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 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
54 struct unix_cache {
55 char *buf;
56 unsigned long block;
57 int access_time;
58 unsigned dirty:1;
59 unsigned 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
66 struct unix_private_data {
67 int magic;
68 int dev;
69 int flags;
70 int access_time;
71 ext2_loff_t offset;
72 struct unix_cache cache[CACHE_SIZE];
73 };
74
75 static errcode_t unix_open(const char *name, int flags, io_channel *channel);
76 static errcode_t unix_close(io_channel channel);
77 static errcode_t unix_set_blksize(io_channel channel, int blksize);
78 static errcode_t unix_read_blk(io_channel channel, unsigned long block,
79 int count, void *data);
80 static errcode_t unix_write_blk(io_channel channel, unsigned long block,
81 int count, const void *data);
82 static errcode_t unix_flush(io_channel channel);
83 static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
84 int size, const void *data);
85 static errcode_t unix_set_option(io_channel channel, const char *option,
86 const char *arg);
87
88 static void reuse_cache(io_channel channel, struct unix_private_data *data,
89 struct unix_cache *cache, unsigned long block);
90
91 /* __FreeBSD_kernel__ is defined by GNU/kFreeBSD - the FreeBSD kernel
92 * does not know buffered block devices - everything is raw. */
93 #if defined(__CYGWIN__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
94 #define NEED_BOUNCE_BUFFER
95 #else
96 #undef NEED_BOUNCE_BUFFER
97 #endif
98
99 static struct struct_io_manager struct_unix_manager = {
100 EXT2_ET_MAGIC_IO_MANAGER,
101 "Unix I/O Manager",
102 unix_open,
103 unix_close,
104 unix_set_blksize,
105 unix_read_blk,
106 unix_write_blk,
107 unix_flush,
108 #ifdef NEED_BOUNCE_BUFFER
109 0,
110 #else
111 unix_write_byte,
112 #endif
113 unix_set_option
114 };
115
116 io_manager unix_io_manager = &struct_unix_manager;
117
118 /*
119 * Here are the raw I/O functions
120 */
121 #ifndef NEED_BOUNCE_BUFFER
raw_read_blk(io_channel channel,struct unix_private_data * data,unsigned long block,int count,void * buf)122 static errcode_t raw_read_blk(io_channel channel,
123 struct unix_private_data *data,
124 unsigned long block,
125 int count, void *buf)
126 {
127 errcode_t retval;
128 ssize_t size;
129 ext2_loff_t location;
130 int actual = 0;
131
132 size = (count < 0) ? -count : count * channel->block_size;
133 location = ((ext2_loff_t) block * channel->block_size) + data->offset;
134 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) {
135 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
136 goto error_out;
137 }
138 actual = read(data->dev, buf, size);
139 if (actual != size) {
140 if (actual < 0)
141 actual = 0;
142 retval = EXT2_ET_SHORT_READ;
143 goto error_out;
144 }
145 return 0;
146
147 error_out:
148 memset((char *) buf+actual, 0, size-actual);
149 if (channel->read_error)
150 retval = (channel->read_error)(channel, block, count, buf,
151 size, actual, retval);
152 return retval;
153 }
154 #else /* NEED_BOUNCE_BUFFER */
155 /*
156 * Windows and FreeBSD block devices only allow sector alignment IO in offset and size
157 */
raw_read_blk(io_channel channel,struct unix_private_data * data,unsigned long block,int count,void * buf)158 static errcode_t raw_read_blk(io_channel channel,
159 struct unix_private_data *data,
160 unsigned long block,
161 int count, void *buf)
162 {
163 errcode_t retval;
164 size_t size, alignsize, fragment;
165 ext2_loff_t location;
166 int total = 0, actual;
167 #define BLOCKALIGN 512
168 char sector[BLOCKALIGN];
169
170 size = (count < 0) ? -count : count * channel->block_size;
171 location = ((ext2_loff_t) block * channel->block_size) + data->offset;
172 #ifdef DEBUG
173 printf("count=%d, size=%d, block=%lu, blk_size=%d, location=%llx\n",
174 count, size, block, channel->block_size, (long long)location);
175 #endif
176 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) {
177 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
