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 #if !defined(__FreeBSD__) && !defined(__NetBSD__) && !defined(__OpenBSD__)
19 #define _XOPEN_SOURCE 600
20 #define _DARWIN_C_SOURCE
21 #define _FILE_OFFSET_BITS 64
22 #ifndef _LARGEFILE_SOURCE
23 #define _LARGEFILE_SOURCE
24 #endif
25 #ifndef _LARGEFILE64_SOURCE
26 #define _LARGEFILE64_SOURCE
27 #endif
28 #ifndef _GNU_SOURCE
29 #define _GNU_SOURCE
30 #endif
31 #endif
32
33 #include "config.h"
34 #include <stdio.h>
35 #include <string.h>
36 #if HAVE_UNISTD_H
37 #include <unistd.h>
38 #endif
39 #if HAVE_ERRNO_H
40 #include <errno.h>
41 #endif
42 #include <fcntl.h>
43 #include <time.h>
44 #ifdef __linux__
45 #include <sys/utsname.h>
46 #endif
47 #if HAVE_SYS_TYPES_H
48 #include <sys/types.h>
49 #endif
50 #ifdef HAVE_SYS_IOCTL_H
51 #include <sys/ioctl.h>
52 #endif
53 #ifdef HAVE_SYS_MOUNT_H
54 #include <sys/mount.h>
55 #endif
56 #ifdef HAVE_SYS_PRCTL_H
57 #include <sys/prctl.h>
58 #else
59 #define PR_GET_DUMPABLE 3
60 #endif
61 #if HAVE_SYS_STAT_H
62 #include <sys/stat.h>
63 #endif
64 #if HAVE_SYS_RESOURCE_H
65 #include <sys/resource.h>
66 #endif
67 #if HAVE_LINUX_FALLOC_H
68 #include <linux/falloc.h>
69 #endif
70 #ifdef HAVE_PTHREAD
71 #include <pthread.h>
72 #endif
73
74 #if defined(__linux__) && defined(_IO) && !defined(BLKROGET)
75 #define BLKROGET _IO(0x12, 94) /* Get read-only status (0 = read_write). */
76 #endif
77
78 #undef ALIGN_DEBUG
79
80 #include "ext2_fs.h"
81 #include "ext2fs.h"
82 #include "ext2fsP.h"
83
84 /*
85 * For checking structure magic numbers...
86 */
87
88 #define EXT2_CHECK_MAGIC(struct, code) \
89 if ((struct)->magic != (code)) return (code)
90
91 struct unix_cache {
92 char *buf;
93 unsigned long long block;
94 int access_time;
95 unsigned dirty:1;
96 unsigned in_use:1;
97 unsigned write_err:1;
98 };
99
100 #define CACHE_SIZE 8
101 #define WRITE_DIRECT_SIZE 4 /* Must be smaller than CACHE_SIZE */
102 #define READ_DIRECT_SIZE 4 /* Should be smaller than CACHE_SIZE */
103
104 struct unix_private_data {
105 int magic;
106 int dev;
107 int flags;
108 int align;
109 int access_time;
110 ext2_loff_t offset;
111 struct unix_cache cache[CACHE_SIZE];
112 void *bounce;
113 struct struct_io_stats io_stats;
114 #ifdef HAVE_PTHREAD
115 pthread_mutex_t cache_mutex;
116 pthread_mutex_t bounce_mutex;
117 pthread_mutex_t stats_mutex;
118 #endif
119 };
120
121 #define IS_ALIGNED(n, align) ((((uintptr_t) n) & \
122 ((uintptr_t) ((align)-1))) == 0)
123
124 typedef enum lock_kind {
125 CACHE_MTX, BOUNCE_MTX, STATS_MTX
126 } kind_t;
127
128 #ifdef HAVE_PTHREAD
get_mutex(struct unix_private_data * data,kind_t kind)129 static inline pthread_mutex_t *get_mutex(struct unix_private_data *data,
130 kind_t kind)
131 {
132 if (data->flags & IO_FLAG_THREADS) {
133 switch (kind) {
134 case CACHE_MTX:
135 return &data->cache_mutex;
136 case BOUNCE_MTX:
137 return &data->bounce_mutex;
138 case STATS_MTX:
139 return &data->stats_mutex;
140 }
141 }
142 return NULL;
143 }
144 #endif
145
mutex_lock(struct unix_private_data * data,kind_t kind)146 static inline void mutex_lock(struct unix_private_data *data, kind_t kind)
147 {
148 #ifdef HAVE_PTHREAD
149 pthread_mutex_t *mtx = get_mutex(data,kind);
150
151 if (mtx)
152 pthread_mutex_lock(mtx);
153 #endif
154 }
155
mutex_unlock(struct unix_private_data * data,kind_t kind)156 static inline void mutex_unlock(struct unix_private_data *data, kind_t kind)
157 {
158 #ifdef HAVE_PTHREAD
159 pthread_mutex_t *mtx = get_mutex(data,kind);
160
161 if (mtx)
162 pthread_mutex_unlock(mtx);
163 #endif
164 }
165
unix_get_stats(io_channel channel,io_stats * stats)166 static errcode_t unix_get_stats(io_channel channel, io_stats *stats)
167 {
168 errcode_t retval = 0;
169
170 struct unix_private_data *data;
171
172 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
173 data = (struct unix_private_data *) channel->private_data;
174 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
175
176 if (stats) {
177 mutex_lock(data, STATS_MTX);
178 *stats = &data->io_stats;
179 mutex_unlock(data, STATS_MTX);
180 }
181
182 return retval;
183 }
184
safe_getenv(const char * arg)185 static char *safe_getenv(const char *arg)
186 {
187 if ((getuid() != geteuid()) || (getgid() != getegid()))
188 return NULL;
189 #ifdef HAVE_PRCTL
190 if (prctl(PR_GET_DUMPABLE, 0, 0, 0, 0) == 0)
191 return NULL;
192 #else
193 #if (defined(linux) && defined(SYS_prctl))
194 if (syscall(SYS_prctl, PR_GET_DUMPABLE, 0, 0, 0, 0) == 0)
195 return NULL;
196 #endif
197 #endif
198
199 #if defined(HAVE_SECURE_GETENV)
200 return secure_getenv(arg);
201 #elif defined(HAVE___SECURE_GETENV)
202 return __secure_getenv(arg);
203 #else
204 return getenv(arg);
205 #endif
206 }
207
208 /*
209 * Here are the raw I/O functions
210 */
raw_read_blk(io_channel channel,struct unix_private_data * data,unsigned long long block,int count,void * bufv)211 static errcode_t raw_read_blk(io_channel channel,
212 struct unix_private_data *data,
213 unsigned long long block,
214 int count, void *bufv)
215 {
216 errcode_t retval;
217 ssize_t size;
218 ext2_loff_t location;
219 int actual = 0;
220 unsigned char *buf = bufv;
221 ssize_t really_read = 0;
222 unsigned long long aligned_blk;
223 int align_size, offset;
224
225 size = (count < 0) ? -count : (ext2_loff_t) count * channel->block_size;
226 mutex_lock(data, STATS_MTX);
227 data->io_stats.bytes_read += size;
228 mutex_unlock(data, STATS_MTX);
229 location = ((ext2_loff_t) block * channel->block_size) + data->offset;
230
231 if (data->flags & IO_FLAG_FORCE_BOUNCE)
232 goto bounce_read;
233
