]>
Commit | Line | Data |
---|---|---|
83f64091 | 1 | /* |
223d4670 | 2 | * Block driver for RAW files (posix) |
5fafdf24 | 3 | * |
83f64091 | 4 | * Copyright (c) 2006 Fabrice Bellard |
5fafdf24 | 5 | * |
83f64091 FB |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
faf07963 | 24 | #include "qemu-common.h" |
1de7afc9 PB |
25 | #include "qemu/timer.h" |
26 | #include "qemu/log.h" | |
737e150e | 27 | #include "block/block_int.h" |
1de7afc9 | 28 | #include "qemu/module.h" |
de81a169 | 29 | #include "trace.h" |
737e150e | 30 | #include "block/thread-pool.h" |
1de7afc9 | 31 | #include "qemu/iov.h" |
9f8540ec | 32 | #include "raw-aio.h" |
83f64091 | 33 | |
83affaa6 | 34 | #if defined(__APPLE__) && (__MACH__) |
83f64091 FB |
35 | #include <paths.h> |
36 | #include <sys/param.h> | |
37 | #include <IOKit/IOKitLib.h> | |
38 | #include <IOKit/IOBSD.h> | |
39 | #include <IOKit/storage/IOMediaBSDClient.h> | |
40 | #include <IOKit/storage/IOMedia.h> | |
41 | #include <IOKit/storage/IOCDMedia.h> | |
42 | //#include <IOKit/storage/IOCDTypes.h> | |
43 | #include <CoreFoundation/CoreFoundation.h> | |
44 | #endif | |
45 | ||
46 | #ifdef __sun__ | |
2e9671da | 47 | #define _POSIX_PTHREAD_SEMANTICS 1 |
83f64091 FB |
48 | #include <sys/dkio.h> |
49 | #endif | |
19cb3738 | 50 | #ifdef __linux__ |
343f8568 JS |
51 | #include <sys/types.h> |
52 | #include <sys/stat.h> | |
19cb3738 | 53 | #include <sys/ioctl.h> |
05acda4d | 54 | #include <sys/param.h> |
19cb3738 FB |
55 | #include <linux/cdrom.h> |
56 | #include <linux/fd.h> | |
5500316d | 57 | #include <linux/fs.h> |
4ab15590 CL |
58 | #ifndef FS_NOCOW_FL |
59 | #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ | |
60 | #endif | |
5500316d PB |
61 | #endif |
62 | #ifdef CONFIG_FIEMAP | |
63 | #include <linux/fiemap.h> | |
19cb3738 | 64 | #endif |
3d4fa43e KK |
65 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
66 | #include <linux/falloc.h> | |
67 | #endif | |
a167ba50 | 68 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
1cb6c3fd | 69 | #include <sys/disk.h> |
9f23011a | 70 | #include <sys/cdio.h> |
1cb6c3fd | 71 | #endif |
83f64091 | 72 | |
128ab2ff BS |
73 | #ifdef __OpenBSD__ |
74 | #include <sys/ioctl.h> | |
75 | #include <sys/disklabel.h> | |
76 | #include <sys/dkio.h> | |
77 | #endif | |
78 | ||
d1f6fd8d CE |
79 | #ifdef __NetBSD__ |
80 | #include <sys/ioctl.h> | |
81 | #include <sys/disklabel.h> | |
82 | #include <sys/dkio.h> | |
83 | #include <sys/disk.h> | |
84 | #endif | |
85 | ||
c5e97233 BS |
86 | #ifdef __DragonFly__ |
87 | #include <sys/ioctl.h> | |
88 | #include <sys/diskslice.h> | |
89 | #endif | |
90 | ||
dce512de CH |
91 | #ifdef CONFIG_XFS |
92 | #include <xfs/xfs.h> | |
93 | #endif | |
94 | ||
19cb3738 | 95 | //#define DEBUG_FLOPPY |
83f64091 | 96 | |
faf07963 | 97 | //#define DEBUG_BLOCK |
03ff3ca3 | 98 | #if defined(DEBUG_BLOCK) |
001faf32 BS |
99 | #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \ |
100 | { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0) | |
8c05dbf9 | 101 | #else |
001faf32 | 102 | #define DEBUG_BLOCK_PRINT(formatCstr, ...) |
8c05dbf9 TS |
103 | #endif |
104 | ||
f6465578 AL |
105 | /* OS X does not have O_DSYNC */ |
106 | #ifndef O_DSYNC | |
1c27a8b3 | 107 | #ifdef O_SYNC |
7ab064d2 | 108 | #define O_DSYNC O_SYNC |
1c27a8b3 JA |
109 | #elif defined(O_FSYNC) |
110 | #define O_DSYNC O_FSYNC | |
111 | #endif | |
f6465578 AL |
112 | #endif |
113 | ||
9f7965c7 AL |
114 | /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ |
115 | #ifndef O_DIRECT | |
116 | #define O_DIRECT O_DSYNC | |
117 | #endif | |
118 | ||
19cb3738 FB |
119 | #define FTYPE_FILE 0 |
120 | #define FTYPE_CD 1 | |
121 | #define FTYPE_FD 2 | |
83f64091 | 122 | |
c57c846a | 123 | /* if the FD is not accessed during that time (in ns), we try to |
19cb3738 | 124 | reopen it to see if the disk has been changed */ |
c57c846a | 125 | #define FD_OPEN_TIMEOUT (1000000000) |
83f64091 | 126 | |
581b9e29 CH |
127 | #define MAX_BLOCKSIZE 4096 |
128 | ||
19cb3738 FB |
129 | typedef struct BDRVRawState { |
130 | int fd; | |
131 | int type; | |
0e1d8f4c | 132 | int open_flags; |
c25f53b0 PB |
133 | size_t buf_align; |
134 | ||
19cb3738 FB |
135 | #if defined(__linux__) |
136 | /* linux floppy specific */ | |
19cb3738 FB |
137 | int64_t fd_open_time; |
138 | int64_t fd_error_time; | |
139 | int fd_got_error; | |
140 | int fd_media_changed; | |
83f64091 | 141 | #endif |
e44bd6fc | 142 | #ifdef CONFIG_LINUX_AIO |
5c6c3a6c | 143 | int use_aio; |
1e5b9d2f | 144 | void *aio_ctx; |
e44bd6fc | 145 | #endif |
dce512de | 146 | #ifdef CONFIG_XFS |
260a82e5 | 147 | bool is_xfs:1; |
dce512de | 148 | #endif |
260a82e5 | 149 | bool has_discard:1; |
97a2ae34 | 150 | bool has_write_zeroes:1; |
260a82e5 | 151 | bool discard_zeroes:1; |
4f11aa8a HR |
152 | #ifdef CONFIG_FIEMAP |
153 | bool skip_fiemap; | |
154 | #endif | |
19cb3738 FB |
155 | } BDRVRawState; |
156 | ||
eeb6b45d JC |
157 | typedef struct BDRVRawReopenState { |
158 | int fd; | |
159 | int open_flags; | |
160 | #ifdef CONFIG_LINUX_AIO | |
161 | int use_aio; | |
162 | #endif | |
163 | } BDRVRawReopenState; | |
164 | ||
19cb3738 | 165 | static int fd_open(BlockDriverState *bs); |
22afa7b5 | 166 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 167 | |
de81a169 PB |
168 | typedef struct RawPosixAIOData { |
169 | BlockDriverState *bs; | |
170 | int aio_fildes; | |
171 | union { | |
172 | struct iovec *aio_iov; | |
173 | void *aio_ioctl_buf; | |
174 | }; | |
175 | int aio_niov; | |
8238010b | 176 | uint64_t aio_nbytes; |
de81a169 PB |
177 | #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */ |
178 | off_t aio_offset; | |
179 | int aio_type; | |
180 | } RawPosixAIOData; | |
181 | ||
a167ba50 | 182 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 183 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
184 | #endif |
185 | ||
1de1ae0a CE |
186 | #if defined(__NetBSD__) |
187 | static int raw_normalize_devicepath(const char **filename) | |
188 | { | |
189 | static char namebuf[PATH_MAX]; | |
190 | const char *dp, *fname; | |
191 | struct stat sb; | |
192 | ||
193 | fname = *filename; | |
194 | dp = strrchr(fname, '/'); | |
195 | if (lstat(fname, &sb) < 0) { | |
196 | fprintf(stderr, "%s: stat failed: %s\n", | |
197 | fname, strerror(errno)); | |
198 | return -errno; | |
199 | } | |
200 | ||
201 | if (!S_ISBLK(sb.st_mode)) { | |
202 | return 0; | |
203 | } | |
204 | ||
205 | if (dp == NULL) { | |
206 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
207 | } else { | |
208 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
209 | (int)(dp - fname), fname, dp + 1); | |
210 | } | |
211 | fprintf(stderr, "%s is a block device", fname); | |
212 | *filename = namebuf; | |
213 | fprintf(stderr, ", using %s\n", *filename); | |
214 | ||
215 | return 0; | |
216 | } | |
217 | #else | |
218 | static int raw_normalize_devicepath(const char **filename) | |
219 | { | |
220 | return 0; | |
221 | } | |
222 | #endif | |
223 | ||
df26a350 | 224 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
c25f53b0 PB |
225 | { |
226 | BDRVRawState *s = bs->opaque; | |
227 | char *buf; | |
228 | unsigned int sector_size; | |
229 | ||
230 | /* For /dev/sg devices the alignment is not really used. | |
231 | With buffered I/O, we don't have any restrictions. */ | |
232 | if (bs->sg || !(s->open_flags & O_DIRECT)) { | |
233 | bs->request_alignment = 1; | |
234 | s->buf_align = 1; | |
235 | return; | |
236 | } | |
237 | ||
238 | /* Try a few ioctls to get the right size */ | |
239 | bs->request_alignment = 0; | |
240 | s->buf_align = 0; | |
241 | ||
242 | #ifdef BLKSSZGET | |
df26a350 | 243 | if (ioctl(fd, BLKSSZGET, §or_size) >= 0) { |
c25f53b0 PB |
244 | bs->request_alignment = sector_size; |
245 | } | |
246 | #endif | |
247 | #ifdef DKIOCGETBLOCKSIZE | |
df26a350 | 248 | if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) { |
c25f53b0 PB |
249 | bs->request_alignment = sector_size; |
250 | } | |
251 | #endif | |
252 | #ifdef DIOCGSECTORSIZE | |
df26a350 | 253 | if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) { |
c25f53b0 PB |
254 | bs->request_alignment = sector_size; |
255 | } | |
256 | #endif | |
257 | #ifdef CONFIG_XFS | |
258 | if (s->is_xfs) { | |
259 | struct dioattr da; | |
df26a350 | 260 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
c25f53b0 PB |
261 | bs->request_alignment = da.d_miniosz; |
262 | /* The kernel returns wrong information for d_mem */ | |
263 | /* s->buf_align = da.d_mem; */ | |
264 | } | |
265 | } | |
266 | #endif | |
267 | ||
268 | /* If we could not get the sizes so far, we can only guess them */ | |
269 | if (!s->buf_align) { | |
270 | size_t align; | |
271 | buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE); | |
272 | for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) { | |
df26a350 | 273 | if (pread(fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) { |
c25f53b0 PB |
274 | s->buf_align = align; |
275 | break; | |
276 | } | |
277 | } | |
278 | qemu_vfree(buf); | |
279 | } | |
280 | ||
281 | if (!bs->request_alignment) { | |
282 | size_t align; | |
283 | buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE); | |
284 | for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) { | |
df26a350 | 285 | if (pread(fd, buf, align, 0) >= 0) { |
c25f53b0 PB |
286 | bs->request_alignment = align; |
287 | break; | |
288 | } | |
289 | } | |
290 | qemu_vfree(buf); | |
291 | } | |
df26a350 KW |
292 | |
293 | if (!s->buf_align || !bs->request_alignment) { | |
294 | error_setg(errp, "Could not find working O_DIRECT alignment. " | |
295 | "Try cache.direct=off."); | |
296 | } | |
c25f53b0 PB |
297 | } |
298 | ||
6a8dc042 JC |
299 | static void raw_parse_flags(int bdrv_flags, int *open_flags) |
300 | { | |
301 | assert(open_flags != NULL); | |
302 | ||
303 | *open_flags |= O_BINARY; | |
304 | *open_flags &= ~O_ACCMODE; | |
305 | if (bdrv_flags & BDRV_O_RDWR) { | |
306 | *open_flags |= O_RDWR; | |
307 | } else { | |
308 | *open_flags |= O_RDONLY; | |
309 | } | |
310 | ||
311 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
312 | * and O_DIRECT for no caching. */ | |
313 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
314 | *open_flags |= O_DIRECT; | |
315 | } | |
6a8dc042 JC |
316 | } |
317 | ||
c2f3426c SH |
318 | static void raw_detach_aio_context(BlockDriverState *bs) |
319 | { | |
320 | #ifdef CONFIG_LINUX_AIO | |
321 | BDRVRawState *s = bs->opaque; | |
322 | ||
323 | if (s->use_aio) { | |
324 | laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs)); | |
325 | } | |
326 | #endif | |
327 | } | |
328 | ||
329 | static void raw_attach_aio_context(BlockDriverState *bs, | |
330 | AioContext *new_context) | |
331 | { | |
332 | #ifdef CONFIG_LINUX_AIO | |
