]>
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 | */ | |
922a01a0 | 24 | |
80c71a24 | 25 | #include "qemu/osdep.h" |
da34e65c | 26 | #include "qapi/error.h" |
f348b6d1 | 27 | #include "qemu/cutils.h" |
d49b6836 | 28 | #include "qemu/error-report.h" |
737e150e | 29 | #include "block/block_int.h" |
1de7afc9 | 30 | #include "qemu/module.h" |
922a01a0 | 31 | #include "qemu/option.h" |
de81a169 | 32 | #include "trace.h" |
737e150e | 33 | #include "block/thread-pool.h" |
1de7afc9 | 34 | #include "qemu/iov.h" |
0187f5c9 | 35 | #include "block/raw-aio.h" |
452fcdbc | 36 | #include "qapi/qmp/qdict.h" |
d49b6836 | 37 | #include "qapi/qmp/qstring.h" |
83f64091 | 38 | |
7c9e5276 PB |
39 | #include "scsi/pr-manager.h" |
40 | #include "scsi/constants.h" | |
41 | ||
83affaa6 | 42 | #if defined(__APPLE__) && (__MACH__) |
83f64091 FB |
43 | #include <paths.h> |
44 | #include <sys/param.h> | |
45 | #include <IOKit/IOKitLib.h> | |
46 | #include <IOKit/IOBSD.h> | |
47 | #include <IOKit/storage/IOMediaBSDClient.h> | |
48 | #include <IOKit/storage/IOMedia.h> | |
49 | #include <IOKit/storage/IOCDMedia.h> | |
50 | //#include <IOKit/storage/IOCDTypes.h> | |
d0855f12 | 51 | #include <IOKit/storage/IODVDMedia.h> |
83f64091 FB |
52 | #include <CoreFoundation/CoreFoundation.h> |
53 | #endif | |
54 | ||
55 | #ifdef __sun__ | |
2e9671da | 56 | #define _POSIX_PTHREAD_SEMANTICS 1 |
83f64091 FB |
57 | #include <sys/dkio.h> |
58 | #endif | |
19cb3738 FB |
59 | #ifdef __linux__ |
60 | #include <sys/ioctl.h> | |
05acda4d | 61 | #include <sys/param.h> |
1efad060 | 62 | #include <sys/syscall.h> |
19cb3738 FB |
63 | #include <linux/cdrom.h> |
64 | #include <linux/fd.h> | |
5500316d | 65 | #include <linux/fs.h> |
1a9335e4 | 66 | #include <linux/hdreg.h> |
3307ed7b | 67 | #include <scsi/sg.h> |
1a9335e4 ET |
68 | #ifdef __s390__ |
69 | #include <asm/dasd.h> | |
70 | #endif | |
4ab15590 CL |
71 | #ifndef FS_NOCOW_FL |
72 | #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ | |
73 | #endif | |
5500316d | 74 | #endif |
b953f075 | 75 | #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) |
3d4fa43e KK |
76 | #include <linux/falloc.h> |
77 | #endif | |
a167ba50 | 78 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
1cb6c3fd | 79 | #include <sys/disk.h> |
9f23011a | 80 | #include <sys/cdio.h> |
1cb6c3fd | 81 | #endif |
83f64091 | 82 | |
128ab2ff BS |
83 | #ifdef __OpenBSD__ |
84 | #include <sys/ioctl.h> | |
85 | #include <sys/disklabel.h> | |
86 | #include <sys/dkio.h> | |
87 | #endif | |
88 | ||
d1f6fd8d CE |
89 | #ifdef __NetBSD__ |
90 | #include <sys/ioctl.h> | |
91 | #include <sys/disklabel.h> | |
92 | #include <sys/dkio.h> | |
93 | #include <sys/disk.h> | |
94 | #endif | |
95 | ||
c5e97233 BS |
96 | #ifdef __DragonFly__ |
97 | #include <sys/ioctl.h> | |
98 | #include <sys/diskslice.h> | |
99 | #endif | |
100 | ||
dce512de CH |
101 | #ifdef CONFIG_XFS |
102 | #include <xfs/xfs.h> | |
103 | #endif | |
104 | ||
faf07963 | 105 | //#define DEBUG_BLOCK |
bcb22555 DA |
106 | |
107 | #ifdef DEBUG_BLOCK | |
108 | # define DEBUG_BLOCK_PRINT 1 | |
8c05dbf9 | 109 | #else |
bcb22555 | 110 | # define DEBUG_BLOCK_PRINT 0 |
8c05dbf9 | 111 | #endif |
bcb22555 DA |
112 | #define DPRINTF(fmt, ...) \ |
113 | do { \ | |
114 | if (DEBUG_BLOCK_PRINT) { \ | |
115 | printf(fmt, ## __VA_ARGS__); \ | |
116 | } \ | |
117 | } while (0) | |
8c05dbf9 | 118 | |
f6465578 AL |
119 | /* OS X does not have O_DSYNC */ |
120 | #ifndef O_DSYNC | |
1c27a8b3 | 121 | #ifdef O_SYNC |
7ab064d2 | 122 | #define O_DSYNC O_SYNC |
1c27a8b3 JA |
123 | #elif defined(O_FSYNC) |
124 | #define O_DSYNC O_FSYNC | |
125 | #endif | |
f6465578 AL |
126 | #endif |
127 | ||
9f7965c7 AL |
128 | /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ |
129 | #ifndef O_DIRECT | |
130 | #define O_DIRECT O_DSYNC | |
131 | #endif | |
132 | ||
19cb3738 FB |
133 | #define FTYPE_FILE 0 |
134 | #define FTYPE_CD 1 | |
83f64091 | 135 | |
581b9e29 CH |
136 | #define MAX_BLOCKSIZE 4096 |
137 | ||
244a5668 FZ |
138 | /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, |
139 | * leaving a few more bytes for its future use. */ | |
140 | #define RAW_LOCK_PERM_BASE 100 | |
141 | #define RAW_LOCK_SHARED_BASE 200 | |
142 | ||
19cb3738 FB |
143 | typedef struct BDRVRawState { |
144 | int fd; | |
244a5668 FZ |
145 | int lock_fd; |
146 | bool use_lock; | |
19cb3738 | 147 | int type; |
0e1d8f4c | 148 | int open_flags; |
c25f53b0 PB |
149 | size_t buf_align; |
150 | ||
244a5668 FZ |
151 | /* The current permissions. */ |
152 | uint64_t perm; | |
153 | uint64_t shared_perm; | |
154 | ||
dce512de | 155 | #ifdef CONFIG_XFS |
260a82e5 | 156 | bool is_xfs:1; |
dce512de | 157 | #endif |
260a82e5 | 158 | bool has_discard:1; |
97a2ae34 | 159 | bool has_write_zeroes:1; |
260a82e5 | 160 | bool discard_zeroes:1; |
0a4279d9 | 161 | bool use_linux_aio:1; |
e5bcf967 | 162 | bool page_cache_inconsistent:1; |
d50d8222 | 163 | bool has_fallocate; |
3cad8307 | 164 | bool needs_alignment; |
31be8a2a | 165 | bool check_cache_dropped; |
7c9e5276 PB |
166 | |
167 | PRManager *pr_mgr; | |
19cb3738 FB |
168 | } BDRVRawState; |
169 | ||
eeb6b45d JC |
170 | typedef struct BDRVRawReopenState { |
171 | int fd; | |
172 | int open_flags; | |
31be8a2a | 173 | bool check_cache_dropped; |
eeb6b45d JC |
174 | } BDRVRawReopenState; |
175 | ||
19cb3738 | 176 | static int fd_open(BlockDriverState *bs); |
22afa7b5 | 177 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 178 | |
de81a169 PB |
179 | typedef struct RawPosixAIOData { |
180 | BlockDriverState *bs; | |
181 | int aio_fildes; | |
182 | union { | |
183 | struct iovec *aio_iov; | |
184 | void *aio_ioctl_buf; | |
185 | }; | |
186 | int aio_niov; | |
8238010b | 187 | uint64_t aio_nbytes; |
de81a169 PB |
188 | #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */ |
189 | off_t aio_offset; | |
190 | int aio_type; | |
1efad060 FZ |
191 | int aio_fd2; |
192 | off_t aio_offset2; | |
de81a169 PB |
193 | } RawPosixAIOData; |
194 | ||
a167ba50 | 195 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 196 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
197 | #endif |
198 | ||
1de1ae0a CE |
199 | #if defined(__NetBSD__) |
200 | static int raw_normalize_devicepath(const char **filename) | |
201 | { | |
202 | static char namebuf[PATH_MAX]; | |
203 | const char *dp, *fname; | |
204 | struct stat sb; | |
205 | ||
206 | fname = *filename; | |
207 | dp = strrchr(fname, '/'); | |
208 | if (lstat(fname, &sb) < 0) { | |
209 | fprintf(stderr, "%s: stat failed: %s\n", | |
210 | fname, strerror(errno)); | |
211 | return -errno; | |
212 | } | |
213 | ||
214 | if (!S_ISBLK(sb.st_mode)) { | |
215 | return 0; | |
216 | } | |
217 | ||
218 | if (dp == NULL) { | |
219 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
220 | } else { | |
221 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
222 | (int)(dp - fname), fname, dp + 1); | |
223 | } | |
224 | fprintf(stderr, "%s is a block device", fname); | |
225 | *filename = namebuf; | |
226 | fprintf(stderr, ", using %s\n", *filename); | |
227 | ||
228 | return 0; | |
229 | } | |
230 | #else | |
231 | static int raw_normalize_devicepath(const char **filename) | |
232 | { | |
233 | return 0; | |
234 | } | |
235 | #endif | |
236 | ||
8a4ed0d1 ET |
237 | /* |
238 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
239 | */ | |
240 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 241 | { |
c25f53b0 | 242 | unsigned int sector_size; |
8a4ed0d1 | 243 | bool success = false; |
700f9ce0 | 244 | int i; |
c25f53b0 | 245 | |
8a4ed0d1 | 246 | errno = ENOTSUP; |
700f9ce0 | 247 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 248 | #ifdef BLKSSZGET |
700f9ce0 | 249 | BLKSSZGET, |
c25f53b0 PB |
250 | #endif |
251 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 252 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
253 | #endif |
254 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 255 | DIOCGSECTORSIZE, |
c25f53b0 | 256 | #endif |
700f9ce0 PM |
257 | }; |
258 | ||
259 | /* Try a few ioctls to get the right size */ | |
260 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
261 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
262 | *sector_size_p = sector_size; | |
263 | success = true; | |
264 | } | |
265 | } | |
8a4ed0d1 ET |
266 | |
267 | return success ? 0 : -errno; | |
268 | } | |
269 | ||
1a9335e4 ET |
270 | /** |
271 | * Get physical block size of @fd. | |
272 | * On success, store it in @blk_size and return 0. | |
273 | * On failure, return -errno. | |
274 | */ | |
275 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
276 | { | |
277 | #ifdef BLKPBSZGET | |
278 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
279 | return -errno; | |
280 | } | |
281 | return 0; | |
282 | #else | |
283 | return -ENOTSUP; | |
284 | #endif | |
285 | } | |
286 | ||
22d182e8 SH |
287 | /* Check if read is allowed with given memory buffer and length. |
288 | * | |
289 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
290 | */ | |
291 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
292 | { | |
293 | ssize_t ret = pread(fd, buf, len, 0); | |
294 | ||
295 | if (ret >= 0) { | |
296 | return true; | |
297 | } | |
298 | ||
299 | #ifdef __linux__ | |
300 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
301 | * other errors (e.g. real I/O error), which could happen on a failed | |
302 | * drive, since we only care about probing alignment. | |
303 | */ | |
304 | if (errno != EINVAL) { | |
305 | return true; | |
306 | } | |
307 | #endif | |
308 | ||
309 | return false; | |
310 | } | |
311 | ||
8a4ed0d1 ET |
312 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
313 | { | |
314 | BDRVRawState *s = bs->opaque; | |
315 | char *buf; | |
459b4e66 | 316 | size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); |
8a4ed0d1 | 317 | |
b192af8a | 318 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 319 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 320 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 321 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
322 | s->buf_align = 1; |
323 | return; | |
324 | } | |
325 | ||
a5b8dd2c | 326 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
327 | s->buf_align = 0; |
328 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
329 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
330 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 331 | } |
c25f53b0 PB |
332 | #ifdef CONFIG_XFS |
333 | if (s->is_xfs) { | |
334 | struct dioattr da; | |
df26a350 | 335 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
a5b8dd2c | 336 | bs->bl.request_alignment = da.d_miniosz; |
c25f53b0 PB |
337 | /* The kernel returns wrong information for d_mem */ |
338 | /* s->buf_align = da.d_mem; */ | |
339 | } | |
340 | } | |
341 | #endif | |
342 | ||
343 | /* If we could not get the sizes so far, we can only guess them */ | |
344 | if (!s->buf_align) { | |
345 | size_t align; | |
459b4e66 DL |
346 | buf = qemu_memalign(max_align, 2 * max_align); |
347 | for (align = 512; align <= max_align; align <<= 1) { | |
348 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
c25f53b0 PB |
349 | s->buf_align = align; |
350 | break; | |
351 | } | |
352 | } | |
353 | qemu_vfree(buf); | |
354 | } | |
355 | ||
a5b8dd2c | 356 | if (!bs->bl.request_alignment) { |
c25f53b0 | 357 | size_t align; |
459b4e66 DL |
358 | buf = qemu_memalign(s->buf_align, max_align); |
359 | for (align = 512; align <= max_align; align <<= 1) { | |
22d182e8 | 360 | if (raw_is_io_aligned(fd, buf, align)) { |
a5b8dd2c | 361 | bs->bl.request_alignment = align; |
c25f53b0 PB |
362 | break; |
363 | } | |
364 | } | |
365 | qemu_vfree(buf); | |
366 | } | |
df26a350 | 367 | |
a5b8dd2c | 368 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
369 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
370 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 371 | } |
c25f53b0 PB |
372 | } |
373 | ||
6a8dc042 JC |
374 | static void raw_parse_flags(int bdrv_flags, int *open_flags) |
375 | { | |
376 | assert(open_flags != NULL); | |
377 | ||
378 | *open_flags |= O_BINARY; | |
379 | *open_flags &= ~O_ACCMODE; | |
380 | if (bdrv_flags & BDRV_O_RDWR) { | |
381 | *open_flags |= O_RDWR; | |
382 | } else { | |
383 | *open_flags |= O_RDONLY; | |
384 | } | |
385 | ||
386 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
387 | * and O_DIRECT for no caching. */ | |
388 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
389 | *open_flags |= O_DIRECT; | |
390 | } | |
6a8dc042 JC |
391 | } |
392 | ||
078896a9 HR |
393 | static void raw_parse_filename(const char *filename, QDict *options, |
394 | Error **errp) | |
395 | { | |
03c320d8 | 396 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
397 | } |
398 | ||
c66a6157 KW |
399 | static QemuOptsList raw_runtime_opts = { |
400 | .name = "raw", | |
401 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
402 | .desc = { | |
403 | { | |
404 | .name = "filename", | |
405 | .type = QEMU_OPT_STRING, | |
406 | .help = "File name of the image", | |
407 | }, | |
0a4279d9 KW |
408 | { |
409 | .name = "aio", | |
410 | .type = QEMU_OPT_STRING, | |
411 | .help = "host AIO implementation (threads, native)", | |
412 | }, | |
16b48d5d FZ |
413 | { |
414 | .name = "locking", | |
415 | .type = QEMU_OPT_STRING, | |
416 | .help = "file locking mode (on/off/auto, default: auto)", | |
417 | }, | |
7c9e5276 PB |
418 | { |
419 | .name = "pr-manager", | |
420 | .type = QEMU_OPT_STRING, | |
421 | .help = "id of persistent reservation manager object (default: none)", | |
422 | }, | |
31be8a2a SH |
423 | { |
424 | .name = "x-check-cache-dropped", | |
425 | .type = QEMU_OPT_BOOL, | |
426 | .help = "check that page cache was dropped on live migration (default: off)" | |
427 | }, | |
c66a6157 KW |
428 | { /* end of list */ } |
429 | }, | |
430 | }; | |
431 | ||
432 | static int raw_open_common(BlockDriverState *bs, QDict *options, | |
e428e439 | 433 | int bdrv_flags, int open_flags, Error **errp) |
83f64091 FB |
434 | { |
435 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
436 | QemuOpts *opts; |
437 | Error *local_err = NULL; | |
8bfea15d | 438 | const char *filename = NULL; |
7c9e5276 | 439 | const char *str; |
0a4279d9 | 440 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 441 | int fd, ret; |
260a82e5 | 442 | struct stat st; |
244a5668 | 443 | OnOffAuto locking; |
83f64091 | 444 | |
87ea75d5 | 445 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
c66a6157 | 446 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 447 | if (local_err) { |
e428e439 | 448 | error_propagate(errp, local_err); |
c66a6157 KW |
449 | ret = -EINVAL; |
450 | goto fail; | |
451 | } | |
452 | ||
453 | filename = qemu_opt_get(opts, "filename"); | |
454 | ||
1de1ae0a CE |
455 | ret = raw_normalize_devicepath(&filename); |
456 | if (ret != 0) { | |
e428e439 | 457 | error_setg_errno(errp, -ret, "Could not normalize device path"); |
