blob: 4809b209fbb195cc5c74ebed73b99558aeb40510 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2003 Sistina Software
3 *
4 * This file is released under the GPL.
5 */
6
7#include "dm-io.h"
8
9#include <linux/bio.h>
10#include <linux/mempool.h>
11#include <linux/module.h>
12#include <linux/sched.h>
13#include <linux/slab.h>
14
15static struct bio_set *_bios;
16
17/* FIXME: can we shrink this ? */
18struct io {
19 unsigned long error;
20 atomic_t count;
21 struct task_struct *sleeper;
22 io_notify_fn callback;
23 void *context;
24};
25
26/*
27 * io contexts are only dynamically allocated for asynchronous
28 * io. Since async io is likely to be the majority of io we'll
29 * have the same number of io contexts as buffer heads ! (FIXME:
30 * must reduce this).
31 */
32static unsigned _num_ios;
33static mempool_t *_io_pool;
34
Al Virodd0fc662005-10-07 07:46:04 +010035static void *alloc_io(gfp_t gfp_mask, void *pool_data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070036{
37 return kmalloc(sizeof(struct io), gfp_mask);
38}
39
40static void free_io(void *element, void *pool_data)
41{
42 kfree(element);
43}
44
45static unsigned int pages_to_ios(unsigned int pages)
46{
47 return 4 * pages; /* too many ? */
48}
49
50static int resize_pool(unsigned int new_ios)
51{
52 int r = 0;
53
54 if (_io_pool) {
55 if (new_ios == 0) {
56 /* free off the pool */
57 mempool_destroy(_io_pool);
58 _io_pool = NULL;
59 bioset_free(_bios);
60
61 } else {
62 /* resize the pool */
63 r = mempool_resize(_io_pool, new_ios, GFP_KERNEL);
64 }
65
66 } else {
67 /* create new pool */
68 _io_pool = mempool_create(new_ios, alloc_io, free_io, NULL);
69 if (!_io_pool)
70 return -ENOMEM;
71
72 _bios = bioset_create(16, 16, 4);
73 if (!_bios) {
74 mempool_destroy(_io_pool);
75 _io_pool = NULL;
76 return -ENOMEM;
77 }
78 }
79
80 if (!r)
81 _num_ios = new_ios;
82
83 return r;
84}
85
86int dm_io_get(unsigned int num_pages)
87{
88 return resize_pool(_num_ios + pages_to_ios(num_pages));
89}
90
91void dm_io_put(unsigned int num_pages)
92{
93 resize_pool(_num_ios - pages_to_ios(num_pages));
94}
95
96/*-----------------------------------------------------------------
97 * We need to keep track of which region a bio is doing io for.
98 * In order to save a memory allocation we store this the last
99 * bvec which we know is unused (blech).
100 * XXX This is ugly and can OOPS with some configs... find another way.
101 *---------------------------------------------------------------*/
102static inline void bio_set_region(struct bio *bio, unsigned region)
103{
104 bio->bi_io_vec[bio->bi_max_vecs - 1].bv_len = region;
105}
106
107static inline unsigned bio_get_region(struct bio *bio)
108{
109 return bio->bi_io_vec[bio->bi_max_vecs - 1].bv_len;
110}
111
112/*-----------------------------------------------------------------
113 * We need an io object to keep track of the number of bios that
114 * have been dispatched for a particular io.
115 *---------------------------------------------------------------*/
116static void dec_count(struct io *io, unsigned int region, int error)
117{
118 if (error)
119 set_bit(region, &io->error);
120
121 if (atomic_dec_and_test(&io->count)) {
122 if (io->sleeper)
123 wake_up_process(io->sleeper);
124
125 else {
126 int r = io->error;
127 io_notify_fn fn = io->callback;
128 void *context = io->context;
129
130 mempool_free(io, _io_pool);
131 fn(r, context);
132 }
133 }
134}
135
136static int endio(struct bio *bio, unsigned int done, int error)
137{
138 struct io *io = (struct io *) bio->bi_private;
139
140 /* keep going until we've finished */
141 if (bio->bi_size)
142 return 1;
143
144 if (error && bio_data_dir(bio) == READ)
145 zero_fill_bio(bio);
146
147 dec_count(io, bio_get_region(bio), error);
148 bio_put(bio);
149
150 return 0;
151}
152
153/*-----------------------------------------------------------------
154 * These little objects provide an abstraction for getting a new
155 * destination page for io.
