempool.h> |
15 | #include <linux/module.h> | |
16 | #include <linux/sched.h> | |
17 | #include <linux/slab.h> | |
a765e20e | 18 | #include <linux/dm-io.h> |
1da177e4 | 19 | |
f1e53987 MP |
20 | #define DM_MSG_PREFIX "io" |
21 | ||
22 | #define DM_IO_MAX_REGIONS BITS_PER_LONG | |
23 | ||
891ce207 HM |
24 | struct dm_io_client { |
25 | mempool_t *pool; | |
26 | struct bio_set *bios; | |
27 | }; | |
28 | ||
f1e53987 MP |
29 | /* |
30 | * Aligning 'struct io' reduces the number of bits required to store | |
31 | * its address. Refer to store_io_and_region_in_bio() below. | |
32 | */ | |
1da177e4 | 33 | struct io { |
e01fd7ee | 34 | unsigned long error_bits; |
1da177e4 | 35 | atomic_t count; |
891ce207 | 36 | struct dm_io_client *client; |
1da177e4 LT |
37 | io_notify_fn callback; |
38 | void *context; | |
bb91bc7b MP |
39 | void *vma_invalidate_address; |
40 | unsigned long vma_invalidate_size; | |
f1e53987 | 41 | } __attribute__((aligned(DM_IO_MAX_REGIONS))); |
1da177e4 | 42 | |
952b3557 MP |
43 | static struct kmem_cache *_dm_io_cache; |
44 | ||
c8b03afe HM |
45 | /* |
46 | * Create a client with mempool and bioset. | |
47 | */ | |
bda8efec | 48 | struct dm_io_client *dm_io_client_create(void) |
c8b03afe | 49 | { |
c8b03afe | 50 | struct dm_io_client *client; |
e8603136 | 51 | unsigned min_ios = dm_get_reserved_bio_based_ios(); |
c8b03afe HM |
52 | |
53 | client = kmalloc(sizeof(*client), GFP_KERNEL); | |
54 | if (!client) | |
55 | return ERR_PTR(-ENOMEM); | |
56 | ||
e8603136 | 57 | client->pool = mempool_create_slab_pool(min_ios, _dm_io_cache); |
c8b03afe HM |
58 | if (!client->pool) |
59 | goto bad; | |
60 | ||
e8603136 | 61 | client->bios = bioset_create(min_ios, 0); |
c8b03afe HM |
62 | if (!client->bios) |
63 | goto bad; | |
64 | ||
65 | return client; | |
66 | ||
67 | bad: | |
68 | if (client->pool) | |
69 | mempool_destroy(client->pool); | |
70 | kfree(client); | |
71 | return ERR_PTR(-ENOMEM); | |
72 | } | |
73 | EXPORT_SYMBOL(dm_io_client_create); | |
74 | ||
c8b03afe HM |
75 | void dm_io_client_destroy(struct dm_io_client *client) |
76 | { | |
77 | mempool_destroy(client->pool); | |
78 | bioset_free(client->bios); | |
79 | kfree(client); | |
80 | } | |
81 | EXPORT_SYMBOL(dm_io_client_destroy); | |
82 | ||
1da177e4 LT |
83 | /*----------------------------------------------------------------- |
84 | * We need to keep track of which region a bio is doing io for. | |
f1e53987 MP |
85 | * To avoid a memory allocation to store just 5 or 6 bits, we |
86 | * ensure the 'struct io' pointer is aligned so enough low bits are | |
87 | * always zero and then combine it with the region number directly in | |
88 | * bi_private. | |
1da177e4 | 89 | *---------------------------------------------------------------*/ |
f1e53987 MP |
90 | static void store_io_and_region_in_bio(struct bio *bio, struct io *io, |
91 | unsigned region) | |
1da177e4 | 92 | { |
f1e53987 MP |
93 | if (unlikely(!IS_ALIGNED((unsigned long)io, DM_IO_MAX_REGIONS))) { |
94 | DMCRIT("Unaligned struct io pointer %p", io); | |
95 | BUG(); | |
96 | } | |
97 | ||
98 | bio->bi_private = (void *)((unsigned long)io | region); | |
1da177e4 LT |
