]>
Commit | Line | Data |
---|---|---|
3839e657 | 1 | /* |
fff45483 | 2 | * unix_io.c --- This is the Unix (well, really POSIX) implementation |
a4613d13 | 3 | * of the I/O manager. |
3839e657 TT |
4 | * |
5 | * Implements a one-block write-through cache. | |
6 | * | |
efc6f628 | 7 | * Includes support for Windows NT support under Cygwin. |
fff45483 | 8 | * |
64e1b274 | 9 | * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001, |
a4613d13 | 10 | * 2002 by Theodore Ts'o. |
19c78dc0 TT |
11 | * |
12 | * %Begin-Header% | |
543547a5 TT |
13 | * This file may be redistributed under the terms of the GNU Library |
14 | * General Public License, version 2. | |
19c78dc0 | 15 | * %End-Header% |
3839e657 TT |
16 | */ |
17 | ||
ca82d22b | 18 | #if !defined(__FreeBSD__) && !defined(__NetBSD__) && !defined(__OpenBSD__) |
ca209dc6 AD |
19 | #define _XOPEN_SOURCE 600 |
20 | #define _DARWIN_C_SOURCE | |
21 | #define _FILE_OFFSET_BITS 64 | |
f1644c32 | 22 | #ifndef _LARGEFILE_SOURCE |
dc5f68ca | 23 | #define _LARGEFILE_SOURCE |
f1644c32 TT |
24 | #endif |
25 | #ifndef _LARGEFILE64_SOURCE | |
dc5f68ca | 26 | #define _LARGEFILE64_SOURCE |
f1644c32 | 27 | #endif |
cf5301d7 | 28 | #ifndef _GNU_SOURCE |
7f1a1fbf | 29 | #define _GNU_SOURCE |
cf5301d7 | 30 | #endif |
ca82d22b | 31 | #endif |
dc5f68ca | 32 | |
d1154eb4 | 33 | #include "config.h" |
3839e657 TT |
34 | #include <stdio.h> |
35 | #include <string.h> | |
4cbe8af4 | 36 | #if HAVE_UNISTD_H |
3839e657 | 37 | #include <unistd.h> |
4cbe8af4 | 38 | #endif |
c4e749ab TT |
39 | #if HAVE_ERRNO_H |
40 | #include <errno.h> | |
41 | #endif | |
3839e657 TT |
42 | #include <fcntl.h> |
43 | #include <time.h> | |
f154d2f6 TT |
44 | #ifdef __linux__ |
45 | #include <sys/utsname.h> | |
46 | #endif | |
ca209dc6 AD |
47 | #if HAVE_SYS_TYPES_H |
48 | #include <sys/types.h> | |
49 | #endif | |
7ed7a4b6 ES |
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 | |
1d2ff46a | 56 | #if HAVE_SYS_STAT_H |
3839e657 | 57 | #include <sys/stat.h> |
1d2ff46a | 58 | #endif |
fff45483 | 59 | #if HAVE_SYS_RESOURCE_H |
8880e759 | 60 | #include <sys/resource.h> |
fff45483 | 61 | #endif |
d2bfdc7f LC |
62 | #if HAVE_LINUX_FALLOC_H |
63 | #include <linux/falloc.h> | |
64 | #endif | |
3839e657 | 65 | |
7f1a1fbf | 66 | #if defined(__linux__) && defined(_IO) && !defined(BLKROGET) |
7ed7a4b6 ES |
67 | #define BLKROGET _IO(0x12, 94) /* Get read-only status (0 = read_write). */ |
68 | #endif | |
69 | ||
7f1a1fbf TT |
70 | #undef ALIGN_DEBUG |
71 | ||
b5abe6fa | 72 | #include "ext2_fs.h" |
7b4e4534 | 73 | #include "ext2fs.h" |
3839e657 | 74 | |
f3db3566 TT |
75 | /* |
76 | * For checking structure magic numbers... | |
77 | */ | |
78 | ||
79 | #define EXT2_CHECK_MAGIC(struct, code) \ | |
80 | if ((struct)->magic != (code)) return (code) | |
adfc8c6c TT |
81 | |
82 | struct unix_cache { | |
40024fdb TT |
83 | char *buf; |
84 | unsigned long long block; | |
85 | int access_time; | |
86 | unsigned dirty:1; | |
87 | unsigned in_use:1; | |
adfc8c6c TT |
88 | }; |
89 | ||
90 | #define CACHE_SIZE 8 | |
82c4660c TT |
91 | #define WRITE_DIRECT_SIZE 4 /* Must be smaller than CACHE_SIZE */ |
92 | #define READ_DIRECT_SIZE 4 /* Should be smaller than CACHE_SIZE */ | |
adfc8c6c | 93 | |
3839e657 | 94 | struct unix_private_data { |
f3db3566 | 95 | int magic; |
3839e657 TT |
96 | int dev; |
97 | int flags; | |
7f1a1fbf | 98 | int align; |
adfc8c6c | 99 | int access_time; |
2e8ca9a2 | 100 | ext2_loff_t offset; |
adfc8c6c | 101 | struct unix_cache cache[CACHE_SIZE]; |
7f1a1fbf | 102 | void *bounce; |
6d96b00d | 103 | struct struct_io_stats io_stats; |
3839e657 TT |
104 | }; |
105 | ||
d4e5abfb AS |
106 | #define IS_ALIGNED(n, align) ((((uintptr_t) n) & \ |
107 | ((uintptr_t) ((align)-1))) == 0) | |
7f1a1fbf | 108 | |
6d96b00d TT |
109 | static errcode_t unix_get_stats(io_channel channel, io_stats *stats) |
110 | { | |
a4613d13 | 111 | errcode_t retval = 0; |
6d96b00d TT |
112 | |
113 | struct unix_private_data *data; | |
114 | ||
115 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); | |
116 | data = (struct unix_private_data *) channel->private_data; | |
117 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); | |
118 | ||
119 | if (stats) | |
120 | *stats = &data->io_stats; | |
121 | ||
122 | return retval; | |
123 | } | |
124 | ||
adfc8c6c TT |
125 | /* |
126 | * Here are the raw I/O functions | |
127 | */ | |
128 | static errcode_t raw_read_blk(io_channel channel, | |
129 | struct unix_private_data *data, | |
59ecd32d | 130 | unsigned long long block, |
d32c915a | 131 | int count, void *bufv) |
adfc8c6c TT |
132 | { |
133 | errcode_t retval; | |
54434927 | 134 | ssize_t size; |
adfc8c6c TT |
135 | ext2_loff_t location; |
136 | int actual = 0; | |
d32c915a | 137 | unsigned char *buf = bufv; |
adfc8c6c TT |
138 | |
139 | size = (count < 0) ? -count : count * channel->block_size; | |
6d96b00d | 140 | data->io_stats.bytes_read += size; |
2e8ca9a2 | 141 | location = ((ext2_loff_t) block * channel->block_size) + data->offset; |
baa35446 | 142 | |
f00948ad | 143 | #ifdef HAVE_PREAD64 |
baa35446 DW |
144 | /* Try an aligned pread */ |
145 | if ((channel->align == 0) || | |
146 | (IS_ALIGNED(buf, channel->align) && | |
147 | IS_ALIGNED(size, channel->align))) { | |
f00948ad TT |
