]>
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" |
a8d25326 | 26 | #include "qemu-common.h" |
da34e65c | 27 | #include "qapi/error.h" |
f348b6d1 | 28 | #include "qemu/cutils.h" |
d49b6836 | 29 | #include "qemu/error-report.h" |
737e150e | 30 | #include "block/block_int.h" |
1de7afc9 | 31 | #include "qemu/module.h" |
922a01a0 | 32 | #include "qemu/option.h" |
de81a169 | 33 | #include "trace.h" |
737e150e | 34 | #include "block/thread-pool.h" |
1de7afc9 | 35 | #include "qemu/iov.h" |
0187f5c9 | 36 | #include "block/raw-aio.h" |
452fcdbc | 37 | #include "qapi/qmp/qdict.h" |
d49b6836 | 38 | #include "qapi/qmp/qstring.h" |
83f64091 | 39 | |
7c9e5276 PB |
40 | #include "scsi/pr-manager.h" |
41 | #include "scsi/constants.h" | |
42 | ||
83affaa6 | 43 | #if defined(__APPLE__) && (__MACH__) |
83f64091 FB |
44 | #include <paths.h> |
45 | #include <sys/param.h> | |
46 | #include <IOKit/IOKitLib.h> | |
47 | #include <IOKit/IOBSD.h> | |
48 | #include <IOKit/storage/IOMediaBSDClient.h> | |
49 | #include <IOKit/storage/IOMedia.h> | |
50 | #include <IOKit/storage/IOCDMedia.h> | |
51 | //#include <IOKit/storage/IOCDTypes.h> | |
d0855f12 | 52 | #include <IOKit/storage/IODVDMedia.h> |
83f64091 FB |
53 | #include <CoreFoundation/CoreFoundation.h> |
54 | #endif | |
55 | ||
56 | #ifdef __sun__ | |
2e9671da | 57 | #define _POSIX_PTHREAD_SEMANTICS 1 |
83f64091 FB |
58 | #include <sys/dkio.h> |
59 | #endif | |
19cb3738 FB |
60 | #ifdef __linux__ |
61 | #include <sys/ioctl.h> | |
05acda4d | 62 | #include <sys/param.h> |
1efad060 | 63 | #include <sys/syscall.h> |
19cb3738 FB |
64 | #include <linux/cdrom.h> |
65 | #include <linux/fd.h> | |
5500316d | 66 | #include <linux/fs.h> |
1a9335e4 | 67 | #include <linux/hdreg.h> |
3307ed7b | 68 | #include <scsi/sg.h> |
1a9335e4 ET |
69 | #ifdef __s390__ |
70 | #include <asm/dasd.h> | |
71 | #endif | |
4ab15590 CL |
72 | #ifndef FS_NOCOW_FL |
73 | #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ | |
74 | #endif | |
5500316d | 75 | #endif |
b953f075 | 76 | #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) |
3d4fa43e KK |
77 | #include <linux/falloc.h> |
78 | #endif | |
a167ba50 | 79 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
1cb6c3fd | 80 | #include <sys/disk.h> |
9f23011a | 81 | #include <sys/cdio.h> |
1cb6c3fd | 82 | #endif |
83f64091 | 83 | |
128ab2ff BS |
84 | #ifdef __OpenBSD__ |
85 | #include <sys/ioctl.h> | |
86 | #include <sys/disklabel.h> | |
87 | #include <sys/dkio.h> | |
88 | #endif | |
89 | ||
d1f6fd8d CE |
90 | #ifdef __NetBSD__ |
91 | #include <sys/ioctl.h> | |
92 | #include <sys/disklabel.h> | |
93 | #include <sys/dkio.h> | |
94 | #include <sys/disk.h> | |
95 | #endif | |
96 | ||
c5e97233 BS |
97 | #ifdef __DragonFly__ |
98 | #include <sys/ioctl.h> | |
99 | #include <sys/diskslice.h> | |
100 | #endif | |
101 | ||
dce512de CH |
102 | #ifdef CONFIG_XFS |
103 | #include <xfs/xfs.h> | |
104 | #endif | |
105 | ||
4f7d28d7 | 106 | #include "trace.h" |
8c05dbf9 | 107 | |
f6465578 AL |
108 | /* OS X does not have O_DSYNC */ |
109 | #ifndef O_DSYNC | |
1c27a8b3 | 110 | #ifdef O_SYNC |
7ab064d2 | 111 | #define O_DSYNC O_SYNC |
1c27a8b3 JA |
112 | #elif defined(O_FSYNC) |
113 | #define O_DSYNC O_FSYNC | |
114 | #endif | |
f6465578 AL |
115 | #endif |
116 | ||
9f7965c7 AL |
117 | /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ |
118 | #ifndef O_DIRECT | |
119 | #define O_DIRECT O_DSYNC | |
120 | #endif | |
121 | ||
19cb3738 FB |
122 | #define FTYPE_FILE 0 |
123 | #define FTYPE_CD 1 | |
83f64091 | 124 | |
581b9e29 CH |
125 | #define MAX_BLOCKSIZE 4096 |
126 | ||
244a5668 FZ |
127 | /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, |
128 | * leaving a few more bytes for its future use. */ | |
129 | #define RAW_LOCK_PERM_BASE 100 | |
130 | #define RAW_LOCK_SHARED_BASE 200 | |
131 | ||
19cb3738 FB |
132 | typedef struct BDRVRawState { |
133 | int fd; | |
244a5668 | 134 | bool use_lock; |
19cb3738 | 135 | int type; |
0e1d8f4c | 136 | int open_flags; |
c25f53b0 PB |
137 | size_t buf_align; |
138 | ||
244a5668 FZ |
139 | /* The current permissions. */ |
140 | uint64_t perm; | |
141 | uint64_t shared_perm; | |
142 | ||
2996ffad | 143 | /* The perms bits whose corresponding bytes are already locked in |
f2e3af29 | 144 | * s->fd. */ |
2996ffad FZ |
145 | uint64_t locked_perm; |
146 | uint64_t locked_shared_perm; | |
147 | ||
6ceabe6f | 148 | int perm_change_fd; |
094e3639 | 149 | int perm_change_flags; |
e0c9cf3a KW |
150 | BDRVReopenState *reopen_state; |
151 | ||
dce512de | 152 | #ifdef CONFIG_XFS |
260a82e5 | 153 | bool is_xfs:1; |
dce512de | 154 | #endif |
260a82e5 | 155 | bool has_discard:1; |
97a2ae34 | 156 | bool has_write_zeroes:1; |
260a82e5 | 157 | bool discard_zeroes:1; |
0a4279d9 | 158 | bool use_linux_aio:1; |
e5bcf967 | 159 | bool page_cache_inconsistent:1; |
d50d8222 | 160 | bool has_fallocate; |
3cad8307 | 161 | bool needs_alignment; |
f357fcd8 | 162 | bool drop_cache; |
31be8a2a | 163 | bool check_cache_dropped; |
7c9e5276 PB |
164 | |
165 | PRManager *pr_mgr; | |
19cb3738 FB |
166 | } BDRVRawState; |
167 | ||
eeb6b45d JC |
168 | typedef struct BDRVRawReopenState { |
169 | int fd; | |
170 | int open_flags; | |
f357fcd8 | 171 | bool drop_cache; |
31be8a2a | 172 | bool check_cache_dropped; |
eeb6b45d JC |
173 | } BDRVRawReopenState; |
174 | ||
19cb3738 | 175 | static int fd_open(BlockDriverState *bs); |
22afa7b5 | 176 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 177 | |
de81a169 PB |
178 | typedef struct RawPosixAIOData { |
179 | BlockDriverState *bs; | |
d57c44d0 | 180 | int aio_type; |
de81a169 | 181 | int aio_fildes; |
d57c44d0 | 182 | |
de81a169 | 183 | off_t aio_offset; |
d57c44d0 KW |
184 | uint64_t aio_nbytes; |
185 | ||
93f4e2ff | 186 | union { |
d57c44d0 KW |
187 | struct { |
188 | struct iovec *iov; | |
189 | int niov; | |
190 | } io; | |
191 | struct { | |
192 | uint64_t cmd; | |
193 | void *buf; | |
194 | } ioctl; | |
93f4e2ff KW |
195 | struct { |
196 | int aio_fd2; | |
197 | off_t aio_offset2; | |
d57c44d0 | 198 | } copy_range; |
93f4e2ff KW |
199 | struct { |
200 | PreallocMode prealloc; | |
201 | Error **errp; | |
d57c44d0 | 202 | } truncate; |
93f4e2ff | 203 | }; |
de81a169 PB |
204 | } RawPosixAIOData; |
205 | ||
a167ba50 | 206 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 207 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
208 | #endif |
209 | ||
1de1ae0a | 210 | #if defined(__NetBSD__) |
db0754df | 211 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
212 | { |
213 | static char namebuf[PATH_MAX]; | |
214 | const char *dp, *fname; | |
215 | struct stat sb; | |
216 | ||
217 | fname = *filename; | |
218 | dp = strrchr(fname, '/'); | |
219 | if (lstat(fname, &sb) < 0) { | |
f6fc1e30 | 220 | error_setg_file_open(errp, errno, fname); |
1de1ae0a CE |
221 | return -errno; |
222 | } | |
223 | ||
224 | if (!S_ISBLK(sb.st_mode)) { | |
225 | return 0; | |
226 | } | |
227 | ||
228 | if (dp == NULL) { | |
229 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
230 | } else { | |
231 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
232 | (int)(dp - fname), fname, dp + 1); | |
233 | } | |
1de1ae0a | 234 | *filename = namebuf; |
db0754df | 235 | warn_report("%s is a block device, using %s", fname, *filename); |
1de1ae0a CE |
236 | |
237 | return 0; | |
238 | } | |
239 | #else | |
db0754df | 240 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
241 | { |
242 | return 0; | |
243 | } | |
244 | #endif | |
245 | ||
8a4ed0d1 ET |
246 | /* |
247 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
248 | */ | |
249 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 250 | { |
c25f53b0 | 251 | unsigned int sector_size; |
8a4ed0d1 | 252 | bool success = false; |
700f9ce0 | 253 | int i; |
c25f53b0 | 254 | |
8a4ed0d1 | 255 | errno = ENOTSUP; |
700f9ce0 | 256 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 257 | #ifdef BLKSSZGET |
700f9ce0 | 258 | BLKSSZGET, |
c25f53b0 PB |
259 | #endif |
260 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 261 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
262 | #endif |
263 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 264 | DIOCGSECTORSIZE, |
c25f53b0 | 265 | #endif |
700f9ce0 PM |
266 | }; |
267 | ||
268 | /* Try a few ioctls to get the right size */ | |
269 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
270 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
271 | *sector_size_p = sector_size; | |
272 | success = true; | |
273 | } | |
274 | } | |
8a4ed0d1 ET |
275 | |
276 | return success ? 0 : -errno; | |
277 | } | |
278 | ||
1a9335e4 ET |
279 | /** |
280 | * Get physical block size of @fd. | |
281 | * On success, store it in @blk_size and return 0. | |
282 | * On failure, return -errno. | |
283 | */ | |
284 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
285 | { | |
286 | #ifdef BLKPBSZGET | |
287 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
288 | return -errno; | |
289 | } | |
290 | return 0; | |
291 | #else | |
292 | return -ENOTSUP; | |
293 | #endif | |
294 | } | |
295 | ||
22d182e8 SH |
296 | /* Check if read is allowed with given memory buffer and length. |
297 | * | |
298 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
299 | */ | |
300 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
301 | { | |
302 | ssize_t ret = pread(fd, buf, len, 0); | |
303 | ||
304 | if (ret >= 0) { | |
305 | return true; | |
306 | } | |
307 | ||
308 | #ifdef __linux__ | |
309 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
310 | * other errors (e.g. real I/O error), which could happen on a failed | |
311 | * drive, since we only care about probing alignment. | |
312 | */ | |
313 | if (errno != EINVAL) { | |
314 | return true; | |
315 | } | |
316 | #endif | |
317 | ||
318 | return false; | |
319 | } | |
320 | ||
8a4ed0d1 ET |
321 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
322 | { | |
323 | BDRVRawState *s = bs->opaque; | |
324 | char *buf; | |
459b4e66 | 325 | size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); |
a6b257a0 | 326 | size_t alignments[] = {1, 512, 1024, 2048, 4096}; |
8a4ed0d1 | 327 | |
b192af8a | 328 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 329 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 330 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 331 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
332 | s->buf_align = 1; |
333 | return; | |
334 | } | |
335 | ||
a5b8dd2c | 336 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
337 | s->buf_align = 0; |
338 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
339 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
340 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 341 | } |
c25f53b0 PB |
342 | #ifdef CONFIG_XFS |
343 | if (s->is_xfs) { | |
344 | struct dioattr da; | |
df26a350 | 345 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
a5b8dd2c | 346 | bs->bl.request_alignment = da.d_miniosz; |
c25f53b0 PB |
347 | /* The kernel returns wrong information for d_mem */ |
348 | /* s->buf_align = da.d_mem; */ | |
349 | } | |
350 | } | |
351 | #endif | |
352 | ||
a6b257a0 NS |
353 | /* |
354 | * If we could not get the sizes so far, we can only guess them. First try | |
355 | * to detect request alignment, since it is more likely to succeed. Then | |
356 | * try to detect buf_align, which cannot be detected in some cases (e.g. | |
357 | * Gluster). If buf_align cannot be detected, we fallback to the value of | |
358 | * request_alignment. | |
359 | */ | |
360 | ||
361 | if (!bs->bl.request_alignment) { | |
362 | int i; | |
c25f53b0 | 363 | size_t align; |
a6b257a0 NS |
364 | buf = qemu_memalign(max_align, max_align); |
365 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
366 | align = alignments[i]; | |
367 | if (raw_is_io_aligned(fd, buf, align)) { | |
368 | /* Fallback to safe value. */ | |
369 | bs->bl.request_alignment = (align != 1) ? align : max_align; | |
c25f53b0 PB |
370 | break; |
371 | } | |
372 | } | |
373 | qemu_vfree(buf); | |
374 | } | |
375 | ||
a6b257a0 NS |
376 | if (!s->buf_align) { |
377 | int i; | |
c25f53b0 | 378 | size_t align; |
a6b257a0 NS |
379 | buf = qemu_memalign(max_align, 2 * max_align); |
380 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
381 | align = alignments[i]; | |
382 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
236094c7 | 383 | /* Fallback to request_alignment. */ |
a6b257a0 | 384 | s->buf_align = (align != 1) ? align : bs->bl.request_alignment; |
c25f53b0 PB |
385 | break; |
386 | } | |
387 | } | |
388 | qemu_vfree(buf); | |
389 | } | |
df26a350 | 390 | |
a5b8dd2c | 391 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
392 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
393 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 394 | } |
c25f53b0 PB |
395 | } |
396 | ||
23dece19 | 397 | static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) |
6a8dc042 | 398 | { |
23dece19 | 399 | bool read_write = false; |
6a8dc042 JC |
400 | assert(open_flags != NULL); |
401 | ||
402 | *open_flags |= O_BINARY; | |
403 | *open_flags &= ~O_ACCMODE; | |
23dece19 KW |
404 | |
405 | if (bdrv_flags & BDRV_O_AUTO_RDONLY) { | |
406 | read_write = has_writers; | |
407 | } else if (bdrv_flags & BDRV_O_RDWR) { | |
408 | read_write = true; | |
409 | } | |
410 | ||
411 | if (read_write) { | |
6a8dc042 JC |
412 | *open_flags |= O_RDWR; |
413 | } else { | |
414 | *open_flags |= O_RDONLY; | |
415 | } | |
416 | ||
417 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
418 | * and O_DIRECT for no caching. */ | |
419 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
420 | *open_flags |= O_DIRECT; | |
421 | } | |
6a8dc042 JC |
422 | } |
423 | ||
078896a9 HR |
424 | static void raw_parse_filename(const char *filename, QDict *options, |
425 | Error **errp) | |
426 | { | |
03c320d8 | 427 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
428 | } |
429 | ||
c66a6157 KW |
430 | static QemuOptsList raw_runtime_opts = { |
431 | .name = "raw", | |
432 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
433 | .desc = { | |
434 | { | |
435 | .name = "filename", | |
436 | .type = QEMU_OPT_STRING, | |
437 | .help = "File name of the image", | |
438 | }, | |
0a4279d9 KW |
439 | { |
440 | .name = "aio", | |
441 | .type = QEMU_OPT_STRING, | |
442 | .help = "host AIO implementation (threads, native)", | |
443 | }, | |
16b48d5d FZ |
444 | { |
445 | .name = "locking", | |
446 | .type = QEMU_OPT_STRING, | |
447 | .help = "file locking mode (on/off/auto, default: auto)", | |
448 | }, | |
7c9e5276 PB |
449 | { |
450 | .name = "pr-manager", | |
451 | .type = QEMU_OPT_STRING, | |
452 | .help = "id of persistent reservation manager object (default: none)", | |
453 | }, | |
f357fcd8 SH |
454 | #if defined(__linux__) |
455 | { | |
456 | .name = "drop-cache", | |
457 | .type = QEMU_OPT_BOOL, | |
458 | .help = "invalidate page cache during live migration (default: on)", | |
459 | }, | |
460 | #endif | |
31be8a2a SH |
461 | { |
462 | .name = "x-check-cache-dropped", | |
463 | .type = QEMU_OPT_BOOL, | |
464 | .help = "check that page cache was dropped on live migration (default: off)" | |
465 | }, | |
c66a6157 KW |
466 | { /* end of list */ } |
467 | }, | |
468 | }; | |
469 | ||
8a2ce0bc AG |
470 | static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; |
471 | ||
c66a6157 | 472 | static int raw_open_common(BlockDriverState *bs, QDict *options, |
230ff739 JS |
473 | int bdrv_flags, int open_flags, |
474 | bool device, Error **errp) | |
83f64091 FB |
475 | { |
476 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
477 | QemuOpts *opts; |
478 | Error *local_err = NULL; | |
8bfea15d | 479 | const char *filename = NULL; |
7c9e5276 | 480 | const char *str; |
0a4279d9 | 481 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 482 | int fd, ret; |
260a82e5 | 483 | struct stat st; |
244a5668 | 484 | OnOffAuto locking; |
83f64091 | 485 | |
87ea75d5 | 486 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
c66a6157 | 487 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 488 | if (local_err) { |
e428e439 | 489 | error_propagate(errp, local_err); |
c66a6157 KW |
490 | ret = -EINVAL; |
491 | goto fail; | |
492 | } | |
493 | ||
494 | filename = qemu_opt_get(opts, "filename"); | |
495 | ||
db0754df | 496 | ret = raw_normalize_devicepath(&filename, errp); |
1de1ae0a | 497 | if (ret != 0) { |
