]>
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; |
e0c9cf3a KW |
149 | BDRVReopenState *reopen_state; |
150 | ||
dce512de | 151 | #ifdef CONFIG_XFS |
260a82e5 | 152 | bool is_xfs:1; |
dce512de | 153 | #endif |
260a82e5 | 154 | bool has_discard:1; |
97a2ae34 | 155 | bool has_write_zeroes:1; |
260a82e5 | 156 | bool discard_zeroes:1; |
0a4279d9 | 157 | bool use_linux_aio:1; |
e5bcf967 | 158 | bool page_cache_inconsistent:1; |
d50d8222 | 159 | bool has_fallocate; |
3cad8307 | 160 | bool needs_alignment; |
f357fcd8 | 161 | bool drop_cache; |
31be8a2a | 162 | bool check_cache_dropped; |
7c9e5276 PB |
163 | |
164 | PRManager *pr_mgr; | |
19cb3738 FB |
165 | } BDRVRawState; |
166 | ||
eeb6b45d JC |
167 | typedef struct BDRVRawReopenState { |
168 | int fd; | |
169 | int open_flags; | |
f357fcd8 | 170 | bool drop_cache; |
31be8a2a | 171 | bool check_cache_dropped; |
eeb6b45d JC |
172 | } BDRVRawReopenState; |
173 | ||
19cb3738 | 174 | static int fd_open(BlockDriverState *bs); |
22afa7b5 | 175 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 176 | |
de81a169 PB |
177 | typedef struct RawPosixAIOData { |
178 | BlockDriverState *bs; | |
d57c44d0 | 179 | int aio_type; |
de81a169 | 180 | int aio_fildes; |
d57c44d0 | 181 | |
de81a169 | 182 | off_t aio_offset; |
d57c44d0 KW |
183 | uint64_t aio_nbytes; |
184 | ||
93f4e2ff | 185 | union { |
d57c44d0 KW |
186 | struct { |
187 | struct iovec *iov; | |
188 | int niov; | |
189 | } io; | |
190 | struct { | |
191 | uint64_t cmd; | |
192 | void *buf; | |
193 | } ioctl; | |
93f4e2ff KW |
194 | struct { |
195 | int aio_fd2; | |
196 | off_t aio_offset2; | |
d57c44d0 | 197 | } copy_range; |
93f4e2ff KW |
198 | struct { |
199 | PreallocMode prealloc; | |
200 | Error **errp; | |
d57c44d0 | 201 | } truncate; |
93f4e2ff | 202 | }; |
de81a169 PB |
203 | } RawPosixAIOData; |
204 | ||
a167ba50 | 205 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 206 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
207 | #endif |
208 | ||
1de1ae0a | 209 | #if defined(__NetBSD__) |
db0754df | 210 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
211 | { |
212 | static char namebuf[PATH_MAX]; | |
213 | const char *dp, *fname; | |
214 | struct stat sb; | |
215 | ||
216 | fname = *filename; | |
217 | dp = strrchr(fname, '/'); | |
218 | if (lstat(fname, &sb) < 0) { | |
db0754df | 219 | error_setg_errno(errp, errno, "%s: stat failed", fname); |
1de1ae0a CE |
220 | return -errno; |
221 | } | |
222 | ||
223 | if (!S_ISBLK(sb.st_mode)) { | |
224 | return 0; | |
225 | } | |
226 | ||
227 | if (dp == NULL) { | |
228 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
229 | } else { | |
230 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
231 | (int)(dp - fname), fname, dp + 1); | |
232 | } | |
1de1ae0a | 233 | *filename = namebuf; |
db0754df | 234 | warn_report("%s is a block device, using %s", fname, *filename); |
1de1ae0a CE |
235 | |
236 | return 0; | |
237 | } | |
238 | #else | |
db0754df | 239 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
240 | { |
241 | return 0; | |
242 | } | |
243 | #endif | |
244 | ||
8a4ed0d1 ET |
245 | /* |
246 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
247 | */ | |
248 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 249 | { |
c25f53b0 | 250 | unsigned int sector_size; |
8a4ed0d1 | 251 | bool success = false; |
700f9ce0 | 252 | int i; |
c25f53b0 | 253 | |
8a4ed0d1 | 254 | errno = ENOTSUP; |
700f9ce0 | 255 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 256 | #ifdef BLKSSZGET |
700f9ce0 | 257 | BLKSSZGET, |
c25f53b0 PB |
258 | #endif |
259 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 260 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
261 | #endif |
262 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 263 | DIOCGSECTORSIZE, |
c25f53b0 | 264 | #endif |
700f9ce0 PM |
265 | }; |
266 | ||
267 | /* Try a few ioctls to get the right size */ | |
268 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
269 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
270 | *sector_size_p = sector_size; | |
271 | success = true; | |
272 | } | |
273 | } | |
8a4ed0d1 ET |
274 | |
275 | return success ? 0 : -errno; | |
276 | } | |
277 | ||
1a9335e4 ET |
278 | /** |
279 | * Get physical block size of @fd. | |
280 | * On success, store it in @blk_size and return 0. | |
281 | * On failure, return -errno. | |
282 | */ | |
283 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
284 | { | |
285 | #ifdef BLKPBSZGET | |
286 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
287 | return -errno; | |
288 | } | |
289 | return 0; | |
290 | #else | |
291 | return -ENOTSUP; | |
292 | #endif | |
293 | } | |
294 | ||
22d182e8 SH |
295 | /* Check if read is allowed with given memory buffer and length. |
296 | * | |
297 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
298 | */ | |
299 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
300 | { | |
301 | ssize_t ret = pread(fd, buf, len, 0); | |
302 | ||
303 | if (ret >= 0) { | |
304 | return true; | |
305 | } | |
306 | ||
307 | #ifdef __linux__ | |
308 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
309 | * other errors (e.g. real I/O error), which could happen on a failed | |
310 | * drive, since we only care about probing alignment. | |
311 | */ | |
312 | if (errno != EINVAL) { | |
313 | return true; | |
314 | } | |
315 | #endif | |
316 | ||
317 | return false; | |
318 | } | |
319 | ||
8a4ed0d1 ET |
320 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
321 | { | |
322 | BDRVRawState *s = bs->opaque; | |
323 | char *buf; | |
459b4e66 | 324 | size_t max_align = MAX(MAX_BLOCKSIZE, getpagesize()); |
8a4ed0d1 | 325 | |
b192af8a | 326 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 327 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 328 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 329 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
330 | s->buf_align = 1; |
331 | return; | |
332 | } | |
333 | ||
a5b8dd2c | 334 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
335 | s->buf_align = 0; |
336 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
337 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
338 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 339 | } |
c25f53b0 PB |
340 | #ifdef CONFIG_XFS |
341 | if (s->is_xfs) { | |
342 | struct dioattr da; | |
df26a350 | 343 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
a5b8dd2c | 344 | bs->bl.request_alignment = da.d_miniosz; |
c25f53b0 PB |
345 | /* The kernel returns wrong information for d_mem */ |
346 | /* s->buf_align = da.d_mem; */ | |
347 | } | |
348 | } | |
349 | #endif | |
350 | ||
351 | /* If we could not get the sizes so far, we can only guess them */ | |
352 | if (!s->buf_align) { | |
353 | size_t align; | |
459b4e66 DL |
354 | buf = qemu_memalign(max_align, 2 * max_align); |
355 | for (align = 512; align <= max_align; align <<= 1) { | |
356 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
c25f53b0 PB |