234 #ifdef HAVE_PREAD64
235 /* Try an aligned pread */
236 if ((channel->align == 0) ||
237 (IS_ALIGNED(buf, channel->align) &&
238 IS_ALIGNED(location, channel->align) &&
239 IS_ALIGNED(size, channel->align))) {
240 actual = pread64(data->dev, buf, size, location);
241 if (actual == size)
242 return 0;
243 actual = 0;
244 }
245 #elif HAVE_PREAD
246 /* Try an aligned pread */
247 if ((sizeof(off_t) >= sizeof(ext2_loff_t)) &&
248 ((channel->align == 0) ||
249 (IS_ALIGNED(buf, channel->align) &&
250 IS_ALIGNED(location, channel->align) &&
251 IS_ALIGNED(size, channel->align)))) {
252 actual = pread(data->dev, buf, size, location);
253 if (actual == size)
254 return 0;
255 actual = 0;
256 }
257 #endif /* HAVE_PREAD */
258
259 if ((channel->align == 0) ||
260 (IS_ALIGNED(buf, channel->align) &&
261 IS_ALIGNED(location, channel->align) &&
262 IS_ALIGNED(size, channel->align))) {
263 mutex_lock(data, BOUNCE_MTX);
264 if (ext2fs_llseek(data->dev, location, SEEK_SET) < 0) {
265 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
266 goto error_unlock;
267 }
268 actual = read(data->dev, buf, size);
269 if (actual != size) {
270 short_read:
271 if (actual < 0) {
272 retval = errno;
273 actual = 0;
274 } else
275 retval = EXT2_ET_SHORT_READ;
276 goto error_unlock;
277 }
278 goto success_unlock;
279 }
280
281 #ifdef ALIGN_DEBUG
282 printf("raw_read_blk: O_DIRECT fallback: %p %lu\n", buf,
283 (unsigned long) size);
284 #endif
285
286 /*
287 * The buffer or size which we're trying to read isn't aligned
288 * to the O_DIRECT rules, so we need to do this the hard way...
289 */
290 bounce_read:
291 if (channel->align == 0)
292 channel->align = 1;
293 if ((channel->block_size > channel->align) &&
294 (channel->block_size % channel->align) == 0)
295 align_size = channel->block_size;
296 else
297 align_size = channel->align;
298 aligned_blk = location / align_size;
299 offset = location % align_size;
300
301 mutex_lock(data, BOUNCE_MTX);
302 if (ext2fs_llseek(data->dev, aligned_blk * align_size, SEEK_SET) < 0) {
303 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
304 goto error_unlock;
305 }
306 while (size > 0) {
307 actual = read(data->dev, data->bounce, align_size);
308 if (actual != align_size) {
309 mutex_unlock(data, BOUNCE_MTX);
310 actual = really_read;
311 buf -= really_read;
312 size += really_read;
313 goto short_read;
314 }
315 if ((actual + offset) > align_size)
316 actual = align_size - offset;
317 if (actual > size)
318 actual = size;
319 memcpy(buf, (char *)data->bounce + offset, actual);
320
321 really_read += actual;
322 size -= actual;
323 buf += actual;
324 offset = 0;
325 aligned_blk++;
326 }
327 success_unlock:
328 mutex_unlock(data, BOUNCE_MTX);
329 return 0;
330
331 error_unlock:
332 mutex_unlock(data, BOUNCE_MTX);
333 if (actual >= 0 && actual < size)
334 memset((char *) buf+actual, 0, size-actual);
335 if (channel->read_error)
336 retval = (channel->read_error)(channel, block, count, buf,
337 size, actual, retval);
338 return retval;
339 }
340
341 #define RAW_WRITE_NO_HANDLER 1
342
raw_write_blk(io_channel channel,struct unix_private_data * data,unsigned long long block,int count,const void * bufv,int flags)343 static errcode_t raw_write_blk(io_channel channel,
344 struct unix_private_data *data,
345 unsigned long long block,
346 int count, const void *bufv,
347 int flags)
348 {
349 ssize_t size;
350 ext2_loff_t location;
351 int actual = 0;
352 errcode_t retval;
353 const unsigned char *buf = bufv;
354 unsigned long long aligned_blk;
355 int align_size, offset;
356
357 if (count == 1)
358 size = channel->block_size;
359 else {
360 if (count < 0)
361 size = -count;
362 else
363 size = (ext2_loff_t) count * channel->block_size;
364 }
365 mutex_lock(data, STATS_MTX);
366 data->io_stats.bytes_written += size;
367 mutex_unlock(data, STATS_MTX);
368
369 location = ((ext2_loff_t) block * channel->block_size) + data->offset;
370
371 if (data->flags & IO_FLAG_FORCE_BOUNCE)
372 goto bounce_write;
373
374 #ifdef HAVE_PWRITE64
375 /* Try an aligned pwrite */
376 if ((channel->align == 0) ||
377 (IS_ALIGNED(buf, channel->align) &&
378 IS_ALIGNED(location, channel->align) &&
379 IS_ALIGNED(size, channel->align))) {
380 actual = pwrite64(data->dev, buf, size, location);
381 if (actual == size)
382 return 0;
383 }
384 #elif HAVE_PWRITE
385 /* Try an aligned pwrite */
386 if ((sizeof(off_t) >= sizeof(ext2_loff_t)) &&
387 ((channel->align == 0) ||
388 (IS_ALIGNED(buf, channel->align) &&
389 IS_ALIGNED(location, channel->align) &&
390 IS_ALIGNED(size, channel->align)))) {
391 actual = pwrite(data->dev, buf, size, location);
392 if (actual == size)
393 return 0;
394 }
395 #endif /* HAVE_PWRITE */
396
397 if ((channel->align == 0) ||
398 (IS_ALIGNED(buf, channel->align) &&
399 IS_ALIGNED(location, channel->align) &&
400 IS_ALIGNED(size, channel->align))) {
401 mutex_lock(data, BOUNCE_MTX);
402 if (ext2fs_llseek(data->dev, location, SEEK_SET) < 0) {
403 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
404 goto error_unlock;
405 }
406 actual = write(data->dev, buf, size);
407 mutex_unlock(data, BOUNCE_MTX);
408 if (actual < 0) {
409 retval = errno;
410 goto error_out;
411 }
412 if (actual != size) {
413 short_write:
414 retval = EXT2_ET_SHORT_WRITE;
415 goto error_out;
416 }
417 return 0;
418 }
419
420 #ifdef ALIGN_DEBUG
421 printf("raw_write_blk: O_DIRECT fallback: %p %lu\n", buf,
422 (unsigned long) size);
423 #endif
424 /*
425 * The buffer or size which we're trying to write isn't aligned
426 * to the O_DIRECT rules, so we need to do this the hard way...