178 goto error_out;
179 }
180 fragment = size % BLOCKALIGN;
181 alignsize = size - fragment;
182 if (alignsize) {
183 actual = read(data->dev, buf, alignsize);
184 if (actual != alignsize)
185 goto short_read;
186 }
187 if (fragment) {
188 actual = read(data->dev, sector, BLOCKALIGN);
189 if (actual != BLOCKALIGN)
190 goto short_read;
191 memcpy(buf+alignsize, sector, fragment);
192 }
193 return 0;
194
195 short_read:
196 if (actual>0)
197 total += actual;
198 retval = EXT2_ET_SHORT_READ;
199
200 error_out:
201 memset((char *) buf+total, 0, size-actual);
202 if (channel->read_error)
203 retval = (channel->read_error)(channel, block, count, buf,
204 size, actual, retval);
205 return retval;
206 }
207 #endif
208
raw_write_blk(io_channel channel,struct unix_private_data * data,unsigned long block,int count,const void * buf)209 static errcode_t raw_write_blk(io_channel channel,
210 struct unix_private_data *data,
211 unsigned long block,
212 int count, const void *buf)
213 {
214 ssize_t size;
215 ext2_loff_t location;
216 int actual = 0;
217 errcode_t retval;
218
219 if (count == 1)
220 size = channel->block_size;
221 else {
222 if (count < 0)
223 size = -count;
224 else
225 size = count * channel->block_size;
226 }
227
228 location = ((ext2_loff_t) block * channel->block_size) + data->offset;
229 if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) {
230 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
231 goto error_out;
232 }
233
234 actual = write(data->dev, buf, size);
235 if (actual != size) {
236 retval = EXT2_ET_SHORT_WRITE;
237 goto error_out;
238 }
239 return 0;
240
241 error_out:
242 if (channel->write_error)
243 retval = (channel->write_error)(channel, block, count, buf,
244 size, actual, retval);
245 return retval;
246 }
247
248
249 /*
250 * Here we implement the cache functions
251 */
252
253 /* Allocate the cache buffers */
alloc_cache(io_channel channel,struct unix_private_data * data)254 static errcode_t alloc_cache(io_channel channel,
255 struct unix_private_data *data)
256 {
257 errcode_t retval;
258 struct unix_cache *cache;
259 int i;
260
261 data->access_time = 0;
262 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
263 cache->block = 0;
264 cache->access_time = 0;
265 cache->dirty = 0;
266 cache->in_use = 0;
267 if ((retval = ext2fs_get_mem(channel->block_size,
268 &cache->buf)))
269 return retval;
270 }
271 return 0;
272 }
273
274 /* Free the cache buffers */
free_cache(struct unix_private_data * data)275 static void free_cache(struct unix_private_data *data)
276 {
277 struct unix_cache *cache;
278 int i;
279
280 data->access_time = 0;
281 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
282 cache->block = 0;
283 cache->access_time = 0;
284 cache->dirty = 0;
285 cache->in_use = 0;
286 if (cache->buf)
287 ext2fs_free_mem(&cache->buf);
288 cache->buf = 0;
289 }
290 }
291
292 #ifndef NO_IO_CACHE
293 /*
294 * Try to find a block in the cache. If the block is not found, and
295 * eldest is a non-zero pointer, then fill in eldest with the cache
296 * entry to that should be reused.
297 */
find_cached_block(struct unix_private_data * data,unsigned long block,struct unix_cache ** eldest)298 static struct unix_cache *find_cached_block(struct unix_private_data *data,
299 unsigned long block,
300 struct unix_cache **eldest)
301 {
302 struct unix_cache *cache, *unused_cache, *oldest_cache;
303 int i;
304
305 unused_cache = oldest_cache = 0;
306 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
307 if (!cache->in_use) {
308 if (!unused_cache)
309 unused_cache = cache;
310 continue;
311 }
312 if (cache->block == block) {
313 cache->access_time = ++data->access_time;
314 return cache;
315 }
316 if (!oldest_cache ||
317 (cache->access_time < oldest_cache->access_time))
318 oldest_cache = cache;
319 }
320 if (eldest)
321 *eldest = (unused_cache) ? unused_cache : oldest_cache;
322 return 0;
323 }
324
325 /*
326 * Reuse a particular cache entry for another block.