99 | } |
100 | ||
f1e53987 MP |
101 | static void retrieve_io_and_region_from_bio(struct bio *bio, struct io **io, |
102 | unsigned *region) | |
1da177e4 | 103 | { |
f1e53987 MP |
104 | unsigned long val = (unsigned long)bio->bi_private; |
105 | ||
106 | *io = (void *)(val & -(unsigned long)DM_IO_MAX_REGIONS); | |
107 | *region = val & (DM_IO_MAX_REGIONS - 1); | |
1da177e4 LT |
108 | } |
109 | ||
110 | /*----------------------------------------------------------------- | |
111 | * We need an io object to keep track of the number of bios that | |
112 | * have been dispatched for a particular io. | |
113 | *---------------------------------------------------------------*/ | |
97e7cdf1 | 114 | static void complete_io(struct io *io) |
1da177e4 | 115 | { |
97e7cdf1 JT |
116 | unsigned long error_bits = io->error_bits; |
117 | io_notify_fn fn = io->callback; | |
118 | void *context = io->context; | |
1da177e4 | 119 | |
97e7cdf1 JT |
120 | if (io->vma_invalidate_size) |
121 | invalidate_kernel_vmap_range(io->vma_invalidate_address, | |
122 | io->vma_invalidate_size); | |
bb91bc7b | 123 | |
97e7cdf1 JT |
124 | mempool_free(io, io->client->pool); |
125 | fn(error_bits, context); | |
126 | } | |
1da177e4 | 127 | |
97e7cdf1 JT |
128 | static void dec_count(struct io *io, unsigned int region, int error) |
129 | { | |
130 | if (error) | |
131 | set_bit(region, &io->error_bits); | |
1da177e4 | 132 | |
97e7cdf1 JT |
133 | if (atomic_dec_and_test(&io->count)) |
134 | complete_io(io); | |
1da177e4 LT |
135 | } |
136 | ||
6712ecf8 | 137 | static void endio(struct bio *bio, int error) |
1da177e4 | 138 | { |
c897feb3 HM |
139 | struct io *io; |
140 | unsigned region; | |
1da177e4 | 141 | |
1da177e4 LT |
142 | if (error && bio_data_dir(bio) == READ) |
143 | zero_fill_bio(bio); | |
144 | ||
c897feb3 HM |
145 | /* |
146 | * The bio destructor in bio_put() may use the io object. | |
147 | */ | |
f1e53987 | 148 | retrieve_io_and_region_from_bio(bio, &io, ®ion); |
c897feb3 | 149 | |
1da177e4 LT |
150 | bio_put(bio); |
151 | ||
c897feb3 | 152 | dec_count(io, region, error); |
1da177e4 LT |
153 | } |
154 | ||
155 | /*----------------------------------------------------------------- | |
156 | * These little objects provide an abstraction for getting a new | |
157 | * destination page for io. | |
158 | *---------------------------------------------------------------*/ | |
159 | struct dpages { | |
160 | void (*get_page)(struct dpages *dp, | |
161 | struct page **p, unsigned long *len, unsigned *offset); | |
162 | void (*next_page)(struct dpages *dp); | |
163 | ||
164 | unsigned context_u; | |
165 | void *context_ptr; | |
bb91bc7b MP |
166 | |
167 | void *vma_invalidate_address; | |
168 | unsigned long vma_invalidate_size; | |
1da177e4 LT |
169 | }; |
170 | ||
171 | /* | |
172 | * Functions for getting the pages from a list. | |
173 | */ | |
174 | static void list_get_page(struct dpages *dp, | |
175 | struct page **p, unsigned long *len, unsigned *offset) | |
176 | { | |
177 | unsigned o = dp->context_u; | |
178 | struct page_list *pl = (struct page_list *) dp->context_ptr; | |
179 | ||
180 | *p = pl->page; | |
181 | *len = PAGE_SIZE - o; | |
