]>
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; | |
93f4e2ff KW |
191 | union { |
192 | struct { | |
193 | int aio_fd2; | |
194 | off_t aio_offset2; | |
195 | }; | |
196 | struct { | |
197 | PreallocMode prealloc; | |
198 | Error **errp; | |
199 | }; | |
200 | }; | |
de81a169 PB |
201 | } RawPosixAIOData; |
202 | ||
a167ba50 | 203 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 204 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
205 | #endif |
206 | ||
1de1ae0a CE |
207 | #if defined(__NetBSD__) |
208 | static int raw_normalize_devicepath(const char **filename) | |
209 | { | |
210 | static char namebuf[PATH_MAX]; | |
211 | const char *dp, *fname; | |
212 | struct stat sb; | |
213 | ||
214 | fname = *filename; | |
215 | dp = strrchr(fname, '/'); | |
216 | if (lstat(fname, &sb) < 0) { | |
217 | fprintf(stderr, "%s: stat failed: %s\n", | |
218 | fname, strerror(errno)); | |
219 | return -errno; | |
220 | } | |
221 | ||
222 | if (!S_ISBLK(sb.st_mode)) { | |
223 | return 0; | |
224 | } | |
225 | ||
226 | if (dp == NULL) { | |
227 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
228 | } else { | |
229 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
230 | (int)(dp - fname), fname, dp + 1); | |
231 | } | |
232 | fprintf(stderr, "%s is a block device", fname); | |
233 | *filename = namebuf; | |
234 | fprintf(stderr, ", using %s\n", *filename); | |
235 | ||
236 | return 0; | |
237 | } | |
238 | #else | |
239 | static int raw_normalize_devicepath(const char **filename) | |
240 | { | |
241 | return 0; | |
242 | } | |
243 | #endif | |
244 | ||
8a4ed0d1 ET |
245 | /* |
246 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
247 | */ | |
248 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 249 | { |
c25f53b0 | 250 | unsigned int sector_size; |
8a4ed0d1 | 251 | bool success = false; |
700f9ce0 | 252 | int i; |
c25f53b0 | 253 | |
8a4ed0d1 | 254 | errno = ENOTSUP; |
700f9ce0 | 255 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 256 | #ifdef BLKSSZGET |
700f9ce0 | 257 | BLKSSZGET, |
c25f53b0 PB |
258 | #endif |
259 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 260 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
261 | #endif |
262 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 263 | DIOCGSECTORSIZE, |
c25f53b0 | 264 | #endif |
700f9ce0 PM |
265 | }; |
266 | ||
267 | /* Try a few ioctls to get the right size */ | |
268 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
269 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
270 | *sector_size_p = sector_size; | |
271 | success = true; | |
272 | } | |
273 | } | |
8a4ed0d1 ET |
274 | |
275 | return success ? 0 : -errno; | |
276 | } | |
277 | ||
1a9335e4 ET |
278 | /** |
279 | * Get physical block size of @fd. | |
280 | * On success, store it in @blk_size and return 0. | |
281 | * On failure, return -errno. | |
282 | */ | |
283 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
284 | { | |
285 | #ifdef BLKPBSZGET | |
286 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
287 | return -errno; | |
288 | } | |
289 | return 0; | |
290 | #else | |
291 | return -ENOTSUP; | |
292 | #endif | |
293 | } | |
294 | ||
22d182e8 SH |
295 | /* Check if read is allowed with given memory buffer and length. |
296 | * | |
297 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
298 | */ | |
299 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
300 | { | |
301 | ssize_t ret = pread(fd, buf, len, 0); | |
302 | ||
303 | if (ret >= 0) { | |
304 | return true; | |
305 | } | |
306 | ||
307 | #ifdef __linux__ | |
308 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
309 | * other errors (e.g. real I/O error), which could happen on a failed | |
310 | * drive, since we only care about probing alignment. | |
311 | */ | |
312 | if (errno != EINVAL) { | |
313 | return true; | |
314 | } | |
315 | #endif | |
316 | ||
317 | return false; | |
318 | } | |
319 | ||
8a4ed0d1 ET |
320 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
321 | { | |
322 | BDRVRawState *s = bs->opaque; | |
323 | char *buf; | |
459b4e66 | 324 | size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); |
8a4ed0d1 | 325 | |
b192af8a | 326 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 327 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 328 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 329 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
330 | s->buf_align = 1; |
331 | return; | |
332 | } | |
333 | ||
a5b8dd2c | 334 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
335 | s->buf_align = 0; |
336 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
337 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
338 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 339 | } |
c25f53b0 PB |
340 | #ifdef CONFIG_XFS |
341 | if (s->is_xfs) { | |
342 | struct dioattr da; | |
df26a350 | 343 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
a5b8dd2c | 344 | bs->bl.request_alignment = da.d_miniosz; |
c25f53b0 PB |
345 | /* The kernel returns wrong information for d_mem */ |
346 | /* s->buf_align = da.d_mem; */ | |
347 | } | |
348 | } | |
349 | #endif | |
350 | ||
351 | /* If we could not get the sizes so far, we can only guess them */ | |
352 | if (!s->buf_align) { | |
353 | size_t align; | |
459b4e66 DL |
354 | buf = qemu_memalign(max_align, 2 * max_align); |
355 | for (align = 512; align <= max_align; align <<= 1) { | |
356 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
c25f53b0 PB |
357 | s->buf_align = align; |
358 | break; | |
359 | } | |
360 | } | |
361 | qemu_vfree(buf); | |
362 | } | |
363 | ||
a5b8dd2c | 364 | if (!bs->bl.request_alignment) { |
c25f53b0 | 365 | size_t align; |
459b4e66 DL |
366 | buf = qemu_memalign(s->buf_align, max_align); |
367 | for (align = 512; align <= max_align; align <<= 1) { | |
22d182e8 | 368 | if (raw_is_io_aligned(fd, buf, align)) { |
a5b8dd2c | 369 | bs->bl.request_alignment = align; |
c25f53b0 PB |
370 | break; |
371 | } | |
372 | } | |
373 | qemu_vfree(buf); | |
374 | } | |
df26a350 | 375 | |
a5b8dd2c | 376 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
377 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
378 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 379 | } |
c25f53b0 PB |
380 | } |
381 | ||
6a8dc042 JC |
382 | static void raw_parse_flags(int bdrv_flags, int *open_flags) |
383 | { | |
384 | assert(open_flags != NULL); | |
385 | ||
386 | *open_flags |= O_BINARY; | |
387 | *open_flags &= ~O_ACCMODE; | |
388 | if (bdrv_flags & BDRV_O_RDWR) { | |
389 | *open_flags |= O_RDWR; | |
390 | } else { | |
391 | *open_flags |= O_RDONLY; | |
392 | } | |
393 | ||
394 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
395 | * and O_DIRECT for no caching. */ | |
396 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
397 | *open_flags |= O_DIRECT; | |
398 | } | |
6a8dc042 JC |
399 | } |
400 | ||
078896a9 HR |
401 | static void raw_parse_filename(const char *filename, QDict *options, |
402 | Error **errp) | |
403 | { | |
03c320d8 | 404 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
405 | } |
406 | ||
c66a6157 KW |
407 | static QemuOptsList raw_runtime_opts = { |
408 | .name = "raw", | |
409 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
410 | .desc = { | |
411 | { | |
412 | .name = "filename", | |
413 | .type = QEMU_OPT_STRING, | |
414 | .help = "File name of the image", | |
415 | }, | |
0a4279d9 KW |
416 | { |
417 | .name = "aio", | |
418 | .type = QEMU_OPT_STRING, | |
419 | .help = "host AIO implementation (threads, native)", | |
420 | }, | |
16b48d5d FZ |
421 | { |
422 | .name = "locking", | |
423 | .type = QEMU_OPT_STRING, | |
424 | .help = "file locking mode (on/off/auto, default: auto)", | |
425 | }, | |
7c9e5276 PB |
426 | { |
427 | .name = "pr-manager", | |
428 | .type = QEMU_OPT_STRING, | |
429 | .help = "id of persistent reservation manager object (default: none)", | |
430 | }, | |
31be8a2a SH |
431 | { |
432 | .name = "x-check-cache-dropped", | |
433 | .type = QEMU_OPT_BOOL, | |
434 | .help = "check that page cache was dropped on live migration (default: off)" | |
435 | }, | |
c66a6157 KW |
436 | { /* end of list */ } |
437 | }, | |
438 | }; | |
439 | ||
440 | static int raw_open_common(BlockDriverState *bs, QDict *options, | |
e428e439 | 441 | int bdrv_flags, int open_flags, Error **errp) |
83f64091 FB |
442 | { |
443 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
444 | QemuOpts *opts; |
445 | Error *local_err = NULL; | |
8bfea15d | 446 | const char *filename = NULL; |
7c9e5276 | 447 | const char *str; |
0a4279d9 | 448 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 449 | int fd, ret; |
260a82e5 | 450 | struct stat st; |
244a5668 | 451 | OnOffAuto locking; |
83f64091 | 452 | |
87ea75d5 | 453 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
c66a6157 | 454 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 455 | if (local_err) { |
e428e439 | 456 | error_propagate(errp, local_err); |
c66a6157 KW |
457 | ret = -EINVAL; |
458 | goto fail; | |
459 | } | |
460 | ||
461 | filename = qemu_opt_get(opts, "filename"); | |
462 | ||
1de1ae0a CE |
463 | ret = raw_normalize_devicepath(&filename); |
464 | if (ret != 0) { | |
e428e439 | 465 | error_setg_errno(errp, -ret, "Could not normalize device path"); |
c66a6157 | 466 | goto fail; |
1de1ae0a CE |
467 | } |
468 | ||
0a4279d9 KW |
469 | aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) |
470 | ? BLOCKDEV_AIO_OPTIONS_NATIVE | |
471 | : BLOCKDEV_AIO_OPTIONS_THREADS; | |
f7abe0ec MAL |
472 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
473 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 474 | aio_default, &local_err); |
0a4279d9 KW |
475 | if (local_err) { |
476 | error_propagate(errp, local_err); | |
477 | ret = -EINVAL; | |
478 | goto fail; | |
479 | } | |
480 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); | |
481 | ||
f7abe0ec MAL |
482 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
483 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 484 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
485 | if (local_err) { |
486 | error_propagate(errp, local_err); | |
487 | ret = -EINVAL; | |
488 | goto fail; | |
489 | } | |
490 | switch (locking) { | |
491 | case ON_OFF_AUTO_ON: | |
492 | s->use_lock = true; | |
2b218f5d FZ |
493 | if (!qemu_has_ofd_lock()) { |
494 | fprintf(stderr, | |
495 | "File lock requested but OFD locking syscall is " | |
496 | "unavailable, falling back to POSIX file locks.\n" | |
497 | "Due to the implementation, locks can be lost " | |
498 | "unexpectedly.\n"); | |
499 | } | |
244a5668 FZ |
500 | break; |
501 | case ON_OFF_AUTO_OFF: | |
502 | s->use_lock = false; | |
503 | break; | |
504 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 505 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
506 | break; |
507 | default: | |
508 | abort(); | |
509 | } | |
510 | ||
7c9e5276 PB |
511 | str = qemu_opt_get(opts, "pr-manager"); |
512 | if (str) { | |
513 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
514 | if (local_err) { | |
515 | error_propagate(errp, local_err); | |
516 | ret = -EINVAL; | |
517 | goto fail; | |
518 | } | |
519 | } | |
520 | ||
31be8a2a SH |
521 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
522 | false); | |
523 | ||
6a8dc042 JC |
524 | s->open_flags = open_flags; |
525 | raw_parse_flags(bdrv_flags, &s->open_flags); | |
83f64091 | 526 | |
90babde0 | 527 | s->fd = -1; |
40ff6d7e | 528 | fd = qemu_open(filename, s->open_flags, 0644); |
19cb3738 FB |
529 | if (fd < 0) { |
530 | ret = -errno; | |
09237757 | 531 | error_setg_errno(errp, errno, "Could not open '%s'", filename); |
c66a6157 | 532 | if (ret == -EROFS) { |
19cb3738 | 533 | ret = -EACCES; |
c66a6157 KW |
534 | } |
535 | goto fail; | |
19cb3738 | 536 | } |
83f64091 | 537 | s->fd = fd; |
9ef91a67 | 538 | |
244a5668 FZ |
539 | s->lock_fd = -1; |
540 | if (s->use_lock) { | |
541 | fd = qemu_open(filename, s->open_flags); | |
542 | if (fd < 0) { | |
543 | ret = -errno; | |
544 | error_setg_errno(errp, errno, "Could not open '%s' for locking", | |
545 | filename); | |
546 | qemu_close(s->fd); | |
547 | goto fail; | |
548 | } | |
549 | s->lock_fd = fd; | |
550 | } | |
551 | s->perm = 0; | |
552 | s->shared_perm = BLK_PERM_ALL; | |
553 | ||
5c6c3a6c | 554 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 555 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
556 | if (s->use_linux_aio) { |
557 | if (!(s->open_flags & O_DIRECT)) { | |
558 | error_setg(errp, "aio=native was specified, but it requires " | |
559 | "cache.direct=on, which was not specified."); | |
560 | ret = -EINVAL; | |
561 | goto fail; | |
562 | } | |
563 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
564 | error_prepend(errp, "Unable to use native AIO: "); | |
565 | goto fail; | |
566 | } | |
96518254 | 567 | } |
1501ecc1 | 568 | #else |
0a4279d9 | 569 | if (s->use_linux_aio) { |
d657c0c2 KW |
570 | error_setg(errp, "aio=native was specified, but is not supported " |
571 | "in this build."); | |
572 | ret = -EINVAL; | |
573 | goto fail; | |
1501ecc1 SH |
574 | } |
575 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 576 | |
7ce21016 | 577 | s->has_discard = true; |
97a2ae34 | 578 | s->has_write_zeroes = true; |
3cad8307 RPM |
579 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { |
580 | s->needs_alignment = true; | |
581 | } | |
260a82e5 PB |
582 | |
583 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 584 | ret = -errno; |
260a82e5 PB |
585 | error_setg_errno(errp, errno, "Could not stat file"); |
586 | goto fail; | |
587 | } | |
588 | if (S_ISREG(st.st_mode)) { | |
589 | s->discard_zeroes = true; | |
d50d8222 | 590 | s->has_fallocate = true; |
260a82e5 | 591 | } |
d0b4503e PB |
592 | if (S_ISBLK(st.st_mode)) { |
593 | #ifdef BLKDISCARDZEROES | |
594 | unsigned int arg; | |
595 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
596 | s->discard_zeroes = true; | |
597 | } | |
598 | #endif | |
599 | #ifdef __linux__ | |
600 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
601 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 602 | * Same for BLKZEROOUT. |
d0b4503e PB |
603 | */ |
604 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
605 | s->discard_zeroes = false; | |
97a2ae34 | 606 | s->has_write_zeroes = false; |
