unix_io.c revision a85e81a2ff4fb1afc05ff74d5da573031c3495e0
15d1f7b1de12d16ceb2c938c56701a3e8bfa558f7Torne (Richard Coles)/* 21320f92c476a1ad9d19dba2a48c72b75566198e9Primiano Tucci * unix_io.c --- This is the Unix (well, really POSIX) implementation 35d1f7b1de12d16ceb2c938c56701a3e8bfa558f7Torne (Richard Coles) * of the I/O manager. 45d1f7b1de12d16ceb2c938c56701a3e8bfa558f7Torne (Richard Coles) * 5 * Implements a one-block write-through cache. 6 * 7 * Includes support for Windows NT support under Cygwin. 8 * 9 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, 10 * 2002 by Theodore Ts'o. 11 * 12 * %Begin-Header% 13 * This file may be redistributed under the terms of the GNU Public 14 * License. 15 * %End-Header% 16 */ 17 18#define _LARGEFILE_SOURCE 19#define _LARGEFILE64_SOURCE 20 21#include <stdio.h> 22#include <string.h> 23#if HAVE_UNISTD_H 24#include <unistd.h> 25#endif 26#if HAVE_ERRNO_H 27#include <errno.h> 28#endif 29#include <fcntl.h> 30#include <time.h> 31#ifdef __linux__ 32#include <sys/utsname.h> 33#endif 34#if HAVE_SYS_STAT_H 35#include <sys/stat.h> 36#endif 37#if HAVE_SYS_TYPES_H 38#include <sys/types.h> 39#endif 40#if HAVE_SYS_RESOURCE_H 41#include <sys/resource.h> 42#endif 43 44#include "ext2_fs.h" 45#include "ext2fs.h" 46 47/* 48 * For checking structure magic numbers... 49 */ 50 51#define EXT2_CHECK_MAGIC(struct, code) \ 52 if ((struct)->magic != (code)) return (code) 53 54struct unix_cache { 55 char *buf; 56 unsigned long block; 57 int access_time; 58 int dirty:1; 59 int in_use:1; 60}; 61 62#define CACHE_SIZE 8 63#define WRITE_DIRECT_SIZE 4 /* Must be smaller than CACHE_SIZE */ 64#define READ_DIRECT_SIZE 4 /* Should be smaller than CACHE_SIZE */ 65 66struct unix_private_data { 67 int magic; 68 int dev; 69 int flags; 70 int access_time; 71 struct unix_cache cache[CACHE_SIZE]; 72}; 73 74static errcode_t unix_open(const char *name, int flags, io_channel *channel); 75static errcode_t unix_close(io_channel channel); 76static errcode_t unix_set_blksize(io_channel channel, int blksize); 77static errcode_t unix_read_blk(io_channel channel, unsigned long block, 78 int count, void *data); 79static errcode_t unix_write_blk(io_channel channel, unsigned long block, 80 int count, const void *data); 81static errcode_t unix_flush(io_channel channel); 82static errcode_t unix_write_byte(io_channel channel, unsigned long offset, 83 int size, const void *data); 84 85static void reuse_cache(io_channel channel, struct unix_private_data *data, 86 struct unix_cache *cache, unsigned long block); 87 88static struct struct_io_manager struct_unix_manager = { 89 EXT2_ET_MAGIC_IO_MANAGER, 90 "Unix I/O Manager", 91 unix_open, 92 unix_close, 93 unix_set_blksize, 94 unix_read_blk, 95 unix_write_blk, 96 unix_flush, 97#ifdef __CYGWIN__ 98 0 99#else 100 unix_write_byte 101#endif 102}; 103 104io_manager unix_io_manager = &struct_unix_manager; 105 106/* 107 * Here are the raw I/O functions 108 */ 109#ifndef __CYGWIN__ 110static errcode_t raw_read_blk(io_channel channel, 111 struct unix_private_data *data, 112 unsigned long block, 113 int count, void *buf) 114{ 115 errcode_t retval; 116 size_t size; 117 ext2_loff_t location; 118 int actual = 0; 119 120 size = (count < 0) ? -count : count * channel->block_size; 121 location = (ext2_loff_t) block * channel->block_size; 122 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { 123 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; 124 goto error_out; 125 } 126 actual = read(data->dev, buf, size); 127 if (actual != size) { 128 if (actual < 0) 129 actual = 0; 130 