156 *---------------------------------------------------------------*/
157struct dpages {
158 void (*get_page)(struct dpages *dp,
159 struct page **p, unsigned long *len, unsigned *offset);
160 void (*next_page)(struct dpages *dp);
161
162 unsigned context_u;
163 void *context_ptr;
164};
165
166/*
167 * Functions for getting the pages from a list.
168 */
169static void list_get_page(struct dpages *dp,
170 struct page **p, unsigned long *len, unsigned *offset)
171{
172 unsigned o = dp->context_u;
173 struct page_list *pl = (struct page_list *) dp->context_ptr;
174
175 *p = pl->page;
176 *len = PAGE_SIZE - o;
177 *offset = o;
178}
179
180static void list_next_page(struct dpages *dp)
181{
182 struct page_list *pl = (struct page_list *) dp->context_ptr;
183 dp->context_ptr = pl->next;
184 dp->context_u = 0;
185}
186
187static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset)
188{
189 dp->get_page = list_get_page;
190 dp->next_page = list_next_page;
191 dp->context_u = offset;
192 dp->context_ptr = pl;
193}
194
195/*
196 * Functions for getting the pages from a bvec.
197 */
198static void bvec_get_page(struct dpages *dp,
199 struct page **p, unsigned long *len, unsigned *offset)
200{
201 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
202 *p = bvec->bv_page;
203 *len = bvec->bv_len;
204 *offset = bvec->bv_offset;
205}
206
207static void bvec_next_page(struct dpages *dp)
208{
209 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
210 dp->context_ptr = bvec + 1;
211}
212
213static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec)
214{
215 dp->get_page = bvec_get_page;
216 dp->next_page = bvec_next_page;
217 dp->context_ptr = bvec;
218}
219
220static void vm_get_page(struct dpages *dp,
221 struct page **p, unsigned long *len, unsigned *offset)
222{
223 *p = vmalloc_to_page(dp->context_ptr);
224 *offset = dp->context_u;
225 *len = PAGE_SIZE - dp->context_u;
226}
227
228static void vm_next_page(struct dpages *dp)
229{
230 dp->context_ptr += PAGE_SIZE - dp->context_u;
231 dp->context_u = 0;
232}
233
234static void vm_dp_init(struct dpages *dp, void *data)
235{
236 dp->get_page = vm_get_page;
237 dp->next_page = vm_next_page;
238 dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
239 dp->context_ptr = data;
240}
241
Peter Osterlund36763472005-09-06 15:16:42 -0700242static void dm_bio_destructor(struct bio *bio)
243{
244 bio_free(bio, _bios);
245}
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/*-----------------------------------------------------------------
248 * IO routines that accept a list of pages.
249 *---------------------------------------------------------------*/
250static void do_region(int rw, unsigned int region, struct io_region *where,
251 struct dpages *dp, struct io *io)
252{
253 struct bio *bio;
254 struct page *page;
255 unsigned long len;
256 unsigned offset;
257 unsigned num_bvecs;
258 sector_t remaining = where->count;
259
260 while (remaining) {
261 /*
262 * Allocate a suitably sized bio, we add an extra
263 * bvec for bio_get/set_region().
264 */
265 num_bvecs = (remaining / (PAGE_SIZE >> 9)) + 2;
266 bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, _bios);
267 bio->bi_sector = where->sector + (where->count - remaining);
268 bio->bi_bdev = where->bdev;
269 bio->bi_end_io = endio;
270 bio->bi_private = io;
Peter Osterlund36763472005-09-06 15:16:42 -0700271 bio->bi_destructor = dm_bio_destructor;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 bio_set_region(bio, region);
273
274 /*
275 * Try and add as many pages as possible.
276 */
277 while (remaining) {
278 dp->get_page(dp, &page, &len, &offset);
279 len = min(len, to_bytes(remaining));
280 if (!bio_add_page(bio, page, len, offset))
281 break;
282
283 offset = 0;
284 remaining -= to_sector(len);
285 dp->next_page(dp);
286 }
287
288 atomic_inc(&io->count);
289 submit_bio(rw, bio);
290 }
291}
292
293static void dispatch_io(int rw, unsigned int num_regions,
294 struct io_region *where, struct dpages *dp,
295 struct io *io, int sync)
296{
297 int i;
298 struct dpages old_pages = *dp;
299
300 if (sync)
301 rw |= (1 << BIO_RW_SYNC);
302
303 /*
304 * For multiple regions we need to be careful to rewind
305 * the dp object for each call to do_region.
