unix_io.c revision dd0a2679ddd0a9bf53e32efc0f67a7e7a5ea5f00
1/* 2 * unix_io.c --- This is the Unix (well, really POSIX) implementation 3 * of the I/O manager. 4 * 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 Library 14 * General Public License, version 2. 15 * %End-Header% 16 */ 17 18#define _LARGEFILE_SOURCE 19#define _LARGEFILE64_SOURCE 20#ifndef _GNU_SOURCE 21#define _GNU_SOURCE 22#endif 23 24#include "config.h" 25#include <stdio.h> 26#include <string.h> 27#if HAVE_UNISTD_H 28#include <unistd.h> 29#endif 30#if HAVE_ERRNO_H 31#include <errno.h> 32#endif 33#include <fcntl.h> 34#include <time.h> 35#ifdef __linux__ 36#include <sys/utsname.h> 37#endif 38#ifdef HAVE_SYS_IOCTL_H 39#include <sys/ioctl.h> 40#endif 41#ifdef HAVE_SYS_MOUNT_H 42#include <sys/mount.h> 43#endif 44#if HAVE_SYS_STAT_H 45#include <sys/stat.h> 46#endif 47#if HAVE_SYS_TYPES_H 48#include <sys/types.h> 49#endif 50#if HAVE_SYS_RESOURCE_H 51#include <sys/resource.h> 52#endif 53#if HAVE_LINUX_FALLOC_H 54#include <linux/falloc.h> 55#endif 56 57#if defined(__linux__) && defined(_IO) && !defined(BLKROGET) 58#define BLKROGET _IO(0x12, 94) /* Get read-only status (0 = read_write). */ 59#endif 60 61#undef ALIGN_DEBUG 62 63#include "ext2_fs.h" 64#include "ext2fs.h" 65 66/* 67 * For checking structure magic numbers... 68 */ 69 70#define EXT2_CHECK_MAGIC(struct, code) \ 71 if ((struct)->magic != (code)) return (code) 72 73struct unix_cache { 74 char *buf; 75 unsigned long block; 76 int access_time; 77 unsigned dirty:1; 78 unsigned in_use:1; 79}; 80 81#define CACHE_SIZE 8 82#define WRITE_DIRECT_SIZE 4 /* Must be smaller than CACHE_SIZE */ 83#define READ_DIRECT_SIZE 4 /* Should be smaller than CACHE_SIZE */ 84 85struct unix_private_data { 86 int magic; 87 int dev; 88 int flags; 89 int align; 90 int access_time; 91 ext2_loff_t offset; 92 struct unix_cache cache[CACHE_SIZE]; 93 void *bounce; 94 struct struct_io_stats io_stats; 95}; 96 97#define IS_ALIGNED(n, align) ((((unsigned long) n) & \ 98 ((unsigned long) ((align)-1))) == 0) 99 100static errcode_t unix_open(const char *name, int flags, io_channel *channel); 101static errcode_t unix_close(io_channel channel); 102static errcode_t unix_set_blksize(io_channel channel, int blksize); 103static errcode_t unix_read_blk(io_channel channel, unsigned long block, 104 int count, void *data); 105static errcode_t unix_write_blk(io_channel channel, unsigned long block, 106 int count, const void *data); 107static errcode_t unix_flush(io_channel channel); 108static errcode_t unix_write_byte(io_channel channel, unsigned long offset, 109 int size, const void *data); 110static errcode_t unix_set_option(io_channel channel, const char *option, 111 const char *arg); 112static errcode_t unix_get_stats(io_channel channel, io_stats *stats) 113; 114static void reuse_cache(io_channel channel, struct unix_private_data *data, 115 struct unix_cache *cache, unsigned long long block); 116static errcode_t unix_read_blk64(io_channel channel, unsigned long long block, 117 int count, void *data); 118static errcode_t unix_write_blk64(io_channel channel, unsigned long long block, 119 int count, const void *data); 120static errcode_t unix_discard(io_channel channel, unsigned long long block, 121 unsigned long long count); 122 123static struct struct_io_manager struct_unix_manager = { 124 EXT2_ET_MAGIC_IO_MANAGER, 125 "Unix I/O Manager", 126 unix_open, 127 unix_close, 128 unix_set_blksize, 129 unix_read_blk, 130 unix_write_blk, 131 unix_flush, 132 unix_write_byte, 133 unix_set_option, 134 