182 | *offset = o; | |
183 | } | |
184 | ||
185 | static void list_next_page(struct dpages *dp) | |
186 | { | |
187 | struct page_list *pl = (struct page_list *) dp->context_ptr; | |
188 | dp->context_ptr = pl->next; | |
189 | dp->context_u = 0; | |
190 | } | |
191 | ||
192 | static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset) | |
193 | { | |
194 | dp->get_page = list_get_page; | |
195 | dp->next_page = list_next_page; | |
196 | dp->context_u = offset; | |
197 | dp->context_ptr = pl; | |
198 | } | |
199 | ||
200 | /* | |
201 | * Functions for getting the pages from a bvec. | |
202 | */ | |
d73f9907 MP |
203 | static void bio_get_page(struct dpages *dp, struct page **p, |
204 | unsigned long *len, unsigned *offset) | |
1da177e4 | 205 | { |
d73f9907 MP |
206 | struct bio_vec *bvec = dp->context_ptr; |
207 | *p = bvec->bv_page; | |
208 | *len = bvec->bv_len - dp->context_u; | |
209 | *offset = bvec->bv_offset + dp->context_u; | |
1da177e4 LT |
210 | } |
211 | ||
003b5c57 | 212 | static void bio_next_page(struct dpages *dp) |
1da177e4 | 213 | { |
d73f9907 MP |
214 | struct bio_vec *bvec = dp->context_ptr; |
215 | dp->context_ptr = bvec + 1; | |
216 | dp->context_u = 0; | |
1da177e4 LT |
217 | } |
218 | ||
003b5c57 | 219 | static void bio_dp_init(struct dpages *dp, struct bio *bio) |
1da177e4 | 220 | { |
003b5c57 KO |
221 | dp->get_page = bio_get_page; |
222 | dp->next_page = bio_next_page; | |
d73f9907 MP |
223 | dp->context_ptr = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); |
224 | dp->context_u = bio->bi_iter.bi_bvec_done; | |
1da177e4 LT |
225 | } |
226 | ||
c8b03afe HM |
227 | /* |
228 | * Functions for getting the pages from a VMA. | |
229 | */ | |
1da177e4 LT |
230 | static void vm_get_page(struct dpages *dp, |
231 | struct page **p, unsigned long *len, unsigned *offset) | |
232 | { | |
233 | *p = vmalloc_to_page(dp->context_ptr); | |
234 | *offset = dp->context_u; | |
235 | *len = PAGE_SIZE - dp->context_u; | |
236 | } | |
237 | ||
238 | static void vm_next_page(struct dpages *dp) | |
239 | { | |
240 | dp->context_ptr += PAGE_SIZE - dp->context_u; | |
241 | dp->context_u = 0; | |
242 | } | |
243 | ||
244 | static void vm_dp_init(struct dpages *dp, void *data) | |
245 | { | |
246 | dp->get_page = vm_get_page; | |
247 | dp->next_page = vm_next_page; | |
248 | dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1); | |
249 | dp->context_ptr = data; | |
250 | } | |
251 | ||
c8b03afe HM |
252 | /* |
253 | * Functions for getting the pages from kernel memory. | |
254 | */ | |
255 | static void km_get_page(struct dpages *dp, struct page **p, unsigned long *len, | |
256 | unsigned *offset) | |
257 | { | |
258 | *p = virt_to_page(dp->context_ptr); | |
259 | *offset = dp->context_u; | |
260 | *len = PAGE_SIZE - dp->context_u; | |
261 | } | |
262 | ||
263 | static void km_next_page(struct dpages *dp) | |
264 | { | |
265 | dp->context_ptr += PAGE_SIZE - dp->context_u; | |
266 | dp->context_u = 0; | |
267 | } | |
268 | ||
269 | static void km_dp_init(struct dpages *dp, void *data) | |
270 | { | |
271 | dp->get_page = km_get_page; | |
272 | dp->next_page = km_next_page; | |