357 | s->buf_align = align; |
358 | break; | |
359 | } | |
360 | } | |
361 | qemu_vfree(buf); | |
362 | } | |
363 | ||
a5b8dd2c | 364 | if (!bs->bl.request_alignment) { |
c25f53b0 | 365 | size_t align; |
459b4e66 DL |
366 | buf = qemu_memalign(s->buf_align, max_align); |
367 | for (align = 512; align <= max_align; align <<= 1) { | |
22d182e8 | 368 | if (raw_is_io_aligned(fd, buf, align)) { |
a5b8dd2c | 369 | bs->bl.request_alignment = align; |
c25f53b0 PB |
370 | break; |
371 | } | |
372 | } | |
373 | qemu_vfree(buf); | |
374 | } | |
df26a350 | 375 | |
a5b8dd2c | 376 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
377 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
378 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 379 | } |
c25f53b0 PB |
380 | } |
381 | ||
23dece19 | 382 | static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) |
6a8dc042 | 383 | { |
23dece19 | 384 | bool read_write = false; |
6a8dc042 JC |
385 | assert(open_flags != NULL); |
386 | ||
387 | *open_flags |= O_BINARY; | |
388 | *open_flags &= ~O_ACCMODE; | |
23dece19 KW |
389 | |
390 | if (bdrv_flags & BDRV_O_AUTO_RDONLY) { | |
391 | read_write = has_writers; | |
392 | } else if (bdrv_flags & BDRV_O_RDWR) { | |
393 | read_write = true; | |
394 | } | |
395 | ||
396 | if (read_write) { | |
6a8dc042 JC |
397 | *open_flags |= O_RDWR; |
398 | } else { | |
399 | *open_flags |= O_RDONLY; | |
400 | } | |
401 | ||
402 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
403 | * and O_DIRECT for no caching. */ | |
404 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
405 | *open_flags |= O_DIRECT; | |
406 | } | |
6a8dc042 JC |
407 | } |
408 | ||
078896a9 HR |
409 | static void raw_parse_filename(const char *filename, QDict *options, |
410 | Error **errp) | |
411 | { | |
03c320d8 | 412 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
413 | } |
414 | ||
c66a6157 KW |
415 | static QemuOptsList raw_runtime_opts = { |
416 | .name = "raw", | |
417 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
418 | .desc = { | |
419 | { | |
420 | .name = "filename", | |
421 | .type = QEMU_OPT_STRING, | |
422 | .help = "File name of the image", | |
423 | }, | |
0a4279d9 KW |
424 | { |
425 | .name = "aio", | |
426 | .type = QEMU_OPT_STRING, | |
427 | .help = "host AIO implementation (threads, native)", | |
428 | }, | |
16b48d5d FZ |
429 | { |
430 | .name = "locking", | |
431 | .type = QEMU_OPT_STRING, | |
432 | .help = "file locking mode (on/off/auto, default: auto)", | |
433 | }, | |
7c9e5276 PB |
434 | { |
435 | .name = "pr-manager", | |
436 | .type = QEMU_OPT_STRING, | |
437 | .help = "id of persistent reservation manager object (default: none)", | |
438 | }, | |
f357fcd8 SH |
439 | #if defined(__linux__) |
440 | { | |
441 | .name = "drop-cache", | |
442 | .type = QEMU_OPT_BOOL, | |
443 | .help = "invalidate page cache during live migration (default: on)", | |
444 | }, | |
445 | #endif | |
31be8a2a SH |
446 | { |
447 | .name = "x-check-cache-dropped", | |
448 | .type = QEMU_OPT_BOOL, | |
449 | .help = "check that page cache was dropped on live migration (default: off)" | |
450 | }, | |
c66a6157 KW |
451 | { /* end of list */ } |
452 | }, | |
453 | }; | |
454 | ||
8a2ce0bc AG |
455 | static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; |
456 | ||
c66a6157 | 457 | static int raw_open_common(BlockDriverState *bs, QDict *options, |
230ff739 JS |
458 | int bdrv_flags, int open_flags, |
459 | bool device, Error **errp) | |
83f64091 FB |
460 | { |
461 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
462 | QemuOpts *opts; |
463 | Error *local_err = NULL; | |
8bfea15d | 464 | const char *filename = NULL; |
7c9e5276 | 465 | const char *str; |
0a4279d9 | 466 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 467 | int fd, ret; |
260a82e5 | 468 | struct stat st; |
244a5668 | 469 | OnOffAuto locking; |
83f64091 | 470 | |
87ea75d5 | 471 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
c66a6157 | 472 | qemu_opts_absorb_qdict(opts, options, &local_err); |
84d18f06 | 473 | if (local_err) { |
e428e439 | 474 | error_propagate(errp, local_err); |
c66a6157 KW |
475 | ret = -EINVAL; |
476 | goto fail; | |
477 | } | |
478 | ||
479 | filename = qemu_opt_get(opts, "filename"); | |
480 | ||
db0754df | 481 | ret = raw_normalize_devicepath(&filename, errp); |
1de1ae0a | 482 | if (ret != 0) { |
c66a6157 | 483 | goto fail; |
1de1ae0a CE |
484 | } |
485 | ||
0a4279d9 KW |
486 | aio_default = (bdrv_flags & BDRV_O_NATIVE_AIO) |
487 | ? BLOCKDEV_AIO_OPTIONS_NATIVE | |
488 | : BLOCKDEV_AIO_OPTIONS_THREADS; | |
f7abe0ec MAL |
489 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
490 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 491 | aio_default, &local_err); |
0a4279d9 KW |
492 | if (local_err) { |
493 | error_propagate(errp, local_err); | |
494 | ret = -EINVAL; | |
495 | goto fail; | |
496 | } | |
497 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); | |
498 | ||
f7abe0ec MAL |
499 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
500 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 501 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
502 | if (local_err) { |
503 | error_propagate(errp, local_err); | |
504 | ret = -EINVAL; | |
505 | goto fail; | |
506 | } | |
507 | switch (locking) { | |
508 | case ON_OFF_AUTO_ON: | |
509 | s->use_lock = true; | |
2b218f5d | 510 | if (!qemu_has_ofd_lock()) { |
db0754df FZ |
511 | warn_report("File lock requested but OFD locking syscall is " |
512 | "unavailable, falling back to POSIX file locks"); | |
513 | error_printf("Due to the implementation, locks can be lost " | |
514 | "unexpectedly.\n"); | |
2b218f5d | 515 | } |
244a5668 FZ |
516 | break; |
517 | case ON_OFF_AUTO_OFF: | |
518 | s->use_lock = false; | |
519 | break; | |
520 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 521 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
522 | break; |
523 | default: | |
524 | abort(); | |
525 | } | |
526 | ||
7c9e5276 PB |
527 | str = qemu_opt_get(opts, "pr-manager"); |
528 | if (str) { | |
529 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
530 | if (local_err) { | |
531 | error_propagate(errp, local_err); | |
532 | ret = -EINVAL; | |
533 | goto fail; | |
534 | } | |
535 | } | |
536 | ||
f357fcd8 | 537 | s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); |
31be8a2a SH |
538 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
539 | false); | |
540 | ||
6a8dc042 | 541 | s->open_flags = open_flags; |
23dece19 | 542 | raw_parse_flags(bdrv_flags, &s->open_flags, false); |
83f64091 | 543 | |
90babde0 | 544 | s->fd = -1; |
40ff6d7e | 545 | fd = qemu_open(filename, s->open_flags, 0644); |
64107dc0 KW |
546 | ret = fd < 0 ? -errno : 0; |
547 | ||
64107dc0 KW |
548 | if (ret < 0) { |
549 | error_setg_errno(errp, -ret, "Could not open '%s'", filename); | |
c66a6157 | 550 | if (ret == -EROFS) { |
19cb3738 | 551 | ret = -EACCES; |
c66a6157 KW |
552 | } |
553 | goto fail; | |
19cb3738 | 554 | } |