333 | BDRVRawState *s = bs->opaque; | |
334 | ||
335 | if (s->use_aio) { | |
336 | laio_attach_aio_context(s->aio_ctx, new_context); | |
337 | } | |
338 | #endif | |
339 | } | |
340 | ||
fc32a72d JC |
341 | #ifdef CONFIG_LINUX_AIO |
342 | static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags) | |
343 | { | |
344 | int ret = -1; | |
345 | assert(aio_ctx != NULL); | |
346 | assert(use_aio != NULL); | |
347 | /* | |
348 | * Currently Linux do AIO only for files opened with O_DIRECT | |
349 | * specified so check NOCACHE flag too | |
350 | */ | |
351 | if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) == | |
352 | (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) { | |
353 | ||
354 | /* if non-NULL, laio_init() has already been run */ | |
355 | if (*aio_ctx == NULL) { | |
356 | *aio_ctx = laio_init(); | |
357 | if (!*aio_ctx) { | |
358 | goto error; | |
359 | } | |
360 | } | |
361 | *use_aio = 1; | |
362 | } else { | |
363 | *use_aio = 0; | |
364 | } | |
365 | ||
366 | ret = 0; | |
367 | ||
368 | error: | |
369 | return ret; | |
370 | } | |
371 | #endif | |
372 | ||
078896a9 HR |
373 | static void raw_parse_filename(const char *filename, QDict *options, |
374 | Error **errp) | |
375 | { | |
376 | /* The filename does not have to be prefixed by the protocol name, since | |
377 | * "file" is the default protocol; therefore, the return value of this | |
378 | * function call can be ignored. */ | |
379 | strstart(filename, "file:", &filename); | |
380 | ||
381 | qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename))); | |
382 | } | |
383 | ||
c66a6157 KW |
384 | static QemuOptsList raw_runtime_opts = { |
385 | .name = "raw", | |
386 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
387 | .desc = { | |
388 | { | |
389 | .name = "filename", | |
390 | .type = QEMU_OPT_STRING, | |
391 | .help = "File name of the image", | |
392 | }, | |
393 | { /* end of list */ } | |
394 | }, | |
395 | }; | |
396 | ||
397 | static int raw_open_common(BlockDriverState *bs, QDict *options, | |
e428e439 | 398 | int bdrv_flags, int open_flags, Error **errp) |
83f64091 FB |
399 | { |
400 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
401 | QemuOpts *opts; |
402 | Error *local_err = NULL; | |
8bfea15d | 403 | const char *filename = NULL; |
0e1d8f4c | 404 | int fd, ret; |
260a82e5 | 405 | struct stat st; |
83f64091 | 406 | |
87ea75d5 | 407 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
c66a6157 | 408 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 409 | if (local_err) { |
e428e439 | 410 | error_propagate(errp, local_err); |
c66a6157 KW |
411 | ret = -EINVAL; |
412 | goto fail; | |
413 | } | |
414 | ||
415 | filename = qemu_opt_get(opts, "filename"); | |
416 | ||
1de1ae0a CE |
417 | ret = raw_normalize_devicepath(&filename); |
418 | if (ret != 0) { | |
e428e439 | 419 | error_setg_errno(errp, -ret, "Could not normalize device path"); |
c66a6157 | 420 | goto fail; |
1de1ae0a CE |
421 | } |
422 | ||
6a8dc042 JC |
423 | s->open_flags = open_flags; |
424 | raw_parse_flags(bdrv_flags, &s->open_flags); | |
83f64091 | 425 | |
90babde0 | 426 | s->fd = -1; |
40ff6d7e | 427 | fd = qemu_open(filename, s->open_flags, 0644); |
19cb3738 FB |
428 | if (fd < 0) { |
429 | ret = -errno; | |
c66a6157 | 430 | if (ret == -EROFS) { |
19cb3738 | 431 | ret = -EACCES; |
c66a6157 KW |
432 | } |
433 | goto fail; | |
19cb3738 | 434 | } |
83f64091 | 435 | s->fd = fd; |
9ef91a67 | 436 | |
5c6c3a6c | 437 | #ifdef CONFIG_LINUX_AIO |
fc32a72d | 438 | if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) { |
47e6b251 | 439 | qemu_close(fd); |
c66a6157 | 440 | ret = -errno; |
e428e439 | 441 | error_setg_errno(errp, -ret, "Could not set AIO state"); |
c66a6157 | 442 | goto fail; |
9ef91a67 | 443 | } |
fc32a72d | 444 | #endif |
9ef91a67 | 445 | |
7ce21016 | 446 | s->has_discard = true; |
97a2ae34 | 447 | s->has_write_zeroes = true; |
260a82e5 PB |
448 | |
449 | if (fstat(s->fd, &st) < 0) { | |
450 | error_setg_errno(errp, errno, "Could not stat file"); | |
451 | goto fail; | |
452 | } | |
453 | if (S_ISREG(st.st_mode)) { | |
454 | s->discard_zeroes = true; | |
455 | } | |
d0b4503e PB |
456 | if (S_ISBLK(st.st_mode)) { |
457 | #ifdef BLKDISCARDZEROES | |
458 | unsigned int arg; | |
459 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
460 | s->discard_zeroes = true; | |
461 | } | |
462 | #endif | |
463 | #ifdef __linux__ | |
464 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
465 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 466 | * Same for BLKZEROOUT. |
d0b4503e PB |
467 | */ |
468 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
469 | s->discard_zeroes = false; | |
97a2ae34 | 470 | s->has_write_zeroes = false; |
d0b4503e PB |
471 | } |
472 | #endif | |
473 | } | |
260a82e5 | 474 | |
dce512de CH |
475 | #ifdef CONFIG_XFS |
476 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 477 | s->is_xfs = true; |
dce512de CH |
478 | } |
479 | #endif | |
480 | ||
c2f3426c SH |
481 | raw_attach_aio_context(bs, bdrv_get_aio_context(bs)); |
482 | ||
c66a6157 KW |
483 | ret = 0; |
484 | fail: | |
8bfea15d KW |
485 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
486 | unlink(filename); | |
487 | } | |
c66a6157 KW |
488 | qemu_opts_del(opts); |
489 | return ret; | |
83f64091 FB |
490 | } |
491 | ||
015a1036 HR |
492 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
493 | Error **errp) | |
90babde0 CH |
494 | { |
495 | BDRVRawState *s = bs->opaque; | |
e428e439 HR |
496 | Error *local_err = NULL; |
497 | int ret; | |
90babde0 CH |
498 | |
499 | s->type = FTYPE_FILE; | |
e428e439 | 500 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
84d18f06 | 501 | if (local_err) { |
e428e439 HR |
502 | error_propagate(errp, local_err); |
503 | } | |
504 | return ret; | |
90babde0 CH |
505 | } |
506 | ||
eeb6b45d JC |
507 | static int raw_reopen_prepare(BDRVReopenState *state, |
508 | BlockReopenQueue *queue, Error **errp) | |
509 | { | |
510 | BDRVRawState *s; | |
511 | BDRVRawReopenState *raw_s; | |
512 | int ret = 0; | |
df26a350 | 513 | Error *local_err = NULL; |
eeb6b45d JC |
514 | |
515 | assert(state != NULL); | |
516 | assert(state->bs != NULL); | |
517 | ||
518 | s = state->bs->opaque; | |
519 | ||
520 | state->opaque = g_malloc0(sizeof(BDRVRawReopenState)); | |
521 | raw_s = state->opaque; | |
522 | ||
523 | #ifdef CONFIG_LINUX_AIO | |
524 | raw_s->use_aio = s->use_aio; | |
525 | ||
526 | /* we can use s->aio_ctx instead of a copy, because the use_aio flag is | |
527 | * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio() | |
528 | * won't override aio_ctx if aio_ctx is non-NULL */ | |
529 | if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) { | |
e428e439 | 530 | error_setg(errp, "Could not set AIO state"); |
eeb6b45d JC |
531 | return -1; |
532 | } | |
533 | #endif | |
534 | ||
1bc6b705 JC |
535 | if (s->type == FTYPE_FD || s->type == FTYPE_CD) { |
536 | raw_s->open_flags |= O_NONBLOCK; | |
537 | } | |
538 | ||
eeb6b45d JC |
539 | raw_parse_flags(state->flags, &raw_s->open_flags); |
540 | ||
541 | raw_s->fd = -1; | |
542 | ||
fdf263f6 | 543 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
eeb6b45d JC |
544 | #ifdef O_NOATIME |
545 | fcntl_flags |= O_NOATIME; | |
546 | #endif | |
547 | ||
fdf263f6 AF |
548 | #ifdef O_ASYNC |
549 | /* Not all operating systems have O_ASYNC, and those that don't | |
550 | * will not let us track the state into raw_s->open_flags (typically | |
551 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
552 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
553 | */ | |
554 | assert((s->open_flags & O_ASYNC) == 0); | |
555 | #endif | |
556 | ||
eeb6b45d JC |
557 | if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
558 | /* dup the original fd */ | |
559 | /* TODO: use qemu fcntl wrapper */ | |
560 | #ifdef F_DUPFD_CLOEXEC | |
561 | raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0); | |
562 | #else | |
563 | raw_s->fd = dup(s->fd); | |
564 | if (raw_s->fd != -1) { | |
565 | qemu_set_cloexec(raw_s->fd); | |
566 | } | |
567 | #endif | |
568 | if (raw_s->fd >= 0) { | |
569 | ret = fcntl_setfl(raw_s->fd, raw_s->open_flags); | |
570 | if (ret) { | |
571 | qemu_close(raw_s->fd); | |
572 | raw_s->fd = -1; | |
573 | } | |
574 | } | |
575 | } | |
576 | ||
577 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ | |
578 | if (raw_s->fd == -1) { | |
579 | assert(!(raw_s->open_flags & O_CREAT)); | |
580 | raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags); | |
581 | if (raw_s->fd == -1) { | |
e428e439 | 582 | error_setg_errno(errp, errno, "Could not reopen file"); |
eeb6b45d JC |
583 | ret = -1; |
584 | } | |
585 | } | |
df26a350 KW |
586 | |
587 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
588 | * alignment with the new fd. */ | |
589 | if (raw_s->fd != -1) { | |
590 | raw_probe_alignment(state->bs, raw_s->fd, &local_err); | |
591 | if (local_err) { | |
592 | qemu_close(raw_s->fd); | |
593 | raw_s->fd = -1; | |
594 | error_propagate(errp, local_err); | |
595 | ret = -EINVAL; | |
596 | } | |
597 | } | |
598 | ||
eeb6b45d JC |
599 | return ret; |
600 | } | |
601 | ||
eeb6b45d JC |
602 | static void raw_reopen_commit(BDRVReopenState *state) |
603 | { | |
604 | BDRVRawReopenState *raw_s = state->opaque; | |
605 | BDRVRawState *s = state->bs->opaque; | |
606 | ||
607 | s->open_flags = raw_s->open_flags; | |
608 | ||
609 | qemu_close(s->fd); | |
610 | s->fd = raw_s->fd; | |
611 | #ifdef CONFIG_LINUX_AIO | |
612 | s->use_aio = raw_s->use_aio; | |
613 | #endif | |
614 | ||
615 | g_free(state->opaque); | |
616 | state->opaque = NULL; | |
617 | } | |
618 | ||
619 | ||
620 | static void raw_reopen_abort(BDRVReopenState *state) | |
621 | { | |
622 | BDRVRawReopenState *raw_s = state->opaque; | |
623 | ||
624 | /* nothing to do if NULL, we didn't get far enough */ | |
625 | if (raw_s == NULL) { | |
626 | return; | |
627 | } | |
628 | ||
629 | if (raw_s->fd >= 0) { | |
630 | qemu_close(raw_s->fd); | |
631 | raw_s->fd = -1; | |
632 | } | |
633 | g_free(state->opaque); | |
634 | state->opaque = NULL; | |
635 | } | |
636 | ||
3baca891 | 637 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
638 | { |
639 | BDRVRawState *s = bs->opaque; | |
eeb6b45d | 640 | |
df26a350 | 641 | raw_probe_alignment(bs, s->fd, errp); |
c25f53b0 | 642 | bs->bl.opt_mem_alignment = s->buf_align; |
c25f53b0 | 643 | } |
83f64091 | 644 | |
de81a169 PB |
645 | static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb) |
646 | { | |
647 | int ret; | |
648 | ||
649 | ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); | |
650 | if (ret == -1) { | |
651 | return -errno; | |
652 | } | |
653 | ||
b608c8dc | 654 | return 0; |
de81a169 PB |
655 | } |
656 | ||
657 | static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb) | |
