dm-io.c revision f00b16ad665a9b489d46f612679181f3f914917b
1/* 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 35static unsigned int pages_to_ios(unsigned int pages) 36{ 37 return 4 * pages; /* too many ? */ 38} 39 40static int resize_pool(unsigned int new_ios) 41{ 42 int r = 0; 43 44 if (_io_pool) { 45 if (new_ios == 0) { 46 /* free off the pool */ 47 mempool_destroy(_io_pool); 48 _io_pool = NULL; 49 bioset_free(_bios); 50 51 } else { 52 /* resize the pool */ 53 r = mempool_resize(_io_pool, new_ios, GFP_KERNEL); 54 } 55 56 } else { 57 /* create new pool */ 58 _io_pool = mempool_create_kmalloc_pool(new_ios, 59 sizeof(struct io)); 60 if (!_io_pool) 61 return -ENOMEM; 62 63 _bios = bioset_create(16, 16, 4); 64 if (!_bios) { 65 mempool_destroy(_io_pool); 66 _io_pool = NULL; 67 return -ENOMEM; 68 } 69 } 70 71 if (!r) 72 _num_ios = new_ios; 73 74 return r; 75} 76 77int dm_io_get(unsigned int num_pages) 78{ 79 return resize_pool(_num_ios + pages_to_ios(num_pages)); 80} 81 82void dm_io_put(unsigned int num_pages) 83{ 84 resize_pool(_num_ios - pages_to_ios(num_pages)); 85} 86 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 *---------------------------------------------------------------*/ 93static inline void bio_set_region(struct bio *bio, unsigned region) 94{ 95 bio->bi_io_vec[bio->bi_max_vecs].bv_len = region; 96} 97 98static inline unsigned bio_get_region(struct bio *bio) 99{ 100 return bio->bi_io_vec[bio->bi_max_vecs].bv_len; 101} 102 103/*----------------------------------------------------------------- 104 * We need an io object to keep track of the number of bios that 105 * have been dispatched for a particular io. 106 *---------------------------------------------------------------*/ 107static void dec_count(struct io *io, unsigned int region, int error) 108{ 109 if (error) 110 set_bit(region, &io->error); 111 112 if (atomic_dec_and_test(&io->count)) { 113 if (io->sleeper) 114 wake_up_process(io->sleeper); 115 116 else { 117 int r = io->error; 118 io_notify_fn fn = io->callback; 119 void *context = io->context; 120 121 mempool_free(io, _io_pool); 122 fn(r, context); 123 } 124 } 125} 126 127static int endio(struct bio *bio, unsigned int done, int error) 128{ 129 struct io *io = (struct io *) bio->bi_private; 130 131 /* keep going until we've finished */ 132 if (bio->bi_size) 133 return 1; 134 135 if (error && bio_data_dir(bio) == READ) 136 zero_fill_bio(bio); 137 138 dec_count(io, bio_get_region(bio), error); 139 bio->bi_max_vecs++; 140 bio_put(bio); 141 142 return 0; 143} 144 145/*----------------------------------------------------------------- 146 * These little objects provide an abstraction for getting a new 147 * destination page for io. 148 *---------------------------------------------------------------*/ 149struct dpages { 150 void (*get_page)(struct dpages *dp, 151 struct page **p, unsigned long *len, unsigned *offset); 152 void (*next_page)(struct dpages *dp); 153 154 unsigned context_u; 155 void *context_ptr; 156}; 157 158/* 159 * Functions for getting the pages from a list. 160 */ 161static void list_get_page(struct dpages *dp, 162 struct page **p, unsigned long *len, unsigned *offset) 163{ 164 unsigned o = dp->context_u; 165 struct page_list *pl = (struct page_list *) dp->context_ptr; 166 167 *p = pl->page; 168 *len = PAGE_SIZE - o; 169 *offset = o; 170} 171 172static void list_next_page(struct dpages *dp) 173{ 174 struct page_list *pl = (struct page_list *) dp->context_ptr; 175 dp->context_ptr = pl->next; 176 dp->context_u = 0; 177} 178 179static void list_dp_init(struct dpages *dp, struct page_list *pl, unsigned offset) 180{ 181 dp->get_page = list_get_page; 182 dp->next_page = list_next_page; 183 dp->context_u = offset; 184 dp->context_ptr = pl; 185} 186 187/* 188 * Functions for getting the pages from a bvec. 189 */ 190static void bvec_get_page(struct dpages *dp, 191 struct page **p, unsigned long *len, unsigned *offset) 192{ 193 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr; 194 *p = bvec->bv_page; 195 *len = bvec->bv_len; 196 *offset = bvec->bv_offset; 197} 198 199static void bvec_next_page(struct dpages *dp) 200{ 201 struct bio_vec *bvec = (struct bio_vec *) dp->context_ptr; 202 dp->context_ptr = bvec + 1; 203} 204 205static void bvec_dp_init(struct dpages *dp, struct bio_vec *bvec) 206{ 207 dp->get_page = bvec_get_page; 208 dp->next_page = bvec_next_page; 209 dp->context_ptr = bvec; 210} 211 212static void vm_get_page(struct dpages *dp, 213 struct page **p, unsigned long *len, unsigned *offset) 214{ 215 *p = vmalloc_to_page(dp->context_ptr); 216 *offset = dp->context_u; 217 *len = PAGE_SIZE - dp->context_u; 218} 219 220static void vm_next_page(struct dpages *dp) 221{ 222 dp->context_ptr += PAGE_SIZE - dp->context_u; 223 dp->context_u = 0; 224} 225 226static void vm_dp_init(struct dpages *dp, void *data) 227{ 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; 232} 233 234static void dm_bio_destructor(struct bio *bio) 235{ 236 bio_free(bio, _bios); 237} 238 239/*----------------------------------------------------------------- 240 * IO routines that accept a list of pages. 