148 | actual = pread64(data->dev, buf, size, location); |
149 | if (actual == size) | |
150 | return 0; | |
151 | } | |
152 | #elif HAVE_PREAD | |
153 | /* Try an aligned pread */ | |
154 | if ((sizeof(off_t) >= sizeof(ext2_loff_t)) && | |
155 | ((channel->align == 0) || | |
156 | (IS_ALIGNED(buf, channel->align) && | |
157 | IS_ALIGNED(size, channel->align)))) { | |
baa35446 DW |
158 | actual = pread(data->dev, buf, size, location); |
159 | if (actual == size) | |
160 | return 0; | |
161 | } | |
162 | #endif /* HAVE_PREAD */ | |
163 | ||
adfc8c6c TT |
164 | if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { |
165 | retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; | |
166 | goto error_out; | |
167 | } | |
0a05b903 TT |
168 | if ((channel->align == 0) || |
169 | (IS_ALIGNED(buf, channel->align) && | |
170 | IS_ALIGNED(size, channel->align))) { | |
7f1a1fbf TT |
171 | actual = read(data->dev, buf, size); |
172 | if (actual != size) { | |
173 | short_read: | |
174 | if (actual < 0) | |
175 | actual = 0; | |
176 | retval = EXT2_ET_SHORT_READ; | |
177 | goto error_out; | |
178 | } | |
179 | return 0; | |
adfc8c6c | 180 | } |
fff45483 | 181 | |
7f1a1fbf TT |
182 | #ifdef ALIGN_DEBUG |
183 | printf("raw_read_blk: O_DIRECT fallback: %p %lu\n", buf, | |
184 | (unsigned long) size); | |
fff45483 | 185 | #endif |
7f1a1fbf TT |
186 | |
187 | /* | |
188 | * The buffer or size which we're trying to read isn't aligned | |
189 | * to the O_DIRECT rules, so we need to do this the hard way... | |
190 | */ | |
191 | while (size > 0) { | |
192 | actual = read(data->dev, data->bounce, channel->block_size); | |
193 | if (actual != channel->block_size) | |
fff45483 | 194 | goto short_read; |
7f1a1fbf TT |
195 | actual = size; |
196 | if (size > channel->block_size) | |
197 | actual = channel->block_size; | |
198 | memcpy(buf, data->bounce, actual); | |
199 | size -= actual; | |
200 | buf += actual; | |
fff45483 TT |
201 | } |
202 | return 0; | |
203 | ||
fff45483 | 204 | error_out: |
7f1a1fbf | 205 | memset((char *) buf+actual, 0, size-actual); |
fff45483 TT |
206 | if (channel->read_error) |
207 | retval = (channel->read_error)(channel, block, count, buf, | |
208 | size, actual, retval); | |
209 | return retval; | |
210 | } | |
adfc8c6c TT |
211 | |
212 | static errcode_t raw_write_blk(io_channel channel, | |
213 | struct unix_private_data *data, | |
59ecd32d | 214 | unsigned long long block, |
d32c915a | 215 | int count, const void *bufv) |
adfc8c6c | 216 | { |
54434927 | 217 | ssize_t size; |
adfc8c6c TT |
218 | ext2_loff_t location; |
219 | int actual = 0; | |
220 | errcode_t retval; | |
d32c915a | 221 | const unsigned char *buf = bufv; |
adfc8c6c TT |
222 | |
223 | if (count == 1) | |
224 | size = channel->block_size; | |
225 | else { | |
226 | if (count < 0) | |
227 | size = -count; | |
228 | else | |
229 | size = count * channel->block_size; | |
230 | } | |
6d96b00d | 231 | data->io_stats.bytes_written += size; |
adfc8c6c | 232 | |
2e8ca9a2 | 233 | location = ((ext2_loff_t) block * channel->block_size) + data->offset; |
baa35446 | 234 | |
f00948ad | 235 | #ifdef HAVE_PWRITE64 |
baa35446 DW |
236 | /* Try an aligned pwrite */ |
237 | if ((channel->align == 0) || | |
238 | (IS_ALIGNED(buf, channel->align) && | |
239 | IS_ALIGNED(size, channel->align))) { | |
f00948ad TT |
240 | actual = pwrite64(data->dev, buf, size, location); |
241 | if (actual == size) | |
242 | return 0; | |
243 | } | |
244 | #elif HAVE_PWRITE | |
245 | /* Try an aligned pwrite */ | |
246 | if ((sizeof(off_t) >= sizeof(ext2_loff_t)) && | |
247 | ((channel->align == 0) || | |
248 | (IS_ALIGNED(buf, channel->align) && | |
249 | IS_ALIGNED(size, channel->align)))) { | |
baa35446 DW |
250 | actual = pwrite(data->dev, buf, size, location); |
251 | if (actual == size) | |
252 | return 0; | |
253 | } | |
254 | #endif /* HAVE_PWRITE */ | |
255 | ||
adfc8c6c TT |
256 | if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { |
257 | retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; | |
258 | goto error_out; | |
259 | } | |
efc6f628 | 260 | |
0a05b903 TT |
261 | if ((channel->align == 0) || |
262 | (IS_ALIGNED(buf, channel->align) && | |
263 | IS_ALIGNED(size, channel->align))) { | |
7f1a1fbf TT |
264 | actual = write(data->dev, buf, size); |
265 | if (actual != size) { | |
266 | short_write: | |
267 | retval = EXT2_ET_SHORT_WRITE; | |
268 | goto error_out; | |
269 | } | |
270 | return 0; | |
271 | } | |
272 | ||
273 | #ifdef ALIGN_DEBUG | |
274 | printf("raw_write_blk: O_DIRECT fallback: %p %lu\n", buf, | |
275 | (unsigned long) size); | |
276 | #endif | |
277 | /* | |
278 | * The buffer or size which we're trying to write isn't aligned | |
279 | * to the O_DIRECT rules, so we need to do this the hard way... | |
280 | */ | |
281 | while (size > 0) { | |
282 | if (size < channel->block_size) { | |
283 | actual = read(data->dev, data->bounce, | |
284 | channel->block_size); | |
285 | if (actual != channel->block_size) { | |
286 | retval = EXT2_ET_SHORT_READ; | |
287 | goto error_out; | |
288 | } | |
289 | } | |
290 | actual = size; | |
291 | if (size > channel->block_size) | |
292 | actual = channel->block_size; | |
293 | memcpy(data->bounce, buf, actual); | |
127e2291 TT |
294 | if (ext2fs_llseek(data->dev, location, SEEK_SET) != location) { |