unix_get_stats, 135 unix_read_blk64, 136 unix_write_blk64, 137 unix_discard, 138}; 139 140io_manager unix_io_manager = &struct_unix_manager; 141 142static errcode_t unix_get_stats(io_channel channel, io_stats *stats) 143{ 144 errcode_t retval = 0; 145 146 struct unix_private_data *data; 147 148 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 149 data = (struct unix_private_data *) channel->private_data; 150 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 151 152 if (stats) 153 *stats = &data->io_stats; 154 155 return retval; 156} 157 158/* 159 * Here are the raw I/O functions 160 */ 161static errcode_t raw_read_blk(io_channel channel, 162 struct unix_private_data *data, 163 unsigned long long block, 164 int count, void *bufv) 165{ 166 errcode_t retval; 167 ssize_t size; 168 ext2_loff_t location; 169 int actual = 0; 170 unsigned char *buf = bufv; 171 172 size = (count < 0) ? -count : count * channel->block_size; 173 data->io_stats.bytes_read += size; 174 location = ((ext2_loff_t) block * channel->block_size) + data->offset; 175 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { 176 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; 177 goto error_out; 178 } 179 if ((channel->align == 0) || 180 (IS_ALIGNED(buf, channel->align) && 181 IS_ALIGNED(size, channel->align))) { 182 actual = read(data->dev, buf, size); 183 if (actual != size) { 184 short_read: 185 if (actual < 0) 186 actual = 0; 187 retval = EXT2_ET_SHORT_READ; 188 goto error_out; 189 } 190 return 0; 191 } 192 193#ifdef ALIGN_DEBUG 194 printf("raw_read_blk: O_DIRECT fallback: %p %lu\n", buf, 195 (unsigned long) size); 196#endif 197 198 /* 199 * The buffer or size which we're trying to read isn't aligned 200 * to the O_DIRECT rules, so we need to do this the hard way... 201 */ 202 while (size > 0) { 203 actual = read(data->dev, data->bounce, channel->block_size); 204 if (actual != channel->block_size) 205 goto short_read; 206 actual = size; 207 if (size > channel->block_size) 208 actual = channel->block_size; 209 memcpy(buf, data->bounce, actual); 210 size -= actual; 211 buf += actual; 212 } 213 return 0; 214 215error_out: 216 memset((char *) buf+actual, 0, size-actual); 217 if (channel->read_error) 218 retval = (channel->read_error)(channel, block, count, buf, 219 size, actual, retval); 220 return retval; 221} 222 223static errcode_t raw_write_blk(io_channel channel, 224 struct unix_private_data *data, 225 unsigned long long block, 226 int count, const void *bufv) 227{ 228 ssize_t size; 229 ext2_loff_t location; 230 int actual = 0; 231 errcode_t retval; 232 const unsigned char *buf = bufv; 233 234 if (count == 1) 235 size = channel->block_size; 236 else { 237 if (count < 0) 238 size = -count; 239 else 240 size = count * channel->block_size; 241 } 242 data->io_stats.bytes_written += size; 243 244 location = ((ext2_loff_t) block * channel->block_size) + data->offset; 245 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { 246 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; 247 goto error_out; 248 } 249 250 if ((channel->align == 0) || 251 (IS_ALIGNED(buf, channel->align) && 252 IS_ALIGNED(size, channel->align))) { 253 actual = write(data->dev, buf, size); 254 if (actual != size) { 255 short_write: 256 retval = EXT2_ET_SHORT_WRITE; 257 goto error_out; 258 } 259 return 0; 260 } 261 262#ifdef ALIGN_DEBUG 263 printf("raw_write_blk: O_DIRECT fallback: %p %lu\n", buf, 264 (unsigned long) size); 265#endif 266 /* 267 * The buffer or size which we're trying to write isn't aligned 268 * to