c66a6157 | 458 | goto fail; |
1de1ae0a CE |
459 | } |
460 | ||
0a4279d9 KW |
461 | aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) |
462 | ? BLOCKDEV_AIO_OPTIONS_NATIVE | |
463 | : BLOCKDEV_AIO_OPTIONS_THREADS; | |
f7abe0ec MAL |
464 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
465 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 466 | aio_default, &local_err); |
0a4279d9 KW |
467 | if (local_err) { |
468 | error_propagate(errp, local_err); | |
469 | ret = -EINVAL; | |
470 | goto fail; | |
471 | } | |
472 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); | |
473 | ||
f7abe0ec MAL |
474 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
475 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 476 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
477 | if (local_err) { |
478 | error_propagate(errp, local_err); | |
479 | ret = -EINVAL; | |
480 | goto fail; | |
481 | } | |
482 | switch (locking) { | |
483 | case ON_OFF_AUTO_ON: | |
484 | s->use_lock = true; | |
2b218f5d FZ |
485 | if (!qemu_has_ofd_lock()) { |
486 | fprintf(stderr, | |
487 | "File lock requested but OFD locking syscall is " | |
488 | "unavailable, falling back to POSIX file locks.\n" | |
489 | "Due to the implementation, locks can be lost " | |
490 | "unexpectedly.\n"); | |
491 | } | |
244a5668 FZ |
492 | break; |
493 | case ON_OFF_AUTO_OFF: | |
494 | s->use_lock = false; | |
495 | break; | |
496 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 497 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
498 | break; |
499 | default: | |
500 | abort(); | |
501 | } | |
502 | ||
7c9e5276 PB |
503 | str = qemu_opt_get(opts, "pr-manager"); |
504 | if (str) { | |
505 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
506 | if (local_err) { | |
507 | error_propagate(errp, local_err); | |
508 | ret = -EINVAL; | |
509 | goto fail; | |
510 | } | |
511 | } | |
512 | ||
31be8a2a SH |
513 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
514 | false); | |
515 | ||
6a8dc042 JC |
516 | s->open_flags = open_flags; |
517 | raw_parse_flags(bdrv_flags, &s->open_flags); | |
83f64091 | 518 | |
90babde0 | 519 | s->fd = -1; |
40ff6d7e | 520 | fd = qemu_open(filename, s->open_flags, 0644); |
19cb3738 FB |
521 | if (fd < 0) { |
522 | ret = -errno; | |
09237757 | 523 | error_setg_errno(errp, errno, "Could not open '%s'", filename); |
c66a6157 | 524 | if (ret == -EROFS) { |
19cb3738 | 525 | ret = -EACCES; |
c66a6157 KW |
526 | } |
527 | goto fail; | |
19cb3738 | 528 | } |
83f64091 | 529 | s->fd = fd; |
9ef91a67 | 530 | |
244a5668 FZ |
531 | s->lock_fd = -1; |
532 | if (s->use_lock) { | |
533 | fd = qemu_open(filename, s->open_flags); | |
534 | if (fd < 0) { | |
535 | ret = -errno; | |
536 | error_setg_errno(errp, errno, "Could not open '%s' for locking", | |
537 | filename); | |
538 | qemu_close(s->fd); | |
539 | goto fail; | |
540 | } | |
541 | s->lock_fd = fd; | |
542 | } | |
543 | s->perm = 0; | |
544 | s->shared_perm = BLK_PERM_ALL; | |
545 | ||
5c6c3a6c | 546 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 547 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
548 | if (s->use_linux_aio) { |
549 | if (!(s->open_flags & O_DIRECT)) { | |
550 | error_setg(errp, "aio=native was specified, but it requires " | |
551 | "cache.direct=on, which was not specified."); | |
552 | ret = -EINVAL; | |
553 | goto fail; | |
554 | } | |
555 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
556 | error_prepend(errp, "Unable to use native AIO: "); | |
557 | goto fail; | |
558 | } | |
96518254 | 559 | } |
1501ecc1 | 560 | #else |
0a4279d9 | 561 | if (s->use_linux_aio) { |
d657c0c2 KW |
562 | error_setg(errp, "aio=native was specified, but is not supported " |
563 | "in this build."); | |
564 | ret = -EINVAL; | |
565 | goto fail; | |
1501ecc1 SH |
566 | } |
567 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 568 | |
7ce21016 | 569 | s->has_discard = true; |
97a2ae34 | 570 | s->has_write_zeroes = true; |
3cad8307 RPM |
571 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { |
572 | s->needs_alignment = true; | |
573 | } | |
260a82e5 PB |
574 | |
575 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 576 | ret = -errno; |
260a82e5 PB |
577 | error_setg_errno(errp, errno, "Could not stat file"); |
578 | goto fail; | |
579 | } | |
580 | if (S_ISREG(st.st_mode)) { | |
581 | s->discard_zeroes = true; | |
d50d8222 | 582 | s->has_fallocate = true; |
260a82e5 | 583 | } |
d0b4503e PB |
584 | if (S_ISBLK(st.st_mode)) { |
585 | #ifdef BLKDISCARDZEROES | |
586 | unsigned int arg; | |
587 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
588 | s->discard_zeroes = true; | |
589 | } | |
590 | #endif | |
591 | #ifdef __linux__ | |
592 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
593 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 594 | * Same for BLKZEROOUT. |
d0b4503e PB |
595 | */ |
596 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
597 | s->discard_zeroes = false; | |
97a2ae34 | 598 | s->has_write_zeroes = false; |
d0b4503e PB |
599 | } |
600 | #endif | |
601 | } | |
3cad8307 RPM |
602 | #ifdef __FreeBSD__ |
603 | if (S_ISCHR(st.st_mode)) { | |
604 | /* | |
605 | * The file is a char device (disk), which on FreeBSD isn't behind | |
606 | * a pager, so force all requests to be aligned. This is needed | |
607 | * so QEMU makes sure all IO operations on the device are aligned | |
608 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
609 | */ | |
610 | s->needs_alignment = true; | |
611 | } | |
612 | #endif | |
260a82e5 | 613 | |
dce512de CH |
614 | #ifdef CONFIG_XFS |
615 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 616 | s->is_xfs = true; |
dce512de CH |
617 | } |
618 | #endif | |
619 | ||
e24d813b | 620 | bs->supported_zero_flags = s->discard_zeroes ? BDRV_REQ_MAY_UNMAP : 0; |
c66a6157 KW |
621 | ret = 0; |
622 | fail: | |
8bfea15d KW |
623 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
624 | unlink(filename); | |
625 | } | |
c66a6157 KW |
626 | qemu_opts_del(opts); |
627 | return ret; | |
83f64091 FB |
628 | } |
629 | ||
015a1036 HR |
630 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
631 | Error **errp) | |
90babde0 CH |
632 | { |
633 | BDRVRawState *s = bs->opaque; | |
634 | ||
635 | s->type = FTYPE_FILE; | |
9be38598 | 636 | return raw_open_common(bs, options, flags, 0, errp); |
90babde0 CH |
637 | } |
638 | ||
244a5668 FZ |
639 | typedef enum { |
640 | RAW_PL_PREPARE, | |
641 | RAW_PL_COMMIT, | |
642 | RAW_PL_ABORT, | |
643 | } RawPermLockOp; | |
644 | ||
645 | #define PERM_FOREACH(i) \ | |
646 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
647 | ||
648 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
649 | * file; if @unlock == true, also unlock the unneeded bytes. | |
650 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
651 | */ | |
d0a96155 | 652 | static int raw_apply_lock_bytes(int fd, |
244a5668 FZ |
653 | uint64_t perm_lock_bits, |
654 | uint64_t shared_perm_lock_bits, | |
655 | bool unlock, Error **errp) | |
656 | { | |
657 | int ret; | |
658 | int i; | |
659 | ||
660 | PERM_FOREACH(i) { | |
661 | int off = RAW_LOCK_PERM_BASE + i; | |
662 | if (perm_lock_bits & (1ULL << i)) { | |
d0a96155 | 663 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
664 | if (ret) { |
665 | error_setg(errp, "Failed to lock byte %d", off); | |
666 | return ret; | |
667 | } | |
668 | } else if (unlock) { | |
d0a96155 | 669 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
670 | if (ret) { |
671 | error_setg(errp, "Failed to unlock byte %d", off); | |
672 | return ret; | |
673 | } | |
674 | } | |
675 | } | |
676 | PERM_FOREACH(i) { | |
677 | int off = RAW_LOCK_SHARED_BASE + i; | |
678 | if (shared_perm_lock_bits & (1ULL << i)) { | |
d0a96155 | 679 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
680 | if (ret) { |
681 | error_setg(errp, "Failed to lock byte %d", off); | |
682 | return ret; | |
683 | } | |
684 | } else if (unlock) { | |
d0a96155 | 685 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
686 | if (ret) { |
687 | error_setg(errp, "Failed to unlock byte %d", off); | |
688 | return ret; | |
689 | } | |
690 | } | |
691 | } | |
692 | return 0; | |
693 | } | |
694 | ||
695 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 696 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
697 | Error **errp) |
698 | { | |
699 | int ret; | |
700 | int i; | |
701 | ||
702 | PERM_FOREACH(i) { | |
703 | int off = RAW_LOCK_SHARED_BASE + i; | |
704 | uint64_t p = 1ULL << i; | |
705 | if (perm & p) { | |
d0a96155 | 706 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
707 | if (ret) { |
708 | char *perm_name = bdrv_perm_names(p); | |
709 | error_setg(errp, | |
710 | "Failed to get \"%s\" lock", | |
711 | perm_name); | |
712 | g_free(perm_name); | |
713 | error_append_hint(errp, | |
714 | "Is another process using the image?\n"); | |
715 | return ret; | |
716 | } | |
717 | } | |
718 | } | |
719 | PERM_FOREACH(i) { | |
720 | int off = RAW_LOCK_PERM_BASE + i; | |
721 | uint64_t p = 1ULL << i; | |
722 | if (!(shared_perm & p)) { | |
d0a96155 | 723 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
724 | if (ret) { |
725 | char *perm_name = bdrv_perm_names(p); | |
726 | error_setg(errp, | |
727 | "Failed to get shared \"%s\" lock", | |
728 | perm_name); | |
729 | g_free(perm_name); | |
730 | error_append_hint(errp, | |
731 | "Is another process using the image?\n"); | |
732 | return ret; | |
733 | } | |
734 | } | |
735 | } | |
736 | return 0; | |
737 | } | |
738 | ||
739 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
740 | RawPermLockOp op, | |
741 | uint64_t new_perm, uint64_t new_shared, | |
742 | Error **errp) | |
743 | { | |
744 | BDRVRawState *s = bs->opaque; | |
745 | int ret = 0; | |
746 | Error *local_err = NULL; | |
747 | ||
748 | if (!s->use_lock) { | |
749 | return 0; | |
750 | } | |
751 | ||
752 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
753 | return 0; | |
754 | } | |
755 | ||
756 | assert(s->lock_fd > 0); | |
757 | ||
758 | switch (op) { | |
759 | case RAW_PL_PREPARE: | |
d0a96155 | 760 | ret = raw_apply_lock_bytes(s->lock_fd, s->perm | new_perm, |
244a5668 FZ |
761 | ~s->shared_perm | ~new_shared, |
762 | false, errp); | |
763 | if (!ret) { | |
d0a96155 | 764 | ret = raw_check_lock_bytes(s->lock_fd, new_perm, new_shared, errp); |
244a5668 FZ |
765 | if (!ret) { |
766 | return 0; | |
767 | } | |
768 | } | |
769 | op = RAW_PL_ABORT; | |
770 | /* fall through to unlock bytes. */ | |
771 | case RAW_PL_ABORT: | |
d0a96155 HR |
772 | raw_apply_lock_bytes(s->lock_fd, s->perm, ~s->shared_perm, |
773 | true, &local_err); | |
244a5668 FZ |
774 | if (local_err) { |
775 | /* Theoretically the above call only unlocks bytes and it cannot | |
776 | * fail. Something weird happened, report it. | |
777 | */ | |
778 | error_report_err(local_err); | |
779 | } | |
780 | break; | |
781 | case RAW_PL_COMMIT: | |
d0a96155 HR |
782 | raw_apply_lock_bytes(s->lock_fd, new_perm, ~new_shared, |
783 | true, &local_err); | |
244a5668 FZ |
784 | if (local_err) { |
785 | /* Theoretically the above call only unlocks bytes and it cannot | |
786 | * fail. Something weird happened, report it. | |
787 | */ | |
788 | error_report_err(local_err); | |
789 | } | |
790 | break; | |
791 | } | |
792 | return ret; | |
793 | } | |
794 | ||
eeb6b45d JC |
795 | static int raw_reopen_prepare(BDRVReopenState *state, |
796 | BlockReopenQueue *queue, Error **errp) | |
797 | { | |
798 | BDRVRawState *s; | |
4e6d13c9 | 799 | BDRVRawReopenState *rs; |
31be8a2a | 800 | QemuOpts *opts; |
eeb6b45d | 801 | int ret = 0; |
df26a350 | 802 | Error *local_err = NULL; |
eeb6b45d JC |
803 | |
804 | assert(state != NULL); | |
805 | assert(state->bs != NULL); | |
806 | ||
807 | s = state->bs->opaque; | |
808 | ||
5839e53b | 809 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 810 | rs = state->opaque; |
31be8a2a SH |
811 | rs->fd = -1; |
812 | ||
813 | /* Handle options changes */ | |
814 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
815 | qemu_opts_absorb_qdict(opts, state->options, &local_err); | |
816 | if (local_err) { | |
817 | error_propagate(errp, local_err); | |
818 | ret = -EINVAL; | |
819 | goto out; | |
820 | } | |
821 | ||
822 | rs->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", | |
823 | s->check_cache_dropped); | |
eeb6b45d | 824 | |
f709623b | 825 | if (s->type == FTYPE_CD) { |
4e6d13c9 | 826 | rs->open_flags |= O_NONBLOCK; |
1bc6b705 JC |
827 | } |
828 | ||
4e6d13c9 | 829 | raw_parse_flags(state->flags, &rs->open_flags); |
eeb6b45d | 830 | |
fdf263f6 | 831 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
eeb6b45d JC |
832 | #ifdef O_NOATIME |
833 | fcntl_flags |= O_NOATIME; | |
834 | #endif | |
835 | ||
fdf263f6 AF |
836 | #ifdef O_ASYNC |
837 | /* Not all operating systems have O_ASYNC, and those that don't | |
4e6d13c9 | 838 | * will not let us track the state into rs->open_flags (typically |
fdf263f6 AF |
839 | * you achieve the same effect with an ioctl, for example I_SETSIG |
840 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
841 | */ | |
842 | assert((s->open_flags & O_ASYNC) == 0); | |
843 | #endif | |
844 | ||
4e6d13c9 | 845 | if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
eeb6b45d | 846 | /* dup the original fd */ |
4e6d13c9 FZ |
847 | rs->fd = qemu_dup(s->fd); |
848 | if (rs->fd >= 0) { | |
849 | ret = fcntl_setfl(rs->fd, rs->open_flags); | |
eeb6b45d | 850 | if (ret) { |
4e6d13c9 FZ |
851 | qemu_close(rs->fd); |
852 | rs->fd = -1; | |
eeb6b45d JC |
853 | } |
854 | } | |
855 | } | |
856 | ||
857 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ | |
4e6d13c9 | 858 | if (rs->fd == -1) { |
bdd03cdf HR |
859 | const char *normalized_filename = state->bs->filename; |
860 | ret = raw_normalize_devicepath(&normalized_filename); | |
861 | if (ret < 0) { | |
862 | error_setg_errno(errp, -ret, "Could not normalize device path"); | |
863 | } else { | |
4e6d13c9 FZ |
864 | assert(!(rs->open_flags & O_CREAT)); |
865 | rs->fd = qemu_open(normalized_filename, rs->open_flags); | |
866 | if (rs->fd == -1) { | |
bdd03cdf HR |
867 | error_setg_errno(errp, errno, "Could not reopen file"); |
868 | ret = -1; | |
869 | } | |
eeb6b45d JC |
870 | } |
871 | } | |
df26a350 KW |
872 | |
873 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
874 | * alignment with the new fd. */ | |
4e6d13c9 FZ |
875 | if (rs->fd != -1) { |
876 | raw_probe_alignment(state->bs, rs->fd, &local_err); | |
df26a350 | 877 | if (local_err) { |
4e6d13c9 FZ |
878 | qemu_close(rs->fd); |
879 | rs->fd = -1; | |
df26a350 KW |
880 | error_propagate(errp, local_err); |
881 | ret = -EINVAL; | |
882 | } | |
883 | } | |
884 | ||
31be8a2a SH |
885 | out: |
886 | qemu_opts_del(opts); | |
eeb6b45d JC |
887 | return ret; |
888 | } | |
889 | ||
eeb6b45d JC |
890 | static void raw_reopen_commit(BDRVReopenState *state) |
891 | { | |
4e6d13c9 | 892 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