295 | retval = errno ? errno : EXT2_ET_LLSEEK_FAILED; | |
296 | goto error_out; | |
297 | } | |
7f1a1fbf TT |
298 | actual = write(data->dev, data->bounce, channel->block_size); |
299 | if (actual != channel->block_size) | |
300 | goto short_write; | |
301 | size -= actual; | |
302 | buf += actual; | |
adfc8c6c TT |
303 | } |
304 | return 0; | |
efc6f628 | 305 | |
adfc8c6c TT |
306 | error_out: |
307 | if (channel->write_error) | |
308 | retval = (channel->write_error)(channel, block, count, buf, | |
309 | size, actual, retval); | |
310 | return retval; | |
311 | } | |
312 | ||
313 | ||
314 | /* | |
315 | * Here we implement the cache functions | |
316 | */ | |
317 | ||
318 | /* Allocate the cache buffers */ | |
319 | static errcode_t alloc_cache(io_channel channel, | |
320 | struct unix_private_data *data) | |
321 | { | |
322 | errcode_t retval; | |
323 | struct unix_cache *cache; | |
324 | int i; | |
efc6f628 | 325 | |
adfc8c6c TT |
326 | data->access_time = 0; |
327 | for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { | |
328 | cache->block = 0; | |
329 | cache->access_time = 0; | |
330 | cache->dirty = 0; | |
331 | cache->in_use = 0; | |
faafdb76 TT |
332 | if (cache->buf) |
333 | ext2fs_free_mem(&cache->buf); | |
fd1c5a06 | 334 | retval = io_channel_alloc_buf(channel, 0, &cache->buf); |
7f1a1fbf | 335 | if (retval) |
adfc8c6c TT |
336 | return retval; |
337 | } | |
0a05b903 | 338 | if (channel->align) { |
7f1a1fbf TT |
339 | if (data->bounce) |
340 | ext2fs_free_mem(&data->bounce); | |
fd1c5a06 | 341 | retval = io_channel_alloc_buf(channel, 0, &data->bounce); |
7f1a1fbf TT |
342 | } |
343 | return retval; | |
adfc8c6c TT |
344 | } |
345 | ||
346 | /* Free the cache buffers */ | |
54434927 | 347 | static void free_cache(struct unix_private_data *data) |
adfc8c6c TT |
348 | { |
349 | struct unix_cache *cache; | |
350 | int i; | |
efc6f628 | 351 | |
adfc8c6c TT |
352 | data->access_time = 0; |
353 | for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { | |
354 | cache->block = 0; | |
355 | cache->access_time = 0; | |
356 | cache->dirty = 0; | |
357 | cache->in_use = 0; | |
358 | if (cache->buf) | |
c4e3d3f3 | 359 | ext2fs_free_mem(&cache->buf); |
adfc8c6c | 360 | } |
7f1a1fbf TT |
361 | if (data->bounce) |
362 | ext2fs_free_mem(&data->bounce); | |
adfc8c6c TT |
363 | } |
364 | ||
b8a95315 | 365 | #ifndef NO_IO_CACHE |
adfc8c6c | 366 | /* |
82c4660c TT |
367 | * Try to find a block in the cache. If the block is not found, and |
368 | * eldest is a non-zero pointer, then fill in eldest with the cache | |
369 | * entry to that should be reused. | |
adfc8c6c | 370 | */ |
54434927 | 371 | static struct unix_cache *find_cached_block(struct unix_private_data *data, |
59ecd32d | 372 | unsigned long long block, |
82c4660c | 373 | struct unix_cache **eldest) |
adfc8c6c | 374 | { |
31dbecd4 | 375 | struct unix_cache *cache, *unused_cache, *oldest_cache; |
adfc8c6c | 376 | int i; |
efc6f628 | 377 | |
31dbecd4 | 378 | unused_cache = oldest_cache = 0; |
adfc8c6c TT |
379 | for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { |
380 | if (!cache->in_use) { | |
82c4660c TT |
381 | if (!unused_cache) |
382 | unused_cache = cache; | |
adfc8c6c TT |
383 | continue; |
384 | } | |
385 | if (cache->block == block) { | |
386 | cache->access_time = ++data->access_time; | |
387 | return cache; | |
388 | } | |
389 | if (!oldest_cache || | |
390 | (cache->access_time < oldest_cache->access_time)) | |
391 | oldest_cache = cache; | |
392 | } | |
82c4660c TT |
393 | if (eldest) |
394 | *eldest = (unused_cache) ? unused_cache : oldest_cache; | |
395 | return 0; | |
396 | } | |
397 | ||
398 | /* | |
399 | * Reuse a particular cache entry for another block. | |
400 | */ | |
23b7c8b8 | 401 | static void reuse_cache(io_channel channel, struct unix_private_data *data, |
59ecd32d | 402 | struct unix_cache *cache, unsigned long long block) |
82c4660c TT |
403 | { |
404 | if (cache->dirty && cache->in_use) | |
405 | raw_write_blk(channel, data, cache->block, 1, cache->buf); | |
406 | ||
adfc8c6c | 407 | cache->in_use = 1; |
1d47dfb9 | 408 | cache->dirty = 0; |
adfc8c6c TT |
409 | cache->block = block; |
410 | cache->access_time = ++data->access_time; | |
adfc8c6c TT |
411 | } |
412 | ||
413 | /* | |
414 | * Flush all of the blocks in the cache | |
415 | */ | |
416 | static errcode_t flush_cached_blocks(io_channel channel, | |
417 | struct unix_private_data *data, | |
418 | int invalidate) | |
419 | ||
420 | { | |
421 | struct unix_cache *cache; | |
422 | errcode_t retval, retval2; | |
423 | int i; | |
efc6f628 | 424 | |
adfc8c6c TT |
425 | retval2 = 0; |
426 | for (i=0, cache = data->cache; i < CACHE_SIZE; i++, cache++) { | |
427 | if (!cache->in_use) | |
428 | continue; | |
efc6f628 | 429 | |
adfc8c6c TT |
430 | if (invalidate) |
431 | cache->in_use = 0; | |
efc6f628 | 432 | |
adfc8c6c TT |
433 | if (!cache->dirty) |
434 | continue; | |
efc6f628 | 435 | |
adfc8c6c TT |
436 | retval = raw_write_blk(channel, data, |
437 | cache->block, 1, cache->buf); | |
438 | if (retval) | |
439 | retval2 = retval; | |
440 | else | |
441 | cache->dirty = 0; | |
442 | } | |
443 | return retval2; | |
444 | } | |
b8a95315 | 445 | #endif /* NO_IO_CACHE */ |
adfc8c6c | 446 | |
d866599a LC |
447 | #ifdef __linux__ |