83f64091 | 555 | s->fd = fd; |
9ef91a67 | 556 | |
244a5668 FZ |
557 | s->perm = 0; |
558 | s->shared_perm = BLK_PERM_ALL; | |
559 | ||
5c6c3a6c | 560 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 561 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
562 | if (s->use_linux_aio) { |
563 | if (!(s->open_flags & O_DIRECT)) { | |
564 | error_setg(errp, "aio=native was specified, but it requires " | |
565 | "cache.direct=on, which was not specified."); | |
566 | ret = -EINVAL; | |
567 | goto fail; | |
568 | } | |
569 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
570 | error_prepend(errp, "Unable to use native AIO: "); | |
571 | goto fail; | |
572 | } | |
96518254 | 573 | } |
1501ecc1 | 574 | #else |
0a4279d9 | 575 | if (s->use_linux_aio) { |
d657c0c2 KW |
576 | error_setg(errp, "aio=native was specified, but is not supported " |
577 | "in this build."); | |
578 | ret = -EINVAL; | |
579 | goto fail; | |
1501ecc1 SH |
580 | } |
581 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 582 | |
7ce21016 | 583 | s->has_discard = true; |
97a2ae34 | 584 | s->has_write_zeroes = true; |
3cad8307 RPM |
585 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0) { |
586 | s->needs_alignment = true; | |
587 | } | |
260a82e5 PB |
588 | |
589 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 590 | ret = -errno; |
260a82e5 PB |
591 | error_setg_errno(errp, errno, "Could not stat file"); |
592 | goto fail; | |
593 | } | |
230ff739 JS |
594 | |
595 | if (!device) { | |
596 | if (S_ISBLK(st.st_mode)) { | |
597 | warn_report("Opening a block device as a file using the '%s' " | |
598 | "driver is deprecated", bs->drv->format_name); | |
599 | } else if (S_ISCHR(st.st_mode)) { | |
600 | warn_report("Opening a character device as a file using the '%s' " | |
601 | "driver is deprecated", bs->drv->format_name); | |
602 | } else if (!S_ISREG(st.st_mode)) { | |
603 | error_setg(errp, "A regular file was expected by the '%s' driver, " | |
604 | "but something else was given", bs->drv->format_name); | |
605 | ret = -EINVAL; | |
606 | goto fail; | |
607 | } else { | |
608 | s->discard_zeroes = true; | |
609 | s->has_fallocate = true; | |
610 | } | |
611 | } else { | |
612 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
613 | error_setg(errp, "'%s' driver expects either " | |
614 | "a character or block device", bs->drv->format_name); | |
615 | ret = -EINVAL; | |
616 | goto fail; | |
617 | } | |
260a82e5 | 618 | } |
230ff739 | 619 | |
d0b4503e PB |
620 | if (S_ISBLK(st.st_mode)) { |
621 | #ifdef BLKDISCARDZEROES | |
622 | unsigned int arg; | |
623 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
624 | s->discard_zeroes = true; | |
625 | } | |
626 | #endif | |
627 | #ifdef __linux__ | |
628 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
629 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 630 | * Same for BLKZEROOUT. |
d0b4503e PB |
631 | */ |
632 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
633 | s->discard_zeroes = false; | |
97a2ae34 | 634 | s->has_write_zeroes = false; |
d0b4503e PB |
635 | } |
636 | #endif | |
637 | } | |
3cad8307 RPM |
638 | #ifdef __FreeBSD__ |
639 | if (S_ISCHR(st.st_mode)) { | |
640 | /* | |
641 | * The file is a char device (disk), which on FreeBSD isn't behind | |
642 | * a pager, so force all requests to be aligned. This is needed | |
643 | * so QEMU makes sure all IO operations on the device are aligned | |
644 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
645 | */ | |
646 | s->needs_alignment = true; | |
647 | } | |
648 | #endif | |
260a82e5 | 649 | |
dce512de CH |
650 | #ifdef CONFIG_XFS |
651 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 652 | s->is_xfs = true; |
dce512de CH |
653 | } |
654 | #endif | |
655 | ||
738301e1 | 656 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
c66a6157 KW |
657 | ret = 0; |
658 | fail: | |
8bfea15d KW |
659 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
660 | unlink(filename); | |
661 | } | |
c66a6157 KW |
662 | qemu_opts_del(opts); |
663 | return ret; | |
83f64091 FB |
664 | } |
665 | ||
015a1036 HR |
666 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
667 | Error **errp) | |
90babde0 CH |
668 | { |
669 | BDRVRawState *s = bs->opaque; | |
670 | ||
671 | s->type = FTYPE_FILE; | |
230ff739 | 672 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
673 | } |
674 | ||
244a5668 FZ |
675 | typedef enum { |
676 | RAW_PL_PREPARE, | |
677 | RAW_PL_COMMIT, | |
678 | RAW_PL_ABORT, | |
679 | } RawPermLockOp; | |
680 | ||
681 | #define PERM_FOREACH(i) \ | |
682 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
683 | ||
684 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
685 | * file; if @unlock == true, also unlock the unneeded bytes. | |
686 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
687 | */ | |
2996ffad | 688 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
689 | uint64_t perm_lock_bits, |
690 | uint64_t shared_perm_lock_bits, | |
691 | bool unlock, Error **errp) | |
692 | { | |
693 | int ret; | |
694 | int i; | |
2996ffad FZ |
695 | uint64_t locked_perm, locked_shared_perm; |
696 | ||
697 | if (s) { | |
698 | locked_perm = s->locked_perm; | |
699 | locked_shared_perm = s->locked_shared_perm; | |
700 | } else { | |
701 | /* | |
702 | * We don't have the previous bits, just lock/unlock for each of the | |
703 | * requested bits. | |
704 | */ | |
705 | if (unlock) { | |
706 | locked_perm = BLK_PERM_ALL; | |
707 | locked_shared_perm = BLK_PERM_ALL; | |
708 | } else { | |
709 | locked_perm = 0; | |
710 | locked_shared_perm = 0; | |
711 | } | |
712 | } | |
244a5668 FZ |
713 | |
714 | PERM_FOREACH(i) { | |
715 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
716 | uint64_t bit = (1ULL << i); |
717 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 718 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
719 | if (ret) { |
720 | error_setg(errp, "Failed to lock byte %d", off); | |
721 | return ret; | |
2996ffad FZ |
722 | } else if (s) { |
723 | s->locked_perm |= bit; | |
244a5668 | 724 | } |
2996ffad | 725 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 726 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
727 | if (ret) { |
728 | error_setg(errp, "Failed to unlock byte %d", off); | |
729 | return ret; | |
2996ffad FZ |
730 | } else if (s) { |
731 | s->locked_perm &= ~bit; | |
244a5668 FZ |
732 | } |
733 | } | |
734 | } | |
735 | PERM_FOREACH(i) { | |
736 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
737 | uint64_t bit = (1ULL << i); |
738 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 739 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
740 | if (ret) { |
741 | error_setg(errp, "Failed to lock byte %d", off); | |
742 | return ret; | |
2996ffad FZ |
743 | } else if (s) { |
744 | s->locked_shared_perm |= bit; | |
244a5668 | 745 | } |
2996ffad FZ |
746 | } else if (unlock && (locked_shared_perm & bit) && |
747 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 748 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
