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)
132 /* keep going until we've finished */
136 if (error && bio_data_dir(bio) == READ)
140 * The bio destructor in bio_put() may use the io object.
142 io = bio->bi_private;
143 region = bio_get_region(bio);
148 dec_count(io, region, error);
153 /*-----------------------------------------------------------------
154 * These little objects provide an abstraction for getting a new
155 * destination page for io.
156 *---------------------------------------------------------------*/
158 void (*get_page)(struct dpages *dp,
159 struct page **p, unsigned long *len, unsigned *offset);
160 void (*next_page)(struct dpages *dp);
167 * Functions for getting the pages from a list.
169 static void list_get_page(struct dpages *dp,
170 struct page **p, unsigned long *len, unsigned *offset)
172 unsigned o = dp->context_u;
173 struct page_list *pl = (struct page_list *) dp->context_ptr;
176 *len = PAGE_SIZE - o;
180 static void list_next_page(struct dpages *dp)
182 struct page_list *pl = (struct page_list *) dp->context_ptr;
183 dp->context_ptr = pl->next;
187 static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset)
189 dp->get_page = list_get_page;
190 dp->next_page = list_next_page;
191 dp->context_u = offset;
192 dp->context_ptr = pl;
196 * Functions for getting the pages from a bvec.
198 static void bvec_get_page(struct dpages *dp,
199 struct page **p, unsigned long *len, unsigned *offset)
201 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
204 *offset = bvec->bv_offset;
207 static void bvec_next_page(struct dpages *dp)
209 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr;
210 dp->context_ptr = bvec + 1;
213 static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec)
215 dp->get_page = bvec_get_page;
216 dp->next_page = bvec_next_page;
217 dp->context_ptr = bvec;
220 static void vm_get_page(struct dpages *dp,
221 struct page **p, unsigned long *len, unsigned *offset)
223 *p = vmalloc_to_page(dp->context_ptr);
224 *offset = dp->context_u;
225 *len = PAGE_SIZE - dp->context_u;
228 static void vm_next_page(struct dpages *dp)
230 dp->context_ptr += PAGE_SIZE - dp->context_u;
234 static void vm_dp_init(struct dpages *dp, void *data)
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;
242 static void dm_bio_destructor(struct bio *bio)
244 bio_free(bio, _bios);
247 /*-----------------------------------------------------------------
248 * IO routines that accept a list of pages.
249 *---------------------------------------------------------------*/
250 static void do_region(int rw, unsigned int region, struct io_region *where,
251 struct dpages *dp, struct io *io)
258 sector_t remaining = where->count;
262 * Allocate a suitably sized-bio: we add an extra
263 * bvec for bio_get/set_region() and decrement bi_max_vecs
264 * to hide it from bio_add_page().
266 num_bvecs = (remaining / (PAGE_SIZE >> SECTOR_SHIFT)) + 2;
267 bio = bio_alloc_bioset(GFP_NOIO, num_bvecs, _bios);
268 bio->bi_sector = where->sector + (where->count - remaining);
269 bio->bi_bdev = where->bdev;
270 bio->bi_end_io = endio;
271 bio->bi_private = io;
272 bio->bi_destructor = dm_bio_destructor;
274 bio_set_region(bio, region);
277 * Try and add as many pages as possible.
280 dp->get_page(dp, &page, &len, &offset);
281 len = min(len, to_bytes(remaining));
282 if (!bio_add_page(bio, page, len, offset))
286 remaining -= to_sector(len);
290 atomic_inc(&io->count);
295 static void dispatch_io(int rw, unsigned int num_regions,
296 struct io_region *where, struct dpages *dp,
297 struct io *io, int sync)
300 struct dpages old_pages = *dp;
303 rw |= (1 << BIO_RW_SYNC);
306 * For multiple regions we need to be careful to rewind
307 * the dp object for each call to do_region.
309 for (i = 0; i < num_regions; i++) {
312 do_region(rw, i, where + i, dp, io);
316 * Drop the extra reference that we were holding to avoid
317 * the io being completed too early.
322 static int sync_io(unsigned int num_regions, struct io_region *where,
323 int rw, struct dpages *dp, unsigned long *error_bits)
327 if (num_regions > 1 && rw != WRITE) {
333 atomic_set(&io.count, 1); /* see dispatch_io() */
334 io.sleeper = current;
336 dispatch_io(rw, num_regions, where, dp, &io, 1);
339 set_current_state(TASK_UNINTERRUPTIBLE);
341 if (!atomic_read(&io.count) || signal_pending(current))
346 set_current_state(TASK_RUNNING);
348 if (atomic_read(&io.count))
351 *error_bits = io.error;
352 return io.error ? -EIO : 0;
355 static int async_io(unsigned int num_regions, struct io_region *where, int rw,
356 struct dpages *dp, io_notify_fn fn, void *context)
360 if (num_regions > 1 && rw != WRITE) {
366 io = mempool_alloc(_io_pool, GFP_NOIO);
368 atomic_set(&io->count, 1); /* see dispatch_io() */
371 io->context = context;
373 dispatch_io(rw, num_regions, where, dp, io, 0);
377 int dm_io_sync(unsigned int num_regions, struct io_region *where, int rw,
378 struct page_list *pl, unsigned int offset,
379 unsigned long *error_bits)
382 list_dp_init(&dp, pl, offset);
383 return sync_io(num_regions, where, rw, &dp, error_bits);
386 int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw,
387 struct bio_vec *bvec, unsigned long *error_bits)
390 bvec_dp_init(&dp, bvec);
391 return sync_io(num_regions, where, rw, &dp, error_bits);
394 int dm_io_sync_vm(unsigned int num_regions, struct io_region *where, int rw,
395 void *data, unsigned long *error_bits)
398 vm_dp_init(&dp, data);
399 return sync_io(num_regions, where, rw, &dp, error_bits);
402 int dm_io_async(unsigned int num_regions, struct io_region *where, int rw,
403 struct page_list *pl, unsigned int offset,
404 io_notify_fn fn, void *context)
407 list_dp_init(&dp, pl, offset);
408 return async_io(num_regions, where, rw, &dp, fn, context);
411 int dm_io_async_bvec(unsigned int num_regions, struct io_region *where, int rw,
412 struct bio_vec *bvec, io_notify_fn fn, void *context)
415 bvec_dp_init(&dp, bvec);
416 return async_io(num_regions, where, rw, &dp, fn, context);
419 int dm_io_async_vm(unsigned int num_regions, struct io_region *where, int rw,
420 void *data, io_notify_fn fn, void *context)
423 vm_dp_init(&dp, data);
424 return async_io(num_regions, where, rw, &dp, fn, context);
427 EXPORT_SYMBOL(dm_io_get);
428 EXPORT_SYMBOL(dm_io_put);
429 EXPORT_SYMBOL(dm_io_sync);
430 EXPORT_SYMBOL(dm_io_async);
431 EXPORT_SYMBOL(dm_io_sync_bvec);
432 EXPORT_SYMBOL(dm_io_async_bvec);
433 EXPORT_SYMBOL(dm_io_sync_vm);
434 EXPORT_SYMBOL(dm_io_async_vm);