2 * Copyright (C) 2003 Sistina Software
4 * This file is released under the GPL.
10 #include <linux/mempool.h>
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/slab.h>
15 static struct bio_set *_bios;
17 /* FIXME: can we shrink this ? */
21 struct task_struct *sleeper;
22 io_notify_fn callback;
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:
32 static unsigned _num_ios;
33 static mempool_t *_io_pool;
35 static unsigned int pages_to_ios(unsigned int pages)
37 return 4 * pages; /* too many ? */
40 static int resize_pool(unsigned int new_ios)
46 /* free off the pool */
47 mempool_destroy(_io_pool);
53 r = mempool_resize(_io_pool, new_ios, GFP_KERNEL);
58 _io_pool = mempool_create_kmalloc_pool(new_ios,
63 _bios = bioset_create(16, 16);
65 mempool_destroy(_io_pool);
77 int dm_io_get(unsigned int num_pages)
79 return resize_pool(_num_ios + pages_to_ios(num_pages));
82 void dm_io_put(unsigned int num_pages)
84 resize_pool(_num_ios - pages_to_ios(num_pages));
87 /*-----------------------------------------------------------------
88 * We need to keep track of which region a bio is doing io for.
89 * In order to save a memory allocation we store this the last
90 * bvec which we know is unused (blech).
91 * XXX This is ugly and can OOPS with some configs... find another way.
92 *---------------------------------------------------------------*/
93 static inline void bio_set_region(struct bio *bio, unsigned region)
95 bio->bi_io_vec[bio->bi_max_vecs].bv_len = region;
98 static inline unsigned bio_get_region(struct bio *bio)
100 return bio->bi_io_vec[bio->bi_max_vecs].bv_len;
103 /*-----------------------------------------------------------------
104 * We need an io object to keep track of the number of bios that
105 * have been dispatched for a particular io.
106 *---------------------------------------------------------------*/
107 static void dec_count(struct io *io, unsigned int region, int error)
110 set_bit(region, &io->error);
112 if (atomic_dec_and_test(&io->count)) {
114 wake_up_process(io->sleeper);
118 io_notify_fn fn = io->callback;
119 void *context = io->context;
121 mempool_free(io, _io_pool);
127 static int endio(struct bio *bio, unsigned int done, int error)
129 struct io *io = (struct io *) bio->bi_private;
131 /* keep going until we've finished */
135 if (error && bio_data_dir(bio) == READ)
138 dec_count(io, bio_get_region(bio), error);
145 /*-----------------------------------------------------------------
146 * These little objects provide an abstraction for getting a new
147 * destination page for io.
148 *---------------------------------------------------------------*/
150 void (*get_page)(struct dpages *dp,
151 struct page **p, unsigned long *len, unsigned *offset);
152 void (*next_page)(struct dpages *dp);
159 * Functions for getting the pages from a list.
161 static void list_get_page(struct dpages *dp,
162 struct page **p, unsigned long *len, unsigned *offset)
164 unsigned o = dp->context_u;
165 struct page_list *pl = (struct page_list *) dp->context_ptr;
168 *len = PAGE_SIZE - o;
172 static void list_next_page(struct dpages *dp)
174 struct page_list *pl = (struct page_list *) dp->context_ptr;
175 dp->context_ptr = pl->next;
179 static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset)
181 dp->get_page = list_get_page;
182 dp->next_page = list_next_page;
183 dp->context_u = offset;
184 dp->context_ptr = pl;
188 * Functions for getting the pages from a bvec.
190 static void bvec_get_page(struct dpages *dp,
191 struct page **p, unsigned long *len, unsigned *offset)
193 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
196 *offset = bvec->bv_offset;
199 static void bvec_next_page(struct dpages *dp)
201 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
202 dp->context_ptr = bvec + 1;
205 static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec)
207 dp->get_page = bvec_get_page;
208 dp->next_page = bvec_next_page;
209 dp->context_ptr = bvec;
212 static void vm_get_page(struct dpages *dp,
213 struct page **p, unsigned long *len, unsigned *offset)
215 *p = vmalloc_to_page(dp->context_ptr);
216 *offset = dp->context_u;
217 *len = PAGE_SIZE - dp->context_u;
220 static void vm_next_page(struct dpages *dp)
222 dp->context_ptr += PAGE_SIZE - dp->context_u;
226 static void vm_dp_init(struct dpages *dp, void *data)
228 dp->get_page = vm_get_page;
229 dp->next_page = vm_next_page;
230 dp->context_u = ((unsigned long) data) & (PAGE_SIZE - 1);
231 dp->context_ptr = data;
234 static void dm_bio_destructor(struct bio *bio)
236 bio_free(bio, _bios);
239 /*-----------------------------------------------------------------
240 * IO routines that accept a list of pages.