273 | dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1); | |
274 | dp->context_ptr = data; | |
275 | } | |
276 | ||
1da177e4 LT |
277 | /*----------------------------------------------------------------- |
278 | * IO routines that accept a list of pages. | |
279 | *---------------------------------------------------------------*/ | |
22a1ceb1 | 280 | static void do_region(int rw, unsigned region, struct dm_io_region *where, |
1da177e4 LT |
281 | struct dpages *dp, struct io *io) |
282 | { | |
283 | struct bio *bio; | |
284 | struct page *page; | |
285 | unsigned long len; | |
286 | unsigned offset; | |
287 | unsigned num_bvecs; | |
288 | sector_t remaining = where->count; | |
0c535e0d | 289 | struct request_queue *q = bdev_get_queue(where->bdev); |
70d6c400 MS |
290 | unsigned short logical_block_size = queue_logical_block_size(q); |
291 | sector_t num_sectors; | |
1da177e4 | 292 | |
12fc0f49 | 293 | /* |
d87f4c14 TH |
294 | * where->count may be zero if rw holds a flush and we need to |
295 | * send a zero-sized flush. | |
12fc0f49 MP |
296 | */ |
297 | do { | |
1da177e4 | 298 | /* |
f1e53987 | 299 | * Allocate a suitably sized-bio. |
1da177e4 | 300 | */ |
70d6c400 | 301 | if ((rw & REQ_DISCARD) || (rw & REQ_WRITE_SAME)) |
0c535e0d MB |
302 | num_bvecs = 1; |
303 | else | |
304 | num_bvecs = min_t(int, bio_get_nr_vecs(where->bdev), | |
305 | dm_sector_div_up(remaining, (PAGE_SIZE >> SECTOR_SHIFT))); | |
306 | ||
bf17ce3a | 307 | bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, io->client->bios); |
4f024f37 | 308 | bio->bi_iter.bi_sector = where->sector + (where->count - remaining); |
1da177e4 LT |
309 | bio->bi_bdev = where->bdev; |
310 | bio->bi_end_io = endio; | |
f1e53987 | 311 | store_io_and_region_in_bio(bio, io, region); |
1da177e4 | 312 | |
0c535e0d | 313 | if (rw & REQ_DISCARD) { |
70d6c400 | 314 | num_sectors = min_t(sector_t, q->limits.max_discard_sectors, remaining); |
4f024f37 | 315 | bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT; |
70d6c400 MS |
316 | remaining -= num_sectors; |
317 | } else if (rw & REQ_WRITE_SAME) { | |
318 | /* | |
319 | * WRITE SAME only uses a single page. | |
320 | */ | |
321 | dp->get_page(dp, &page, &len, &offset); | |
322 | bio_add_page(bio, page, logical_block_size, offset); | |
323 | num_sectors = min_t(sector_t, q->limits.max_write_same_sectors, remaining); | |
4f024f37 | 324 | bio->bi_iter.bi_size = num_sectors << SECTOR_SHIFT; |
70d6c400 MS |
325 | |
326 | offset = 0; | |
327 | remaining -= num_sectors; | |
328 | dp->next_page(dp); | |
0c535e0d MB |
329 | } else while (remaining) { |
330 | /* | |
331 | * Try and add as many pages as possible. | |
332 | */ | |
1da177e4 LT |
333 | dp->get_page(dp, &page, &len, &offset); |
334 | len = min(len, to_bytes(remaining)); | |
335 | if (!bio_add_page(bio, page, len, offset)) | |
336 | break; | |
337 | ||
338 | offset = 0; | |
339 | remaining -= to_sector(len); | |
340 | dp->next_page(dp); | |
341 | } | |
342 | ||
343 | atomic_inc(&io->count); | |
344 | submit_bio(rw, bio); | |
12fc0f49 | 345 | } while (remaining); |
1da177e4 LT |
346 | } |
347 | ||
348 | static void dispatch_io(int rw, unsigned int num_regions, | |