658 | { | |
659 | int ret; | |
660 | ||
661 | ret = qemu_fdatasync(aiocb->aio_fildes); | |
662 | if (ret == -1) { | |
663 | return -errno; | |
664 | } | |
665 | return 0; | |
666 | } | |
667 | ||
668 | #ifdef CONFIG_PREADV | |
669 | ||
670 | static bool preadv_present = true; | |
671 | ||
672 | static ssize_t | |
673 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
674 | { | |
675 | return preadv(fd, iov, nr_iov, offset); | |
676 | } | |
677 | ||
678 | static ssize_t | |
679 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
680 | { | |
681 | return pwritev(fd, iov, nr_iov, offset); | |
682 | } | |
683 | ||
684 | #else | |
685 | ||
686 | static bool preadv_present = false; | |
687 | ||
688 | static ssize_t | |
689 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
690 | { | |
691 | return -ENOSYS; | |
692 | } | |
693 | ||
694 | static ssize_t | |
695 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
696 | { | |
697 | return -ENOSYS; | |
698 | } | |
699 | ||
700 | #endif | |
701 | ||
702 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
703 | { | |
704 | ssize_t len; | |
705 | ||
706 | do { | |
707 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
708 | len = qemu_pwritev(aiocb->aio_fildes, | |
709 | aiocb->aio_iov, | |
710 | aiocb->aio_niov, | |
711 | aiocb->aio_offset); | |
712 | else | |
713 | len = qemu_preadv(aiocb->aio_fildes, | |
714 | aiocb->aio_iov, | |
715 | aiocb->aio_niov, | |
716 | aiocb->aio_offset); | |
717 | } while (len == -1 && errno == EINTR); | |
718 | ||
719 | if (len == -1) { | |
720 | return -errno; | |
721 | } | |
722 | return len; | |
723 | } | |
724 | ||
725 | /* | |
726 | * Read/writes the data to/from a given linear buffer. | |
727 | * | |
728 | * Returns the number of bytes handles or -errno in case of an error. Short | |
729 | * reads are only returned if the end of the file is reached. | |
730 | */ | |
731 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
732 | { | |
733 | ssize_t offset = 0; | |
734 | ssize_t len; | |
735 | ||
736 | while (offset < aiocb->aio_nbytes) { | |
737 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
738 | len = pwrite(aiocb->aio_fildes, | |
739 | (const char *)buf + offset, | |
740 | aiocb->aio_nbytes - offset, | |
741 | aiocb->aio_offset + offset); | |
742 | } else { | |
743 | len = pread(aiocb->aio_fildes, | |
744 | buf + offset, | |
745 | aiocb->aio_nbytes - offset, | |
746 | aiocb->aio_offset + offset); | |
747 | } | |
748 | if (len == -1 && errno == EINTR) { | |
749 | continue; | |
750 | } else if (len == -1) { | |
751 | offset = -errno; | |
752 | break; | |
753 | } else if (len == 0) { | |
754 | break; | |
755 | } | |
756 | offset += len; | |
757 | } | |
758 | ||
759 | return offset; | |
760 | } | |
761 | ||
762 | static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb) | |
763 | { | |
764 | ssize_t nbytes; | |
765 | char *buf; | |
766 | ||
767 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
768 | /* | |
769 | * If there is just a single buffer, and it is properly aligned | |
770 | * we can just use plain pread/pwrite without any problems. | |
771 | */ | |
772 | if (aiocb->aio_niov == 1) { | |
773 | return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); | |
774 | } | |
775 | /* | |
776 | * We have more than one iovec, and all are properly aligned. | |
777 | * | |
778 | * Try preadv/pwritev first and fall back to linearizing the | |
779 | * buffer if it's not supported. | |
780 | */ | |
781 | if (preadv_present) { | |
782 | nbytes = handle_aiocb_rw_vector(aiocb); | |
783 | if (nbytes == aiocb->aio_nbytes || | |
784 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
785 | return nbytes; | |
786 | } | |
787 | preadv_present = false; | |
788 | } | |
789 | ||
790 | /* | |
791 | * XXX(hch): short read/write. no easy way to handle the reminder | |
792 | * using these interfaces. For now retry using plain | |
793 | * pread/pwrite? | |
794 | */ | |
795 | } | |
796 | ||
797 | /* | |
798 | * Ok, we have to do it the hard way, copy all segments into | |
799 | * a single aligned buffer. | |
800 | */ | |
801 | buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes); | |
802 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
803 | char *p = buf; | |
804 | int i; | |
805 | ||
806 | for (i = 0; i < aiocb->aio_niov; ++i) { | |
807 | memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); | |
808 | p += aiocb->aio_iov[i].iov_len; | |
809 | } | |
8eb029c2 | 810 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
811 | } |
812 | ||
813 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
814 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
815 | char *p = buf; | |
816 | size_t count = aiocb->aio_nbytes, copy; | |
817 | int i; | |
818 | ||
819 | for (i = 0; i < aiocb->aio_niov && count; ++i) { | |
820 | copy = count; | |
821 | if (copy > aiocb->aio_iov[i].iov_len) { | |
822 | copy = aiocb->aio_iov[i].iov_len; | |
823 | } | |
824 | memcpy(aiocb->aio_iov[i].iov_base, p, copy); | |
8eb029c2 | 825 | assert(count >= copy); |
de81a169 PB |
826 | p += copy; |
827 | count -= copy; | |
828 | } | |
8eb029c2 | 829 | assert(count == 0); |
de81a169 PB |
830 | } |
831 | qemu_vfree(buf); | |
832 | ||
833 | return nbytes; | |
834 | } | |
835 | ||
8238010b | 836 | #ifdef CONFIG_XFS |
97a2ae34 PB |
837 | static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) |
838 | { | |
839 | struct xfs_flock64 fl; | |
840 | ||
841 | memset(&fl, 0, sizeof(fl)); | |
842 | fl.l_whence = SEEK_SET; | |
843 | fl.l_start = offset; | |
844 | fl.l_len = bytes; | |
845 | ||
846 | if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { | |
847 | DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno)); | |
848 | return -errno; | |
849 | } | |
850 | ||
851 | return 0; | |
852 | } | |
853 | ||
8238010b PB |
854 | static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) |
855 | { | |
856 | struct xfs_flock64 fl; | |
857 | ||
858 | memset(&fl, 0, sizeof(fl)); | |
859 | fl.l_whence = SEEK_SET; | |
860 | fl.l_start = offset; | |
861 | fl.l_len = bytes; | |
862 | ||
863 | if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { | |
864 | DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno)); | |
865 | return -errno; | |
866 | } | |
867 | ||
868 | return 0; | |
869 | } | |
870 | #endif | |
871 | ||
97a2ae34 PB |
872 | static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb) |
873 | { | |
874 | int ret = -EOPNOTSUPP; | |
875 | BDRVRawState *s = aiocb->bs->opaque; | |
876 | ||
877 | if (s->has_write_zeroes == 0) { | |
878 | return -ENOTSUP; | |
879 | } | |
880 | ||
881 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
882 | #ifdef BLKZEROOUT | |
883 | do { | |
884 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
885 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
886 | return 0; | |
887 | } | |
888 | } while (errno == EINTR); | |
889 | ||
890 | ret = -errno; | |
891 | #endif | |
892 | } else { | |
893 | #ifdef CONFIG_XFS | |
894 | if (s->is_xfs) { | |
895 | return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
896 | } | |
897 | #endif | |
898 | } | |
899 | ||
900 | if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP || | |
901 | ret == -ENOTTY) { | |
902 | s->has_write_zeroes = false; | |
903 | ret = -ENOTSUP; | |
904 | } | |
905 | return ret; | |
906 | } | |
907 | ||
8238010b PB |
908 | static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb) |
909 | { | |
910 | int ret = -EOPNOTSUPP; | |
911 | BDRVRawState *s = aiocb->bs->opaque; | |
912 | ||
7ce21016 PB |
913 | if (!s->has_discard) { |
914 | return -ENOTSUP; | |
8238010b PB |
915 | } |
916 | ||
917 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
918 | #ifdef BLKDISCARD | |
919 | do { | |
920 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
921 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
922 | return 0; | |
923 | } | |
924 | } while (errno == EINTR); | |
925 | ||
926 | ret = -errno; | |
927 | #endif | |
928 | } else { | |
929 | #ifdef CONFIG_XFS | |
930 | if (s->is_xfs) { | |
931 | return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
932 | } | |
933 | #endif | |
934 | ||
935 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
936 | do { | |
937 | if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
938 | aiocb->aio_offset, aiocb->aio_nbytes) == 0) { | |
939 | return 0; | |
940 | } | |
941 | } while (errno == EINTR); | |
942 | ||
943 | ret = -errno; | |
944 | #endif | |
945 | } | |
946 | ||
947 | if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP || | |
948 | ret == -ENOTTY) { | |
7ce21016 PB |
949 | s->has_discard = false; |
950 | ret = -ENOTSUP; | |
8238010b PB |
951 | } |
952 | return ret; | |
953 | } | |
954 | ||
de81a169 PB |
955 | static int aio_worker(void *arg) |
956 | { | |
957 | RawPosixAIOData *aiocb = arg; | |
958 | ssize_t ret = 0; | |
959 | ||
960 | switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { | |
961 | case QEMU_AIO_READ: | |
962 | ret = handle_aiocb_rw(aiocb); | |
963 | if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) { | |
964 | iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret, | |
965 | 0, aiocb->aio_nbytes - ret); | |
966 | ||
967 | ret = aiocb->aio_nbytes; | |
968 | } | |
969 | if (ret == aiocb->aio_nbytes) { | |
970 | ret = 0; | |
971 | } else if (ret >= 0 && ret < aiocb->aio_nbytes) { | |
972 | ret = -EINVAL; | |
973 | } | |
974 | break; | |
975 | case QEMU_AIO_WRITE: | |
976 | ret = handle_aiocb_rw(aiocb); | |
977 | if (ret == aiocb->aio_nbytes) { | |
978 | ret = 0; | |
979 | } else if (ret >= 0 && ret < aiocb->aio_nbytes) { | |
980 | ret = -EINVAL; | |
981 | } | |
982 | break; | |
983 | case QEMU_AIO_FLUSH: | |
984 | ret = handle_aiocb_flush(aiocb); | |
985 | break; | |
986 | case QEMU_AIO_IOCTL: | |
987 | ret = handle_aiocb_ioctl(aiocb); | |
988 | break; | |
8238010b PB |
989 | case QEMU_AIO_DISCARD: |
990 | ret = handle_aiocb_discard(aiocb); | |
991 | break; | |
97a2ae34 PB |
992 | case QEMU_AIO_WRITE_ZEROES: |
993 | ret = handle_aiocb_write_zeroes(aiocb); | |
994 | break; | |
de81a169 PB |
995 | default: |
996 | fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); | |
997 | ret = -EINVAL; | |
998 | break; | |
999 | } | |
1000 | ||
1001 | g_slice_free(RawPosixAIOData, aiocb); | |
1002 | return ret; | |
1003 | } | |
1004 | ||
260a82e5 PB |
1005 | static int paio_submit_co(BlockDriverState *bs, int fd, |
1006 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
1007 | int type) | |
1008 | { | |
1009 | RawPosixAIOData *acb = g_slice_new(RawPosixAIOData); | |
1010 | ThreadPool *pool; | |
1011 | ||
1012 | acb->bs = bs; | |
1013 | acb->aio_type = type; | |
1014 | acb->aio_fildes = fd; | |
1015 | ||
8eb029c2 KW |