d0b4503e PB |
607 | } |
608 | #endif | |
609 | } | |
3cad8307 RPM |
610 | #ifdef __FreeBSD__ |
611 | if (S_ISCHR(st.st_mode)) { | |
612 | /* | |
613 | * The file is a char device (disk), which on FreeBSD isn't behind | |
614 | * a pager, so force all requests to be aligned. This is needed | |
615 | * so QEMU makes sure all IO operations on the device are aligned | |
616 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
617 | */ | |
618 | s->needs_alignment = true; | |
619 | } | |
620 | #endif | |
260a82e5 | 621 | |
dce512de CH |
622 | #ifdef CONFIG_XFS |
623 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 624 | s->is_xfs = true; |
dce512de CH |
625 | } |
626 | #endif | |
627 | ||
e24d813b | 628 | bs->supported_zero_flags = s->discard_zeroes ? BDRV_REQ_MAY_UNMAP : 0; |
c66a6157 KW |
629 | ret = 0; |
630 | fail: | |
8bfea15d KW |
631 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
632 | unlink(filename); | |
633 | } | |
c66a6157 KW |
634 | qemu_opts_del(opts); |
635 | return ret; | |
83f64091 FB |
636 | } |
637 | ||
015a1036 HR |
638 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
639 | Error **errp) | |
90babde0 CH |
640 | { |
641 | BDRVRawState *s = bs->opaque; | |
642 | ||
643 | s->type = FTYPE_FILE; | |
9be38598 | 644 | return raw_open_common(bs, options, flags, 0, errp); |
90babde0 CH |
645 | } |
646 | ||
244a5668 FZ |
647 | typedef enum { |
648 | RAW_PL_PREPARE, | |
649 | RAW_PL_COMMIT, | |
650 | RAW_PL_ABORT, | |
651 | } RawPermLockOp; | |
652 | ||
653 | #define PERM_FOREACH(i) \ | |
654 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
655 | ||
656 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
657 | * file; if @unlock == true, also unlock the unneeded bytes. | |
658 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
659 | */ | |
d0a96155 | 660 | static int raw_apply_lock_bytes(int fd, |
244a5668 FZ |
661 | uint64_t perm_lock_bits, |
662 | uint64_t shared_perm_lock_bits, | |
663 | bool unlock, Error **errp) | |
664 | { | |
665 | int ret; | |
666 | int i; | |
667 | ||
668 | PERM_FOREACH(i) { | |
669 | int off = RAW_LOCK_PERM_BASE + i; | |
670 | if (perm_lock_bits & (1ULL << i)) { | |
d0a96155 | 671 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
672 | if (ret) { |
673 | error_setg(errp, "Failed to lock byte %d", off); | |
674 | return ret; | |
675 | } | |
676 | } else if (unlock) { | |
d0a96155 | 677 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
678 | if (ret) { |
679 | error_setg(errp, "Failed to unlock byte %d", off); | |
680 | return ret; | |
681 | } | |
682 | } | |
683 | } | |
684 | PERM_FOREACH(i) { | |
685 | int off = RAW_LOCK_SHARED_BASE + i; | |
686 | if (shared_perm_lock_bits & (1ULL << i)) { | |
d0a96155 | 687 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
688 | if (ret) { |
689 | error_setg(errp, "Failed to lock byte %d", off); | |
690 | return ret; | |
691 | } | |
692 | } else if (unlock) { | |
d0a96155 | 693 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
694 | if (ret) { |
695 | error_setg(errp, "Failed to unlock byte %d", off); | |
696 | return ret; | |
697 | } | |
698 | } | |
699 | } | |
700 | return 0; | |
701 | } | |
702 | ||
703 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 704 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
705 | Error **errp) |
706 | { | |
707 | int ret; | |
708 | int i; | |
709 | ||
710 | PERM_FOREACH(i) { | |
711 | int off = RAW_LOCK_SHARED_BASE + i; | |
712 | uint64_t p = 1ULL << i; | |
713 | if (perm & p) { | |
d0a96155 | 714 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
715 | if (ret) { |
716 | char *perm_name = bdrv_perm_names(p); | |
717 | error_setg(errp, | |
718 | "Failed to get \"%s\" lock", | |
719 | perm_name); | |
720 | g_free(perm_name); | |
721 | error_append_hint(errp, | |
722 | "Is another process using the image?\n"); | |
723 | return ret; | |
724 | } | |
725 | } | |
726 | } | |
727 | PERM_FOREACH(i) { | |
728 | int off = RAW_LOCK_PERM_BASE + i; | |
729 | uint64_t p = 1ULL << i; | |
730 | if (!(shared_perm & p)) { | |
d0a96155 | 731 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
732 | if (ret) { |
733 | char *perm_name = bdrv_perm_names(p); | |
734 | error_setg(errp, | |
735 | "Failed to get shared \"%s\" lock", | |
736 | perm_name); | |
737 | g_free(perm_name); | |
738 | error_append_hint(errp, | |
739 | "Is another process using the image?\n"); | |
740 | return ret; | |
741 | } | |
742 | } | |
743 | } | |
744 | return 0; | |
745 | } | |
746 | ||
747 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
748 | RawPermLockOp op, | |
749 | uint64_t new_perm, uint64_t new_shared, | |
750 | Error **errp) | |
751 | { | |
752 | BDRVRawState *s = bs->opaque; | |
753 | int ret = 0; | |
754 | Error *local_err = NULL; | |
755 | ||
756 | if (!s->use_lock) { | |
757 | return 0; | |
758 | } | |
759 | ||
760 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
761 | return 0; | |
762 | } | |
763 | ||
764 | assert(s->lock_fd > 0); | |
765 | ||
766 | switch (op) { | |
767 | case RAW_PL_PREPARE: | |
d0a96155 | 768 | ret = raw_apply_lock_bytes(s->lock_fd, s->perm | new_perm, |
244a5668 FZ |
769 | ~s->shared_perm | ~new_shared, |
770 | false, errp); | |
771 | if (!ret) { | |
d0a96155 | 772 | ret = raw_check_lock_bytes(s->lock_fd, new_perm, new_shared, errp); |
244a5668 FZ |
773 | if (!ret) { |
774 | return 0; | |
775 | } | |
776 | } | |
777 | op = RAW_PL_ABORT; | |
778 | /* fall through to unlock bytes. */ | |
779 | case RAW_PL_ABORT: | |
d0a96155 HR |
780 | raw_apply_lock_bytes(s->lock_fd, s->perm, ~s->shared_perm, |
781 | true, &local_err); | |
244a5668 FZ |
782 | if (local_err) { |
783 | /* Theoretically the above call only unlocks bytes and it cannot | |
784 | * fail. Something weird happened, report it. | |
785 | */ | |
786 | error_report_err(local_err); | |
787 | } | |
788 | break; | |
789 | case RAW_PL_COMMIT: | |
d0a96155 HR |
790 | raw_apply_lock_bytes(s->lock_fd, new_perm, ~new_shared, |
791 | true, &local_err); | |
244a5668 FZ |
792 | if (local_err) { |
793 | /* Theoretically the above call only unlocks bytes and it cannot | |
794 | * fail. Something weird happened, report it. | |
795 | */ | |
796 | error_report_err(local_err); | |
797 | } | |
798 | break; | |
799 | } | |
800 | return ret; | |
801 | } | |
802 | ||
eeb6b45d JC |
803 | static int raw_reopen_prepare(BDRVReopenState *state, |
804 | BlockReopenQueue *queue, Error **errp) | |
805 | { | |
806 | BDRVRawState *s; | |
4e6d13c9 | 807 | BDRVRawReopenState *rs; |
31be8a2a | 808 | QemuOpts *opts; |
eeb6b45d | 809 | int ret = 0; |
df26a350 | 810 | Error *local_err = NULL; |
eeb6b45d JC |
811 | |
812 | assert(state != NULL); | |
813 | assert(state->bs != NULL); | |
814 | ||
815 | s = state->bs->opaque; | |
816 | ||
5839e53b | 817 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 818 | rs = state->opaque; |
31be8a2a SH |
819 | rs->fd = -1; |
820 | ||
821 | /* Handle options changes */ | |
822 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
823 | qemu_opts_absorb_qdict(opts, state->options, &local_err); | |
824 | if (local_err) { | |
825 | error_propagate(errp, local_err); | |
826 | ret = -EINVAL; | |
827 | goto out; | |
828 | } | |
829 | ||
830 | rs->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", | |
831 | s->check_cache_dropped); | |
eeb6b45d | 832 | |
f709623b | 833 | if (s->type == FTYPE_CD) { |
4e6d13c9 | 834 | rs->open_flags |= O_NONBLOCK; |
1bc6b705 JC |
835 | } |
836 | ||
4e6d13c9 | 837 | raw_parse_flags(state->flags, &rs->open_flags); |
eeb6b45d | 838 | |
fdf263f6 | 839 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
eeb6b45d JC |
840 | #ifdef O_NOATIME |
841 | fcntl_flags |= O_NOATIME; | |
842 | #endif | |
843 | ||
fdf263f6 AF |
844 | #ifdef O_ASYNC |
845 | /* Not all operating systems have O_ASYNC, and those that don't | |
4e6d13c9 | 846 | * will not let us track the state into rs->open_flags (typically |
fdf263f6 AF |
847 | * you achieve the same effect with an ioctl, for example I_SETSIG |
848 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
849 | */ | |
850 | assert((s->open_flags & O_ASYNC) == 0); | |
851 | #endif | |
852 | ||
4e6d13c9 | 853 | if ((rs->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
eeb6b45d | 854 | /* dup the original fd */ |
4e6d13c9 FZ |
855 | rs->fd = qemu_dup(s->fd); |
856 | if (rs->fd >= 0) { | |
857 | ret = fcntl_setfl(rs->fd, rs->open_flags); | |
eeb6b45d | 858 | if (ret) { |
4e6d13c9 FZ |
859 | qemu_close(rs->fd); |
860 | rs->fd = -1; | |
eeb6b45d JC |
861 | } |
862 | } | |
863 | } | |
864 | ||
865 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ | |
4e6d13c9 | 866 | if (rs->fd == -1) { |
bdd03cdf HR |
867 | const char *normalized_filename = state->bs->filename; |
868 | ret = raw_normalize_devicepath(&normalized_filename); | |
869 | if (ret < 0) { | |
870 | error_setg_errno(errp, -ret, "Could not normalize device path"); | |
871 | } else { | |
4e6d13c9 FZ |
872 | assert(!(rs->open_flags & O_CREAT)); |
873 | rs->fd = qemu_open(normalized_filename, rs->open_flags); | |
874 | if (rs->fd == -1) { | |
bdd03cdf HR |
875 | error_setg_errno(errp, errno, "Could not reopen file"); |
876 | ret = -1; | |
877 | } | |
eeb6b45d JC |
878 | } |
879 | } | |
df26a350 KW |
880 | |
881 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
882 | * alignment with the new fd. */ | |
4e6d13c9 FZ |
883 | if (rs->fd != -1) { |
884 | raw_probe_alignment(state->bs, rs->fd, &local_err); | |
df26a350 | 885 | if (local_err) { |
4e6d13c9 FZ |
886 | qemu_close(rs->fd); |
887 | rs->fd = -1; | |
df26a350 KW |
888 | error_propagate(errp, local_err); |
889 | ret = -EINVAL; | |
890 | } | |
891 | } | |
892 | ||
31be8a2a SH |
893 | out: |
894 | qemu_opts_del(opts); | |
eeb6b45d JC |
895 | return ret; |
896 | } | |
897 | ||
eeb6b45d JC |
898 | static void raw_reopen_commit(BDRVReopenState *state) |
899 | { | |
4e6d13c9 | 900 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
901 | BDRVRawState *s = state->bs->opaque; |
902 | ||
31be8a2a | 903 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 904 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
905 | |
906 | qemu_close(s->fd); | |
4e6d13c9 | 907 | s->fd = rs->fd; |
eeb6b45d JC |
908 | |
909 | g_free(state->opaque); | |
910 | state->opaque = NULL; | |
911 | } | |
912 | ||
913 | ||
914 | static void raw_reopen_abort(BDRVReopenState *state) | |
915 | { | |
4e6d13c9 | 916 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
917 | |
918 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 919 | if (rs == NULL) { |
eeb6b45d JC |
920 | return; |
921 | } | |
922 | ||
4e6d13c9 FZ |
923 | if (rs->fd >= 0) { |
924 | qemu_close(rs->fd); | |
925 | rs->fd = -1; | |
eeb6b45d JC |
926 | } |
927 | g_free(state->opaque); | |
928 | state->opaque = NULL; | |
929 | } | |
930 | ||
48265250 | 931 | static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd) |
6f607174 FZ |
932 | { |
933 | #ifdef BLKSECTGET | |
48265250 EF |
934 | int max_bytes = 0; |
935 | short max_sectors = 0; | |
936 | if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
937 | return max_bytes; | |
938 | } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) { | |
939 | return max_sectors << BDRV_SECTOR_BITS; | |
6f607174 FZ |
940 | } else { |
941 | return -errno; | |
942 | } | |
943 | #else | |
944 | return -ENOSYS; | |
945 | #endif | |
946 | } | |
947 | ||
9103f1ce FZ |
948 | static int hdev_get_max_segments(const struct stat *st) |
949 | { | |
950 | #ifdef CONFIG_LINUX | |
951 | char buf[32]; | |
952 | const char *end; | |
953 | char *sysfspath; | |
954 | int ret; | |
955 | int fd = -1; | |
956 | long max_segments; | |
957 | ||
958 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", | |
959 | major(st->st_rdev), minor(st->st_rdev)); | |
960 | fd = open(sysfspath, O_RDONLY); | |
961 | if (fd == -1) { | |
962 | ret = -errno; | |
963 | goto out; | |
964 | } | |
965 | do { | |
69583490 | 966 | ret = read(fd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
967 | } while (ret == -1 && errno == EINTR); |
968 | if (ret < 0) { | |
969 | ret = -errno; | |
970 | goto out; | |
971 | } else if (ret == 0) { | |
972 | ret = -EIO; | |
973 | goto out; | |
974 | } | |
975 | buf[ret] = 0; | |
976 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
977 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
978 | if (ret == 0 && end && *end == '\n') { | |
979 | ret = max_segments; | |
980 | } | |
981 | ||
982 | out: | |
fed414df FZ |
983 | if (fd != -1) { |
984 | close(fd); | |
985 | } | |
9103f1ce FZ |
986 | g_free(sysfspath); |
987 | return ret; | |
988 | #else | |
989 | return -ENOTSUP; | |
990 | #endif | |
991 | } | |
992 | ||
3baca891 | 993 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
994 | { |
995 | BDRVRawState *s = bs->opaque; | |
6f607174 FZ |
996 | struct stat st; |
997 | ||
998 | if (!fstat(s->fd, &st)) { | |
c4c41a0a | 999 | if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) { |
48265250 EF |
1000 | int ret = hdev_get_max_transfer_length(bs, s->fd); |
1001 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
1002 | bs->bl.max_transfer = pow2floor(ret); | |
6f607174 | 1003 | } |
9103f1ce FZ |
1004 | ret = hdev_get_max_segments(&st); |
1005 | if (ret > 0) { | |
1006 | bs->bl.max_transfer = MIN(bs->bl.max_transfer, | |
1007 | ret * getpagesize()); | |
1008 | } | |
6f607174 FZ |
1009 | } |
1010 | } | |
eeb6b45d | 1011 | |
df26a350 | 1012 | raw_probe_alignment(bs, s->fd, errp); |
4196d2f0 | 1013 | bs->bl.min_mem_alignment = s->buf_align; |
459b4e66 | 1014 | bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); |
c25f53b0 | 1015 | } |
83f64091 | 1016 | |
1a9335e4 ET |
1017 | static int check_for_dasd(int fd) |
1018 | { | |
1019 | #ifdef BIODASDINFO2 | |
1020 | struct dasd_information2_t info = {0}; | |
1021 | ||
1022 | return ioctl(fd, BIODASDINFO2, &info); | |
1023 | #else | |
1024 | return -1; | |
1025 | #endif | |
1026 | } | |
1027 | ||
1028 | /** | |
1029 | * Try to get @bs's logical and physical block size. | |
1030 | * On success, store them in @bsz and return zero. | |
1031 | * On failure, return negative errno. | |
1032 | */ | |
1033 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1034 | { | |
1035 | BDRVRawState *s = bs->opaque; | |
1036 | int ret; | |
1037 | ||
1038 | /* If DASD, get blocksizes */ | |