327 */
reuse_cache(io_channel channel,struct unix_private_data * data,struct unix_cache * cache,unsigned long block)328 static void reuse_cache(io_channel channel, struct unix_private_data *data,
329 struct unix_cache *cache, unsigned long block)
330 {
331 if (cache->dirty && cache->in_use)
332 raw_write_blk(channel, data, cache->block, 1, cache->buf);
333
334 cache->in_use = 1;
335 cache->dirty = 0;
336 cache->block = block;
337 cache->access_time = ++data->access_time;
338 }
339
340 /*
341 * Flush all of the blocks in the cache
342 */
flush_cached_blocks(io_channel channel,struct unix_private_data * data,int invalidate)343 static errcode_t flush_cached_blocks(io_channel channel,
344 struct unix_private_data *data,
345 int invalidate)
346
347 {
348 struct unix_cache *cache;
349 errcode_t retval, retval2;
350 int i;
351
352 retval2 = 0;
353 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
354 if (!cache->in_use)
355 continue;
356
357 if (invalidate)
358 cache->in_use = 0;
359
360 if (!cache->dirty)
361 continue;
362
363 retval = raw_write_blk(channel, data,
364 cache->block, 1, cache->buf);
365 if (retval)
366 retval2 = retval;
367 else
368 cache->dirty = 0;
369 }
370 return retval2;
371 }
372 #endif /* NO_IO_CACHE */
373
unix_open(const char * name,int flags,io_channel * channel)374 static errcode_t unix_open(const char *name, int flags, io_channel *channel)
375 {
376 io_channel io = NULL;
377 struct unix_private_data *data = NULL;
378 errcode_t retval;
379 int open_flags;
380 struct stat st;
381 #ifdef __linux__
382 struct utsname ut;
383 #endif
384
385 if (name == 0)
386 return EXT2_ET_BAD_DEVICE_NAME;
387 retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io);
388 if (retval)
389 return retval;
390 memset(io, 0, sizeof(struct struct_io_channel));
391 io->magic = EXT2_ET_MAGIC_IO_CHANNEL;
392 retval = ext2fs_get_mem(sizeof(struct unix_private_data), &data);
393 if (retval)
394 goto cleanup;
395
396 io->manager = unix_io_manager;
397 retval = ext2fs_get_mem(strlen(name)+1, &io->name);
398 if (retval)
399 goto cleanup;
400
401 strcpy(io->name, name);
402 io->private_data = data;
403 io->block_size = 1024;
404 io->read_error = 0;
405 io->write_error = 0;
406 io->refcount = 1;
407
408 memset(data, 0, sizeof(struct unix_private_data));
409 data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL;
410
411 if ((retval = alloc_cache(io, data)))
412 goto cleanup;
413
414 open_flags = (flags & IO_FLAG_RW) ? O_RDWR : O_RDONLY;
415 if (flags & IO_FLAG_EXCLUSIVE)
416 open_flags |= O_EXCL;
417 #ifdef HAVE_OPEN64
418 data->dev = open64(io->name, open_flags);
419 #else
420 data->dev = open(io->name, open_flags);
421 #endif
422 if (data->dev < 0) {
423 retval = errno;
424 goto cleanup;
425 }
426
427 #ifdef __linux__
428 #undef RLIM_INFINITY
429 #if (defined(__alpha__) || ((defined(__sparc__) || defined(__mips__)) && (SIZEOF_LONG == 4)))
430 #define RLIM_INFINITY ((unsigned long)(~0UL>>1))
431 #else
432 #define RLIM_INFINITY (~0UL)
433 #endif
434 /*
435 * Work around a bug in 2.4.10-2.4.18 kernels where writes to
436 * block devices are wrongly getting hit by the filesize
437 * limit. This workaround isn't perfect, since it won't work
438 * if glibc wasn't built against 2.2 header files. (Sigh.)