427 */
428 bounce_write:
429 if (channel->align == 0)
430 channel->align = 1;
431 if ((channel->block_size > channel->align) &&
432 (channel->block_size % channel->align) == 0)
433 align_size = channel->block_size;
434 else
435 align_size = channel->align;
436 aligned_blk = location / align_size;
437 offset = location % align_size;
438
439 while (size > 0) {
440 int actual_w;
441
442 mutex_lock(data, BOUNCE_MTX);
443 if (size < align_size || offset) {
444 if (ext2fs_llseek(data->dev, aligned_blk * align_size,
445 SEEK_SET) < 0) {
446 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
447 goto error_unlock;
448 }
449 actual = read(data->dev, data->bounce,
450 align_size);
451 if (actual != align_size) {
452 if (actual < 0) {
453 retval = errno;
454 goto error_unlock;
455 }
456 memset((char *) data->bounce + actual, 0,
457 align_size - actual);
458 }
459 }
460 actual = size;
461 if ((actual + offset) > align_size)
462 actual = align_size - offset;
463 if (actual > size)
464 actual = size;
465 memcpy(((char *)data->bounce) + offset, buf, actual);
466 if (ext2fs_llseek(data->dev, aligned_blk * align_size, SEEK_SET) < 0) {
467 retval = errno ? errno : EXT2_ET_LLSEEK_FAILED;
468 goto error_unlock;
469 }
470 actual_w = write(data->dev, data->bounce, align_size);
471 mutex_unlock(data, BOUNCE_MTX);
472 if (actual_w < 0) {
473 retval = errno;
474 goto error_out;
475 }
476 if (actual_w != align_size)
477 goto short_write;
478 size -= actual;
479 buf += actual;
480 location += actual;
481 aligned_blk++;
482 offset = 0;
483 }
484 return 0;
485
486 error_unlock:
487 mutex_unlock(data, BOUNCE_MTX);
488 error_out:
489 if (((flags & RAW_WRITE_NO_HANDLER) == 0) && channel->write_error)
490 retval = (channel->write_error)(channel, block, count, buf,
491 size, actual, retval);
492 return retval;
493 }
494
495
496 /*
497 * Here we implement the cache functions
498 */
499
500 /* Allocate the cache buffers */
alloc_cache(io_channel channel,struct unix_private_data * data)501 static errcode_t alloc_cache(io_channel channel,
502 struct unix_private_data *data)
503 {
504 errcode_t retval;
505 struct unix_cache *cache;
506 int i;
507
508 data->access_time = 0;
509 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
510 cache->block = 0;
511 cache->access_time = 0;
512 cache->dirty = 0;
513 cache->in_use = 0;
514 if (cache->buf)
515 ext2fs_free_mem(&cache->buf);
516 retval = io_channel_alloc_buf(channel, 0, &cache->buf);
517 if (retval)
518 return retval;
519 }
520 if (channel->align || data->flags & IO_FLAG_FORCE_BOUNCE) {
521 if (data->bounce)
522 ext2fs_free_mem(&data->bounce);
523 retval = io_channel_alloc_buf(channel, 0, &data->bounce);
524 }
525 return retval;
526 }
527
528 /* Free the cache buffers */
free_cache(struct unix_private_data * data)529 static void free_cache(struct unix_private_data *data)
530 {
531 struct unix_cache *cache;
532 int i;
533
534 data->access_time = 0;
535 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
536 cache->block = 0;
537 cache->access_time = 0;
538 cache->dirty = 0;
539 cache->in_use = 0;
540 if (cache->buf)
541 ext2fs_free_mem(&cache->buf);
542 }
543 if (data->bounce)
544 ext2fs_free_mem(&data->bounce);
545 }
546
547 #ifndef NO_IO_CACHE
548 /*
549 * Try to find a block in the cache. If the block is not found, and
550 * eldest is a non-zero pointer, then fill in eldest with the cache
551 * entry to that should be reused.
552 */
find_cached_block(struct unix_private_data * data,unsigned long long block,struct unix_cache ** eldest)553 static struct unix_cache *find_cached_block(struct unix_private_data *data,
554 unsigned long long block,
555 struct unix_cache **eldest)
556 {
557 struct unix_cache *cache, *unused_cache, *oldest_cache;
558 int i;
559
560 unused_cache = oldest_cache = 0;
561 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
562 if (!cache->in_use) {
563 if (!unused_cache)
564 unused_cache = cache;
565 continue;
566 }
567 if (cache->block == block) {
568 cache->access_time = ++data->access_time;
569 return cache;
570 }
571 if (!oldest_cache ||
572 (cache->access_time < oldest_cache->access_time))
573 oldest_cache = cache;
574 }
575 if (eldest)
576 *eldest = (unused_cache) ? unused_cache : oldest_cache;
577 return 0;
578 }
579
580 /*
581 * Reuse a particular cache entry for another block.