1016 | acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE; |
1017 | acb->aio_offset = sector_num * BDRV_SECTOR_SIZE; | |
1018 | ||
260a82e5 PB |
1019 | if (qiov) { |
1020 | acb->aio_iov = qiov->iov; | |
1021 | acb->aio_niov = qiov->niov; | |
8eb029c2 | 1022 | assert(qiov->size == acb->aio_nbytes); |
260a82e5 | 1023 | } |
260a82e5 PB |
1024 | |
1025 | trace_paio_submit_co(sector_num, nb_sectors, type); | |
1026 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
1027 | return thread_pool_submit_co(pool, aio_worker, acb); | |
1028 | } | |
1029 | ||
de81a169 PB |
1030 | static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd, |
1031 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
1032 | BlockDriverCompletionFunc *cb, void *opaque, int type) | |
1033 | { | |
1034 | RawPosixAIOData *acb = g_slice_new(RawPosixAIOData); | |
c4d9d196 | 1035 | ThreadPool *pool; |
de81a169 PB |
1036 | |
1037 | acb->bs = bs; | |
1038 | acb->aio_type = type; | |
1039 | acb->aio_fildes = fd; | |
1040 | ||
8eb029c2 KW |
1041 | acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE; |
1042 | acb->aio_offset = sector_num * BDRV_SECTOR_SIZE; | |
1043 | ||
de81a169 PB |
1044 | if (qiov) { |
1045 | acb->aio_iov = qiov->iov; | |
1046 | acb->aio_niov = qiov->niov; | |
8eb029c2 | 1047 | assert(qiov->size == acb->aio_nbytes); |
de81a169 | 1048 | } |
de81a169 PB |
1049 | |
1050 | trace_paio_submit(acb, opaque, sector_num, nb_sectors, type); | |
c4d9d196 SH |
1051 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
1052 | return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); | |
de81a169 PB |
1053 | } |
1054 | ||
9ef91a67 CH |
1055 | static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs, |
1056 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
1057 | BlockDriverCompletionFunc *cb, void *opaque, int type) | |
83f64091 | 1058 | { |
ce1a14dc | 1059 | BDRVRawState *s = bs->opaque; |
ce1a14dc | 1060 | |
19cb3738 FB |
1061 | if (fd_open(bs) < 0) |
1062 | return NULL; | |
1063 | ||
f141eafe AL |
1064 | /* |
1065 | * If O_DIRECT is used the buffer needs to be aligned on a sector | |
c1ee7d56 | 1066 | * boundary. Check if this is the case or tell the low-level |
9ef91a67 | 1067 | * driver that it needs to copy the buffer. |
f141eafe | 1068 | */ |
9acc5a06 | 1069 | if ((bs->open_flags & BDRV_O_NOCACHE)) { |
c53b1c51 | 1070 | if (!bdrv_qiov_is_aligned(bs, qiov)) { |
5c6c3a6c | 1071 | type |= QEMU_AIO_MISALIGNED; |
e44bd6fc | 1072 | #ifdef CONFIG_LINUX_AIO |
5c6c3a6c CH |
1073 | } else if (s->use_aio) { |
1074 | return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov, | |
e44bd6fc SW |
1075 | nb_sectors, cb, opaque, type); |
1076 | #endif | |
5c6c3a6c | 1077 | } |
9ef91a67 | 1078 | } |
f141eafe | 1079 | |
1e5b9d2f | 1080 | return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors, |
9ef91a67 | 1081 | cb, opaque, type); |
83f64091 FB |
1082 | } |
1083 | ||
1b3abdcc ML |
1084 | static void raw_aio_plug(BlockDriverState *bs) |
1085 | { | |
1086 | #ifdef CONFIG_LINUX_AIO | |
1087 | BDRVRawState *s = bs->opaque; | |
1088 | if (s->use_aio) { | |
1089 | laio_io_plug(bs, s->aio_ctx); | |
1090 | } | |
1091 | #endif | |
1092 | } | |
1093 | ||
1094 | static void raw_aio_unplug(BlockDriverState *bs) | |
1095 | { | |
1096 | #ifdef CONFIG_LINUX_AIO | |
1097 | BDRVRawState *s = bs->opaque; | |
1098 | if (s->use_aio) { | |
1099 | laio_io_unplug(bs, s->aio_ctx, true); | |
1100 | } | |
1101 | #endif | |
1102 | } | |
1103 | ||
1104 | static void raw_aio_flush_io_queue(BlockDriverState *bs) | |
1105 | { | |
1106 | #ifdef CONFIG_LINUX_AIO | |
1107 | BDRVRawState *s = bs->opaque; | |
1108 | if (s->use_aio) { | |
1109 | laio_io_unplug(bs, s->aio_ctx, false); | |
1110 | } | |
1111 | #endif | |
1112 | } | |
1113 | ||
f141eafe AL |
1114 | static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs, |
1115 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
ce1a14dc | 1116 | BlockDriverCompletionFunc *cb, void *opaque) |
83f64091 | 1117 | { |
9ef91a67 CH |
1118 | return raw_aio_submit(bs, sector_num, qiov, nb_sectors, |
1119 | cb, opaque, QEMU_AIO_READ); | |
83f64091 FB |
1120 | } |
1121 | ||
f141eafe AL |
1122 | static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs, |
1123 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
ce1a14dc | 1124 | BlockDriverCompletionFunc *cb, void *opaque) |
83f64091 | 1125 | { |
9ef91a67 CH |
1126 | return raw_aio_submit(bs, sector_num, qiov, nb_sectors, |
1127 | cb, opaque, QEMU_AIO_WRITE); | |
83f64091 | 1128 | } |
53538725 | 1129 | |
b2e12bc6 CH |
1130 | static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs, |
1131 | BlockDriverCompletionFunc *cb, void *opaque) | |
1132 | { | |
1133 | BDRVRawState *s = bs->opaque; | |
1134 | ||
1135 | if (fd_open(bs) < 0) | |
1136 | return NULL; | |
1137 | ||
1e5b9d2f | 1138 | return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH); |
b2e12bc6 CH |
1139 | } |
1140 | ||
83f64091 FB |
1141 | static void raw_close(BlockDriverState *bs) |
1142 | { | |
1143 | BDRVRawState *s = bs->opaque; | |
c2f3426c SH |
1144 | |
1145 | raw_detach_aio_context(bs); | |
1146 | ||
abd269b7 SH |
1147 | #ifdef CONFIG_LINUX_AIO |
1148 | if (s->use_aio) { | |
1149 | laio_cleanup(s->aio_ctx); | |
1150 | } | |
1151 | #endif | |
19cb3738 | 1152 | if (s->fd >= 0) { |
2e1e79da | 1153 | qemu_close(s->fd); |
19cb3738 FB |
1154 | s->fd = -1; |
1155 | } | |
83f64091 FB |
1156 | } |
1157 | ||
1158 | static int raw_truncate(BlockDriverState *bs, int64_t offset) | |
1159 | { | |
1160 | BDRVRawState *s = bs->opaque; | |
55b949c8 CH |
1161 | struct stat st; |
1162 | ||
1163 | if (fstat(s->fd, &st)) { | |
83f64091 | 1164 | return -errno; |
55b949c8 CH |
1165 | } |
1166 | ||
1167 | if (S_ISREG(st.st_mode)) { | |
1168 | if (ftruncate(s->fd, offset) < 0) { | |
1169 | return -errno; | |
1170 | } | |
1171 | } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
1172 | if (offset > raw_getlength(bs)) { | |
1173 | return -EINVAL; | |
1174 | } | |
1175 | } else { | |
1176 | return -ENOTSUP; | |
1177 | } | |
1178 | ||
83f64091 FB |
1179 | return 0; |
1180 | } | |
1181 | ||
128ab2ff BS |
1182 | #ifdef __OpenBSD__ |
1183 | static int64_t raw_getlength(BlockDriverState *bs) | |
1184 | { | |
1185 | BDRVRawState *s = bs->opaque; | |
1186 | int fd = s->fd; | |
1187 | struct stat st; | |
1188 | ||
1189 | if (fstat(fd, &st)) | |
aa729704 | 1190 | return -errno; |
128ab2ff BS |
1191 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
1192 | struct disklabel dl; | |
1193 | ||
1194 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 1195 | return -errno; |
128ab2ff BS |
1196 | return (uint64_t)dl.d_secsize * |
1197 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
1198 | } else | |
1199 | return st.st_size; | |
1200 | } | |
d1f6fd8d CE |
1201 | #elif defined(__NetBSD__) |
1202 | static int64_t raw_getlength(BlockDriverState *bs) | |
1203 | { | |
1204 | BDRVRawState *s = bs->opaque; | |
1205 | int fd = s->fd; | |
1206 | struct stat st; | |
1207 | ||
1208 | if (fstat(fd, &st)) | |
aa729704 | 1209 | return -errno; |
d1f6fd8d CE |
1210 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
1211 | struct dkwedge_info dkw; | |
1212 | ||
1213 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
1214 | return dkw.dkw_size * 512; | |
1215 | } else { | |
1216 | struct disklabel dl; | |
1217 | ||
1218 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 1219 | return -errno; |
d1f6fd8d CE |
1220 | return (uint64_t)dl.d_secsize * |
1221 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
1222 | } | |
1223 | } else | |
1224 | return st.st_size; | |
1225 | } | |
50779cc2 CH |
1226 | #elif defined(__sun__) |
1227 | static int64_t raw_getlength(BlockDriverState *bs) | |
1228 | { | |
1229 | BDRVRawState *s = bs->opaque; | |
1230 | struct dk_minfo minfo; | |
1231 | int ret; | |
aa729704 | 1232 | int64_t size; |
50779cc2 CH |
1233 | |
1234 | ret = fd_open(bs); | |
1235 | if (ret < 0) { | |
1236 | return ret; | |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
1241 | */ | |
1242 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
1243 | if (ret != -1) { | |
1244 | return minfo.dki_lbsize * minfo.dki_capacity; | |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * There are reports that lseek on some devices fails, but | |
1249 | * irc discussion said that contingency on contingency was overkill. | |
1250 | */ | |
aa729704 MA |
1251 | size = lseek(s->fd, 0, SEEK_END); |
1252 | if (size < 0) { | |
1253 | return -errno; | |
1254 | } | |
1255 | return size; | |
50779cc2 CH |
1256 | } |
1257 | #elif defined(CONFIG_BSD) | |
1258 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
1259 | { |
1260 | BDRVRawState *s = bs->opaque; | |
1261 | int fd = s->fd; | |
1262 | int64_t size; | |
83f64091 | 1263 | struct stat sb; |
a167ba50 | 1264 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 1265 | int reopened = 0; |
83f64091 | 1266 | #endif |
19cb3738 FB |
1267 | int ret; |
1268 | ||
1269 | ret = fd_open(bs); | |
1270 | if (ret < 0) | |
1271 | return ret; | |
83f64091 | 1272 | |
a167ba50 | 1273 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
1274 | again: |
1275 | #endif | |
83f64091 FB |
1276 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
1277 | #ifdef DIOCGMEDIASIZE | |
1278 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) | |
c5e97233 BS |
1279 | #elif defined(DIOCGPART) |
1280 | { | |
1281 | struct partinfo pi; | |
1282 | if (ioctl(fd, DIOCGPART, &pi) == 0) | |
1283 | size = pi.media_size; | |
1284 | else | |
1285 | size = 0; | |
1286 | } | |
1287 | if (size == 0) | |
83f64091 | 1288 | #endif |
83affaa6 | 1289 | #if defined(__APPLE__) && defined(__MACH__) |
675036e4 | 1290 | size = LLONG_MAX; |
83f64091 FB |
1291 | #else |
1292 | size = lseek(fd, 0LL, SEEK_END); | |
aa729704 MA |
1293 | if (size < 0) { |
1294 | return -errno; | |
1295 | } | |
9f23011a | 1296 | #endif |
a167ba50 | 1297 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
1298 | switch(s->type) { |
1299 | case FTYPE_CD: | |
1300 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
1301 | if (size == 2048LL * (unsigned)-1) | |
1302 | size = 0; | |
1303 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 1304 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
1305 | reopened = 1; |
1306 | goto again; | |
1307 | } | |
1308 | } | |
83f64091 | 1309 | #endif |
50779cc2 | 1310 | } else { |
83f64091 | 1311 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
1312 | if (size < 0) { |
1313 | return -errno; | |
1314 | } | |
83f64091 | 1315 | } |
83f64091 FB |
1316 | return size; |
1317 | } | |
50779cc2 CH |
1318 | #else |
1319 | static int64_t raw_getlength(BlockDriverState *bs) | |
1320 | { | |
1321 | BDRVRawState *s = bs->opaque; | |
1322 | int ret; | |