448 | #ifndef BLKDISCARDZEROES | |
449 | #define BLKDISCARDZEROES _IO(0x12,124) | |
450 | #endif | |
451 | #endif | |
452 | ||
182acd17 AD |
453 | int ext2fs_open_file(const char *pathname, int flags, mode_t mode) |
454 | { | |
455 | if (mode) | |
456 | #if defined(HAVE_OPEN64) && !defined(__OSX_AVAILABLE_BUT_DEPRECATED) | |
457 | return open64(pathname, flags, mode); | |
458 | else | |
459 | return open64(pathname, flags); | |
460 | #else | |
461 | return open(pathname, flags, mode); | |
462 | else | |
463 | return open(pathname, flags); | |
464 | #endif | |
465 | } | |
466 | ||
467 | int ext2fs_stat(const char *path, ext2fs_struct_stat *buf) | |
468 | { | |
469 | #if defined(HAVE_FSTAT64) && !defined(__OSX_AVAILABLE_BUT_DEPRECATED) | |
470 | return stat64(path, buf); | |
471 | #else | |
472 | return stat(path, buf); | |
473 | #endif | |
474 | } | |
475 | ||
476 | int ext2fs_fstat(int fd, ext2fs_struct_stat *buf) | |
477 | { | |
478 | #if defined(HAVE_FSTAT64) && !defined(__OSX_AVAILABLE_BUT_DEPRECATED) | |
479 | return fstat64(fd, buf); | |
480 | #else | |
481 | return fstat(fd, buf); | |
482 | #endif | |
483 | } | |
484 | ||
4ccf9e4f AS |
485 | |
486 | static errcode_t unix_open_channel(const char *name, int fd, | |
487 | int flags, io_channel *channel, | |
488 | io_manager io_mgr) | |
3839e657 TT |
489 | { |
490 | io_channel io = NULL; | |
491 | struct unix_private_data *data = NULL; | |
492 | errcode_t retval; | |
c859cb1d | 493 | ext2fs_struct_stat st; |
f154d2f6 | 494 | #ifdef __linux__ |
a4613d13 | 495 | struct utsname ut; |
f154d2f6 | 496 | #endif |
3839e657 | 497 | |
c4e3d3f3 | 498 | retval = ext2fs_get_mem(sizeof(struct struct_io_channel), &io); |
7b4e4534 | 499 | if (retval) |
624e8ebe | 500 | goto cleanup; |
f3db3566 TT |
501 | memset(io, 0, sizeof(struct struct_io_channel)); |
502 | io->magic = EXT2_ET_MAGIC_IO_CHANNEL; | |
c4e3d3f3 | 503 | retval = ext2fs_get_mem(sizeof(struct unix_private_data), &data); |
7b4e4534 | 504 | if (retval) |
3839e657 | 505 | goto cleanup; |
7b4e4534 | 506 | |
4ccf9e4f | 507 | io->manager = io_mgr; |
c4e3d3f3 | 508 | retval = ext2fs_get_mem(strlen(name)+1, &io->name); |
7b4e4534 | 509 | if (retval) |
3839e657 | 510 | goto cleanup; |
7b4e4534 | 511 | |
3839e657 TT |
512 | strcpy(io->name, name); |
513 | io->private_data = data; | |
f3db3566 TT |
514 | io->block_size = 1024; |
515 | io->read_error = 0; | |
516 | io->write_error = 0; | |
a29f4d30 | 517 | io->refcount = 1; |
3839e657 TT |
518 | |
519 | memset(data, 0, sizeof(struct unix_private_data)); | |
f3db3566 | 520 | data->magic = EXT2_ET_MAGIC_UNIX_IO_CHANNEL; |
6d96b00d | 521 | data->io_stats.num_fields = 2; |
4ccf9e4f AS |
522 | data->flags = flags; |
523 | data->dev = fd; | |
7b4e4534 | 524 | |
d9a5d375 | 525 | #if defined(O_DIRECT) |
4ccf9e4f | 526 | if (flags & IO_FLAG_DIRECT_IO) |
dd0a2679 | 527 | io->align = ext2fs_get_dio_alignment(data->dev); |
d9a5d375 | 528 | #elif defined(F_NOCACHE) |
4ccf9e4f | 529 | if (flags & IO_FLAG_DIRECT_IO) |
dd0a2679 | 530 | io->align = 4096; |
534a4c3d | 531 | #endif |
64e1b274 | 532 | |
d2bfdc7f LC |
533 | /* |
534 | * If the device is really a block device, then set the | |
535 | * appropriate flag, otherwise we can set DISCARD_ZEROES flag | |
536 | * because we are going to use punch hole instead of discard | |
537 | * and if it succeed, subsequent read from sparse area returns | |
538 | * zero. | |
539 | */ | |
4ccf9e4f | 540 | if (ext2fs_fstat(data->dev, &st) == 0) { |
d2bfdc7f LC |
541 | if (S_ISBLK(st.st_mode)) |
542 | io->flags |= CHANNEL_FLAGS_BLOCK_DEVICE; | |
543 | else | |
544 | io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES; | |
545 | } | |
546 | ||
d866599a | 547 | #ifdef BLKDISCARDZEROES |
1d6fd6d0 AD |
548 | { |
549 | int zeroes = 0; | |
550 | if (ioctl(data->dev, BLKDISCARDZEROES, &zeroes) == 0 && | |
551 | zeroes) | |
552 | io->flags |= CHANNEL_FLAGS_DISCARD_ZEROES; | |
553 | } | |
d866599a LC |
554 | #endif |
555 | ||
7f1a1fbf TT |
556 | #if defined(__CYGWIN__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
557 | /* | |
558 | * Some operating systems require that the buffers be aligned, | |
559 | * regardless of O_DIRECT | |
560 | */ | |
dd0a2679 TT |
561 | if (!io->align) |
562 | io->align = 512; | |
7f1a1fbf TT |
563 | #endif |
564 | ||
565 | ||
566 | if ((retval = alloc_cache(io, data))) | |
567 | goto cleanup; | |
568 | ||
7ed7a4b6 ES |
569 | #ifdef BLKROGET |
570 | if (flags & IO_FLAG_RW) { | |
571 | int error; | |
572 | int readonly = 0; | |
573 | ||
574 | /* Is the block device actually writable? */ | |
575 | error = ioctl(data->dev, BLKROGET, &readonly); | |
576 | if (!error && readonly) { | |
7ed7a4b6 ES |
577 | retval = EPERM; |
578 | goto cleanup; | |
579 | } | |
580 | } | |
581 | #endif | |
582 | ||
64e1b274 TT |
583 | #ifdef __linux__ |
584 | #undef RLIM_INFINITY | |
585 | #if (defined(__alpha__) || ((defined(__sparc__) || defined(__mips__)) && (SIZEOF_LONG == 4))) | |
586 | #define RLIM_INFINITY ((unsigned long)(~0UL>>1)) | |
587 | #else | |
588 | #define RLIM_INFINITY (~0UL) | |
589 | #endif | |
8880e759 | 590 | /* |
f154d2f6 TT |
591 | * Work around a bug in 2.4.10-2.4.18 kernels where writes to |
592 | * block devices are wrongly getting hit by the filesize | |
593 | * limit. This workaround isn't perfect, since it won't work | |