the O_DIRECT rules, so we need to do this the hard way... 269 */ 270 while (size > 0) { 271 if (size < channel->block_size) { 272 actual = read(data->dev, data->bounce, 273 channel->block_size); 274 if (actual != channel->block_size) { 275 retval = EXT2_ET_SHORT_READ; 276 goto error_out; 277 } 278 } 279 actual = size; 280 if (size > channel->block_size) 281 actual = channel->block_size; 282 memcpy(data->bounce, buf, actual); 283 actual = write(data->dev, data->bounce, channel->block_size); 284 if (actual != channel->block_size) 285 goto short_write; 286 size -= actual; 287 buf += actual; 288 } 289 return 0; 290 291error_out: 292 if (channel->write_error) 293 retval = (channel->write_error)(channel, block, count, buf, 294 size, actual, retval); 295 return retval; 296} 297 298 299/* 300 * Here we implement the cache functions 301 */ 302 303/* Allocate the cache buffers */ 304static errcode_t alloc_cache(io_channel channel, 305 struct unix_private_data *data) 306{ 307 errcode_t retval; 308 struct unix_cache *cache; 309 int i; 310 311 data->access_time = 0; 312 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 313 cache->block = 0; 314 cache->access_time = 0; 315 cache->dirty = 0; 316 cache->in_use = 0; 317 if (cache->buf) 318 ext2fs_free_mem(&cache->buf); 319 retval = ext2fs_get_memalign(channel->block_size, 320 channel->align, &cache->buf); 321 if (retval) 322 return retval; 323 } 324 if (channel->align) { 325 if (data->bounce) 326 ext2fs_free_mem(&data->bounce); 327 retval = ext2fs_get_memalign(channel->block_size, 328 channel->align, 329 &data->bounce); 330 } 331 return retval; 332} 333 334/* Free the cache buffers */ 335static void free_cache(struct unix_private_data *data) 336{ 337 struct unix_cache *cache; 338 int i; 339 340 data->access_time = 0; 341 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 342 cache->block = 0; 343 cache->access_time = 0; 344 cache->dirty = 0; 345 cache->in_use = 0; 346 if (cache->buf) 347 ext2fs_free_mem(&cache->buf); 348 } 349 if (data->bounce) 350 ext2fs_free_mem(&data->bounce); 351} 352 353#ifndef NO_IO_CACHE 354/* 355 * Try to find a block in the cache. If the block is not found, and 356 * eldest is a non-zero pointer, then fill in eldest with the cache 357 * entry to that should be reused. 358 */ 359static struct unix_cache *find_cached_block(struct unix_private_data *data, 360 unsigned long long block, 361 struct unix_cache **eldest) 362{ 363 struct unix_cache *cache, *unused_cache, *oldest_cache; 364 int i; 365 366 unused_cache = oldest_cache = 0; 367 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 368 if (!cache->in_use) { 369 if (!unused_cache) 370 unused_cache = cache; 371 continue; 372 } 373 if (cache->block == block) { 374 cache->access_time = ++data->access_time; 375 return cache; 376 } 377 if (!oldest_cache || 378 (cache->access_time < oldest_cache->access_time)) 379 oldest_cache = cache; 380 } 381 if (eldest) 382 *eldest = (unused_cache) ? unused_cache : oldest_cache; 383 return 0; 384} 385 386/* 387 * Reuse a particular cache entry for another block. 