aa729704 | 1323 | int64_t size; |
50779cc2 CH |
1324 | |
1325 | ret = fd_open(bs); | |
1326 | if (ret < 0) { | |
1327 | return ret; | |
1328 | } | |
1329 | ||
aa729704 MA |
1330 | size = lseek(s->fd, 0, SEEK_END); |
1331 | if (size < 0) { | |
1332 | return -errno; | |
1333 | } | |
1334 | return size; | |
50779cc2 | 1335 | } |
128ab2ff | 1336 | #endif |
83f64091 | 1337 | |
4a1d5e1f FZ |
1338 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
1339 | { | |
1340 | struct stat st; | |
1341 | BDRVRawState *s = bs->opaque; | |
1342 | ||
1343 | if (fstat(s->fd, &st) < 0) { | |
1344 | return -errno; | |
1345 | } | |
1346 | return (int64_t)st.st_blocks * 512; | |
1347 | } | |
1348 | ||
6f482f74 | 1349 | static int raw_create(const char *filename, QemuOpts *opts, Error **errp) |
83f64091 FB |
1350 | { |
1351 | int fd; | |
1e37d059 | 1352 | int result = 0; |
0e7e1989 | 1353 | int64_t total_size = 0; |
4ab15590 | 1354 | bool nocow = false; |
83f64091 | 1355 | |
464d9f64 HR |
1356 | strstart(filename, "file:", &filename); |
1357 | ||
0e7e1989 | 1358 | /* Read out options */ |
6f482f74 CL |
1359 | total_size = |
1360 | qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / BDRV_SECTOR_SIZE; | |
4ab15590 | 1361 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); |
83f64091 | 1362 | |
6165f4d8 CB |
1363 | fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, |
1364 | 0644); | |
1e37d059 SW |
1365 | if (fd < 0) { |
1366 | result = -errno; | |
e428e439 | 1367 | error_setg_errno(errp, -result, "Could not create file"); |
1e37d059 | 1368 | } else { |
4ab15590 CL |
1369 | if (nocow) { |
1370 | #ifdef __linux__ | |
1371 | /* Set NOCOW flag to solve performance issue on fs like btrfs. | |
1372 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
1373 | * will be ignored since any failure of this operation should not | |
1374 | * block the left work. | |
1375 | */ | |
1376 | int attr; | |
1377 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
1378 | attr |= FS_NOCOW_FL; | |
1379 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
1380 | } | |
1381 | #endif | |
1382 | } | |
1383 | ||
9040385d | 1384 | if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) { |
1e37d059 | 1385 | result = -errno; |
e428e439 | 1386 | error_setg_errno(errp, -result, "Could not resize file"); |
1e37d059 | 1387 | } |
2e1e79da | 1388 | if (qemu_close(fd) != 0) { |
1e37d059 | 1389 | result = -errno; |
e428e439 | 1390 | error_setg_errno(errp, -result, "Could not close the new file"); |
1e37d059 SW |
1391 | } |
1392 | } | |
1393 | return result; | |
83f64091 FB |
1394 | } |
1395 | ||
4f11aa8a HR |
1396 | static int64_t try_fiemap(BlockDriverState *bs, off_t start, off_t *data, |
1397 | off_t *hole, int nb_sectors, int *pnum) | |
5500316d | 1398 | { |
5500316d | 1399 | #ifdef CONFIG_FIEMAP |
94282e71 | 1400 | BDRVRawState *s = bs->opaque; |
4f11aa8a | 1401 | int64_t ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start; |
5500316d PB |
1402 | struct { |
1403 | struct fiemap fm; | |
1404 | struct fiemap_extent fe; | |
1405 | } f; | |
94282e71 | 1406 | |
4f11aa8a HR |
1407 | if (s->skip_fiemap) { |
1408 | return -ENOTSUP; | |
1409 | } | |
1410 | ||
5500316d PB |
1411 | f.fm.fm_start = start; |
1412 | f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE; | |
1413 | f.fm.fm_flags = 0; | |
1414 | f.fm.fm_extent_count = 1; | |
1415 | f.fm.fm_reserved = 0; | |
1416 | if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) { | |
4f11aa8a HR |
1417 | s->skip_fiemap = true; |
1418 | return -errno; | |
5500316d PB |
1419 | } |
1420 | ||
1421 | if (f.fm.fm_mapped_extents == 0) { | |
1422 | /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length. | |
1423 | * f.fm.fm_start + f.fm.fm_length must be clamped to the file size! | |
1424 | */ | |
1425 | off_t length = lseek(s->fd, 0, SEEK_END); | |
4f11aa8a HR |
1426 | *hole = f.fm.fm_start; |
1427 | *data = MIN(f.fm.fm_start + f.fm.fm_length, length); | |
5500316d | 1428 | } else { |
4f11aa8a HR |
1429 | *data = f.fe.fe_logical; |
1430 | *hole = f.fe.fe_logical + f.fe.fe_length; | |
f5f7abcf PB |
1431 | if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) { |
1432 | ret |= BDRV_BLOCK_ZERO; | |
1433 | } | |
5500316d | 1434 | } |
94282e71 | 1435 | |
4f11aa8a HR |
1436 | return ret; |
1437 | #else | |
1438 | return -ENOTSUP; | |
1439 | #endif | |
1440 | } | |
94282e71 | 1441 | |
4f11aa8a HR |
1442 | static int64_t try_seek_hole(BlockDriverState *bs, off_t start, off_t *data, |
1443 | off_t *hole, int *pnum) | |
1444 | { | |
1445 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 KW |
1446 | BDRVRawState *s = bs->opaque; |
1447 | ||
4f11aa8a HR |
1448 | *hole = lseek(s->fd, start, SEEK_HOLE); |
1449 | if (*hole == -1) { | |
5500316d PB |
1450 | /* -ENXIO indicates that sector_num was past the end of the file. |
1451 | * There is a virtual hole there. */ | |
1452 | assert(errno != -ENXIO); | |
1453 | ||
4f11aa8a | 1454 | return -errno; |
5500316d PB |
1455 | } |
1456 | ||
4f11aa8a HR |
1457 | if (*hole > start) { |
1458 | *data = start; | |
5500316d PB |
1459 | } else { |
1460 | /* On a hole. We need another syscall to find its end. */ | |
4f11aa8a HR |
1461 | *data = lseek(s->fd, start, SEEK_DATA); |
1462 | if (*data == -1) { | |
1463 | *data = lseek(s->fd, 0, SEEK_END); | |
5500316d PB |
1464 | } |
1465 | } | |
4f11aa8a HR |
1466 | |
1467 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start; | |
5500316d | 1468 | #else |
4f11aa8a | 1469 | return -ENOTSUP; |
5500316d | 1470 | #endif |
4f11aa8a HR |
1471 | } |
1472 | ||
1473 | /* | |
1474 | * Returns true iff the specified sector is present in the disk image. Drivers | |
1475 | * not implementing the functionality are assumed to not support backing files, | |
1476 | * hence all their sectors are reported as allocated. | |
1477 | * | |
1478 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
1479 | * and 'pnum' is set to 0. | |
1480 | * | |
1481 | * 'pnum' is set to the number of sectors (including and immediately following | |
1482 | * the specified sector) that are known to be in the same | |
1483 | * allocated/unallocated state. | |
1484 | * | |
1485 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
1486 | * beyond the end of the disk image it will be clamped. | |
1487 | */ | |
1488 | static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs, | |
1489 | int64_t sector_num, | |
1490 | int nb_sectors, int *pnum) | |
1491 | { | |
1492 | off_t start, data = 0, hole = 0; | |
1493 | int64_t ret; | |
1494 | ||
1495 | ret = fd_open(bs); | |
1496 | if (ret < 0) { | |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | start = sector_num * BDRV_SECTOR_SIZE; | |
1501 | ||
1502 | ret = try_fiemap(bs, start, &data, &hole, nb_sectors, pnum); | |
1503 | if (ret < 0) { | |
1504 | ret = try_seek_hole(bs, start, &data, &hole, pnum); | |
1505 | if (ret < 0) { | |
1506 | /* Assume everything is allocated. */ | |
1507 | data = 0; | |
1508 | hole = start + nb_sectors * BDRV_SECTOR_SIZE; | |
1509 | ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start; | |
1510 | } | |
1511 | } | |
5500316d PB |
1512 | |
1513 | if (data <= start) { | |
1514 | /* On a data extent, compute sectors to the end of the extent. */ | |
1515 | *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE); | |
5500316d PB |
1516 | } else { |
1517 | /* On a hole, compute sectors to the beginning of the next extent. */ | |
1518 | *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE); | |
63390a8d PB |
1519 | ret &= ~BDRV_BLOCK_DATA; |
1520 | ret |= BDRV_BLOCK_ZERO; | |
5500316d | 1521 | } |
63390a8d PB |
1522 | |
1523 | return ret; | |
5500316d PB |
1524 | } |
1525 | ||
8238010b PB |
1526 | static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs, |
1527 | int64_t sector_num, int nb_sectors, | |
1528 | BlockDriverCompletionFunc *cb, void *opaque) | |
dce512de | 1529 | { |
dce512de CH |
1530 | BDRVRawState *s = bs->opaque; |
1531 | ||
8238010b PB |
1532 | return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors, |
1533 | cb, opaque, QEMU_AIO_DISCARD); | |
dce512de | 1534 | } |
0e7e1989 | 1535 | |
260a82e5 PB |
1536 | static int coroutine_fn raw_co_write_zeroes( |
1537 | BlockDriverState *bs, int64_t sector_num, | |
1538 | int nb_sectors, BdrvRequestFlags flags) | |
1539 | { | |
1540 | BDRVRawState *s = bs->opaque; | |
1541 | ||
1542 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
97a2ae34 PB |
1543 | return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors, |
1544 | QEMU_AIO_WRITE_ZEROES); | |
1545 | } else if (s->discard_zeroes) { | |
1546 | return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors, | |
1547 | QEMU_AIO_DISCARD); | |
260a82e5 | 1548 | } |
97a2ae34 | 1549 | return -ENOTSUP; |
260a82e5 PB |
1550 | } |
1551 | ||
1552 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1553 | { | |
1554 | BDRVRawState *s = bs->opaque; | |
1555 | ||
1556 | bdi->unallocated_blocks_are_zero = s->discard_zeroes; | |
1557 | bdi->can_write_zeroes_with_unmap = s->discard_zeroes; | |
1558 | return 0; | |
1559 | } | |
1560 | ||
6f482f74 CL |
1561 | static QemuOptsList raw_create_opts = { |
1562 | .name = "raw-create-opts", | |
1563 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
1564 | .desc = { | |
1565 | { | |
1566 | .name = BLOCK_OPT_SIZE, | |
1567 | .type = QEMU_OPT_SIZE, | |
1568 | .help = "Virtual disk size" | |
1569 | }, | |
4ab15590 CL |
1570 | { |
1571 | .name = BLOCK_OPT_NOCOW, | |
1572 | .type = QEMU_OPT_BOOL, | |
1573 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
1574 | }, | |
6f482f74 CL |
1575 | { /* end of list */ } |
1576 | } | |
0e7e1989 KW |
1577 | }; |
1578 | ||
84a12e66 CH |
1579 | static BlockDriver bdrv_file = { |
1580 | .format_name = "file", | |
1581 | .protocol_name = "file", | |
856ae5c3 | 1582 | .instance_size = sizeof(BDRVRawState), |
030be321 | 1583 | .bdrv_needs_filename = true, |
856ae5c3 | 1584 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 1585 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 1586 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
1587 | .bdrv_reopen_prepare = raw_reopen_prepare, |
1588 | .bdrv_reopen_commit = raw_reopen_commit, | |
1589 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 1590 | .bdrv_close = raw_close, |
c282e1fd | 1591 | .bdrv_create = raw_create, |
3ac21627 | 1592 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
b6b8a333 | 1593 | .bdrv_co_get_block_status = raw_co_get_block_status, |
260a82e5 | 1594 | .bdrv_co_write_zeroes = raw_co_write_zeroes, |
3b46e624 | 1595 | |
f141eafe AL |
1596 | .bdrv_aio_readv = raw_aio_readv, |
1597 | .bdrv_aio_writev = raw_aio_writev, | |
b2e12bc6 | 1598 | .bdrv_aio_flush = raw_aio_flush, |
8238010b | 1599 | .bdrv_aio_discard = raw_aio_discard, |