594 | * if glibc wasn't built against 2.2 header files. (Sigh.) | |
efc6f628 | 595 | * |
8880e759 | 596 | */ |
f154d2f6 TT |
597 | if ((flags & IO_FLAG_RW) && |
598 | (uname(&ut) == 0) && | |
599 | ((ut.release[0] == '2') && (ut.release[1] == '.') && | |
600 | (ut.release[2] == '4') && (ut.release[3] == '.') && | |
601 | (ut.release[4] == '1') && (ut.release[5] >= '0') && | |
602 | (ut.release[5] < '8')) && | |
4ccf9e4f | 603 | (ext2fs_fstat(data->dev, &st) == 0) && |
8880e759 TT |
604 | (S_ISBLK(st.st_mode))) { |
605 | struct rlimit rlim; | |
efc6f628 | 606 | |
64e1b274 | 607 | rlim.rlim_cur = rlim.rlim_max = (unsigned long) RLIM_INFINITY; |
8880e759 TT |
608 | setrlimit(RLIMIT_FSIZE, &rlim); |
609 | getrlimit(RLIMIT_FSIZE, &rlim); | |
bd27880b TT |
610 | if (((unsigned long) rlim.rlim_cur) < |
611 | ((unsigned long) rlim.rlim_max)) { | |
8880e759 TT |
612 | rlim.rlim_cur = rlim.rlim_max; |
613 | setrlimit(RLIMIT_FSIZE, &rlim); | |
614 | } | |
615 | } | |
64e1b274 | 616 | #endif |
3839e657 TT |
617 | *channel = io; |
618 | return 0; | |
619 | ||
620 | cleanup: | |
3839e657 | 621 | if (data) { |
4e0bb5eb TT |
622 | if (data->dev >= 0) |
623 | close(data->dev); | |
54434927 | 624 | free_cache(data); |
c4e3d3f3 | 625 | ext2fs_free_mem(&data); |
3839e657 | 626 | } |
4e0bb5eb TT |
627 | if (io) { |
628 | if (io->name) { | |
629 | ext2fs_free_mem(&io->name); | |
630 | } | |
c4e3d3f3 | 631 | ext2fs_free_mem(&io); |
4e0bb5eb | 632 | } |
3839e657 TT |
633 | return retval; |
634 | } | |
635 | ||
4ccf9e4f AS |
636 | static errcode_t unixfd_open(const char *str_fd, int flags, |
637 | io_channel *channel) | |
638 | { | |
639 | int fd; | |
640 | int fd_flags; | |
641 | ||
642 | fd = atoi(str_fd); | |
643 | fd_flags = fcntl(fd, F_GETFD); | |
644 | if (fd_flags == -1) | |
645 | return -EBADF; | |
646 | ||
647 | flags = 0; | |
648 | if (fd_flags & O_RDWR) | |
649 | flags |= IO_FLAG_RW; | |
650 | if (fd_flags & O_EXCL) | |
651 | flags |= IO_FLAG_EXCLUSIVE; | |
652 | #if defined(O_DIRECT) | |
653 | if (fd_flags & O_DIRECT) | |
654 | flags |= IO_FLAG_DIRECT_IO; | |
655 | #endif | |
656 | ||
657 | return unix_open_channel(str_fd, fd, flags, channel, unixfd_io_manager); | |
658 | } | |
659 | ||
660 | static errcode_t unix_open(const char *name, int flags, | |
661 | io_channel *channel) | |
662 | { | |
663 | int fd = -1; | |
664 | int open_flags; | |
665 | ||
666 | if (name == 0) | |
667 | return EXT2_ET_BAD_DEVICE_NAME; | |
668 | ||
669 | open_flags = (flags & IO_FLAG_RW) ? O_RDWR : O_RDONLY; | |
670 | if (flags & IO_FLAG_EXCLUSIVE) | |
671 | open_flags |= O_EXCL; | |
672 | #if defined(O_DIRECT) | |
673 | if (flags & IO_FLAG_DIRECT_IO) | |
674 | open_flags |= O_DIRECT; | |
675 | #endif | |
676 | fd = ext2fs_open_file(name, open_flags, 0); | |
677 | if (fd < 0) | |
678 | return errno; | |
679 | #if defined(F_NOCACHE) && !defined(IO_DIRECT) | |
680 | if (flags & IO_FLAG_DIRECT_IO) { | |
681 | if (fcntl(fd, F_NOCACHE, 1) < 0) | |
682 | return errno; | |
683 | } | |
684 | #endif | |
685 | return unix_open_channel(name, fd, flags, channel, unix_io_manager); | |
686 | } | |
687 | ||
3839e657 TT |
688 | static errcode_t unix_close(io_channel channel) |
689 | { | |
690 | struct unix_private_data *data; | |
691 | errcode_t retval = 0; | |
692 | ||
f3db3566 | 693 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); |
3839e657 | 694 | data = (struct unix_private_data *) channel->private_data; |
f3db3566 | 695 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); |
a29f4d30 TT |
696 | |
697 | if (--channel->refcount > 0) | |
698 | return 0; | |
adfc8c6c | 699 | |
b8a95315 | 700 | #ifndef NO_IO_CACHE |
adfc8c6c | 701 | retval = flush_cached_blocks(channel, data, 0); |
b8a95315 | 702 | #endif |
adfc8c6c | 703 | |
3839e657 TT |
704 | if (close(data->dev) < 0) |
705 | retval = errno; | |
54434927 | 706 | free_cache(data); |
f12e285f | 707 | |
c4e3d3f3 | 708 | ext2fs_free_mem(&channel->private_data); |
3839e657 | 709 | if (channel->name) |
c4e3d3f3 TT |
710 | ext2fs_free_mem(&channel->name); |
711 | ext2fs_free_mem(&channel); | |
3839e657 TT |
712 | return retval; |
713 | } | |
714 | ||
715 | static errcode_t unix_set_blksize(io_channel channel, int blksize) | |
716 | { | |
717 | struct unix_private_data *data; | |
7b4e4534 | 718 | errcode_t retval; |
3839e657 | 719 | |
f3db3566 | 720 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); |
3839e657 | 721 | data = (struct unix_private_data *) channel->private_data; |
f3db3566 TT |
722 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); |
723 | ||
3839e657 | 724 | if (channel->block_size != blksize) { |
b8a95315 | 725 | #ifndef NO_IO_CACHE |
adfc8c6c TT |
726 | if ((retval = flush_cached_blocks(channel, data, 0))) |
727 | return retval; | |
b8a95315 | 728 | #endif |
efc6f628 | 729 | |
3839e657 | 730 | channel->block_size = blksize; |
54434927 | 731 | free_cache(data); |
adfc8c6c | 732 | if ((retval = alloc_cache(channel, data))) |
7b4e4534 | 733 | return retval; |
3839e657 TT |
734 | } |
735 | return 0; | |
736 | } | |
737 | ||
59ecd32d | 738 | static errcode_t unix_read_blk64(io_channel channel, unsigned long long block, |
3839e657 TT |
739 | int count, void *buf) |
740 | { | |
741 | struct unix_private_data *data; | |