c66a6157 | 498 | goto fail; |
1de1ae0a CE |
499 | } |
500 | ||
0a4279d9 KW |
501 | aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) |
502 | ? BLOCKDEV_AIO_OPTIONS_NATIVE | |
503 | : BLOCKDEV_AIO_OPTIONS_THREADS; | |
f7abe0ec MAL |
504 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
505 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 506 | aio_default, &local_err); |
0a4279d9 KW |
507 | if (local_err) { |
508 | error_propagate(errp, local_err); | |
509 | ret = -EINVAL; | |
510 | goto fail; | |
511 | } | |
512 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); | |
513 | ||
f7abe0ec MAL |
514 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
515 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 516 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
517 | if (local_err) { |
518 | error_propagate(errp, local_err); | |
519 | ret = -EINVAL; | |
520 | goto fail; | |
521 | } | |
522 | switch (locking) { | |
523 | case ON_OFF_AUTO_ON: | |
524 | s->use_lock = true; | |
2b218f5d | 525 | if (!qemu_has_ofd_lock()) { |
db0754df FZ |
526 | warn_report("File lock requested but OFD locking syscall is " |
527 | "unavailable, falling back to POSIX file locks"); | |
528 | error_printf("Due to the implementation, locks can be lost " | |
529 | "unexpectedly.\n"); | |
2b218f5d | 530 | } |
244a5668 FZ |
531 | break; |
532 | case ON_OFF_AUTO_OFF: | |
533 | s->use_lock = false; | |
534 | break; | |
535 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 536 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
537 | break; |
538 | default: | |
539 | abort(); | |
540 | } | |
541 | ||
7c9e5276 PB |
542 | str = qemu_opt_get(opts, "pr-manager"); |
543 | if (str) { | |
544 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
545 | if (local_err) { | |
546 | error_propagate(errp, local_err); | |
547 | ret = -EINVAL; | |
548 | goto fail; | |
549 | } | |
550 | } | |
551 | ||
f357fcd8 | 552 | s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); |
31be8a2a SH |
553 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
554 | false); | |
555 | ||
6a8dc042 | 556 | s->open_flags = open_flags; |
23dece19 | 557 | raw_parse_flags(bdrv_flags, &s->open_flags, false); |
83f64091 | 558 | |
90babde0 | 559 | s->fd = -1; |
40ff6d7e | 560 | fd = qemu_open(filename, s->open_flags, 0644); |
64107dc0 KW |
561 | ret = fd < 0 ? -errno : 0; |
562 | ||
64107dc0 | 563 | if (ret < 0) { |
f6fc1e30 | 564 | error_setg_file_open(errp, -ret, filename); |
c66a6157 | 565 | if (ret == -EROFS) { |
19cb3738 | 566 | ret = -EACCES; |
c66a6157 KW |
567 | } |
568 | goto fail; | |
19cb3738 | 569 | } |
83f64091 | 570 | s->fd = fd; |
9ef91a67 | 571 | |
244a5668 FZ |
572 | s->perm = 0; |
573 | s->shared_perm = BLK_PERM_ALL; | |
574 | ||
5c6c3a6c | 575 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 576 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
577 | if (s->use_linux_aio) { |
578 | if (!(s->open_flags & O_DIRECT)) { | |
579 | error_setg(errp, "aio=native was specified, but it requires " | |
580 | "cache.direct=on, which was not specified."); | |
581 | ret = -EINVAL; | |
582 | goto fail; | |
583 | } | |
584 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
585 | error_prepend(errp, "Unable to use native AIO: "); | |
586 | goto fail; | |
587 | } | |
96518254 | 588 | } |
1501ecc1 | 589 | #else |
0a4279d9 | 590 | if (s->use_linux_aio) { |
d657c0c2 KW |
591 | error_setg(errp, "aio=native was specified, but is not supported " |
592 | "in this build."); | |
593 | ret = -EINVAL; | |
594 | goto fail; | |
1501ecc1 SH |
595 | } |
596 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 597 | |
7ce21016 | 598 | s->has_discard = true; |
97a2ae34 | 599 | s->has_write_zeroes = true; |
3cad8307 RPM |
600 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { |
601 | s->needs_alignment = true; | |
602 | } | |
260a82e5 PB |
603 | |
604 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 605 | ret = -errno; |
260a82e5 PB |
606 | error_setg_errno(errp, errno, "Could not stat file"); |
607 | goto fail; | |
608 | } | |
230ff739 JS |
609 | |
610 | if (!device) { | |
611 | if (S_ISBLK(st.st_mode)) { | |
612 | warn_report("Opening a block device as a file using the '%s' " | |
613 | "driver is deprecated", bs->drv->format_name); | |
614 | } else if (S_ISCHR(st.st_mode)) { | |
615 | warn_report("Opening a character device as a file using the '%s' " | |
616 | "driver is deprecated", bs->drv->format_name); | |
617 | } else if (!S_ISREG(st.st_mode)) { | |
618 | error_setg(errp, "A regular file was expected by the '%s' driver, " | |
619 | "but something else was given", bs->drv->format_name); | |
620 | ret = -EINVAL; | |
621 | goto fail; | |
622 | } else { | |
623 | s->discard_zeroes = true; | |
624 | s->has_fallocate = true; | |
625 | } | |
626 | } else { | |
627 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
628 | error_setg(errp, "'%s' driver expects either " | |
629 | "a character or block device", bs->drv->format_name); | |
630 | ret = -EINVAL; | |
631 | goto fail; | |
632 | } | |
260a82e5 | 633 | } |
230ff739 | 634 | |
d0b4503e PB |
635 | if (S_ISBLK(st.st_mode)) { |
636 | #ifdef BLKDISCARDZEROES | |
637 | unsigned int arg; | |
638 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
639 | s->discard_zeroes = true; | |
640 | } | |
641 | #endif | |
642 | #ifdef __linux__ | |
643 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
644 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 645 | * Same for BLKZEROOUT. |
d0b4503e PB |
646 | */ |
647 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
648 | s->discard_zeroes = false; | |
97a2ae34 | 649 | s->has_write_zeroes = false; |
d0b4503e PB |
650 | } |
651 | #endif | |
652 | } | |
3cad8307 RPM |
653 | #ifdef __FreeBSD__ |
654 | if (S_ISCHR(st.st_mode)) { | |
655 | /* | |
656 | * The file is a char device (disk), which on FreeBSD isn't behind | |
657 | * a pager, so force all requests to be aligned. This is needed | |
658 | * so QEMU makes sure all IO operations on the device are aligned | |
659 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
660 | */ | |
661 | s->needs_alignment = true; | |
662 | } | |
663 | #endif | |
260a82e5 | 664 | |
dce512de CH |
665 | #ifdef CONFIG_XFS |
666 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 667 | s->is_xfs = true; |
dce512de CH |
668 | } |
669 | #endif | |
670 | ||
738301e1 | 671 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
c66a6157 KW |
672 | ret = 0; |
673 | fail: | |
8bfea15d KW |
674 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
675 | unlink(filename); | |
676 | } | |
c66a6157 KW |
677 | qemu_opts_del(opts); |
678 | return ret; | |
83f64091 FB |
679 | } |
680 | ||
015a1036 HR |
681 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
682 | Error **errp) | |
90babde0 CH |
683 | { |
684 | BDRVRawState *s = bs->opaque; | |
685 | ||
686 | s->type = FTYPE_FILE; | |
230ff739 | 687 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
688 | } |
689 | ||
244a5668 FZ |
690 | typedef enum { |
691 | RAW_PL_PREPARE, | |
692 | RAW_PL_COMMIT, | |
693 | RAW_PL_ABORT, | |
694 | } RawPermLockOp; | |
695 | ||
696 | #define PERM_FOREACH(i) \ | |
697 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
698 | ||
699 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
700 | * file; if @unlock == true, also unlock the unneeded bytes. | |
701 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
702 | */ | |
2996ffad | 703 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
704 | uint64_t perm_lock_bits, |
705 | uint64_t shared_perm_lock_bits, | |
706 | bool unlock, Error **errp) | |
707 | { | |
708 | int ret; | |
709 | int i; | |
2996ffad FZ |
710 | uint64_t locked_perm, locked_shared_perm; |
711 | ||
712 | if (s) { | |
713 | locked_perm = s->locked_perm; | |
714 | locked_shared_perm = s->locked_shared_perm; | |
715 | } else { | |
716 | /* | |
717 | * We don't have the previous bits, just lock/unlock for each of the | |
718 | * requested bits. | |
719 | */ | |
720 | if (unlock) { | |
721 | locked_perm = BLK_PERM_ALL; | |
722 | locked_shared_perm = BLK_PERM_ALL; | |
723 | } else { | |
724 | locked_perm = 0; | |
725 | locked_shared_perm = 0; | |
726 | } | |
727 | } | |
244a5668 FZ |
728 | |
729 | PERM_FOREACH(i) { | |
730 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
731 | uint64_t bit = (1ULL << i); |
732 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 733 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
734 | if (ret) { |
735 | error_setg(errp, "Failed to lock byte %d", off); | |
736 | return ret; | |
2996ffad FZ |
737 | } else if (s) { |
738 | s->locked_perm |= bit; | |
244a5668 | 739 | } |
2996ffad | 740 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 741 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
742 | if (ret) { |
743 | error_setg(errp, "Failed to unlock byte %d", off); | |
744 | return ret; | |
2996ffad FZ |
745 | } else if (s) { |
746 | s->locked_perm &= ~bit; | |
244a5668 FZ |
747 | } |
748 | } | |
749 | } | |
750 | PERM_FOREACH(i) { | |
751 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
752 | uint64_t bit = (1ULL << i); |
753 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 754 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
755 | if (ret) { |
756 | error_setg(errp, "Failed to lock byte %d", off); | |
757 | return ret; | |
2996ffad FZ |
758 | } else if (s) { |
759 | s->locked_shared_perm |= bit; | |
244a5668 | 760 | } |
2996ffad FZ |
761 | } else if (unlock && (locked_shared_perm & bit) && |
762 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 763 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
764 | if (ret) { |
765 | error_setg(errp, "Failed to unlock byte %d", off); | |
766 | return ret; | |
2996ffad FZ |
767 | } else if (s) { |
768 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
769 | } |
770 | } | |
771 | } | |
772 | return 0; | |
773 | } | |
774 | ||
775 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 776 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
777 | Error **errp) |
778 | { | |
779 | int ret; | |
780 | int i; | |
781 | ||
782 | PERM_FOREACH(i) { | |
783 | int off = RAW_LOCK_SHARED_BASE + i; | |
784 | uint64_t p = 1ULL << i; | |
785 | if (perm & p) { | |
d0a96155 | 786 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
787 | if (ret) { |
788 | char *perm_name = bdrv_perm_names(p); | |
789 | error_setg(errp, | |
790 | "Failed to get \"%s\" lock", | |
791 | perm_name); | |
792 | g_free(perm_name); | |
244a5668 FZ |
793 | return ret; |
794 | } | |
795 | } | |
796 | } | |
797 | PERM_FOREACH(i) { | |
798 | int off = RAW_LOCK_PERM_BASE + i; | |
799 | uint64_t p = 1ULL << i; | |
800 | if (!(shared_perm & p)) { | |
d0a96155 | 801 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
802 | if (ret) { |
803 | char *perm_name = bdrv_perm_names(p); | |
804 | error_setg(errp, | |
805 | "Failed to get shared \"%s\" lock", | |
806 | perm_name); | |
807 | g_free(perm_name); | |
244a5668 FZ |
808 | return ret; |
809 | } | |
810 | } | |
811 | } | |
812 | return 0; | |
813 | } | |
814 | ||
815 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
816 | RawPermLockOp op, | |
817 | uint64_t new_perm, uint64_t new_shared, | |
818 | Error **errp) | |
819 | { | |
820 | BDRVRawState *s = bs->opaque; | |
821 | int ret = 0; | |
822 | Error *local_err = NULL; | |
823 | ||
824 | if (!s->use_lock) { | |
825 | return 0; | |
826 | } | |
827 | ||
828 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
829 | return 0; | |
830 | } | |
831 | ||
244a5668 FZ |
832 | switch (op) { |
833 | case RAW_PL_PREPARE: | |
696aaaed VSO |
834 | if ((s->perm | new_perm) == s->perm && |
835 | (s->shared_perm & new_shared) == s->shared_perm) | |
836 | { | |
837 | /* | |
838 | * We are going to unlock bytes, it should not fail. If it fail due | |
839 | * to some fs-dependent permission-unrelated reasons (which occurs | |
840 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
841 | * can't prevent such errors by any check here. And we ignore them | |
842 | * anyway in ABORT and COMMIT. | |
843 | */ | |
844 | return 0; | |
845 | } | |
f2e3af29 | 846 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
847 | ~s->shared_perm | ~new_shared, |
848 | false, errp); | |
849 | if (!ret) { | |
f2e3af29 | 850 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
851 | if (!ret) { |
852 | return 0; | |
853 | } | |
b857431d FZ |
854 | error_append_hint(errp, |
855 | "Is another process using the image [%s]?\n", | |
856 | bs->filename); | |
244a5668 FZ |
857 | } |
858 | op = RAW_PL_ABORT; | |
859 | /* fall through to unlock bytes. */ | |
860 | case RAW_PL_ABORT: | |
f2e3af29 | 861 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 862 | true, &local_err); |
244a5668 FZ |
863 | if (local_err) { |
864 | /* Theoretically the above call only unlocks bytes and it cannot | |
865 | * fail. Something weird happened, report it. | |
866 | */ | |
db0754df | 867 | warn_report_err(local_err); |
244a5668 FZ |
868 | } |
869 | break; | |
870 | case RAW_PL_COMMIT: | |
f2e3af29 | 871 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 872 | true, &local_err); |
244a5668 FZ |
873 | if (local_err) { |
874 | /* Theoretically the above call only unlocks bytes and it cannot | |
875 | * fail. Something weird happened, report it. | |
876 | */ | |
db0754df | 877 | warn_report_err(local_err); |
244a5668 FZ |
878 | } |
879 | break; | |
880 | } | |
881 | return ret; | |
882 | } | |
883 | ||
5cec2870 | 884 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 885 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 886 | Error **errp) |
5cec2870 KW |
887 | { |
888 | BDRVRawState *s = bs->opaque; | |
889 | int fd = -1; | |
890 | int ret; | |
23dece19 KW |
891 | bool has_writers = perm & |
892 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
893 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
894 | #ifdef O_NOATIME | |
895 | fcntl_flags |= O_NOATIME; | |
896 | #endif | |
897 | ||
898 | *open_flags = 0; | |
899 | if (s->type == FTYPE_CD) { | |
900 | *open_flags |= O_NONBLOCK; | |
901 | } | |
902 | ||
23dece19 | 903 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
904 | |
905 | #ifdef O_ASYNC | |
906 | /* Not all operating systems have O_ASYNC, and those that don't | |
907 | * will not let us track the state into rs->open_flags (typically | |
908 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
909 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
910 | */ | |
911 | assert((s->open_flags & O_ASYNC) == 0); | |
912 | #endif | |
913 | ||
6ceabe6f KW |
914 | if (!force_dup && *open_flags == s->open_flags) { |
915 | /* We're lucky, the existing fd is fine */ | |
916 | return s->fd; | |
917 | } | |
918 | ||
5cec2870 KW |
919 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
920 | /* dup the original fd */ | |
921 | fd = qemu_dup(s->fd); | |
922 | if (fd >= 0) { | |
923 | ret = fcntl_setfl(fd, *open_flags); | |
924 | if (ret) { | |
925 | qemu_close(fd); | |
926 | fd = -1; | |
927 | } | |
928 | } | |
929 | } | |
930 | ||
931 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ | |
932 | if (fd == -1) { | |
933 | const char *normalized_filename = bs->filename; | |
934 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
935 | if (ret >= 0) { | |
936 | assert(!(*open_flags & O_CREAT)); | |
937 | fd = qemu_open(normalized_filename, *open_flags); | |
938 | if (fd == -1) { | |
939 | error_setg_errno(errp, errno, "Could not reopen file"); | |
940 | return -1; | |
941 | } | |
942 | } | |
943 | } | |
944 | ||
945 | return fd; | |
946 | } | |
947 | ||
eeb6b45d JC |
948 | static int raw_reopen_prepare(BDRVReopenState *state, |
949 | BlockReopenQueue *queue, Error **errp) | |
950 | { | |
951 | BDRVRawState *s; | |
4e6d13c9 | 952 | BDRVRawReopenState *rs; |
31be8a2a | 953 | QemuOpts *opts; |
a6aeca0c | 954 | int ret; |
df26a350 | 955 | Error *local_err = NULL; |
eeb6b45d JC |
956 | |
957 | assert(state != NULL); | |
958 | assert(state->bs != NULL); | |
959 | ||
960 | s = state->bs->opaque; | |
961 | ||
5839e53b | 962 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 963 | rs = state->opaque; |
31be8a2a SH |
964 | |
965 | /* Handle options changes */ | |
966 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
967 | qemu_opts_absorb_qdict(opts, state->options, &local_err); | |