893 | BDRVRawState *s = state->bs->opaque; |
894 | ||
31be8a2a | 895 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 896 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
897 | |
898 | qemu_close(s->fd); | |
4e6d13c9 | 899 | s->fd = rs->fd; |
eeb6b45d JC |
900 | |
901 | g_free(state->opaque); | |
902 | state->opaque = NULL; | |
903 | } | |
904 | ||
905 | ||
906 | static void raw_reopen_abort(BDRVReopenState *state) | |
907 | { | |
4e6d13c9 | 908 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
909 | |
910 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 911 | if (rs == NULL) { |
eeb6b45d JC |
912 | return; |
913 | } | |
914 | ||
4e6d13c9 FZ |
915 | if (rs->fd >= 0) { |
916 | qemu_close(rs->fd); | |
917 | rs->fd = -1; | |
eeb6b45d JC |
918 | } |
919 | g_free(state->opaque); | |
920 | state->opaque = NULL; | |
921 | } | |
922 | ||
48265250 | 923 | static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd) |
6f607174 FZ |
924 | { |
925 | #ifdef BLKSECTGET | |
48265250 EF |
926 | int max_bytes = 0; |
927 | short max_sectors = 0; | |
928 | if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
929 | return max_bytes; | |
930 | } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) { | |
931 | return max_sectors << BDRV_SECTOR_BITS; | |
6f607174 FZ |
932 | } else { |
933 | return -errno; | |
934 | } | |
935 | #else | |
936 | return -ENOSYS; | |
937 | #endif | |
938 | } | |
939 | ||
9103f1ce FZ |
940 | static int hdev_get_max_segments(const struct stat *st) |
941 | { | |
942 | #ifdef CONFIG_LINUX | |
943 | char buf[32]; | |
944 | const char *end; | |
945 | char *sysfspath; | |
946 | int ret; | |
947 | int fd = -1; | |
948 | long max_segments; | |
949 | ||
950 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", | |
951 | major(st->st_rdev), minor(st->st_rdev)); | |
952 | fd = open(sysfspath, O_RDONLY); | |
953 | if (fd == -1) { | |
954 | ret = -errno; | |
955 | goto out; | |
956 | } | |
957 | do { | |
69583490 | 958 | ret = read(fd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
959 | } while (ret == -1 && errno == EINTR); |
960 | if (ret < 0) { | |
961 | ret = -errno; | |
962 | goto out; | |
963 | } else if (ret == 0) { | |
964 | ret = -EIO; | |
965 | goto out; | |
966 | } | |
967 | buf[ret] = 0; | |
968 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
969 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
970 | if (ret == 0 && end && *end == '\n') { | |
971 | ret = max_segments; | |
972 | } | |
973 | ||
974 | out: | |
fed414df FZ |
975 | if (fd != -1) { |
976 | close(fd); | |
977 | } | |
9103f1ce FZ |
978 | g_free(sysfspath); |
979 | return ret; | |
980 | #else | |
981 | return -ENOTSUP; | |
982 | #endif | |
983 | } | |
984 | ||
3baca891 | 985 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
986 | { |
987 | BDRVRawState *s = bs->opaque; | |
6f607174 FZ |
988 | struct stat st; |
989 | ||
990 | if (!fstat(s->fd, &st)) { | |
c4c41a0a | 991 | if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) { |
48265250 EF |
992 | int ret = hdev_get_max_transfer_length(bs, s->fd); |
993 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
994 | bs->bl.max_transfer = pow2floor(ret); | |
6f607174 | 995 | } |
9103f1ce FZ |
996 | ret = hdev_get_max_segments(&st); |
997 | if (ret > 0) { | |
998 | bs->bl.max_transfer = MIN(bs->bl.max_transfer, | |
999 | ret * getpagesize()); | |
1000 | } | |
6f607174 FZ |
1001 | } |
1002 | } | |
eeb6b45d | 1003 | |
df26a350 | 1004 | raw_probe_alignment(bs, s->fd, errp); |
4196d2f0 | 1005 | bs->bl.min_mem_alignment = s->buf_align; |
459b4e66 | 1006 | bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); |
c25f53b0 | 1007 | } |
83f64091 | 1008 | |
1a9335e4 ET |
1009 | static int check_for_dasd(int fd) |
1010 | { | |
1011 | #ifdef BIODASDINFO2 | |
1012 | struct dasd_information2_t info = {0}; | |
1013 | ||
1014 | return ioctl(fd, BIODASDINFO2, &info); | |
1015 | #else | |
1016 | return -1; | |
1017 | #endif | |
1018 | } | |
1019 | ||
1020 | /** | |
1021 | * Try to get @bs's logical and physical block size. | |
1022 | * On success, store them in @bsz and return zero. | |
1023 | * On failure, return negative errno. | |
1024 | */ | |
1025 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1026 | { | |
1027 | BDRVRawState *s = bs->opaque; | |
1028 | int ret; | |
1029 | ||
1030 | /* If DASD, get blocksizes */ | |
1031 | if (check_for_dasd(s->fd) < 0) { | |
1032 | return -ENOTSUP; | |
1033 | } | |
1034 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1035 | if (ret < 0) { | |
1036 | return ret; | |
1037 | } | |
1038 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1039 | } | |
1040 | ||
1041 | /** | |
1042 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1043 | * On success, store them in @geo and return 0. | |
1044 | * On failure return -errno. | |
1045 | * (Allows block driver to assign default geometry values that guest sees) | |
1046 | */ | |
1047 | #ifdef __linux__ | |
1048 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1049 | { | |
1050 | BDRVRawState *s = bs->opaque; | |
1051 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1052 | |
1053 | /* If DASD, get its geometry */ | |
1054 | if (check_for_dasd(s->fd) < 0) { | |
1055 | return -ENOTSUP; | |
1056 | } | |
1057 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1058 | return -errno; | |
1059 | } | |
1060 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1061 | (e.g. certain multipath setups) */ | |
1062 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1063 | return -ENOTSUP; | |
1064 | } | |
1065 | /* Do not return a geometry for partition */ | |
1066 | if (ioctl_geo.start != 0) { | |
1067 | return -ENOTSUP; | |
1068 | } | |
1069 | geo->heads = ioctl_geo.heads; | |
1070 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1071 | geo->cylinders = ioctl_geo.cylinders; |
1072 | ||
1073 | return 0; | |
1074 | } | |
1075 | #else /* __linux__ */ | |
1076 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1077 | { | |
1078 | return -ENOTSUP; | |
1079 | } | |
1080 | #endif | |
1081 | ||
de81a169 PB |
1082 | static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb) |
1083 | { | |
1084 | int ret; | |
1085 | ||
1086 | ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); | |
1087 | if (ret == -1) { | |
1088 | return -errno; | |
1089 | } | |
1090 | ||
b608c8dc | 1091 | return 0; |
de81a169 PB |
1092 | } |
1093 | ||
1094 | static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb) | |
1095 | { | |
e5bcf967 | 1096 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1097 | int ret; |
1098 | ||
e5bcf967 KW |
1099 | if (s->page_cache_inconsistent) { |
1100 | return -EIO; | |
1101 | } | |
1102 | ||
de81a169 PB |
1103 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1104 | if (ret == -1) { | |
e5bcf967 KW |
1105 | /* There is no clear definition of the semantics of a failing fsync(), |
1106 | * so we may have to assume the worst. The sad truth is that this | |
1107 | * assumption is correct for Linux. Some pages are now probably marked | |
1108 | * clean in the page cache even though they are inconsistent with the | |
1109 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1110 | * further writeback attempt will be made. We can't get back to a state | |
1111 | * in which we know what is on disk (we would have to rewrite | |
1112 | * everything that was touched since the last fdatasync() at least), so | |
1113 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1114 | * really defined, I have little hope that other OSes are doing better. | |
1115 | * | |
1116 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1117 | * cache. */ | |
1118 | if ((s->open_flags & O_DIRECT) == 0) { | |
1119 | s->page_cache_inconsistent = true; | |
1120 | } | |
de81a169 PB |
1121 | return -errno; |
1122 | } | |
1123 | return 0; | |
1124 | } | |
1125 | ||
1126 | #ifdef CONFIG_PREADV | |
1127 | ||
1128 | static bool preadv_present = true; | |
1129 | ||
1130 | static ssize_t | |
1131 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1132 | { | |
1133 | return preadv(fd, iov, nr_iov, offset); | |
1134 | } | |
1135 | ||
1136 | static ssize_t | |
1137 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1138 | { | |
1139 | return pwritev(fd, iov, nr_iov, offset); | |
1140 | } | |
1141 | ||
1142 | #else | |
1143 | ||
1144 | static bool preadv_present = false; | |
1145 | ||
1146 | static ssize_t | |
1147 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1148 | { | |
1149 | return -ENOSYS; | |
1150 | } | |
1151 | ||
1152 | static ssize_t | |
1153 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1154 | { | |
1155 | return -ENOSYS; | |
1156 | } | |
1157 | ||
1158 | #endif | |
1159 | ||
1160 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1161 | { | |
1162 | ssize_t len; | |
1163 | ||
1164 | do { | |
1165 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1166 | len = qemu_pwritev(aiocb->aio_fildes, | |
1167 | aiocb->aio_iov, | |
1168 | aiocb->aio_niov, | |
1169 | aiocb->aio_offset); | |
1170 | else | |
1171 | len = qemu_preadv(aiocb->aio_fildes, | |
1172 | aiocb->aio_iov, | |
1173 | aiocb->aio_niov, | |
1174 | aiocb->aio_offset); | |
1175 | } while (len == -1 && errno == EINTR); | |
1176 | ||
1177 | if (len == -1) { | |
1178 | return -errno; | |
1179 | } | |
1180 | return len; | |
1181 | } | |
1182 | ||
1183 | /* | |
1184 | * Read/writes the data to/from a given linear buffer. | |
1185 | * | |
1186 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1187 | * reads are only returned if the end of the file is reached. | |
1188 | */ | |
1189 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1190 | { | |
1191 | ssize_t offset = 0; | |
1192 | ssize_t len; | |
1193 | ||
1194 | while (offset < aiocb->aio_nbytes) { | |
1195 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1196 | len = pwrite(aiocb->aio_fildes, | |
1197 | (const char *)buf + offset, | |
1198 | aiocb->aio_nbytes - offset, | |
1199 | aiocb->aio_offset + offset); | |
1200 | } else { | |
1201 | len = pread(aiocb->aio_fildes, | |
1202 | buf + offset, | |
1203 | aiocb->aio_nbytes - offset, | |
1204 | aiocb->aio_offset + offset); | |
1205 | } | |
1206 | if (len == -1 && errno == EINTR) { | |
1207 | continue; | |
61ed73cf SH |
1208 | } else if (len == -1 && errno == EINVAL && |
1209 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1210 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1211 | offset > 0) { | |
1212 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1213 | * after a short read. Assume that O_DIRECT short reads only occur | |
1214 | * at EOF. Therefore this is a short read, not an I/O error. | |
1215 | */ | |
1216 | break; | |
de81a169 PB |
1217 | } else if (len == -1) { |
1218 | offset = -errno; | |
1219 | break; | |
1220 | } else if (len == 0) { | |
1221 | break; | |
1222 | } | |
1223 | offset += len; | |
1224 | } | |
1225 | ||
1226 | return offset; | |
1227 | } | |
1228 | ||
1229 | static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb) | |
1230 | { | |
1231 | ssize_t nbytes; | |
1232 | char *buf; | |
1233 | ||
1234 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1235 | /* | |
1236 | * If there is just a single buffer, and it is properly aligned | |
1237 | * we can just use plain pread/pwrite without any problems. | |
1238 | */ | |
1239 | if (aiocb->aio_niov == 1) { | |
1240 | return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); | |
1241 | } | |
1242 | /* | |
1243 | * We have more than one iovec, and all are properly aligned. | |
1244 | * | |
1245 | * Try preadv/pwritev first and fall back to linearizing the | |
1246 | * buffer if it's not supported. | |
1247 | */ | |
1248 | if (preadv_present) { | |
1249 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1250 | if (nbytes == aiocb->aio_nbytes || | |
1251 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
1252 | return nbytes; | |
1253 | } | |
1254 | preadv_present = false; | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1259 | * using these interfaces. For now retry using plain | |
1260 | * pread/pwrite? | |
1261 | */ | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * Ok, we have to do it the hard way, copy all segments into | |
1266 | * a single aligned buffer. | |
1267 | */ | |
50d4a858 KW |
1268 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1269 | if (buf == NULL) { | |
1270 | return -ENOMEM; | |
1271 | } | |
1272 | ||
de81a169 PB |
1273 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1274 | char *p = buf; | |
1275 | int i; | |
1276 | ||
1277 | for (i = 0; i < aiocb->aio_niov; ++i) { | |
1278 | memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); | |
1279 | p += aiocb->aio_iov[i].iov_len; | |
1280 | } | |
8eb029c2 | 1281 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1282 | } |
1283 | ||
1284 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1285 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1286 | char *p = buf; | |
1287 | size_t count = aiocb->aio_nbytes, copy; | |
1288 | int i; | |
1289 | ||
1290 | for (i = 0; i < aiocb->aio_niov && count; ++i) { | |
1291 | copy = count; | |
1292 | if (copy > aiocb->aio_iov[i].iov_len) { | |
1293 | copy = aiocb->aio_iov[i].iov_len; | |
1294 | } | |
1295 | memcpy(aiocb->aio_iov[i].iov_base, p, copy); | |
8eb029c2 | 1296 | assert(count >= copy); |
de81a169 PB |
1297 | p += copy; |
1298 | count -= copy; | |
1299 | } | |
8eb029c2 | 1300 | assert(count == 0); |
de81a169 PB |
1301 | } |
1302 | qemu_vfree(buf); | |
1303 | ||
1304 | return nbytes; | |
1305 | } | |
1306 | ||
8238010b | 1307 | #ifdef CONFIG_XFS |
97a2ae34 PB |
1308 | static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) |
1309 | { | |
1310 | struct xfs_flock64 fl; | |
bcb22555 | 1311 | int err; |
97a2ae34 PB |
1312 | |
1313 | memset(&fl, 0, sizeof(fl)); | |
1314 | fl.l_whence = SEEK_SET; | |
1315 | fl.l_start = offset; | |
1316 | fl.l_len = bytes; | |
1317 | ||
1318 | if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { | |
bcb22555 DA |
1319 | err = errno; |
1320 | DPRINTF("cannot write zero range (%s)\n", strerror(errno)); | |
1321 | return -err; | |
97a2ae34 PB |
1322 | } |
1323 | ||
1324 | return 0; | |
1325 | } | |
1326 | ||
8238010b PB |
1327 | static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) |
1328 | { | |
1329 | struct xfs_flock64 fl; | |
bcb22555 | 1330 | int err; |
8238010b PB |
1331 | |
1332 | memset(&fl, 0, sizeof(fl)); | |
1333 | fl.l_whence = SEEK_SET; | |
1334 | fl.l_start = offset; | |
1335 | fl.l_len = bytes; | |
1336 | ||
1337 | if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { | |
bcb22555 DA |
1338 | err = errno; |
1339 | DPRINTF("cannot punch hole (%s)\n", strerror(errno)); | |
1340 | return -err; | |
8238010b PB |
1341 | } |
1342 | ||
1343 | return 0; | |
1344 | } | |
1345 | #endif | |
1346 | ||
1486df0e DL |
1347 | static int translate_err(int err) |
1348 | { | |
1349 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1350 | err == -ENOTTY) { | |