582 */
reuse_cache(io_channel channel,struct unix_private_data * data,struct unix_cache * cache,unsigned long long block)583 static errcode_t reuse_cache(io_channel channel,
584 struct unix_private_data *data, struct unix_cache *cache,
585 unsigned long long block)
586 {
587 if (cache->dirty && cache->in_use) {
588 errcode_t retval;
589
590 retval = raw_write_blk(channel, data, cache->block, 1,
591 cache->buf, RAW_WRITE_NO_HANDLER);
592 if (retval) {
593 cache->write_err = 1;
594 return retval;
595 }
596 }
597
598 cache->in_use = 1;
599 cache->dirty = 0;
600 cache->write_err = 0;
601 cache->block = block;
602 cache->access_time = ++data->access_time;
603 return 0;
604 }
605
606 #define FLUSH_INVALIDATE 0x01
607 #define FLUSH_NOLOCK 0x02
608
609 /*
610 * Flush all of the blocks in the cache
611 */
flush_cached_blocks(io_channel channel,struct unix_private_data * data,int flags)612 static errcode_t flush_cached_blocks(io_channel channel,
613 struct unix_private_data *data,
614 int flags)
615 {
616 struct unix_cache *cache;
617 errcode_t retval, retval2 = 0;
618 int i;
619 int errors_found = 0;
620
621 if ((flags & FLUSH_NOLOCK) == 0)
622 mutex_lock(data, CACHE_MTX);
623 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
624 if (!cache->in_use || !cache->dirty)
625 continue;
626 retval = raw_write_blk(channel, data,
627 cache->block, 1, cache->buf,
628 RAW_WRITE_NO_HANDLER);
629 if (retval) {
630 cache->write_err = 1;
631 errors_found = 1;
632 retval2 = retval;
633 } else {
634 cache->dirty = 0;
635 cache->write_err = 0;
636 if (flags & FLUSH_INVALIDATE)
637 cache->in_use = 0;
638 }
639 }
640 if ((flags & FLUSH_NOLOCK) == 0)
641 mutex_unlock(data, CACHE_MTX);
642 retry:
643 while (errors_found) {
644 if ((flags & FLUSH_NOLOCK) == 0)
645 mutex_lock(data, CACHE_MTX);
646 errors_found = 0;
647 for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) {
648 if (!cache->in_use || !cache->write_err)
649 continue;
650 errors_found = 1;
651 if (cache->write_err && channel->write_error) {
652 char *err_buf = NULL;
653 unsigned long long err_block = cache->block;
654
655 cache->dirty = 0;
656 cache->in_use = 0;
657 cache->write_err = 0;
658 if (io_channel_alloc_buf(channel, 0,
659 &err_buf))
660 err_buf = NULL;
661 else
662 memcpy(err_buf, cache->buf,
663 channel->block_size);
664 mutex_unlock(data, CACHE_MTX);
665 (channel->write_error)(channel, err_block,
666 1, err_buf, channel->block_size, -1,
667 retval2);
668 if (err_buf)
669 ext2fs_free_mem(&err_buf);
670 goto retry;
671 } else
672 cache->write_err = 0;
673 }
674 if ((flags & FLUSH_NOLOCK) == 0)
675 mutex_unlock(data, CACHE_MTX);
676 }
677 return retval2;
678 }
679 #endif /* NO_IO_CACHE */
680
681 #ifdef __linux__
682 #ifndef BLKDISCARDZEROES
683 #define BLKDISCARDZEROES _IO(0x12,124)
684 #endif
685 #endif
686
ext2fs_open_file(const char * pathname,int flags,mode_t mode)687 int ext2fs_open_file(const char *pathname, int flags, mode_t mode)
688 {
689 if (mode)
690 #if defined(HAVE_OPEN64) && !defined(__OSX_AVAILABLE_BUT_DEPRECATED)
691 return open64(pathname, flags, mode);
692 else
693 return open64(pathname, flags);
694 #else
695 return open(pathname, flags, mode);
696 else
697 return open(pathname, flags);
698 #endif
699 }
700
ext2fs_stat(const char * path,ext2fs_struct_stat * buf)701 int ext2fs_stat(const char *path, ext2fs_struct_stat *buf)
702 {
703 #if defined(HAVE_FSTAT64) && !defined(__OSX_AVAILABLE_BUT_DEPRECATED)
704 return stat64(path, buf);
705 #else
706 return stat(path, buf);
707 #endif
708 }
709
ext2fs_fstat(int fd,ext2fs_struct_stat * buf)710 int ext2fs_fstat(int fd, ext2fs_struct_stat *buf)
711 {
712 #if defined(HAVE_FSTAT64) && !defined(__OSX_AVAILABLE_BUT_DEPRECATED)
713 return fstat64(fd, buf);
714 #else
715 return fstat(fd, buf);
716 #endif
717 }
718
719
unix_open_channel(const char * name,int fd,int flags,io_channel * channel,io_manager io_mgr)720 static errcode_t unix_open_channel(const char *name, int fd,
721 int flags, io_channel *channel,
722 io_manager io_mgr)
723 {
724 io_channel io = NULL;
725 struct unix_private_data *data = NULL;
726 errcode_t retval;
727 ext2fs_struct_stat st;
728 #ifdef __linux__
729 struct utsname ut;
730 #endif
731
732 if (safe_getenv("UNIX_IO_FORCE_BOUNCE"))
733 flags |= IO_FLAG_FORCE_BOUNCE;
734
735 #ifdef __linux__
736 /*
737 * We need to make sure any previous errors in the block
738 * device are thrown away, sigh.
739 */
740 (void) fsync(fd);
741 #endif
742
743 retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io);
744 if (retval)
745 goto cleanup;
746 memset(io, 0, sizeof(struct struct_io_channel));
747 io->magic = EXT2_ET_MAGIC_IO_CHANNEL;
748 retval = ext2fs_get_mem(sizeof(struct unix_private_data), &data);
749 if (retval)
750 goto cleanup;
751
752 io->manager = io_mgr;
753 retval = ext2fs_get_mem(strlen(name)+1, &io->name);
754 if (retval)
755 goto cleanup;
756
757 strcpy(io->name, name);
758 io->private_data = data;
759 io->block_size = 1024;
760 io->read_error = 0;
761 io->write_error = 0;
762 io->refcount = 1;
763 io->flags = 0;
764
765 memset(data, 0, sizeof(struct unix_private_data));
766 data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL;
767 data->io_stats.num_fields = 2;
768 data->flags = flags;
769 data->dev = fd;
770
771 #if defined(O_DIRECT)
772 if (flags & IO_FLAG_DIRECT_IO)
773 io->align = ext2fs_get_dio_alignment(data->dev);
774 #elif defined(F_NOCACHE)
775 if (flags & IO_FLAG_DIRECT_IO)
776 io->align = 4096;
777 #endif
778
779 /*
780 * If the device is really a block device, then set the
781 * appropriate flag, otherwise we can set DISCARD_ZEROES flag
782 * because we are going to use punch hole instead of discard
783 * and if it succeed, subsequent read from sparse area returns
784 * zero.