968 | if (local_err) { | |
969 | error_propagate(errp, local_err); | |
970 | ret = -EINVAL; | |
971 | goto out; | |
972 | } | |
973 | ||
f357fcd8 | 974 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
975 | rs->check_cache_dropped = |
976 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
977 | ||
978 | /* This driver's reopen function doesn't currently allow changing | |
979 | * other options, so let's put them back in the original QDict and | |
980 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
981 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 982 | |
5cec2870 | 983 | rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, |
23dece19 | 984 | state->perm, true, &local_err); |
5cec2870 KW |
985 | if (local_err) { |
986 | error_propagate(errp, local_err); | |
987 | ret = -1; | |
988 | goto out; | |
eeb6b45d | 989 | } |
df26a350 KW |
990 | |
991 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
992 | * alignment with the new fd. */ | |
4e6d13c9 FZ |
993 | if (rs->fd != -1) { |
994 | raw_probe_alignment(state->bs, rs->fd, &local_err); | |
df26a350 | 995 | if (local_err) { |
df26a350 KW |
996 | error_propagate(errp, local_err); |
997 | ret = -EINVAL; | |
a6aeca0c KW |
998 | goto out_fd; |
999 | } | |
df26a350 KW |
1000 | } |
1001 | ||
e0c9cf3a | 1002 | s->reopen_state = state; |
a6aeca0c KW |
1003 | ret = 0; |
1004 | out_fd: | |
1005 | if (ret < 0) { | |
1006 | qemu_close(rs->fd); | |
1007 | rs->fd = -1; | |
1008 | } | |
31be8a2a SH |
1009 | out: |
1010 | qemu_opts_del(opts); | |
eeb6b45d JC |
1011 | return ret; |
1012 | } | |
1013 | ||
eeb6b45d JC |
1014 | static void raw_reopen_commit(BDRVReopenState *state) |
1015 | { | |
4e6d13c9 | 1016 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1017 | BDRVRawState *s = state->bs->opaque; |
1018 | ||
f357fcd8 | 1019 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1020 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1021 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1022 | |
1023 | qemu_close(s->fd); | |
4e6d13c9 | 1024 | s->fd = rs->fd; |
eeb6b45d JC |
1025 | |
1026 | g_free(state->opaque); | |
1027 | state->opaque = NULL; | |
e0c9cf3a KW |
1028 | |
1029 | assert(s->reopen_state == state); | |
1030 | s->reopen_state = NULL; | |
eeb6b45d JC |
1031 | } |
1032 | ||
1033 | ||
1034 | static void raw_reopen_abort(BDRVReopenState *state) | |
1035 | { | |
4e6d13c9 | 1036 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1037 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1038 | |
1039 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1040 | if (rs == NULL) { |
eeb6b45d JC |
1041 | return; |
1042 | } | |
1043 | ||
4e6d13c9 FZ |
1044 | if (rs->fd >= 0) { |
1045 | qemu_close(rs->fd); | |
1046 | rs->fd = -1; | |
eeb6b45d JC |
1047 | } |
1048 | g_free(state->opaque); | |
1049 | state->opaque = NULL; | |
e0c9cf3a KW |
1050 | |
1051 | assert(s->reopen_state == state); | |
1052 | s->reopen_state = NULL; | |
eeb6b45d JC |
1053 | } |
1054 | ||
867eccfe | 1055 | static int sg_get_max_transfer_length(int fd) |
6f607174 FZ |
1056 | { |
1057 | #ifdef BLKSECTGET | |
48265250 | 1058 | int max_bytes = 0; |
867eccfe ML |
1059 | |
1060 | if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
48265250 | 1061 | return max_bytes; |
6f607174 FZ |
1062 | } else { |
1063 | return -errno; | |
1064 | } | |
1065 | #else | |
1066 | return -ENOSYS; | |
1067 | #endif | |
1068 | } | |
1069 | ||
867eccfe | 1070 | static int sg_get_max_segments(int fd) |
9103f1ce FZ |
1071 | { |
1072 | #ifdef CONFIG_LINUX | |
1073 | char buf[32]; | |
1074 | const char *end; | |
867eccfe | 1075 | char *sysfspath = NULL; |
9103f1ce | 1076 | int ret; |
867eccfe | 1077 | int sysfd = -1; |
9103f1ce | 1078 | long max_segments; |
867eccfe ML |
1079 | struct stat st; |
1080 | ||
1081 | if (fstat(fd, &st)) { | |
1082 | ret = -errno; | |
1083 | goto out; | |
1084 | } | |
9103f1ce FZ |
1085 | |
1086 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", | |
867eccfe ML |
1087 | major(st.st_rdev), minor(st.st_rdev)); |
1088 | sysfd = open(sysfspath, O_RDONLY); | |
1089 | if (sysfd == -1) { | |
9103f1ce FZ |
1090 | ret = -errno; |
1091 | goto out; | |
1092 | } | |
1093 | do { | |
867eccfe | 1094 | ret = read(sysfd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1095 | } while (ret == -1 && errno == EINTR); |
1096 | if (ret < 0) { | |
1097 | ret = -errno; | |
1098 | goto out; | |
1099 | } else if (ret == 0) { | |
1100 | ret = -EIO; | |
1101 | goto out; | |
1102 | } | |
1103 | buf[ret] = 0; | |
1104 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1105 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1106 | if (ret == 0 && end && *end == '\n') { | |
1107 | ret = max_segments; | |
1108 | } | |
1109 | ||
1110 | out: | |
867eccfe ML |
1111 | if (sysfd != -1) { |
1112 | close(sysfd); | |
fed414df | 1113 | } |
9103f1ce FZ |
1114 | g_free(sysfspath); |
1115 | return ret; | |
1116 | #else | |
1117 | return -ENOTSUP; | |
1118 | #endif | |
1119 | } | |
1120 | ||
3baca891 | 1121 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1122 | { |
1123 | BDRVRawState *s = bs->opaque; | |
6f607174 | 1124 | |
867eccfe ML |
1125 | if (bs->sg) { |
1126 | int ret = sg_get_max_transfer_length(s->fd); | |
1127 | ||
1128 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
1129 | bs->bl.max_transfer = pow2floor(ret); | |
1130 | } | |
1131 | ||
1132 | ret = sg_get_max_segments(s->fd); | |
1133 | if (ret > 0) { | |
1134 | bs->bl.max_transfer = MIN(bs->bl.max_transfer, ret * getpagesize()); | |
6f607174 FZ |
1135 | } |
1136 | } | |
eeb6b45d | 1137 | |
df26a350 | 1138 | raw_probe_alignment(bs, s->fd, errp); |
4196d2f0 | 1139 | bs->bl.min_mem_alignment = s->buf_align; |
459b4e66 | 1140 | bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); |
c25f53b0 | 1141 | } |
83f64091 | 1142 | |
1a9335e4 ET |
1143 | static int check_for_dasd(int fd) |
1144 | { | |
1145 | #ifdef BIODASDINFO2 | |
1146 | struct dasd_information2_t info = {0}; | |
1147 | ||
1148 | return ioctl(fd, BIODASDINFO2, &info); | |
1149 | #else | |
1150 | return -1; | |
1151 | #endif | |
1152 | } | |
1153 | ||
1154 | /** | |
1155 | * Try to get @bs's logical and physical block size. | |
1156 | * On success, store them in @bsz and return zero. | |
1157 | * On failure, return negative errno. | |
1158 | */ | |
1159 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1160 | { | |
1161 | BDRVRawState *s = bs->opaque; | |
1162 | int ret; | |
1163 | ||
1164 | /* If DASD, get blocksizes */ | |
1165 | if (check_for_dasd(s->fd) < 0) { | |
1166 | return -ENOTSUP; | |
1167 | } | |
1168 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1169 | if (ret < 0) { | |
1170 | return ret; | |
1171 | } | |
1172 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1173 | } | |
1174 | ||
1175 | /** | |
1176 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1177 | * On success, store them in @geo and return 0. | |
1178 | * On failure return -errno. | |
1179 | * (Allows block driver to assign default geometry values that guest sees) | |
1180 | */ | |
1181 | #ifdef __linux__ | |
1182 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1183 | { | |
1184 | BDRVRawState *s = bs->opaque; | |
1185 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1186 | |
1187 | /* If DASD, get its geometry */ | |
1188 | if (check_for_dasd(s->fd) < 0) { | |
1189 | return -ENOTSUP; | |
1190 | } | |
1191 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1192 | return -errno; | |
1193 | } | |
1194 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1195 | (e.g. certain multipath setups) */ | |
1196 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1197 | return -ENOTSUP; | |
1198 | } | |
1199 | /* Do not return a geometry for partition */ | |
1200 | if (ioctl_geo.start != 0) { | |
1201 | return -ENOTSUP; | |
1202 | } | |
1203 | geo->heads = ioctl_geo.heads; | |
1204 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1205 | geo->cylinders = ioctl_geo.cylinders; |
1206 | ||
1207 | return 0; | |
1208 | } | |
1209 | #else /* __linux__ */ | |
1210 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1211 | { | |
1212 | return -ENOTSUP; | |
1213 | } | |
1214 | #endif | |
1215 | ||
03425671 KW |
1216 | #if defined(__linux__) |
1217 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1218 | { |
03425671 | 1219 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1220 | int ret; |
1221 | ||
d57c44d0 | 1222 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); |
de81a169 PB |
1223 | if (ret == -1) { |
1224 | return -errno; | |
1225 | } | |
1226 | ||
b608c8dc | 1227 | return 0; |
de81a169 | 1228 | } |
03425671 | 1229 | #endif /* linux */ |
de81a169 | 1230 | |
06dc9bd5 | 1231 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1232 | { |
06dc9bd5 | 1233 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1234 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1235 | int ret; |
1236 | ||
e5bcf967 KW |
1237 | if (s->page_cache_inconsistent) { |
1238 | return -EIO; | |
1239 | } | |
1240 | ||
de81a169 PB |
1241 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1242 | if (ret == -1) { | |
e5bcf967 KW |
1243 | /* There is no clear definition of the semantics of a failing fsync(), |
1244 | * so we may have to assume the worst. The sad truth is that this | |
1245 | * assumption is correct for Linux. Some pages are now probably marked | |
1246 | * clean in the page cache even though they are inconsistent with the | |
1247 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1248 | * further writeback attempt will be made. We can't get back to a state | |
1249 | * in which we know what is on disk (we would have to rewrite | |
1250 | * everything that was touched since the last fdatasync() at least), so | |
1251 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1252 | * really defined, I have little hope that other OSes are doing better. | |
1253 | * | |
1254 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1255 | * cache. */ | |
1256 | if ((s->open_flags & O_DIRECT) == 0) { | |
1257 | s->page_cache_inconsistent = true; | |
1258 | } | |
de81a169 PB |
1259 | return -errno; |
1260 | } | |
1261 | return 0; | |
1262 | } | |
1263 | ||
1264 | #ifdef CONFIG_PREADV | |
1265 | ||
1266 | static bool preadv_present = true; | |
1267 | ||
1268 | static ssize_t | |
1269 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1270 | { | |
1271 | return preadv(fd, iov, nr_iov, offset); | |
1272 | } | |
1273 | ||
1274 | static ssize_t | |
1275 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1276 | { | |
1277 | return pwritev(fd, iov, nr_iov, offset); | |
1278 | } | |
1279 | ||
1280 | #else | |
1281 | ||
1282 | static bool preadv_present = false; | |
1283 | ||
1284 | static ssize_t | |
1285 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1286 | { | |
1287 | return -ENOSYS; | |
1288 | } | |
1289 | ||
1290 | static ssize_t | |
1291 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1292 | { | |
1293 | return -ENOSYS; | |
1294 | } | |
1295 | ||
1296 | #endif | |
1297 | ||
1298 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1299 | { | |
1300 | ssize_t len; | |
1301 | ||
1302 | do { | |
1303 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1304 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1305 | aiocb->io.iov, |
1306 | aiocb->io.niov, | |
de81a169 PB |
1307 | aiocb->aio_offset); |
1308 | else | |
1309 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1310 | aiocb->io.iov, |
1311 | aiocb->io.niov, | |
de81a169 PB |
1312 | aiocb->aio_offset); |
1313 | } while (len == -1 && errno == EINTR); | |
1314 | ||
1315 | if (len == -1) { | |
1316 | return -errno; | |
1317 | } | |
1318 | return len; | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Read/writes the data to/from a given linear buffer. | |
1323 | * | |
1324 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1325 | * reads are only returned if the end of the file is reached. | |
1326 | */ | |
1327 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1328 | { | |
1329 | ssize_t offset = 0; | |
1330 | ssize_t len; | |
1331 | ||
1332 | while (offset < aiocb->aio_nbytes) { | |
1333 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1334 | len = pwrite(aiocb->aio_fildes, | |
1335 | (const char *)buf + offset, | |
1336 | aiocb->aio_nbytes - offset, | |
1337 | aiocb->aio_offset + offset); | |
1338 | } else { | |
1339 | len = pread(aiocb->aio_fildes, | |
1340 | buf + offset, | |
1341 | aiocb->aio_nbytes - offset, | |
1342 | aiocb->aio_offset + offset); | |
1343 | } | |
1344 | if (len == -1 && errno == EINTR) { | |
1345 | continue; | |
61ed73cf SH |
1346 | } else if (len == -1 && errno == EINVAL && |
1347 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1348 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1349 | offset > 0) { | |
1350 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1351 | * after a short read. Assume that O_DIRECT short reads only occur | |
1352 | * at EOF. Therefore this is a short read, not an I/O error. | |
1353 | */ | |
1354 | break; | |
de81a169 PB |
1355 | } else if (len == -1) { |
1356 | offset = -errno; | |
1357 | break; | |
1358 | } else if (len == 0) { | |
1359 | break; | |
1360 | } | |
1361 | offset += len; | |
1362 | } | |
1363 | ||
1364 | return offset; | |
1365 | } | |
1366 | ||
999e6b69 | 1367 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1368 | { |
999e6b69 | 1369 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1370 | ssize_t nbytes; |
1371 | char *buf; | |
1372 | ||
1373 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1374 | /* | |
1375 | * If there is just a single buffer, and it is properly aligned | |
1376 | * we can just use plain pread/pwrite without any problems. | |
1377 | */ | |
d57c44d0 | 1378 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1379 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1380 | goto out; | |
de81a169 PB |
1381 | } |
1382 | /* | |
1383 | * We have more than one iovec, and all are properly aligned. | |
1384 | * | |
1385 | * Try preadv/pwritev first and fall back to linearizing the | |
1386 | * buffer if it's not supported. | |
1387 | */ | |
1388 | if (preadv_present) { | |
1389 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1390 | if (nbytes == aiocb->aio_nbytes || | |
1391 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1392 | goto out; |
de81a169 PB |
1393 | } |
1394 | preadv_present = false; | |
1395 | } | |
1396 | ||
1397 | /* | |
1398 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1399 | * using these interfaces. For now retry using plain | |
1400 | * pread/pwrite? | |
1401 | */ | |
1402 | } | |
1403 | ||
1404 | /* | |
1405 | * Ok, we have to do it the hard way, copy all segments into | |
1406 | * a single aligned buffer. | |
1407 | */ | |
50d4a858 KW |
1408 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1409 | if (buf == NULL) { | |
54c7ca1b KW |
1410 | nbytes = -ENOMEM; |
1411 | goto out; | |
50d4a858 KW |
1412 | } |
1413 | ||
de81a169 PB |
1414 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1415 | char *p = buf; | |
1416 | int i; | |
1417 | ||
d57c44d0 KW |
1418 | for (i = 0; i < aiocb->io.niov; ++i) { |
1419 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1420 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1421 | } |
8eb029c2 | 1422 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1423 | } |
1424 | ||
1425 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1426 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1427 | char *p = buf; | |
1428 | size_t count = aiocb->aio_nbytes, copy; | |
1429 | int i; | |
1430 | ||
d57c44d0 | 1431 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1432 | copy = count; |
d57c44d0 KW |
1433 | if (copy > aiocb->io.iov[i].iov_len) { |
1434 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1435 | } |
d57c44d0 | 1436 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1437 | assert(count >= copy); |
de81a169 PB |
1438 | p += copy; |
1439 | count -= copy; | |
1440 | } | |
8eb029c2 | 1441 | assert(count == 0); |
de81a169 PB |
1442 | } |
1443 | qemu_vfree(buf); | |
1444 | ||
54c7ca1b KW |
1445 | out: |
1446 | if (nbytes == aiocb->aio_nbytes) { | |
1447 | return 0; | |
1448 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1449 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1450 | return -EINVAL; | |