retval = EXT2_ET_SHORT_READ; 131 goto error_out; 132 } 133 return 0; 134 135error_out: 136 memset((char *) buf+actual, 0, size-actual); 137 if (channel->read_error) 138 retval = (channel->read_error)(channel, block, count, buf, 139 size, actual, retval); 140 return retval; 141} 142#else /* __CYGWIN__ */ 143/* 144 * Windows block devices only allow sector alignment IO in offset and size 145 */ 146static errcode_t raw_read_blk(io_channel channel, 147 struct unix_private_data *data, 148 unsigned long block, 149 int count, void *buf) 150{ 151 errcode_t retval; 152 size_t size, alignsize, fragment; 153 ext2_loff_t location; 154 int total = 0, actual; 155#define BLOCKALIGN 512 156 char sector[BLOCKALIGN]; 157 158 size = (count < 0) ? -count : count * channel->block_size; 159 location = (ext2_loff_t) block * channel->block_size; 160#ifdef DEBUG 161 printf("count=%d, size=%d, block=%d, blk_size=%d, location=%lx\n", 162 count, size, block, channel->block_size, location); 163#endif 164 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { 165 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; 166 goto error_out; 167 } 168 fragment = size % BLOCKALIGN; 169 alignsize = size - fragment; 170 if (alignsize) { 171 actual = read(data->dev, buf, alignsize); 172 if (actual != alignsize) 173 goto short_read; 174 } 175 if (fragment) { 176 actual = read(data->dev, sector, BLOCKALIGN); 177 if (actual != BLOCKALIGN) 178 goto short_read; 179 memcpy(buf+alignsize, sector, fragment); 180 } 181 return 0; 182 183short_read: 184 if (actual>0) 185 total += actual; 186 retval = EXT2_ET_SHORT_READ; 187 188error_out: 189 memset((char *) buf+total, 0, size-actual); 190 if (channel->read_error) 191 retval = (channel->read_error)(channel, block, count, buf, 192 size, actual, retval); 193 return retval; 194} 195#endif 196 197static errcode_t raw_write_blk(io_channel channel, 198 struct unix_private_data *data, 199 unsigned long block, 200 int count, const void *buf) 201{ 202 size_t size; 203 ext2_loff_t location; 204 int actual = 0; 205 errcode_t retval; 206 207 if (count == 1) 208 size = channel->block_size; 209 else { 210 if (count < 0) 211 size = -count; 212 else 213 size = count * channel->block_size; 214 } 215 216 location = (ext2_loff_t) block * channel->block_size; 217 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { 218 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; 219 goto error_out; 220 } 221 222 actual = write(data->dev, buf, size); 223 if (actual != size) { 224 retval = EXT2_ET_SHORT_WRITE; 225 goto error_out; 226 } 227 return 0; 228 229error_out: 230 if (channel->write_error) 231 retval = (channel->write_error)(channel, block, count, buf, 232 size, actual, retval); 233 return retval; 234} 235 236 237/* 238 * Here we implement the cache functions 239 */ 240 241/* Allocate the cache buffers */ 242static errcode_t alloc_cache(io_channel channel, 243 struct unix_private_data *data) 244{ 245 errcode_t retval; 246 struct unix_cache *cache; 247 int i; 248 249 data->access_time = 0; 250 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 251 cache->block = 0; 252 cache->access_time = 0; 253 cache->dirty = 0; 254 cache->in_use = 0; 255 if ((retval = ext2fs_get_mem(channel->block_size, 256 (void **) &cache->buf))) 257 return retval; 258 } 259 return 0; 260} 261 262/* Free the cache buffers */ 263static void free_cache(io_channel channel, 264 struct unix_private_data *data) 265{ 266 struct unix_cache *cache; 267 int i; 268 269 data->access_time = 0; 270 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 271 cache->block = 0; 272 cache->access_time = 0; 273 cache->dirty = 0; 274 cache->in_use = 0; 275 if (cache->buf) 276 ext2fs_free_mem((void **) &cache->buf); 277 cache->buf = 0; 278 } 279} 280 281/* 282 * Try to find a block in the cache. If the block is not found, and 283 * eldest is a non-zero pointer, then fill in eldest with the cache 284 * entry to that should be reused. 