785 */
786 if (ext2fs_fstat(data->dev, &st) == 0) {
787 if (ext2fsP_is_disk_device(st.st_mode))
788 io->flags |= CHANNEL_FLAGS_BLOCK_DEVICE;
789 else
790 io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES;
791 }
792
793 #ifdef BLKDISCARDZEROES
794 {
795 int zeroes = 0;
796 if (ioctl(data->dev, BLKDISCARDZEROES, &zeroes) == 0 &&
797 zeroes)
798 io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES;
799 }
800 #endif
801
802 #if defined(__CYGWIN__)
803 /*
804 * Some operating systems require that the buffers be aligned,
805 * regardless of O_DIRECT
806 */
807 if (!io->align)
808 io->align = 512;
809 #endif
810
811 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
812 if (io->flags & CHANNEL_FLAGS_BLOCK_DEVICE) {
813 int dio_align = ext2fs_get_dio_alignment(fd);
814
815 if (io->align < dio_align)
816 io->align = dio_align;
817 }
818 #endif
819
820 if ((retval = alloc_cache(io, data)))
821 goto cleanup;
822
823 #ifdef BLKROGET
824 if (flags & IO_FLAG_RW) {
825 int error;
826 int readonly = 0;
827
828 /* Is the block device actually writable? */
829 error = ioctl(data->dev, BLKROGET, &readonly);
830 if (!error && readonly) {
831 retval = EPERM;
832 goto cleanup;
833 }
834 }
835 #endif
836
837 #ifdef __linux__
838 #undef RLIM_INFINITY
839 #if (defined(__alpha__) || ((defined(__sparc__) || defined(__mips__)) && (SIZEOF_LONG == 4)))
840 #define RLIM_INFINITY ((unsigned long)(~0UL>>1))
841 #else
842 #define RLIM_INFINITY (~0UL)
843 #endif
844 /*
845 * Work around a bug in 2.4.10-2.4.18 kernels where writes to
846 * block devices are wrongly getting hit by the filesize
847 * limit. This workaround isn't perfect, since it won't work
848 * if glibc wasn't built against 2.2 header files. (Sigh.)
849 *
850 */
851 if ((flags & IO_FLAG_RW) &&
852 (uname(&ut) == 0) &&
853 ((ut.release[0] == '2') && (ut.release[1] == '.') &&
854 (ut.release[2] == '4') && (ut.release[3] == '.') &&
855 (ut.release[4] == '1') && (ut.release[5] >= '0') &&
856 (ut.release[5] < '8')) &&
857 (ext2fs_fstat(data->dev, &st) == 0) &&
858 (ext2fsP_is_disk_device(st.st_mode))) {
859 struct rlimit rlim;
860
861 rlim.rlim_cur = rlim.rlim_max = (unsigned long) RLIM_INFINITY;
862 setrlimit(RLIMIT_FSIZE, &rlim);
863 getrlimit(RLIMIT_FSIZE, &rlim);
864 if (((unsigned long) rlim.rlim_cur) <
865 ((unsigned long) rlim.rlim_max)) {
866 rlim.rlim_cur = rlim.rlim_max;
867 setrlimit(RLIMIT_FSIZE, &rlim);
868 }
869 }
870 #endif
871 #ifdef HAVE_PTHREAD
872 if (flags & IO_FLAG_THREADS) {
873 io->flags |= CHANNEL_FLAGS_THREADS;
874 retval = pthread_mutex_init(&data->cache_mutex, NULL);
875 if (retval)
876 goto cleanup;
877 retval = pthread_mutex_init(&data->bounce_mutex, NULL);
878 if (retval) {
879 pthread_mutex_destroy(&data->cache_mutex);
880 goto cleanup;
881 }
882 retval = pthread_mutex_init(&data->stats_mutex, NULL);
883 if (retval) {
884 pthread_mutex_destroy(&data->cache_mutex);
885 pthread_mutex_destroy(&data->bounce_mutex);
886 goto cleanup;
887 }
888 }
889 #endif
890 *channel = io;
891 return 0;
892
893 cleanup:
894 if (data) {
895 if (data->dev >= 0)
896 close(data->dev);
897 free_cache(data);
898 ext2fs_free_mem(&data);
899 }
900 if (io) {
901 if (io->name) {
902 ext2fs_free_mem(&io->name);
903 }
904 ext2fs_free_mem(&io);
905 }
906 return retval;
907 }
908
unixfd_open(const char * str_fd,int flags,io_channel * channel)909 static errcode_t unixfd_open(const char *str_fd, int flags,
910 io_channel *channel)
911 {
912 int fd;
913 int fd_flags;
914
915 fd = atoi(str_fd);
916 #if defined(HAVE_FCNTL)
917 fd_flags = fcntl(fd, F_GETFD);
918 if (fd_flags == -1)
919 return EBADF;
920
921 flags = 0;
922 if (fd_flags & O_RDWR)
923 flags |= IO_FLAG_RW;
924 if (fd_flags & O_EXCL)
925 flags |= IO_FLAG_EXCLUSIVE;
926 #if defined(O_DIRECT)
927 if (fd_flags & O_DIRECT)
928 flags |= IO_FLAG_DIRECT_IO;
929 #endif
930 #endif /* HAVE_FCNTL */
931
932 return unix_open_channel(str_fd, fd, flags, channel, unixfd_io_manager);
933 }
934
unix_open(const char * name,int flags,io_channel * channel)935 static errcode_t unix_open(const char *name, int flags,
936 io_channel *channel)
937 {
938 int fd = -1;
939 int open_flags;
940
941 if (name == 0)
942 return EXT2_ET_BAD_DEVICE_NAME;
943
944 open_flags = (flags & IO_FLAG_RW) ? O_RDWR : O_RDONLY;
945 if (flags & IO_FLAG_EXCLUSIVE)
946 open_flags |= O_EXCL;
947 #if defined(O_DIRECT)
948 if (flags & IO_FLAG_DIRECT_IO)
949 open_flags |= O_DIRECT;
950 #endif
951 fd = ext2fs_open_file(name, open_flags, 0);
952 if (fd < 0)
953 return errno;
954 #if defined(F_NOCACHE) && !defined(IO_DIRECT)
955 if (flags & IO_FLAG_DIRECT_IO) {
956 if (fcntl(fd, F_NOCACHE, 1) < 0)
957 return errno;
958 }
959 #endif
960 return unix_open_channel(name, fd, flags, channel, unix_io_manager);
961 }
962
unix_close(io_channel channel)963 static errcode_t unix_close(io_channel channel)
964 {
965 struct unix_private_data *data;
966 errcode_t retval = 0;
967
968 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
969 data = (struct unix_private_data *) channel->private_data;
970 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
971
972 if (--channel->refcount > 0)
973 return 0;
974
975 #ifndef NO_IO_CACHE
976 retval = flush_cached_blocks(channel, data, 0);
977 #endif
978
979 if (close(data->dev) < 0)
980 retval = errno;
981 free_cache(data);
982 #ifdef HAVE_PTHREAD
983 if (data->flags & IO_FLAG_THREADS) {