241 *---------------------------------------------------------------*/ 242static void do_region(int rw, unsigned int region, struct io_region *where, 243 struct dpages *dp, struct io *io) 244{ 245 struct bio *bio; 246 struct page *page; 247 unsigned long len; 248 unsigned offset; 249 unsigned num_bvecs; 250 sector_t remaining = where->count; 251 252 while (remaining) { 253 /* 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(). 257 */ 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; 265 bio->bi_max_vecs--; 266 bio_set_region(bio, region); 267 268 /* 269 * Try and add as many pages as possible. 270 */ 271 while (remaining) { 272 dp->get_page(dp, &page, &len, &offset); 273 len = min(len, to_bytes(remaining)); 274 if (!bio_add_page(bio, page, len, offset)) 275 break; 276 277 offset = 0; 278 remaining -= to_sector(len); 279 dp->next_page(dp); 280 } 281 282 atomic_inc(&io->count); 283 submit_bio(rw, bio); 284 } 285} 286 287static void dispatch_io(int rw, unsigned int num_regions, 288 struct io_region *where, struct dpages *dp, 289 struct io *io, int sync) 290{ 291 int i; 292 struct dpages old_pages = *dp; 293 294 if (sync) 295 rw |= (1 << BIO_RW_SYNC); 296 297 /* 298 * For multiple regions we need to be careful to rewind 299 * the dp object for each call to do_region. 300 */ 301 for (i = 0; i < num_regions; i++) { 302 *dp = old_pages; 303 if (where[i].count) 304 do_region(rw, i, where + i, dp, io); 305 } 306 307 /* 308 * Drop the extra reference that we were holding to avoid 309 * the io being completed too early. 310 */ 311 dec_count(io, 0, 0); 312} 313 314static int sync_io(unsigned int num_regions, struct io_region *where, 315 int rw, struct dpages *dp, unsigned long *error_bits) 316{ 317 struct io io; 318 319 if (num_regions > 1 && rw != WRITE) { 320 WARN_ON(1); 321 return -EIO; 322 } 323 324 io.error = 0; 325 atomic_set(&io.count, 1); /* see dispatch_io() */ 326 io.sleeper = current; 327 328 dispatch_io(rw, num_regions, where, dp, &io, 1); 329 330 while (1) { 331 set_current_state(TASK_UNINTERRUPTIBLE); 332 333 if (!atomic_read(&io.count) || signal_pending(current)) 334 break; 335 336 io_schedule(); 337 } 338 set_current_state(TASK_RUNNING); 339 340 if (atomic_read(&io.count)) 341 return -EINTR; 342 343 *error_bits = io.error; 344 return io.error ? -EIO : 0; 345} 346 347static int async_io(unsigned int num_regions, struct io_region *where, int rw, 348 struct dpages *dp, io_notify_fn fn, void *context) 349{ 350 struct io *io; 351 352 if (num_regions > 1 && rw != WRITE) { 353 WARN_ON(1); 354 fn(1, context); 355 return -EIO; 356 } 357 358 io = mempool_alloc(_io_pool, GFP_NOIO); 359 io->error = 0; 360 atomic_set(&io->count, 1); /* see dispatch_io() */ 361 io->sleeper = NULL; 362 io->callback = fn; 363 io->context = context; 364 365 dispatch_io(rw, num_regions, where, dp, io, 0); 366 return 0; 367} 368 369int 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) 372{ 373 struct dpages dp; 374 list_dp_init(&dp, pl, offset); 375 return sync_io(num_regions, where, rw, &dp, error_bits); 376} 377 378int dm_io_sync_bvec(unsigned int num_regions, struct io_region *where, int rw, 379 struct bio_vec *bvec, unsigned long *error_bits) 380{ 381 struct dpages dp; 382 bvec_dp_init(&dp, bvec); 383 return sync_io(num_regions, where, rw, &dp, error_bits); 384} 385 386int dm_io_sync_vm(unsigned int num_regions, struct io_region *where, int rw, 387 void *data, unsigned long *error_bits) 388{ 389 struct dpages dp; 390 vm_dp_init(&dp, data); 391 return sync_io(num_regions, where, rw, &dp, error_bits); 392} 393 394int 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) 397{ 398 struct dpages dp; 399 list_dp_init(&dp, pl, offset); 400 return async_io(num_regions, where, rw, &dp, fn, context); 401} 402 403int 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) 405{ 406 struct dpages dp; 407 bvec_dp_init(&dp, bvec); 408 return async_io(num_regions, where, rw, &dp, fn, context); 409} 410 411int dm_io_async_vm(unsigned int num_regions, struct io_region *where, int rw, 412 void *data, io_notify_fn fn, void *context) 413{ 414 struct dpages dp; 415 vm_dp_init(&dp, data); 416 return async_io(num_regions, where, rw, &dp, fn, context); 417} 418 419EXPORT_SYMBOL(dm_io_get); 420EXPORT_SYMBOL(dm_io_put); 421EXPORT_SYMBOL(dm_io_sync); 422EXPORT_SYMBOL(dm_io_async); 423EXPORT_SYMBOL(dm_io_sync_bvec); 424EXPORT_SYMBOL(dm_io_async_bvec); 425EXPORT_SYMBOL(dm_io_sync_vm); 426EXPORT_SYMBOL(dm_io_async_vm); 427