439 *
440 */
441 if ((flags & IO_FLAG_RW) &&
442 (uname(&ut) == 0) &&
443 ((ut.release[0] == '2') && (ut.release[1] == '.') &&
444 (ut.release[2] == '4') && (ut.release[3] == '.') &&
445 (ut.release[4] == '1') && (ut.release[5] >= '0') &&
446 (ut.release[5] < '8')) &&
447 (fstat(data->dev, &st) == 0) &&
448 (S_ISBLK(st.st_mode))) {
449 struct rlimit rlim;
450
451 rlim.rlim_cur = rlim.rlim_max = (unsigned long) RLIM_INFINITY;
452 setrlimit(RLIMIT_FSIZE, &rlim);
453 getrlimit(RLIMIT_FSIZE, &rlim);
454 if (((unsigned long) rlim.rlim_cur) <
455 ((unsigned long) rlim.rlim_max)) {
456 rlim.rlim_cur = rlim.rlim_max;
457 setrlimit(RLIMIT_FSIZE, &rlim);
458 }
459 }
460 #endif
461 *channel = io;
462 return 0;
463
464 cleanup:
465 if (data) {
466 free_cache(data);
467 ext2fs_free_mem(&data);
468 }
469 if (io)
470 ext2fs_free_mem(&io);
471 return retval;
472 }
473
unix_close(io_channel channel)474 static errcode_t unix_close(io_channel channel)
475 {
476 struct unix_private_data *data;
477 errcode_t retval = 0;
478
479 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
480 data = (struct unix_private_data *) channel->private_data;
481 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
482
483 if (--channel->refcount > 0)
484 return 0;
485
486 #ifndef NO_IO_CACHE
487 retval = flush_cached_blocks(channel, data, 0);
488 #endif
489
490 if (close(data->dev) < 0)
491 retval = errno;
492 free_cache(data);
493
494 ext2fs_free_mem(&channel->private_data);
495 if (channel->name)
496 ext2fs_free_mem(&channel->name);
497 ext2fs_free_mem(&channel);
498 return retval;
499 }
500
unix_set_blksize(io_channel channel,int blksize)501 static errcode_t unix_set_blksize(io_channel channel, int blksize)
502 {
503 struct unix_private_data *data;
504 errcode_t retval;
505
506 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
507 data = (struct unix_private_data *) channel->private_data;
508 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
509
510 if (channel->block_size != blksize) {
511 #ifndef NO_IO_CACHE
512 if ((retval = flush_cached_blocks(channel, data, 0)))
513 return retval;
514 #endif
515
516 channel->block_size = blksize;
517 free_cache(data);
518 if ((retval = alloc_cache(channel, data)))
519 return retval;
520 }
521 return 0;
522 }
523
524
unix_read_blk(io_channel channel,unsigned long block,int count,void * buf)525 static errcode_t unix_read_blk(io_channel channel, unsigned long block,
526 int count, void *buf)
527 {
528 struct unix_private_data *data;
529 struct unix_cache *cache, *reuse[READ_DIRECT_SIZE];
530 errcode_t retval;
531 char *cp;
532 int i, j;
533
534 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
535 data = (struct unix_private_data *) channel->private_data;
536 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
537
538 #ifdef NO_IO_CACHE
539 return raw_read_blk(channel, data, block, count, buf);
540 #else
541 /*
542 * If we're doing an odd-sized read or a very large read,
543 * flush out the cache and then do a direct read.
544 */
545 if (count < 0 || count > WRITE_DIRECT_SIZE) {
546 if ((retval = flush_cached_blocks(channel, data, 0)))
547 return retval;
548 return raw_read_blk(channel, data, block, count, buf);
549 }
550
551 cp = buf;
552 while (count > 0) {
553 /* If it's in the cache, use it! */
554 if ((cache = find_cached_block(data, block, &reuse[0]))) {
555 #ifdef DEBUG
556 printf("Using cached block %lu\n", block);
557 #endif
558 memcpy(cp, cache->buf, channel->block_size);
559 count--;
560 block++;
561 cp += channel->block_size;
562 continue;
563 }
564 /*
565 * Find the number of uncached blocks so we can do a
566 * single read request
567 */
568 for (i=1; i < count; i++)
569 if (find_cached_block(data, block+i, &reuse[i]))
570 break;
571 #ifdef DEBUG
572 printf("Reading %d blocks starting at %lu\n", i, block);
573 #endif
574 if ((retval = raw_read_blk(channel, data, block, i, cp)))
575 return retval;
576
577 /* Save the results in the cache */
578 for (j=0; j < i; j++) {
579 count--;
580 cache = reuse[j];
581 reuse_cache(channel, data, cache, block++);
582 memcpy(cache->buf, cp, channel->block_size);
583 cp += channel->block_size;
584 }
585 }
586 return 0;
587 #endif /* NO_IO_CACHE */
588 }
589
unix_write_blk(io_channel channel,unsigned long block,int count,const void * buf)590 static errcode_t unix_write_blk(io_channel channel, unsigned long block,
591 int count, const void *buf)
592 {
593 struct unix_private_data *data;
594 struct unix_cache *cache, *reuse;
595 errcode_t retval = 0;
596 const char *cp;
597 int writethrough;
598
599 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
600 data = (struct unix_private_data *) channel->private_data;
601 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
602
603 #ifdef NO_IO_CACHE
604 return raw_write_blk(channel, data, block, count, buf);
605 #else
606 /*
607 * If we're doing an odd-sized write or a very large write,
608 * flush out the cache completely and then do a direct write.