c25f53b0 | 1600 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
1601 | .bdrv_io_plug = raw_aio_plug, |
1602 | .bdrv_io_unplug = raw_aio_unplug, | |
1603 | .bdrv_flush_io_queue = raw_aio_flush_io_queue, | |
3c529d93 | 1604 | |
83f64091 FB |
1605 | .bdrv_truncate = raw_truncate, |
1606 | .bdrv_getlength = raw_getlength, | |
260a82e5 | 1607 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
1608 | .bdrv_get_allocated_file_size |
1609 | = raw_get_allocated_file_size, | |
0e7e1989 | 1610 | |
c2f3426c SH |
1611 | .bdrv_detach_aio_context = raw_detach_aio_context, |
1612 | .bdrv_attach_aio_context = raw_attach_aio_context, | |
1613 | ||
6f482f74 | 1614 | .create_opts = &raw_create_opts, |
83f64091 FB |
1615 | }; |
1616 | ||
19cb3738 FB |
1617 | /***********************************************/ |
1618 | /* host device */ | |
1619 | ||
83affaa6 | 1620 | #if defined(__APPLE__) && defined(__MACH__) |
19cb3738 FB |
1621 | static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator ); |
1622 | static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize ); | |
1623 | ||
1624 | kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator ) | |
1625 | { | |
5fafdf24 | 1626 | kern_return_t kernResult; |
19cb3738 FB |
1627 | mach_port_t masterPort; |
1628 | CFMutableDictionaryRef classesToMatch; | |
1629 | ||
1630 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
1631 | if ( KERN_SUCCESS != kernResult ) { | |
1632 | printf( "IOMasterPort returned %d\n", kernResult ); | |
1633 | } | |
3b46e624 | 1634 | |
5fafdf24 | 1635 | classesToMatch = IOServiceMatching( kIOCDMediaClass ); |
19cb3738 FB |
1636 | if ( classesToMatch == NULL ) { |
1637 | printf( "IOServiceMatching returned a NULL dictionary.\n" ); | |
1638 | } else { | |
1639 | CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue ); | |
1640 | } | |
1641 | kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator ); | |
1642 | if ( KERN_SUCCESS != kernResult ) | |
1643 | { | |
1644 | printf( "IOServiceGetMatchingServices returned %d\n", kernResult ); | |
1645 | } | |
3b46e624 | 1646 | |
19cb3738 FB |
1647 | return kernResult; |
1648 | } | |
1649 | ||
1650 | kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize ) | |
1651 | { | |
1652 | io_object_t nextMedia; | |
1653 | kern_return_t kernResult = KERN_FAILURE; | |
1654 | *bsdPath = '\0'; | |
1655 | nextMedia = IOIteratorNext( mediaIterator ); | |
1656 | if ( nextMedia ) | |
1657 | { | |
1658 | CFTypeRef bsdPathAsCFString; | |
1659 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
1660 | if ( bsdPathAsCFString ) { | |
1661 | size_t devPathLength; | |
1662 | strcpy( bsdPath, _PATH_DEV ); | |
1663 | strcat( bsdPath, "r" ); | |
1664 | devPathLength = strlen( bsdPath ); | |
1665 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
1666 | kernResult = KERN_SUCCESS; | |
1667 | } | |
1668 | CFRelease( bsdPathAsCFString ); | |
1669 | } | |
1670 | IOObjectRelease( nextMedia ); | |
1671 | } | |
3b46e624 | 1672 | |
19cb3738 FB |
1673 | return kernResult; |
1674 | } | |
1675 | ||
1676 | #endif | |
1677 | ||
508c7cb3 CH |
1678 | static int hdev_probe_device(const char *filename) |
1679 | { | |
1680 | struct stat st; | |
1681 | ||
1682 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
1683 | if (strstart(filename, "/dev/cdrom", NULL)) | |
1684 | return 50; | |
1685 | ||
1686 | if (stat(filename, &st) >= 0 && | |
1687 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
1688 | return 100; | |
1689 | } | |
1690 | ||
1691 | return 0; | |
1692 | } | |
1693 | ||
da888d37 SH |
1694 | static int check_hdev_writable(BDRVRawState *s) |
1695 | { | |
1696 | #if defined(BLKROGET) | |
1697 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
1698 | * This is independent of device node permissions and therefore open(2) | |
1699 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
1700 | * | |
1701 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
1702 | * check for read-only block devices so that Linux block devices behave | |
1703 | * properly. | |
1704 | */ | |
1705 | struct stat st; | |
1706 | int readonly = 0; | |
1707 | ||
1708 | if (fstat(s->fd, &st)) { | |
1709 | return -errno; | |
1710 | } | |
1711 | ||
1712 | if (!S_ISBLK(st.st_mode)) { | |
1713 | return 0; | |
1714 | } | |
1715 | ||
1716 | if (ioctl(s->fd, BLKROGET, &readonly) < 0) { | |
1717 | return -errno; | |
1718 | } | |
1719 | ||
1720 | if (readonly) { | |
1721 | return -EACCES; | |
1722 | } | |
1723 | #endif /* defined(BLKROGET) */ | |
1724 | return 0; | |
1725 | } | |
1726 | ||
7af803d4 HR |
1727 | static void hdev_parse_filename(const char *filename, QDict *options, |
1728 | Error **errp) | |
1729 | { | |
1730 | /* The prefix is optional, just as for "file". */ | |
1731 | strstart(filename, "host_device:", &filename); | |
1732 | ||
1733 | qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename))); | |
1734 | } | |
1735 | ||
015a1036 HR |
1736 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
1737 | Error **errp) | |
19cb3738 FB |
1738 | { |
1739 | BDRVRawState *s = bs->opaque; | |
e428e439 | 1740 | Error *local_err = NULL; |
da888d37 | 1741 | int ret; |
c66a6157 | 1742 | const char *filename = qdict_get_str(options, "filename"); |
a76bab49 | 1743 | |
83affaa6 | 1744 | #if defined(__APPLE__) && defined(__MACH__) |
19cb3738 FB |
1745 | if (strstart(filename, "/dev/cdrom", NULL)) { |
1746 | kern_return_t kernResult; | |
1747 | io_iterator_t mediaIterator; | |
1748 | char bsdPath[ MAXPATHLEN ]; | |
1749 | int fd; | |
5fafdf24 | 1750 | |
19cb3738 FB |
1751 | kernResult = FindEjectableCDMedia( &mediaIterator ); |
1752 | kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) ); | |
3b46e624 | 1753 | |
19cb3738 FB |
1754 | if ( bsdPath[ 0 ] != '\0' ) { |
1755 | strcat(bsdPath,"s0"); | |
1756 | /* some CDs don't have a partition 0 */ | |
6165f4d8 | 1757 | fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE); |
19cb3738 FB |
1758 | if (fd < 0) { |
1759 | bsdPath[strlen(bsdPath)-1] = '1'; | |
1760 | } else { | |
2e1e79da | 1761 | qemu_close(fd); |
19cb3738 FB |
1762 | } |
1763 | filename = bsdPath; | |
a5c5ea3f | 1764 | qdict_put(options, "filename", qstring_from_str(filename)); |
19cb3738 | 1765 | } |
3b46e624 | 1766 | |
19cb3738 FB |
1767 | if ( mediaIterator ) |
1768 | IOObjectRelease( mediaIterator ); | |
1769 | } | |
1770 | #endif | |
19cb3738 FB |
1771 | |
1772 | s->type = FTYPE_FILE; | |
4dd75c70 | 1773 | #if defined(__linux__) |
05acda4d BK |
1774 | { |
1775 | char resolved_path[ MAXPATHLEN ], *temp; | |
1776 | ||
1777 | temp = realpath(filename, resolved_path); | |
1778 | if (temp && strstart(temp, "/dev/sg", NULL)) { | |
1779 | bs->sg = 1; | |
1780 | } | |
19cb3738 FB |
1781 | } |
1782 | #endif | |
90babde0 | 1783 | |
e428e439 | 1784 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
da888d37 | 1785 | if (ret < 0) { |
84d18f06 | 1786 | if (local_err) { |
e428e439 HR |
1787 | error_propagate(errp, local_err); |
1788 | } | |
da888d37 SH |
1789 | return ret; |
1790 | } | |
1791 | ||
1792 | if (flags & BDRV_O_RDWR) { | |
1793 | ret = check_hdev_writable(s); | |
1794 | if (ret < 0) { | |
1795 | raw_close(bs); | |
e428e439 | 1796 | error_setg_errno(errp, -ret, "The device is not writable"); |
da888d37 SH |
1797 | return ret; |
1798 | } | |
1799 | } | |
1800 | ||
1801 | return ret; | |
19cb3738 FB |
1802 | } |
1803 | ||
03ff3ca3 | 1804 | #if defined(__linux__) |
19cb3738 FB |
1805 | /* Note: we do not have a reliable method to detect if the floppy is |
1806 | present. The current method is to try to open the floppy at every | |
1807 | I/O and to keep it opened during a few hundreds of ms. */ | |
1808 | static int fd_open(BlockDriverState *bs) | |
1809 | { | |
1810 | BDRVRawState *s = bs->opaque; | |
1811 | int last_media_present; | |
1812 | ||
1813 | if (s->type != FTYPE_FD) | |
1814 | return 0; | |
1815 | last_media_present = (s->fd >= 0); | |
5fafdf24 | 1816 | if (s->fd >= 0 && |
c57c846a | 1817 | (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) { |
2e1e79da | 1818 | qemu_close(s->fd); |
19cb3738 FB |
1819 | s->fd = -1; |
1820 | #ifdef DEBUG_FLOPPY | |
1821 | printf("Floppy closed\n"); | |
1822 | #endif | |
1823 | } | |
1824 | if (s->fd < 0) { | |
5fafdf24 | 1825 | if (s->fd_got_error && |
c57c846a | 1826 | (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) { |
19cb3738 FB |
1827 | #ifdef DEBUG_FLOPPY |
1828 | printf("No floppy (open delayed)\n"); | |
1829 | #endif | |
1830 | return -EIO; | |
1831 | } | |
6165f4d8 | 1832 | s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK); |
19cb3738 | 1833 | if (s->fd < 0) { |
c57c846a | 1834 | s->fd_error_time = get_clock(); |
19cb3738 FB |
1835 | s->fd_got_error = 1; |
1836 | if (last_media_present) | |
1837 | s->fd_media_changed = 1; | |
1838 | #ifdef DEBUG_FLOPPY | |
1839 | printf("No floppy\n"); | |
1840 | #endif | |
1841 | return -EIO; | |
1842 | } | |
1843 | #ifdef DEBUG_FLOPPY | |
1844 | printf("Floppy opened\n"); | |
1845 | #endif | |
1846 | } | |
1847 | if (!last_media_present) | |
1848 | s->fd_media_changed = 1; | |
c57c846a | 1849 | s->fd_open_time = get_clock(); |
19cb3738 FB |
1850 | s->fd_got_error = 0; |
1851 | return 0; | |
1852 | } | |
19cb3738 | 1853 | |
63ec93db | 1854 | static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) |
985a03b0 TS |
1855 | { |
1856 | BDRVRawState *s = bs->opaque; | |
1857 | ||
1858 | return ioctl(s->fd, req, buf); | |
1859 | } | |
221f715d | 1860 | |
63ec93db | 1861 | static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs, |
221f715d AL |
1862 | unsigned long int req, void *buf, |
1863 | BlockDriverCompletionFunc *cb, void *opaque) | |
1864 | { | |
f141eafe | 1865 | BDRVRawState *s = bs->opaque; |
c208e8c2 | 1866 | RawPosixAIOData *acb; |
c4d9d196 | 1867 | ThreadPool *pool; |
221f715d | 1868 | |
f141eafe AL |
1869 | if (fd_open(bs) < 0) |
1870 | return NULL; | |
c208e8c2 PB |
1871 | |
1872 | acb = g_slice_new(RawPosixAIOData); | |
1873 | acb->bs = bs; | |
1874 | acb->aio_type = QEMU_AIO_IOCTL; | |
1875 | acb->aio_fildes = s->fd; | |
1876 | acb->aio_offset = 0; | |
1877 | acb->aio_ioctl_buf = buf; | |
1878 | acb->aio_ioctl_cmd = req; | |
c4d9d196 SH |
1879 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
1880 | return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); | |
221f715d AL |
1881 | } |
1882 | ||
a167ba50 | 1883 | #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
1884 | static int fd_open(BlockDriverState *bs) |
1885 | { | |
1886 | BDRVRawState *s = bs->opaque; | |
1887 | ||
1888 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
1889 | if (s->fd >= 0) | |
1890 | return 0; | |
1891 | return -EIO; | |
1892 | } | |
9f23011a | 1893 | #else /* !linux && !FreeBSD */ |
19cb3738 | 1894 | |
08af02e2 AL |