1351 | err = -ENOTSUP; | |
1352 | } | |
1353 | return err; | |
1354 | } | |
1355 | ||
d50d8222 | 1356 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1357 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1358 | { | |
1359 | do { | |
1360 | if (fallocate(fd, mode, offset, len) == 0) { | |
1361 | return 0; | |
1362 | } | |
1363 | } while (errno == EINTR); | |
1364 | return translate_err(-errno); | |
1365 | } | |
1366 | #endif | |
1367 | ||
37cc9f7f | 1368 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1369 | { |
37cc9f7f | 1370 | int ret = -ENOTSUP; |
97a2ae34 PB |
1371 | BDRVRawState *s = aiocb->bs->opaque; |
1372 | ||
37cc9f7f | 1373 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1374 | return -ENOTSUP; |
1375 | } | |
1376 | ||
97a2ae34 | 1377 | #ifdef BLKZEROOUT |
37cc9f7f DL |
1378 | do { |
1379 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1380 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1381 | return 0; | |
97a2ae34 | 1382 | } |
37cc9f7f DL |
1383 | } while (errno == EINTR); |
1384 | ||
1385 | ret = translate_err(-errno); | |
97a2ae34 | 1386 | #endif |
97a2ae34 | 1387 | |
1486df0e | 1388 | if (ret == -ENOTSUP) { |
97a2ae34 | 1389 | s->has_write_zeroes = false; |
97a2ae34 PB |
1390 | } |
1391 | return ret; | |
1392 | } | |
1393 | ||
37cc9f7f DL |
1394 | static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb) |
1395 | { | |
a6dcf097 | 1396 | #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS) |
37cc9f7f | 1397 | BDRVRawState *s = aiocb->bs->opaque; |
a6dcf097 | 1398 | #endif |
70d9110b DL |
1399 | #ifdef CONFIG_FALLOCATE |
1400 | int64_t len; | |
1401 | #endif | |
37cc9f7f DL |
1402 | |
1403 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1404 | return handle_aiocb_write_zeroes_block(aiocb); | |
1405 | } | |
1406 | ||
1407 | #ifdef CONFIG_XFS | |
1408 | if (s->is_xfs) { | |
1409 | return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1410 | } | |
1411 | #endif | |
1412 | ||
b953f075 DL |
1413 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1414 | if (s->has_write_zeroes) { | |
1415 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1416 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1417 | if (ret == 0 || ret != -ENOTSUP) { | |
1418 | return ret; | |
1419 | } | |
1420 | s->has_write_zeroes = false; | |
1421 | } | |
1422 | #endif | |
1423 | ||
1cdc3239 DL |
1424 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1425 | if (s->has_discard && s->has_fallocate) { | |
1426 | int ret = do_fallocate(s->fd, | |
1427 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1428 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1429 | if (ret == 0) { | |
1430 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1431 | if (ret == 0 || ret != -ENOTSUP) { | |
1432 | return ret; | |
1433 | } | |
1434 | s->has_fallocate = false; | |
1435 | } else if (ret != -ENOTSUP) { | |
1436 | return ret; | |
1437 | } else { | |
1438 | s->has_discard = false; | |
1439 | } | |
1440 | } | |
1441 | #endif | |
1442 | ||
d50d8222 | 1443 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1444 | /* Last resort: we are trying to extend the file with zeroed data. This |
1445 | * can be done via fallocate(fd, 0) */ | |
1446 | len = bdrv_getlength(aiocb->bs); | |
1447 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1448 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1449 | if (ret == 0 || ret != -ENOTSUP) { | |
1450 | return ret; | |
1451 | } | |
1452 | s->has_fallocate = false; | |
1453 | } | |
1454 | #endif | |
1455 | ||
37cc9f7f DL |
1456 | return -ENOTSUP; |
1457 | } | |
1458 | ||
1efad060 FZ |
1459 | #ifndef HAVE_COPY_FILE_RANGE |
1460 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1461 | off_t *out_off, size_t len, unsigned int flags) | |
1462 | { | |
1463 | #ifdef __NR_copy_file_range | |
1464 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1465 | out_off, len, flags); | |
1466 | #else | |
1467 | errno = ENOSYS; | |
1468 | return -1; | |
1469 | #endif | |
1470 | } | |
1471 | #endif | |
1472 | ||
1473 | static ssize_t handle_aiocb_copy_range(RawPosixAIOData *aiocb) | |
1474 | { | |
1475 | uint64_t bytes = aiocb->aio_nbytes; | |
1476 | off_t in_off = aiocb->aio_offset; | |
1477 | off_t out_off = aiocb->aio_offset2; | |
1478 | ||
1479 | while (bytes) { | |
1480 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
1481 | aiocb->aio_fd2, &out_off, | |
1482 | bytes, 0); | |
1483 | if (ret == -EINTR) { | |
1484 | continue; | |
1485 | } | |
1486 | if (ret < 0) { | |
1487 | if (errno == ENOSYS) { | |
1488 | return -ENOTSUP; | |
1489 | } else { | |
1490 | return -errno; | |
1491 | } | |
1492 | } | |
1493 | if (!ret) { | |
1494 | /* No progress (e.g. when beyond EOF), fall back to buffer I/O. */ | |
1495 | return -ENOTSUP; | |
1496 | } | |
1497 | bytes -= ret; | |
1498 | } | |
1499 | return 0; | |
1500 | } | |
1501 | ||
8238010b PB |
1502 | static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb) |
1503 | { | |
1504 | int ret = -EOPNOTSUPP; | |
1505 | BDRVRawState *s = aiocb->bs->opaque; | |
1506 | ||
7ce21016 PB |
1507 | if (!s->has_discard) { |
1508 | return -ENOTSUP; | |
8238010b PB |
1509 | } |
1510 | ||
1511 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1512 | #ifdef BLKDISCARD | |
1513 | do { | |
1514 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1515 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1516 | return 0; | |
1517 | } | |
1518 | } while (errno == EINTR); | |
1519 | ||
1520 | ret = -errno; | |
1521 | #endif | |
1522 | } else { | |
1523 | #ifdef CONFIG_XFS | |
1524 | if (s->is_xfs) { | |
1525 | return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1526 | } | |
1527 | #endif | |
1528 | ||
1529 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
0b991712 DL |
1530 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1531 | aiocb->aio_offset, aiocb->aio_nbytes); | |
8238010b PB |
1532 | #endif |
1533 | } | |
1534 | ||
1486df0e DL |
1535 | ret = translate_err(ret); |
1536 | if (ret == -ENOTSUP) { | |
7ce21016 | 1537 | s->has_discard = false; |
8238010b PB |
1538 | } |
1539 | return ret; | |
1540 | } | |
1541 | ||
de81a169 PB |
1542 | static int aio_worker(void *arg) |
1543 | { | |
1544 | RawPosixAIOData *aiocb = arg; | |
1545 | ssize_t ret = 0; | |
1546 | ||
1547 | switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { | |
1548 | case QEMU_AIO_READ: | |
1549 | ret = handle_aiocb_rw(aiocb); | |
c0191e76 | 1550 | if (ret >= 0 && ret < aiocb->aio_nbytes) { |
de81a169 PB |
1551 | iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret, |
1552 | 0, aiocb->aio_nbytes - ret); | |
1553 | ||
1554 | ret = aiocb->aio_nbytes; | |
1555 | } | |
1556 | if (ret == aiocb->aio_nbytes) { | |
1557 | ret = 0; | |
1558 | } else if (ret >= 0 && ret < aiocb->aio_nbytes) { | |
1559 | ret = -EINVAL; | |
1560 | } | |
1561 | break; | |
1562 | case QEMU_AIO_WRITE: | |
1563 | ret = handle_aiocb_rw(aiocb); | |
1564 | if (ret == aiocb->aio_nbytes) { | |
1565 | ret = 0; | |
1566 | } else if (ret >= 0 && ret < aiocb->aio_nbytes) { | |
1567 | ret = -EINVAL; | |
1568 | } | |
1569 | break; | |
1570 | case QEMU_AIO_FLUSH: | |
1571 | ret = handle_aiocb_flush(aiocb); | |
1572 | break; | |
1573 | case QEMU_AIO_IOCTL: | |
1574 | ret = handle_aiocb_ioctl(aiocb); | |
1575 | break; | |
8238010b PB |
1576 | case QEMU_AIO_DISCARD: |
1577 | ret = handle_aiocb_discard(aiocb); | |
1578 | break; | |
97a2ae34 PB |
1579 | case QEMU_AIO_WRITE_ZEROES: |
1580 | ret = handle_aiocb_write_zeroes(aiocb); | |
1581 | break; | |
1efad060 FZ |
1582 | case QEMU_AIO_COPY_RANGE: |
1583 | ret = handle_aiocb_copy_range(aiocb); | |
1584 | break; | |
de81a169 PB |
1585 | default: |
1586 | fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); | |
1587 | ret = -EINVAL; | |
1588 | break; | |
1589 | } | |
1590 | ||
c84b3192 | 1591 | g_free(aiocb); |
de81a169 PB |
1592 | return ret; |
1593 | } | |
1594 | ||
1efad060 FZ |
1595 | static int paio_submit_co_full(BlockDriverState *bs, int fd, |
1596 | int64_t offset, int fd2, int64_t offset2, | |
1597 | QEMUIOVector *qiov, | |
1598 | int bytes, int type) | |
260a82e5 | 1599 | { |
c84b3192 | 1600 | RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); |
260a82e5 PB |
1601 | ThreadPool *pool; |
1602 | ||
1603 | acb->bs = bs; | |
1604 | acb->aio_type = type; | |
1605 | acb->aio_fildes = fd; | |
1efad060 FZ |
1606 | acb->aio_fd2 = fd2; |
1607 | acb->aio_offset2 = offset2; | |
260a82e5 | 1608 | |
f5a5ca79 | 1609 | acb->aio_nbytes = bytes; |
2ffa76c2 | 1610 | acb->aio_offset = offset; |
8eb029c2 | 1611 | |
260a82e5 PB |
1612 | if (qiov) { |
1613 | acb->aio_iov = qiov->iov; | |
1614 | acb->aio_niov = qiov->niov; | |
f5a5ca79 | 1615 | assert(qiov->size == bytes); |
260a82e5 | 1616 | } |
260a82e5 | 1617 | |
f5a5ca79 | 1618 | trace_paio_submit_co(offset, bytes, type); |
260a82e5 PB |
1619 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
1620 | return thread_pool_submit_co(pool, aio_worker, acb); | |
1621 | } | |
1622 | ||
1efad060 FZ |
1623 | static inline int paio_submit_co(BlockDriverState *bs, int fd, |
1624 | int64_t offset, QEMUIOVector *qiov, | |
1625 | int bytes, int type) | |
1626 | { | |
1627 | return paio_submit_co_full(bs, fd, offset, -1, 0, qiov, bytes, type); | |
1628 | } | |
1629 | ||
7c84b1b8 | 1630 | static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd, |
f5a5ca79 | 1631 | int64_t offset, QEMUIOVector *qiov, int bytes, |
097310b5 | 1632 | BlockCompletionFunc *cb, void *opaque, int type) |
de81a169 | 1633 | { |
c84b3192 | 1634 | RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); |
c4d9d196 | 1635 | ThreadPool *pool; |
de81a169 PB |
1636 | |
1637 | acb->bs = bs; | |
1638 | acb->aio_type = type; | |
1639 | acb->aio_fildes = fd; | |
1640 | ||
f5a5ca79 | 1641 | acb->aio_nbytes = bytes; |
36e3b2e7 | 1642 | acb->aio_offset = offset; |
8eb029c2 | 1643 | |
de81a169 PB |
1644 | if (qiov) { |
1645 | acb->aio_iov = qiov->iov; | |
1646 | acb->aio_niov = qiov->niov; | |
8eb029c2 | 1647 | assert(qiov->size == acb->aio_nbytes); |
de81a169 | 1648 | } |
de81a169 | 1649 | |
f5a5ca79 | 1650 | trace_paio_submit(acb, opaque, offset, bytes, type); |
c4d9d196 SH |
1651 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
1652 | return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); | |
de81a169 PB |
1653 | } |
1654 | ||
9d52aa3c KW |
1655 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
1656 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 1657 | { |
ce1a14dc | 1658 | BDRVRawState *s = bs->opaque; |
ce1a14dc | 1659 | |
19cb3738 | 1660 | if (fd_open(bs) < 0) |
2174f12b | 1661 | return -EIO; |
19cb3738 | 1662 | |
f141eafe | 1663 | /* |
3cad8307 RPM |
1664 | * Check if the underlying device requires requests to be aligned, |
1665 | * and if the request we are trying to submit is aligned or not. | |
1666 | * If this is the case tell the low-level driver that it needs | |
1667 | * to copy the buffer. | |
f141eafe | 1668 | */ |
3cad8307 | 1669 | if (s->needs_alignment) { |
c53b1c51 | 1670 | if (!bdrv_qiov_is_aligned(bs, qiov)) { |
5c6c3a6c | 1671 | type |= QEMU_AIO_MISALIGNED; |
e44bd6fc | 1672 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 1673 | } else if (s->use_linux_aio) { |
0187f5c9 | 1674 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
9d52aa3c | 1675 | assert(qiov->size == bytes); |
0187f5c9 | 1676 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type); |
e44bd6fc | 1677 | #endif |
5c6c3a6c | 1678 | } |
9ef91a67 | 1679 | } |
f141eafe | 1680 | |
9d52aa3c | 1681 | return paio_submit_co(bs, s->fd, offset, qiov, bytes, type); |
2174f12b KW |
1682 | } |
1683 | ||
9d52aa3c KW |
1684 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, |
1685 | uint64_t bytes, QEMUIOVector *qiov, | |
1686 | int flags) | |
2174f12b | 1687 | { |
9d52aa3c | 1688 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
1689 | } |
1690 | ||
9d52aa3c KW |
1691 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, |
1692 | uint64_t bytes, QEMUIOVector *qiov, | |
1693 | int flags) | |
2174f12b | 1694 | { |
9d52aa3c KW |
1695 | assert(flags == 0); |
1696 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
1697 | } |
1698 | ||
1b3abdcc ML |
1699 | static void raw_aio_plug(BlockDriverState *bs) |
1700 | { | |
1701 | #ifdef CONFIG_LINUX_AIO | |
0a4279d9 KW |
1702 | BDRVRawState *s = bs->opaque; |
1703 | if (s->use_linux_aio) { | |
0187f5c9 PB |
1704 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1705 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
1706 | } |
1707 | #endif | |
1708 | } | |
1709 | ||
1710 | static void raw_aio_unplug(BlockDriverState *bs) | |
1711 | { | |
1712 | #ifdef CONFIG_LINUX_AIO | |
0a4279d9 KW |
1713 | BDRVRawState *s = bs->opaque; |
1714 | if (s->use_linux_aio) { | |
0187f5c9 PB |
1715 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1716 | laio_io_unplug(bs, aio); | |
1b3abdcc ML |
1717 | } |
1718 | #endif | |
1719 | } | |
1720 | ||
7c84b1b8 | 1721 | static BlockAIOCB *raw_aio_flush(BlockDriverState *bs, |
097310b5 | 1722 | BlockCompletionFunc *cb, void *opaque) |
b2e12bc6 CH |
1723 | { |
1724 | BDRVRawState *s = bs->opaque; | |
1725 | ||
1726 | if (fd_open(bs) < 0) | |
1727 | return NULL; | |
1728 | ||
1e5b9d2f | 1729 | return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH); |
b2e12bc6 CH |
1730 | } |
1731 | ||
ed6e2161 NA |
1732 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
1733 | AioContext *new_context) | |
1734 | { | |
1735 | #ifdef CONFIG_LINUX_AIO | |
1736 | BDRVRawState *s = bs->opaque; | |
1737 | if (s->use_linux_aio) { | |
1738 | Error *local_err; | |
1739 | if (!aio_setup_linux_aio(new_context, &local_err)) { | |
1740 | error_reportf_err(local_err, "Unable to use native AIO, " | |
1741 | "falling back to thread pool: "); | |
1742 | s->use_linux_aio = false; | |
1743 | } | |
1744 | } | |
1745 | #endif | |
1746 | } | |
1747 | ||
83f64091 FB |
1748 | static void raw_close(BlockDriverState *bs) |
1749 | { | |
1750 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 1751 | |
19cb3738 | 1752 | if (s->fd >= 0) { |
2e1e79da | 1753 | qemu_close(s->fd); |
19cb3738 FB |
1754 | s->fd = -1; |
1755 | } | |
244a5668 FZ |
1756 | if (s->lock_fd >= 0) { |
1757 | qemu_close(s->lock_fd); | |
1758 | s->lock_fd = -1; | |
1759 | } | |
83f64091 FB |
1760 | } |
1761 | ||
d0bc9e5d HR |
1762 | /** |
1763 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
1764 | * new space according to @prealloc. | |
1765 | * | |
1766 | * Returns: 0 on success, -errno on failure. | |
1767 | */ | |
9f63b07e HR |
1768 | static int raw_regular_truncate(int fd, int64_t offset, PreallocMode prealloc, |
1769 | Error **errp) | |
1770 | { | |
1771 | int result = 0; | |
d0bc9e5d HR |
1772 | int64_t current_length = 0; |
1773 | char *buf = NULL; | |
1774 | struct stat st; | |
1775 | ||
1776 | if (fstat(fd, &st) < 0) { | |
1777 | result = -errno; | |