609 */
610 if (count < 0 || count > WRITE_DIRECT_SIZE) {
611 if ((retval = flush_cached_blocks(channel, data, 1)))
612 return retval;
613 return raw_write_blk(channel, data, block, count, buf);
614 }
615
616 /*
617 * For a moderate-sized multi-block write, first force a write
618 * if we're in write-through cache mode, and then fill the
619 * cache with the blocks.
620 */
621 writethrough = channel->flags & CHANNEL_FLAGS_WRITETHROUGH;
622 if (writethrough)
623 retval = raw_write_blk(channel, data, block, count, buf);
624
625 cp = buf;
626 while (count > 0) {
627 cache = find_cached_block(data, block, &reuse);
628 if (!cache) {
629 cache = reuse;
630 reuse_cache(channel, data, cache, block);
631 }
632 memcpy(cache->buf, cp, channel->block_size);
633 cache->dirty = !writethrough;
634 count--;
635 block++;
636 cp += channel->block_size;
637 }
638 return retval;
639 #endif /* NO_IO_CACHE */
640 }
641
unix_write_byte(io_channel channel,unsigned long offset,int size,const void * buf)642 static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
643 int size, const void *buf)
644 {
645 struct unix_private_data *data;
646 errcode_t retval = 0;
647 ssize_t actual;
648
649 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
650 data = (struct unix_private_data *) channel->private_data;
651 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
652
653 #ifndef NO_IO_CACHE
654 /*
655 * Flush out the cache completely
656 */
657 if ((retval = flush_cached_blocks(channel, data, 1)))
658 return retval;
659 #endif
660
661 if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0)
662 return errno;
663
664 actual = write(data->dev, buf, size);
665 if (actual != size)
666 return EXT2_ET_SHORT_WRITE;
667
668 return 0;
669 }
670
671 /*
672 * Flush data buffers to disk.
673 */
unix_flush(io_channel channel)674 static errcode_t unix_flush(io_channel channel)
675 {
676 struct unix_private_data *data;
677 errcode_t retval = 0;
678
679 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
680 data = (struct unix_private_data *) channel->private_data;
681 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
682
683 #ifndef NO_IO_CACHE
684 retval = flush_cached_blocks(channel, data, 0);
685 #endif
686 fsync(data->dev);
687 return retval;
688 }
689
unix_set_option(io_channel channel,const char * option,const char * arg)690 static errcode_t unix_set_option(io_channel channel, const char *option,
691 const char *arg)
692 {
693 struct unix_private_data *data;
694 unsigned long long tmp;
695 char *end;
696
697 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
698 data = (struct unix_private_data *) channel->private_data;
699 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
700
701 if (!strcmp(option, "offset")) {
702 if (!arg)
703 return EXT2_ET_INVALID_ARGUMENT;
704
705 tmp = strtoull(arg, &end, 0);
706 if (*end)
707 return EXT2_ET_INVALID_ARGUMENT;
708 data->offset = tmp;
709 if (data->offset < 0)
710 return EXT2_ET_INVALID_ARGUMENT;
711 return 0;
712 }
713 return EXT2_ET_INVALID_ARGUMENT;
714 }
715