1451 | } else { | |
1452 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1453 | 0, aiocb->aio_nbytes - nbytes); | |
1454 | return 0; | |
1455 | } | |
1456 | } else { | |
1457 | assert(nbytes < 0); | |
1458 | return nbytes; | |
1459 | } | |
de81a169 PB |
1460 | } |
1461 | ||
1486df0e DL |
1462 | static int translate_err(int err) |
1463 | { | |
1464 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1465 | err == -ENOTTY) { | |
1466 | err = -ENOTSUP; | |
1467 | } | |
1468 | return err; | |
1469 | } | |
1470 | ||
d50d8222 | 1471 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1472 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1473 | { | |
1474 | do { | |
1475 | if (fallocate(fd, mode, offset, len) == 0) { | |
1476 | return 0; | |
1477 | } | |
1478 | } while (errno == EINTR); | |
1479 | return translate_err(-errno); | |
1480 | } | |
1481 | #endif | |
1482 | ||
37cc9f7f | 1483 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1484 | { |
37cc9f7f | 1485 | int ret = -ENOTSUP; |
97a2ae34 PB |
1486 | BDRVRawState *s = aiocb->bs->opaque; |
1487 | ||
37cc9f7f | 1488 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1489 | return -ENOTSUP; |
1490 | } | |
1491 | ||
97a2ae34 | 1492 | #ifdef BLKZEROOUT |
738301e1 KW |
1493 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1494 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1495 | * fallbacks. */ | |
1496 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1497 | do { | |
1498 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1499 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1500 | return 0; | |
1501 | } | |
1502 | } while (errno == EINTR); | |
37cc9f7f | 1503 | |
738301e1 | 1504 | ret = translate_err(-errno); |
effecce6 KW |
1505 | if (ret == -ENOTSUP) { |
1506 | s->has_write_zeroes = false; | |
1507 | } | |
738301e1 | 1508 | } |
97a2ae34 | 1509 | #endif |
97a2ae34 | 1510 | |
97a2ae34 PB |
1511 | return ret; |
1512 | } | |
1513 | ||
7154d8ae | 1514 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1515 | { |
7154d8ae | 1516 | RawPosixAIOData *aiocb = opaque; |
70d9110b | 1517 | #ifdef CONFIG_FALLOCATE |
b2c6f23f | 1518 | BDRVRawState *s = aiocb->bs->opaque; |
70d9110b DL |
1519 | int64_t len; |
1520 | #endif | |
37cc9f7f DL |
1521 | |
1522 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1523 | return handle_aiocb_write_zeroes_block(aiocb); | |
1524 | } | |
1525 | ||
b953f075 DL |
1526 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1527 | if (s->has_write_zeroes) { | |
1528 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1529 | aiocb->aio_offset, aiocb->aio_nbytes); | |
294682cc AS |
1530 | if (ret == -EINVAL) { |
1531 | /* | |
1532 | * Allow falling back to pwrite for file systems that | |
1533 | * do not support fallocate() for an unaligned byte range. | |
1534 | */ | |
1535 | return -ENOTSUP; | |
1536 | } | |
b953f075 DL |
1537 | if (ret == 0 || ret != -ENOTSUP) { |
1538 | return ret; | |
1539 | } | |
1540 | s->has_write_zeroes = false; | |
1541 | } | |
1542 | #endif | |
1543 | ||
1cdc3239 DL |
1544 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1545 | if (s->has_discard && s->has_fallocate) { | |
1546 | int ret = do_fallocate(s->fd, | |
1547 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1548 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1549 | if (ret == 0) { | |
1550 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1551 | if (ret == 0 || ret != -ENOTSUP) { | |
1552 | return ret; | |
1553 | } | |
1554 | s->has_fallocate = false; | |
1555 | } else if (ret != -ENOTSUP) { | |
1556 | return ret; | |
1557 | } else { | |
1558 | s->has_discard = false; | |
1559 | } | |
1560 | } | |
1561 | #endif | |
1562 | ||
d50d8222 | 1563 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1564 | /* Last resort: we are trying to extend the file with zeroed data. This |
1565 | * can be done via fallocate(fd, 0) */ | |
1566 | len = bdrv_getlength(aiocb->bs); | |
1567 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1568 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1569 | if (ret == 0 || ret != -ENOTSUP) { | |
1570 | return ret; | |
1571 | } | |
1572 | s->has_fallocate = false; | |
1573 | } | |
1574 | #endif | |
1575 | ||
37cc9f7f DL |
1576 | return -ENOTSUP; |
1577 | } | |
1578 | ||
7154d8ae | 1579 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1580 | { |
7154d8ae | 1581 | RawPosixAIOData *aiocb = opaque; |
34fa110e KW |
1582 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
1583 | int ret; | |
1584 | ||
1585 | /* First try to write zeros and unmap at the same time */ | |
1586 | ||
1587 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
1588 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1589 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1590 | if (ret != -ENOTSUP) { | |
1591 | return ret; | |
1592 | } | |
1593 | #endif | |
1594 | ||
34fa110e KW |
1595 | /* If we couldn't manage to unmap while guaranteed that the area reads as |
1596 | * all-zero afterwards, just write zeroes without unmapping */ | |
1597 | ret = handle_aiocb_write_zeroes(aiocb); | |
1598 | return ret; | |
1599 | } | |
1600 | ||
1efad060 FZ |
1601 | #ifndef HAVE_COPY_FILE_RANGE |
1602 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1603 | off_t *out_off, size_t len, unsigned int flags) | |
1604 | { | |
1605 | #ifdef __NR_copy_file_range | |
1606 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1607 | out_off, len, flags); | |
1608 | #else | |
1609 | errno = ENOSYS; | |
1610 | return -1; | |
1611 | #endif | |
1612 | } | |
1613 | #endif | |
1614 | ||
58a209c4 | 1615 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1616 | { |
58a209c4 | 1617 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1618 | uint64_t bytes = aiocb->aio_nbytes; |
1619 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1620 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1621 | |
1622 | while (bytes) { | |
1623 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1624 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1625 | bytes, 0); |
ecc983a5 | 1626 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1627 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1628 | 0, ret); | |
c436e3d0 FZ |
1629 | if (ret == 0) { |
1630 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1631 | * buffer I/O. */ | |
1632 | return -ENOSPC; | |
1efad060 FZ |
1633 | } |
1634 | if (ret < 0) { | |
c436e3d0 FZ |
1635 | switch (errno) { |
1636 | case ENOSYS: | |
1efad060 | 1637 | return -ENOTSUP; |
c436e3d0 FZ |
1638 | case EINTR: |
1639 | continue; | |
1640 | default: | |
1efad060 FZ |
1641 | return -errno; |
1642 | } | |
1643 | } | |
1efad060 FZ |
1644 | bytes -= ret; |
1645 | } | |
1646 | return 0; | |
1647 | } | |
1648 | ||
46ee0f46 | 1649 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1650 | { |
46ee0f46 | 1651 | RawPosixAIOData *aiocb = opaque; |
8238010b PB |
1652 | int ret = -EOPNOTSUPP; |
1653 | BDRVRawState *s = aiocb->bs->opaque; | |
1654 | ||
7ce21016 PB |
1655 | if (!s->has_discard) { |
1656 | return -ENOTSUP; | |
8238010b PB |
1657 | } |
1658 | ||
1659 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1660 | #ifdef BLKDISCARD | |
1661 | do { | |
1662 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1663 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1664 | return 0; | |
1665 | } | |
1666 | } while (errno == EINTR); | |
1667 | ||
1668 | ret = -errno; | |
1669 | #endif | |
1670 | } else { | |
8238010b | 1671 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
0b991712 DL |
1672 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1673 | aiocb->aio_offset, aiocb->aio_nbytes); | |
8238010b PB |
1674 | #endif |
1675 | } | |
1676 | ||
1486df0e DL |
1677 | ret = translate_err(ret); |
1678 | if (ret == -ENOTSUP) { | |
7ce21016 | 1679 | s->has_discard = false; |
8238010b PB |
1680 | } |
1681 | return ret; | |
1682 | } | |
1683 | ||
3a20013f NS |
1684 | /* |
1685 | * Help alignment probing by allocating the first block. | |
1686 | * | |
1687 | * When reading with direct I/O from unallocated area on Gluster backed by XFS, | |
1688 | * reading succeeds regardless of request length. In this case we fallback to | |
1689 | * safe alignment which is not optimal. Allocating the first block avoids this | |
1690 | * fallback. | |
1691 | * | |
1692 | * fd may be opened with O_DIRECT, but we don't know the buffer alignment or | |
1693 | * request alignment, so we use safe values. | |
1694 | * | |
1695 | * Returns: 0 on success, -errno on failure. Since this is an optimization, | |
1696 | * caller may ignore failures. | |
1697 | */ | |
1698 | static int allocate_first_block(int fd, size_t max_size) | |
1699 | { | |
1700 | size_t write_size = (max_size < MAX_BLOCKSIZE) | |
1701 | ? BDRV_SECTOR_SIZE | |
1702 | : MAX_BLOCKSIZE; | |
1703 | size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); | |
1704 | void *buf; | |
1705 | ssize_t n; | |
1706 | int ret; | |
1707 | ||
1708 | buf = qemu_memalign(max_align, write_size); | |
1709 | memset(buf, 0, write_size); | |
1710 | ||
1711 | do { | |
1712 | n = pwrite(fd, buf, write_size, 0); | |
1713 | } while (n == -1 && errno == EINTR); | |
1714 | ||
1715 | ret = (n == -1) ? -errno : 0; | |
1716 | ||
1717 | qemu_vfree(buf); | |
1718 | return ret; | |
1719 | } | |
1720 | ||
29cb4c01 | 1721 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1722 | { |
29cb4c01 | 1723 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1724 | int result = 0; |
1725 | int64_t current_length = 0; | |
1726 | char *buf = NULL; | |
1727 | struct stat st; | |
1728 | int fd = aiocb->aio_fildes; | |
1729 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1730 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1731 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1732 | |
1733 | if (fstat(fd, &st) < 0) { | |
1734 | result = -errno; | |
1735 | error_setg_errno(errp, -result, "Could not stat file"); | |
1736 | return result; | |
1737 | } | |
1738 | ||
1739 | current_length = st.st_size; | |
d57c44d0 | 1740 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1741 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1742 | return -ENOTSUP; | |
1743 | } | |
1744 | ||
d57c44d0 | 1745 | switch (prealloc) { |
93f4e2ff KW |
1746 | #ifdef CONFIG_POSIX_FALLOCATE |
1747 | case PREALLOC_MODE_FALLOC: | |
1748 | /* | |
1749 | * Truncating before posix_fallocate() makes it about twice slower on | |
1750 | * file systems that do not support fallocate(), trying to check if a | |
1751 | * block is allocated before allocating it, so don't do that here. | |
1752 | */ | |
1753 | if (offset != current_length) { | |
1754 | result = -posix_fallocate(fd, current_length, | |
1755 | offset - current_length); | |
1756 | if (result != 0) { | |
1757 | /* posix_fallocate() doesn't set errno. */ | |
1758 | error_setg_errno(errp, -result, | |
1759 | "Could not preallocate new data"); | |
3a20013f NS |
1760 | } else if (current_length == 0) { |
1761 | /* | |
1762 | * posix_fallocate() uses fallocate() if the filesystem | |
1763 | * supports it, or fallback to manually writing zeroes. If | |
1764 | * fallocate() was used, unaligned reads from the fallocated | |
1765 | * area in raw_probe_alignment() will succeed, hence we need to | |
1766 | * allocate the first block. | |
1767 | * | |
1768 | * Optimize future alignment probing; ignore failures. | |
1769 | */ | |
1770 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1771 | } |
1772 | } else { | |
1773 | result = 0; | |
1774 | } | |
1775 | goto out; | |
1776 | #endif | |
1777 | case PREALLOC_MODE_FULL: | |
1778 | { | |
1779 | int64_t num = 0, left = offset - current_length; | |
1780 | off_t seek_result; | |
1781 | ||
1782 | /* | |
1783 | * Knowing the final size from the beginning could allow the file | |
1784 | * system driver to do less allocations and possibly avoid | |
1785 | * fragmentation of the file. | |
1786 | */ | |
1787 | if (ftruncate(fd, offset) != 0) { | |
1788 | result = -errno; | |
1789 | error_setg_errno(errp, -result, "Could not resize file"); | |
1790 | goto out; | |
1791 | } | |
1792 | ||
1793 | buf = g_malloc0(65536); | |
1794 | ||
1795 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1796 | if (seek_result < 0) { | |
1797 | result = -errno; | |
1798 | error_setg_errno(errp, -result, | |
1799 | "Failed to seek to the old end of file"); | |
1800 | goto out; | |
1801 | } | |
1802 | ||
1803 | while (left > 0) { | |
1804 | num = MIN(left, 65536); | |
1805 | result = write(fd, buf, num); | |
1806 | if (result < 0) { | |
a1c81f4f FZ |
1807 | if (errno == EINTR) { |
1808 | continue; | |
1809 | } | |
93f4e2ff KW |
1810 | result = -errno; |
1811 | error_setg_errno(errp, -result, | |
1812 | "Could not write zeros for preallocation"); | |
1813 | goto out; | |
1814 | } | |
1815 | left -= result; | |
1816 | } | |
1817 | if (result >= 0) { | |
1818 | result = fsync(fd); | |
1819 | if (result < 0) { | |
1820 | result = -errno; | |
1821 | error_setg_errno(errp, -result, | |
1822 | "Could not flush file to disk"); | |
1823 | goto out; | |
1824 | } | |
1825 | } | |
1826 | goto out; | |
1827 | } | |
1828 | case PREALLOC_MODE_OFF: | |
1829 | if (ftruncate(fd, offset) != 0) { | |
1830 | result = -errno; | |
1831 | error_setg_errno(errp, -result, "Could not resize file"); | |
3a20013f NS |
1832 | } else if (current_length == 0 && offset > current_length) { |
1833 | /* Optimize future alignment probing; ignore failures. */ | |
1834 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1835 | } |
1836 | return result; | |
1837 | default: | |
1838 | result = -ENOTSUP; | |
1839 | error_setg(errp, "Unsupported preallocation mode: %s", | |
d57c44d0 | 1840 | PreallocMode_str(prealloc)); |
93f4e2ff KW |
1841 | return result; |
1842 | } | |
1843 | ||
1844 | out: | |
1845 | if (result < 0) { | |
1846 | if (ftruncate(fd, current_length) < 0) { | |
1847 | error_report("Failed to restore old file length: %s", | |
1848 | strerror(errno)); | |
1849 | } | |
1850 | } | |
1851 | ||
1852 | g_free(buf); | |
1853 | return result; | |
1854 | } | |
1855 | ||
5d5de250 KW |
1856 | static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, |
1857 | ThreadPoolFunc func, void *arg) | |
1858 | { | |
1859 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ | |
1860 | ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
1861 | return thread_pool_submit_co(pool, func, arg); | |
1862 | } | |
1863 | ||
9d52aa3c KW |
1864 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
1865 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 1866 | { |
ce1a14dc | 1867 | BDRVRawState *s = bs->opaque; |
999e6b69 | 1868 | RawPosixAIOData acb; |
ce1a14dc | 1869 | |
19cb3738 | 1870 | if (fd_open(bs) < 0) |
2174f12b | 1871 | return -EIO; |
19cb3738 | 1872 | |
f141eafe | 1873 | /* |
3cad8307 RPM |
1874 | * Check if the underlying device requires requests to be aligned, |
1875 | * and if the request we are trying to submit is aligned or not. | |
1876 | * If this is the case tell the low-level driver that it needs | |
1877 | * to copy the buffer. | |
f141eafe | 1878 | */ |
3cad8307 | 1879 | if (s->needs_alignment) { |
c53b1c51 | 1880 | if (!bdrv_qiov_is_aligned(bs, qiov)) { |
5c6c3a6c | 1881 | type |= QEMU_AIO_MISALIGNED; |
e44bd6fc | 1882 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 1883 | } else if (s->use_linux_aio) { |
0187f5c9 | 1884 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
9d52aa3c | 1885 | assert(qiov->size == bytes); |
0187f5c9 | 1886 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type); |
e44bd6fc | 1887 | #endif |
5c6c3a6c | 1888 | } |
9ef91a67 | 1889 | } |
f141eafe | 1890 | |
999e6b69 KW |
1891 | acb = (RawPosixAIOData) { |
1892 | .bs = bs, | |
1893 | .aio_fildes = s->fd, | |
1894 | .aio_type = type, | |
1895 | .aio_offset = offset, | |
1896 | .aio_nbytes = bytes, | |
1897 | .io = { | |
1898 | .iov = qiov->iov, | |
1899 | .niov = qiov->niov, | |
1900 | }, | |
1901 | }; | |
1902 | ||
1903 | assert(qiov->size == bytes); | |
1904 | return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); | |
2174f12b KW |
1905 | } |
1906 | ||
9d52aa3c KW |
1907 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, |
1908 | uint64_t bytes, QEMUIOVector *qiov, | |
1909 | int flags) | |
2174f12b | 1910 | { |
9d52aa3c | 1911 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