388 */ 389static void reuse_cache(io_channel channel, struct unix_private_data *data, 390 struct unix_cache *cache, unsigned long long block) 391{ 392 if (cache->dirty && cache->in_use) 393 raw_write_blk(channel, data, cache->block, 1, cache->buf); 394 395 cache->in_use = 1; 396 cache->dirty = 0; 397 cache->block = block; 398 cache->access_time = ++data->access_time; 399} 400 401/* 402 * Flush all of the blocks in the cache 403 */ 404static errcode_t flush_cached_blocks(io_channel channel, 405 struct unix_private_data *data, 406 int invalidate) 407 408{ 409 struct unix_cache *cache; 410 errcode_t retval, retval2; 411 int i; 412 413 retval2 = 0; 414 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { 415 if (!cache->in_use) 416 continue; 417 418 if (invalidate) 419 cache->in_use = 0; 420 421 if (!cache->dirty) 422 continue; 423 424 retval = raw_write_blk(channel, data, 425 cache->block, 1, cache->buf); 426 if (retval) 427 retval2 = retval; 428 else 429 cache->dirty = 0; 430 } 431 return retval2; 432} 433#endif /* NO_IO_CACHE */ 434 435#ifdef __linux__ 436#ifndef BLKDISCARDZEROES 437#define BLKDISCARDZEROES _IO(0x12,124) 438#endif 439#endif 440 441static errcode_t unix_open(const char *name, int flags, io_channel *channel) 442{ 443 io_channel io = NULL; 444 struct unix_private_data *data = NULL; 445 errcode_t retval; 446 int open_flags, zeroes = 0; 447 int f_nocache = 0; 448 ext2fs_struct_stat st; 449#ifdef __linux__ 450 struct utsname ut; 451#endif 452 453 if (name == 0) 454 return EXT2_ET_BAD_DEVICE_NAME; 455 retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io); 456 if (retval) 457 goto cleanup; 458 memset(io, 0, sizeof(struct struct_io_channel)); 459 io->magic = EXT2_ET_MAGIC_IO_CHANNEL; 460 retval = ext2fs_get_mem(sizeof(struct unix_private_data), &data); 461 if (retval) 462 goto cleanup; 463 464 io->manager = unix_io_manager; 465 retval = ext2fs_get_mem(strlen(name)+1, &io->name); 466 if (retval) 467 goto cleanup; 468 469 strcpy(io->name, name); 470 io->private_data = data; 471 io->block_size = 1024; 472 io->read_error = 0; 473 io->write_error = 0; 474 io->refcount = 1; 475 476 memset(data, 0, sizeof(struct unix_private_data)); 477 data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL; 478 data->io_stats.num_fields = 2; 479 480 open_flags = (flags & IO_FLAG_RW) ? O_RDWR : O_RDONLY; 481 if (flags & IO_FLAG_EXCLUSIVE) 482 open_flags |= O_EXCL; 483#if defined(O_DIRECT) 484 if (flags & IO_FLAG_DIRECT_IO) { 485 open_flags |= O_DIRECT; 486 io->align = ext2fs_get_dio_alignment(data->dev); 487 } 488#elif defined(F_NOCACHE) 489 if (flags & IO_FLAG_DIRECT_IO) { 490 f_nocache = F_NOCACHE; 491 io->align = 4096; 492 } 493#endif 494 data->flags = flags; 495 496 data->dev = ext2fs_open_file(io->name, open_flags, 0); 497 if (data->dev < 0) { 498 retval = errno; 499 goto cleanup; 500 } 501 if (f_nocache) { 502 if (fcntl(data->dev, f_nocache, 1) < 0) { 503 retval = errno; 504 goto cleanup; 505 } 506 } 507 508 /* 509 * If the device is really a block device, then set the 510 * appropriate flag, otherwise we can set DISCARD_ZEROES flag 511 * because we are going to use punch hole instead of discard 512 * and if it succeed, subsequent read from sparse area returns 513 * zero. 514 */ 515 if (ext2fs_stat(io->name, &st) == 0) { 516 if (S_ISBLK(st.st_mode)) 517 io->flags |= CHANNEL_FLAGS_BLOCK_DEVICE; 518 else 519 io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES; 520 } 521 522#ifdef BLKDISCARDZEROES 523 ioctl(data->dev, BLKDISCARDZEROES, &zeroes); 524 if (zeroes) 525 io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES; 526#endif 527 528#if defined(__CYGWIN__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) 529 /* 530 * Some operating systems require that the buffers be aligned, 531 * regardless of O_DIRECT 532 */ 533 if (!io->align) 534 io->align = 512; 535#endif 536 537 538 if ((retval = alloc_cache(io, data))) 539 goto cleanup; 540 541#ifdef BLKROGET 542 if (flags & IO_FLAG_RW) { 543 int error; 