749 | if (ret) { |
750 | error_setg(errp, "Failed to unlock byte %d", off); | |
751 | return ret; | |
2996ffad FZ |
752 | } else if (s) { |
753 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
754 | } |
755 | } | |
756 | } | |
757 | return 0; | |
758 | } | |
759 | ||
760 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 761 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
762 | Error **errp) |
763 | { | |
764 | int ret; | |
765 | int i; | |
766 | ||
767 | PERM_FOREACH(i) { | |
768 | int off = RAW_LOCK_SHARED_BASE + i; | |
769 | uint64_t p = 1ULL << i; | |
770 | if (perm & p) { | |
d0a96155 | 771 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
772 | if (ret) { |
773 | char *perm_name = bdrv_perm_names(p); | |
774 | error_setg(errp, | |
775 | "Failed to get \"%s\" lock", | |
776 | perm_name); | |
777 | g_free(perm_name); | |
244a5668 FZ |
778 | return ret; |
779 | } | |
780 | } | |
781 | } | |
782 | PERM_FOREACH(i) { | |
783 | int off = RAW_LOCK_PERM_BASE + i; | |
784 | uint64_t p = 1ULL << i; | |
785 | if (!(shared_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 shared \"%s\" lock", | |
791 | perm_name); | |
792 | g_free(perm_name); | |
244a5668 FZ |
793 | return ret; |
794 | } | |
795 | } | |
796 | } | |
797 | return 0; | |
798 | } | |
799 | ||
800 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
801 | RawPermLockOp op, | |
802 | uint64_t new_perm, uint64_t new_shared, | |
803 | Error **errp) | |
804 | { | |
805 | BDRVRawState *s = bs->opaque; | |
806 | int ret = 0; | |
807 | Error *local_err = NULL; | |
808 | ||
809 | if (!s->use_lock) { | |
810 | return 0; | |
811 | } | |
812 | ||
813 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
814 | return 0; | |
815 | } | |
816 | ||
244a5668 FZ |
817 | switch (op) { |
818 | case RAW_PL_PREPARE: | |
696aaaed VSO |
819 | if ((s->perm | new_perm) == s->perm && |
820 | (s->shared_perm & new_shared) == s->shared_perm) | |
821 | { | |
822 | /* | |
823 | * We are going to unlock bytes, it should not fail. If it fail due | |
824 | * to some fs-dependent permission-unrelated reasons (which occurs | |
825 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
826 | * can't prevent such errors by any check here. And we ignore them | |
827 | * anyway in ABORT and COMMIT. | |
828 | */ | |
829 | return 0; | |
830 | } | |
f2e3af29 | 831 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
832 | ~s->shared_perm | ~new_shared, |
833 | false, errp); | |
834 | if (!ret) { | |
f2e3af29 | 835 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
836 | if (!ret) { |
837 | return 0; | |
838 | } | |
b857431d FZ |
839 | error_append_hint(errp, |
840 | "Is another process using the image [%s]?\n", | |
841 | bs->filename); | |
244a5668 FZ |
842 | } |
843 | op = RAW_PL_ABORT; | |
844 | /* fall through to unlock bytes. */ | |
845 | case RAW_PL_ABORT: | |
f2e3af29 | 846 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 847 | true, &local_err); |
244a5668 FZ |
848 | if (local_err) { |
849 | /* Theoretically the above call only unlocks bytes and it cannot | |
850 | * fail. Something weird happened, report it. | |
851 | */ | |
db0754df | 852 | warn_report_err(local_err); |
244a5668 FZ |
853 | } |
854 | break; | |
855 | case RAW_PL_COMMIT: | |
f2e3af29 | 856 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 857 | true, &local_err); |
244a5668 FZ |
858 | if (local_err) { |
859 | /* Theoretically the above call only unlocks bytes and it cannot | |
860 | * fail. Something weird happened, report it. | |
861 | */ | |
db0754df | 862 | warn_report_err(local_err); |
244a5668 FZ |
863 | } |
864 | break; | |
865 | } | |
866 | return ret; | |
867 | } | |
868 | ||
5cec2870 | 869 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 870 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 871 | Error **errp) |
5cec2870 KW |
872 | { |
873 | BDRVRawState *s = bs->opaque; | |
874 | int fd = -1; | |
875 | int ret; | |
23dece19 KW |
876 | bool has_writers = perm & |
877 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
878 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
879 | #ifdef O_NOATIME | |
880 | fcntl_flags |= O_NOATIME; | |
881 | #endif | |
882 | ||
883 | *open_flags = 0; | |
884 | if (s->type == FTYPE_CD) { | |
885 | *open_flags |= O_NONBLOCK; | |
886 | } | |
887 | ||
23dece19 | 888 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
889 | |
890 | #ifdef O_ASYNC | |
891 | /* Not all operating systems have O_ASYNC, and those that don't | |
892 | * will not let us track the state into rs->open_flags (typically | |
893 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
894 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
895 | */ | |
896 | assert((s->open_flags & O_ASYNC) == 0); | |
897 | #endif | |
898 | ||
6ceabe6f KW |
899 | if (!force_dup && *open_flags == s->open_flags) { |
900 | /* We're lucky, the existing fd is fine */ | |
901 | return s->fd; | |
902 | } | |
903 | ||
5cec2870 KW |
904 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
905 | /* dup the original fd */ | |
906 | fd = qemu_dup(s->fd); | |
907 | if (fd >= 0) { | |
908 | ret = fcntl_setfl(fd, *open_flags); | |
909 | if (ret) { | |
910 | qemu_close(fd); | |
911 | fd = -1; | |
912 | } | |
913 | } | |
914 | } | |
915 | ||
916 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ | |
917 | if (fd == -1) { | |
918 | const char *normalized_filename = bs->filename; | |
919 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
920 | if (ret >= 0) { | |
921 | assert(!(*open_flags & O_CREAT)); | |
922 | fd = qemu_open(normalized_filename, *open_flags); | |
923 | if (fd == -1) { | |
924 | error_setg_errno(errp, errno, "Could not reopen file"); | |
925 | return -1; | |
926 | } | |
927 | } | |
928 | } | |
929 | ||
930 | return fd; | |
931 | } | |
932 | ||
eeb6b45d JC |
933 | static int raw_reopen_prepare(BDRVReopenState *state, |
934 | BlockReopenQueue *queue, Error **errp) | |
935 | { | |
936 | BDRVRawState *s; | |
4e6d13c9 | 937 | BDRVRawReopenState *rs; |
31be8a2a | 938 | QemuOpts *opts; |
a6aeca0c | 939 | int ret; |
df26a350 | 940 | Error *local_err = NULL; |
eeb6b45d JC |
941 | |
942 | assert(state != NULL); | |
943 | assert(state->bs != NULL); | |
944 | ||
945 | s = state->bs->opaque; | |
946 | ||
5839e53b | 947 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 948 | rs = state->opaque; |
31be8a2a SH |
949 | |
950 | /* Handle options changes */ | |
951 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
952 | qemu_opts_absorb_qdict(opts, state->options, &local_err); | |
953 | if (local_err) { | |
954 | error_propagate(errp, local_err); | |
955 | ret = -EINVAL; | |
956 | goto out; | |
957 | } | |
958 | ||
f357fcd8 | 959 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
960 | rs->check_cache_dropped = |
961 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
962 | ||
963 | /* This driver's reopen function doesn't currently allow changing | |