1039 | if (check_for_dasd(s->fd) < 0) { | |
1040 | return -ENOTSUP; | |
1041 | } | |
1042 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1043 | if (ret < 0) { | |
1044 | return ret; | |
1045 | } | |
1046 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1047 | } | |
1048 | ||
1049 | /** | |
1050 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1051 | * On success, store them in @geo and return 0. | |
1052 | * On failure return -errno. | |
1053 | * (Allows block driver to assign default geometry values that guest sees) | |
1054 | */ | |
1055 | #ifdef __linux__ | |
1056 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1057 | { | |
1058 | BDRVRawState *s = bs->opaque; | |
1059 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1060 | |
1061 | /* If DASD, get its geometry */ | |
1062 | if (check_for_dasd(s->fd) < 0) { | |
1063 | return -ENOTSUP; | |
1064 | } | |
1065 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1066 | return -errno; | |
1067 | } | |
1068 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1069 | (e.g. certain multipath setups) */ | |
1070 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1071 | return -ENOTSUP; | |
1072 | } | |
1073 | /* Do not return a geometry for partition */ | |
1074 | if (ioctl_geo.start != 0) { | |
1075 | return -ENOTSUP; | |
1076 | } | |
1077 | geo->heads = ioctl_geo.heads; | |
1078 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1079 | geo->cylinders = ioctl_geo.cylinders; |
1080 | ||
1081 | return 0; | |
1082 | } | |
1083 | #else /* __linux__ */ | |
1084 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1085 | { | |
1086 | return -ENOTSUP; | |
1087 | } | |
1088 | #endif | |
1089 | ||
de81a169 PB |
1090 | static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb) |
1091 | { | |
1092 | int ret; | |
1093 | ||
1094 | ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); | |
1095 | if (ret == -1) { | |
1096 | return -errno; | |
1097 | } | |
1098 | ||
b608c8dc | 1099 | return 0; |
de81a169 PB |
1100 | } |
1101 | ||
1102 | static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb) | |
1103 | { | |
e5bcf967 | 1104 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1105 | int ret; |
1106 | ||
e5bcf967 KW |
1107 | if (s->page_cache_inconsistent) { |
1108 | return -EIO; | |
1109 | } | |
1110 | ||
de81a169 PB |
1111 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1112 | if (ret == -1) { | |
e5bcf967 KW |
1113 | /* There is no clear definition of the semantics of a failing fsync(), |
1114 | * so we may have to assume the worst. The sad truth is that this | |
1115 | * assumption is correct for Linux. Some pages are now probably marked | |
1116 | * clean in the page cache even though they are inconsistent with the | |
1117 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1118 | * further writeback attempt will be made. We can't get back to a state | |
1119 | * in which we know what is on disk (we would have to rewrite | |
1120 | * everything that was touched since the last fdatasync() at least), so | |
1121 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1122 | * really defined, I have little hope that other OSes are doing better. | |
1123 | * | |
1124 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1125 | * cache. */ | |
1126 | if ((s->open_flags & O_DIRECT) == 0) { | |
1127 | s->page_cache_inconsistent = true; | |
1128 | } | |
de81a169 PB |
1129 | return -errno; |
1130 | } | |
1131 | return 0; | |
1132 | } | |
1133 | ||
1134 | #ifdef CONFIG_PREADV | |
1135 | ||
1136 | static bool preadv_present = true; | |
1137 | ||
1138 | static ssize_t | |
1139 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1140 | { | |
1141 | return preadv(fd, iov, nr_iov, offset); | |
1142 | } | |
1143 | ||
1144 | static ssize_t | |
1145 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1146 | { | |
1147 | return pwritev(fd, iov, nr_iov, offset); | |
1148 | } | |
1149 | ||
1150 | #else | |
1151 | ||
1152 | static bool preadv_present = false; | |
1153 | ||
1154 | static ssize_t | |
1155 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1156 | { | |
1157 | return -ENOSYS; | |
1158 | } | |
1159 | ||
1160 | static ssize_t | |
1161 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1162 | { | |
1163 | return -ENOSYS; | |
1164 | } | |
1165 | ||
1166 | #endif | |
1167 | ||
1168 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1169 | { | |
1170 | ssize_t len; | |
1171 | ||
1172 | do { | |
1173 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1174 | len = qemu_pwritev(aiocb->aio_fildes, | |
1175 | aiocb->aio_iov, | |
1176 | aiocb->aio_niov, | |
1177 | aiocb->aio_offset); | |
1178 | else | |
1179 | len = qemu_preadv(aiocb->aio_fildes, | |
1180 | aiocb->aio_iov, | |
1181 | aiocb->aio_niov, | |
1182 | aiocb->aio_offset); | |
1183 | } while (len == -1 && errno == EINTR); | |
1184 | ||
1185 | if (len == -1) { | |
1186 | return -errno; | |
1187 | } | |
1188 | return len; | |
1189 | } | |
1190 | ||
1191 | /* | |
1192 | * Read/writes the data to/from a given linear buffer. | |
1193 | * | |
1194 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1195 | * reads are only returned if the end of the file is reached. | |
1196 | */ | |
1197 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1198 | { | |
1199 | ssize_t offset = 0; | |
1200 | ssize_t len; | |
1201 | ||
1202 | while (offset < aiocb->aio_nbytes) { | |
1203 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1204 | len = pwrite(aiocb->aio_fildes, | |
1205 | (const char *)buf + offset, | |
1206 | aiocb->aio_nbytes - offset, | |
1207 | aiocb->aio_offset + offset); | |
1208 | } else { | |
1209 | len = pread(aiocb->aio_fildes, | |
1210 | buf + offset, | |
1211 | aiocb->aio_nbytes - offset, | |
1212 | aiocb->aio_offset + offset); | |
1213 | } | |
1214 | if (len == -1 && errno == EINTR) { | |
1215 | continue; | |
61ed73cf SH |
1216 | } else if (len == -1 && errno == EINVAL && |
1217 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1218 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1219 | offset > 0) { | |
1220 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1221 | * after a short read. Assume that O_DIRECT short reads only occur | |
1222 | * at EOF. Therefore this is a short read, not an I/O error. | |
1223 | */ | |
1224 | break; | |
de81a169 PB |
1225 | } else if (len == -1) { |
1226 | offset = -errno; | |
1227 | break; | |
1228 | } else if (len == 0) { | |
1229 | break; | |
1230 | } | |
1231 | offset += len; | |
1232 | } | |
1233 | ||
1234 | return offset; | |
1235 | } | |
1236 | ||
1237 | static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb) | |
1238 | { | |
1239 | ssize_t nbytes; | |
1240 | char *buf; | |
1241 | ||
1242 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1243 | /* | |
1244 | * If there is just a single buffer, and it is properly aligned | |
1245 | * we can just use plain pread/pwrite without any problems. | |
1246 | */ | |
1247 | if (aiocb->aio_niov == 1) { | |
1248 | return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); | |
1249 | } | |
1250 | /* | |
1251 | * We have more than one iovec, and all are properly aligned. | |
1252 | * | |
1253 | * Try preadv/pwritev first and fall back to linearizing the | |
1254 | * buffer if it's not supported. | |
1255 | */ | |
1256 | if (preadv_present) { | |
1257 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1258 | if (nbytes == aiocb->aio_nbytes || | |
1259 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
1260 | return nbytes; | |
1261 | } | |
1262 | preadv_present = false; | |
1263 | } | |
1264 | ||
1265 | /* | |
1266 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1267 | * using these interfaces. For now retry using plain | |
1268 | * pread/pwrite? | |
1269 | */ | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * Ok, we have to do it the hard way, copy all segments into | |
1274 | * a single aligned buffer. | |
1275 | */ | |
50d4a858 KW |
1276 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1277 | if (buf == NULL) { | |
1278 | return -ENOMEM; | |
1279 | } | |
1280 | ||
de81a169 PB |
1281 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1282 | char *p = buf; | |
1283 | int i; | |
1284 | ||
1285 | for (i = 0; i < aiocb->aio_niov; ++i) { | |
1286 | memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); | |
1287 | p += aiocb->aio_iov[i].iov_len; | |
1288 | } | |
8eb029c2 | 1289 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1290 | } |
1291 | ||
1292 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1293 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1294 | char *p = buf; | |
1295 | size_t count = aiocb->aio_nbytes, copy; | |
1296 | int i; | |
1297 | ||
1298 | for (i = 0; i < aiocb->aio_niov && count; ++i) { | |
1299 | copy = count; | |
1300 | if (copy > aiocb->aio_iov[i].iov_len) { | |
1301 | copy = aiocb->aio_iov[i].iov_len; | |
1302 | } | |
1303 | memcpy(aiocb->aio_iov[i].iov_base, p, copy); | |
8eb029c2 | 1304 | assert(count >= copy); |
de81a169 PB |
1305 | p += copy; |
1306 | count -= copy; | |
1307 | } | |
8eb029c2 | 1308 | assert(count == 0); |
de81a169 PB |
1309 | } |
1310 | qemu_vfree(buf); | |
1311 | ||
1312 | return nbytes; | |
1313 | } | |
1314 | ||
8238010b | 1315 | #ifdef CONFIG_XFS |
97a2ae34 PB |
1316 | static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) |
1317 | { | |
1318 | struct xfs_flock64 fl; | |
bcb22555 | 1319 | int err; |
97a2ae34 PB |
1320 | |
1321 | memset(&fl, 0, sizeof(fl)); | |
1322 | fl.l_whence = SEEK_SET; | |
1323 | fl.l_start = offset; | |
1324 | fl.l_len = bytes; | |
1325 | ||
1326 | if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { | |
bcb22555 DA |
1327 | err = errno; |
1328 | DPRINTF("cannot write zero range (%s)\n", strerror(errno)); | |
1329 | return -err; | |
97a2ae34 PB |
1330 | } |
1331 | ||
1332 | return 0; | |
1333 | } | |
1334 | ||
8238010b PB |
1335 | static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) |
1336 | { | |
1337 | struct xfs_flock64 fl; | |
bcb22555 | 1338 | int err; |
8238010b PB |
1339 | |
1340 | memset(&fl, 0, sizeof(fl)); | |
1341 | fl.l_whence = SEEK_SET; | |
1342 | fl.l_start = offset; | |
1343 | fl.l_len = bytes; | |
1344 | ||
1345 | if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { | |
bcb22555 DA |
1346 | err = errno; |
1347 | DPRINTF("cannot punch hole (%s)\n", strerror(errno)); | |
1348 | return -err; | |
8238010b PB |
1349 | } |
1350 | ||
1351 | return 0; | |
1352 | } | |
1353 | #endif | |
1354 | ||
1486df0e DL |
1355 | static int translate_err(int err) |
1356 | { | |
1357 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1358 | err == -ENOTTY) { | |
1359 | err = -ENOTSUP; | |
1360 | } | |
1361 | return err; | |
1362 | } | |
1363 | ||
d50d8222 | 1364 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1365 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1366 | { | |
1367 | do { | |
1368 | if (fallocate(fd, mode, offset, len) == 0) { | |
1369 | return 0; | |
1370 | } | |
1371 | } while (errno == EINTR); | |
1372 | return translate_err(-errno); | |
1373 | } | |
1374 | #endif | |
1375 | ||
37cc9f7f | 1376 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1377 | { |
37cc9f7f | 1378 | int ret = -ENOTSUP; |
97a2ae34 PB |
1379 | BDRVRawState *s = aiocb->bs->opaque; |
1380 | ||
37cc9f7f | 1381 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1382 | return -ENOTSUP; |
1383 | } | |
1384 | ||
97a2ae34 | 1385 | #ifdef BLKZEROOUT |
37cc9f7f DL |
1386 | do { |
1387 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1388 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1389 | return 0; | |
97a2ae34 | 1390 | } |
37cc9f7f DL |
1391 | } while (errno == EINTR); |
1392 | ||
1393 | ret = translate_err(-errno); | |
97a2ae34 | 1394 | #endif |
97a2ae34 | 1395 | |
1486df0e | 1396 | if (ret == -ENOTSUP) { |
97a2ae34 | 1397 | s->has_write_zeroes = false; |
97a2ae34 PB |
1398 | } |
1399 | return ret; | |
1400 | } | |
1401 | ||
37cc9f7f DL |
1402 | static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb) |
1403 | { | |
a6dcf097 | 1404 | #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS) |
37cc9f7f | 1405 | BDRVRawState *s = aiocb->bs->opaque; |
a6dcf097 | 1406 | #endif |
70d9110b DL |
1407 | #ifdef CONFIG_FALLOCATE |
1408 | int64_t len; | |
1409 | #endif | |
37cc9f7f DL |
1410 | |
1411 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1412 | return handle_aiocb_write_zeroes_block(aiocb); | |
1413 | } | |
1414 | ||
1415 | #ifdef CONFIG_XFS | |
1416 | if (s->is_xfs) { | |
1417 | return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1418 | } | |
1419 | #endif | |
1420 | ||
b953f075 DL |
1421 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1422 | if (s->has_write_zeroes) { | |
1423 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1424 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1425 | if (ret == 0 || ret != -ENOTSUP) { | |
1426 | return ret; | |
1427 | } | |
1428 | s->has_write_zeroes = false; | |
1429 | } | |
1430 | #endif | |
1431 | ||
1cdc3239 DL |
1432 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1433 | if (s->has_discard && s->has_fallocate) { | |
1434 | int ret = do_fallocate(s->fd, | |
1435 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1436 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1437 | if (ret == 0) { | |
1438 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1439 | if (ret == 0 || ret != -ENOTSUP) { | |
1440 | return ret; | |
1441 | } | |
1442 | s->has_fallocate = false; | |
1443 | } else if (ret != -ENOTSUP) { | |
1444 | return ret; | |
1445 | } else { | |
1446 | s->has_discard = false; | |
1447 | } | |
1448 | } | |
1449 | #endif | |
1450 | ||
d50d8222 | 1451 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1452 | /* Last resort: we are trying to extend the file with zeroed data. This |
1453 | * can be done via fallocate(fd, 0) */ | |
1454 | len = bdrv_getlength(aiocb->bs); | |
1455 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1456 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1457 | if (ret == 0 || ret != -ENOTSUP) { | |
1458 | return ret; | |
1459 | } | |
1460 | s->has_fallocate = false; | |
1461 | } | |
1462 | #endif | |
1463 | ||
37cc9f7f DL |
1464 | return -ENOTSUP; |
1465 | } | |
1466 | ||
1efad060 FZ |
1467 | #ifndef HAVE_COPY_FILE_RANGE |