544 int readonly = 0; 545 546 /* Is the block device actually writable? */ 547 error = ioctl(data->dev, BLKROGET, &readonly); 548 if (!error && readonly) { 549 close(data->dev); 550 retval = EPERM; 551 goto cleanup; 552 } 553 } 554#endif 555 556#ifdef __linux__ 557#undef RLIM_INFINITY 558#if (defined(__alpha__) || ((defined(__sparc__) || defined(__mips__)) && (SIZEOF_LONG == 4))) 559#define RLIM_INFINITY ((unsigned long)(~0UL>>1)) 560#else 561#define RLIM_INFINITY (~0UL) 562#endif 563 /* 564 * Work around a bug in 2.4.10-2.4.18 kernels where writes to 565 * block devices are wrongly getting hit by the filesize 566 * limit. This workaround isn't perfect, since it won't work 567 * if glibc wasn't built against 2.2 header files. (Sigh.) 568 * 569 */ 570 if ((flags & IO_FLAG_RW) && 571 (uname(&ut) == 0) && 572 ((ut.release[0] == '2') && (ut.release[1] == '.') && 573 (ut.release[2] == '4') && (ut.release[3] == '.') && 574 (ut.release[4] == '1') && (ut.release[5] >= '0') && 575 (ut.release[5] < '8')) && 576 (ext2fs_stat(io->name, &st) == 0) && 577 (S_ISBLK(st.st_mode))) { 578 struct rlimit rlim; 579 580 rlim.rlim_cur = rlim.rlim_max = (unsigned long) RLIM_INFINITY; 581 setrlimit(RLIMIT_FSIZE, &rlim); 582 getrlimit(RLIMIT_FSIZE, &rlim); 583 if (((unsigned long) rlim.rlim_cur) < 584 ((unsigned long) rlim.rlim_max)) { 585 rlim.rlim_cur = rlim.rlim_max; 586 setrlimit(RLIMIT_FSIZE, &rlim); 587 } 588 } 589#endif 590 *channel = io; 591 return 0; 592 593cleanup: 594 if (data) { 595 free_cache(data); 596 ext2fs_free_mem(&data); 597 } 598 if (io) 599 ext2fs_free_mem(&io); 600 return retval; 601} 602 603static errcode_t unix_close(io_channel channel) 604{ 605 struct unix_private_data *data; 606 errcode_t retval = 0; 607 608 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 609 data = (struct unix_private_data *) channel->private_data; 610 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 611 612 if (--channel->refcount > 0) 613 return 0; 614 615#ifndef NO_IO_CACHE 616 retval = flush_cached_blocks(channel, data, 0); 617#endif 618 619 if (close(data->dev) < 0) 620 retval = errno; 621 free_cache(data); 622 623 ext2fs_free_mem(&channel->private_data); 624 if (channel->name) 625 ext2fs_free_mem(&channel->name); 626 ext2fs_free_mem(&channel); 627 return retval; 628} 629 630static errcode_t unix_set_blksize(io_channel channel, int blksize) 631{ 632 struct unix_private_data *data; 633 errcode_t retval; 634 635 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 636 data = (struct unix_private_data *) channel->private_data; 637 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 638 639 if (channel->block_size != blksize) { 640#ifndef NO_IO_CACHE 641 if ((retval = flush_cached_blocks(channel, data, 0))) 642 return retval; 643#endif 644 645 channel->block_size = blksize; 646 free_cache(data); 647 if ((retval = alloc_cache(channel, data))) 648 return retval; 649 } 650 return 0; 651} 652 653 654static errcode_t unix_read_blk64(io_channel channel, unsigned long long block, 655 int count, void *buf) 656{ 657 struct unix_private_data *data; 658 struct unix_cache *cache, *reuse[READ_DIRECT_SIZE]; 659 errcode_t retval; 660 char *cp; 661 int i, j; 662 663 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 664 data = (struct unix_private_data *) channel->private_data; 665 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 666 667#ifdef NO_IO_CACHE 668 return raw_read_blk(channel, data, block, count, buf); 669#else 670 /* 671 * If we're doing an odd-sized read or a very large read, 