1912 | } |
1913 | ||
9d52aa3c KW |
1914 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, |
1915 | uint64_t bytes, QEMUIOVector *qiov, | |
1916 | int flags) | |
2174f12b | 1917 | { |
9d52aa3c KW |
1918 | assert(flags == 0); |
1919 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
1920 | } |
1921 | ||
1b3abdcc ML |
1922 | static void raw_aio_plug(BlockDriverState *bs) |
1923 | { | |
1924 | #ifdef CONFIG_LINUX_AIO | |
0a4279d9 KW |
1925 | BDRVRawState *s = bs->opaque; |
1926 | if (s->use_linux_aio) { | |
0187f5c9 PB |
1927 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1928 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
1929 | } |
1930 | #endif | |
1931 | } | |
1932 | ||
1933 | static void raw_aio_unplug(BlockDriverState *bs) | |
1934 | { | |
1935 | #ifdef CONFIG_LINUX_AIO | |
0a4279d9 KW |
1936 | BDRVRawState *s = bs->opaque; |
1937 | if (s->use_linux_aio) { | |
0187f5c9 PB |
1938 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1939 | laio_io_unplug(bs, aio); | |
1b3abdcc ML |
1940 | } |
1941 | #endif | |
1942 | } | |
1943 | ||
33d70fb6 | 1944 | static int raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
1945 | { |
1946 | BDRVRawState *s = bs->opaque; | |
06dc9bd5 | 1947 | RawPosixAIOData acb; |
33d70fb6 | 1948 | int ret; |
b2e12bc6 | 1949 | |
33d70fb6 KW |
1950 | ret = fd_open(bs); |
1951 | if (ret < 0) { | |
1952 | return ret; | |
1953 | } | |
b2e12bc6 | 1954 | |
06dc9bd5 KW |
1955 | acb = (RawPosixAIOData) { |
1956 | .bs = bs, | |
1957 | .aio_fildes = s->fd, | |
1958 | .aio_type = QEMU_AIO_FLUSH, | |
1959 | }; | |
1960 | ||
1961 | return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); | |
b2e12bc6 CH |
1962 | } |
1963 | ||
ed6e2161 NA |
1964 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
1965 | AioContext *new_context) | |
1966 | { | |
1967 | #ifdef CONFIG_LINUX_AIO | |
1968 | BDRVRawState *s = bs->opaque; | |
1969 | if (s->use_linux_aio) { | |
1970 | Error *local_err; | |
1971 | if (!aio_setup_linux_aio(new_context, &local_err)) { | |
1972 | error_reportf_err(local_err, "Unable to use native AIO, " | |
1973 | "falling back to thread pool: "); | |
1974 | s->use_linux_aio = false; | |
1975 | } | |
1976 | } | |
1977 | #endif | |
1978 | } | |
1979 | ||
83f64091 FB |
1980 | static void raw_close(BlockDriverState *bs) |
1981 | { | |
1982 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 1983 | |
19cb3738 | 1984 | if (s->fd >= 0) { |
2e1e79da | 1985 | qemu_close(s->fd); |
19cb3738 FB |
1986 | s->fd = -1; |
1987 | } | |
83f64091 FB |
1988 | } |
1989 | ||
d0bc9e5d HR |
1990 | /** |
1991 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
1992 | * new space according to @prealloc. | |
1993 | * | |
1994 | * Returns: 0 on success, -errno on failure. | |
1995 | */ | |
93f4e2ff KW |
1996 | static int coroutine_fn |
1997 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
1998 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 1999 | { |
29cb4c01 | 2000 | RawPosixAIOData acb; |
d0bc9e5d | 2001 | |
29cb4c01 | 2002 | acb = (RawPosixAIOData) { |
93f4e2ff KW |
2003 | .bs = bs, |
2004 | .aio_fildes = fd, | |
2005 | .aio_type = QEMU_AIO_TRUNCATE, | |
2006 | .aio_offset = offset, | |
d57c44d0 KW |
2007 | .truncate = { |
2008 | .prealloc = prealloc, | |
2009 | .errp = errp, | |
2010 | }, | |
93f4e2ff | 2011 | }; |
d0bc9e5d | 2012 | |
29cb4c01 | 2013 | return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); |
9f63b07e HR |
2014 | } |
2015 | ||
061ca8a3 KW |
2016 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
2017 | PreallocMode prealloc, Error **errp) | |
83f64091 FB |
2018 | { |
2019 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 2020 | struct stat st; |
f59adb32 | 2021 | int ret; |
55b949c8 CH |
2022 | |
2023 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
2024 | ret = -errno; |
2025 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
2026 | return ret; | |
55b949c8 CH |
2027 | } |
2028 | ||
2029 | if (S_ISREG(st.st_mode)) { | |
93f4e2ff | 2030 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
2031 | } |
2032 | ||
2033 | if (prealloc != PREALLOC_MODE_OFF) { | |
2034 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 2035 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
2036 | return -ENOTSUP; |
2037 | } | |
2038 | ||
2039 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
f59adb32 HR |
2040 | if (offset > raw_getlength(bs)) { |
2041 | error_setg(errp, "Cannot grow device files"); | |
2042 | return -EINVAL; | |
2043 | } | |
55b949c8 | 2044 | } else { |
f59adb32 | 2045 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
2046 | return -ENOTSUP; |
2047 | } | |
2048 | ||
83f64091 FB |
2049 | return 0; |
2050 | } | |
2051 | ||
128ab2ff BS |
2052 | #ifdef __OpenBSD__ |
2053 | static int64_t raw_getlength(BlockDriverState *bs) | |
2054 | { | |
2055 | BDRVRawState *s = bs->opaque; | |
2056 | int fd = s->fd; | |
2057 | struct stat st; | |
2058 | ||
2059 | if (fstat(fd, &st)) | |
aa729704 | 2060 | return -errno; |
128ab2ff BS |
2061 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2062 | struct disklabel dl; | |
2063 | ||
2064 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2065 | return -errno; |
128ab2ff BS |
2066 | return (uint64_t)dl.d_secsize * |
2067 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2068 | } else | |
2069 | return st.st_size; | |
2070 | } | |
d1f6fd8d CE |
2071 | #elif defined(__NetBSD__) |
2072 | static int64_t raw_getlength(BlockDriverState *bs) | |
2073 | { | |
2074 | BDRVRawState *s = bs->opaque; | |
2075 | int fd = s->fd; | |
2076 | struct stat st; | |
2077 | ||
2078 | if (fstat(fd, &st)) | |
aa729704 | 2079 | return -errno; |
d1f6fd8d CE |
2080 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2081 | struct dkwedge_info dkw; | |
2082 | ||
2083 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
2084 | return dkw.dkw_size * 512; | |
2085 | } else { | |
2086 | struct disklabel dl; | |
2087 | ||
2088 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2089 | return -errno; |
d1f6fd8d CE |
2090 | return (uint64_t)dl.d_secsize * |
2091 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2092 | } | |
2093 | } else | |
2094 | return st.st_size; | |
2095 | } | |
50779cc2 CH |
2096 | #elif defined(__sun__) |
2097 | static int64_t raw_getlength(BlockDriverState *bs) | |
2098 | { | |
2099 | BDRVRawState *s = bs->opaque; | |
2100 | struct dk_minfo minfo; | |
2101 | int ret; | |
aa729704 | 2102 | int64_t size; |
50779cc2 CH |
2103 | |
2104 | ret = fd_open(bs); | |
2105 | if (ret < 0) { | |
2106 | return ret; | |
2107 | } | |
2108 | ||
2109 | /* | |
2110 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
2111 | */ | |
2112 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
2113 | if (ret != -1) { | |
2114 | return minfo.dki_lbsize * minfo.dki_capacity; | |
2115 | } | |
2116 | ||
2117 | /* | |
2118 | * There are reports that lseek on some devices fails, but | |
2119 | * irc discussion said that contingency on contingency was overkill. | |
2120 | */ | |
aa729704 MA |
2121 | size = lseek(s->fd, 0, SEEK_END); |
2122 | if (size < 0) { | |
2123 | return -errno; | |
2124 | } | |
2125 | return size; | |
50779cc2 CH |
2126 | } |
2127 | #elif defined(CONFIG_BSD) | |
2128 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2129 | { |
2130 | BDRVRawState *s = bs->opaque; | |
2131 | int fd = s->fd; | |
2132 | int64_t size; | |
83f64091 | 2133 | struct stat sb; |
a167ba50 | 2134 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2135 | int reopened = 0; |
83f64091 | 2136 | #endif |
19cb3738 FB |
2137 | int ret; |
2138 | ||
2139 | ret = fd_open(bs); | |
2140 | if (ret < 0) | |
2141 | return ret; | |
83f64091 | 2142 | |
a167ba50 | 2143 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2144 | again: |
2145 | #endif | |
83f64091 FB |
2146 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
2147 | #ifdef DIOCGMEDIASIZE | |
7d37435b | 2148 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) |
c5e97233 BS |
2149 | #elif defined(DIOCGPART) |
2150 | { | |
2151 | struct partinfo pi; | |
2152 | if (ioctl(fd, DIOCGPART, &pi) == 0) | |
2153 | size = pi.media_size; | |
2154 | else | |
2155 | size = 0; | |
2156 | } | |
2157 | if (size == 0) | |
83f64091 | 2158 | #endif |
83affaa6 | 2159 | #if defined(__APPLE__) && defined(__MACH__) |
728dacbd JA |
2160 | { |
2161 | uint64_t sectors = 0; | |
2162 | uint32_t sector_size = 0; | |
2163 | ||
2164 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2165 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2166 | size = sectors * sector_size; | |
2167 | } else { | |
2168 | size = lseek(fd, 0LL, SEEK_END); | |
2169 | if (size < 0) { | |
2170 | return -errno; | |
2171 | } | |
2172 | } | |
2173 | } | |
83f64091 FB |
2174 | #else |
2175 | size = lseek(fd, 0LL, SEEK_END); | |
aa729704 MA |
2176 | if (size < 0) { |
2177 | return -errno; | |
2178 | } | |
9f23011a | 2179 | #endif |
a167ba50 | 2180 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2181 | switch(s->type) { |
2182 | case FTYPE_CD: | |
2183 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2184 | if (size == 2048LL * (unsigned)-1) | |
2185 | size = 0; | |
2186 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2187 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2188 | reopened = 1; |
2189 | goto again; | |
2190 | } | |
2191 | } | |
83f64091 | 2192 | #endif |
50779cc2 | 2193 | } else { |
83f64091 | 2194 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2195 | if (size < 0) { |
2196 | return -errno; | |
2197 | } | |
83f64091 | 2198 | } |
83f64091 FB |
2199 | return size; |
2200 | } | |
50779cc2 CH |
2201 | #else |
2202 | static int64_t raw_getlength(BlockDriverState *bs) | |
2203 | { | |
2204 | BDRVRawState *s = bs->opaque; | |
2205 | int ret; | |
aa729704 | 2206 | int64_t size; |
50779cc2 CH |
2207 | |
2208 | ret = fd_open(bs); | |
2209 | if (ret < 0) { | |
2210 | return ret; | |
2211 | } | |
2212 | ||
aa729704 MA |
2213 | size = lseek(s->fd, 0, SEEK_END); |
2214 | if (size < 0) { | |
2215 | return -errno; | |
2216 | } | |
2217 | return size; | |
50779cc2 | 2218 | } |
128ab2ff | 2219 | #endif |
83f64091 | 2220 | |
4a1d5e1f FZ |
2221 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2222 | { | |
2223 | struct stat st; | |
2224 | BDRVRawState *s = bs->opaque; | |
2225 | ||
2226 | if (fstat(s->fd, &st) < 0) { | |
2227 | return -errno; | |
2228 | } | |
2229 | return (int64_t)st.st_blocks * 512; | |
2230 | } | |
2231 | ||
93f4e2ff KW |
2232 | static int coroutine_fn |
2233 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2234 | { |
927f11e1 | 2235 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2236 | Error *local_err = NULL; |
83f64091 | 2237 | int fd; |
d815efca | 2238 | uint64_t perm, shared; |
1e37d059 | 2239 | int result = 0; |
83f64091 | 2240 | |
927f11e1 KW |
2241 | /* Validate options and set default values */ |
2242 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2243 | file_opts = &options->u.file; | |
464d9f64 | 2244 | |
927f11e1 KW |
2245 | if (!file_opts->has_nocow) { |
2246 | file_opts->nocow = false; | |
2247 | } | |
2248 | if (!file_opts->has_preallocation) { | |
2249 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2250 | } |
83f64091 | 2251 | |
927f11e1 | 2252 | /* Create file */ |
b8cf1913 | 2253 | fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); |
1e37d059 SW |
2254 | if (fd < 0) { |
2255 | result = -errno; | |
e428e439 | 2256 | error_setg_errno(errp, -result, "Could not create file"); |
06247428 HT |
2257 | goto out; |
2258 | } | |
2259 | ||
b8cf1913 HR |
2260 | /* Take permissions: We want to discard everything, so we need |
2261 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2262 | * BLK_PERM_RESIZE. | |
2263 | * On the other hand, we cannot share the RESIZE permission | |
2264 | * because we promise that after this function, the file has the | |
2265 | * size given in the options. If someone else were to resize it | |
2266 | * concurrently, we could not guarantee that. | |
2267 | * Note that after this function, we can no longer guarantee that | |
2268 | * the file is not touched by a third party, so it may be resized | |
2269 | * then. */ | |
2270 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2271 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2272 | ||
2273 | /* Step one: Take locks */ | |
2996ffad | 2274 | result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2275 | if (result < 0) { |
2276 | goto out_close; | |
2277 | } | |
2278 | ||
2279 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2280 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2281 | if (result < 0) { | |
b857431d FZ |
2282 | error_append_hint(errp, |
2283 | "Is another process using the image [%s]?\n", | |
2284 | file_opts->filename); | |
7c20c808 | 2285 | goto out_unlock; |
b8cf1913 HR |
2286 | } |
2287 | ||
2288 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2289 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2290 | if (result < 0) { |
7c20c808 | 2291 | goto out_unlock; |
b8cf1913 HR |
2292 | } |
2293 | ||
927f11e1 | 2294 | if (file_opts->nocow) { |
4ab15590 | 2295 | #ifdef __linux__ |
06247428 HT |
2296 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2297 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2298 | * will be ignored since any failure of this operation should not | |
2299 | * block the left work. | |
2300 | */ | |
2301 | int attr; | |
2302 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2303 | attr |= FS_NOCOW_FL; | |
2304 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2305 | } |
06247428 HT |
2306 | #endif |
2307 | } | |
2308 | ||
b8cf1913 HR |
2309 | /* Resize and potentially preallocate the file to the desired |
2310 | * final size */ | |
93f4e2ff KW |
2311 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2312 | file_opts->preallocation, errp); | |
9f63b07e | 2313 | if (result < 0) { |
7c20c808 HR |
2314 | goto out_unlock; |
2315 | } | |
2316 | ||
2317 | out_unlock: | |
2996ffad | 2318 | raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); |
7c20c808 HR |
2319 | if (local_err) { |
2320 | /* The above call should not fail, and if it does, that does | |
2321 | * not mean the whole creation operation has failed. So | |
2322 | * report it the user for their convenience, but do not report | |
2323 | * it to the caller. */ | |
db0754df | 2324 | warn_report_err(local_err); |
1e37d059 | 2325 | } |
06247428 | 2326 | |
5a1dad9d | 2327 | out_close: |
06247428 HT |
2328 | if (qemu_close(fd) != 0 && result == 0) { |
2329 | result = -errno; | |
2330 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2331 | } | |
2332 | out: | |
1e37d059 | 2333 | return result; |
83f64091 FB |
2334 | } |
2335 | ||
927f11e1 KW |
2336 | static int coroutine_fn raw_co_create_opts(const char *filename, QemuOpts *opts, |
2337 | Error **errp) | |
2338 | { | |
2339 | BlockdevCreateOptions options; | |
2340 | int64_t total_size = 0; | |
2341 | bool nocow = false; | |
2342 | PreallocMode prealloc; | |
2343 | char *buf = NULL; | |
2344 | Error *local_err = NULL; | |
2345 | ||
2346 | /* Skip file: protocol prefix */ | |
2347 | strstart(filename, "file:", &filename); | |
2348 | ||
2349 | /* Read out options */ | |
2350 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2351 | BDRV_SECTOR_SIZE); | |
2352 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); | |
2353 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2354 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2355 | PREALLOC_MODE_OFF, &local_err); | |
2356 | g_free(buf); | |
2357 | if (local_err) { | |
2358 | error_propagate(errp, local_err); | |
2359 | return -EINVAL; | |
2360 | } | |
2361 | ||
2362 | options = (BlockdevCreateOptions) { | |
2363 | .driver = BLOCKDEV_DRIVER_FILE, | |
2364 | .u.file = { | |
2365 | .filename = (char *) filename, | |
2366 | .size = total_size, | |
2367 | .has_preallocation = true, | |
2368 | .preallocation = prealloc, | |
2369 | .has_nocow = true, | |
2370 | .nocow = nocow, | |
2371 | }, | |
2372 | }; | |
2373 | return raw_co_create(&options, errp); | |
2374 | } | |
2375 | ||
d1f06fe6 MA |
2376 | /* |
2377 | * Find allocation range in @bs around offset @start. | |
2378 | * May change underlying file descriptor's file offset. | |
2379 | * If @start is not in a hole, store @start in @data, and the | |