1468 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1469 | off_t *out_off, size_t len, unsigned int flags) | |
1470 | { | |
1471 | #ifdef __NR_copy_file_range | |
1472 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1473 | out_off, len, flags); | |
1474 | #else | |
1475 | errno = ENOSYS; | |
1476 | return -1; | |
1477 | #endif | |
1478 | } | |
1479 | #endif | |
1480 | ||
1481 | static ssize_t handle_aiocb_copy_range(RawPosixAIOData *aiocb) | |
1482 | { | |
1483 | uint64_t bytes = aiocb->aio_nbytes; | |
1484 | off_t in_off = aiocb->aio_offset; | |
1485 | off_t out_off = aiocb->aio_offset2; | |
1486 | ||
1487 | while (bytes) { | |
1488 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
1489 | aiocb->aio_fd2, &out_off, | |
1490 | bytes, 0); | |
ecc983a5 FZ |
1491 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
1492 | aiocb->aio_fd2, out_off, bytes, 0, ret); | |
c436e3d0 FZ |
1493 | if (ret == 0) { |
1494 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1495 | * buffer I/O. */ | |
1496 | return -ENOSPC; | |
1efad060 FZ |
1497 | } |
1498 | if (ret < 0) { | |
c436e3d0 FZ |
1499 | switch (errno) { |
1500 | case ENOSYS: | |
1efad060 | 1501 | return -ENOTSUP; |
c436e3d0 FZ |
1502 | case EINTR: |
1503 | continue; | |
1504 | default: | |
1efad060 FZ |
1505 | return -errno; |
1506 | } | |
1507 | } | |
1efad060 FZ |
1508 | bytes -= ret; |
1509 | } | |
1510 | return 0; | |
1511 | } | |
1512 | ||
8238010b PB |
1513 | static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb) |
1514 | { | |
1515 | int ret = -EOPNOTSUPP; | |
1516 | BDRVRawState *s = aiocb->bs->opaque; | |
1517 | ||
7ce21016 PB |
1518 | if (!s->has_discard) { |
1519 | return -ENOTSUP; | |
8238010b PB |
1520 | } |
1521 | ||
1522 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1523 | #ifdef BLKDISCARD | |
1524 | do { | |
1525 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1526 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1527 | return 0; | |
1528 | } | |
1529 | } while (errno == EINTR); | |
1530 | ||
1531 | ret = -errno; | |
1532 | #endif | |
1533 | } else { | |
1534 | #ifdef CONFIG_XFS | |
1535 | if (s->is_xfs) { | |
1536 | return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1537 | } | |
1538 | #endif | |
1539 | ||
1540 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
0b991712 DL |
1541 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1542 | aiocb->aio_offset, aiocb->aio_nbytes); | |
8238010b PB |
1543 | #endif |
1544 | } | |
1545 | ||
1486df0e DL |
1546 | ret = translate_err(ret); |
1547 | if (ret == -ENOTSUP) { | |
7ce21016 | 1548 | s->has_discard = false; |
8238010b PB |
1549 | } |
1550 | return ret; | |
1551 | } | |
1552 | ||
93f4e2ff KW |
1553 | static int handle_aiocb_truncate(RawPosixAIOData *aiocb) |
1554 | { | |
1555 | int result = 0; | |
1556 | int64_t current_length = 0; | |
1557 | char *buf = NULL; | |
1558 | struct stat st; | |
1559 | int fd = aiocb->aio_fildes; | |
1560 | int64_t offset = aiocb->aio_offset; | |
1561 | Error **errp = aiocb->errp; | |
1562 | ||
1563 | if (fstat(fd, &st) < 0) { | |
1564 | result = -errno; | |
1565 | error_setg_errno(errp, -result, "Could not stat file"); | |
1566 | return result; | |
1567 | } | |
1568 | ||
1569 | current_length = st.st_size; | |
1570 | if (current_length > offset && aiocb->prealloc != PREALLOC_MODE_OFF) { | |
1571 | error_setg(errp, "Cannot use preallocation for shrinking files"); | |
1572 | return -ENOTSUP; | |
1573 | } | |
1574 | ||
1575 | switch (aiocb->prealloc) { | |
1576 | #ifdef CONFIG_POSIX_FALLOCATE | |
1577 | case PREALLOC_MODE_FALLOC: | |
1578 | /* | |
1579 | * Truncating before posix_fallocate() makes it about twice slower on | |
1580 | * file systems that do not support fallocate(), trying to check if a | |
1581 | * block is allocated before allocating it, so don't do that here. | |
1582 | */ | |
1583 | if (offset != current_length) { | |
1584 | result = -posix_fallocate(fd, current_length, | |
1585 | offset - current_length); | |
1586 | if (result != 0) { | |
1587 | /* posix_fallocate() doesn't set errno. */ | |
1588 | error_setg_errno(errp, -result, | |
1589 | "Could not preallocate new data"); | |
1590 | } | |
1591 | } else { | |
1592 | result = 0; | |
1593 | } | |
1594 | goto out; | |
1595 | #endif | |
1596 | case PREALLOC_MODE_FULL: | |
1597 | { | |
1598 | int64_t num = 0, left = offset - current_length; | |
1599 | off_t seek_result; | |
1600 | ||
1601 | /* | |
1602 | * Knowing the final size from the beginning could allow the file | |
1603 | * system driver to do less allocations and possibly avoid | |
1604 | * fragmentation of the file. | |
1605 | */ | |
1606 | if (ftruncate(fd, offset) != 0) { | |
1607 | result = -errno; | |
1608 | error_setg_errno(errp, -result, "Could not resize file"); | |
1609 | goto out; | |
1610 | } | |
1611 | ||
1612 | buf = g_malloc0(65536); | |
1613 | ||
1614 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1615 | if (seek_result < 0) { | |
1616 | result = -errno; | |
1617 | error_setg_errno(errp, -result, | |
1618 | "Failed to seek to the old end of file"); | |
1619 | goto out; | |
1620 | } | |
1621 | ||
1622 | while (left > 0) { | |
1623 | num = MIN(left, 65536); | |
1624 | result = write(fd, buf, num); | |
1625 | if (result < 0) { | |
1626 | result = -errno; | |
1627 | error_setg_errno(errp, -result, | |
1628 | "Could not write zeros for preallocation"); | |
1629 | goto out; | |
1630 | } | |
1631 | left -= result; | |
1632 | } | |
1633 | if (result >= 0) { | |
1634 | result = fsync(fd); | |
1635 | if (result < 0) { | |
1636 | result = -errno; | |
1637 | error_setg_errno(errp, -result, | |
1638 | "Could not flush file to disk"); | |
1639 | goto out; | |
1640 | } | |
1641 | } | |
1642 | goto out; | |
1643 | } | |
1644 | case PREALLOC_MODE_OFF: | |
1645 | if (ftruncate(fd, offset) != 0) { | |
1646 | result = -errno; | |
1647 | error_setg_errno(errp, -result, "Could not resize file"); | |
1648 | } | |
1649 | return result; | |
1650 | default: | |
1651 | result = -ENOTSUP; | |
1652 | error_setg(errp, "Unsupported preallocation mode: %s", | |
1653 | PreallocMode_str(aiocb->prealloc)); | |
1654 | return result; | |
1655 | } | |
1656 | ||
1657 | out: | |
1658 | if (result < 0) { | |
1659 | if (ftruncate(fd, current_length) < 0) { | |
1660 | error_report("Failed to restore old file length: %s", | |
1661 | strerror(errno)); | |
1662 | } | |
1663 | } | |
1664 | ||
1665 | g_free(buf); | |
1666 | return result; | |
1667 | } | |
1668 | ||
de81a169 PB |
1669 | static int aio_worker(void *arg) |
1670 | { | |
1671 | RawPosixAIOData *aiocb = arg; | |
1672 | ssize_t ret = 0; | |
1673 | ||
1674 | switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { | |
1675 | case QEMU_AIO_READ: | |
1676 | ret = handle_aiocb_rw(aiocb); | |
c0191e76 | 1677 | if (ret >= 0 && ret < aiocb->aio_nbytes) { |
de81a169 PB |
1678 | iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret, |
1679 | 0, aiocb->aio_nbytes - ret); | |
1680 | ||
1681 | ret = aiocb->aio_nbytes; | |
1682 | } | |
1683 | if (ret == aiocb->aio_nbytes) { | |
1684 | ret = 0; | |
1685 | } else if (ret >= 0 && ret < aiocb->aio_nbytes) { | |
1686 | ret = -EINVAL; | |
1687 | } | |
1688 | break; | |
1689 | case QEMU_AIO_WRITE: | |
1690 | ret = handle_aiocb_rw(aiocb); | |
1691 | if (ret == aiocb->aio_nbytes) { | |
1692 | ret = 0; | |
1693 | } else if (ret >= 0 && ret < aiocb->aio_nbytes) { | |
1694 | ret = -EINVAL; | |
1695 | } | |
1696 | break; | |
1697 | case QEMU_AIO_FLUSH: | |
1698 | ret = handle_aiocb_flush(aiocb); | |
1699 | break; | |
1700 | case QEMU_AIO_IOCTL: | |
1701 | ret = handle_aiocb_ioctl(aiocb); | |
1702 | break; | |
8238010b PB |
1703 | case QEMU_AIO_DISCARD: |
1704 | ret = handle_aiocb_discard(aiocb); | |
1705 | break; | |
97a2ae34 PB |
1706 | case QEMU_AIO_WRITE_ZEROES: |
1707 | ret = handle_aiocb_write_zeroes(aiocb); | |
1708 | break; | |
1efad060 FZ |
1709 | case QEMU_AIO_COPY_RANGE: |
1710 | ret = handle_aiocb_copy_range(aiocb); | |
1711 | break; | |
93f4e2ff KW |
1712 | case QEMU_AIO_TRUNCATE: |
1713 | ret = handle_aiocb_truncate(aiocb); | |
1714 | break; | |
de81a169 PB |
1715 | default: |
1716 | fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); | |
1717 | ret = -EINVAL; | |
1718 | break; | |
1719 | } | |
1720 | ||
c84b3192 | 1721 | g_free(aiocb); |
de81a169 PB |
1722 | return ret; |
1723 | } | |
1724 | ||
1efad060 FZ |
1725 | static int paio_submit_co_full(BlockDriverState *bs, int fd, |
1726 | int64_t offset, int fd2, int64_t offset2, | |
1727 | QEMUIOVector *qiov, | |
1728 | int bytes, int type) | |
260a82e5 | 1729 | { |
c84b3192 | 1730 | RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); |
260a82e5 PB |
1731 | ThreadPool *pool; |
1732 | ||
1733 | acb->bs = bs; | |
1734 | acb->aio_type = type; | |
1735 | acb->aio_fildes = fd; | |
1efad060 FZ |
1736 | acb->aio_fd2 = fd2; |
1737 | acb->aio_offset2 = offset2; | |
260a82e5 | 1738 | |
f5a5ca79 | 1739 | acb->aio_nbytes = bytes; |
2ffa76c2 | 1740 | acb->aio_offset = offset; |
8eb029c2 | 1741 | |
260a82e5 PB |
1742 | if (qiov) { |
1743 | acb->aio_iov = qiov->iov; | |
1744 | acb->aio_niov = qiov->niov; | |
f5a5ca79 | 1745 | assert(qiov->size == bytes); |
260a82e5 | 1746 | } |
260a82e5 | 1747 | |
f8a30874 | 1748 | trace_file_paio_submit_co(offset, bytes, type); |
260a82e5 PB |
1749 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
1750 | return thread_pool_submit_co(pool, aio_worker, acb); | |
1751 | } | |
1752 | ||
1efad060 FZ |
1753 | static inline int paio_submit_co(BlockDriverState *bs, int fd, |
1754 | int64_t offset, QEMUIOVector *qiov, | |
1755 | int bytes, int type) | |
1756 | { | |
1757 | return paio_submit_co_full(bs, fd, offset, -1, 0, qiov, bytes, type); | |
1758 | } | |
1759 | ||
9d52aa3c KW |
1760 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
1761 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 1762 | { |
ce1a14dc | 1763 | BDRVRawState *s = bs->opaque; |
ce1a14dc | 1764 | |
19cb3738 | 1765 | if (fd_open(bs) < 0) |
2174f12b | 1766 | return -EIO; |
19cb3738 | 1767 | |
f141eafe | 1768 | /* |
3cad8307 RPM |
1769 | * Check if the underlying device requires requests to be aligned, |
1770 | * and if the request we are trying to submit is aligned or not. | |
1771 | * If this is the case tell the low-level driver that it needs | |
1772 | * to copy the buffer. | |
f141eafe | 1773 | */ |
3cad8307 | 1774 | if (s->needs_alignment) { |
c53b1c51 | 1775 | if (!bdrv_qiov_is_aligned(bs, qiov)) { |
5c6c3a6c | 1776 | type |= QEMU_AIO_MISALIGNED; |
e44bd6fc | 1777 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 1778 | } else if (s->use_linux_aio) { |
0187f5c9 | 1779 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
9d52aa3c | 1780 | assert(qiov->size == bytes); |
0187f5c9 | 1781 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type); |
e44bd6fc | 1782 | #endif |
5c6c3a6c | 1783 | } |
9ef91a67 | 1784 | } |
f141eafe | 1785 | |
9d52aa3c | 1786 | return paio_submit_co(bs, s->fd, offset, qiov, bytes, type); |
2174f12b KW |
1787 | } |
1788 | ||
9d52aa3c KW |
1789 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, |
1790 | uint64_t bytes, QEMUIOVector *qiov, | |
1791 | int flags) | |
2174f12b | 1792 | { |
9d52aa3c | 1793 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
1794 | } |
1795 | ||
9d52aa3c KW |
1796 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, |
1797 | uint64_t bytes, QEMUIOVector *qiov, | |
1798 | int flags) | |
2174f12b | 1799 | { |
9d52aa3c KW |
1800 | assert(flags == 0); |
1801 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
1802 | } |
1803 | ||
1b3abdcc ML |
1804 | static void raw_aio_plug(BlockDriverState *bs) |
1805 | { | |
1806 | #ifdef CONFIG_LINUX_AIO | |
0a4279d9 KW |
1807 | BDRVRawState *s = bs->opaque; |
1808 | if (s->use_linux_aio) { | |
0187f5c9 PB |
1809 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1810 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
1811 | } |
1812 | #endif | |
1813 | } | |
1814 | ||
1815 | static void raw_aio_unplug(BlockDriverState *bs) | |
1816 | { | |
1817 | #ifdef CONFIG_LINUX_AIO | |
0a4279d9 KW |
1818 | BDRVRawState *s = bs->opaque; |
1819 | if (s->use_linux_aio) { | |
0187f5c9 PB |
1820 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1821 | laio_io_unplug(bs, aio); | |
1b3abdcc ML |
1822 | } |
1823 | #endif | |
1824 | } | |
1825 | ||
33d70fb6 | 1826 | static int raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
1827 | { |
1828 | BDRVRawState *s = bs->opaque; | |
33d70fb6 | 1829 | int ret; |
b2e12bc6 | 1830 | |
33d70fb6 KW |
1831 | ret = fd_open(bs); |
1832 | if (ret < 0) { | |
1833 | return ret; | |
1834 | } | |
b2e12bc6 | 1835 | |
33d70fb6 | 1836 | return paio_submit_co(bs, s->fd, 0, NULL, 0, QEMU_AIO_FLUSH); |
b2e12bc6 CH |
1837 | } |
1838 | ||
ed6e2161 NA |
1839 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
1840 | AioContext *new_context) | |
1841 | { | |
1842 | #ifdef CONFIG_LINUX_AIO | |
1843 | BDRVRawState *s = bs->opaque; | |
1844 | if (s->use_linux_aio) { | |
1845 | Error *local_err; | |
1846 | if (!aio_setup_linux_aio(new_context, &local_err)) { | |
1847 | error_reportf_err(local_err, "Unable to use native AIO, " | |
1848 | "falling back to thread pool: "); | |
1849 | s->use_linux_aio = false; | |
1850 | } | |
1851 | } | |
1852 | #endif | |
1853 | } | |
1854 | ||
83f64091 FB |
1855 | static void raw_close(BlockDriverState *bs) |
1856 | { | |
1857 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 1858 | |
19cb3738 | 1859 | if (s->fd >= 0) { |
2e1e79da | 1860 | qemu_close(s->fd); |
19cb3738 FB |
1861 | s->fd = -1; |
1862 | } | |
244a5668 FZ |
1863 | if (s->lock_fd >= 0) { |
1864 | qemu_close(s->lock_fd); | |
1865 | s->lock_fd = -1; | |
1866 | } | |
83f64091 FB |
1867 | } |
1868 | ||
d0bc9e5d HR |
1869 | /** |
1870 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
1871 | * new space according to @prealloc. | |
1872 | * | |
1873 | * Returns: 0 on success, -errno on failure. | |
1874 | */ | |
93f4e2ff KW |
1875 | static int coroutine_fn |
1876 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
1877 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 1878 | { |
93f4e2ff KW |
1879 | RawPosixAIOData *acb = g_new(RawPosixAIOData, 1); |
1880 | ThreadPool *pool; | |
d0bc9e5d | 1881 | |
93f4e2ff KW |