306 */
307 for (i = 0; i < num_regions; i++) {
308 *dp = old_pages;
309 if (where[i].count)
310 do_region(rw, i, where + i, dp, io);
311 }
312
313 /*
314 * Drop the extra refence that we were holding to avoid
315 * the io being completed too early.
316 */
317 dec_count(io, 0, 0);
318}
319
320static int sync_io(unsigned int num_regions, struct io_region *where,
321 int rw, struct dpages *dp, unsigned long *error_bits)
322{
323 struct io io;
324
325 if (num_regions > 1 && rw != WRITE) {
326 WARN_ON(1);
327 return -EIO;
328 }
329
330 io.error = 0;
331 atomic_set(&io.count, 1); /* see dispatch_io() */
332 io.sleeper = current;
333
334 dispatch_io(rw, num_regions, where, dp, &io, 1);
335
336 while (1) {
337 set_current_state(TASK_UNINTERRUPTIBLE);
338
339 if (!atomic_read(&io.count) || signal_pending(current))
340 break;
341
342 io_schedule();
343 }
344 set_current_state(TASK_RUNNING);
345
346 if (atomic_read(&io.count))
347 return -EINTR;
348
349 *error_bits = io.error;
350 return io.error ? -EIO : 0;
351}
352
353static int async_io(unsigned int num_regions, struct io_region *where, int rw,
354 struct dpages *dp, io_notify_fn fn, void *context)
355{
356 struct io *io;
357
358 if (num_regions > 1 && rw != WRITE) {
359 WARN_ON(1);
360 fn(1, context);
361 return -EIO;
362 }
363
364 io = mempool_alloc(_io_pool, GFP_NOIO);
365 io->error = 0;
366 atomic_set(&io->count, 1); /* see dispatch_io() */
367 io->sleeper = NULL;
368 io->callback = fn;
369 io->context = context;
370
371 dispatch_io(rw, num_regions, where, dp, io, 0);
372 return 0;
373}
374
375int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw,
376 struct page_list *pl, unsigned int offset,
377 unsigned long *error_bits)
378{
379 struct dpages dp;
380 list_dp_init(&dp, pl, offset);
381 return sync_io(num_regions, where, rw, &dp, error_bits);
382}
383
384int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw,
385 struct bio_vec *bvec, unsigned long *error_bits)
386{
387 struct dpages dp;
388 bvec_dp_init(&dp, bvec);
389 return sync_io(num_regions, where, rw, &dp, error_bits);
390}
391
392int dm_io_sync_vm(unsigned int num_regions, struct io_region *where, int rw,
393 void *data, unsigned long *error_bits)
394{
395 struct dpages dp;
396 vm_dp_init(&dp, data);
397 return sync_io(num_regions, where, rw, &dp, error_bits);
398}
399
400int dm_io_async(unsigned int num_regions, struct io_region *where, int rw,
401 struct page_list *pl, unsigned int offset,
402 io_notify_fn fn, void *context)
403{
404 struct dpages dp;
405 list_dp_init(&dp, pl, offset);
406 return async_io(num_regions, where, rw, &dp, fn, context);
407}
408
409int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw,
410 struct bio_vec *bvec, io_notify_fn fn, void *context)
411{
412 struct dpages dp;
413 bvec_dp_init(&dp, bvec);
414 return async_io(num_regions, where, rw, &dp, fn, context);
415}
416
417int dm_io_async_vm(unsigned int num_regions, struct io_region *where, int rw,
418 void *data, io_notify_fn fn, void *context)
419{
420 struct dpages dp;
421 vm_dp_init(&dp, data);
422 return async_io(num_regions, where, rw, &dp, fn, context);
423}
424
425EXPORT_SYMBOL(dm_io_get);
426EXPORT_SYMBOL(dm_io_put);
427EXPORT_SYMBOL(dm_io_sync);
428EXPORT_SYMBOL(dm_io_async);
429EXPORT_SYMBOL(dm_io_sync_bvec);
430EXPORT_SYMBOL(dm_io_async_bvec);
431EXPORT_SYMBOL(dm_io_sync_vm);
432EXPORT_SYMBOL(dm_io_async_vm);