2380 | * beginning of the next hole in @hole, and return 0. | |
2381 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2382 | * beginning of the next non-hole in @data, and return 0. | |
2383 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2384 | * If we can't find out, return a negative errno other than -ENXIO. | |
2385 | */ | |
2386 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2387 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2388 | { |
2389 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2390 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2391 | off_t offs; |
94282e71 | 2392 | |
d1f06fe6 MA |
2393 | /* |
2394 | * SEEK_DATA cases: | |
2395 | * D1. offs == start: start is in data | |
2396 | * D2. offs > start: start is in a hole, next data at offs | |
2397 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2398 | * or start is beyond EOF | |
2399 | * If the latter happens, the file has been truncated behind | |
2400 | * our back since we opened it. All bets are off then. | |
2401 | * Treating like a trailing hole is simplest. | |
2402 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2403 | */ | |
2404 | offs = lseek(s->fd, start, SEEK_DATA); | |
2405 | if (offs < 0) { | |
2406 | return -errno; /* D3 or D4 */ | |
2407 | } | |
a03083a0 JC |
2408 | |
2409 | if (offs < start) { | |
2410 | /* This is not a valid return by lseek(). We are safe to just return | |
2411 | * -EIO in this case, and we'll treat it like D4. */ | |
2412 | return -EIO; | |
2413 | } | |
d1f06fe6 MA |
2414 | |
2415 | if (offs > start) { | |
2416 | /* D2: in hole, next data at offs */ | |
2417 | *hole = start; | |
2418 | *data = offs; | |
2419 | return 0; | |
5500316d PB |
2420 | } |
2421 | ||
d1f06fe6 MA |
2422 | /* D1: in data, end not yet known */ |
2423 | ||
2424 | /* | |
2425 | * SEEK_HOLE cases: | |
2426 | * H1. offs == start: start is in a hole | |
2427 | * If this happens here, a hole has been dug behind our back | |
2428 | * since the previous lseek(). | |
2429 | * H2. offs > start: either start is in data, next hole at offs, | |
2430 | * or start is in trailing hole, EOF at offs | |
2431 | * Linux treats trailing holes like any other hole: offs == | |
2432 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2433 | * If that happens here, a hole has been dug behind our back | |
2434 | * since the previous lseek(). | |
2435 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2436 | * If this happens, the file has been truncated behind our | |
2437 | * back since we opened it. Treat it like a trailing hole. | |
2438 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2439 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2440 | */ | |
2441 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2442 | if (offs < 0) { | |
2443 | return -errno; /* D1 and (H3 or H4) */ | |
2444 | } | |
a03083a0 JC |
2445 | |
2446 | if (offs < start) { | |
2447 | /* This is not a valid return by lseek(). We are safe to just return | |
2448 | * -EIO in this case, and we'll treat it like H4. */ | |
2449 | return -EIO; | |
2450 | } | |
d1f06fe6 MA |
2451 | |
2452 | if (offs > start) { | |
2453 | /* | |
2454 | * D1 and H2: either in data, next hole at offs, or it was in | |
2455 | * data but is now in a trailing hole. In the latter case, | |
2456 | * all bets are off. Treating it as if it there was data all | |
2457 | * the way to EOF is safe, so simply do that. | |
2458 | */ | |
4f11aa8a | 2459 | *data = start; |
d1f06fe6 MA |
2460 | *hole = offs; |
2461 | return 0; | |
5500316d | 2462 | } |
4f11aa8a | 2463 | |
d1f06fe6 MA |
2464 | /* D1 and H1 */ |
2465 | return -EBUSY; | |
5500316d | 2466 | #else |
4f11aa8a | 2467 | return -ENOTSUP; |
5500316d | 2468 | #endif |
4f11aa8a HR |
2469 | } |
2470 | ||
2471 | /* | |
a290f085 | 2472 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2473 | * |
a290f085 | 2474 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2475 | * |
a290f085 EB |
2476 | * 'pnum' is set to the number of bytes (including and immediately following |
2477 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2478 | * allocated/unallocated state. |
2479 | * | |
a290f085 | 2480 | * 'bytes' is the max value 'pnum' should be set to. |
4f11aa8a | 2481 | */ |
a290f085 EB |
2482 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2483 | bool want_zero, | |
2484 | int64_t offset, | |
2485 | int64_t bytes, int64_t *pnum, | |
2486 | int64_t *map, | |
2487 | BlockDriverState **file) | |
2488 | { | |
2489 | off_t data = 0, hole = 0; | |
d7f62751 | 2490 | int ret; |
4f11aa8a | 2491 | |
9c3db310 HR |
2492 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); |
2493 | ||
4f11aa8a HR |
2494 | ret = fd_open(bs); |
2495 | if (ret < 0) { | |
2496 | return ret; | |
2497 | } | |
2498 | ||
a290f085 EB |
2499 | if (!want_zero) { |
2500 | *pnum = bytes; | |
2501 | *map = offset; | |
2502 | *file = bs; | |
2503 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2504 | } |
4f11aa8a | 2505 | |
a290f085 | 2506 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2507 | if (ret == -ENXIO) { |
2508 | /* Trailing hole */ | |
a290f085 | 2509 | *pnum = bytes; |
d1f06fe6 MA |
2510 | ret = BDRV_BLOCK_ZERO; |
2511 | } else if (ret < 0) { | |
2512 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2513 | *pnum = bytes; |
d1f06fe6 | 2514 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2515 | } else if (data == offset) { |
2516 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2517 | * possibly including a partial sector at EOF. */ |
a290f085 | 2518 | *pnum = MIN(bytes, hole - offset); |
9c3db310 HR |
2519 | |
2520 | /* | |
2521 | * We are not allowed to return partial sectors, though, so | |
2522 | * round up if necessary. | |
2523 | */ | |
2524 | if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { | |
2525 | int64_t file_length = raw_getlength(bs); | |
2526 | if (file_length > 0) { | |
2527 | /* Ignore errors, this is just a safeguard */ | |
2528 | assert(hole == file_length); | |
2529 | } | |
2530 | *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); | |
2531 | } | |
2532 | ||
d1f06fe6 | 2533 | ret = BDRV_BLOCK_DATA; |
5500316d | 2534 | } else { |
a290f085 EB |
2535 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2536 | assert(hole == offset); | |
2537 | *pnum = MIN(bytes, data - offset); | |
d1f06fe6 | 2538 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2539 | } |
a290f085 | 2540 | *map = offset; |
02650acb | 2541 | *file = bs; |
a290f085 | 2542 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2543 | } |
2544 | ||
31be8a2a SH |
2545 | #if defined(__linux__) |
2546 | /* Verify that the file is not in the page cache */ | |
2547 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2548 | { | |
2549 | const size_t window_size = 128 * 1024 * 1024; | |
2550 | BDRVRawState *s = bs->opaque; | |
2551 | void *window = NULL; | |
2552 | size_t length = 0; | |
2553 | unsigned char *vec; | |
2554 | size_t page_size; | |
2555 | off_t offset; | |
2556 | off_t end; | |
2557 | ||
2558 | /* mincore(2) page status information requires 1 byte per page */ | |
2559 | page_size = sysconf(_SC_PAGESIZE); | |
2560 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2561 | ||
2562 | end = raw_getlength(bs); | |
2563 | ||
2564 | for (offset = 0; offset < end; offset += window_size) { | |
2565 | void *new_window; | |
2566 | size_t new_length; | |
2567 | size_t vec_end; | |
2568 | size_t i; | |
2569 | int ret; | |
2570 | ||
2571 | /* Unmap previous window if size has changed */ | |
2572 | new_length = MIN(end - offset, window_size); | |
2573 | if (new_length != length) { | |
2574 | munmap(window, length); | |
2575 | window = NULL; | |
2576 | length = 0; | |
2577 | } | |
2578 | ||
2579 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2580 | s->fd, offset); | |
2581 | if (new_window == MAP_FAILED) { | |
2582 | error_setg_errno(errp, errno, "mmap failed"); | |
2583 | break; | |
2584 | } | |
2585 | ||
2586 | window = new_window; | |
2587 | length = new_length; | |
2588 | ||
2589 | ret = mincore(window, length, vec); | |
2590 | if (ret < 0) { | |
2591 | error_setg_errno(errp, errno, "mincore failed"); | |
2592 | break; | |
2593 | } | |
2594 | ||
2595 | vec_end = DIV_ROUND_UP(length, page_size); | |
2596 | for (i = 0; i < vec_end; i++) { | |
2597 | if (vec[i] & 0x1) { | |
2598 | error_setg(errp, "page cache still in use!"); | |
2599 | break; | |
2600 | } | |
2601 | } | |
2602 | } | |
2603 | ||
2604 | if (window) { | |
2605 | munmap(window, length); | |
2606 | } | |
2607 | ||
2608 | g_free(vec); | |
2609 | } | |
2610 | #endif /* __linux__ */ | |
2611 | ||
dd577a26 SH |
2612 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2613 | Error **errp) | |
2614 | { | |
2615 | BDRVRawState *s = bs->opaque; | |
2616 | int ret; | |
2617 | ||
2618 | ret = fd_open(bs); | |
2619 | if (ret < 0) { | |
2620 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2621 | return; | |
2622 | } | |
2623 | ||
f357fcd8 SH |
2624 | if (!s->drop_cache) { |
2625 | return; | |
2626 | } | |
2627 | ||
dd577a26 SH |
2628 | if (s->open_flags & O_DIRECT) { |
2629 | return; /* No host kernel page cache */ | |
2630 | } | |
2631 | ||
2632 | #if defined(__linux__) | |
2633 | /* This sets the scene for the next syscall... */ | |
2634 | ret = bdrv_co_flush(bs); | |
2635 | if (ret < 0) { | |
2636 | error_setg_errno(errp, -ret, "flush failed"); | |
2637 | return; | |
2638 | } | |
2639 | ||
2640 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2641 | * process. These limitations are okay because we just fsynced the file, | |
2642 | * we don't use mmap, and the file should not be in use by other processes. | |
2643 | */ | |
2644 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2645 | if (ret != 0) { /* the return value is a positive errno */ | |
2646 | error_setg_errno(errp, ret, "fadvise failed"); | |
2647 | return; | |
2648 | } | |
31be8a2a SH |
2649 | |
2650 | if (s->check_cache_dropped) { | |
2651 | check_cache_dropped(bs, errp); | |
2652 | } | |
dd577a26 SH |
2653 | #else /* __linux__ */ |
2654 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2655 | * unsupported on other host operating systems. Cache consistency issues | |
2656 | * may occur but no error is reported here, partly because that's the | |
2657 | * historical behavior and partly because it's hard to differentiate valid | |
2658 | * configurations that should not cause errors. | |
2659 | */ | |
2660 | #endif /* !__linux__ */ | |
2661 | } | |
2662 | ||
33d70fb6 | 2663 | static coroutine_fn int |
46ee0f46 | 2664 | raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) |
dce512de | 2665 | { |
dce512de | 2666 | BDRVRawState *s = bs->opaque; |
46ee0f46 KW |
2667 | RawPosixAIOData acb; |
2668 | ||
2669 | acb = (RawPosixAIOData) { | |
2670 | .bs = bs, | |
2671 | .aio_fildes = s->fd, | |
2672 | .aio_type = QEMU_AIO_DISCARD, | |
2673 | .aio_offset = offset, | |
2674 | .aio_nbytes = bytes, | |
2675 | }; | |
dce512de | 2676 | |
46ee0f46 KW |
2677 | if (blkdev) { |
2678 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2679 | } | |
2680 | ||
2681 | return raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); | |
2682 | } | |
2683 | ||
2684 | static coroutine_fn int | |
2685 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
2686 | { | |
2687 | return raw_do_pdiscard(bs, offset, bytes, false); | |
dce512de | 2688 | } |
0e7e1989 | 2689 | |
7154d8ae KW |
2690 | static int coroutine_fn |
2691 | raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, | |
2692 | BdrvRequestFlags flags, bool blkdev) | |
260a82e5 PB |
2693 | { |
2694 | BDRVRawState *s = bs->opaque; | |
7154d8ae KW |
2695 | RawPosixAIOData acb; |
2696 | ThreadPoolFunc *handler; | |
2697 | ||
2698 | acb = (RawPosixAIOData) { | |
2699 | .bs = bs, | |
2700 | .aio_fildes = s->fd, | |
2701 | .aio_type = QEMU_AIO_WRITE_ZEROES, | |
2702 | .aio_offset = offset, | |
2703 | .aio_nbytes = bytes, | |
2704 | }; | |
2705 | ||
2706 | if (blkdev) { | |
2707 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2708 | } | |
738301e1 KW |
2709 | if (flags & BDRV_REQ_NO_FALLBACK) { |
2710 | acb.aio_type |= QEMU_AIO_NO_FALLBACK; | |
2711 | } | |
260a82e5 | 2712 | |
34fa110e | 2713 | if (flags & BDRV_REQ_MAY_UNMAP) { |
7154d8ae KW |
2714 | acb.aio_type |= QEMU_AIO_DISCARD; |
2715 | handler = handle_aiocb_write_zeroes_unmap; | |
2716 | } else { | |
2717 | handler = handle_aiocb_write_zeroes; | |
260a82e5 | 2718 | } |
34fa110e | 2719 | |
7154d8ae KW |
2720 | return raw_thread_pool_submit(bs, handler, &acb); |
2721 | } | |
2722 | ||
2723 | static int coroutine_fn raw_co_pwrite_zeroes( | |
2724 | BlockDriverState *bs, int64_t offset, | |
2725 | int bytes, BdrvRequestFlags flags) | |
2726 | { | |
2727 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); | |
260a82e5 PB |
2728 | } |
2729 | ||
2730 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
2731 | { | |
2732 | BDRVRawState *s = bs->opaque; | |
2733 | ||
2734 | bdi->unallocated_blocks_are_zero = s->discard_zeroes; | |
260a82e5 PB |
2735 | return 0; |
2736 | } | |
2737 | ||
6f482f74 CL |
2738 | static QemuOptsList raw_create_opts = { |
2739 | .name = "raw-create-opts", | |
2740 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
2741 | .desc = { | |
2742 | { | |
2743 | .name = BLOCK_OPT_SIZE, | |
2744 | .type = QEMU_OPT_SIZE, | |
2745 | .help = "Virtual disk size" | |
2746 | }, | |
4ab15590 CL |
2747 | { |
2748 | .name = BLOCK_OPT_NOCOW, | |
2749 | .type = QEMU_OPT_BOOL, | |
2750 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
2751 | }, | |
06247428 HT |
2752 | { |
2753 | .name = BLOCK_OPT_PREALLOC, | |
2754 | .type = QEMU_OPT_STRING, | |
abea0053 SG |
2755 | .help = "Preallocation mode (allowed values: off" |
2756 | #ifdef CONFIG_POSIX_FALLOCATE | |
2757 | ", falloc" | |
2758 | #endif | |
2759 | ", full)" | |
06247428 | 2760 | }, |
6f482f74 CL |
2761 | { /* end of list */ } |
2762 | } | |
0e7e1989 KW |
2763 | }; |
2764 | ||
244a5668 FZ |
2765 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
2766 | Error **errp) | |
2767 | { | |
6ceabe6f KW |
2768 | BDRVRawState *s = bs->opaque; |
2769 | BDRVRawReopenState *rs = NULL; | |
2770 | int open_flags; | |
2771 | int ret; | |
2772 | ||
2773 | if (s->perm_change_fd) { | |
2774 | /* | |
2775 | * In the context of reopen, this function may be called several times | |
2776 | * (directly and recursively while change permissions of the parent). | |
2777 | * This is even true for children that don't inherit from the original | |
2778 | * reopen node, so s->reopen_state is not set. | |
2779 | * | |
2780 | * Ignore all but the first call. | |
2781 | */ | |
2782 | return 0; | |
2783 | } | |
2784 | ||
2785 | if (s->reopen_state) { | |
2786 | /* We already have a new file descriptor to set permissions for */ | |
2787 | assert(s->reopen_state->perm == perm); | |
2788 | assert(s->reopen_state->shared_perm == shared); | |
2789 | rs = s->reopen_state->opaque; | |
2790 | s->perm_change_fd = rs->fd; | |
094e3639 | 2791 | s->perm_change_flags = rs->open_flags; |
6ceabe6f KW |
2792 | } else { |
2793 | /* We may need a new fd if auto-read-only switches the mode */ | |
23dece19 | 2794 | ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, |
6ceabe6f KW |
2795 | false, errp); |
2796 | if (ret < 0) { | |
2797 | return ret; | |
2798 | } else if (ret != s->fd) { | |
2799 | s->perm_change_fd = ret; | |
094e3639 | 2800 | s->perm_change_flags = open_flags; |
6ceabe6f KW |
2801 | } |
2802 | } | |
2803 | ||
2804 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
2805 | * but keep everything locked that new will need. */ | |
2806 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
2807 | if (ret < 0) { | |
2808 | goto fail; | |
2809 | } | |
2810 | ||
2811 | /* Copy locks to the new fd */ | |
2812 | if (s->perm_change_fd) { | |
2813 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, | |
2814 | false, errp); | |
2815 | if (ret < 0) { | |
2816 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
2817 | goto fail; | |
2818 | } | |
2819 | } | |
2820 | return 0; | |
2821 | ||
2822 | fail: | |
2823 | if (s->perm_change_fd && !s->reopen_state) { | |
2824 | qemu_close(s->perm_change_fd); | |
2825 | } | |
2826 | s->perm_change_fd = 0; | |
2827 | return ret; | |
244a5668 FZ |
2828 | } |
2829 | ||
2830 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
2831 | { | |