82c4660c | 742 | struct unix_cache *cache, *reuse[READ_DIRECT_SIZE]; |
3839e657 | 743 | errcode_t retval; |
31dbecd4 | 744 | char *cp; |
adfc8c6c | 745 | int i, j; |
3839e657 | 746 | |
f3db3566 | 747 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); |
3839e657 | 748 | data = (struct unix_private_data *) channel->private_data; |
f3db3566 | 749 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); |
3839e657 | 750 | |
b8a95315 TT |
751 | #ifdef NO_IO_CACHE |
752 | return raw_read_blk(channel, data, block, count, buf); | |
753 | #else | |
3839e657 | 754 | /* |
82c4660c TT |
755 | * If we're doing an odd-sized read or a very large read, |
756 | * flush out the cache and then do a direct read. | |
3839e657 | 757 | */ |
82c4660c | 758 | if (count < 0 || count > WRITE_DIRECT_SIZE) { |
adfc8c6c TT |
759 | if ((retval = flush_cached_blocks(channel, data, 0))) |
760 | return retval; | |
761 | return raw_read_blk(channel, data, block, count, buf); | |
3839e657 | 762 | } |
adfc8c6c | 763 | |
31dbecd4 | 764 | cp = buf; |
adfc8c6c TT |
765 | while (count > 0) { |
766 | /* If it's in the cache, use it! */ | |
54434927 | 767 | if ((cache = find_cached_block(data, block, &reuse[0]))) { |
adfc8c6c | 768 | #ifdef DEBUG |
d0ff90d5 | 769 | printf("Using cached block %lu\n", block); |
f3db3566 | 770 | #endif |
31dbecd4 | 771 | memcpy(cp, cache->buf, channel->block_size); |
adfc8c6c TT |
772 | count--; |
773 | block++; | |
31dbecd4 | 774 | cp += channel->block_size; |
adfc8c6c TT |
775 | continue; |
776 | } | |
7f1a1fbf TT |
777 | if (count == 1) { |
778 | /* | |
779 | * Special case where we read directly into the | |
780 | * cache buffer; important in the O_DIRECT case | |
781 | */ | |
782 | cache = reuse[0]; | |
783 | reuse_cache(channel, data, cache, block); | |
784 | if ((retval = raw_read_blk(channel, data, block, 1, | |
785 | cache->buf))) { | |
786 | cache->in_use = 0; | |
787 | return retval; | |
788 | } | |
789 | memcpy(cp, cache->buf, channel->block_size); | |
790 | return 0; | |
791 | } | |
792 | ||
adfc8c6c TT |
793 | /* |
794 | * Find the number of uncached blocks so we can do a | |
795 | * single read request | |
796 | */ | |
797 | for (i=1; i < count; i++) | |
54434927 | 798 | if (find_cached_block(data, block+i, &reuse[i])) |
adfc8c6c TT |
799 | break; |
800 | #ifdef DEBUG | |
d0ff90d5 | 801 | printf("Reading %d blocks starting at %lu\n", i, block); |
adfc8c6c | 802 | #endif |
31dbecd4 | 803 | if ((retval = raw_read_blk(channel, data, block, i, cp))) |
adfc8c6c | 804 | return retval; |
efc6f628 | 805 | |
adfc8c6c TT |
806 | /* Save the results in the cache */ |
807 | for (j=0; j < i; j++) { | |
808 | count--; | |
82c4660c TT |
809 | cache = reuse[j]; |
810 | reuse_cache(channel, data, cache, block++); | |
811 | memcpy(cache->buf, cp, channel->block_size); | |
31dbecd4 | 812 | cp += channel->block_size; |
adfc8c6c | 813 | } |
3839e657 TT |
814 | } |
815 | return 0; | |
b8a95315 | 816 | #endif /* NO_IO_CACHE */ |
3839e657 TT |
817 | } |
818 | ||
59ecd32d JS |
819 | static errcode_t unix_read_blk(io_channel channel, unsigned long block, |
820 | int count, void *buf) | |
821 | { | |
822 | return unix_read_blk64(channel, block, count, buf); | |
823 | } | |
824 | ||
825 | static errcode_t unix_write_blk64(io_channel channel, unsigned long long block, | |
3839e657 TT |
826 | int count, const void *buf) |
827 | { | |
828 | struct unix_private_data *data; | |
82c4660c | 829 | struct unix_cache *cache, *reuse; |
23b7c8b8 | 830 | errcode_t retval = 0; |
31dbecd4 TT |
831 | const char *cp; |
832 | int writethrough; | |
3839e657 | 833 | |
f3db3566 | 834 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); |
3839e657 | 835 | data = (struct unix_private_data *) channel->private_data; |
f3db3566 | 836 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); |
3839e657 | 837 | |
b8a95315 TT |
838 | #ifdef NO_IO_CACHE |
839 | return raw_write_blk(channel, data, block, count, buf); | |
efc6f628 | 840 | #else |
adfc8c6c TT |
841 | /* |
842 | * If we're doing an odd-sized write or a very large write, | |
843 | * flush out the cache completely and then do a direct write. | |
844 | */ | |
82c4660c | 845 | if (count < 0 || count > WRITE_DIRECT_SIZE) { |
adfc8c6c TT |
846 | if ((retval = flush_cached_blocks(channel, data, 1))) |
847 | return retval; | |
848 | return raw_write_blk(channel, data, block, count, buf); | |
3839e657 TT |
849 | } |
850 | ||
adfc8c6c TT |
851 | /* |
852 | * For a moderate-sized multi-block write, first force a write | |
853 | * if we're in write-through cache mode, and then fill the | |
854 | * cache with the blocks. | |
855 | */ | |
856 | writethrough = channel->flags & CHANNEL_FLAGS_WRITETHROUGH; | |
857 | if (writethrough) | |
858 | retval = raw_write_blk(channel, data, block, count, buf); | |
efc6f628 | 859 | |
31dbecd4 | 860 | cp = buf; |
adfc8c6c | 861 | while (count > 0) { |
54434927 | 862 | cache = find_cached_block(data, block, &reuse); |
adfc8c6c | 863 | if (!cache) { |
82c4660c TT |
864 | cache = reuse; |
865 | reuse_cache(channel, data, cache, block); | |
adfc8c6c | 866 | } |
8d5324c4 DW |
867 | if (cache->buf != cp) |
868 | memcpy(cache->buf, cp, channel->block_size); | |
82c4660c | 869 | cache->dirty = !writethrough; |
adfc8c6c TT |
870 | count--; |
871 | block++; | |
31dbecd4 | 872 | cp += channel->block_size; |