1895 | static int fd_open(BlockDriverState *bs) |
1896 | { | |
1897 | return 0; | |
1898 | } | |
1899 | ||
221f715d | 1900 | #endif /* !linux && !FreeBSD */ |
04eeb8b6 | 1901 | |
c36dd8a0 AF |
1902 | static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs, |
1903 | int64_t sector_num, int nb_sectors, | |
1904 | BlockDriverCompletionFunc *cb, void *opaque) | |
1905 | { | |
1906 | BDRVRawState *s = bs->opaque; | |
1907 | ||
1908 | if (fd_open(bs) < 0) { | |
1909 | return NULL; | |
1910 | } | |
1911 | return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors, | |
1912 | cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); | |
1913 | } | |
1914 | ||
d0b4503e PB |
1915 | static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs, |
1916 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) | |
1917 | { | |
1918 | BDRVRawState *s = bs->opaque; | |
1919 | int rc; | |
1920 | ||
1921 | rc = fd_open(bs); | |
1922 | if (rc < 0) { | |
1923 | return rc; | |
1924 | } | |
1925 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
97a2ae34 PB |
1926 | return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors, |
1927 | QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV); | |
1928 | } else if (s->discard_zeroes) { | |
1929 | return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors, | |
1930 | QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); | |
d0b4503e | 1931 | } |
97a2ae34 | 1932 | return -ENOTSUP; |
d0b4503e PB |
1933 | } |
1934 | ||
6f482f74 | 1935 | static int hdev_create(const char *filename, QemuOpts *opts, |
d5124c00 | 1936 | Error **errp) |
93c65b47 AL |
1937 | { |
1938 | int fd; | |
1939 | int ret = 0; | |
1940 | struct stat stat_buf; | |
0e7e1989 | 1941 | int64_t total_size = 0; |
cc28c6aa HR |
1942 | bool has_prefix; |
1943 | ||
1944 | /* This function is used by all three protocol block drivers and therefore | |
1945 | * any of these three prefixes may be given. | |
1946 | * The return value has to be stored somewhere, otherwise this is an error | |
1947 | * due to -Werror=unused-value. */ | |
1948 | has_prefix = | |
1949 | strstart(filename, "host_device:", &filename) || | |
1950 | strstart(filename, "host_cdrom:" , &filename) || | |
1951 | strstart(filename, "host_floppy:", &filename); | |
1952 | ||
1953 | (void)has_prefix; | |
93c65b47 | 1954 | |
0e7e1989 | 1955 | /* Read out options */ |
6f482f74 CL |
1956 | total_size = |
1957 | qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0) / BDRV_SECTOR_SIZE; | |
93c65b47 | 1958 | |
6165f4d8 | 1959 | fd = qemu_open(filename, O_WRONLY | O_BINARY); |
e428e439 HR |
1960 | if (fd < 0) { |
1961 | ret = -errno; | |
1962 | error_setg_errno(errp, -ret, "Could not open device"); | |
1963 | return ret; | |
1964 | } | |
93c65b47 | 1965 | |
e428e439 | 1966 | if (fstat(fd, &stat_buf) < 0) { |
57e69b7d | 1967 | ret = -errno; |
e428e439 HR |
1968 | error_setg_errno(errp, -ret, "Could not stat device"); |
1969 | } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { | |
1970 | error_setg(errp, | |
1971 | "The given file is neither a block nor a character device"); | |
57e69b7d | 1972 | ret = -ENODEV; |
e428e439 HR |
1973 | } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) { |
1974 | error_setg(errp, "Device is too small"); | |
93c65b47 | 1975 | ret = -ENOSPC; |
e428e439 | 1976 | } |
93c65b47 | 1977 | |
2e1e79da | 1978 | qemu_close(fd); |
93c65b47 AL |
1979 | return ret; |
1980 | } | |
1981 | ||
5efa9d5a | 1982 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 1983 | .format_name = "host_device", |
84a12e66 | 1984 | .protocol_name = "host_device", |
0b4ce02e | 1985 | .instance_size = sizeof(BDRVRawState), |
030be321 | 1986 | .bdrv_needs_filename = true, |
0b4ce02e | 1987 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 1988 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 1989 | .bdrv_file_open = hdev_open, |
0b4ce02e | 1990 | .bdrv_close = raw_close, |
1bc6b705 JC |
1991 | .bdrv_reopen_prepare = raw_reopen_prepare, |
1992 | .bdrv_reopen_commit = raw_reopen_commit, | |
1993 | .bdrv_reopen_abort = raw_reopen_abort, | |
c282e1fd | 1994 | .bdrv_create = hdev_create, |
6f482f74 | 1995 | .create_opts = &raw_create_opts, |
d0b4503e | 1996 | .bdrv_co_write_zeroes = hdev_co_write_zeroes, |
3b46e624 | 1997 | |
f141eafe AL |
1998 | .bdrv_aio_readv = raw_aio_readv, |
1999 | .bdrv_aio_writev = raw_aio_writev, | |
b2e12bc6 | 2000 | .bdrv_aio_flush = raw_aio_flush, |
8238010b | 2001 | .bdrv_aio_discard = hdev_aio_discard, |
c25f53b0 | 2002 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2003 | .bdrv_io_plug = raw_aio_plug, |
2004 | .bdrv_io_unplug = raw_aio_unplug, | |
2005 | .bdrv_flush_io_queue = raw_aio_flush_io_queue, | |
3c529d93 | 2006 | |
55b949c8 | 2007 | .bdrv_truncate = raw_truncate, |
e60f469c | 2008 | .bdrv_getlength = raw_getlength, |
260a82e5 | 2009 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
2010 | .bdrv_get_allocated_file_size |
2011 | = raw_get_allocated_file_size, | |
19cb3738 | 2012 | |
c2f3426c SH |
2013 | .bdrv_detach_aio_context = raw_detach_aio_context, |
2014 | .bdrv_attach_aio_context = raw_attach_aio_context, | |
2015 | ||
f3a5d3f8 | 2016 | /* generic scsi device */ |
63ec93db CH |
2017 | #ifdef __linux__ |
2018 | .bdrv_ioctl = hdev_ioctl, | |
63ec93db CH |
2019 | .bdrv_aio_ioctl = hdev_aio_ioctl, |
2020 | #endif | |
f3a5d3f8 CH |
2021 | }; |
2022 | ||
2023 | #ifdef __linux__ | |
d3f49845 HR |
2024 | static void floppy_parse_filename(const char *filename, QDict *options, |
2025 | Error **errp) | |
2026 | { | |
2027 | /* The prefix is optional, just as for "file". */ | |
2028 | strstart(filename, "host_floppy:", &filename); | |
2029 | ||
2030 | qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename))); | |
2031 | } | |
2032 | ||
015a1036 HR |
2033 | static int floppy_open(BlockDriverState *bs, QDict *options, int flags, |
2034 | Error **errp) | |
f3a5d3f8 CH |
2035 | { |
2036 | BDRVRawState *s = bs->opaque; | |
e428e439 | 2037 | Error *local_err = NULL; |
f3a5d3f8 CH |
2038 | int ret; |
2039 | ||
f3a5d3f8 | 2040 | s->type = FTYPE_FD; |
f3a5d3f8 | 2041 | |
19a3da7f | 2042 | /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */ |
e428e439 HR |
2043 | ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err); |
2044 | if (ret) { | |
84d18f06 | 2045 | if (local_err) { |
e428e439 HR |
2046 | error_propagate(errp, local_err); |
2047 | } | |
f3a5d3f8 | 2048 | return ret; |
e428e439 | 2049 | } |
f3a5d3f8 CH |
2050 | |
2051 | /* close fd so that we can reopen it as needed */ | |
2e1e79da | 2052 | qemu_close(s->fd); |
f3a5d3f8 CH |
2053 | s->fd = -1; |
2054 | s->fd_media_changed = 1; | |
2055 | ||
2056 | return 0; | |
2057 | } | |
2058 | ||
508c7cb3 CH |
2059 | static int floppy_probe_device(const char *filename) |
2060 | { | |
2ebf7c4b CR |
2061 | int fd, ret; |
2062 | int prio = 0; | |
2063 | struct floppy_struct fdparam; | |
343f8568 | 2064 | struct stat st; |
2ebf7c4b | 2065 | |
e1740828 CB |
2066 | if (strstart(filename, "/dev/fd", NULL) && |
2067 | !strstart(filename, "/dev/fdset/", NULL)) { | |
2ebf7c4b | 2068 | prio = 50; |
e1740828 | 2069 | } |
2ebf7c4b | 2070 | |
6165f4d8 | 2071 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
2ebf7c4b CR |
2072 | if (fd < 0) { |
2073 | goto out; | |
2074 | } | |
343f8568 JS |
2075 | ret = fstat(fd, &st); |
2076 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
2077 | goto outc; | |
2078 | } | |
2ebf7c4b CR |
2079 | |
2080 | /* Attempt to detect via a floppy specific ioctl */ | |
2081 | ret = ioctl(fd, FDGETPRM, &fdparam); | |
2082 | if (ret >= 0) | |
2083 | prio = 100; | |
2084 | ||
343f8568 | 2085 | outc: |
2e1e79da | 2086 | qemu_close(fd); |
2ebf7c4b CR |
2087 | out: |
2088 | return prio; | |
508c7cb3 CH |
2089 | } |
2090 | ||
2091 | ||
f3a5d3f8 CH |
2092 | static int floppy_is_inserted(BlockDriverState *bs) |
2093 | { | |
2094 | return fd_open(bs) >= 0; | |
2095 | } | |
2096 | ||
2097 | static int floppy_media_changed(BlockDriverState *bs) | |
2098 | { | |
2099 | BDRVRawState *s = bs->opaque; | |
2100 | int ret; | |
2101 | ||
2102 | /* | |
2103 | * XXX: we do not have a true media changed indication. | |
2104 | * It does not work if the floppy is changed without trying to read it. | |
2105 | */ | |
2106 | fd_open(bs); | |
2107 | ret = s->fd_media_changed; | |
2108 | s->fd_media_changed = 0; | |
2109 | #ifdef DEBUG_FLOPPY | |
2110 | printf("Floppy changed=%d\n", ret); | |
2111 | #endif | |
2112 | return ret; | |
2113 | } | |
2114 | ||
f36f3949 | 2115 | static void floppy_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
2116 | { |
2117 | BDRVRawState *s = bs->opaque; | |
2118 | int fd; | |
2119 | ||
2120 | if (s->fd >= 0) { | |
2e1e79da | 2121 | qemu_close(s->fd); |
f3a5d3f8 CH |
2122 | s->fd = -1; |
2123 | } | |
6165f4d8 | 2124 | fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK); |
f3a5d3f8 CH |
2125 | if (fd >= 0) { |
2126 | if (ioctl(fd, FDEJECT, 0) < 0) | |
2127 | perror("FDEJECT"); | |
2e1e79da | 2128 | qemu_close(fd); |
f3a5d3f8 | 2129 | } |
f3a5d3f8 CH |
2130 | } |
2131 | ||
2132 | static BlockDriver bdrv_host_floppy = { | |
2133 | .format_name = "host_floppy", | |
84a12e66 | 2134 | .protocol_name = "host_floppy", |
f3a5d3f8 | 2135 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2136 | .bdrv_needs_filename = true, |
508c7cb3 | 2137 | .bdrv_probe_device = floppy_probe_device, |
d3f49845 | 2138 | .bdrv_parse_filename = floppy_parse_filename, |
66f82cee | 2139 | .bdrv_file_open = floppy_open, |
f3a5d3f8 | 2140 | .bdrv_close = raw_close, |
1bc6b705 JC |
2141 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2142 | .bdrv_reopen_commit = raw_reopen_commit, | |
2143 | .bdrv_reopen_abort = raw_reopen_abort, | |
c282e1fd | 2144 | .bdrv_create = hdev_create, |
6f482f74 | 2145 | .create_opts = &raw_create_opts, |
f3a5d3f8 | 2146 | |
f3a5d3f8 CH |
2147 | .bdrv_aio_readv = raw_aio_readv, |
2148 | .bdrv_aio_writev = raw_aio_writev, | |
b2e12bc6 | 2149 | .bdrv_aio_flush = raw_aio_flush, |
c25f53b0 | 2150 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2151 | .bdrv_io_plug = raw_aio_plug, |
2152 | .bdrv_io_unplug = raw_aio_unplug, | |
2153 | .bdrv_flush_io_queue = raw_aio_flush_io_queue, | |
f3a5d3f8 | 2154 | |
55b949c8 | 2155 | .bdrv_truncate = raw_truncate, |
b94a2610 KW |
2156 | .bdrv_getlength = raw_getlength, |
2157 | .has_variable_length = true, | |
4a1d5e1f FZ |
2158 | .bdrv_get_allocated_file_size |
2159 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 2160 | |
c2f3426c SH |
2161 | .bdrv_detach_aio_context = raw_detach_aio_context, |
2162 | .bdrv_attach_aio_context = raw_attach_aio_context, | |
2163 | ||
f3a5d3f8 CH |
2164 | /* removable device support */ |
2165 | .bdrv_is_inserted = floppy_is_inserted, | |
2166 | .bdrv_media_changed = floppy_media_changed, | |