1778 | error_setg_errno(errp, -result, "Could not stat file"); | |
1779 | return result; | |
1780 | } | |
1781 | ||
1782 | current_length = st.st_size; | |
1783 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { | |
1784 | error_setg(errp, "Cannot use preallocation for shrinking files"); | |
1785 | return -ENOTSUP; | |
1786 | } | |
9f63b07e HR |
1787 | |
1788 | switch (prealloc) { | |
1789 | #ifdef CONFIG_POSIX_FALLOCATE | |
1790 | case PREALLOC_MODE_FALLOC: | |
1791 | /* | |
1792 | * Truncating before posix_fallocate() makes it about twice slower on | |
1793 | * file systems that do not support fallocate(), trying to check if a | |
1794 | * block is allocated before allocating it, so don't do that here. | |
1795 | */ | |
89b259ee KW |
1796 | if (offset != current_length) { |
1797 | result = -posix_fallocate(fd, current_length, offset - current_length); | |
1798 | if (result != 0) { | |
1799 | /* posix_fallocate() doesn't set errno. */ | |
1800 | error_setg_errno(errp, -result, | |
1801 | "Could not preallocate new data"); | |
1802 | } | |
1803 | } else { | |
1804 | result = 0; | |
9f63b07e | 1805 | } |
d0bc9e5d | 1806 | goto out; |
9f63b07e HR |
1807 | #endif |
1808 | case PREALLOC_MODE_FULL: | |
1809 | { | |
d0bc9e5d | 1810 | int64_t num = 0, left = offset - current_length; |
82b45e0a | 1811 | off_t seek_result; |
9f63b07e HR |
1812 | |
1813 | /* | |
1814 | * Knowing the final size from the beginning could allow the file | |
1815 | * system driver to do less allocations and possibly avoid | |
1816 | * fragmentation of the file. | |
1817 | */ | |
1818 | if (ftruncate(fd, offset) != 0) { | |
1819 | result = -errno; | |
1820 | error_setg_errno(errp, -result, "Could not resize file"); | |
d0bc9e5d | 1821 | goto out; |
9f63b07e HR |
1822 | } |
1823 | ||
1824 | buf = g_malloc0(65536); | |
1825 | ||
82b45e0a HR |
1826 | seek_result = lseek(fd, current_length, SEEK_SET); |
1827 | if (seek_result < 0) { | |
d0bc9e5d HR |
1828 | result = -errno; |
1829 | error_setg_errno(errp, -result, | |
1830 | "Failed to seek to the old end of file"); | |
1831 | goto out; | |
1832 | } | |
1833 | ||
9f63b07e HR |
1834 | while (left > 0) { |
1835 | num = MIN(left, 65536); | |
1836 | result = write(fd, buf, num); | |
1837 | if (result < 0) { | |
1838 | result = -errno; | |
1839 | error_setg_errno(errp, -result, | |
d0bc9e5d HR |
1840 | "Could not write zeros for preallocation"); |
1841 | goto out; | |
9f63b07e HR |
1842 | } |
1843 | left -= result; | |
1844 | } | |
1845 | if (result >= 0) { | |
1846 | result = fsync(fd); | |
1847 | if (result < 0) { | |
1848 | result = -errno; | |
1849 | error_setg_errno(errp, -result, | |
d0bc9e5d HR |
1850 | "Could not flush file to disk"); |
1851 | goto out; | |
9f63b07e HR |
1852 | } |
1853 | } | |
d0bc9e5d | 1854 | goto out; |
9f63b07e HR |
1855 | } |
1856 | case PREALLOC_MODE_OFF: | |
1857 | if (ftruncate(fd, offset) != 0) { | |
1858 | result = -errno; | |
1859 | error_setg_errno(errp, -result, "Could not resize file"); | |
1860 | } | |
1861 | return result; | |
1862 | default: | |
1863 | result = -ENOTSUP; | |
1864 | error_setg(errp, "Unsupported preallocation mode: %s", | |
977c736f | 1865 | PreallocMode_str(prealloc)); |
9f63b07e HR |
1866 | return result; |
1867 | } | |
d0bc9e5d HR |
1868 | |
1869 | out: | |
1870 | if (result < 0) { | |
1871 | if (ftruncate(fd, current_length) < 0) { | |
1872 | error_report("Failed to restore old file length: %s", | |
1873 | strerror(errno)); | |
1874 | } | |
1875 | } | |
1876 | ||
1877 | g_free(buf); | |
1878 | return result; | |
9f63b07e HR |
1879 | } |
1880 | ||
8243ccb7 HR |
1881 | static int raw_truncate(BlockDriverState *bs, int64_t offset, |
1882 | PreallocMode prealloc, Error **errp) | |
83f64091 FB |
1883 | { |
1884 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 1885 | struct stat st; |
f59adb32 | 1886 | int ret; |
55b949c8 CH |
1887 | |
1888 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
1889 | ret = -errno; |
1890 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
1891 | return ret; | |
55b949c8 CH |
1892 | } |
1893 | ||
1894 | if (S_ISREG(st.st_mode)) { | |
35d72602 HR |
1895 | return raw_regular_truncate(s->fd, offset, prealloc, errp); |
1896 | } | |
1897 | ||
1898 | if (prealloc != PREALLOC_MODE_OFF) { | |
1899 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 1900 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
1901 | return -ENOTSUP; |
1902 | } | |
1903 | ||
1904 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
f59adb32 HR |
1905 | if (offset > raw_getlength(bs)) { |
1906 | error_setg(errp, "Cannot grow device files"); | |
1907 | return -EINVAL; | |
1908 | } | |
55b949c8 | 1909 | } else { |
f59adb32 | 1910 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
1911 | return -ENOTSUP; |
1912 | } | |
1913 | ||
83f64091 FB |
1914 | return 0; |
1915 | } | |
1916 | ||
128ab2ff BS |
1917 | #ifdef __OpenBSD__ |
1918 | static int64_t raw_getlength(BlockDriverState *bs) | |
1919 | { | |
1920 | BDRVRawState *s = bs->opaque; | |
1921 | int fd = s->fd; | |
1922 | struct stat st; | |
1923 | ||
1924 | if (fstat(fd, &st)) | |
aa729704 | 1925 | return -errno; |
128ab2ff BS |
1926 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
1927 | struct disklabel dl; | |
1928 | ||
1929 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 1930 | return -errno; |
128ab2ff BS |
1931 | return (uint64_t)dl.d_secsize * |
1932 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
1933 | } else | |
1934 | return st.st_size; | |
1935 | } | |
d1f6fd8d CE |
1936 | #elif defined(__NetBSD__) |
1937 | static int64_t raw_getlength(BlockDriverState *bs) | |
1938 | { | |
1939 | BDRVRawState *s = bs->opaque; | |
1940 | int fd = s->fd; | |
1941 | struct stat st; | |
1942 | ||
1943 | if (fstat(fd, &st)) | |
aa729704 | 1944 | return -errno; |
d1f6fd8d CE |
1945 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
1946 | struct dkwedge_info dkw; | |
1947 | ||
1948 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
1949 | return dkw.dkw_size * 512; | |
1950 | } else { | |
1951 | struct disklabel dl; | |
1952 | ||
1953 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 1954 | return -errno; |
d1f6fd8d CE |
1955 | return (uint64_t)dl.d_secsize * |
1956 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
1957 | } | |
1958 | } else | |
1959 | return st.st_size; | |
1960 | } | |
50779cc2 CH |
1961 | #elif defined(__sun__) |
1962 | static int64_t raw_getlength(BlockDriverState *bs) | |
1963 | { | |
1964 | BDRVRawState *s = bs->opaque; | |
1965 | struct dk_minfo minfo; | |
1966 | int ret; | |
aa729704 | 1967 | int64_t size; |
50779cc2 CH |
1968 | |
1969 | ret = fd_open(bs); | |
1970 | if (ret < 0) { | |
1971 | return ret; | |
1972 | } | |
1973 | ||
1974 | /* | |
1975 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
1976 | */ | |
1977 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
1978 | if (ret != -1) { | |
1979 | return minfo.dki_lbsize * minfo.dki_capacity; | |
1980 | } | |
1981 | ||
1982 | /* | |
1983 | * There are reports that lseek on some devices fails, but | |
1984 | * irc discussion said that contingency on contingency was overkill. | |
1985 | */ | |
aa729704 MA |
1986 | size = lseek(s->fd, 0, SEEK_END); |
1987 | if (size < 0) { | |
1988 | return -errno; | |
1989 | } | |
1990 | return size; | |
50779cc2 CH |
1991 | } |
1992 | #elif defined(CONFIG_BSD) | |
1993 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
1994 | { |
1995 | BDRVRawState *s = bs->opaque; | |
1996 | int fd = s->fd; | |
1997 | int64_t size; | |
83f64091 | 1998 | struct stat sb; |
a167ba50 | 1999 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2000 | int reopened = 0; |
83f64091 | 2001 | #endif |
19cb3738 FB |
2002 | int ret; |
2003 | ||
2004 | ret = fd_open(bs); | |
2005 | if (ret < 0) | |
2006 | return ret; | |
83f64091 | 2007 | |
a167ba50 | 2008 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2009 | again: |
2010 | #endif | |
83f64091 FB |
2011 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
2012 | #ifdef DIOCGMEDIASIZE | |
2013 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) | |
c5e97233 BS |
2014 | #elif defined(DIOCGPART) |
2015 | { | |
2016 | struct partinfo pi; | |
2017 | if (ioctl(fd, DIOCGPART, &pi) == 0) | |
2018 | size = pi.media_size; | |
2019 | else | |
2020 | size = 0; | |
2021 | } | |
2022 | if (size == 0) | |
83f64091 | 2023 | #endif |
83affaa6 | 2024 | #if defined(__APPLE__) && defined(__MACH__) |
728dacbd JA |
2025 | { |
2026 | uint64_t sectors = 0; | |
2027 | uint32_t sector_size = 0; | |
2028 | ||
2029 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2030 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2031 | size = sectors * sector_size; | |
2032 | } else { | |
2033 | size = lseek(fd, 0LL, SEEK_END); | |
2034 | if (size < 0) { | |
2035 | return -errno; | |
2036 | } | |
2037 | } | |
2038 | } | |
83f64091 FB |
2039 | #else |
2040 | size = lseek(fd, 0LL, SEEK_END); | |
aa729704 MA |
2041 | if (size < 0) { |
2042 | return -errno; | |
2043 | } | |
9f23011a | 2044 | #endif |
a167ba50 | 2045 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2046 | switch(s->type) { |
2047 | case FTYPE_CD: | |
2048 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2049 | if (size == 2048LL * (unsigned)-1) | |
2050 | size = 0; | |
2051 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2052 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2053 | reopened = 1; |
2054 | goto again; | |
2055 | } | |
2056 | } | |
83f64091 | 2057 | #endif |
50779cc2 | 2058 | } else { |
83f64091 | 2059 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2060 | if (size < 0) { |
2061 | return -errno; | |
2062 | } | |
83f64091 | 2063 | } |
83f64091 FB |
2064 | return size; |
2065 | } | |
50779cc2 CH |
2066 | #else |
2067 | static int64_t raw_getlength(BlockDriverState *bs) | |
2068 | { | |
2069 | BDRVRawState *s = bs->opaque; | |
2070 | int ret; | |
aa729704 | 2071 | int64_t size; |
50779cc2 CH |
2072 | |
2073 | ret = fd_open(bs); | |
2074 | if (ret < 0) { | |
2075 | return ret; | |
2076 | } | |
2077 | ||
aa729704 MA |
2078 | size = lseek(s->fd, 0, SEEK_END); |
2079 | if (size < 0) { | |
2080 | return -errno; | |
2081 | } | |
2082 | return size; | |
50779cc2 | 2083 | } |
128ab2ff | 2084 | #endif |
83f64091 | 2085 | |
4a1d5e1f FZ |
2086 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2087 | { | |
2088 | struct stat st; | |
2089 | BDRVRawState *s = bs->opaque; | |
2090 | ||
2091 | if (fstat(s->fd, &st) < 0) { | |
2092 | return -errno; | |
2093 | } | |
2094 | return (int64_t)st.st_blocks * 512; | |
2095 | } | |
2096 | ||
927f11e1 | 2097 | static int raw_co_create(BlockdevCreateOptions *options, Error **errp) |
83f64091 | 2098 | { |
927f11e1 | 2099 | BlockdevCreateOptionsFile *file_opts; |
83f64091 | 2100 | int fd; |
b8cf1913 | 2101 | int perm, shared; |
1e37d059 | 2102 | int result = 0; |
83f64091 | 2103 | |
927f11e1 KW |
2104 | /* Validate options and set default values */ |
2105 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2106 | file_opts = &options->u.file; | |
464d9f64 | 2107 | |
927f11e1 KW |
2108 | if (!file_opts->has_nocow) { |
2109 | file_opts->nocow = false; | |
2110 | } | |
2111 | if (!file_opts->has_preallocation) { | |
2112 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2113 | } |
83f64091 | 2114 | |
927f11e1 | 2115 | /* Create file */ |
b8cf1913 | 2116 | fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); |
1e37d059 SW |
2117 | if (fd < 0) { |
2118 | result = -errno; | |
e428e439 | 2119 | error_setg_errno(errp, -result, "Could not create file"); |
06247428 HT |
2120 | goto out; |
2121 | } | |
2122 | ||
b8cf1913 HR |
2123 | /* Take permissions: We want to discard everything, so we need |
2124 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2125 | * BLK_PERM_RESIZE. | |
2126 | * On the other hand, we cannot share the RESIZE permission | |
2127 | * because we promise that after this function, the file has the | |
2128 | * size given in the options. If someone else were to resize it | |
2129 | * concurrently, we could not guarantee that. | |
2130 | * Note that after this function, we can no longer guarantee that | |
2131 | * the file is not touched by a third party, so it may be resized | |
2132 | * then. */ | |
2133 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2134 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2135 | ||
2136 | /* Step one: Take locks */ | |
2137 | result = raw_apply_lock_bytes(fd, perm, shared, false, errp); | |
2138 | if (result < 0) { | |
2139 | goto out_close; | |
2140 | } | |
2141 | ||
2142 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2143 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2144 | if (result < 0) { | |
2145 | goto out_close; | |
2146 | } | |
2147 | ||
2148 | /* Clear the file by truncating it to 0 */ | |
2149 | result = raw_regular_truncate(fd, 0, PREALLOC_MODE_OFF, errp); | |
2150 | if (result < 0) { | |
2151 | goto out_close; | |
2152 | } | |
2153 | ||
927f11e1 | 2154 | if (file_opts->nocow) { |
4ab15590 | 2155 | #ifdef __linux__ |
06247428 HT |
2156 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2157 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2158 | * will be ignored since any failure of this operation should not | |
2159 | * block the left work. | |
2160 | */ | |
2161 | int attr; | |
2162 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2163 | attr |= FS_NOCOW_FL; | |
2164 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2165 | } |
06247428 HT |
2166 | #endif |
2167 | } | |
2168 | ||
b8cf1913 HR |
2169 | /* Resize and potentially preallocate the file to the desired |
2170 | * final size */ | |
927f11e1 KW |
2171 | result = raw_regular_truncate(fd, file_opts->size, file_opts->preallocation, |
2172 | errp); | |
9f63b07e HR |
2173 | if (result < 0) { |
2174 | goto out_close; | |
1e37d059 | 2175 | } |
06247428 | 2176 | |
5a1dad9d | 2177 | out_close: |
06247428 HT |
2178 | if (qemu_close(fd) != 0 && result == 0) { |
2179 | result = -errno; | |
2180 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2181 | } | |
2182 | out: | |
1e37d059 | 2183 | return result; |
83f64091 FB |
2184 | } |
2185 | ||
927f11e1 KW |
2186 | static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, |
2187 | Error **errp) | |
2188 | { | |
2189 | BlockdevCreateOptions options; | |
2190 | int64_t total_size = 0; | |
2191 | bool nocow = false; | |
2192 | PreallocMode prealloc; | |
2193 | char *buf = NULL; | |
2194 | Error *local_err = NULL; | |
2195 | ||
2196 | /* Skip file: protocol prefix */ | |
2197 | strstart(filename, "file:", &filename); | |
2198 | ||
2199 | /* Read out options */ | |
2200 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2201 | BDRV_SECTOR_SIZE); | |
2202 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); | |
2203 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2204 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2205 | PREALLOC_MODE_OFF, &local_err); | |