964 | * other options, so let's put them back in the original QDict and | |
965 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
966 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 967 | |
5cec2870 | 968 | rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, |
23dece19 | 969 | state->perm, true, &local_err); |
5cec2870 KW |
970 | if (local_err) { |
971 | error_propagate(errp, local_err); | |
972 | ret = -1; | |
973 | goto out; | |
eeb6b45d | 974 | } |
df26a350 KW |
975 | |
976 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
977 | * alignment with the new fd. */ | |
4e6d13c9 FZ |
978 | if (rs->fd != -1) { |
979 | raw_probe_alignment(state->bs, rs->fd, &local_err); | |
df26a350 | 980 | if (local_err) { |
df26a350 KW |
981 | error_propagate(errp, local_err); |
982 | ret = -EINVAL; | |
a6aeca0c KW |
983 | goto out_fd; |
984 | } | |
df26a350 KW |
985 | } |
986 | ||
e0c9cf3a | 987 | s->reopen_state = state; |
a6aeca0c KW |
988 | ret = 0; |
989 | out_fd: | |
990 | if (ret < 0) { | |
991 | qemu_close(rs->fd); | |
992 | rs->fd = -1; | |
993 | } | |
31be8a2a SH |
994 | out: |
995 | qemu_opts_del(opts); | |
eeb6b45d JC |
996 | return ret; |
997 | } | |
998 | ||
eeb6b45d JC |
999 | static void raw_reopen_commit(BDRVReopenState *state) |
1000 | { | |
4e6d13c9 | 1001 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1002 | BDRVRawState *s = state->bs->opaque; |
1003 | ||
f357fcd8 | 1004 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1005 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1006 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1007 | |
1008 | qemu_close(s->fd); | |
4e6d13c9 | 1009 | s->fd = rs->fd; |
eeb6b45d JC |
1010 | |
1011 | g_free(state->opaque); | |
1012 | state->opaque = NULL; | |
e0c9cf3a KW |
1013 | |
1014 | assert(s->reopen_state == state); | |
1015 | s->reopen_state = NULL; | |
eeb6b45d JC |
1016 | } |
1017 | ||
1018 | ||
1019 | static void raw_reopen_abort(BDRVReopenState *state) | |
1020 | { | |
4e6d13c9 | 1021 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1022 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1023 | |
1024 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1025 | if (rs == NULL) { |
eeb6b45d JC |
1026 | return; |
1027 | } | |
1028 | ||
4e6d13c9 FZ |
1029 | if (rs->fd >= 0) { |
1030 | qemu_close(rs->fd); | |
1031 | rs->fd = -1; | |
eeb6b45d JC |
1032 | } |
1033 | g_free(state->opaque); | |
1034 | state->opaque = NULL; | |
e0c9cf3a KW |
1035 | |
1036 | assert(s->reopen_state == state); | |
1037 | s->reopen_state = NULL; | |
eeb6b45d JC |
1038 | } |
1039 | ||
48265250 | 1040 | static int hdev_get_max_transfer_length(BlockDriverState *bs, int fd) |
6f607174 FZ |
1041 | { |
1042 | #ifdef BLKSECTGET | |
48265250 EF |
1043 | int max_bytes = 0; |
1044 | short max_sectors = 0; | |
1045 | if (bs->sg && ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
1046 | return max_bytes; | |
1047 | } else if (!bs->sg && ioctl(fd, BLKSECTGET, &max_sectors) == 0) { | |
1048 | return max_sectors << BDRV_SECTOR_BITS; | |
6f607174 FZ |
1049 | } else { |
1050 | return -errno; | |
1051 | } | |
1052 | #else | |
1053 | return -ENOSYS; | |
1054 | #endif | |
1055 | } | |
1056 | ||
9103f1ce FZ |
1057 | static int hdev_get_max_segments(const struct stat *st) |
1058 | { | |
1059 | #ifdef CONFIG_LINUX | |
1060 | char buf[32]; | |
1061 | const char *end; | |
1062 | char *sysfspath; | |
1063 | int ret; | |
1064 | int fd = -1; | |
1065 | long max_segments; | |
1066 | ||
1067 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", | |
1068 | major(st->st_rdev), minor(st->st_rdev)); | |
1069 | fd = open(sysfspath, O_RDONLY); | |
1070 | if (fd == -1) { | |
1071 | ret = -errno; | |
1072 | goto out; | |
1073 | } | |
1074 | do { | |
69583490 | 1075 | ret = read(fd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1076 | } while (ret == -1 && errno == EINTR); |
1077 | if (ret < 0) { | |
1078 | ret = -errno; | |
1079 | goto out; | |
1080 | } else if (ret == 0) { | |
1081 | ret = -EIO; | |
1082 | goto out; | |
1083 | } | |
1084 | buf[ret] = 0; | |
1085 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1086 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1087 | if (ret == 0 && end && *end == '\n') { | |
1088 | ret = max_segments; | |
1089 | } | |
1090 | ||
1091 | out: | |
fed414df FZ |
1092 | if (fd != -1) { |
1093 | close(fd); | |
1094 | } | |
9103f1ce FZ |
1095 | g_free(sysfspath); |
1096 | return ret; | |
1097 | #else | |
1098 | return -ENOTSUP; | |
1099 | #endif | |
1100 | } | |
1101 | ||
3baca891 | 1102 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1103 | { |
1104 | BDRVRawState *s = bs->opaque; | |
6f607174 FZ |
1105 | struct stat st; |
1106 | ||
1107 | if (!fstat(s->fd, &st)) { | |
c4c41a0a | 1108 | if (S_ISBLK(st.st_mode) || S_ISCHR(st.st_mode)) { |
48265250 EF |
1109 | int ret = hdev_get_max_transfer_length(bs, s->fd); |
1110 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
1111 | bs->bl.max_transfer = pow2floor(ret); | |
6f607174 | 1112 | } |
9103f1ce FZ |
1113 | ret = hdev_get_max_segments(&st); |
1114 | if (ret > 0) { | |
1115 | bs->bl.max_transfer = MIN(bs->bl.max_transfer, | |
1116 | ret * getpagesize()); | |
1117 | } | |
6f607174 FZ |
1118 | } |
1119 | } | |
eeb6b45d | 1120 | |
df26a350 | 1121 | raw_probe_alignment(bs, s->fd, errp); |
4196d2f0 | 1122 | bs->bl.min_mem_alignment = s->buf_align; |
459b4e66 | 1123 | bs->bl.opt_mem_alignment = MAX(s->buf_align, getpagesize()); |
c25f53b0 | 1124 | } |
83f64091 | 1125 | |
1a9335e4 ET |
1126 | static int check_for_dasd(int fd) |
1127 | { | |
1128 | #ifdef BIODASDINFO2 | |
1129 | struct dasd_information2_t info = {0}; | |
1130 | ||
1131 | return ioctl(fd, BIODASDINFO2, &info); | |
1132 | #else | |
1133 | return -1; | |
1134 | #endif | |
1135 | } | |
1136 | ||
1137 | /** | |
1138 | * Try to get @bs's logical and physical block size. | |
1139 | * On success, store them in @bsz and return zero. | |
1140 | * On failure, return negative errno. | |
1141 | */ | |
1142 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1143 | { | |
1144 | BDRVRawState *s = bs->opaque; | |
1145 | int ret; | |
1146 | ||
1147 | /* If DASD, get blocksizes */ | |
1148 | if (check_for_dasd(s->fd) < 0) { | |
1149 | return -ENOTSUP; | |
1150 | } | |
1151 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1152 | if (ret < 0) { | |
1153 | return ret; | |
1154 | } | |
1155 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1156 | } | |
1157 | ||
1158 | /** | |
1159 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1160 | * On success, store them in @geo and return 0. | |
1161 | * On failure return -errno. | |
1162 | * (Allows block driver to assign default geometry values that guest sees) | |
1163 | */ | |
1164 | #ifdef __linux__ | |
1165 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1166 | { | |
1167 | BDRVRawState *s = bs->opaque; | |
1168 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1169 | |
1170 | /* If DASD, get its geometry */ | |
1171 | if (check_for_dasd(s->fd) < 0) { | |