1882 | *acb = (RawPosixAIOData) { |
1883 | .bs = bs, | |
1884 | .aio_fildes = fd, | |
1885 | .aio_type = QEMU_AIO_TRUNCATE, | |
1886 | .aio_offset = offset, | |
1887 | .prealloc = prealloc, | |
1888 | .errp = errp, | |
1889 | }; | |
d0bc9e5d | 1890 | |
93f4e2ff KW |
1891 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ |
1892 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
1893 | return thread_pool_submit_co(pool, aio_worker, acb); | |
9f63b07e HR |
1894 | } |
1895 | ||
061ca8a3 KW |
1896 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
1897 | PreallocMode prealloc, Error **errp) | |
83f64091 FB |
1898 | { |
1899 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 1900 | struct stat st; |
f59adb32 | 1901 | int ret; |
55b949c8 CH |
1902 | |
1903 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
1904 | ret = -errno; |
1905 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
1906 | return ret; | |
55b949c8 CH |
1907 | } |
1908 | ||
1909 | if (S_ISREG(st.st_mode)) { | |
93f4e2ff | 1910 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
1911 | } |
1912 | ||
1913 | if (prealloc != PREALLOC_MODE_OFF) { | |
1914 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 1915 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
1916 | return -ENOTSUP; |
1917 | } | |
1918 | ||
1919 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
f59adb32 HR |
1920 | if (offset > raw_getlength(bs)) { |
1921 | error_setg(errp, "Cannot grow device files"); | |
1922 | return -EINVAL; | |
1923 | } | |
55b949c8 | 1924 | } else { |
f59adb32 | 1925 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
1926 | return -ENOTSUP; |
1927 | } | |
1928 | ||
83f64091 FB |
1929 | return 0; |
1930 | } | |
1931 | ||
128ab2ff BS |
1932 | #ifdef __OpenBSD__ |
1933 | static int64_t raw_getlength(BlockDriverState *bs) | |
1934 | { | |
1935 | BDRVRawState *s = bs->opaque; | |
1936 | int fd = s->fd; | |
1937 | struct stat st; | |
1938 | ||
1939 | if (fstat(fd, &st)) | |
aa729704 | 1940 | return -errno; |
128ab2ff BS |
1941 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
1942 | struct disklabel dl; | |
1943 | ||
1944 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 1945 | return -errno; |
128ab2ff BS |
1946 | return (uint64_t)dl.d_secsize * |
1947 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
1948 | } else | |
1949 | return st.st_size; | |
1950 | } | |
d1f6fd8d CE |
1951 | #elif defined(__NetBSD__) |
1952 | static int64_t raw_getlength(BlockDriverState *bs) | |
1953 | { | |
1954 | BDRVRawState *s = bs->opaque; | |
1955 | int fd = s->fd; | |
1956 | struct stat st; | |
1957 | ||
1958 | if (fstat(fd, &st)) | |
aa729704 | 1959 | return -errno; |
d1f6fd8d CE |
1960 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
1961 | struct dkwedge_info dkw; | |
1962 | ||
1963 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
1964 | return dkw.dkw_size * 512; | |
1965 | } else { | |
1966 | struct disklabel dl; | |
1967 | ||
1968 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 1969 | return -errno; |
d1f6fd8d CE |
1970 | return (uint64_t)dl.d_secsize * |
1971 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
1972 | } | |
1973 | } else | |
1974 | return st.st_size; | |
1975 | } | |
50779cc2 CH |
1976 | #elif defined(__sun__) |
1977 | static int64_t raw_getlength(BlockDriverState *bs) | |
1978 | { | |
1979 | BDRVRawState *s = bs->opaque; | |
1980 | struct dk_minfo minfo; | |
1981 | int ret; | |
aa729704 | 1982 | int64_t size; |
50779cc2 CH |
1983 | |
1984 | ret = fd_open(bs); | |
1985 | if (ret < 0) { | |
1986 | return ret; | |
1987 | } | |
1988 | ||
1989 | /* | |
1990 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
1991 | */ | |
1992 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
1993 | if (ret != -1) { | |
1994 | return minfo.dki_lbsize * minfo.dki_capacity; | |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * There are reports that lseek on some devices fails, but | |
1999 | * irc discussion said that contingency on contingency was overkill. | |
2000 | */ | |
aa729704 MA |
2001 | size = lseek(s->fd, 0, SEEK_END); |
2002 | if (size < 0) { | |
2003 | return -errno; | |
2004 | } | |
2005 | return size; | |
50779cc2 CH |
2006 | } |
2007 | #elif defined(CONFIG_BSD) | |
2008 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2009 | { |
2010 | BDRVRawState *s = bs->opaque; | |
2011 | int fd = s->fd; | |
2012 | int64_t size; | |
83f64091 | 2013 | struct stat sb; |
a167ba50 | 2014 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2015 | int reopened = 0; |
83f64091 | 2016 | #endif |
19cb3738 FB |
2017 | int ret; |
2018 | ||
2019 | ret = fd_open(bs); | |
2020 | if (ret < 0) | |
2021 | return ret; | |
83f64091 | 2022 | |
a167ba50 | 2023 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2024 | again: |
2025 | #endif | |
83f64091 FB |
2026 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
2027 | #ifdef DIOCGMEDIASIZE | |
2028 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) | |
c5e97233 BS |
2029 | #elif defined(DIOCGPART) |
2030 | { | |
2031 | struct partinfo pi; | |
2032 | if (ioctl(fd, DIOCGPART, &pi) == 0) | |
2033 | size = pi.media_size; | |
2034 | else | |
2035 | size = 0; | |
2036 | } | |
2037 | if (size == 0) | |
83f64091 | 2038 | #endif |
83affaa6 | 2039 | #if defined(__APPLE__) && defined(__MACH__) |
728dacbd JA |
2040 | { |
2041 | uint64_t sectors = 0; | |
2042 | uint32_t sector_size = 0; | |
2043 | ||
2044 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2045 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2046 | size = sectors * sector_size; | |
2047 | } else { | |
2048 | size = lseek(fd, 0LL, SEEK_END); | |
2049 | if (size < 0) { | |
2050 | return -errno; | |
2051 | } | |
2052 | } | |
2053 | } | |
83f64091 FB |
2054 | #else |
2055 | size = lseek(fd, 0LL, SEEK_END); | |
aa729704 MA |
2056 | if (size < 0) { |
2057 | return -errno; | |
2058 | } | |
9f23011a | 2059 | #endif |
a167ba50 | 2060 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2061 | switch(s->type) { |
2062 | case FTYPE_CD: | |
2063 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2064 | if (size == 2048LL * (unsigned)-1) | |
2065 | size = 0; | |
2066 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2067 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2068 | reopened = 1; |
2069 | goto again; | |
2070 | } | |
2071 | } | |
83f64091 | 2072 | #endif |
50779cc2 | 2073 | } else { |
83f64091 | 2074 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2075 | if (size < 0) { |
2076 | return -errno; | |
2077 | } | |
83f64091 | 2078 | } |
83f64091 FB |
2079 | return size; |
2080 | } | |
50779cc2 CH |
2081 | #else |
2082 | static int64_t raw_getlength(BlockDriverState *bs) | |
2083 | { | |
2084 | BDRVRawState *s = bs->opaque; | |
2085 | int ret; | |
aa729704 | 2086 | int64_t size; |
50779cc2 CH |
2087 | |
2088 | ret = fd_open(bs); | |
2089 | if (ret < 0) { | |
2090 | return ret; | |
2091 | } | |
2092 | ||
aa729704 MA |
2093 | size = lseek(s->fd, 0, SEEK_END); |
2094 | if (size < 0) { | |
2095 | return -errno; | |
2096 | } | |
2097 | return size; | |
50779cc2 | 2098 | } |
128ab2ff | 2099 | #endif |
83f64091 | 2100 | |
4a1d5e1f FZ |
2101 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2102 | { | |
2103 | struct stat st; | |
2104 | BDRVRawState *s = bs->opaque; | |
2105 | ||
2106 | if (fstat(s->fd, &st) < 0) { | |
2107 | return -errno; | |
2108 | } | |
2109 | return (int64_t)st.st_blocks * 512; | |
2110 | } | |
2111 | ||
93f4e2ff KW |
2112 | static int coroutine_fn |
2113 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2114 | { |
927f11e1 | 2115 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2116 | Error *local_err = NULL; |
83f64091 | 2117 | int fd; |
d815efca | 2118 | uint64_t perm, shared; |
1e37d059 | 2119 | int result = 0; |
83f64091 | 2120 | |
927f11e1 KW |
2121 | /* Validate options and set default values */ |
2122 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2123 | file_opts = &options->u.file; | |
464d9f64 | 2124 | |
927f11e1 KW |
2125 | if (!file_opts->has_nocow) { |
2126 | file_opts->nocow = false; | |
2127 | } | |
2128 | if (!file_opts->has_preallocation) { | |
2129 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2130 | } |
83f64091 | 2131 | |
927f11e1 | 2132 | /* Create file */ |
b8cf1913 | 2133 | fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); |
1e37d059 SW |
2134 | if (fd < 0) { |
2135 | result = -errno; | |
e428e439 | 2136 | error_setg_errno(errp, -result, "Could not create file"); |
06247428 HT |
2137 | goto out; |
2138 | } | |
2139 | ||
b8cf1913 HR |
2140 | /* Take permissions: We want to discard everything, so we need |
2141 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2142 | * BLK_PERM_RESIZE. | |
2143 | * On the other hand, we cannot share the RESIZE permission | |
2144 | * because we promise that after this function, the file has the | |
2145 | * size given in the options. If someone else were to resize it | |
2146 | * concurrently, we could not guarantee that. | |
2147 | * Note that after this function, we can no longer guarantee that | |
2148 | * the file is not touched by a third party, so it may be resized | |
2149 | * then. */ | |
2150 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2151 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2152 | ||
2153 | /* Step one: Take locks */ | |
d815efca | 2154 | result = raw_apply_lock_bytes(fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2155 | if (result < 0) { |
2156 | goto out_close; | |
2157 | } | |
2158 | ||
2159 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2160 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2161 | if (result < 0) { | |
7c20c808 | 2162 | goto out_unlock; |
b8cf1913 HR |
2163 | } |
2164 | ||
2165 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2166 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2167 | if (result < 0) { |
7c20c808 | 2168 | goto out_unlock; |
b8cf1913 HR |
2169 | } |
2170 | ||
927f11e1 | 2171 | if (file_opts->nocow) { |
4ab15590 | 2172 | #ifdef __linux__ |
06247428 HT |
2173 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2174 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2175 | * will be ignored since any failure of this operation should not | |
2176 | * block the left work. | |
2177 | */ | |
2178 | int attr; | |
2179 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2180 | attr |= FS_NOCOW_FL; | |
2181 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2182 | } |
06247428 HT |
2183 | #endif |
2184 | } | |
2185 | ||
b8cf1913 HR |
2186 | /* Resize and potentially preallocate the file to the desired |
2187 | * final size */ | |
93f4e2ff KW |
2188 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2189 | file_opts->preallocation, errp); | |
9f63b07e | 2190 | if (result < 0) { |
7c20c808 HR |
2191 | goto out_unlock; |
2192 | } | |
2193 | ||
2194 | out_unlock: | |
2195 | raw_apply_lock_bytes(fd, 0, 0, true, &local_err); | |
2196 | if (local_err) { | |
2197 | /* The above call should not fail, and if it does, that does | |
2198 | * not mean the whole creation operation has failed. So | |
2199 | * report it the user for their convenience, but do not report | |
2200 | * it to the caller. */ | |
2201 | error_report_err(local_err); | |
1e37d059 | 2202 | } |
06247428 | 2203 | |
5a1dad9d | 2204 | out_close: |
06247428 HT |
2205 | if (qemu_close(fd) != 0 && result == 0) { |
2206 | result = -errno; | |
2207 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2208 | } | |
2209 | out: | |
1e37d059 | 2210 | return result; |
83f64091 FB |
2211 | } |
2212 | ||
927f11e1 KW |
2213 | static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, |
2214 | Error **errp) | |
2215 | { | |
2216 | BlockdevCreateOptions options; | |
2217 | int64_t total_size = 0; | |
2218 | bool nocow = false; | |
2219 | PreallocMode prealloc; | |
2220 | char *buf = NULL; | |
2221 | Error *local_err = NULL; | |
2222 | ||
2223 | /* Skip file: protocol prefix */ | |
2224 | strstart(filename, "file:", &filename); | |
2225 | ||
2226 | /* Read out options */ | |
2227 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2228 | BDRV_SECTOR_SIZE); | |
2229 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); | |
2230 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2231 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2232 | PREALLOC_MODE_OFF, &local_err); | |
2233 | g_free(buf); | |
2234 | if (local_err) { | |
2235 | error_propagate(errp, local_err); | |
2236 | return -EINVAL; | |
2237 | } | |
2238 | ||
2239 | options = (BlockdevCreateOptions) { | |
2240 | .driver = BLOCKDEV_DRIVER_FILE, | |
2241 | .u.file = { | |
2242 | .filename = (char *) filename, | |
2243 | .size = total_size, | |
2244 | .has_preallocation = true, | |
2245 | .preallocation = prealloc, | |
2246 | .has_nocow = true, | |
2247 | .nocow = nocow, | |
2248 | }, | |
2249 | }; | |
2250 | return raw_co_create(&options, errp); | |
2251 | } | |
2252 | ||
d1f06fe6 MA |
2253 | /* |
2254 | * Find allocation range in @bs around offset @start. | |
2255 | * May change underlying file descriptor's file offset. | |
2256 | * If @start is not in a hole, store @start in @data, and the | |
2257 | * beginning of the next hole in @hole, and return 0. | |
2258 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2259 | * beginning of the next non-hole in @data, and return 0. | |
2260 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2261 | * If we can't find out, return a negative errno other than -ENXIO. | |
2262 | */ | |
2263 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2264 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2265 | { |
2266 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2267 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2268 | off_t offs; |
94282e71 | 2269 | |
d1f06fe6 MA |
2270 | /* |
2271 | * SEEK_DATA cases: | |
2272 | * D1. offs == start: start is in data | |
2273 | * D2. offs > start: start is in a hole, next data at offs | |
2274 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2275 | * or start is beyond EOF | |
2276 | * If the latter happens, the file has been truncated behind | |
2277 | * our back since we opened it. All bets are off then. | |
2278 | * Treating like a trailing hole is simplest. | |