22a1ceb1 | 349 | struct dm_io_region *where, struct dpages *dp, |
1da177e4 LT |
350 | struct io *io, int sync) |
351 | { | |
352 | int i; | |
353 | struct dpages old_pages = *dp; | |
354 | ||
f1e53987 MP |
355 | BUG_ON(num_regions > DM_IO_MAX_REGIONS); |
356 | ||
1da177e4 | 357 | if (sync) |
721a9602 | 358 | rw |= REQ_SYNC; |
1da177e4 LT |
359 | |
360 | /* | |
361 | * For multiple regions we need to be careful to rewind | |
362 | * the dp object for each call to do_region. | |
363 | */ | |
364 | for (i = 0; i < num_regions; i++) { | |
365 | *dp = old_pages; | |
d87f4c14 | 366 | if (where[i].count || (rw & REQ_FLUSH)) |
1da177e4 LT |
367 | do_region(rw, i, where + i, dp, io); |
368 | } | |
369 | ||
370 | /* | |
f00b16ad | 371 | * Drop the extra reference that we were holding to avoid |
1da177e4 LT |
372 | * the io being completed too early. |
373 | */ | |
374 | dec_count(io, 0, 0); | |
375 | } | |
376 | ||
97e7cdf1 JT |
377 | struct sync_io { |
378 | unsigned long error_bits; | |
379 | struct completion wait; | |
380 | }; | |
381 | ||
382 | static void sync_io_complete(unsigned long error, void *context) | |
383 | { | |
384 | struct sync_io *sio = context; | |
385 | ||
386 | sio->error_bits = error; | |
387 | complete(&sio->wait); | |
388 | } | |
389 | ||
891ce207 | 390 | static int sync_io(struct dm_io_client *client, unsigned int num_regions, |
22a1ceb1 | 391 | struct dm_io_region *where, int rw, struct dpages *dp, |
891ce207 | 392 | unsigned long *error_bits) |
1da177e4 | 393 | { |
97e7cdf1 JT |
394 | struct io *io; |
395 | struct sync_io sio; | |
1da177e4 | 396 | |
7ff14a36 | 397 | if (num_regions > 1 && (rw & RW_MASK) != WRITE) { |
1da177e4 LT |
398 | WARN_ON(1); |
399 | return -EIO; | |
400 | } | |
401 | ||
97e7cdf1 JT |
402 | init_completion(&sio.wait); |
403 | ||
404 | io = mempool_alloc(client->pool, GFP_NOIO); | |
f1e53987 | 405 | io->error_bits = 0; |
f1e53987 | 406 | atomic_set(&io->count, 1); /* see dispatch_io() */ |
f1e53987 | 407 | io->client = client; |
97e7cdf1 JT |
408 | io->callback = sync_io_complete; |
409 | io->context = &sio; | |
1da177e4 | 410 | |
bb91bc7b MP |
411 | io->vma_invalidate_address = dp->vma_invalidate_address; |
412 | io->vma_invalidate_size = dp->vma_invalidate_size; | |
413 | ||
f1e53987 | 414 | dispatch_io(rw, num_regions, where, dp, io, 1); |
1da177e4 | 415 | |
97e7cdf1 | 416 | wait_for_completion_io(&sio.wait); |
1da177e4 | 417 | |
891ce207 | 418 | if (error_bits) |
97e7cdf1 | 419 | *error_bits = sio.error_bits; |
891ce207 | 420 | |
97e7cdf1 | 421 | return sio.error_bits ? -EIO : 0; |
1da177e4 LT |
422 | } |
423 | ||
891ce207 | 424 | static int async_io(struct dm_io_client *client, unsigned int num_regions, |
22a1ceb1 | 425 | struct dm_io_region *where, int rw, struct dpages *dp, |
891ce207 | 426 | io_notify_fn fn, void *context) |
1da177e4 LT |
427 | { |
428 | struct io *io; | |
429 | ||
7ff14a36 | 430 | if (num_regions > 1 && (rw & RW_MASK) != WRITE) { |
1da177e4 LT |
431 | WARN_ON(1); |
432 | fn(1, context); | |
433 | return -EIO; | |
434 | } | |
435 | ||
bf17ce3a | 436 | io = mempool_alloc(client->pool, GFP_NOIO); |