984 pthread_mutex_destroy(&data->cache_mutex);
985 pthread_mutex_destroy(&data->bounce_mutex);
986 pthread_mutex_destroy(&data->stats_mutex);
987 }
988 #endif
989
990 ext2fs_free_mem(&channel->private_data);
991 if (channel->name)
992 ext2fs_free_mem(&channel->name);
993 ext2fs_free_mem(&channel);
994 return retval;
995 }
996
unix_set_blksize(io_channel channel,int blksize)997 static errcode_t unix_set_blksize(io_channel channel, int blksize)
998 {
999 struct unix_private_data *data;
1000 errcode_t retval = 0;
1001
1002 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1003 data = (struct unix_private_data *) channel->private_data;
1004 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1005
1006 if (channel->block_size != blksize) {
1007 mutex_lock(data, CACHE_MTX);
1008 mutex_lock(data, BOUNCE_MTX);
1009 #ifndef NO_IO_CACHE
1010 if ((retval = flush_cached_blocks(channel, data, FLUSH_NOLOCK))){
1011 mutex_unlock(data, BOUNCE_MTX);
1012 mutex_unlock(data, CACHE_MTX);
1013 return retval;
1014 }
1015 #endif
1016
1017 channel->block_size = blksize;
1018 free_cache(data);
1019 retval = alloc_cache(channel, data);
1020 mutex_unlock(data, BOUNCE_MTX);
1021 mutex_unlock(data, CACHE_MTX);
1022 }
1023 return retval;
1024 }
1025
unix_read_blk64(io_channel channel,unsigned long long block,int count,void * buf)1026 static errcode_t unix_read_blk64(io_channel channel, unsigned long long block,
1027 int count, void *buf)
1028 {
1029 struct unix_private_data *data;
1030 struct unix_cache *cache;
1031 errcode_t retval;
1032 char *cp;
1033 int i, j;
1034
1035 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1036 data = (struct unix_private_data *) channel->private_data;
1037 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1038
1039 #ifdef NO_IO_CACHE
1040 return raw_read_blk(channel, data, block, count, buf);
1041 #else
1042 if (data->flags & IO_FLAG_NOCACHE)
1043 return raw_read_blk(channel, data, block, count, buf);
1044 /*
1045 * If we're doing an odd-sized read or a very large read,
1046 * flush out the cache and then do a direct read.
1047 */
1048 if (count < 0 || count > WRITE_DIRECT_SIZE) {
1049 if ((retval = flush_cached_blocks(channel, data, 0)))
1050 return retval;
1051 return raw_read_blk(channel, data, block, count, buf);
1052 }
1053
1054 cp = buf;
1055 mutex_lock(data, CACHE_MTX);
1056 while (count > 0) {
1057 /* If it's in the cache, use it! */
1058 if ((cache = find_cached_block(data, block, NULL))) {
1059 #ifdef DEBUG
1060 printf("Using cached block %lu\n", block);
1061 #endif
1062 memcpy(cp, cache->buf, channel->block_size);
1063 count--;
1064 block++;
1065 cp += channel->block_size;
1066 continue;
1067 }
1068
1069 /*
1070 * Find the number of uncached blocks so we can do a
1071 * single read request
1072 */
1073 for (i=1; i < count; i++)
1074 if (find_cached_block(data, block+i, NULL))
1075 break;
1076 #ifdef DEBUG
1077 printf("Reading %d blocks starting at %lu\n", i, block);
1078 #endif
1079 mutex_unlock(data, CACHE_MTX);
1080 if ((retval = raw_read_blk(channel, data, block, i, cp)))
1081 return retval;
1082 mutex_lock(data, CACHE_MTX);
1083
1084 /* Save the results in the cache */
1085 for (j=0; j < i; j++) {
1086 if (!find_cached_block(data, block, &cache)) {
1087 retval = reuse_cache(channel, data,
1088 cache, block);
1089 if (retval)
1090 goto call_write_handler;
1091 memcpy(cache->buf, cp, channel->block_size);
1092 }
1093 count--;
1094 block++;
1095 cp += channel->block_size;
1096 }
1097 }
1098 mutex_unlock(data, CACHE_MTX);
1099 return 0;
1100
1101 call_write_handler:
1102 if (cache->write_err && channel->write_error) {
1103 char *err_buf = NULL;
1104 unsigned long long err_block = cache->block;
1105
1106 cache->dirty = 0;
1107 cache->in_use = 0;
1108 cache->write_err = 0;
1109 if (io_channel_alloc_buf(channel, 0, &err_buf))
1110 err_buf = NULL;
1111 else
1112 memcpy(err_buf, cache->buf, channel->block_size);
1113 mutex_unlock(data, CACHE_MTX);
1114 (channel->write_error)(channel, err_block, 1, err_buf,
1115 channel->block_size, -1,
1116 retval);
1117 if (err_buf)
1118 ext2fs_free_mem(&err_buf);
1119 } else
1120 mutex_unlock(data, CACHE_MTX);
1121 return retval;
1122 #endif /* NO_IO_CACHE */
1123 }
1124
unix_read_blk(io_channel channel,unsigned long block,int count,void * buf)1125 static errcode_t unix_read_blk(io_channel channel, unsigned long block,
1126 int count, void *buf)
1127 {
1128 return unix_read_blk64(channel, block, count, buf);
1129 }
1130
unix_write_blk64(io_channel channel,unsigned long long block,int count,const void * buf)1131 static errcode_t unix_write_blk64(io_channel channel, unsigned long long block,
1132 int count, const void *buf)
1133 {
1134 struct unix_private_data *data;
1135 struct unix_cache *cache, *reuse;
1136 errcode_t retval = 0;
1137 const char *cp;
1138 int writethrough;
1139
1140 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1141 data = (struct unix_private_data *) channel->private_data;
1142 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1143
1144 #ifdef NO_IO_CACHE
1145 return raw_write_blk(channel, data, block, count, buf, 0);
1146 #else
1147 if (data->flags & IO_FLAG_NOCACHE)
1148 return raw_write_blk(channel, data, block, count, buf, 0);
1149 /*
1150 * If we're doing an odd-sized write or a very large write,
1151 * flush out the cache completely and then do a direct write.