2206 | g_free(buf); | |
2207 | if (local_err) { | |
2208 | error_propagate(errp, local_err); | |
2209 | return -EINVAL; | |
2210 | } | |
2211 | ||
2212 | options = (BlockdevCreateOptions) { | |
2213 | .driver = BLOCKDEV_DRIVER_FILE, | |
2214 | .u.file = { | |
2215 | .filename = (char *) filename, | |
2216 | .size = total_size, | |
2217 | .has_preallocation = true, | |
2218 | .preallocation = prealloc, | |
2219 | .has_nocow = true, | |
2220 | .nocow = nocow, | |
2221 | }, | |
2222 | }; | |
2223 | return raw_co_create(&options, errp); | |
2224 | } | |
2225 | ||
d1f06fe6 MA |
2226 | /* |
2227 | * Find allocation range in @bs around offset @start. | |
2228 | * May change underlying file descriptor's file offset. | |
2229 | * If @start is not in a hole, store @start in @data, and the | |
2230 | * beginning of the next hole in @hole, and return 0. | |
2231 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2232 | * beginning of the next non-hole in @data, and return 0. | |
2233 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2234 | * If we can't find out, return a negative errno other than -ENXIO. | |
2235 | */ | |
2236 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2237 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2238 | { |
2239 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2240 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2241 | off_t offs; |
94282e71 | 2242 | |
d1f06fe6 MA |
2243 | /* |
2244 | * SEEK_DATA cases: | |
2245 | * D1. offs == start: start is in data | |
2246 | * D2. offs > start: start is in a hole, next data at offs | |
2247 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2248 | * or start is beyond EOF | |
2249 | * If the latter happens, the file has been truncated behind | |
2250 | * our back since we opened it. All bets are off then. | |
2251 | * Treating like a trailing hole is simplest. | |
2252 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2253 | */ | |
2254 | offs = lseek(s->fd, start, SEEK_DATA); | |
2255 | if (offs < 0) { | |
2256 | return -errno; /* D3 or D4 */ | |
2257 | } | |
a03083a0 JC |
2258 | |
2259 | if (offs < start) { | |
2260 | /* This is not a valid return by lseek(). We are safe to just return | |
2261 | * -EIO in this case, and we'll treat it like D4. */ | |
2262 | return -EIO; | |
2263 | } | |
d1f06fe6 MA |
2264 | |
2265 | if (offs > start) { | |
2266 | /* D2: in hole, next data at offs */ | |
2267 | *hole = start; | |
2268 | *data = offs; | |
2269 | return 0; | |
5500316d PB |
2270 | } |
2271 | ||
d1f06fe6 MA |
2272 | /* D1: in data, end not yet known */ |
2273 | ||
2274 | /* | |
2275 | * SEEK_HOLE cases: | |
2276 | * H1. offs == start: start is in a hole | |
2277 | * If this happens here, a hole has been dug behind our back | |
2278 | * since the previous lseek(). | |
2279 | * H2. offs > start: either start is in data, next hole at offs, | |
2280 | * or start is in trailing hole, EOF at offs | |
2281 | * Linux treats trailing holes like any other hole: offs == | |
2282 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2283 | * If that happens here, a hole has been dug behind our back | |
2284 | * since the previous lseek(). | |
2285 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2286 | * If this happens, the file has been truncated behind our | |
2287 | * back since we opened it. Treat it like a trailing hole. | |
2288 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2289 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2290 | */ | |
2291 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2292 | if (offs < 0) { | |
2293 | return -errno; /* D1 and (H3 or H4) */ | |
2294 | } | |
a03083a0 JC |
2295 | |
2296 | if (offs < start) { | |
2297 | /* This is not a valid return by lseek(). We are safe to just return | |
2298 | * -EIO in this case, and we'll treat it like H4. */ | |
2299 | return -EIO; | |
2300 | } | |
d1f06fe6 MA |
2301 | |
2302 | if (offs > start) { | |
2303 | /* | |
2304 | * D1 and H2: either in data, next hole at offs, or it was in | |
2305 | * data but is now in a trailing hole. In the latter case, | |
2306 | * all bets are off. Treating it as if it there was data all | |
2307 | * the way to EOF is safe, so simply do that. | |
2308 | */ | |
4f11aa8a | 2309 | *data = start; |
d1f06fe6 MA |
2310 | *hole = offs; |
2311 | return 0; | |
5500316d | 2312 | } |
4f11aa8a | 2313 | |
d1f06fe6 MA |
2314 | /* D1 and H1 */ |
2315 | return -EBUSY; | |
5500316d | 2316 | #else |
4f11aa8a | 2317 | return -ENOTSUP; |
5500316d | 2318 | #endif |
4f11aa8a HR |
2319 | } |
2320 | ||
2321 | /* | |
a290f085 | 2322 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2323 | * |
a290f085 | 2324 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2325 | * |
a290f085 EB |
2326 | * 'pnum' is set to the number of bytes (including and immediately following |
2327 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2328 | * allocated/unallocated state. |
2329 | * | |
a290f085 | 2330 | * 'bytes' is the max value 'pnum' should be set to. |
4f11aa8a | 2331 | */ |
a290f085 EB |
2332 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2333 | bool want_zero, | |
2334 | int64_t offset, | |
2335 | int64_t bytes, int64_t *pnum, | |
2336 | int64_t *map, | |
2337 | BlockDriverState **file) | |
2338 | { | |
2339 | off_t data = 0, hole = 0; | |
d7f62751 | 2340 | int ret; |
4f11aa8a HR |
2341 | |
2342 | ret = fd_open(bs); | |
2343 | if (ret < 0) { | |
2344 | return ret; | |
2345 | } | |
2346 | ||
a290f085 EB |
2347 | if (!want_zero) { |
2348 | *pnum = bytes; | |
2349 | *map = offset; | |
2350 | *file = bs; | |
2351 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2352 | } |
4f11aa8a | 2353 | |
a290f085 | 2354 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2355 | if (ret == -ENXIO) { |
2356 | /* Trailing hole */ | |
a290f085 | 2357 | *pnum = bytes; |
d1f06fe6 MA |
2358 | ret = BDRV_BLOCK_ZERO; |
2359 | } else if (ret < 0) { | |
2360 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2361 | *pnum = bytes; |
d1f06fe6 | 2362 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2363 | } else if (data == offset) { |
2364 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2365 | * possibly including a partial sector at EOF. */ |
a290f085 | 2366 | *pnum = MIN(bytes, hole - offset); |
d1f06fe6 | 2367 | ret = BDRV_BLOCK_DATA; |
5500316d | 2368 | } else { |
a290f085 EB |
2369 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2370 | assert(hole == offset); | |
2371 | *pnum = MIN(bytes, data - offset); | |
d1f06fe6 | 2372 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2373 | } |
a290f085 | 2374 | *map = offset; |
02650acb | 2375 | *file = bs; |
a290f085 | 2376 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2377 | } |
2378 | ||
31be8a2a SH |
2379 | #if defined(__linux__) |
2380 | /* Verify that the file is not in the page cache */ | |
2381 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2382 | { | |
2383 | const size_t window_size = 128 * 1024 * 1024; | |
2384 | BDRVRawState *s = bs->opaque; | |
2385 | void *window = NULL; | |
2386 | size_t length = 0; | |
2387 | unsigned char *vec; | |
2388 | size_t page_size; | |
2389 | off_t offset; | |
2390 | off_t end; | |
2391 | ||
2392 | /* mincore(2) page status information requires 1 byte per page */ | |
2393 | page_size = sysconf(_SC_PAGESIZE); | |
2394 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2395 | ||
2396 | end = raw_getlength(bs); | |
2397 | ||
2398 | for (offset = 0; offset < end; offset += window_size) { | |
2399 | void *new_window; | |
2400 | size_t new_length; | |
2401 | size_t vec_end; | |
2402 | size_t i; | |
2403 | int ret; | |
2404 | ||
2405 | /* Unmap previous window if size has changed */ | |
2406 | new_length = MIN(end - offset, window_size); | |
2407 | if (new_length != length) { | |
2408 | munmap(window, length); | |
2409 | window = NULL; | |
2410 | length = 0; | |
2411 | } | |
2412 | ||
2413 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2414 | s->fd, offset); | |
2415 | if (new_window == MAP_FAILED) { | |
2416 | error_setg_errno(errp, errno, "mmap failed"); | |
2417 | break; | |
2418 | } | |
2419 | ||
2420 | window = new_window; | |
2421 | length = new_length; | |
2422 | ||
2423 | ret = mincore(window, length, vec); | |
2424 | if (ret < 0) { | |
2425 | error_setg_errno(errp, errno, "mincore failed"); | |
2426 | break; | |
2427 | } | |
2428 | ||
2429 | vec_end = DIV_ROUND_UP(length, page_size); | |
2430 | for (i = 0; i < vec_end; i++) { | |
2431 | if (vec[i] & 0x1) { | |
2432 | error_setg(errp, "page cache still in use!"); | |
2433 | break; | |
2434 | } | |
2435 | } | |
2436 | } | |
2437 | ||
2438 | if (window) { | |
2439 | munmap(window, length); | |
2440 | } | |
2441 | ||
2442 | g_free(vec); | |
2443 | } | |
2444 | #endif /* __linux__ */ | |
2445 | ||
dd577a26 SH |
2446 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2447 | Error **errp) | |
2448 | { | |
2449 | BDRVRawState *s = bs->opaque; | |
2450 | int ret; | |
2451 | ||
2452 | ret = fd_open(bs); | |
2453 | if (ret < 0) { | |
2454 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2455 | return; | |
2456 | } | |
2457 | ||
2458 | if (s->open_flags & O_DIRECT) { | |
2459 | return; /* No host kernel page cache */ | |
2460 | } | |
2461 | ||
2462 | #if defined(__linux__) | |
2463 | /* This sets the scene for the next syscall... */ | |
2464 | ret = bdrv_co_flush(bs); | |
2465 | if (ret < 0) { | |
2466 | error_setg_errno(errp, -ret, "flush failed"); | |
2467 | return; | |
2468 | } | |
2469 | ||
2470 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2471 | * process. These limitations are okay because we just fsynced the file, | |
2472 | * we don't use mmap, and the file should not be in use by other processes. | |
2473 | */ | |
2474 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2475 | if (ret != 0) { /* the return value is a positive errno */ | |
2476 | error_setg_errno(errp, ret, "fadvise failed"); | |
2477 | return; | |
2478 | } | |
31be8a2a SH |
2479 | |
2480 | if (s->check_cache_dropped) { | |
2481 | check_cache_dropped(bs, errp); | |
2482 | } | |
dd577a26 SH |
2483 | #else /* __linux__ */ |
2484 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2485 | * unsupported on other host operating systems. Cache consistency issues | |
2486 | * may occur but no error is reported here, partly because that's the | |
2487 | * historical behavior and partly because it's hard to differentiate valid | |
2488 | * configurations that should not cause errors. | |
2489 | */ | |
2490 | #endif /* !__linux__ */ | |
2491 | } | |
2492 | ||
4da444a0 | 2493 | static coroutine_fn BlockAIOCB *raw_aio_pdiscard(BlockDriverState *bs, |
f5a5ca79 | 2494 | int64_t offset, int bytes, |
097310b5 | 2495 | BlockCompletionFunc *cb, void *opaque) |
dce512de | 2496 | { |
dce512de CH |
2497 | BDRVRawState *s = bs->opaque; |
2498 | ||
f5a5ca79 | 2499 | return paio_submit(bs, s->fd, offset, NULL, bytes, |
8238010b | 2500 | cb, opaque, QEMU_AIO_DISCARD); |
dce512de | 2501 | } |
0e7e1989 | 2502 | |
2ffa76c2 EB |
2503 | static int coroutine_fn raw_co_pwrite_zeroes( |
2504 | BlockDriverState *bs, int64_t offset, | |
f5a5ca79 | 2505 | int bytes, BdrvRequestFlags flags) |
260a82e5 PB |
2506 | { |
2507 | BDRVRawState *s = bs->opaque; | |
2508 | ||
2509 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
f5a5ca79 | 2510 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 PB |
2511 | QEMU_AIO_WRITE_ZEROES); |
2512 | } else if (s->discard_zeroes) { | |
f5a5ca79 | 2513 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 | 2514 | QEMU_AIO_DISCARD); |
260a82e5 | 2515 | } |
97a2ae34 | 2516 | return -ENOTSUP; |
260a82e5 PB |
2517 | } |
2518 | ||
2519 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
2520 | { | |
2521 | BDRVRawState *s = bs->opaque; | |
2522 | ||
2523 | bdi->unallocated_blocks_are_zero = s->discard_zeroes; | |
260a82e5 PB |
2524 | return 0; |
2525 | } | |
2526 | ||
6f482f74 CL |
2527 | static QemuOptsList raw_create_opts = { |
2528 | .name = "raw-create-opts", | |
2529 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
2530 | .desc = { | |
2531 | { | |
2532 | .name = BLOCK_OPT_SIZE, | |
2533 | .type = QEMU_OPT_SIZE, | |
2534 | .help = "Virtual disk size" | |
2535 | }, | |
4ab15590 CL |
2536 | { |
2537 | .name = BLOCK_OPT_NOCOW, | |
2538 | .type = QEMU_OPT_BOOL, | |
2539 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
2540 | }, | |
06247428 HT |
2541 | { |
2542 | .name = BLOCK_OPT_PREALLOC, | |
2543 | .type = QEMU_OPT_STRING, | |
2544 | .help = "Preallocation mode (allowed values: off, falloc, full)" | |
2545 | }, | |
6f482f74 CL |
2546 | { /* end of list */ } |
2547 | } | |
0e7e1989 KW |
2548 | }; |
2549 | ||
244a5668 FZ |
2550 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
2551 | Error **errp) | |
2552 | { | |
2553 | return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
2554 | } | |
2555 | ||
2556 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
2557 | { | |
2558 | BDRVRawState *s = bs->opaque; | |
2559 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); | |
2560 | s->perm = perm; | |
2561 | s->shared_perm = shared; | |
2562 | } | |
2563 | ||
2564 | static void raw_abort_perm_update(BlockDriverState *bs) | |
2565 | { | |
2566 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
2567 | } | |
2568 | ||
1efad060 FZ |
2569 | static int coroutine_fn raw_co_copy_range_from(BlockDriverState *bs, |
2570 | BdrvChild *src, uint64_t src_offset, | |
2571 | BdrvChild *dst, uint64_t dst_offset, | |
2572 | uint64_t bytes, BdrvRequestFlags flags) | |
2573 | { | |
2574 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, flags); | |
2575 | } | |
2576 | ||
2577 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
2578 | BdrvChild *src, uint64_t src_offset, | |
2579 | BdrvChild *dst, uint64_t dst_offset, | |
2580 | uint64_t bytes, BdrvRequestFlags flags) | |
2581 | { | |
2582 | BDRVRawState *s = bs->opaque; | |
2583 | BDRVRawState *src_s; | |
2584 | ||
2585 | assert(dst->bs == bs); | |
2586 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
2587 | return -ENOTSUP; | |
2588 | } | |
2589 | ||
2590 | src_s = src->bs->opaque; | |
2591 | if (fd_open(bs) < 0 || fd_open(bs) < 0) { | |
2592 | return -EIO; | |
2593 | } | |
2594 | return paio_submit_co_full(bs, src_s->fd, src_offset, s->fd, dst_offset, | |
2595 | NULL, bytes, QEMU_AIO_COPY_RANGE); | |
2596 | } | |
2597 | ||
5f535a94 | 2598 | BlockDriver bdrv_file = { |
84a12e66 CH |
2599 | .format_name = "file", |
2600 | .protocol_name = "file", | |
856ae5c3 | 2601 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2602 | .bdrv_needs_filename = true, |
856ae5c3 | 2603 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 2604 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 2605 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
2606 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2607 | .bdrv_reopen_commit = raw_reopen_commit, | |