e01fd7ee | 437 | io->error_bits = 0; |
1da177e4 | 438 | atomic_set(&io->count, 1); /* see dispatch_io() */ |
891ce207 | 439 | io->client = client; |
1da177e4 LT |
440 | io->callback = fn; |
441 | io->context = context; | |
442 | ||
bb91bc7b MP |
443 | io->vma_invalidate_address = dp->vma_invalidate_address; |
444 | io->vma_invalidate_size = dp->vma_invalidate_size; | |
445 | ||
1da177e4 LT |
446 | dispatch_io(rw, num_regions, where, dp, io, 0); |
447 | return 0; | |
448 | } | |
449 | ||
bb91bc7b MP |
450 | static int dp_init(struct dm_io_request *io_req, struct dpages *dp, |
451 | unsigned long size) | |
c8b03afe HM |
452 | { |
453 | /* Set up dpages based on memory type */ | |
bb91bc7b MP |
454 | |
455 | dp->vma_invalidate_address = NULL; | |
456 | dp->vma_invalidate_size = 0; | |
457 | ||
c8b03afe HM |
458 | switch (io_req->mem.type) { |
459 | case DM_IO_PAGE_LIST: | |
460 | list_dp_init(dp, io_req->mem.ptr.pl, io_req->mem.offset); | |
461 | break; | |
462 | ||
003b5c57 KO |
463 | case DM_IO_BIO: |
464 | bio_dp_init(dp, io_req->mem.ptr.bio); | |
c8b03afe HM |
465 | break; |
466 | ||
467 | case DM_IO_VMA: | |
bb91bc7b MP |
468 | flush_kernel_vmap_range(io_req->mem.ptr.vma, size); |
469 | if ((io_req->bi_rw & RW_MASK) == READ) { | |
470 | dp->vma_invalidate_address = io_req->mem.ptr.vma; | |
471 | dp->vma_invalidate_size = size; | |
472 | } | |
c8b03afe HM |
473 | vm_dp_init(dp, io_req->mem.ptr.vma); |
474 | break; | |
475 | ||
476 | case DM_IO_KMEM: | |
477 | km_dp_init(dp, io_req->mem.ptr.addr); | |
478 | break; | |
479 | ||
480 | default: | |
481 | return -EINVAL; | |
482 | } | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | /* | |
7ff14a36 MP |
488 | * New collapsed (a)synchronous interface. |
489 | * | |
490 | * If the IO is asynchronous (i.e. it has notify.fn), you must either unplug | |
97e7cdf1 JT |
491 | * the queue with blk_unplug() some time later or set REQ_SYNC in io_req->bi_rw. |
492 | * If you fail to do one of these, the IO will be submitted to the disk after | |
493 | * q->unplug_delay, which defaults to 3ms in blk-settings.c. | |
c8b03afe HM |
494 | */ |
495 | int dm_io(struct dm_io_request *io_req, unsigned num_regions, | |
22a1ceb1 | 496 | struct dm_io_region *where, unsigned long *sync_error_bits) |
c8b03afe HM |
497 | { |
498 | int r; | |
499 | struct dpages dp; | |
500 | ||
bb91bc7b | 501 | r = dp_init(io_req, &dp, (unsigned long)where->count << SECTOR_SHIFT); |
c8b03afe HM |
502 | if (r) |
503 | return r; | |
504 | ||
505 | if (!io_req->notify.fn) | |
506 | return sync_io(io_req->client, num_regions, where, | |
507 | io_req->bi_rw, &dp, sync_error_bits); | |
508 | ||
509 | return async_io(io_req->client, num_regions, where, io_req->bi_rw, | |
510 | &dp, io_req->notify.fn, io_req->notify.context); | |
511 | } | |
512 | EXPORT_SYMBOL(dm_io); | |
952b3557 MP |
513 | |
514 | int __init dm_io_init(void) | |
515 | { | |
516 | _dm_io_cache = KMEM_CACHE(io, 0); | |
517 | if (!_dm_io_cache) | |
518 | return -ENOMEM; | |
519 | ||
520 | return 0; | |
521 | } | |
522 | ||
523 | void dm_io_exit(void) | |
524 | { | |
525 | kmem_cache_destroy(_dm_io_cache); | |
526 | _dm_io_cache = NULL; | |
527 | } |