2832 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
2833 | |
2834 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
2835 | * called after .bdrv_reopen_commit) */ | |
2836 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
2837 | qemu_close(s->fd); | |
2838 | s->fd = s->perm_change_fd; | |
094e3639 | 2839 | s->open_flags = s->perm_change_flags; |
6ceabe6f KW |
2840 | } |
2841 | s->perm_change_fd = 0; | |
2842 | ||
244a5668 FZ |
2843 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
2844 | s->perm = perm; | |
2845 | s->shared_perm = shared; | |
2846 | } | |
2847 | ||
2848 | static void raw_abort_perm_update(BlockDriverState *bs) | |
2849 | { | |
6ceabe6f KW |
2850 | BDRVRawState *s = bs->opaque; |
2851 | ||
2852 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
2853 | * the file descriptor. */ | |
2854 | if (s->perm_change_fd && !s->reopen_state) { | |
2855 | qemu_close(s->perm_change_fd); | |
2856 | } | |
2857 | s->perm_change_fd = 0; | |
2858 | ||
244a5668 FZ |
2859 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
2860 | } | |
2861 | ||
67b51fb9 VSO |
2862 | static int coroutine_fn raw_co_copy_range_from( |
2863 | BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, | |
2864 | BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, | |
2865 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) | |
1efad060 | 2866 | { |
67b51fb9 VSO |
2867 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
2868 | read_flags, write_flags); | |
1efad060 FZ |
2869 | } |
2870 | ||
2871 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 VSO |
2872 | BdrvChild *src, |
2873 | uint64_t src_offset, | |
2874 | BdrvChild *dst, | |
2875 | uint64_t dst_offset, | |
2876 | uint64_t bytes, | |
2877 | BdrvRequestFlags read_flags, | |
2878 | BdrvRequestFlags write_flags) | |
1efad060 | 2879 | { |
58a209c4 | 2880 | RawPosixAIOData acb; |
1efad060 FZ |
2881 | BDRVRawState *s = bs->opaque; |
2882 | BDRVRawState *src_s; | |
2883 | ||
2884 | assert(dst->bs == bs); | |
2885 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
2886 | return -ENOTSUP; | |
2887 | } | |
2888 | ||
2889 | src_s = src->bs->opaque; | |
9f850f67 | 2890 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
2891 | return -EIO; |
2892 | } | |
58a209c4 KW |
2893 | |
2894 | acb = (RawPosixAIOData) { | |
2895 | .bs = bs, | |
2896 | .aio_type = QEMU_AIO_COPY_RANGE, | |
2897 | .aio_fildes = src_s->fd, | |
2898 | .aio_offset = src_offset, | |
2899 | .aio_nbytes = bytes, | |
2900 | .copy_range = { | |
2901 | .aio_fd2 = s->fd, | |
2902 | .aio_offset2 = dst_offset, | |
2903 | }, | |
2904 | }; | |
2905 | ||
2906 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
2907 | } |
2908 | ||
5f535a94 | 2909 | BlockDriver bdrv_file = { |
84a12e66 CH |
2910 | .format_name = "file", |
2911 | .protocol_name = "file", | |
856ae5c3 | 2912 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2913 | .bdrv_needs_filename = true, |
856ae5c3 | 2914 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 2915 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 2916 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
2917 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2918 | .bdrv_reopen_commit = raw_reopen_commit, | |
2919 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 2920 | .bdrv_close = raw_close, |
927f11e1 | 2921 | .bdrv_co_create = raw_co_create, |
efc75e2a | 2922 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 2923 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
1dcaf527 | 2924 | .bdrv_has_zero_init_truncate = bdrv_has_zero_init_1, |
a290f085 | 2925 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 2926 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 2927 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
3b46e624 | 2928 | |
9d52aa3c KW |
2929 | .bdrv_co_preadv = raw_co_preadv, |
2930 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
2931 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
2932 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
2933 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
2934 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 2935 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2936 | .bdrv_io_plug = raw_aio_plug, |
2937 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 2938 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 2939 | |
061ca8a3 | 2940 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 2941 | .bdrv_getlength = raw_getlength, |
260a82e5 | 2942 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
2943 | .bdrv_get_allocated_file_size |
2944 | = raw_get_allocated_file_size, | |
244a5668 FZ |
2945 | .bdrv_check_perm = raw_check_perm, |
2946 | .bdrv_set_perm = raw_set_perm, | |
2947 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 2948 | .create_opts = &raw_create_opts, |
8a2ce0bc | 2949 | .mutable_opts = mutable_opts, |
83f64091 FB |
2950 | }; |
2951 | ||
19cb3738 FB |
2952 | /***********************************************/ |
2953 | /* host device */ | |
2954 | ||
83affaa6 | 2955 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
2956 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2957 | CFIndex maxPathSize, int flags); | |
d0855f12 | 2958 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 2959 | { |
d0855f12 | 2960 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
2961 | mach_port_t masterPort; |
2962 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
2963 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
2964 | char *mediaType = NULL; | |
19cb3738 FB |
2965 | |
2966 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
2967 | if ( KERN_SUCCESS != kernResult ) { | |
2968 | printf( "IOMasterPort returned %d\n", kernResult ); | |
2969 | } | |
3b46e624 | 2970 | |
d0855f12 JA |
2971 | int index; |
2972 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
2973 | classesToMatch = IOServiceMatching(matching_array[index]); | |
2974 | if (classesToMatch == NULL) { | |
2975 | error_report("IOServiceMatching returned NULL for %s", | |
2976 | matching_array[index]); | |
2977 | continue; | |
2978 | } | |
2979 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
2980 | kCFBooleanTrue); | |
2981 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
2982 | mediaIterator); | |
2983 | if (kernResult != KERN_SUCCESS) { | |
2984 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
2985 | kernResult); | |
2986 | continue; | |
2987 | } | |
3b46e624 | 2988 | |
d0855f12 JA |
2989 | /* If a match was found, leave the loop */ |
2990 | if (*mediaIterator != 0) { | |
4f7d28d7 | 2991 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
2992 | mediaType = g_strdup(matching_array[index]); |
2993 | break; | |
2994 | } | |
2995 | } | |
2996 | return mediaType; | |
19cb3738 FB |
2997 | } |
2998 | ||
98caa5bc JA |
2999 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3000 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
3001 | { |
3002 | io_object_t nextMedia; | |
3003 | kern_return_t kernResult = KERN_FAILURE; | |
3004 | *bsdPath = '\0'; | |
3005 | nextMedia = IOIteratorNext( mediaIterator ); | |
3006 | if ( nextMedia ) | |
3007 | { | |
3008 | CFTypeRef bsdPathAsCFString; | |
3009 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3010 | if ( bsdPathAsCFString ) { | |
3011 | size_t devPathLength; | |
3012 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3013 | if (flags & BDRV_O_NOCACHE) { |
3014 | strcat(bsdPath, "r"); | |
3015 | } | |
19cb3738 FB |
3016 | devPathLength = strlen( bsdPath ); |
3017 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3018 | kernResult = KERN_SUCCESS; | |
3019 | } | |
3020 | CFRelease( bsdPathAsCFString ); | |
3021 | } | |
3022 | IOObjectRelease( nextMedia ); | |
3023 | } | |
3b46e624 | 3024 | |
19cb3738 FB |
3025 | return kernResult; |
3026 | } | |
3027 | ||
d0855f12 JA |
3028 | /* Sets up a real cdrom for use in QEMU */ |
3029 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3030 | { | |
3031 | int index, num_of_test_partitions = 2, fd; | |
3032 | char test_partition[MAXPATHLEN]; | |
3033 | bool partition_found = false; | |
3034 | ||
3035 | /* look for a working partition */ | |
3036 | for (index = 0; index < num_of_test_partitions; index++) { | |
3037 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3038 | index); | |
3039 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); | |
3040 | if (fd >= 0) { | |
3041 | partition_found = true; | |
3042 | qemu_close(fd); | |
3043 | break; | |
3044 | } | |
3045 | } | |
3046 | ||
3047 | /* if a working partition on the device was not found */ | |
3048 | if (partition_found == false) { | |
3049 | error_setg(errp, "Failed to find a working partition on disc"); | |
3050 | } else { | |
4f7d28d7 | 3051 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3052 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3053 | } | |
3054 | return partition_found; | |
3055 | } | |
3056 | ||
3057 | /* Prints directions on mounting and unmounting a device */ | |
3058 | static void print_unmounting_directions(const char *file_name) | |
3059 | { | |
3060 | error_report("If device %s is mounted on the desktop, unmount" | |
3061 | " it first before using it in QEMU", file_name); | |
3062 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3063 | file_name); | |
3064 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3065 | } | |
3066 | ||
3067 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3068 | |
508c7cb3 CH |
3069 | static int hdev_probe_device(const char *filename) |
3070 | { | |
3071 | struct stat st; | |
3072 | ||
3073 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3074 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3075 | return 50; | |
3076 | ||
3077 | if (stat(filename, &st) >= 0 && | |
3078 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3079 | return 100; | |
3080 | } | |
3081 | ||
3082 | return 0; | |
3083 | } | |
3084 | ||
da888d37 SH |
3085 | static int check_hdev_writable(BDRVRawState *s) |
3086 | { | |
3087 | #if defined(BLKROGET) | |
3088 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
3089 | * This is independent of device node permissions and therefore open(2) | |
3090 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
3091 | * | |
3092 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
3093 | * check for read-only block devices so that Linux block devices behave | |
3094 | * properly. | |
3095 | */ | |
3096 | struct stat st; | |
3097 | int readonly = 0; | |
3098 | ||
3099 | if (fstat(s->fd, &st)) { | |
3100 | return -errno; | |
3101 | } | |
3102 | ||
3103 | if (!S_ISBLK(st.st_mode)) { | |
3104 | return 0; | |
3105 | } | |
3106 | ||
3107 | if (ioctl(s->fd, BLKROGET, &readonly) < 0) { | |
3108 | return -errno; | |
3109 | } | |
3110 | ||
3111 | if (readonly) { | |
3112 | return -EACCES; | |
3113 | } | |
3114 | #endif /* defined(BLKROGET) */ | |
3115 | return 0; | |
3116 | } | |
3117 | ||
7af803d4 HR |
3118 | static void hdev_parse_filename(const char *filename, QDict *options, |
3119 | Error **errp) | |
3120 | { | |
03c320d8 | 3121 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3122 | } |
3123 | ||
3307ed7b DA |
3124 | static bool hdev_is_sg(BlockDriverState *bs) |
3125 | { | |
3126 | ||
3127 | #if defined(__linux__) | |
3128 | ||
0d4377b3 | 3129 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3130 | struct stat st; |
3131 | struct sg_scsi_id scsiid; | |
3132 | int sg_version; | |
0d4377b3 KW |
3133 | int ret; |
3134 | ||
3135 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3136 | return false; | |
3137 | } | |
3307ed7b | 3138 | |
0d4377b3 KW |
3139 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3140 | if (ret < 0) { | |
3141 | return false; | |
3142 | } | |
3143 | ||
3144 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3145 | if (ret >= 0) { | |
4f7d28d7 | 3146 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3147 | return true; |
3148 | } | |
3149 | ||
3150 | #endif | |
3151 | ||
3152 | return false; | |
3153 | } | |
3154 | ||
015a1036 HR |
3155 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3156 | Error **errp) | |
19cb3738 FB |
3157 | { |
3158 | BDRVRawState *s = bs->opaque; | |
e428e439 | 3159 | Error *local_err = NULL; |
da888d37 | 3160 | int ret; |
a76bab49 | 3161 | |
83affaa6 | 3162 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3163 | /* |
3164 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3165 | * would require more care. When @options come from -blockdev or | |
3166 | * blockdev_add, its members are typed according to the QAPI | |
3167 | * schema, but when they come from -drive, they're all QString. | |
3168 | */ | |
3307ed7b | 3169 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3170 | char bsd_path[MAXPATHLEN] = ""; |
3171 | bool error_occurred = false; | |
3172 | ||
3173 | /* If using a real cdrom */ | |
3174 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3175 | char *mediaType = NULL; | |
3176 | kern_return_t ret_val; | |
3177 | io_iterator_t mediaIterator = 0; | |
3178 | ||
3179 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3180 | if (mediaType == NULL) { | |
3181 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3182 | " drive"); | |
3183 | error_occurred = true; | |
3184 | goto hdev_open_Mac_error; | |
3185 | } | |
3307ed7b | 3186 | |
d0855f12 JA |
3187 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3188 | if (ret_val != KERN_SUCCESS) { | |
3189 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3190 | error_occurred = true; | |
3191 | goto hdev_open_Mac_error; | |
3192 | } | |
3193 | ||
3194 | /* If a real optical drive was not found */ | |
3195 | if (bsd_path[0] == '\0') { | |
3196 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3197 | error_occurred = true; | |
3198 | goto hdev_open_Mac_error; | |
3199 | } | |
3200 | ||
3201 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3202 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3203 | setup_cdrom(bsd_path, errp) == false) { | |
3204 | print_unmounting_directions(bsd_path); | |
3205 | error_occurred = true; | |
3206 | goto hdev_open_Mac_error; | |
19cb3738 | 3207 | } |
3b46e624 | 3208 | |
46f5ac20 | 3209 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3210 | |
3211 | hdev_open_Mac_error: | |
3212 | g_free(mediaType); | |
3213 | if (mediaIterator) { | |
3214 | IOObjectRelease(mediaIterator); | |
3215 | } | |
3216 | if (error_occurred) { | |
3217 | return -ENOENT; | |
3218 | } | |
19cb3738 | 3219 | } |
d0855f12 | 3220 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3221 | |
3222 | s->type = FTYPE_FILE; | |
90babde0 | 3223 | |
230ff739 | 3224 | ret = raw_open_common(bs, options, flags, 0, true, &local_err); |
da888d37 | 3225 | if (ret < 0) { |
621ff94d | 3226 | error_propagate(errp, local_err); |
d0855f12 JA |
3227 | #if defined(__APPLE__) && defined(__MACH__) |
3228 | if (*bsd_path) { | |
3229 | filename = bsd_path; | |
3230 | } | |
3231 | /* if a physical device experienced an error while being opened */ | |
3232 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3233 | print_unmounting_directions(filename); | |
3234 | } | |
3235 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3236 | return ret; |
3237 | } | |
3238 | ||
3307ed7b DA |
3239 | /* Since this does ioctl the device must be already opened */ |
3240 | bs->sg = hdev_is_sg(bs); | |
3241 | ||
da888d37 SH |
3242 | if (flags & BDRV_O_RDWR) { |
3243 | ret = check_hdev_writable(s); | |
3244 | if (ret < 0) { | |
3245 | raw_close(bs); | |
e428e439 | 3246 | error_setg_errno(errp, -ret, "The device is not writable"); |
da888d37 SH |
3247 | return ret; |
3248 | } | |
3249 | } | |
3250 | ||
3251 | return ret; | |
19cb3738 FB |
3252 | } |
3253 | ||
03ff3ca3 | 3254 | #if defined(__linux__) |
2f3a7ab3 KW |
3255 | static int coroutine_fn |
3256 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3257 | { |
f141eafe | 3258 | BDRVRawState *s = bs->opaque; |
03425671 | 3259 | RawPosixAIOData acb; |
2f3a7ab3 | 3260 | int ret; |
221f715d | 3261 | |
2f3a7ab3 KW |
3262 | ret = fd_open(bs); |
3263 | if (ret < 0) { | |
3264 | return ret; | |
3265 | } | |
c208e8c2 | 3266 | |
7c9e5276 PB |
3267 | if (req == SG_IO && s->pr_mgr) { |
3268 | struct sg_io_hdr *io_hdr = buf; | |
3269 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3270 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3271 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3272 | s->fd, io_hdr); |