2608 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 2609 | .bdrv_close = raw_close, |
927f11e1 | 2610 | .bdrv_co_create = raw_co_create, |
efc75e2a | 2611 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 2612 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 2613 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 2614 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 2615 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
3b46e624 | 2616 | |
9d52aa3c KW |
2617 | .bdrv_co_preadv = raw_co_preadv, |
2618 | .bdrv_co_pwritev = raw_co_pwritev, | |
b2e12bc6 | 2619 | .bdrv_aio_flush = raw_aio_flush, |
4da444a0 | 2620 | .bdrv_aio_pdiscard = raw_aio_pdiscard, |
1efad060 FZ |
2621 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
2622 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 2623 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2624 | .bdrv_io_plug = raw_aio_plug, |
2625 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 2626 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 2627 | |
83f64091 FB |
2628 | .bdrv_truncate = raw_truncate, |
2629 | .bdrv_getlength = raw_getlength, | |
260a82e5 | 2630 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
2631 | .bdrv_get_allocated_file_size |
2632 | = raw_get_allocated_file_size, | |
244a5668 FZ |
2633 | .bdrv_check_perm = raw_check_perm, |
2634 | .bdrv_set_perm = raw_set_perm, | |
2635 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 2636 | .create_opts = &raw_create_opts, |
83f64091 FB |
2637 | }; |
2638 | ||
19cb3738 FB |
2639 | /***********************************************/ |
2640 | /* host device */ | |
2641 | ||
83affaa6 | 2642 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
2643 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2644 | CFIndex maxPathSize, int flags); | |
d0855f12 | 2645 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 2646 | { |
d0855f12 | 2647 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
2648 | mach_port_t masterPort; |
2649 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
2650 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
2651 | char *mediaType = NULL; | |
19cb3738 FB |
2652 | |
2653 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
2654 | if ( KERN_SUCCESS != kernResult ) { | |
2655 | printf( "IOMasterPort returned %d\n", kernResult ); | |
2656 | } | |
3b46e624 | 2657 | |
d0855f12 JA |
2658 | int index; |
2659 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
2660 | classesToMatch = IOServiceMatching(matching_array[index]); | |
2661 | if (classesToMatch == NULL) { | |
2662 | error_report("IOServiceMatching returned NULL for %s", | |
2663 | matching_array[index]); | |
2664 | continue; | |
2665 | } | |
2666 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
2667 | kCFBooleanTrue); | |
2668 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
2669 | mediaIterator); | |
2670 | if (kernResult != KERN_SUCCESS) { | |
2671 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
2672 | kernResult); | |
2673 | continue; | |
2674 | } | |
3b46e624 | 2675 | |
d0855f12 JA |
2676 | /* If a match was found, leave the loop */ |
2677 | if (*mediaIterator != 0) { | |
2678 | DPRINTF("Matching using %s\n", matching_array[index]); | |
2679 | mediaType = g_strdup(matching_array[index]); | |
2680 | break; | |
2681 | } | |
2682 | } | |
2683 | return mediaType; | |
19cb3738 FB |
2684 | } |
2685 | ||
98caa5bc JA |
2686 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2687 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
2688 | { |
2689 | io_object_t nextMedia; | |
2690 | kern_return_t kernResult = KERN_FAILURE; | |
2691 | *bsdPath = '\0'; | |
2692 | nextMedia = IOIteratorNext( mediaIterator ); | |
2693 | if ( nextMedia ) | |
2694 | { | |
2695 | CFTypeRef bsdPathAsCFString; | |
2696 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
2697 | if ( bsdPathAsCFString ) { | |
2698 | size_t devPathLength; | |
2699 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
2700 | if (flags & BDRV_O_NOCACHE) { |
2701 | strcat(bsdPath, "r"); | |
2702 | } | |
19cb3738 FB |
2703 | devPathLength = strlen( bsdPath ); |
2704 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
2705 | kernResult = KERN_SUCCESS; | |
2706 | } | |
2707 | CFRelease( bsdPathAsCFString ); | |
2708 | } | |
2709 | IOObjectRelease( nextMedia ); | |
2710 | } | |
3b46e624 | 2711 | |
19cb3738 FB |
2712 | return kernResult; |
2713 | } | |
2714 | ||
d0855f12 JA |
2715 | /* Sets up a real cdrom for use in QEMU */ |
2716 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
2717 | { | |
2718 | int index, num_of_test_partitions = 2, fd; | |
2719 | char test_partition[MAXPATHLEN]; | |
2720 | bool partition_found = false; | |
2721 | ||
2722 | /* look for a working partition */ | |
2723 | for (index = 0; index < num_of_test_partitions; index++) { | |
2724 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
2725 | index); | |
2726 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); | |
2727 | if (fd >= 0) { | |
2728 | partition_found = true; | |
2729 | qemu_close(fd); | |
2730 | break; | |
2731 | } | |
2732 | } | |
2733 | ||
2734 | /* if a working partition on the device was not found */ | |
2735 | if (partition_found == false) { | |
2736 | error_setg(errp, "Failed to find a working partition on disc"); | |
2737 | } else { | |
2738 | DPRINTF("Using %s as optical disc\n", test_partition); | |
2739 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); | |
2740 | } | |
2741 | return partition_found; | |
2742 | } | |
2743 | ||
2744 | /* Prints directions on mounting and unmounting a device */ | |
2745 | static void print_unmounting_directions(const char *file_name) | |
2746 | { | |
2747 | error_report("If device %s is mounted on the desktop, unmount" | |
2748 | " it first before using it in QEMU", file_name); | |
2749 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
2750 | file_name); | |
2751 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
2752 | } | |
2753 | ||
2754 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 2755 | |
508c7cb3 CH |
2756 | static int hdev_probe_device(const char *filename) |
2757 | { | |
2758 | struct stat st; | |
2759 | ||
2760 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
2761 | if (strstart(filename, "/dev/cdrom", NULL)) | |
2762 | return 50; | |
2763 | ||
2764 | if (stat(filename, &st) >= 0 && | |
2765 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
2766 | return 100; | |
2767 | } | |
2768 | ||
2769 | return 0; | |
2770 | } | |
2771 | ||
da888d37 SH |
2772 | static int check_hdev_writable(BDRVRawState *s) |
2773 | { | |
2774 | #if defined(BLKROGET) | |
2775 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
2776 | * This is independent of device node permissions and therefore open(2) | |
2777 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
2778 | * | |
2779 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
2780 | * check for read-only block devices so that Linux block devices behave | |
2781 | * properly. | |
2782 | */ | |
2783 | struct stat st; | |
2784 | int readonly = 0; | |
2785 | ||
2786 | if (fstat(s->fd, &st)) { | |
2787 | return -errno; | |
2788 | } | |
2789 | ||
2790 | if (!S_ISBLK(st.st_mode)) { | |
2791 | return 0; | |
2792 | } | |
2793 | ||
2794 | if (ioctl(s->fd, BLKROGET, &readonly) < 0) { | |
2795 | return -errno; | |
2796 | } | |
2797 | ||
2798 | if (readonly) { | |
2799 | return -EACCES; | |
2800 | } | |
2801 | #endif /* defined(BLKROGET) */ | |
2802 | return 0; | |
2803 | } | |
2804 | ||
7af803d4 HR |
2805 | static void hdev_parse_filename(const char *filename, QDict *options, |
2806 | Error **errp) | |
2807 | { | |
03c320d8 | 2808 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
2809 | } |
2810 | ||
3307ed7b DA |
2811 | static bool hdev_is_sg(BlockDriverState *bs) |
2812 | { | |
2813 | ||
2814 | #if defined(__linux__) | |
2815 | ||
0d4377b3 | 2816 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
2817 | struct stat st; |
2818 | struct sg_scsi_id scsiid; | |
2819 | int sg_version; | |
0d4377b3 KW |
2820 | int ret; |
2821 | ||
2822 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
2823 | return false; | |
2824 | } | |
3307ed7b | 2825 | |
0d4377b3 KW |
2826 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
2827 | if (ret < 0) { | |
2828 | return false; | |
2829 | } | |
2830 | ||
2831 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
2832 | if (ret >= 0) { | |
3307ed7b DA |
2833 | DPRINTF("SG device found: type=%d, version=%d\n", |
2834 | scsiid.scsi_type, sg_version); | |
2835 | return true; | |
2836 | } | |
2837 | ||
2838 | #endif | |
2839 | ||
2840 | return false; | |
2841 | } | |
2842 | ||
015a1036 HR |
2843 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
2844 | Error **errp) | |
19cb3738 FB |
2845 | { |
2846 | BDRVRawState *s = bs->opaque; | |
e428e439 | 2847 | Error *local_err = NULL; |
da888d37 | 2848 | int ret; |
a76bab49 | 2849 | |
83affaa6 | 2850 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
2851 | /* |
2852 | * Caution: while qdict_get_str() is fine, getting non-string types | |
2853 | * would require more care. When @options come from -blockdev or | |
2854 | * blockdev_add, its members are typed according to the QAPI | |
2855 | * schema, but when they come from -drive, they're all QString. | |
2856 | */ | |
3307ed7b | 2857 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
2858 | char bsd_path[MAXPATHLEN] = ""; |
2859 | bool error_occurred = false; | |
2860 | ||
2861 | /* If using a real cdrom */ | |
2862 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
2863 | char *mediaType = NULL; | |
2864 | kern_return_t ret_val; | |
2865 | io_iterator_t mediaIterator = 0; | |
2866 | ||
2867 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
2868 | if (mediaType == NULL) { | |
2869 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
2870 | " drive"); | |
2871 | error_occurred = true; | |
2872 | goto hdev_open_Mac_error; | |
2873 | } | |
3307ed7b | 2874 | |
d0855f12 JA |
2875 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
2876 | if (ret_val != KERN_SUCCESS) { | |
2877 | error_setg(errp, "Could not get BSD path for optical drive"); | |
2878 | error_occurred = true; | |
2879 | goto hdev_open_Mac_error; | |
2880 | } | |
2881 | ||
2882 | /* If a real optical drive was not found */ | |
2883 | if (bsd_path[0] == '\0') { | |
2884 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
2885 | error_occurred = true; | |
2886 | goto hdev_open_Mac_error; | |
2887 | } | |
2888 | ||
2889 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
2890 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
2891 | setup_cdrom(bsd_path, errp) == false) { | |
2892 | print_unmounting_directions(bsd_path); | |
2893 | error_occurred = true; | |
2894 | goto hdev_open_Mac_error; | |
19cb3738 | 2895 | } |
3b46e624 | 2896 | |
46f5ac20 | 2897 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
2898 | |
2899 | hdev_open_Mac_error: | |
2900 | g_free(mediaType); | |
2901 | if (mediaIterator) { | |
2902 | IOObjectRelease(mediaIterator); | |
2903 | } | |
2904 | if (error_occurred) { | |
2905 | return -ENOENT; | |
2906 | } | |
19cb3738 | 2907 | } |
d0855f12 | 2908 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
2909 | |
2910 | s->type = FTYPE_FILE; | |
90babde0 | 2911 | |
e428e439 | 2912 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
da888d37 | 2913 | if (ret < 0) { |
621ff94d | 2914 | error_propagate(errp, local_err); |
d0855f12 JA |
2915 | #if defined(__APPLE__) && defined(__MACH__) |
2916 | if (*bsd_path) { | |
2917 | filename = bsd_path; | |
2918 | } | |
2919 | /* if a physical device experienced an error while being opened */ | |
2920 | if (strncmp(filename, "/dev/", 5) == 0) { | |
2921 | print_unmounting_directions(filename); | |
2922 | } | |
2923 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
2924 | return ret; |
2925 | } | |
2926 | ||
3307ed7b DA |
2927 | /* Since this does ioctl the device must be already opened */ |
2928 | bs->sg = hdev_is_sg(bs); | |
2929 | ||
da888d37 SH |
2930 | if (flags & BDRV_O_RDWR) { |
2931 | ret = check_hdev_writable(s); | |
2932 | if (ret < 0) { | |
2933 | raw_close(bs); | |
e428e439 | 2934 | error_setg_errno(errp, -ret, "The device is not writable"); |
da888d37 SH |
2935 | return ret; |
2936 | } | |
2937 | } | |
2938 | ||
2939 | return ret; | |
19cb3738 FB |
2940 | } |
2941 | ||
03ff3ca3 | 2942 | #if defined(__linux__) |
221f715d | 2943 | |
7c84b1b8 | 2944 | static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs, |
221f715d | 2945 | unsigned long int req, void *buf, |
097310b5 | 2946 | BlockCompletionFunc *cb, void *opaque) |
221f715d | 2947 | { |
f141eafe | 2948 | BDRVRawState *s = bs->opaque; |
c208e8c2 | 2949 | RawPosixAIOData *acb; |
c4d9d196 | 2950 | ThreadPool *pool; |
221f715d | 2951 | |
f141eafe AL |
2952 | if (fd_open(bs) < 0) |
2953 | return NULL; | |
c208e8c2 | 2954 | |
7c9e5276 PB |
2955 | if (req == SG_IO && s->pr_mgr) { |
2956 | struct sg_io_hdr *io_hdr = buf; | |
2957 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
2958 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
2959 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2960 | s->fd, io_hdr, cb, opaque); | |
2961 | } | |
2962 | } | |
2963 | ||
c84b3192 | 2964 | acb = g_new(RawPosixAIOData, 1); |
c208e8c2 PB |
2965 | acb->bs = bs; |
2966 | acb->aio_type = QEMU_AIO_IOCTL; | |
2967 | acb->aio_fildes = s->fd; | |
2968 | acb->aio_offset = 0; | |
2969 | acb->aio_ioctl_buf = buf; | |
2970 | acb->aio_ioctl_cmd = req; | |
c4d9d196 SH |
2971 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
2972 | return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); | |
221f715d | 2973 | } |
f709623b | 2974 | #endif /* linux */ |
221f715d | 2975 | |
9f23011a BS |
2976 | static int fd_open(BlockDriverState *bs) |
2977 | { | |
2978 | BDRVRawState *s = bs->opaque; | |
2979 | ||
2980 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
2981 | if (s->fd >= 0) | |
2982 | return 0; | |
2983 | return -EIO; | |
2984 | } | |
04eeb8b6 | 2985 | |
4da444a0 | 2986 | static coroutine_fn BlockAIOCB *hdev_aio_pdiscard(BlockDriverState *bs, |
f5a5ca79 | 2987 | int64_t offset, int bytes, |
097310b5 | 2988 | BlockCompletionFunc *cb, void *opaque) |
c36dd8a0 AF |
2989 | { |
2990 | BDRVRawState *s = bs->opaque; | |
2991 | ||
2992 | if (fd_open(bs) < 0) { | |
2993 | return NULL; | |
2994 | } | |
f5a5ca79 | 2995 | return paio_submit(bs, s->fd, offset, NULL, bytes, |
c36dd8a0 AF |
2996 | cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); |
2997 | } | |
2998 | ||
2ffa76c2 | 2999 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3000 | int64_t offset, int bytes, BdrvRequestFlags flags) |
d0b4503e PB |
3001 | { |
3002 | BDRVRawState *s = bs->opaque; | |
3003 | int rc; | |
3004 | ||
3005 | rc = fd_open(bs); | |
3006 | if (rc < 0) { | |
3007 | return rc; | |
3008 | } | |