285 */ 286static struct unix_cache *find_cached_block(io_channel channel, 287 struct unix_private_data *data, 288 unsigned long block, 289 struct unix_cache **eldest) 290{ 291 struct unix_cache *cache, *unused_cache, *oldest_cache; 292 int i; 293 294 unused_cache = oldest_cache = 0; 295 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 296 if (!cache->in_use) { 297 if (!unused_cache) 298 unused_cache = cache; 299 continue; 300 } 301 if (cache->block == block) { 302 cache->access_time = ++data->access_time; 303 return cache; 304 } 305 if (!oldest_cache || 306 (cache->access_time < oldest_cache->access_time)) 307 oldest_cache = cache; 308 } 309 if (eldest) 310 *eldest = (unused_cache) ? unused_cache : oldest_cache; 311 return 0; 312} 313 314/* 315 * Reuse a particular cache entry for another block. 316 */ 317static void reuse_cache(io_channel channel, struct unix_private_data *data, 318 struct unix_cache *cache, unsigned long block) 319{ 320 if (cache->dirty && cache->in_use) 321 raw_write_blk(channel, data, cache->block, 1, cache->buf); 322 323 cache->in_use = 1; 324 cache->dirty = 0; 325 cache->block = block; 326 cache->access_time = ++data->access_time; 327} 328 329/* 330 * Flush all of the blocks in the cache 331 */ 332static errcode_t flush_cached_blocks(io_channel channel, 333 struct unix_private_data *data, 334 int invalidate) 335 336{ 337 struct unix_cache *cache; 338 errcode_t retval, retval2; 339 int i; 340 341 retval2 = 0; 342 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 343 if (!cache->in_use) 344 continue; 345 346 if (invalidate) 347 cache->in_use = 0; 348 349 if (!cache->dirty) 350 continue; 351 352 retval = raw_write_blk(channel, data, 353 cache->block, 1, cache->buf); 354 if (retval) 355 retval2 = retval; 356 else 357 cache->dirty = 0; 358 } 359 return retval2; 360} 361 362static errcode_t unix_open(const char *name, int flags, io_channel *channel) 363{ 364 io_channel io = NULL; 365 struct unix_private_data *data = NULL; 366 errcode_t retval; 367 int open_flags; 368 struct stat st; 369#ifdef __linux__ 370 struct utsname ut; 371#endif 372 373 if (name == 0) 374 return EXT2_ET_BAD_DEVICE_NAME; 375 retval = ext2fs_get_mem(sizeof(struct struct_io_channel), 376 (void **) &io); 377 if (retval) 378 return retval; 379 memset(io, 0, sizeof(struct struct_io_channel)); 380 io->magic = EXT2_ET_MAGIC_IO_CHANNEL; 381 retval = ext2fs_get_mem(sizeof(struct unix_private_data), 382 (void **) &data); 383 if (retval) 384 goto cleanup; 385 386 io->manager = unix_io_manager; 387 retval = ext2fs_get_mem(strlen(name)+1, (void **) &io->name); 388 if (retval) 389 goto cleanup; 390 391 strcpy(io->name, name); 392 io->private_data = data; 393 io->block_size = 1024; 394 io->read_error = 0; 395 io->write_error = 0; 396 io->refcount = 1; 397 398 memset(data, 0, sizeof(struct unix_private_data)); 399 data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL; 400 401 if ((retval = alloc_cache(io, data))) 402 goto cleanup; 403 404 open_flags = (flags & IO_FLAG_RW) ? O_RDWR : O_RDONLY; 405#ifdef HAVE_OPEN64 406 data->dev = open64(name, open_flags); 407#else 408 data->dev = open(name, open_flags); 409#endif 410 if (data->dev < 0) { 