2167 | .bdrv_eject = floppy_eject, | |
f3a5d3f8 | 2168 | }; |
18fa1c42 | 2169 | #endif |
f3a5d3f8 | 2170 | |
18fa1c42 HR |
2171 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
2172 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
2173 | Error **errp) | |
2174 | { | |
2175 | /* The prefix is optional, just as for "file". */ | |
2176 | strstart(filename, "host_cdrom:", &filename); | |
2177 | ||
2178 | qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename))); | |
2179 | } | |
2180 | #endif | |
2181 | ||
2182 | #ifdef __linux__ | |
015a1036 HR |
2183 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
2184 | Error **errp) | |
f3a5d3f8 CH |
2185 | { |
2186 | BDRVRawState *s = bs->opaque; | |
e428e439 HR |
2187 | Error *local_err = NULL; |
2188 | int ret; | |
f3a5d3f8 | 2189 | |
f3a5d3f8 CH |
2190 | s->type = FTYPE_CD; |
2191 | ||
19a3da7f | 2192 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
e428e439 | 2193 | ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err); |
84d18f06 | 2194 | if (local_err) { |
e428e439 HR |
2195 | error_propagate(errp, local_err); |
2196 | } | |
2197 | return ret; | |
f3a5d3f8 CH |
2198 | } |
2199 | ||
508c7cb3 CH |
2200 | static int cdrom_probe_device(const char *filename) |
2201 | { | |
3baf720e CR |
2202 | int fd, ret; |
2203 | int prio = 0; | |
343f8568 | 2204 | struct stat st; |
3baf720e | 2205 | |
6165f4d8 | 2206 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
3baf720e CR |
2207 | if (fd < 0) { |
2208 | goto out; | |
2209 | } | |
343f8568 JS |
2210 | ret = fstat(fd, &st); |
2211 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
2212 | goto outc; | |
2213 | } | |
3baf720e CR |
2214 | |
2215 | /* Attempt to detect via a CDROM specific ioctl */ | |
2216 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
2217 | if (ret >= 0) | |
2218 | prio = 100; | |
2219 | ||
343f8568 | 2220 | outc: |
2e1e79da | 2221 | qemu_close(fd); |
3baf720e CR |
2222 | out: |
2223 | return prio; | |
508c7cb3 CH |
2224 | } |
2225 | ||
f3a5d3f8 CH |
2226 | static int cdrom_is_inserted(BlockDriverState *bs) |
2227 | { | |
2228 | BDRVRawState *s = bs->opaque; | |
2229 | int ret; | |
2230 | ||
2231 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
2232 | if (ret == CDS_DISC_OK) | |
2233 | return 1; | |
2234 | return 0; | |
2235 | } | |
2236 | ||
f36f3949 | 2237 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
2238 | { |
2239 | BDRVRawState *s = bs->opaque; | |
2240 | ||
2241 | if (eject_flag) { | |
2242 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
2243 | perror("CDROMEJECT"); | |
2244 | } else { | |
2245 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
2246 | perror("CDROMEJECT"); | |
2247 | } | |
f3a5d3f8 CH |
2248 | } |
2249 | ||
025e849a | 2250 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
2251 | { |
2252 | BDRVRawState *s = bs->opaque; | |
2253 | ||
2254 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
2255 | /* | |
2256 | * Note: an error can happen if the distribution automatically | |
2257 | * mounts the CD-ROM | |
2258 | */ | |
2259 | /* perror("CDROM_LOCKDOOR"); */ | |
2260 | } | |
f3a5d3f8 CH |
2261 | } |
2262 | ||
2263 | static BlockDriver bdrv_host_cdrom = { | |
2264 | .format_name = "host_cdrom", | |
84a12e66 | 2265 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 2266 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2267 | .bdrv_needs_filename = true, |
508c7cb3 | 2268 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 2269 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 2270 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 2271 | .bdrv_close = raw_close, |
1bc6b705 JC |
2272 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2273 | .bdrv_reopen_commit = raw_reopen_commit, | |
2274 | .bdrv_reopen_abort = raw_reopen_abort, | |
c282e1fd | 2275 | .bdrv_create = hdev_create, |
6f482f74 | 2276 | .create_opts = &raw_create_opts, |
f3a5d3f8 | 2277 | |
f3a5d3f8 CH |
2278 | .bdrv_aio_readv = raw_aio_readv, |
2279 | .bdrv_aio_writev = raw_aio_writev, | |
b2e12bc6 | 2280 | .bdrv_aio_flush = raw_aio_flush, |
c25f53b0 | 2281 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2282 | .bdrv_io_plug = raw_aio_plug, |
2283 | .bdrv_io_unplug = raw_aio_unplug, | |
2284 | .bdrv_flush_io_queue = raw_aio_flush_io_queue, | |
f3a5d3f8 | 2285 | |
55b949c8 | 2286 | .bdrv_truncate = raw_truncate, |
b94a2610 KW |
2287 | .bdrv_getlength = raw_getlength, |
2288 | .has_variable_length = true, | |
4a1d5e1f FZ |
2289 | .bdrv_get_allocated_file_size |
2290 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 2291 | |
c2f3426c SH |
2292 | .bdrv_detach_aio_context = raw_detach_aio_context, |
2293 | .bdrv_attach_aio_context = raw_attach_aio_context, | |
2294 | ||
f3a5d3f8 CH |
2295 | /* removable device support */ |
2296 | .bdrv_is_inserted = cdrom_is_inserted, | |
2297 | .bdrv_eject = cdrom_eject, | |
025e849a | 2298 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
2299 | |
2300 | /* generic scsi device */ | |
63ec93db | 2301 | .bdrv_ioctl = hdev_ioctl, |
63ec93db | 2302 | .bdrv_aio_ioctl = hdev_aio_ioctl, |
f3a5d3f8 CH |
2303 | }; |
2304 | #endif /* __linux__ */ | |
2305 | ||
a167ba50 | 2306 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
2307 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
2308 | Error **errp) | |
f3a5d3f8 CH |
2309 | { |
2310 | BDRVRawState *s = bs->opaque; | |
e428e439 | 2311 | Error *local_err = NULL; |
f3a5d3f8 CH |
2312 | int ret; |
2313 | ||
2314 | s->type = FTYPE_CD; | |
2315 | ||
e428e439 HR |
2316 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
2317 | if (ret) { | |
84d18f06 | 2318 | if (local_err) { |
e428e439 HR |
2319 | error_propagate(errp, local_err); |
2320 | } | |
f3a5d3f8 | 2321 | return ret; |
e428e439 | 2322 | } |
f3a5d3f8 | 2323 | |
9b2260cb | 2324 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
2325 | ioctl(s->fd, CDIOCALLOW); |
2326 | return 0; | |
2327 | } | |
2328 | ||
508c7cb3 CH |
2329 | static int cdrom_probe_device(const char *filename) |
2330 | { | |
2331 | if (strstart(filename, "/dev/cd", NULL) || | |
2332 | strstart(filename, "/dev/acd", NULL)) | |
2333 | return 100; | |
2334 | return 0; | |
2335 | } | |
2336 | ||
f3a5d3f8 CH |
2337 | static int cdrom_reopen(BlockDriverState *bs) |
2338 | { | |
2339 | BDRVRawState *s = bs->opaque; | |
2340 | int fd; | |
2341 | ||
2342 | /* | |
2343 | * Force reread of possibly changed/newly loaded disc, | |
2344 | * FreeBSD seems to not notice sometimes... | |
2345 | */ | |
2346 | if (s->fd >= 0) | |
2e1e79da | 2347 | qemu_close(s->fd); |
6165f4d8 | 2348 | fd = qemu_open(bs->filename, s->open_flags, 0644); |
f3a5d3f8 CH |
2349 | if (fd < 0) { |
2350 | s->fd = -1; | |
2351 | return -EIO; | |
2352 | } | |
2353 | s->fd = fd; | |
2354 | ||
9b2260cb | 2355 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
2356 | ioctl(s->fd, CDIOCALLOW); |
2357 | return 0; | |
2358 | } | |
2359 | ||
2360 | static int cdrom_is_inserted(BlockDriverState *bs) | |
2361 | { | |
2362 | return raw_getlength(bs) > 0; | |
2363 | } | |
2364 | ||
f36f3949 | 2365 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
2366 | { |
2367 | BDRVRawState *s = bs->opaque; | |
2368 | ||
2369 | if (s->fd < 0) | |
822e1cd1 | 2370 | return; |
f3a5d3f8 CH |
2371 | |
2372 | (void) ioctl(s->fd, CDIOCALLOW); | |
2373 | ||
2374 | if (eject_flag) { | |
2375 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
2376 | perror("CDIOCEJECT"); | |
2377 | } else { | |
2378 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
2379 | perror("CDIOCCLOSE"); | |
2380 | } | |
2381 | ||
822e1cd1 | 2382 | cdrom_reopen(bs); |
f3a5d3f8 CH |
2383 | } |
2384 | ||
025e849a | 2385 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
2386 | { |
2387 | BDRVRawState *s = bs->opaque; | |
2388 | ||
2389 | if (s->fd < 0) | |
7bf37fed | 2390 | return; |
f3a5d3f8 CH |
2391 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
2392 | /* | |
2393 | * Note: an error can happen if the distribution automatically | |
2394 | * mounts the CD-ROM | |
2395 | */ | |
2396 | /* perror("CDROM_LOCKDOOR"); */ | |
2397 | } | |
f3a5d3f8 CH |
2398 | } |
2399 | ||
2400 | static BlockDriver bdrv_host_cdrom = { | |
2401 | .format_name = "host_cdrom", | |
84a12e66 | 2402 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 2403 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2404 | .bdrv_needs_filename = true, |
508c7cb3 | 2405 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 2406 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 2407 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 2408 | .bdrv_close = raw_close, |
1bc6b705 JC |
2409 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2410 | .bdrv_reopen_commit = raw_reopen_commit, | |
2411 | .bdrv_reopen_abort = raw_reopen_abort, | |
c282e1fd | 2412 | .bdrv_create = hdev_create, |
6f482f74 | 2413 | .create_opts = &raw_create_opts, |
f3a5d3f8 | 2414 | |
f3a5d3f8 CH |
2415 | .bdrv_aio_readv = raw_aio_readv, |
2416 | .bdrv_aio_writev = raw_aio_writev, | |
b2e12bc6 | 2417 | .bdrv_aio_flush = raw_aio_flush, |
c25f53b0 | 2418 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2419 | .bdrv_io_plug = raw_aio_plug, |
2420 | .bdrv_io_unplug = raw_aio_unplug, | |
2421 | .bdrv_flush_io_queue = raw_aio_flush_io_queue, | |
f3a5d3f8 | 2422 | |
55b949c8 | 2423 | .bdrv_truncate = raw_truncate, |
b94a2610 KW |
2424 | .bdrv_getlength = raw_getlength, |
2425 | .has_variable_length = true, | |
4a1d5e1f FZ |
2426 | .bdrv_get_allocated_file_size |
2427 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 2428 | |
c2f3426c SH |
2429 | .bdrv_detach_aio_context = raw_detach_aio_context, |
2430 | .bdrv_attach_aio_context = raw_attach_aio_context, | |
2431 | ||
19cb3738 | 2432 | /* removable device support */ |
f3a5d3f8 CH |
2433 | .bdrv_is_inserted = cdrom_is_inserted, |
2434 | .bdrv_eject = cdrom_eject, | |
025e849a | 2435 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 2436 | }; |
f3a5d3f8 | 2437 | #endif /* __FreeBSD__ */ |
5efa9d5a | 2438 | |
84a12e66 | 2439 | static void bdrv_file_init(void) |
5efa9d5a | 2440 | { |
508c7cb3 CH |
2441 | /* |
2442 | * Register all the drivers. Note that order is important, the driver | |
2443 | * registered last will get probed first. | |
2444 | */ | |
84a12e66 | 2445 | bdrv_register(&bdrv_file); |
5efa9d5a | 2446 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 CH |
2447 | #ifdef __linux__ |
2448 | bdrv_register(&bdrv_host_floppy); | |
2449 | bdrv_register(&bdrv_host_cdrom); | |
2450 | #endif | |
a167ba50 | 2451 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
2452 | bdrv_register(&bdrv_host_cdrom); |
2453 | #endif | |
5efa9d5a AL |
2454 | } |
2455 | ||
84a12e66 | 2456 | block_init(bdrv_file_init); |