2279 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2280 | */ | |
2281 | offs = lseek(s->fd, start, SEEK_DATA); | |
2282 | if (offs < 0) { | |
2283 | return -errno; /* D3 or D4 */ | |
2284 | } | |
a03083a0 JC |
2285 | |
2286 | if (offs < start) { | |
2287 | /* This is not a valid return by lseek(). We are safe to just return | |
2288 | * -EIO in this case, and we'll treat it like D4. */ | |
2289 | return -EIO; | |
2290 | } | |
d1f06fe6 MA |
2291 | |
2292 | if (offs > start) { | |
2293 | /* D2: in hole, next data at offs */ | |
2294 | *hole = start; | |
2295 | *data = offs; | |
2296 | return 0; | |
5500316d PB |
2297 | } |
2298 | ||
d1f06fe6 MA |
2299 | /* D1: in data, end not yet known */ |
2300 | ||
2301 | /* | |
2302 | * SEEK_HOLE cases: | |
2303 | * H1. offs == start: start is in a hole | |
2304 | * If this happens here, a hole has been dug behind our back | |
2305 | * since the previous lseek(). | |
2306 | * H2. offs > start: either start is in data, next hole at offs, | |
2307 | * or start is in trailing hole, EOF at offs | |
2308 | * Linux treats trailing holes like any other hole: offs == | |
2309 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2310 | * If that happens here, a hole has been dug behind our back | |
2311 | * since the previous lseek(). | |
2312 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2313 | * If this happens, the file has been truncated behind our | |
2314 | * back since we opened it. Treat it like a trailing hole. | |
2315 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2316 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2317 | */ | |
2318 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2319 | if (offs < 0) { | |
2320 | return -errno; /* D1 and (H3 or H4) */ | |
2321 | } | |
a03083a0 JC |
2322 | |
2323 | if (offs < start) { | |
2324 | /* This is not a valid return by lseek(). We are safe to just return | |
2325 | * -EIO in this case, and we'll treat it like H4. */ | |
2326 | return -EIO; | |
2327 | } | |
d1f06fe6 MA |
2328 | |
2329 | if (offs > start) { | |
2330 | /* | |
2331 | * D1 and H2: either in data, next hole at offs, or it was in | |
2332 | * data but is now in a trailing hole. In the latter case, | |
2333 | * all bets are off. Treating it as if it there was data all | |
2334 | * the way to EOF is safe, so simply do that. | |
2335 | */ | |
4f11aa8a | 2336 | *data = start; |
d1f06fe6 MA |
2337 | *hole = offs; |
2338 | return 0; | |
5500316d | 2339 | } |
4f11aa8a | 2340 | |
d1f06fe6 MA |
2341 | /* D1 and H1 */ |
2342 | return -EBUSY; | |
5500316d | 2343 | #else |
4f11aa8a | 2344 | return -ENOTSUP; |
5500316d | 2345 | #endif |
4f11aa8a HR |
2346 | } |
2347 | ||
2348 | /* | |
a290f085 | 2349 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2350 | * |
a290f085 | 2351 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2352 | * |
a290f085 EB |
2353 | * 'pnum' is set to the number of bytes (including and immediately following |
2354 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2355 | * allocated/unallocated state. |
2356 | * | |
a290f085 | 2357 | * 'bytes' is the max value 'pnum' should be set to. |
4f11aa8a | 2358 | */ |
a290f085 EB |
2359 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2360 | bool want_zero, | |
2361 | int64_t offset, | |
2362 | int64_t bytes, int64_t *pnum, | |
2363 | int64_t *map, | |
2364 | BlockDriverState **file) | |
2365 | { | |
2366 | off_t data = 0, hole = 0; | |
d7f62751 | 2367 | int ret; |
4f11aa8a HR |
2368 | |
2369 | ret = fd_open(bs); | |
2370 | if (ret < 0) { | |
2371 | return ret; | |
2372 | } | |
2373 | ||
a290f085 EB |
2374 | if (!want_zero) { |
2375 | *pnum = bytes; | |
2376 | *map = offset; | |
2377 | *file = bs; | |
2378 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2379 | } |
4f11aa8a | 2380 | |
a290f085 | 2381 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2382 | if (ret == -ENXIO) { |
2383 | /* Trailing hole */ | |
a290f085 | 2384 | *pnum = bytes; |
d1f06fe6 MA |
2385 | ret = BDRV_BLOCK_ZERO; |
2386 | } else if (ret < 0) { | |
2387 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2388 | *pnum = bytes; |
d1f06fe6 | 2389 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2390 | } else if (data == offset) { |
2391 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2392 | * possibly including a partial sector at EOF. */ |
a290f085 | 2393 | *pnum = MIN(bytes, hole - offset); |
d1f06fe6 | 2394 | ret = BDRV_BLOCK_DATA; |
5500316d | 2395 | } else { |
a290f085 EB |
2396 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2397 | assert(hole == offset); | |
2398 | *pnum = MIN(bytes, data - offset); | |
d1f06fe6 | 2399 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2400 | } |
a290f085 | 2401 | *map = offset; |
02650acb | 2402 | *file = bs; |
a290f085 | 2403 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2404 | } |
2405 | ||
31be8a2a SH |
2406 | #if defined(__linux__) |
2407 | /* Verify that the file is not in the page cache */ | |
2408 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2409 | { | |
2410 | const size_t window_size = 128 * 1024 * 1024; | |
2411 | BDRVRawState *s = bs->opaque; | |
2412 | void *window = NULL; | |
2413 | size_t length = 0; | |
2414 | unsigned char *vec; | |
2415 | size_t page_size; | |
2416 | off_t offset; | |
2417 | off_t end; | |
2418 | ||
2419 | /* mincore(2) page status information requires 1 byte per page */ | |
2420 | page_size = sysconf(_SC_PAGESIZE); | |
2421 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2422 | ||
2423 | end = raw_getlength(bs); | |
2424 | ||
2425 | for (offset = 0; offset < end; offset += window_size) { | |
2426 | void *new_window; | |
2427 | size_t new_length; | |
2428 | size_t vec_end; | |
2429 | size_t i; | |
2430 | int ret; | |
2431 | ||
2432 | /* Unmap previous window if size has changed */ | |
2433 | new_length = MIN(end - offset, window_size); | |
2434 | if (new_length != length) { | |
2435 | munmap(window, length); | |
2436 | window = NULL; | |
2437 | length = 0; | |
2438 | } | |
2439 | ||
2440 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2441 | s->fd, offset); | |
2442 | if (new_window == MAP_FAILED) { | |
2443 | error_setg_errno(errp, errno, "mmap failed"); | |
2444 | break; | |
2445 | } | |
2446 | ||
2447 | window = new_window; | |
2448 | length = new_length; | |
2449 | ||
2450 | ret = mincore(window, length, vec); | |
2451 | if (ret < 0) { | |
2452 | error_setg_errno(errp, errno, "mincore failed"); | |
2453 | break; | |
2454 | } | |
2455 | ||
2456 | vec_end = DIV_ROUND_UP(length, page_size); | |
2457 | for (i = 0; i < vec_end; i++) { | |
2458 | if (vec[i] & 0x1) { | |
2459 | error_setg(errp, "page cache still in use!"); | |
2460 | break; | |
2461 | } | |
2462 | } | |
2463 | } | |
2464 | ||
2465 | if (window) { | |
2466 | munmap(window, length); | |
2467 | } | |
2468 | ||
2469 | g_free(vec); | |
2470 | } | |
2471 | #endif /* __linux__ */ | |
2472 | ||
dd577a26 SH |
2473 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2474 | Error **errp) | |
2475 | { | |
2476 | BDRVRawState *s = bs->opaque; | |
2477 | int ret; | |
2478 | ||
2479 | ret = fd_open(bs); | |
2480 | if (ret < 0) { | |
2481 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2482 | return; | |
2483 | } | |
2484 | ||
2485 | if (s->open_flags & O_DIRECT) { | |
2486 | return; /* No host kernel page cache */ | |
2487 | } | |
2488 | ||
2489 | #if defined(__linux__) | |
2490 | /* This sets the scene for the next syscall... */ | |
2491 | ret = bdrv_co_flush(bs); | |
2492 | if (ret < 0) { | |
2493 | error_setg_errno(errp, -ret, "flush failed"); | |
2494 | return; | |
2495 | } | |
2496 | ||
2497 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2498 | * process. These limitations are okay because we just fsynced the file, | |
2499 | * we don't use mmap, and the file should not be in use by other processes. | |
2500 | */ | |
2501 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2502 | if (ret != 0) { /* the return value is a positive errno */ | |
2503 | error_setg_errno(errp, ret, "fadvise failed"); | |
2504 | return; | |
2505 | } | |
31be8a2a SH |
2506 | |
2507 | if (s->check_cache_dropped) { | |
2508 | check_cache_dropped(bs, errp); | |
2509 | } | |
dd577a26 SH |
2510 | #else /* __linux__ */ |
2511 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2512 | * unsupported on other host operating systems. Cache consistency issues | |
2513 | * may occur but no error is reported here, partly because that's the | |
2514 | * historical behavior and partly because it's hard to differentiate valid | |
2515 | * configurations that should not cause errors. | |
2516 | */ | |
2517 | #endif /* !__linux__ */ | |
2518 | } | |
2519 | ||
33d70fb6 KW |
2520 | static coroutine_fn int |
2521 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
dce512de | 2522 | { |
dce512de CH |
2523 | BDRVRawState *s = bs->opaque; |
2524 | ||
33d70fb6 | 2525 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, QEMU_AIO_DISCARD); |
dce512de | 2526 | } |
0e7e1989 | 2527 | |
2ffa76c2 EB |
2528 | static int coroutine_fn raw_co_pwrite_zeroes( |
2529 | BlockDriverState *bs, int64_t offset, | |
f5a5ca79 | 2530 | int bytes, BdrvRequestFlags flags) |
260a82e5 PB |
2531 | { |
2532 | BDRVRawState *s = bs->opaque; | |
2533 | ||
2534 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
f5a5ca79 | 2535 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 PB |
2536 | QEMU_AIO_WRITE_ZEROES); |
2537 | } else if (s->discard_zeroes) { | |
f5a5ca79 | 2538 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 | 2539 | QEMU_AIO_DISCARD); |
260a82e5 | 2540 | } |
97a2ae34 | 2541 | return -ENOTSUP; |
260a82e5 PB |
2542 | } |
2543 | ||
2544 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
2545 | { | |
2546 | BDRVRawState *s = bs->opaque; | |
2547 | ||
2548 | bdi->unallocated_blocks_are_zero = s->discard_zeroes; | |
260a82e5 PB |
2549 | return 0; |
2550 | } | |
2551 | ||
6f482f74 CL |
2552 | static QemuOptsList raw_create_opts = { |
2553 | .name = "raw-create-opts", | |
2554 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
2555 | .desc = { | |
2556 | { | |
2557 | .name = BLOCK_OPT_SIZE, | |
2558 | .type = QEMU_OPT_SIZE, | |
2559 | .help = "Virtual disk size" | |
2560 | }, | |
4ab15590 CL |
2561 | { |
2562 | .name = BLOCK_OPT_NOCOW, | |
2563 | .type = QEMU_OPT_BOOL, | |
2564 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
2565 | }, | |
06247428 HT |
2566 | { |
2567 | .name = BLOCK_OPT_PREALLOC, | |
2568 | .type = QEMU_OPT_STRING, | |
2569 | .help = "Preallocation mode (allowed values: off, falloc, full)" | |
2570 | }, | |
6f482f74 CL |
2571 | { /* end of list */ } |
2572 | } | |
0e7e1989 KW |
2573 | }; |
2574 | ||
244a5668 FZ |
2575 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
2576 | Error **errp) | |
2577 | { | |
2578 | return raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
2579 | } | |
2580 | ||
2581 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
2582 | { | |
2583 | BDRVRawState *s = bs->opaque; | |
2584 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); | |
2585 | s->perm = perm; | |
2586 | s->shared_perm = shared; | |
2587 | } | |
2588 | ||
2589 | static void raw_abort_perm_update(BlockDriverState *bs) | |
2590 | { | |
2591 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
2592 | } | |
2593 | ||
67b51fb9 VSO |
2594 | static int coroutine_fn raw_co_copy_range_from( |
2595 | BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, | |
2596 | BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, | |
2597 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) | |
1efad060 | 2598 | { |
67b51fb9 VSO |
2599 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
2600 | read_flags, write_flags); | |
1efad060 FZ |
2601 | } |
2602 | ||
2603 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 VSO |
2604 | BdrvChild *src, |
2605 | uint64_t src_offset, | |
2606 | BdrvChild *dst, | |
2607 | uint64_t dst_offset, | |
2608 | uint64_t bytes, | |
2609 | BdrvRequestFlags read_flags, | |
2610 | BdrvRequestFlags write_flags) | |
1efad060 FZ |
2611 | { |
2612 | BDRVRawState *s = bs->opaque; | |
2613 | BDRVRawState *src_s; | |
2614 | ||
2615 | assert(dst->bs == bs); | |
2616 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
2617 | return -ENOTSUP; | |
2618 | } | |
2619 | ||
2620 | src_s = src->bs->opaque; | |
9f850f67 | 2621 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
2622 | return -EIO; |
2623 | } | |
2624 | return paio_submit_co_full(bs, src_s->fd, src_offset, s->fd, dst_offset, | |
2625 | NULL, bytes, QEMU_AIO_COPY_RANGE); | |
2626 | } | |
2627 | ||
5f535a94 | 2628 | BlockDriver bdrv_file = { |
84a12e66 CH |
2629 | .format_name = "file", |
2630 | .protocol_name = "file", | |
856ae5c3 | 2631 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2632 | .bdrv_needs_filename = true, |
856ae5c3 | 2633 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 2634 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 2635 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
2636 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2637 | .bdrv_reopen_commit = raw_reopen_commit, | |
2638 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 2639 | .bdrv_close = raw_close, |
927f11e1 | 2640 | .bdrv_co_create = raw_co_create, |
efc75e2a | 2641 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 2642 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 2643 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 2644 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 2645 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
3b46e624 | 2646 | |
9d52aa3c KW |
2647 | .bdrv_co_preadv = raw_co_preadv, |
2648 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
2649 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
2650 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
2651 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
2652 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 2653 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2654 | .bdrv_io_plug = raw_aio_plug, |
2655 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 2656 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 2657 | |
061ca8a3 | 2658 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 2659 | .bdrv_getlength = raw_getlength, |