1172 | return -ENOTSUP; | |
1173 | } | |
1174 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1175 | return -errno; | |
1176 | } | |
1177 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1178 | (e.g. certain multipath setups) */ | |
1179 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1180 | return -ENOTSUP; | |
1181 | } | |
1182 | /* Do not return a geometry for partition */ | |
1183 | if (ioctl_geo.start != 0) { | |
1184 | return -ENOTSUP; | |
1185 | } | |
1186 | geo->heads = ioctl_geo.heads; | |
1187 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1188 | geo->cylinders = ioctl_geo.cylinders; |
1189 | ||
1190 | return 0; | |
1191 | } | |
1192 | #else /* __linux__ */ | |
1193 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1194 | { | |
1195 | return -ENOTSUP; | |
1196 | } | |
1197 | #endif | |
1198 | ||
03425671 KW |
1199 | #if defined(__linux__) |
1200 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1201 | { |
03425671 | 1202 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1203 | int ret; |
1204 | ||
d57c44d0 | 1205 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); |
de81a169 PB |
1206 | if (ret == -1) { |
1207 | return -errno; | |
1208 | } | |
1209 | ||
b608c8dc | 1210 | return 0; |
de81a169 | 1211 | } |
03425671 | 1212 | #endif /* linux */ |
de81a169 | 1213 | |
06dc9bd5 | 1214 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1215 | { |
06dc9bd5 | 1216 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1217 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1218 | int ret; |
1219 | ||
e5bcf967 KW |
1220 | if (s->page_cache_inconsistent) { |
1221 | return -EIO; | |
1222 | } | |
1223 | ||
de81a169 PB |
1224 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1225 | if (ret == -1) { | |
e5bcf967 KW |
1226 | /* There is no clear definition of the semantics of a failing fsync(), |
1227 | * so we may have to assume the worst. The sad truth is that this | |
1228 | * assumption is correct for Linux. Some pages are now probably marked | |
1229 | * clean in the page cache even though they are inconsistent with the | |
1230 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1231 | * further writeback attempt will be made. We can't get back to a state | |
1232 | * in which we know what is on disk (we would have to rewrite | |
1233 | * everything that was touched since the last fdatasync() at least), so | |
1234 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1235 | * really defined, I have little hope that other OSes are doing better. | |
1236 | * | |
1237 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1238 | * cache. */ | |
1239 | if ((s->open_flags & O_DIRECT) == 0) { | |
1240 | s->page_cache_inconsistent = true; | |
1241 | } | |
de81a169 PB |
1242 | return -errno; |
1243 | } | |
1244 | return 0; | |
1245 | } | |
1246 | ||
1247 | #ifdef CONFIG_PREADV | |
1248 | ||
1249 | static bool preadv_present = true; | |
1250 | ||
1251 | static ssize_t | |
1252 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1253 | { | |
1254 | return preadv(fd, iov, nr_iov, offset); | |
1255 | } | |
1256 | ||
1257 | static ssize_t | |
1258 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1259 | { | |
1260 | return pwritev(fd, iov, nr_iov, offset); | |
1261 | } | |
1262 | ||
1263 | #else | |
1264 | ||
1265 | static bool preadv_present = false; | |
1266 | ||
1267 | static ssize_t | |
1268 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1269 | { | |
1270 | return -ENOSYS; | |
1271 | } | |
1272 | ||
1273 | static ssize_t | |
1274 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1275 | { | |
1276 | return -ENOSYS; | |
1277 | } | |
1278 | ||
1279 | #endif | |
1280 | ||
1281 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1282 | { | |
1283 | ssize_t len; | |
1284 | ||
1285 | do { | |
1286 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1287 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1288 | aiocb->io.iov, |
1289 | aiocb->io.niov, | |
de81a169 PB |
1290 | aiocb->aio_offset); |
1291 | else | |
1292 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1293 | aiocb->io.iov, |
1294 | aiocb->io.niov, | |
de81a169 PB |
1295 | aiocb->aio_offset); |
1296 | } while (len == -1 && errno == EINTR); | |
1297 | ||
1298 | if (len == -1) { | |
1299 | return -errno; | |
1300 | } | |
1301 | return len; | |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * Read/writes the data to/from a given linear buffer. | |
1306 | * | |
1307 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1308 | * reads are only returned if the end of the file is reached. | |
1309 | */ | |
1310 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1311 | { | |
1312 | ssize_t offset = 0; | |
1313 | ssize_t len; | |
1314 | ||
1315 | while (offset < aiocb->aio_nbytes) { | |
1316 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1317 | len = pwrite(aiocb->aio_fildes, | |
1318 | (const char *)buf + offset, | |
1319 | aiocb->aio_nbytes - offset, | |
1320 | aiocb->aio_offset + offset); | |
1321 | } else { | |
1322 | len = pread(aiocb->aio_fildes, | |
1323 | buf + offset, | |
1324 | aiocb->aio_nbytes - offset, | |
1325 | aiocb->aio_offset + offset); | |
1326 | } | |
1327 | if (len == -1 && errno == EINTR) { | |
1328 | continue; | |
61ed73cf SH |
1329 | } else if (len == -1 && errno == EINVAL && |
1330 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1331 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1332 | offset > 0) { | |
1333 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1334 | * after a short read. Assume that O_DIRECT short reads only occur | |
1335 | * at EOF. Therefore this is a short read, not an I/O error. | |
1336 | */ | |
1337 | break; | |
de81a169 PB |
1338 | } else if (len == -1) { |
1339 | offset = -errno; | |
1340 | break; | |
1341 | } else if (len == 0) { | |
1342 | break; | |
1343 | } | |
1344 | offset += len; | |
1345 | } | |
1346 | ||
1347 | return offset; | |
1348 | } | |
1349 | ||
999e6b69 | 1350 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1351 | { |
999e6b69 | 1352 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1353 | ssize_t nbytes; |
1354 | char *buf; | |
1355 | ||
1356 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1357 | /* | |
1358 | * If there is just a single buffer, and it is properly aligned | |
1359 | * we can just use plain pread/pwrite without any problems. | |
1360 | */ | |
d57c44d0 | 1361 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1362 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1363 | goto out; | |
de81a169 PB |
1364 | } |
1365 | /* | |
1366 | * We have more than one iovec, and all are properly aligned. | |
1367 | * | |
1368 | * Try preadv/pwritev first and fall back to linearizing the | |
1369 | * buffer if it's not supported. | |
1370 | */ | |
1371 | if (preadv_present) { | |
1372 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1373 | if (nbytes == aiocb->aio_nbytes || | |