1152 */
1153 if (count < 0 || count > WRITE_DIRECT_SIZE) {
1154 if ((retval = flush_cached_blocks(channel, data,
1155 FLUSH_INVALIDATE)))
1156 return retval;
1157 return raw_write_blk(channel, data, block, count, buf, 0);
1158 }
1159
1160 /*
1161 * For a moderate-sized multi-block write, first force a write
1162 * if we're in write-through cache mode, and then fill the
1163 * cache with the blocks.
1164 */
1165 writethrough = channel->flags & CHANNEL_FLAGS_WRITETHROUGH;
1166 if (writethrough)
1167 retval = raw_write_blk(channel, data, block, count, buf, 0);
1168
1169 cp = buf;
1170 mutex_lock(data, CACHE_MTX);
1171 while (count > 0) {
1172 cache = find_cached_block(data, block, &reuse);
1173 if (!cache) {
1174 errcode_t err;
1175
1176 cache = reuse;
1177 err = reuse_cache(channel, data, cache, block);
1178 if (err)
1179 goto call_write_handler;
1180 }
1181 if (cache->buf != cp)
1182 memcpy(cache->buf, cp, channel->block_size);
1183 cache->dirty = !writethrough;
1184 count--;
1185 block++;
1186 cp += channel->block_size;
1187 }
1188 mutex_unlock(data, CACHE_MTX);
1189 return retval;
1190
1191 call_write_handler:
1192 if (cache->write_err && channel->write_error) {
1193 char *err_buf = NULL;
1194 unsigned long long err_block = cache->block;
1195
1196 cache->dirty = 0;
1197 cache->in_use = 0;
1198 cache->write_err = 0;
1199 if (io_channel_alloc_buf(channel, 0, &err_buf))
1200 err_buf = NULL;
1201 else
1202 memcpy(err_buf, cache->buf, channel->block_size);
1203 mutex_unlock(data, CACHE_MTX);
1204 (channel->write_error)(channel, err_block, 1, err_buf,
1205 channel->block_size, -1,
1206 retval);
1207 if (err_buf)
1208 ext2fs_free_mem(&err_buf);
1209 } else
1210 mutex_unlock(data, CACHE_MTX);
1211 return retval;
1212 #endif /* NO_IO_CACHE */
1213 }
1214
unix_cache_readahead(io_channel channel,unsigned long long block,unsigned long long count)1215 static errcode_t unix_cache_readahead(io_channel channel,
1216 unsigned long long block,
1217 unsigned long long count)
1218 {
1219 #ifdef POSIX_FADV_WILLNEED
1220 struct unix_private_data *data;
1221
1222 data = (struct unix_private_data *)channel->private_data;
1223 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1224 return posix_fadvise(data->dev,
1225 (ext2_loff_t)block * channel->block_size + data->offset,
1226 (ext2_loff_t)count * channel->block_size,
1227 POSIX_FADV_WILLNEED);
1228 #else
1229 return EXT2_ET_OP_NOT_SUPPORTED;
1230 #endif
1231 }
1232
unix_write_blk(io_channel channel,unsigned long block,int count,const void * buf)1233 static errcode_t unix_write_blk(io_channel channel, unsigned long block,
1234 int count, const void *buf)
1235 {
1236 return unix_write_blk64(channel, block, count, buf);
1237 }
1238
unix_write_byte(io_channel channel,unsigned long offset,int size,const void * buf)1239 static errcode_t unix_write_byte(io_channel channel, unsigned long offset,
1240 int size, const void *buf)
1241 {
1242 struct unix_private_data *data;
1243 errcode_t retval = 0;
1244 ssize_t actual;
1245
1246 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1247 data = (struct unix_private_data *) channel->private_data;
1248 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1249
1250 if (channel->align != 0) {
1251 #ifdef ALIGN_DEBUG
1252 printf("unix_write_byte: O_DIRECT fallback\n");
1253 #endif
1254 return EXT2_ET_UNIMPLEMENTED;
1255 }
1256
1257 #ifndef NO_IO_CACHE
1258 /*
1259 * Flush out the cache completely
1260 */
1261 if ((retval = flush_cached_blocks(channel, data, FLUSH_INVALIDATE)))
1262 return retval;
1263 #endif
1264
1265 if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0)
1266 return errno;
1267
1268 actual = write(data->dev, buf, size);
1269 if (actual < 0)
1270 return errno;
1271 if (actual != size)
1272 return EXT2_ET_SHORT_WRITE;
1273
1274 return 0;
1275 }
1276
1277 /*
1278 * Flush data buffers to disk.
1279 */
unix_flush(io_channel channel)1280 static errcode_t unix_flush(io_channel channel)
1281 {
1282 struct unix_private_data *data;
1283 errcode_t retval = 0;
1284
1285 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1286 data = (struct unix_private_data *) channel->private_data;
1287 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1288
1289 #ifndef NO_IO_CACHE
1290 retval = flush_cached_blocks(channel, data, 0);
1291 #endif
1292 #ifdef HAVE_FSYNC
1293 if (!retval && fsync(data->dev) != 0)
1294 return errno;
1295 #endif
1296 return retval;
1297 }
1298
unix_set_option(io_channel channel,const char * option,const char * arg)1299 static errcode_t unix_set_option(io_channel channel, const char *option,
1300 const char *arg)
1301 {
1302 struct unix_private_data *data;
1303 unsigned long long tmp;
1304 errcode_t retval;
1305 char *end;
1306
1307 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1308 data = (struct unix_private_data *) channel->private_data;
1309 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1310
1311 if (!strcmp(option, "offset")) {
1312 if (!arg)
1313 return EXT2_ET_INVALID_ARGUMENT;
1314
1315 tmp = strtoull(arg, &end, 0);
1316 if (*end)
1317 return EXT2_ET_INVALID_ARGUMENT;
1318 data->offset = tmp;
1319 if (data->offset < 0)
1320 return EXT2_ET_INVALID_ARGUMENT;
1321 return 0;
1322 }
1323 if (!strcmp(option, "cache")) {
1324 if (!arg)
1325 return EXT2_ET_INVALID_ARGUMENT;
1326 if (!strcmp(arg, "on")) {
1327 data->flags &= ~IO_FLAG_NOCACHE;
1328 return 0;
1329 }
1330 if (!strcmp(arg, "off")) {
1331 retval = flush_cached_blocks(channel, data, 0);
1332 data->flags |= IO_FLAG_NOCACHE;
1333 return retval;
1334 }
1335 return EXT2_ET_INVALID_ARGUMENT;
1336 }
1337 return EXT2_ET_INVALID_ARGUMENT;
1338 }
1339
1340 #if defined(__linux__) && !defined(BLKDISCARD)
1341 #define BLKDISCARD _IO(0x12,119)
1342 #endif
1343
unix_discard(io_channel channel,unsigned long long block,unsigned long long count)1344 static errcode_t unix_discard(io_channel channel, unsigned long long block,
1345 unsigned long long count)
1346 {
1347 struct unix_private_data *data;
1348 int ret;
1349
1350 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1351 data = (struct unix_private_data *) channel->private_data;
1352 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1353
1354 if (channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE) {
1355 #ifdef BLKDISCARD
1356 __u64 range[2];
1357
1358 range[0] = (__u64)(block) * channel->block_size + data->offset;
1359 range[1] = (__u64)(count) * channel->block_size;
1360
1361 ret = ioctl(data->dev, BLKDISCARD, &range);
1362 #else
1363 goto unimplemented;
1364 #endif
1365 } else {
1366 #if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_PUNCH_HOLE)
1367 /*
1368 * If we are not on block device, try to use punch hole
1369 * to reclaim free space.