3009 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
f5a5ca79 | 3010 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 PB |
3011 | QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV); |
3012 | } else if (s->discard_zeroes) { | |
f5a5ca79 | 3013 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 | 3014 | QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); |
d0b4503e | 3015 | } |
97a2ae34 | 3016 | return -ENOTSUP; |
d0b4503e PB |
3017 | } |
3018 | ||
efc75e2a SH |
3019 | static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, |
3020 | Error **errp) | |
93c65b47 AL |
3021 | { |
3022 | int fd; | |
3023 | int ret = 0; | |
3024 | struct stat stat_buf; | |
0e7e1989 | 3025 | int64_t total_size = 0; |
cc28c6aa HR |
3026 | bool has_prefix; |
3027 | ||
f709623b HR |
3028 | /* This function is used by both protocol block drivers and therefore either |
3029 | * of these prefixes may be given. | |
cc28c6aa HR |
3030 | * The return value has to be stored somewhere, otherwise this is an error |
3031 | * due to -Werror=unused-value. */ | |
3032 | has_prefix = | |
3033 | strstart(filename, "host_device:", &filename) || | |
f709623b | 3034 | strstart(filename, "host_cdrom:" , &filename); |
cc28c6aa HR |
3035 | |
3036 | (void)has_prefix; | |
93c65b47 | 3037 | |
bdd03cdf HR |
3038 | ret = raw_normalize_devicepath(&filename); |
3039 | if (ret < 0) { | |
3040 | error_setg_errno(errp, -ret, "Could not normalize device path"); | |
3041 | return ret; | |
3042 | } | |
3043 | ||
0e7e1989 | 3044 | /* Read out options */ |
180e9526 HT |
3045 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
3046 | BDRV_SECTOR_SIZE); | |
93c65b47 | 3047 | |
6165f4d8 | 3048 | fd = qemu_open(filename, O_WRONLY | O_BINARY); |
e428e439 HR |
3049 | if (fd < 0) { |
3050 | ret = -errno; | |
3051 | error_setg_errno(errp, -ret, "Could not open device"); | |
3052 | return ret; | |
3053 | } | |
93c65b47 | 3054 | |
e428e439 | 3055 | if (fstat(fd, &stat_buf) < 0) { |
57e69b7d | 3056 | ret = -errno; |
e428e439 HR |
3057 | error_setg_errno(errp, -ret, "Could not stat device"); |
3058 | } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { | |
3059 | error_setg(errp, | |
3060 | "The given file is neither a block nor a character device"); | |
57e69b7d | 3061 | ret = -ENODEV; |
180e9526 | 3062 | } else if (lseek(fd, 0, SEEK_END) < total_size) { |
e428e439 | 3063 | error_setg(errp, "Device is too small"); |
93c65b47 | 3064 | ret = -ENOSPC; |
e428e439 | 3065 | } |
93c65b47 | 3066 | |
97ec9117 FZ |
3067 | if (!ret && total_size) { |
3068 | uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; | |
3069 | int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); | |
3070 | if (lseek(fd, 0, SEEK_SET) == -1) { | |
3071 | ret = -errno; | |
3072 | } else { | |
3073 | ret = qemu_write_full(fd, buf, zero_size); | |
3074 | ret = ret == zero_size ? 0 : -errno; | |
3075 | } | |
3076 | } | |
2e1e79da | 3077 | qemu_close(fd); |
93c65b47 AL |
3078 | return ret; |
3079 | } | |
3080 | ||
5efa9d5a | 3081 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3082 | .format_name = "host_device", |
84a12e66 | 3083 | .protocol_name = "host_device", |
0b4ce02e | 3084 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3085 | .bdrv_needs_filename = true, |
0b4ce02e | 3086 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3087 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3088 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3089 | .bdrv_close = raw_close, |
1bc6b705 JC |
3090 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3091 | .bdrv_reopen_commit = raw_reopen_commit, | |
3092 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3093 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3094 | .create_opts = &raw_create_opts, |
dd577a26 | 3095 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3096 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3097 | |
9d52aa3c KW |
3098 | .bdrv_co_preadv = raw_co_preadv, |
3099 | .bdrv_co_pwritev = raw_co_pwritev, | |
b2e12bc6 | 3100 | .bdrv_aio_flush = raw_aio_flush, |
4da444a0 | 3101 | .bdrv_aio_pdiscard = hdev_aio_pdiscard, |
1efad060 FZ |
3102 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3103 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3104 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3105 | .bdrv_io_plug = raw_aio_plug, |
3106 | .bdrv_io_unplug = raw_aio_unplug, | |
3c529d93 | 3107 | |
55b949c8 | 3108 | .bdrv_truncate = raw_truncate, |
e60f469c | 3109 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3110 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3111 | .bdrv_get_allocated_file_size |
3112 | = raw_get_allocated_file_size, | |
244a5668 FZ |
3113 | .bdrv_check_perm = raw_check_perm, |
3114 | .bdrv_set_perm = raw_set_perm, | |
3115 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3116 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3117 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3118 | |
f3a5d3f8 | 3119 | /* generic scsi device */ |
63ec93db | 3120 | #ifdef __linux__ |
63ec93db CH |
3121 | .bdrv_aio_ioctl = hdev_aio_ioctl, |
3122 | #endif | |
f3a5d3f8 CH |
3123 | }; |
3124 | ||
18fa1c42 HR |
3125 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3126 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3127 | Error **errp) | |
3128 | { | |
03c320d8 | 3129 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3130 | } |
3131 | #endif | |
3132 | ||
3133 | #ifdef __linux__ | |
015a1036 HR |
3134 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3135 | Error **errp) | |
f3a5d3f8 CH |
3136 | { |
3137 | BDRVRawState *s = bs->opaque; | |
3138 | ||
f3a5d3f8 CH |
3139 | s->type = FTYPE_CD; |
3140 | ||
19a3da7f | 3141 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
9be38598 | 3142 | return raw_open_common(bs, options, flags, O_NONBLOCK, errp); |
f3a5d3f8 CH |
3143 | } |
3144 | ||
508c7cb3 CH |
3145 | static int cdrom_probe_device(const char *filename) |
3146 | { | |
3baf720e CR |
3147 | int fd, ret; |
3148 | int prio = 0; | |
343f8568 | 3149 | struct stat st; |
3baf720e | 3150 | |
6165f4d8 | 3151 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
3baf720e CR |
3152 | if (fd < 0) { |
3153 | goto out; | |
3154 | } | |
343f8568 JS |
3155 | ret = fstat(fd, &st); |
3156 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3157 | goto outc; | |
3158 | } | |
3baf720e CR |
3159 | |
3160 | /* Attempt to detect via a CDROM specific ioctl */ | |
3161 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3162 | if (ret >= 0) | |
3163 | prio = 100; | |
3164 | ||
343f8568 | 3165 | outc: |
2e1e79da | 3166 | qemu_close(fd); |
3baf720e CR |
3167 | out: |
3168 | return prio; | |
508c7cb3 CH |
3169 | } |
3170 | ||
e031f750 | 3171 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3172 | { |
3173 | BDRVRawState *s = bs->opaque; | |
3174 | int ret; | |
3175 | ||
3176 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3177 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3178 | } |
3179 | ||
f36f3949 | 3180 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3181 | { |
3182 | BDRVRawState *s = bs->opaque; | |
3183 | ||
3184 | if (eject_flag) { | |
3185 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3186 | perror("CDROMEJECT"); | |
3187 | } else { | |
3188 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3189 | perror("CDROMEJECT"); | |
3190 | } | |
f3a5d3f8 CH |
3191 | } |
3192 | ||
025e849a | 3193 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3194 | { |
3195 | BDRVRawState *s = bs->opaque; | |
3196 | ||
3197 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3198 | /* | |
3199 | * Note: an error can happen if the distribution automatically | |
3200 | * mounts the CD-ROM | |
3201 | */ | |
3202 | /* perror("CDROM_LOCKDOOR"); */ | |
3203 | } | |
f3a5d3f8 CH |
3204 | } |
3205 | ||
3206 | static BlockDriver bdrv_host_cdrom = { | |
3207 | .format_name = "host_cdrom", | |
84a12e66 | 3208 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3209 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3210 | .bdrv_needs_filename = true, |
508c7cb3 | 3211 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3212 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3213 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3214 | .bdrv_close = raw_close, |
1bc6b705 JC |
3215 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3216 | .bdrv_reopen_commit = raw_reopen_commit, | |
3217 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3218 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3219 | .create_opts = &raw_create_opts, |
dd577a26 | 3220 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3221 | |
9d52aa3c KW |
3222 | |
3223 | .bdrv_co_preadv = raw_co_preadv, | |
3224 | .bdrv_co_pwritev = raw_co_pwritev, | |
b2e12bc6 | 3225 | .bdrv_aio_flush = raw_aio_flush, |
c25f53b0 | 3226 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3227 | .bdrv_io_plug = raw_aio_plug, |
3228 | .bdrv_io_unplug = raw_aio_unplug, | |
f3a5d3f8 | 3229 | |
55b949c8 | 3230 | .bdrv_truncate = raw_truncate, |
b94a2610 KW |
3231 | .bdrv_getlength = raw_getlength, |
3232 | .has_variable_length = true, | |
4a1d5e1f FZ |
3233 | .bdrv_get_allocated_file_size |
3234 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3235 | |
3236 | /* removable device support */ | |
3237 | .bdrv_is_inserted = cdrom_is_inserted, | |
3238 | .bdrv_eject = cdrom_eject, | |
025e849a | 3239 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3240 | |
3241 | /* generic scsi device */ | |
63ec93db | 3242 | .bdrv_aio_ioctl = hdev_aio_ioctl, |
f3a5d3f8 CH |
3243 | }; |
3244 | #endif /* __linux__ */ | |
3245 | ||
a167ba50 | 3246 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3247 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3248 | Error **errp) | |
f3a5d3f8 CH |
3249 | { |
3250 | BDRVRawState *s = bs->opaque; | |
e428e439 | 3251 | Error *local_err = NULL; |
f3a5d3f8 CH |
3252 | int ret; |
3253 | ||
3254 | s->type = FTYPE_CD; | |
3255 | ||
e428e439 HR |
3256 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
3257 | if (ret) { | |
621ff94d | 3258 | error_propagate(errp, local_err); |
f3a5d3f8 | 3259 | return ret; |
e428e439 | 3260 | } |
f3a5d3f8 | 3261 | |
9b2260cb | 3262 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3263 | ioctl(s->fd, CDIOCALLOW); |
3264 | return 0; | |
3265 | } | |
3266 | ||
508c7cb3 CH |
3267 | static int cdrom_probe_device(const char *filename) |
3268 | { | |
3269 | if (strstart(filename, "/dev/cd", NULL) || | |
3270 | strstart(filename, "/dev/acd", NULL)) | |
3271 | return 100; | |
3272 | return 0; | |
3273 | } | |
3274 | ||
f3a5d3f8 CH |
3275 | static int cdrom_reopen(BlockDriverState *bs) |
3276 | { | |
3277 | BDRVRawState *s = bs->opaque; | |
3278 | int fd; | |
3279 | ||
3280 | /* | |
3281 | * Force reread of possibly changed/newly loaded disc, | |
3282 | * FreeBSD seems to not notice sometimes... | |
3283 | */ | |
3284 | if (s->fd >= 0) | |
2e1e79da | 3285 | qemu_close(s->fd); |
6165f4d8 | 3286 | fd = qemu_open(bs->filename, s->open_flags, 0644); |
f3a5d3f8 CH |
3287 | if (fd < 0) { |
3288 | s->fd = -1; | |
3289 | return -EIO; | |
3290 | } | |
3291 | s->fd = fd; | |
3292 | ||
9b2260cb | 3293 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3294 | ioctl(s->fd, CDIOCALLOW); |
3295 | return 0; | |
3296 | } | |
3297 | ||
e031f750 | 3298 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3299 | { |
3300 | return raw_getlength(bs) > 0; | |
3301 | } | |
3302 | ||
f36f3949 | 3303 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3304 | { |
3305 | BDRVRawState *s = bs->opaque; | |
3306 | ||
3307 | if (s->fd < 0) | |
822e1cd1 | 3308 | return; |
f3a5d3f8 CH |
3309 | |
3310 | (void) ioctl(s->fd, CDIOCALLOW); | |
3311 | ||
3312 | if (eject_flag) { | |
3313 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3314 | perror("CDIOCEJECT"); | |
3315 | } else { | |
3316 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3317 | perror("CDIOCCLOSE"); | |
3318 | } | |
3319 | ||
822e1cd1 | 3320 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3321 | } |
3322 | ||
025e849a | 3323 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3324 | { |
3325 | BDRVRawState *s = bs->opaque; | |
3326 | ||
3327 | if (s->fd < 0) | |
7bf37fed | 3328 | return; |
f3a5d3f8 CH |
3329 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3330 | /* | |
3331 | * Note: an error can happen if the distribution automatically | |
3332 | * mounts the CD-ROM | |
3333 | */ | |
3334 | /* perror("CDROM_LOCKDOOR"); */ | |
3335 | } | |
f3a5d3f8 CH |
3336 | } |
3337 | ||
3338 | static BlockDriver bdrv_host_cdrom = { | |
3339 | .format_name = "host_cdrom", | |
84a12e66 | 3340 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3341 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3342 | .bdrv_needs_filename = true, |
508c7cb3 | 3343 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3344 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3345 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3346 | .bdrv_close = raw_close, |
1bc6b705 JC |
3347 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3348 | .bdrv_reopen_commit = raw_reopen_commit, | |
3349 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3350 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3351 | .create_opts = &raw_create_opts, |
f3a5d3f8 | 3352 | |
9d52aa3c KW |
3353 | .bdrv_co_preadv = raw_co_preadv, |
3354 | .bdrv_co_pwritev = raw_co_pwritev, | |
b2e12bc6 | 3355 | .bdrv_aio_flush = raw_aio_flush, |
c25f53b0 | 3356 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3357 | .bdrv_io_plug = raw_aio_plug, |
3358 | .bdrv_io_unplug = raw_aio_unplug, | |
f3a5d3f8 | 3359 | |
55b949c8 | 3360 | .bdrv_truncate = raw_truncate, |
b94a2610 KW |
3361 | .bdrv_getlength = raw_getlength, |
3362 | .has_variable_length = true, | |
4a1d5e1f FZ |
3363 | .bdrv_get_allocated_file_size |
3364 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3365 | |
19cb3738 | 3366 | /* removable device support */ |
f3a5d3f8 CH |
3367 | .bdrv_is_inserted = cdrom_is_inserted, |
3368 | .bdrv_eject = cdrom_eject, | |
025e849a | 3369 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3370 | }; |
f3a5d3f8 | 3371 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3372 | |
84a12e66 | 3373 | static void bdrv_file_init(void) |
5efa9d5a | 3374 | { |
508c7cb3 CH |
3375 | /* |
3376 | * Register all the drivers. Note that order is important, the driver | |
3377 | * registered last will get probed first. | |
3378 | */ | |
84a12e66 | 3379 | bdrv_register(&bdrv_file); |
5efa9d5a | 3380 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3381 | #ifdef __linux__ |
f3a5d3f8 CH |
3382 | bdrv_register(&bdrv_host_cdrom); |
3383 | #endif | |
a167ba50 | 3384 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3385 | bdrv_register(&bdrv_host_cdrom); |
3386 | #endif | |
5efa9d5a AL |
3387 | } |
3388 | ||
84a12e66 | 3389 | block_init(bdrv_file_init); |