1374 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1375 | goto out; |
de81a169 PB |
1376 | } |
1377 | preadv_present = false; | |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1382 | * using these interfaces. For now retry using plain | |
1383 | * pread/pwrite? | |
1384 | */ | |
1385 | } | |
1386 | ||
1387 | /* | |
1388 | * Ok, we have to do it the hard way, copy all segments into | |
1389 | * a single aligned buffer. | |
1390 | */ | |
50d4a858 KW |
1391 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1392 | if (buf == NULL) { | |
54c7ca1b KW |
1393 | nbytes = -ENOMEM; |
1394 | goto out; | |
50d4a858 KW |
1395 | } |
1396 | ||
de81a169 PB |
1397 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1398 | char *p = buf; | |
1399 | int i; | |
1400 | ||
d57c44d0 KW |
1401 | for (i = 0; i < aiocb->io.niov; ++i) { |
1402 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1403 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1404 | } |
8eb029c2 | 1405 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1406 | } |
1407 | ||
1408 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1409 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1410 | char *p = buf; | |
1411 | size_t count = aiocb->aio_nbytes, copy; | |
1412 | int i; | |
1413 | ||
d57c44d0 | 1414 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1415 | copy = count; |
d57c44d0 KW |
1416 | if (copy > aiocb->io.iov[i].iov_len) { |
1417 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1418 | } |
d57c44d0 | 1419 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1420 | assert(count >= copy); |
de81a169 PB |
1421 | p += copy; |
1422 | count -= copy; | |
1423 | } | |
8eb029c2 | 1424 | assert(count == 0); |
de81a169 PB |
1425 | } |
1426 | qemu_vfree(buf); | |
1427 | ||
54c7ca1b KW |
1428 | out: |
1429 | if (nbytes == aiocb->aio_nbytes) { | |
1430 | return 0; | |
1431 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1432 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1433 | return -EINVAL; | |
1434 | } else { | |
1435 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1436 | 0, aiocb->aio_nbytes - nbytes); | |
1437 | return 0; | |
1438 | } | |
1439 | } else { | |
1440 | assert(nbytes < 0); | |
1441 | return nbytes; | |
1442 | } | |
de81a169 PB |
1443 | } |
1444 | ||
8238010b | 1445 | #ifdef CONFIG_XFS |
97a2ae34 PB |
1446 | static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes) |
1447 | { | |
50ba5b2d | 1448 | int64_t len; |
97a2ae34 | 1449 | struct xfs_flock64 fl; |
bcb22555 | 1450 | int err; |
97a2ae34 | 1451 | |
50ba5b2d HR |
1452 | len = lseek(s->fd, 0, SEEK_END); |
1453 | if (len < 0) { | |
1454 | return -errno; | |
1455 | } | |
1456 | ||
1457 | if (offset + bytes > len) { | |
1458 | /* XFS_IOC_ZERO_RANGE does not increase the file length */ | |
1459 | if (ftruncate(s->fd, offset + bytes) < 0) { | |
1460 | return -errno; | |
1461 | } | |
1462 | } | |
1463 | ||
97a2ae34 PB |
1464 | memset(&fl, 0, sizeof(fl)); |
1465 | fl.l_whence = SEEK_SET; | |
1466 | fl.l_start = offset; | |
1467 | fl.l_len = bytes; | |
1468 | ||
1469 | if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) { | |
bcb22555 | 1470 | err = errno; |
4f7d28d7 | 1471 | trace_file_xfs_write_zeroes(strerror(errno)); |
bcb22555 | 1472 | return -err; |
97a2ae34 PB |
1473 | } |
1474 | ||
1475 | return 0; | |
1476 | } | |
1477 | ||
8238010b PB |
1478 | static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes) |
1479 | { | |
1480 | struct xfs_flock64 fl; | |
bcb22555 | 1481 | int err; |
8238010b PB |
1482 | |
1483 | memset(&fl, 0, sizeof(fl)); | |
1484 | fl.l_whence = SEEK_SET; | |
1485 | fl.l_start = offset; | |
1486 | fl.l_len = bytes; | |
1487 | ||
1488 | if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) { | |
bcb22555 | 1489 | err = errno; |
4f7d28d7 | 1490 | trace_file_xfs_discard(strerror(errno)); |
bcb22555 | 1491 | return -err; |
8238010b PB |
1492 | } |
1493 | ||
1494 | return 0; | |
1495 | } | |
1496 | #endif | |
1497 | ||
1486df0e DL |
1498 | static int translate_err(int err) |
1499 | { | |
1500 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1501 | err == -ENOTTY) { | |
1502 | err = -ENOTSUP; | |
1503 | } | |
1504 | return err; | |
1505 | } | |
1506 | ||
d50d8222 | 1507 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1508 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1509 | { | |
1510 | do { | |
1511 | if (fallocate(fd, mode, offset, len) == 0) { | |
1512 | return 0; | |
1513 | } | |
1514 | } while (errno == EINTR); | |
1515 | return translate_err(-errno); | |
1516 | } | |
1517 | #endif | |
1518 | ||
37cc9f7f | 1519 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1520 | { |
37cc9f7f | 1521 | int ret = -ENOTSUP; |
97a2ae34 PB |
1522 | BDRVRawState *s = aiocb->bs->opaque; |
1523 | ||
37cc9f7f | 1524 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1525 | return -ENOTSUP; |
1526 | } | |
1527 | ||
97a2ae34 | 1528 | #ifdef BLKZEROOUT |
738301e1 KW |
1529 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1530 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1531 | * fallbacks. */ | |
1532 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1533 | do { | |
1534 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1535 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1536 | return 0; | |
1537 | } | |
1538 | } while (errno == EINTR); | |
37cc9f7f | 1539 | |
738301e1 KW |
1540 | ret = translate_err(-errno); |
1541 | } | |
97a2ae34 | 1542 | #endif |
97a2ae34 | 1543 | |
1486df0e | 1544 | if (ret == -ENOTSUP) { |
97a2ae34 | 1545 | s->has_write_zeroes = false; |
97a2ae34 PB |
1546 | } |
1547 | return ret; | |
1548 | } | |
1549 | ||
7154d8ae | 1550 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1551 | { |
7154d8ae | 1552 | RawPosixAIOData *aiocb = opaque; |
a6dcf097 | 1553 | #if defined(CONFIG_FALLOCATE) || defined(CONFIG_XFS) |
37cc9f7f | 1554 | BDRVRawState *s = aiocb->bs->opaque; |
a6dcf097 | 1555 | #endif |
70d9110b DL |
1556 | #ifdef CONFIG_FALLOCATE |
1557 | int64_t len; | |
1558 | #endif | |
37cc9f7f DL |
1559 | |
1560 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1561 | return handle_aiocb_write_zeroes_block(aiocb); | |
1562 | } | |
1563 | ||
1564 | #ifdef CONFIG_XFS | |
1565 | if (s->is_xfs) { | |
1566 | return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1567 | } | |
1568 | #endif | |
1569 | ||
b953f075 DL |
1570 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1571 | if (s->has_write_zeroes) { | |
1572 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1573 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1574 | if (ret == 0 || ret != -ENOTSUP) { | |
1575 | return ret; | |
1576 | } | |
1577 | s->has_write_zeroes = false; | |
1578 | } | |
1579 | #endif | |
1580 | ||
1cdc3239 DL |
1581 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1582 | if (s->has_discard && s->has_fallocate) { | |
1583 | int ret = do_fallocate(s->fd, | |
1584 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1585 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1586 | if (ret == 0) { | |