411 retval = errno; 412 goto cleanup; 413 } 414 415#ifdef __linux__ 416#undef RLIM_INFINITY 417#if (defined(__alpha__) || ((defined(__sparc__) || defined(__mips__)) && (SIZEOF_LONG == 4))) 418#define RLIM_INFINITY ((unsigned long)(~0UL>>1)) 419#else 420#define RLIM_INFINITY (~0UL) 421#endif 422 /* 423 * Work around a bug in 2.4.10-2.4.18 kernels where writes to 424 * block devices are wrongly getting hit by the filesize 425 * limit. This workaround isn't perfect, since it won't work 426 * if glibc wasn't built against 2.2 header files. (Sigh.) 427 * 428 */ 429 if ((flags & IO_FLAG_RW) && 430 (uname(&ut) == 0) && 431 ((ut.release[0] == '2') && (ut.release[1] == '.') && 432 (ut.release[2] == '4') && (ut.release[3] == '.') && 433 (ut.release[4] == '1') && (ut.release[5] >= '0') && 434 (ut.release[5] < '8')) && 435 (fstat(data->dev, &st) == 0) && 436 (S_ISBLK(st.st_mode))) { 437 struct rlimit rlim; 438 439 rlim.rlim_cur = rlim.rlim_max = (unsigned long) RLIM_INFINITY; 440 setrlimit(RLIMIT_FSIZE, &rlim); 441 getrlimit(RLIMIT_FSIZE, &rlim); 442 if (((unsigned long) rlim.rlim_cur) < 443 ((unsigned long) rlim.rlim_max)) { 444 rlim.rlim_cur = rlim.rlim_max; 445 setrlimit(RLIMIT_FSIZE, &rlim); 446 } 447 } 448#endif 449 *channel = io; 450 return 0; 451 452cleanup: 453 if (data) { 454 free_cache(io, data); 455 ext2fs_free_mem((void **) &data); 456 } 457 if (io) 458 ext2fs_free_mem((void **) &io); 459 return retval; 460} 461 462static errcode_t unix_close(io_channel channel) 463{ 464 struct unix_private_data *data; 465 errcode_t retval = 0; 466 467 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 468 data = (struct unix_private_data *) channel->private_data; 469 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 470 471 if (--channel->refcount > 0) 472 return 0; 473 474 retval = flush_cached_blocks(channel, data, 0); 475 476 if (close(data->dev) < 0) 477 retval = errno; 478 free_cache(channel, data); 479 480 ext2fs_free_mem((void **) &channel->private_data); 481 if (channel->name) 482 ext2fs_free_mem((void **) &channel->name); 483 ext2fs_free_mem((void **) &channel); 484 return retval; 485} 486 487static errcode_t unix_set_blksize(io_channel channel, int blksize) 488{ 489 struct unix_private_data *data; 490 errcode_t retval; 491 492 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 493 data = (struct unix_private_data *) channel->private_data; 494 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 495 496 if (channel->block_size != blksize) { 497 if ((retval = flush_cached_blocks(channel, data, 0))) 498 return retval; 499 500 channel->block_size = blksize; 501 free_cache(channel, data); 502 if ((retval = alloc_cache(channel, data))) 503 return retval; 504 } 505 return 0; 506} 507 508 509static errcode_t unix_read_blk(io_channel channel, unsigned long block, 510 int count, void *buf) 511{ 512 struct unix_private_data *data; 513 struct unix_cache *cache, *reuse[READ_DIRECT_SIZE]; 514 errcode_t retval; 515 char *cp; 516 int i, j; 517 518 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 519 data = (struct unix_private_data *) channel->private_data; 520 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 521 522 /* 523 * If we're doing an odd-sized read or a very large read, 524 * flush out the cache and then do a direct read. 525 */ 526 if (count < 0 || count > WRITE_DIRECT_SIZE) { 527 if ((retval = flush_cached_blocks(channel, data, 0))) 528 return retval; 529 return raw_read_blk(channel, data, block, count, buf); 530 } 531 532 cp = buf; 533 while (count > 0) { 534 /* If it's in the