1370 */
1371 ret = fallocate(data->dev,
1372 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1373 (off_t)(block) * channel->block_size + data->offset,
1374 (off_t)(count) * channel->block_size);
1375 #else
1376 goto unimplemented;
1377 #endif
1378 }
1379 if (ret < 0) {
1380 if (errno == EOPNOTSUPP)
1381 goto unimplemented;
1382 return errno;
1383 }
1384 return 0;
1385 unimplemented:
1386 return EXT2_ET_UNIMPLEMENTED;
1387 }
1388
1389 /*
1390 * If we know about ZERO_RANGE, try that before we try PUNCH_HOLE because
1391 * ZERO_RANGE doesn't unmap preallocated blocks. We prefer fallocate because
1392 * it always invalidates page cache, and libext2fs requires that reads after
1393 * ZERO_RANGE return zeroes.
1394 */
__unix_zeroout(int fd,off_t offset,off_t len)1395 static int __unix_zeroout(int fd, off_t offset, off_t len)
1396 {
1397 int ret = -1;
1398
1399 #if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_ZERO_RANGE)
1400 ret = fallocate(fd, FALLOC_FL_ZERO_RANGE, offset, len);
1401 if (ret == 0)
1402 return 0;
1403 #endif
1404 #if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
1405 ret = fallocate(fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
1406 offset, len);
1407 if (ret == 0)
1408 return 0;
1409 #endif
1410 errno = EOPNOTSUPP;
1411 return ret;
1412 }
1413
1414 /* parameters might not be used if OS doesn't support zeroout */
1415 #if __GNUC_PREREQ (4, 6)
1416 #pragma GCC diagnostic push
1417 #pragma GCC diagnostic ignored "-Wunused-parameter"
1418 #endif
unix_zeroout(io_channel channel,unsigned long long block,unsigned long long count)1419 static errcode_t unix_zeroout(io_channel channel, unsigned long long block,
1420 unsigned long long count)
1421 {
1422 struct unix_private_data *data;
1423 int ret;
1424
1425 EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
1426 data = (struct unix_private_data *) channel->private_data;
1427 EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
1428
1429 if (safe_getenv("UNIX_IO_NOZEROOUT"))
1430 goto unimplemented;
1431
1432 if (!(channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE)) {
1433 /* Regular file, try to use truncate/punch/zero. */
1434 struct stat statbuf;
1435
1436 if (count == 0)
1437 return 0;
1438 /*
1439 * If we're trying to zero a range past the end of the file,
1440 * extend the file size, then truncate everything.
1441 */
1442 ret = fstat(data->dev, &statbuf);
1443 if (ret)
1444 goto err;
1445 if ((unsigned long long) statbuf.st_size <
1446 (block + count) * channel->block_size + data->offset) {
1447 ret = ftruncate(data->dev,
1448 (block + count) * channel->block_size + data->offset);
1449 if (ret)
1450 goto err;
1451 }
1452 }
1453
1454 ret = __unix_zeroout(data->dev,
1455 (off_t)(block) * channel->block_size + data->offset,
1456 (off_t)(count) * channel->block_size);
1457 err:
1458 if (ret < 0) {
1459 if (errno == EOPNOTSUPP)
1460 goto unimplemented;
1461 return errno;
1462 }
1463 return 0;
1464 unimplemented:
1465 return EXT2_ET_UNIMPLEMENTED;
1466 }
1467 #if __GNUC_PREREQ (4, 6)
1468 #pragma GCC diagnostic pop
1469 #endif
1470
1471 static struct struct_io_manager struct_unix_manager = {
1472 .magic = EXT2_ET_MAGIC_IO_MANAGER,
1473 .name = "Unix I/O Manager",
1474 .open = unix_open,
1475 .close = unix_close,
1476 .set_blksize = unix_set_blksize,
1477 .read_blk = unix_read_blk,
1478 .write_blk = unix_write_blk,
1479 .flush = unix_flush,
1480 .write_byte = unix_write_byte,
1481 .set_option = unix_set_option,
1482 .get_stats = unix_get_stats,
1483 .read_blk64 = unix_read_blk64,
1484 .write_blk64 = unix_write_blk64,
1485 .discard = unix_discard,
1486 .cache_readahead = unix_cache_readahead,
1487 .zeroout = unix_zeroout,
1488 };
1489
1490 io_manager unix_io_manager = &struct_unix_manager;
1491
1492 static struct struct_io_manager struct_unixfd_manager = {
1493 .magic = EXT2_ET_MAGIC_IO_MANAGER,
1494 .name = "Unix fd I/O Manager",
1495 .open = unixfd_open,
1496 .close = unix_close,
1497 .set_blksize = unix_set_blksize,
1498 .read_blk = unix_read_blk,
1499 .write_blk = unix_write_blk,
1500 .flush = unix_flush,
1501 .write_byte = unix_write_byte,
1502 .set_option = unix_set_option,
1503 .get_stats = unix_get_stats,
1504 .read_blk64 = unix_read_blk64,
1505 .write_blk64 = unix_write_blk64,
1506 .discard = unix_discard,
1507 .cache_readahead = unix_cache_readahead,
1508 .zeroout = unix_zeroout,
1509 };
1510
1511 io_manager unixfd_io_manager = &struct_unixfd_manager;
1512