241 *---------------------------------------------------------------*/
242 static void do_region(int rw, unsigned int region, struct io_region *where,
243 struct dpages *dp, struct io *io)
250 sector_t remaining = where->count;
254 * Allocate a suitably sized-bio: we add an extra
255 * bvec for bio_get/set_region() and decrement bi_max_vecs
256 * to hide it from bio_add_page().
258 num_bvecs = (remaining / (PAGE_SIZE >> SECTOR_SHIFT)) + 2;
259 bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, _bios);
260 bio->bi_sector = where->sector + (where->count - remaining);
261 bio->bi_bdev = where->bdev;
262 bio->bi_end_io = endio;
263 bio->bi_private = io;
264 bio->bi_destructor = dm_bio_destructor;
266 bio_set_region(bio, region);
269 * Try and add as many pages as possible.
272 dp->get_page(dp, &page, &len, &offset);
273 len = min(len, to_bytes(remaining));
274 if (!bio_add_page(bio, page, len, offset))
278 remaining -= to_sector(len);
282 atomic_inc(&io->count);
287 static void dispatch_io(int rw, unsigned int num_regions,
288 struct io_region *where, struct dpages *dp,
289 struct io *io, int sync)
292 struct dpages old_pages = *dp;
295 rw |= (1 << BIO_RW_SYNC);
298 * For multiple regions we need to be careful to rewind
299 * the dp object for each call to do_region.
301 for (i = 0; i < num_regions; i++) {
304 do_region(rw, i, where + i, dp, io);
308 * Drop the extra reference that we were holding to avoid
309 * the io being completed too early.
314 static int sync_io(unsigned int num_regions, struct io_region *where,
315 int rw, struct dpages *dp, unsigned long *error_bits)
319 if (num_regions > 1 && rw != WRITE) {
325 atomic_set(&io.count, 1); /* see dispatch_io() */
326 io.sleeper = current;
328 dispatch_io(rw, num_regions, where, dp, &io, 1);
331 set_current_state(TASK_UNINTERRUPTIBLE);
333 if (!atomic_read(&io.count) || signal_pending(current))
338 set_current_state(TASK_RUNNING);
340 if (atomic_read(&io.count))
343 *error_bits = io.error;
344 return io.error ? -EIO : 0;
347 static int async_io(unsigned int num_regions, struct io_region *where, int rw,
348 struct dpages *dp, io_notify_fn fn, void *context)
352 if (num_regions > 1 && rw != WRITE) {
358 io = mempool_alloc(_io_pool, GFP_NOIO);
360 atomic_set(&io->count, 1); /* see dispatch_io() */
363 io->context = context;
365 dispatch_io(rw, num_regions, where, dp, io, 0);
369 int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw,
370 struct page_list *pl, unsigned int offset,
371 unsigned long *error_bits)
374 list_dp_init(&dp, pl, offset);
375 return sync_io(num_regions, where, rw, &dp, error_bits);
378 int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw,
379 struct bio_vec *bvec, unsigned long *error_bits)
382 bvec_dp_init(&dp, bvec);
383 return sync_io(num_regions, where, rw, &dp, error_bits);
386 int dm_io_sync_vm(unsigned int num_regions, struct io_region *where, int rw,
387 void *data, unsigned long *error_bits)
390 vm_dp_init(&dp, data);
391 return sync_io(num_regions, where, rw, &dp, error_bits);
394 int dm_io_async(unsigned int num_regions, struct io_region *where, int rw,
395 struct page_list *pl, unsigned int offset,
396 io_notify_fn fn, void *context)
399 list_dp_init(&dp, pl, offset);
400 return async_io(num_regions, where, rw, &dp, fn, context);
403 int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw,
404 struct bio_vec *bvec, io_notify_fn fn, void *context)
407 bvec_dp_init(&dp, bvec);
408 return async_io(num_regions, where, rw, &dp, fn, context);
411 int dm_io_async_vm(unsigned int num_regions, struct io_region *where, int rw,
412 void *data, io_notify_fn fn, void *context)
415 vm_dp_init(&dp, data);
416 return async_io(num_regions, where, rw, &dp, fn, context);
419 EXPORT_SYMBOL(dm_io_get);
420 EXPORT_SYMBOL(dm_io_put);
421 EXPORT_SYMBOL(dm_io_sync);
422 EXPORT_SYMBOL(dm_io_async);
423 EXPORT_SYMBOL(dm_io_sync_bvec);
424 EXPORT_SYMBOL(dm_io_async_bvec);
425 EXPORT_SYMBOL(dm_io_sync_vm);
426 EXPORT_SYMBOL(dm_io_async_vm);