cache, use it! */ 535 if ((cache = find_cached_block(channel, data, block, 536 &reuse[0]))) { 537#ifdef DEBUG 538 printf("Using cached block %d\n", block); 539#endif 540 memcpy(cp, cache->buf, channel->block_size); 541 count--; 542 block++; 543 cp += channel->block_size; 544 continue; 545 } 546 /* 547 * Find the number of uncached blocks so we can do a 548 * single read request 549 */ 550 for (i=1; i < count; i++) 551 if (find_cached_block(channel, data, block+i, 552 &reuse[i])) 553 break; 554#ifdef DEBUG 555 printf("Reading %d blocks starting at %d\n", i, block); 556#endif 557 if ((retval = raw_read_blk(channel, data, block, i, cp))) 558 return retval; 559 560 /* Save the results in the cache */ 561 for (j=0; j < i; j++) { 562 count--; 563 cache = reuse[j]; 564 reuse_cache(channel, data, cache, block++); 565 memcpy(cache->buf, cp, channel->block_size); 566 cp += channel->block_size; 567 } 568 } 569 return 0; 570} 571 572static errcode_t unix_write_blk(io_channel channel, unsigned long block, 573 int count, const void *buf) 574{ 575 struct unix_private_data *data; 576 struct unix_cache *cache, *reuse; 577 errcode_t retval = 0; 578 const char *cp; 579 int writethrough; 580 581 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 582 data = (struct unix_private_data *) channel->private_data; 583 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 584 585 /* 586 * If we're doing an odd-sized write or a very large write, 587 * flush out the cache completely and then do a direct write. 588 */ 589 if (count < 0 || count > WRITE_DIRECT_SIZE) { 590 if ((retval = flush_cached_blocks(channel, data, 1))) 591 return retval; 592 return raw_write_blk(channel, data, block, count, buf); 593 } 594 595 /* 596 * For a moderate-sized multi-block write, first force a write 597 * if we're in write-through cache mode, and then fill the 598 * cache with the blocks. 599 */ 600 writethrough = channel->flags & CHANNEL_FLAGS_WRITETHROUGH; 601 if (writethrough) 602 retval = raw_write_blk(channel, data, block, count, buf); 603 604 cp = buf; 605 while (count > 0) { 606 cache = find_cached_block(channel, data, block, &reuse); 607 if (!cache) { 608 cache = reuse; 609 reuse_cache(channel, data, cache, block); 610 } 611 memcpy(cache->buf, cp, channel->block_size); 612 cache->dirty = !writethrough; 613 count--; 614 block++; 615 cp += channel->block_size; 616 } 617 return retval; 618} 619 620static errcode_t unix_write_byte(io_channel channel, unsigned long offset, 621 int size, const void *buf) 622{ 623 struct unix_private_data *data; 624 errcode_t retval = 0; 625 size_t actual; 626 627 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 628 data = (struct unix_private_data *) channel->private_data; 629 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 630 631 /* 632 * Flush out the cache completely 633 */ 634 if ((retval = flush_cached_blocks(channel, data, 1))) 635 return retval; 636 637 if (lseek(data->dev, offset, SEEK_SET) < 0) 638 return errno; 639 640 actual = write(data->dev, buf, size); 641 if (actual != size) 642 return EXT2_ET_SHORT_WRITE; 643 644 return 0; 645} 646 647/* 648 * Flush data buffers to disk. 649 */ 650static errcode_t unix_flush(io_channel channel) 651{ 652 struct unix_private_data *data; 653 errcode_t retval = 0; 654 655 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 656 data = (struct unix_private_data *) channel->private_data; 657 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 658 659 retval = flush_cached_blocks(channel, data, 0); 660 fsync(data->dev); 661 return retval; 662} 663 664