260a82e5 | 2660 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
2661 | .bdrv_get_allocated_file_size |
2662 | = raw_get_allocated_file_size, | |
244a5668 FZ |
2663 | .bdrv_check_perm = raw_check_perm, |
2664 | .bdrv_set_perm = raw_set_perm, | |
2665 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 2666 | .create_opts = &raw_create_opts, |
83f64091 FB |
2667 | }; |
2668 | ||
19cb3738 FB |
2669 | /***********************************************/ |
2670 | /* host device */ | |
2671 | ||
83affaa6 | 2672 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
2673 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2674 | CFIndex maxPathSize, int flags); | |
d0855f12 | 2675 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 2676 | { |
d0855f12 | 2677 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
2678 | mach_port_t masterPort; |
2679 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
2680 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
2681 | char *mediaType = NULL; | |
19cb3738 FB |
2682 | |
2683 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
2684 | if ( KERN_SUCCESS != kernResult ) { | |
2685 | printf( "IOMasterPort returned %d\n", kernResult ); | |
2686 | } | |
3b46e624 | 2687 | |
d0855f12 JA |
2688 | int index; |
2689 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
2690 | classesToMatch = IOServiceMatching(matching_array[index]); | |
2691 | if (classesToMatch == NULL) { | |
2692 | error_report("IOServiceMatching returned NULL for %s", | |
2693 | matching_array[index]); | |
2694 | continue; | |
2695 | } | |
2696 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
2697 | kCFBooleanTrue); | |
2698 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
2699 | mediaIterator); | |
2700 | if (kernResult != KERN_SUCCESS) { | |
2701 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
2702 | kernResult); | |
2703 | continue; | |
2704 | } | |
3b46e624 | 2705 | |
d0855f12 JA |
2706 | /* If a match was found, leave the loop */ |
2707 | if (*mediaIterator != 0) { | |
2708 | DPRINTF("Matching using %s\n", matching_array[index]); | |
2709 | mediaType = g_strdup(matching_array[index]); | |
2710 | break; | |
2711 | } | |
2712 | } | |
2713 | return mediaType; | |
19cb3738 FB |
2714 | } |
2715 | ||
98caa5bc JA |
2716 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2717 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
2718 | { |
2719 | io_object_t nextMedia; | |
2720 | kern_return_t kernResult = KERN_FAILURE; | |
2721 | *bsdPath = '\0'; | |
2722 | nextMedia = IOIteratorNext( mediaIterator ); | |
2723 | if ( nextMedia ) | |
2724 | { | |
2725 | CFTypeRef bsdPathAsCFString; | |
2726 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
2727 | if ( bsdPathAsCFString ) { | |
2728 | size_t devPathLength; | |
2729 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
2730 | if (flags & BDRV_O_NOCACHE) { |
2731 | strcat(bsdPath, "r"); | |
2732 | } | |
19cb3738 FB |
2733 | devPathLength = strlen( bsdPath ); |
2734 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
2735 | kernResult = KERN_SUCCESS; | |
2736 | } | |
2737 | CFRelease( bsdPathAsCFString ); | |
2738 | } | |
2739 | IOObjectRelease( nextMedia ); | |
2740 | } | |
3b46e624 | 2741 | |
19cb3738 FB |
2742 | return kernResult; |
2743 | } | |
2744 | ||
d0855f12 JA |
2745 | /* Sets up a real cdrom for use in QEMU */ |
2746 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
2747 | { | |
2748 | int index, num_of_test_partitions = 2, fd; | |
2749 | char test_partition[MAXPATHLEN]; | |
2750 | bool partition_found = false; | |
2751 | ||
2752 | /* look for a working partition */ | |
2753 | for (index = 0; index < num_of_test_partitions; index++) { | |
2754 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
2755 | index); | |
2756 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); | |
2757 | if (fd >= 0) { | |
2758 | partition_found = true; | |
2759 | qemu_close(fd); | |
2760 | break; | |
2761 | } | |
2762 | } | |
2763 | ||
2764 | /* if a working partition on the device was not found */ | |
2765 | if (partition_found == false) { | |
2766 | error_setg(errp, "Failed to find a working partition on disc"); | |
2767 | } else { | |
2768 | DPRINTF("Using %s as optical disc\n", test_partition); | |
2769 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); | |
2770 | } | |
2771 | return partition_found; | |
2772 | } | |
2773 | ||
2774 | /* Prints directions on mounting and unmounting a device */ | |
2775 | static void print_unmounting_directions(const char *file_name) | |
2776 | { | |
2777 | error_report("If device %s is mounted on the desktop, unmount" | |
2778 | " it first before using it in QEMU", file_name); | |
2779 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
2780 | file_name); | |
2781 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
2782 | } | |
2783 | ||
2784 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 2785 | |
508c7cb3 CH |
2786 | static int hdev_probe_device(const char *filename) |
2787 | { | |
2788 | struct stat st; | |
2789 | ||
2790 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
2791 | if (strstart(filename, "/dev/cdrom", NULL)) | |
2792 | return 50; | |
2793 | ||
2794 | if (stat(filename, &st) >= 0 && | |
2795 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
2796 | return 100; | |
2797 | } | |
2798 | ||
2799 | return 0; | |
2800 | } | |
2801 | ||
da888d37 SH |
2802 | static int check_hdev_writable(BDRVRawState *s) |
2803 | { | |
2804 | #if defined(BLKROGET) | |
2805 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
2806 | * This is independent of device node permissions and therefore open(2) | |
2807 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
2808 | * | |
2809 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
2810 | * check for read-only block devices so that Linux block devices behave | |
2811 | * properly. | |
2812 | */ | |
2813 | struct stat st; | |
2814 | int readonly = 0; | |
2815 | ||
2816 | if (fstat(s->fd, &st)) { | |
2817 | return -errno; | |
2818 | } | |
2819 | ||
2820 | if (!S_ISBLK(st.st_mode)) { | |
2821 | return 0; | |
2822 | } | |
2823 | ||
2824 | if (ioctl(s->fd, BLKROGET, &readonly) < 0) { | |
2825 | return -errno; | |
2826 | } | |
2827 | ||
2828 | if (readonly) { | |
2829 | return -EACCES; | |
2830 | } | |
2831 | #endif /* defined(BLKROGET) */ | |
2832 | return 0; | |
2833 | } | |
2834 | ||
7af803d4 HR |
2835 | static void hdev_parse_filename(const char *filename, QDict *options, |
2836 | Error **errp) | |
2837 | { | |
03c320d8 | 2838 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
2839 | } |
2840 | ||
3307ed7b DA |
2841 | static bool hdev_is_sg(BlockDriverState *bs) |
2842 | { | |
2843 | ||
2844 | #if defined(__linux__) | |
2845 | ||
0d4377b3 | 2846 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
2847 | struct stat st; |
2848 | struct sg_scsi_id scsiid; | |
2849 | int sg_version; | |
0d4377b3 KW |
2850 | int ret; |
2851 | ||
2852 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
2853 | return false; | |
2854 | } | |
3307ed7b | 2855 | |
0d4377b3 KW |
2856 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
2857 | if (ret < 0) { | |
2858 | return false; | |
2859 | } | |
2860 | ||
2861 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
2862 | if (ret >= 0) { | |
3307ed7b DA |
2863 | DPRINTF("SG device found: type=%d, version=%d\n", |
2864 | scsiid.scsi_type, sg_version); | |
2865 | return true; | |
2866 | } | |
2867 | ||
2868 | #endif | |
2869 | ||
2870 | return false; | |
2871 | } | |
2872 | ||
015a1036 HR |
2873 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
2874 | Error **errp) | |
19cb3738 FB |
2875 | { |
2876 | BDRVRawState *s = bs->opaque; | |
e428e439 | 2877 | Error *local_err = NULL; |
da888d37 | 2878 | int ret; |
a76bab49 | 2879 | |
83affaa6 | 2880 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
2881 | /* |
2882 | * Caution: while qdict_get_str() is fine, getting non-string types | |
2883 | * would require more care. When @options come from -blockdev or | |
2884 | * blockdev_add, its members are typed according to the QAPI | |
2885 | * schema, but when they come from -drive, they're all QString. | |
2886 | */ | |
3307ed7b | 2887 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
2888 | char bsd_path[MAXPATHLEN] = ""; |
2889 | bool error_occurred = false; | |
2890 | ||
2891 | /* If using a real cdrom */ | |
2892 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
2893 | char *mediaType = NULL; | |
2894 | kern_return_t ret_val; | |
2895 | io_iterator_t mediaIterator = 0; | |
2896 | ||
2897 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
2898 | if (mediaType == NULL) { | |
2899 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
2900 | " drive"); | |
2901 | error_occurred = true; | |
2902 | goto hdev_open_Mac_error; | |
2903 | } | |
3307ed7b | 2904 | |
d0855f12 JA |
2905 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
2906 | if (ret_val != KERN_SUCCESS) { | |
2907 | error_setg(errp, "Could not get BSD path for optical drive"); | |
2908 | error_occurred = true; | |
2909 | goto hdev_open_Mac_error; | |
2910 | } | |
2911 | ||
2912 | /* If a real optical drive was not found */ | |
2913 | if (bsd_path[0] == '\0') { | |
2914 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
2915 | error_occurred = true; | |
2916 | goto hdev_open_Mac_error; | |
2917 | } | |
2918 | ||
2919 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
2920 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
2921 | setup_cdrom(bsd_path, errp) == false) { | |
2922 | print_unmounting_directions(bsd_path); | |
2923 | error_occurred = true; | |
2924 | goto hdev_open_Mac_error; | |
19cb3738 | 2925 | } |
3b46e624 | 2926 | |
46f5ac20 | 2927 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
2928 | |
2929 | hdev_open_Mac_error: | |
2930 | g_free(mediaType); | |
2931 | if (mediaIterator) { | |
2932 | IOObjectRelease(mediaIterator); | |
2933 | } | |
2934 | if (error_occurred) { | |
2935 | return -ENOENT; | |
2936 | } | |
19cb3738 | 2937 | } |
d0855f12 | 2938 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
2939 | |
2940 | s->type = FTYPE_FILE; | |
90babde0 | 2941 | |
e428e439 | 2942 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
da888d37 | 2943 | if (ret < 0) { |
621ff94d | 2944 | error_propagate(errp, local_err); |
d0855f12 JA |
2945 | #if defined(__APPLE__) && defined(__MACH__) |
2946 | if (*bsd_path) { | |
2947 | filename = bsd_path; | |
2948 | } | |
2949 | /* if a physical device experienced an error while being opened */ | |
2950 | if (strncmp(filename, "/dev/", 5) == 0) { | |
2951 | print_unmounting_directions(filename); | |
2952 | } | |
2953 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
2954 | return ret; |
2955 | } | |
2956 | ||
3307ed7b DA |
2957 | /* Since this does ioctl the device must be already opened */ |
2958 | bs->sg = hdev_is_sg(bs); | |
2959 | ||
da888d37 SH |
2960 | if (flags & BDRV_O_RDWR) { |
2961 | ret = check_hdev_writable(s); | |
2962 | if (ret < 0) { | |
2963 | raw_close(bs); | |
e428e439 | 2964 | error_setg_errno(errp, -ret, "The device is not writable"); |
da888d37 SH |
2965 | return ret; |
2966 | } | |
2967 | } | |
2968 | ||
2969 | return ret; | |
19cb3738 FB |
2970 | } |
2971 | ||
03ff3ca3 | 2972 | #if defined(__linux__) |
221f715d | 2973 | |
7c84b1b8 | 2974 | static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs, |
221f715d | 2975 | unsigned long int req, void *buf, |
097310b5 | 2976 | BlockCompletionFunc *cb, void *opaque) |
221f715d | 2977 | { |
f141eafe | 2978 | BDRVRawState *s = bs->opaque; |
c208e8c2 | 2979 | RawPosixAIOData *acb; |
c4d9d196 | 2980 | ThreadPool *pool; |
221f715d | 2981 | |
f141eafe AL |
2982 | if (fd_open(bs) < 0) |
2983 | return NULL; | |
c208e8c2 | 2984 | |
7c9e5276 PB |
2985 | if (req == SG_IO && s->pr_mgr) { |
2986 | struct sg_io_hdr *io_hdr = buf; | |
2987 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
2988 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
2989 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2990 | s->fd, io_hdr, cb, opaque); | |
2991 | } | |
2992 | } | |
2993 | ||
c84b3192 | 2994 | acb = g_new(RawPosixAIOData, 1); |
c208e8c2 PB |
2995 | acb->bs = bs; |
2996 | acb->aio_type = QEMU_AIO_IOCTL; | |
2997 | acb->aio_fildes = s->fd; | |
2998 | acb->aio_offset = 0; | |
2999 | acb->aio_ioctl_buf = buf; | |
3000 | acb->aio_ioctl_cmd = req; | |
c4d9d196 SH |
3001 | pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); |
3002 | return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque); | |
221f715d | 3003 | } |
f709623b | 3004 | #endif /* linux */ |
221f715d | 3005 | |
9f23011a BS |
3006 | static int fd_open(BlockDriverState *bs) |
3007 | { | |
3008 | BDRVRawState *s = bs->opaque; | |
3009 | ||
3010 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
3011 | if (s->fd >= 0) | |
3012 | return 0; | |
3013 | return -EIO; | |
3014 | } | |
04eeb8b6 | 3015 | |
33d70fb6 KW |
3016 | static coroutine_fn int |
3017 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
c36dd8a0 AF |
3018 | { |
3019 | BDRVRawState *s = bs->opaque; | |
33d70fb6 | 3020 | int ret; |
c36dd8a0 | 3021 | |
33d70fb6 KW |
3022 | ret = fd_open(bs); |
3023 | if (ret < 0) { | |
3024 | return ret; | |
c36dd8a0 | 3025 | } |
33d70fb6 KW |
3026 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
3027 | QEMU_AIO_DISCARD | QEMU_AIO_BLKDEV); | |
c36dd8a0 AF |
3028 | } |
3029 | ||
2ffa76c2 | 3030 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3031 | int64_t offset, int bytes, BdrvRequestFlags flags) |
d0b4503e PB |
3032 | { |
3033 | BDRVRawState *s = bs->opaque; | |
3034 | int rc; | |
3035 | ||
3036 | rc = fd_open(bs); | |
3037 | if (rc < 0) { | |
3038 | return rc; | |
3039 | } | |
3040 | if (!(flags & BDRV_REQ_MAY_UNMAP)) { | |
f5a5ca79 | 3041 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 PB |
3042 | QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV); |
3043 | } else if (s->discard_zeroes) { | |
f5a5ca79 | 3044 | return paio_submit_co(bs, s->fd, offset, NULL, bytes, |
97a2ae34 | 3045 | QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV); |
d0b4503e | 3046 | } |
97a2ae34 | 3047 | return -ENOTSUP; |
d0b4503e PB |
3048 | } |
3049 | ||
efc75e2a SH |
3050 | static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, |