adfc8c6c | 873 | } |
3839e657 | 874 | return retval; |
b8a95315 | 875 | #endif /* NO_IO_CACHE */ |
3839e657 TT |
876 | } |
877 | ||
ca209dc6 AD |
878 | static errcode_t unix_cache_readahead(io_channel channel, |
879 | unsigned long long block, | |
880 | unsigned long long count) | |
881 | { | |
882 | #ifdef POSIX_FADV_WILLNEED | |
883 | struct unix_private_data *data; | |
884 | ||
885 | data = (struct unix_private_data *)channel->private_data; | |
a7ac9c22 | 886 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); |
ca209dc6 | 887 | return posix_fadvise(data->dev, |
a7ac9c22 | 888 | (ext2_loff_t)block * channel->block_size + data->offset, |
ca209dc6 AD |
889 | (ext2_loff_t)count * channel->block_size, |
890 | POSIX_FADV_WILLNEED); | |
891 | #else | |
892 | return EXT2_ET_OP_NOT_SUPPORTED; | |
893 | #endif | |
894 | } | |
895 | ||
59ecd32d JS |
896 | static errcode_t unix_write_blk(io_channel channel, unsigned long block, |
897 | int count, const void *buf) | |
898 | { | |
899 | return unix_write_blk64(channel, block, count, buf); | |
900 | } | |
901 | ||
c180ac86 TT |
902 | static errcode_t unix_write_byte(io_channel channel, unsigned long offset, |
903 | int size, const void *buf) | |
904 | { | |
905 | struct unix_private_data *data; | |
31dbecd4 | 906 | errcode_t retval = 0; |
54434927 | 907 | ssize_t actual; |
c180ac86 TT |
908 | |
909 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); | |
910 | data = (struct unix_private_data *) channel->private_data; | |
911 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); | |
912 | ||
0a05b903 | 913 | if (channel->align != 0) { |
7f1a1fbf TT |
914 | #ifdef ALIGN_DEBUG |
915 | printf("unix_write_byte: O_DIRECT fallback\n"); | |
916 | #endif | |
917 | return EXT2_ET_UNIMPLEMENTED; | |
918 | } | |
919 | ||
b8a95315 | 920 | #ifndef NO_IO_CACHE |
c180ac86 TT |
921 | /* |
922 | * Flush out the cache completely | |
923 | */ | |
924 | if ((retval = flush_cached_blocks(channel, data, 1))) | |
925 | return retval; | |
b8a95315 | 926 | #endif |
c180ac86 | 927 | |
2e8ca9a2 | 928 | if (lseek(data->dev, offset + data->offset, SEEK_SET) < 0) |
c180ac86 | 929 | return errno; |
efc6f628 | 930 | |
c180ac86 TT |
931 | actual = write(data->dev, buf, size); |
932 | if (actual != size) | |
933 | return EXT2_ET_SHORT_WRITE; | |
934 | ||
935 | return 0; | |
936 | } | |
937 | ||
3839e657 | 938 | /* |
efc6f628 | 939 | * Flush data buffers to disk. |
3839e657 TT |
940 | */ |
941 | static errcode_t unix_flush(io_channel channel) | |
942 | { | |
f3db3566 | 943 | struct unix_private_data *data; |
adfc8c6c | 944 | errcode_t retval = 0; |
efc6f628 | 945 | |
f3db3566 TT |
946 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); |
947 | data = (struct unix_private_data *) channel->private_data; | |
948 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); | |
adfc8c6c | 949 | |
b8a95315 | 950 | #ifndef NO_IO_CACHE |
adfc8c6c | 951 | retval = flush_cached_blocks(channel, data, 0); |
b8a95315 | 952 | #endif |
36f21439 | 953 | fsync(data->dev); |
adfc8c6c | 954 | return retval; |
3839e657 TT |
955 | } |
956 | ||
efc6f628 | 957 | static errcode_t unix_set_option(io_channel channel, const char *option, |
2e8ca9a2 TT |
958 | const char *arg) |
959 | { | |
960 | struct unix_private_data *data; | |
2aee23f3 | 961 | unsigned long long tmp; |
2e8ca9a2 TT |
962 | char *end; |
963 | ||
964 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); | |
965 | data = (struct unix_private_data *) channel->private_data; | |
966 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); | |
967 | ||
968 | if (!strcmp(option, "offset")) { | |
969 | if (!arg) | |
970 | return EXT2_ET_INVALID_ARGUMENT; | |
971 | ||
2aee23f3 | 972 | tmp = strtoull(arg, &end, 0); |
2e8ca9a2 TT |
973 | if (*end) |
974 | return EXT2_ET_INVALID_ARGUMENT; | |
975 | data->offset = tmp; | |
2aee23f3 TT |
976 | if (data->offset < 0) |
977 | return EXT2_ET_INVALID_ARGUMENT; | |
2e8ca9a2 TT |
978 | return 0; |
979 | } | |
980 | return EXT2_ET_INVALID_ARGUMENT; | |
981 | } | |
e90a59ed LC |
982 | |
983 | #if defined(__linux__) && !defined(BLKDISCARD) | |
d2bfdc7f | 984 | #define BLKDISCARD _IO(0x12,119) |
e90a59ed LC |
985 | #endif |
986 | ||
987 | static errcode_t unix_discard(io_channel channel, unsigned long long block, | |
988 | unsigned long long count) | |
989 | { | |
e90a59ed | 990 | struct unix_private_data *data; |
e90a59ed LC |
991 | int ret; |
992 | ||
993 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); | |
994 | data = (struct unix_private_data *) channel->private_data; | |
995 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); | |
996 | ||
d2bfdc7f LC |
997 | if (channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE) { |
998 | #ifdef BLKDISCARD | |
795c02de | 999 | __u64 range[2]; |
1d6fd6d0 | 1000 | |
a7ac9c22 | 1001 | range[0] = (__u64)(block) * channel->block_size + data->offset; |
795c02de | 1002 | range[1] = (__u64)(count) * channel->block_size; |
e90a59ed | 1003 | |
d2bfdc7f LC |
1004 | ret = ioctl(data->dev, BLKDISCARD, &range); |
1005 | #else | |
1006 | goto unimplemented; | |
1007 | #endif | |
1008 | } else { | |
800766ee | 1009 | #if defined(HAVE_FALLOCATE) && defined(FALLOC_FL_PUNCH_HOLE) |
d2bfdc7f LC |
1010 | /* |
1011 | * If we are not on block device, try to use punch hole | |