672 * flush out the cache and then do a direct read. 673 */ 674 if (count < 0 || count > WRITE_DIRECT_SIZE) { 675 if ((retval = flush_cached_blocks(channel, data, 0))) 676 return retval; 677 return raw_read_blk(channel, data, block, count, buf); 678 } 679 680 cp = buf; 681 while (count > 0) { 682 /* If it's in the cache, use it! */ 683 if ((cache = find_cached_block(data, block, &reuse[0]))) { 684#ifdef DEBUG 685 printf("Using cached block %lu\n", block); 686#endif 687 memcpy(cp, cache->buf, channel->block_size); 688 count--; 689 block++; 690 cp += channel->block_size; 691 continue; 692 } 693 if (count == 1) { 694 /* 695 * Special case where we read directly into the 696 * cache buffer; important in the O_DIRECT case 697 */ 698 cache = reuse[0]; 699 reuse_cache(channel, data, cache, block); 700 if ((retval = raw_read_blk(channel, data, block, 1, 701 cache->buf))) { 702 cache->in_use = 0; 703 return retval; 704 } 705 memcpy(cp, cache->buf, channel->block_size); 706 return 0; 707 } 708 709 /* 710 * Find the number of uncached blocks so we can do a 711 * single read request 712 */ 713 for (i=1; i < count; i++) 714 if (find_cached_block(data, block+i, &reuse[i])) 715 break; 716#ifdef DEBUG 717 printf("Reading %d blocks starting at %lu\n", i, block); 718#endif 719 if ((retval = raw_read_blk(channel, data, block, i, cp))) 720 return retval; 721 722 /* Save the results in the cache */ 723 for (j=0; j < i; j++) { 724 count--; 725 cache = reuse[j]; 726 reuse_cache(channel, data, cache, block++); 727 memcpy(cache->buf, cp, channel->block_size); 728 cp += channel->block_size; 729 } 730 } 731 return 0; 732#endif /* NO_IO_CACHE */ 733} 734 735static errcode_t unix_read_blk(io_channel channel, unsigned long block, 736 int count, void *buf) 737{ 738 return unix_read_blk64(channel, block, count, buf); 739} 740 741static errcode_t unix_write_blk64(io_channel channel, unsigned long long block, 742 int count, const void *buf) 743{ 744 struct unix_private_data *data; 745 struct unix_cache *cache, *reuse; 746 errcode_t retval = 0; 747 const char *cp; 748 int writethrough; 749 750 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 751 data = (struct unix_private_data *) channel->private_data; 752 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 753 754#ifdef NO_IO_CACHE 755 return raw_write_blk(channel, data, block, count, buf); 756#else 757 /* 758 * If we're doing an odd-sized write or a very large write, 759 * flush out the cache completely and then do a direct write. 760 */ 761 if (count < 0 || count > WRITE_DIRECT_SIZE) { 762 if ((retval = flush_cached_blocks(channel, data, 1))) 763 return retval; 764 return raw_write_blk(channel, data, block, count, buf); 765 } 766 767 /* 768 * For a moderate-sized multi-block write, first force a write 769 * if we're in write-through cache mode, and then fill the 770 * cache with the blocks. 771 */ 772 writethrough = channel->flags & CHANNEL_FLAGS_WRITETHROUGH; 773 if (writethrough) 774 retval = raw_write_blk(channel, data, block, count, buf); 775 776 cp = buf; 777 while (count > 0) { 778 cache = find_cached_block(data, block, &reuse); 779 if (!cache) { 780 cache = reuse; 781 reuse_cache(channel, data, cache, block); 782 } 783 memcpy(cache->buf, cp, channel->block_size); 784 cache->dirty = !writethrough; 785 count--; 786 block++; 787 cp += channel->block_size; 788 } 789 return retval; 790#endif /* NO_IO_CACHE */ 791} 792 793static errcode_t unix_write_blk(io_channel channel, unsigned long block, 794 int count, const void *buf) 795{ 796 