3051 | Error **errp) | |
93c65b47 AL |
3052 | { |
3053 | int fd; | |
3054 | int ret = 0; | |
3055 | struct stat stat_buf; | |
0e7e1989 | 3056 | int64_t total_size = 0; |
cc28c6aa HR |
3057 | bool has_prefix; |
3058 | ||
f709623b HR |
3059 | /* This function is used by both protocol block drivers and therefore either |
3060 | * of these prefixes may be given. | |
cc28c6aa HR |
3061 | * The return value has to be stored somewhere, otherwise this is an error |
3062 | * due to -Werror=unused-value. */ | |
3063 | has_prefix = | |
3064 | strstart(filename, "host_device:", &filename) || | |
f709623b | 3065 | strstart(filename, "host_cdrom:" , &filename); |
cc28c6aa HR |
3066 | |
3067 | (void)has_prefix; | |
93c65b47 | 3068 | |
bdd03cdf HR |
3069 | ret = raw_normalize_devicepath(&filename); |
3070 | if (ret < 0) { | |
3071 | error_setg_errno(errp, -ret, "Could not normalize device path"); | |
3072 | return ret; | |
3073 | } | |
3074 | ||
0e7e1989 | 3075 | /* Read out options */ |
180e9526 HT |
3076 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
3077 | BDRV_SECTOR_SIZE); | |
93c65b47 | 3078 | |
6165f4d8 | 3079 | fd = qemu_open(filename, O_WRONLY | O_BINARY); |
e428e439 HR |
3080 | if (fd < 0) { |
3081 | ret = -errno; | |
3082 | error_setg_errno(errp, -ret, "Could not open device"); | |
3083 | return ret; | |
3084 | } | |
93c65b47 | 3085 | |
e428e439 | 3086 | if (fstat(fd, &stat_buf) < 0) { |
57e69b7d | 3087 | ret = -errno; |
e428e439 HR |
3088 | error_setg_errno(errp, -ret, "Could not stat device"); |
3089 | } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { | |
3090 | error_setg(errp, | |
3091 | "The given file is neither a block nor a character device"); | |
57e69b7d | 3092 | ret = -ENODEV; |
180e9526 | 3093 | } else if (lseek(fd, 0, SEEK_END) < total_size) { |
e428e439 | 3094 | error_setg(errp, "Device is too small"); |
93c65b47 | 3095 | ret = -ENOSPC; |
e428e439 | 3096 | } |
93c65b47 | 3097 | |
97ec9117 FZ |
3098 | if (!ret && total_size) { |
3099 | uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; | |
3100 | int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); | |
3101 | if (lseek(fd, 0, SEEK_SET) == -1) { | |
3102 | ret = -errno; | |
3103 | } else { | |
3104 | ret = qemu_write_full(fd, buf, zero_size); | |
3105 | ret = ret == zero_size ? 0 : -errno; | |
3106 | } | |
3107 | } | |
2e1e79da | 3108 | qemu_close(fd); |
93c65b47 AL |
3109 | return ret; |
3110 | } | |
3111 | ||
5efa9d5a | 3112 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3113 | .format_name = "host_device", |
84a12e66 | 3114 | .protocol_name = "host_device", |
0b4ce02e | 3115 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3116 | .bdrv_needs_filename = true, |
0b4ce02e | 3117 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3118 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3119 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3120 | .bdrv_close = raw_close, |
1bc6b705 JC |
3121 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3122 | .bdrv_reopen_commit = raw_reopen_commit, | |
3123 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3124 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3125 | .create_opts = &raw_create_opts, |
dd577a26 | 3126 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3127 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3128 | |
9d52aa3c KW |
3129 | .bdrv_co_preadv = raw_co_preadv, |
3130 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3131 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3132 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3133 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3134 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3135 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3136 | .bdrv_io_plug = raw_aio_plug, |
3137 | .bdrv_io_unplug = raw_aio_unplug, | |
3c529d93 | 3138 | |
061ca8a3 | 3139 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3140 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3141 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3142 | .bdrv_get_allocated_file_size |
3143 | = raw_get_allocated_file_size, | |
244a5668 FZ |
3144 | .bdrv_check_perm = raw_check_perm, |
3145 | .bdrv_set_perm = raw_set_perm, | |
3146 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3147 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3148 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3149 | |
f3a5d3f8 | 3150 | /* generic scsi device */ |
63ec93db | 3151 | #ifdef __linux__ |
63ec93db CH |
3152 | .bdrv_aio_ioctl = hdev_aio_ioctl, |
3153 | #endif | |
f3a5d3f8 CH |
3154 | }; |
3155 | ||
18fa1c42 HR |
3156 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3157 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3158 | Error **errp) | |
3159 | { | |
03c320d8 | 3160 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3161 | } |
3162 | #endif | |
3163 | ||
3164 | #ifdef __linux__ | |
015a1036 HR |
3165 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3166 | Error **errp) | |
f3a5d3f8 CH |
3167 | { |
3168 | BDRVRawState *s = bs->opaque; | |
3169 | ||
f3a5d3f8 CH |
3170 | s->type = FTYPE_CD; |
3171 | ||
19a3da7f | 3172 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
9be38598 | 3173 | return raw_open_common(bs, options, flags, O_NONBLOCK, errp); |
f3a5d3f8 CH |
3174 | } |
3175 | ||
508c7cb3 CH |
3176 | static int cdrom_probe_device(const char *filename) |
3177 | { | |
3baf720e CR |
3178 | int fd, ret; |
3179 | int prio = 0; | |
343f8568 | 3180 | struct stat st; |
3baf720e | 3181 | |
6165f4d8 | 3182 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
3baf720e CR |
3183 | if (fd < 0) { |
3184 | goto out; | |
3185 | } | |
343f8568 JS |
3186 | ret = fstat(fd, &st); |
3187 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3188 | goto outc; | |
3189 | } | |
3baf720e CR |
3190 | |
3191 | /* Attempt to detect via a CDROM specific ioctl */ | |
3192 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3193 | if (ret >= 0) | |
3194 | prio = 100; | |
3195 | ||
343f8568 | 3196 | outc: |
2e1e79da | 3197 | qemu_close(fd); |
3baf720e CR |
3198 | out: |
3199 | return prio; | |
508c7cb3 CH |
3200 | } |
3201 | ||
e031f750 | 3202 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3203 | { |
3204 | BDRVRawState *s = bs->opaque; | |
3205 | int ret; | |
3206 | ||
3207 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3208 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3209 | } |
3210 | ||
f36f3949 | 3211 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3212 | { |
3213 | BDRVRawState *s = bs->opaque; | |
3214 | ||
3215 | if (eject_flag) { | |
3216 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3217 | perror("CDROMEJECT"); | |
3218 | } else { | |
3219 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3220 | perror("CDROMEJECT"); | |
3221 | } | |
f3a5d3f8 CH |
3222 | } |
3223 | ||
025e849a | 3224 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3225 | { |
3226 | BDRVRawState *s = bs->opaque; | |
3227 | ||
3228 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3229 | /* | |
3230 | * Note: an error can happen if the distribution automatically | |
3231 | * mounts the CD-ROM | |
3232 | */ | |
3233 | /* perror("CDROM_LOCKDOOR"); */ | |
3234 | } | |
f3a5d3f8 CH |
3235 | } |
3236 | ||
3237 | static BlockDriver bdrv_host_cdrom = { | |
3238 | .format_name = "host_cdrom", | |
84a12e66 | 3239 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3240 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3241 | .bdrv_needs_filename = true, |
508c7cb3 | 3242 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3243 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3244 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3245 | .bdrv_close = raw_close, |
1bc6b705 JC |
3246 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3247 | .bdrv_reopen_commit = raw_reopen_commit, | |
3248 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3249 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3250 | .create_opts = &raw_create_opts, |
dd577a26 | 3251 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3252 | |
9d52aa3c KW |
3253 | |
3254 | .bdrv_co_preadv = raw_co_preadv, | |
3255 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3256 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3257 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3258 | .bdrv_io_plug = raw_aio_plug, |
3259 | .bdrv_io_unplug = raw_aio_unplug, | |
f3a5d3f8 | 3260 | |
061ca8a3 | 3261 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3262 | .bdrv_getlength = raw_getlength, |
3263 | .has_variable_length = true, | |
4a1d5e1f FZ |
3264 | .bdrv_get_allocated_file_size |
3265 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3266 | |
3267 | /* removable device support */ | |
3268 | .bdrv_is_inserted = cdrom_is_inserted, | |
3269 | .bdrv_eject = cdrom_eject, | |
025e849a | 3270 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3271 | |
3272 | /* generic scsi device */ | |
63ec93db | 3273 | .bdrv_aio_ioctl = hdev_aio_ioctl, |
f3a5d3f8 CH |
3274 | }; |
3275 | #endif /* __linux__ */ | |
3276 | ||
a167ba50 | 3277 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3278 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3279 | Error **errp) | |
f3a5d3f8 CH |
3280 | { |
3281 | BDRVRawState *s = bs->opaque; | |
e428e439 | 3282 | Error *local_err = NULL; |
f3a5d3f8 CH |
3283 | int ret; |
3284 | ||
3285 | s->type = FTYPE_CD; | |
3286 | ||
e428e439 HR |
3287 | ret = raw_open_common(bs, options, flags, 0, &local_err); |
3288 | if (ret) { | |
621ff94d | 3289 | error_propagate(errp, local_err); |
f3a5d3f8 | 3290 | return ret; |
e428e439 | 3291 | } |
f3a5d3f8 | 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 | ||
508c7cb3 CH |
3298 | static int cdrom_probe_device(const char *filename) |
3299 | { | |
3300 | if (strstart(filename, "/dev/cd", NULL) || | |
3301 | strstart(filename, "/dev/acd", NULL)) | |
3302 | return 100; | |
3303 | return 0; | |
3304 | } | |
3305 | ||
f3a5d3f8 CH |
3306 | static int cdrom_reopen(BlockDriverState *bs) |
3307 | { | |
3308 | BDRVRawState *s = bs->opaque; | |
3309 | int fd; | |
3310 | ||
3311 | /* | |
3312 | * Force reread of possibly changed/newly loaded disc, | |
3313 | * FreeBSD seems to not notice sometimes... | |
3314 | */ | |
3315 | if (s->fd >= 0) | |
2e1e79da | 3316 | qemu_close(s->fd); |
6165f4d8 | 3317 | fd = qemu_open(bs->filename, s->open_flags, 0644); |
f3a5d3f8 CH |
3318 | if (fd < 0) { |
3319 | s->fd = -1; | |
3320 | return -EIO; | |
3321 | } | |
3322 | s->fd = fd; | |
3323 | ||
9b2260cb | 3324 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3325 | ioctl(s->fd, CDIOCALLOW); |
3326 | return 0; | |
3327 | } | |
3328 | ||
e031f750 | 3329 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3330 | { |
3331 | return raw_getlength(bs) > 0; | |
3332 | } | |
3333 | ||
f36f3949 | 3334 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3335 | { |
3336 | BDRVRawState *s = bs->opaque; | |
3337 | ||
3338 | if (s->fd < 0) | |
822e1cd1 | 3339 | return; |
f3a5d3f8 CH |
3340 | |
3341 | (void) ioctl(s->fd, CDIOCALLOW); | |
3342 | ||
3343 | if (eject_flag) { | |
3344 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3345 | perror("CDIOCEJECT"); | |
3346 | } else { | |
3347 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3348 | perror("CDIOCCLOSE"); | |
3349 | } | |
3350 | ||
822e1cd1 | 3351 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3352 | } |
3353 | ||
025e849a | 3354 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3355 | { |
3356 | BDRVRawState *s = bs->opaque; | |
3357 | ||
3358 | if (s->fd < 0) | |
7bf37fed | 3359 | return; |
f3a5d3f8 CH |
3360 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3361 | /* | |
3362 | * Note: an error can happen if the distribution automatically | |
3363 | * mounts the CD-ROM | |
3364 | */ | |
3365 | /* perror("CDROM_LOCKDOOR"); */ | |
3366 | } | |
f3a5d3f8 CH |
3367 | } |
3368 | ||
3369 | static BlockDriver bdrv_host_cdrom = { | |
3370 | .format_name = "host_cdrom", | |
84a12e66 | 3371 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3372 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3373 | .bdrv_needs_filename = true, |
508c7cb3 | 3374 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3375 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3376 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3377 | .bdrv_close = raw_close, |
1bc6b705 JC |
3378 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3379 | .bdrv_reopen_commit = raw_reopen_commit, | |
3380 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3381 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3382 | .create_opts = &raw_create_opts, |
f3a5d3f8 | 3383 | |
9d52aa3c KW |
3384 | .bdrv_co_preadv = raw_co_preadv, |
3385 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3386 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3387 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3388 | .bdrv_io_plug = raw_aio_plug, |
3389 | .bdrv_io_unplug = raw_aio_unplug, | |
f3a5d3f8 | 3390 | |
061ca8a3 | 3391 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3392 | .bdrv_getlength = raw_getlength, |
3393 | .has_variable_length = true, | |
4a1d5e1f FZ |
3394 | .bdrv_get_allocated_file_size |
3395 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3396 | |
19cb3738 | 3397 | /* removable device support */ |
f3a5d3f8 CH |
3398 | .bdrv_is_inserted = cdrom_is_inserted, |
3399 | .bdrv_eject = cdrom_eject, | |
025e849a | 3400 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3401 | }; |
f3a5d3f8 | 3402 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3403 | |
84a12e66 | 3404 | static void bdrv_file_init(void) |
5efa9d5a | 3405 | { |
508c7cb3 CH |
3406 | /* |
3407 | * Register all the drivers. Note that order is important, the driver | |
3408 | * registered last will get probed first. | |
3409 | */ | |
84a12e66 | 3410 | bdrv_register(&bdrv_file); |
5efa9d5a | 3411 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3412 | #ifdef __linux__ |
f3a5d3f8 CH |
3413 | bdrv_register(&bdrv_host_cdrom); |
3414 | #endif | |
a167ba50 | 3415 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3416 | bdrv_register(&bdrv_host_cdrom); |
3417 | #endif | |
5efa9d5a AL |
3418 | } |
3419 | ||
84a12e66 | 3420 | block_init(bdrv_file_init); |