1012 | * to reclaim free space. | |
1013 | */ | |
1014 | ret = fallocate(data->dev, | |
1015 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
a7ac9c22 | 1016 | (off_t)(block) * channel->block_size + data->offset, |
d2bfdc7f LC |
1017 | (off_t)(count) * channel->block_size); |
1018 | #else | |
1019 | goto unimplemented; | |
1020 | #endif | |
1021 | } | |
1022 | if (ret < 0) { | |
1023 | if (errno == EOPNOTSUPP) | |
1024 | goto unimplemented; | |
e90a59ed | 1025 | return errno; |
d2bfdc7f | 1026 | } |
e90a59ed | 1027 | return 0; |
d2bfdc7f | 1028 | unimplemented: |
e90a59ed | 1029 | return EXT2_ET_UNIMPLEMENTED; |
e90a59ed | 1030 | } |
a4613d13 | 1031 | |
25f291c9 TT |
1032 | /* parameters might not be used if OS doesn't support zeroout */ |
1033 | #pragma GCC diagnostic push | |
1034 | #pragma GCC diagnostic ignored "-Wunused-parameter" | |
3d28f545 DW |
1035 | static errcode_t unix_zeroout(io_channel channel, unsigned long long block, |
1036 | unsigned long long count) | |
1037 | { | |
1038 | struct unix_private_data *data; | |
1039 | int ret; | |
1040 | ||
1041 | EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL); | |
1042 | data = (struct unix_private_data *) channel->private_data; | |
1043 | EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL); | |
1044 | ||
1045 | if (getenv("UNIX_IO_NOZEROOUT")) | |
1046 | goto unimplemented; | |
1047 | ||
1048 | if (channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE) { | |
1049 | /* Not implemented until the BLKZEROOUT mess is fixed */ | |
1050 | goto unimplemented; | |
1051 | } else { | |
1052 | /* Regular file, try to use truncate/punch/zero. */ | |
3d28f545 DW |
1053 | struct stat statbuf; |
1054 | ||
1055 | if (count == 0) | |
1056 | return 0; | |
1057 | /* | |
1058 | * If we're trying to zero a range past the end of the file, | |
cf491d3a | 1059 | * extend the file size, then truncate everything. |
3d28f545 DW |
1060 | */ |
1061 | ret = fstat(data->dev, &statbuf); | |
1062 | if (ret) | |
1063 | goto err; | |
a7ac9c22 MH |
1064 | if ((unsigned long long) statbuf.st_size < |
1065 | (block + count) * channel->block_size + data->offset) { | |
3d28f545 | 1066 | ret = ftruncate(data->dev, |
a7ac9c22 | 1067 | (block + count) * channel->block_size + data->offset); |
3d28f545 DW |
1068 | if (ret) |
1069 | goto err; | |
1070 | } | |
cf491d3a TT |
1071 | #if defined(HAVE_FALLOCATE) && (defined(FALLOC_FL_ZERO_RANGE) || \ |
1072 | (defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE))) | |
3d28f545 DW |
1073 | #if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE) |
1074 | ret = fallocate(data->dev, | |
1075 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
a7ac9c22 | 1076 | (off_t)(block) * channel->block_size + data->offset, |
3d28f545 DW |
1077 | (off_t)(count) * channel->block_size); |
1078 | if (ret == 0) | |
1079 | goto err; | |
1080 | #endif | |
1081 | #ifdef FALLOC_FL_ZERO_RANGE | |
1082 | ret = fallocate(data->dev, | |
1083 | FALLOC_FL_ZERO_RANGE, | |
a7ac9c22 | 1084 | (off_t)(block) * channel->block_size + data->offset, |
3d28f545 DW |
1085 | (off_t)(count) * channel->block_size); |
1086 | #endif | |
1087 | #else | |
1088 | goto unimplemented; | |
1089 | #endif /* HAVE_FALLOCATE && (ZERO_RANGE || (PUNCH_HOLE && KEEP_SIZE)) */ | |
1090 | } | |
1091 | err: | |
1092 | if (ret < 0) { | |
1093 | if (errno == EOPNOTSUPP) | |
1094 | goto unimplemented; | |
1095 | return errno; | |
1096 | } | |
1097 | return 0; | |
1098 | unimplemented: | |
1099 | return EXT2_ET_UNIMPLEMENTED; | |
1100 | } | |
25f291c9 | 1101 | #pragma GCC diagnostic pop |
3d28f545 | 1102 | |
a4613d13 | 1103 | static struct struct_io_manager struct_unix_manager = { |
d4ecec45 TT |
1104 | .magic = EXT2_ET_MAGIC_IO_MANAGER, |
1105 | .name = "Unix I/O Manager", | |
1106 | .open = unix_open, | |
1107 | .close = unix_close, | |
1108 | .set_blksize = unix_set_blksize, | |
1109 | .read_blk = unix_read_blk, | |
1110 | .write_blk = unix_write_blk, | |
1111 | .flush = unix_flush, | |
1112 | .write_byte = unix_write_byte, | |
1113 | .set_option = unix_set_option, | |
1114 | .get_stats = unix_get_stats, | |
1115 | .read_blk64 = unix_read_blk64, | |
1116 | .write_blk64 = unix_write_blk64, | |
1117 | .discard = unix_discard, | |
ca209dc6 | 1118 | .cache_readahead = unix_cache_readahead, |
3d28f545 | 1119 | .zeroout = unix_zeroout, |
a4613d13 AD |
1120 | }; |
1121 | ||
1122 | io_manager unix_io_manager = &struct_unix_manager; | |
4ccf9e4f AS |
1123 | |
1124 | static struct struct_io_manager struct_unixfd_manager = { | |
1125 | .magic = EXT2_ET_MAGIC_IO_MANAGER, | |
1126 | .name = "Unix fd I/O Manager", | |
1127 | .open = unixfd_open, | |
1128 | .close = unix_close, | |
1129 | .set_blksize = unix_set_blksize, | |
1130 | .read_blk = unix_read_blk, | |
1131 | .write_blk = unix_write_blk, | |
1132 | .flush = unix_flush, | |
1133 | .write_byte = unix_write_byte, | |
1134 | .set_option = unix_set_option, | |
1135 | .get_stats = unix_get_stats, | |
1136 | .read_blk64 = unix_read_blk64, | |
1137 | .write_blk64 = unix_write_blk64, | |
1138 | .discard = unix_discard, | |
1139 | .cache_readahead = unix_cache_readahead, | |
1140 | .zeroout = unix_zeroout, | |
1141 | }; | |
1142 | ||
1143 | io_manager unixfd_io_manager = &struct_unixfd_manager; |