return unix_write_blk64(channel, block, count, buf); 797} 798 799static errcode_t unix_write_byte(io_channel channel, unsigned long offset, 800 int size, const void *buf) 801{ 802 struct unix_private_data *data; 803 errcode_t retval = 0; 804 ssize_t actual; 805 806 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 807 data = (struct unix_private_data *) channel->private_data; 808 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 809 810 if (channel->align != 0) { 811#ifdef ALIGN_DEBUG 812 printf("unix_write_byte: O_DIRECT fallback\n"); 813#endif 814 return EXT2_ET_UNIMPLEMENTED; 815 } 816 817#ifndef NO_IO_CACHE 818 /* 819 * Flush out the cache completely 820 */ 821 if ((retval = flush_cached_blocks(channel, data, 1))) 822 return retval; 823#endif 824 825 if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0) 826 return errno; 827 828 actual = write(data->dev, buf, size); 829 if (actual != size) 830 return EXT2_ET_SHORT_WRITE; 831 832 return 0; 833} 834 835/* 836 * Flush data buffers to disk. 837 */ 838static errcode_t unix_flush(io_channel channel) 839{ 840 struct unix_private_data *data; 841 errcode_t retval = 0; 842 843 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 844 data = (struct unix_private_data *) channel->private_data; 845 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 846 847#ifndef NO_IO_CACHE 848 retval = flush_cached_blocks(channel, data, 0); 849#endif 850 fsync(data->dev); 851 return retval; 852} 853 854static errcode_t unix_set_option(io_channel channel, const char *option, 855 const char *arg) 856{ 857 struct unix_private_data *data; 858 unsigned long long tmp; 859 char *end; 860 861 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 862 data = (struct unix_private_data *) channel->private_data; 863 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 864 865 if (!strcmp(option, "offset")) { 866 if (!arg) 867 return EXT2_ET_INVALID_ARGUMENT; 868 869 tmp = strtoull(arg, &end, 0); 870 if (*end) 871 return EXT2_ET_INVALID_ARGUMENT; 872 data->offset = tmp; 873 if (data->offset < 0) 874 return EXT2_ET_INVALID_ARGUMENT; 875 return 0; 876 } 877 return EXT2_ET_INVALID_ARGUMENT; 878} 879 880#if defined(__linux__) && !defined(BLKDISCARD) 881#define BLKDISCARD _IO(0x12,119) 882#endif 883 884static errcode_t unix_discard(io_channel channel, unsigned long long block, 885 unsigned long long count) 886{ 887 struct unix_private_data *data; 888 __uint64_t range[2]; 889 int ret; 890 891 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); 892 data = (struct unix_private_data *) channel->private_data; 893 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); 894 895 if (channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE) { 896#ifdef BLKDISCARD 897 range[0] = (__uint64_t)(block) * channel->block_size; 898 range[1] = (__uint64_t)(count) * channel->block_size; 899 900 ret = ioctl(data->dev, BLKDISCARD, &range); 901#else 902 goto unimplemented; 903#endif 904 } else { 905#if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_PUNCH_HOLE) 906 /* 907 * If we are not on block device, try to use punch hole 908 * to reclaim free space. 909 */ 910 ret = fallocate(data->dev, 911 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, 912 (off_t)(block) * channel->block_size, 913 (off_t)(count) * channel->block_size); 914#else 915 goto unimplemented; 916#endif 917 } 918 if (ret < 0) { 919 if (errno == EOPNOTSUPP) 920 goto unimplemented; 921 return errno; 922 } 923 return 0; 924unimplemented: 925 return EXT2_ET_UNIMPLEMENTED; 926} 927