7c9e5276 PB |
3273 | } |
3274 | } | |
3275 | ||
03425671 KW |
3276 | acb = (RawPosixAIOData) { |
3277 | .bs = bs, | |
3278 | .aio_type = QEMU_AIO_IOCTL, | |
3279 | .aio_fildes = s->fd, | |
3280 | .aio_offset = 0, | |
3281 | .ioctl = { | |
3282 | .buf = buf, | |
3283 | .cmd = req, | |
3284 | }, | |
3285 | }; | |
3286 | ||
3287 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3288 | } |
f709623b | 3289 | #endif /* linux */ |
221f715d | 3290 | |
9f23011a BS |
3291 | static int fd_open(BlockDriverState *bs) |
3292 | { | |
3293 | BDRVRawState *s = bs->opaque; | |
3294 | ||
3295 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
3296 | if (s->fd >= 0) | |
3297 | return 0; | |
3298 | return -EIO; | |
3299 | } | |
04eeb8b6 | 3300 | |
33d70fb6 KW |
3301 | static coroutine_fn int |
3302 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
c36dd8a0 | 3303 | { |
33d70fb6 | 3304 | int ret; |
c36dd8a0 | 3305 | |
33d70fb6 KW |
3306 | ret = fd_open(bs); |
3307 | if (ret < 0) { | |
3308 | return ret; | |
c36dd8a0 | 3309 | } |
46ee0f46 | 3310 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3311 | } |
3312 | ||
2ffa76c2 | 3313 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3314 | int64_t offset, int bytes, BdrvRequestFlags flags) |
d0b4503e | 3315 | { |
d0b4503e PB |
3316 | int rc; |
3317 | ||
3318 | rc = fd_open(bs); | |
3319 | if (rc < 0) { | |
3320 | return rc; | |
3321 | } | |
34fa110e | 3322 | |
7154d8ae | 3323 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3324 | } |
3325 | ||
efc75e2a SH |
3326 | static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, |
3327 | Error **errp) | |
93c65b47 AL |
3328 | { |
3329 | int fd; | |
3330 | int ret = 0; | |
3331 | struct stat stat_buf; | |
0e7e1989 | 3332 | int64_t total_size = 0; |
cc28c6aa HR |
3333 | bool has_prefix; |
3334 | ||
f709623b HR |
3335 | /* This function is used by both protocol block drivers and therefore either |
3336 | * of these prefixes may be given. | |
cc28c6aa HR |
3337 | * The return value has to be stored somewhere, otherwise this is an error |
3338 | * due to -Werror=unused-value. */ | |
3339 | has_prefix = | |
3340 | strstart(filename, "host_device:", &filename) || | |
f709623b | 3341 | strstart(filename, "host_cdrom:" , &filename); |
cc28c6aa HR |
3342 | |
3343 | (void)has_prefix; | |
93c65b47 | 3344 | |
db0754df | 3345 | ret = raw_normalize_devicepath(&filename, errp); |
bdd03cdf | 3346 | if (ret < 0) { |
bdd03cdf HR |
3347 | return ret; |
3348 | } | |
3349 | ||
0e7e1989 | 3350 | /* Read out options */ |
180e9526 HT |
3351 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
3352 | BDRV_SECTOR_SIZE); | |
93c65b47 | 3353 | |
6165f4d8 | 3354 | fd = qemu_open(filename, O_WRONLY | O_BINARY); |
e428e439 HR |
3355 | if (fd < 0) { |
3356 | ret = -errno; | |
3357 | error_setg_errno(errp, -ret, "Could not open device"); | |
3358 | return ret; | |
3359 | } | |
93c65b47 | 3360 | |
e428e439 | 3361 | if (fstat(fd, &stat_buf) < 0) { |
57e69b7d | 3362 | ret = -errno; |
e428e439 HR |
3363 | error_setg_errno(errp, -ret, "Could not stat device"); |
3364 | } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { | |
3365 | error_setg(errp, | |
3366 | "The given file is neither a block nor a character device"); | |
57e69b7d | 3367 | ret = -ENODEV; |
180e9526 | 3368 | } else if (lseek(fd, 0, SEEK_END) < total_size) { |
e428e439 | 3369 | error_setg(errp, "Device is too small"); |
93c65b47 | 3370 | ret = -ENOSPC; |
e428e439 | 3371 | } |
93c65b47 | 3372 | |
97ec9117 FZ |
3373 | if (!ret && total_size) { |
3374 | uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; | |
3375 | int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); | |
3376 | if (lseek(fd, 0, SEEK_SET) == -1) { | |
3377 | ret = -errno; | |
3378 | } else { | |
3379 | ret = qemu_write_full(fd, buf, zero_size); | |
3380 | ret = ret == zero_size ? 0 : -errno; | |
3381 | } | |
3382 | } | |
2e1e79da | 3383 | qemu_close(fd); |
93c65b47 AL |
3384 | return ret; |
3385 | } | |
3386 | ||
5efa9d5a | 3387 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3388 | .format_name = "host_device", |
84a12e66 | 3389 | .protocol_name = "host_device", |
0b4ce02e | 3390 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3391 | .bdrv_needs_filename = true, |
0b4ce02e | 3392 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3393 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3394 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3395 | .bdrv_close = raw_close, |
1bc6b705 JC |
3396 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3397 | .bdrv_reopen_commit = raw_reopen_commit, | |
3398 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3399 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3400 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3401 | .mutable_opts = mutable_opts, |
dd577a26 | 3402 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3403 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3404 | |
9d52aa3c KW |
3405 | .bdrv_co_preadv = raw_co_preadv, |
3406 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3407 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3408 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3409 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3410 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3411 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3412 | .bdrv_io_plug = raw_aio_plug, |
3413 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3414 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3415 | |
061ca8a3 | 3416 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3417 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3418 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3419 | .bdrv_get_allocated_file_size |
3420 | = raw_get_allocated_file_size, | |
244a5668 FZ |
3421 | .bdrv_check_perm = raw_check_perm, |
3422 | .bdrv_set_perm = raw_set_perm, | |
3423 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3424 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3425 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3426 | |
f3a5d3f8 | 3427 | /* generic scsi device */ |
63ec93db | 3428 | #ifdef __linux__ |
2f3a7ab3 | 3429 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3430 | #endif |
f3a5d3f8 CH |
3431 | }; |
3432 | ||
18fa1c42 HR |
3433 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3434 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3435 | Error **errp) | |
3436 | { | |
03c320d8 | 3437 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3438 | } |
3439 | #endif | |
3440 | ||
3441 | #ifdef __linux__ | |
015a1036 HR |
3442 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3443 | Error **errp) | |
f3a5d3f8 CH |
3444 | { |
3445 | BDRVRawState *s = bs->opaque; | |
3446 | ||
f3a5d3f8 CH |
3447 | s->type = FTYPE_CD; |
3448 | ||
19a3da7f | 3449 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3450 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3451 | } |
3452 | ||
508c7cb3 CH |
3453 | static int cdrom_probe_device(const char *filename) |
3454 | { | |
3baf720e CR |
3455 | int fd, ret; |
3456 | int prio = 0; | |
343f8568 | 3457 | struct stat st; |
3baf720e | 3458 | |
6165f4d8 | 3459 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
3baf720e CR |
3460 | if (fd < 0) { |
3461 | goto out; | |
3462 | } | |
343f8568 JS |
3463 | ret = fstat(fd, &st); |
3464 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3465 | goto outc; | |
3466 | } | |
3baf720e CR |
3467 | |
3468 | /* Attempt to detect via a CDROM specific ioctl */ | |
3469 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3470 | if (ret >= 0) | |
3471 | prio = 100; | |
3472 | ||
343f8568 | 3473 | outc: |
2e1e79da | 3474 | qemu_close(fd); |
3baf720e CR |
3475 | out: |
3476 | return prio; | |
508c7cb3 CH |
3477 | } |
3478 | ||
e031f750 | 3479 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3480 | { |
3481 | BDRVRawState *s = bs->opaque; | |
3482 | int ret; | |
3483 | ||
3484 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3485 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3486 | } |
3487 | ||
f36f3949 | 3488 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3489 | { |
3490 | BDRVRawState *s = bs->opaque; | |
3491 | ||
3492 | if (eject_flag) { | |
3493 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3494 | perror("CDROMEJECT"); | |
3495 | } else { | |
3496 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3497 | perror("CDROMEJECT"); | |
3498 | } | |
f3a5d3f8 CH |
3499 | } |
3500 | ||
025e849a | 3501 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3502 | { |
3503 | BDRVRawState *s = bs->opaque; | |
3504 | ||
3505 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3506 | /* | |
3507 | * Note: an error can happen if the distribution automatically | |
3508 | * mounts the CD-ROM | |
3509 | */ | |
3510 | /* perror("CDROM_LOCKDOOR"); */ | |
3511 | } | |
f3a5d3f8 CH |
3512 | } |
3513 | ||
3514 | static BlockDriver bdrv_host_cdrom = { | |
3515 | .format_name = "host_cdrom", | |
84a12e66 | 3516 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3517 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3518 | .bdrv_needs_filename = true, |
508c7cb3 | 3519 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3520 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3521 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3522 | .bdrv_close = raw_close, |
1bc6b705 JC |
3523 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3524 | .bdrv_reopen_commit = raw_reopen_commit, | |
3525 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3526 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3527 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3528 | .mutable_opts = mutable_opts, |
dd577a26 | 3529 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3530 | |
9d52aa3c KW |
3531 | |
3532 | .bdrv_co_preadv = raw_co_preadv, | |
3533 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3534 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3535 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3536 | .bdrv_io_plug = raw_aio_plug, |
3537 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3538 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3539 | |
061ca8a3 | 3540 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3541 | .bdrv_getlength = raw_getlength, |
3542 | .has_variable_length = true, | |
4a1d5e1f FZ |
3543 | .bdrv_get_allocated_file_size |
3544 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3545 | |
3546 | /* removable device support */ | |
3547 | .bdrv_is_inserted = cdrom_is_inserted, | |
3548 | .bdrv_eject = cdrom_eject, | |
025e849a | 3549 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3550 | |
3551 | /* generic scsi device */ | |
2f3a7ab3 | 3552 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3553 | }; |
3554 | #endif /* __linux__ */ | |
3555 | ||
a167ba50 | 3556 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3557 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3558 | Error **errp) | |
f3a5d3f8 CH |
3559 | { |
3560 | BDRVRawState *s = bs->opaque; | |
e428e439 | 3561 | Error *local_err = NULL; |
f3a5d3f8 CH |
3562 | int ret; |
3563 | ||
3564 | s->type = FTYPE_CD; | |
3565 | ||
230ff739 | 3566 | ret = raw_open_common(bs, options, flags, 0, true, &local_err); |
e428e439 | 3567 | if (ret) { |
621ff94d | 3568 | error_propagate(errp, local_err); |
f3a5d3f8 | 3569 | return ret; |
e428e439 | 3570 | } |
f3a5d3f8 | 3571 | |
9b2260cb | 3572 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3573 | ioctl(s->fd, CDIOCALLOW); |
3574 | return 0; | |
3575 | } | |
3576 | ||
508c7cb3 CH |
3577 | static int cdrom_probe_device(const char *filename) |
3578 | { | |
3579 | if (strstart(filename, "/dev/cd", NULL) || | |
3580 | strstart(filename, "/dev/acd", NULL)) | |
3581 | return 100; | |
3582 | return 0; | |
3583 | } | |
3584 | ||
f3a5d3f8 CH |
3585 | static int cdrom_reopen(BlockDriverState *bs) |
3586 | { | |
3587 | BDRVRawState *s = bs->opaque; | |
3588 | int fd; | |
3589 | ||
3590 | /* | |
3591 | * Force reread of possibly changed/newly loaded disc, | |
3592 | * FreeBSD seems to not notice sometimes... | |
3593 | */ | |
3594 | if (s->fd >= 0) | |
2e1e79da | 3595 | qemu_close(s->fd); |
6165f4d8 | 3596 | fd = qemu_open(bs->filename, s->open_flags, 0644); |
f3a5d3f8 CH |
3597 | if (fd < 0) { |
3598 | s->fd = -1; | |
3599 | return -EIO; | |
3600 | } | |
3601 | s->fd = fd; | |
3602 | ||
9b2260cb | 3603 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3604 | ioctl(s->fd, CDIOCALLOW); |
3605 | return 0; | |
3606 | } | |
3607 | ||
e031f750 | 3608 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3609 | { |
3610 | return raw_getlength(bs) > 0; | |
3611 | } | |
3612 | ||
f36f3949 | 3613 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3614 | { |
3615 | BDRVRawState *s = bs->opaque; | |
3616 | ||
3617 | if (s->fd < 0) | |
822e1cd1 | 3618 | return; |
f3a5d3f8 CH |
3619 | |
3620 | (void) ioctl(s->fd, CDIOCALLOW); | |
3621 | ||
3622 | if (eject_flag) { | |
3623 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3624 | perror("CDIOCEJECT"); | |
3625 | } else { | |
3626 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3627 | perror("CDIOCCLOSE"); | |
3628 | } | |
3629 | ||
822e1cd1 | 3630 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3631 | } |
3632 | ||
025e849a | 3633 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3634 | { |
3635 | BDRVRawState *s = bs->opaque; | |
3636 | ||
3637 | if (s->fd < 0) | |
7bf37fed | 3638 | return; |
f3a5d3f8 CH |
3639 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3640 | /* | |
3641 | * Note: an error can happen if the distribution automatically | |
3642 | * mounts the CD-ROM | |
3643 | */ | |
3644 | /* perror("CDROM_LOCKDOOR"); */ | |
3645 | } | |
f3a5d3f8 CH |
3646 | } |
3647 | ||
3648 | static BlockDriver bdrv_host_cdrom = { | |
3649 | .format_name = "host_cdrom", | |
84a12e66 | 3650 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3651 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3652 | .bdrv_needs_filename = true, |
508c7cb3 | 3653 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3654 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3655 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3656 | .bdrv_close = raw_close, |
1bc6b705 JC |
3657 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3658 | .bdrv_reopen_commit = raw_reopen_commit, | |
3659 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3660 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3661 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3662 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3663 | |
9d52aa3c KW |
3664 | .bdrv_co_preadv = raw_co_preadv, |
3665 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3666 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3667 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3668 | .bdrv_io_plug = raw_aio_plug, |
3669 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3670 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3671 | |
061ca8a3 | 3672 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3673 | .bdrv_getlength = raw_getlength, |
3674 | .has_variable_length = true, | |
4a1d5e1f FZ |
3675 | .bdrv_get_allocated_file_size |
3676 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3677 | |
19cb3738 | 3678 | /* removable device support */ |
f3a5d3f8 CH |
3679 | .bdrv_is_inserted = cdrom_is_inserted, |
3680 | .bdrv_eject = cdrom_eject, | |
025e849a | 3681 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3682 | }; |
f3a5d3f8 | 3683 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3684 | |
84a12e66 | 3685 | static void bdrv_file_init(void) |
5efa9d5a | 3686 | { |
508c7cb3 CH |
3687 | /* |
3688 | * Register all the drivers. Note that order is important, the driver | |
3689 | * registered last will get probed first. | |
3690 | */ | |
84a12e66 | 3691 | bdrv_register(&bdrv_file); |
5efa9d5a | 3692 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3693 | #ifdef __linux__ |
f3a5d3f8 CH |
3694 | bdrv_register(&bdrv_host_cdrom); |
3695 | #endif | |
a167ba50 | 3696 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3697 | bdrv_register(&bdrv_host_cdrom); |
3698 | #endif | |
5efa9d5a AL |
3699 | } |
3700 | ||
84a12e66 | 3701 | block_init(bdrv_file_init); |