1587 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1588 | if (ret == 0 || ret != -ENOTSUP) { | |
1589 | return ret; | |
1590 | } | |
1591 | s->has_fallocate = false; | |
1592 | } else if (ret != -ENOTSUP) { | |
1593 | return ret; | |
1594 | } else { | |
1595 | s->has_discard = false; | |
1596 | } | |
1597 | } | |
1598 | #endif | |
1599 | ||
d50d8222 | 1600 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1601 | /* Last resort: we are trying to extend the file with zeroed data. This |
1602 | * can be done via fallocate(fd, 0) */ | |
1603 | len = bdrv_getlength(aiocb->bs); | |
1604 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1605 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1606 | if (ret == 0 || ret != -ENOTSUP) { | |
1607 | return ret; | |
1608 | } | |
1609 | s->has_fallocate = false; | |
1610 | } | |
1611 | #endif | |
1612 | ||
37cc9f7f DL |
1613 | return -ENOTSUP; |
1614 | } | |
1615 | ||
7154d8ae | 1616 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1617 | { |
7154d8ae | 1618 | RawPosixAIOData *aiocb = opaque; |
34fa110e KW |
1619 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
1620 | int ret; | |
1621 | ||
1622 | /* First try to write zeros and unmap at the same time */ | |
1623 | ||
1624 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
1625 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1626 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1627 | if (ret != -ENOTSUP) { | |
1628 | return ret; | |
1629 | } | |
1630 | #endif | |
1631 | ||
1632 | #ifdef CONFIG_XFS | |
1633 | if (s->is_xfs) { | |
1634 | /* xfs_discard() guarantees that the discarded area reads as all-zero | |
1635 | * afterwards, so we can use it here. */ | |
1636 | return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1637 | } | |
1638 | #endif | |
1639 | ||
1640 | /* If we couldn't manage to unmap while guaranteed that the area reads as | |
1641 | * all-zero afterwards, just write zeroes without unmapping */ | |
1642 | ret = handle_aiocb_write_zeroes(aiocb); | |
1643 | return ret; | |
1644 | } | |
1645 | ||
1efad060 FZ |
1646 | #ifndef HAVE_COPY_FILE_RANGE |
1647 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1648 | off_t *out_off, size_t len, unsigned int flags) | |
1649 | { | |
1650 | #ifdef __NR_copy_file_range | |
1651 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1652 | out_off, len, flags); | |
1653 | #else | |
1654 | errno = ENOSYS; | |
1655 | return -1; | |
1656 | #endif | |
1657 | } | |
1658 | #endif | |
1659 | ||
58a209c4 | 1660 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1661 | { |
58a209c4 | 1662 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1663 | uint64_t bytes = aiocb->aio_nbytes; |
1664 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1665 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1666 | |
1667 | while (bytes) { | |
1668 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1669 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1670 | bytes, 0); |
ecc983a5 | 1671 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1672 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1673 | 0, ret); | |
c436e3d0 FZ |
1674 | if (ret == 0) { |
1675 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1676 | * buffer I/O. */ | |
1677 | return -ENOSPC; | |
1efad060 FZ |
1678 | } |
1679 | if (ret < 0) { | |
c436e3d0 FZ |
1680 | switch (errno) { |
1681 | case ENOSYS: | |
1efad060 | 1682 | return -ENOTSUP; |
c436e3d0 FZ |
1683 | case EINTR: |
1684 | continue; | |
1685 | default: | |
1efad060 FZ |
1686 | return -errno; |
1687 | } | |
1688 | } | |
1efad060 FZ |
1689 | bytes -= ret; |
1690 | } | |
1691 | return 0; | |
1692 | } | |
1693 | ||
46ee0f46 | 1694 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1695 | { |
46ee0f46 | 1696 | RawPosixAIOData *aiocb = opaque; |
8238010b PB |
1697 | int ret = -EOPNOTSUPP; |
1698 | BDRVRawState *s = aiocb->bs->opaque; | |
1699 | ||
7ce21016 PB |
1700 | if (!s->has_discard) { |
1701 | return -ENOTSUP; | |
8238010b PB |
1702 | } |
1703 | ||
1704 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1705 | #ifdef BLKDISCARD | |
1706 | do { | |
1707 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1708 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1709 | return 0; | |
1710 | } | |
1711 | } while (errno == EINTR); | |
1712 | ||
1713 | ret = -errno; | |
1714 | #endif | |
1715 | } else { | |
1716 | #ifdef CONFIG_XFS | |
1717 | if (s->is_xfs) { | |
1718 | return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes); | |
1719 | } | |
1720 | #endif | |
1721 | ||
1722 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
0b991712 DL |
1723 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1724 | aiocb->aio_offset, aiocb->aio_nbytes); | |
8238010b PB |
1725 | #endif |
1726 | } | |
1727 | ||
1486df0e DL |
1728 | ret = translate_err(ret); |
1729 | if (ret == -ENOTSUP) { | |
7ce21016 | 1730 | s->has_discard = false; |
8238010b PB |
1731 | } |
1732 | return ret; | |
1733 | } | |
1734 | ||
29cb4c01 | 1735 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1736 | { |
29cb4c01 | 1737 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1738 | int result = 0; |
1739 | int64_t current_length = 0; | |
1740 | char *buf = NULL; | |
1741 | struct stat st; | |
1742 | int fd = aiocb->aio_fildes; | |
1743 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1744 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1745 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1746 | |
1747 | if (fstat(fd, &st) < 0) { | |
1748 | result = -errno; | |
1749 | error_setg_errno(errp, -result, "Could not stat file"); | |
1750 | return result; | |
1751 | } | |
1752 | ||
1753 | current_length = st.st_size; | |
d57c44d0 | 1754 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1755 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1756 | return -ENOTSUP; | |
1757 | } | |
1758 | ||
d57c44d0 | 1759 | switch (prealloc) { |
93f4e2ff KW |
1760 | #ifdef CONFIG_POSIX_FALLOCATE |
1761 | case PREALLOC_MODE_FALLOC: | |
1762 | /* | |
1763 | * Truncating before posix_fallocate() makes it about twice slower on | |
1764 | * file systems that do not support fallocate(), trying to check if a | |
1765 | * block is allocated before allocating it, so don't do that here. | |
1766 | */ | |
1767 | if (offset != current_length) { | |
1768 | result = -posix_fallocate(fd, current_length, | |
1769 | offset - current_length); | |
1770 | if (result != 0) { | |
1771 | /* posix_fallocate() doesn't set errno. */ | |
1772 | error_setg_errno(errp, -result, | |
1773 | "Could not preallocate new data"); | |
1774 | } | |
1775 | } else { | |
1776 | result = 0; | |
1777 | } | |
1778 | goto out; | |
1779 | #endif | |
1780 | case PREALLOC_MODE_FULL: | |
1781 | { | |
1782 | int64_t num = 0, left = offset - current_length; | |
1783 | off_t seek_result; | |
1784 | ||
1785 | /* | |
1786 | * Knowing the final size from the beginning could allow the file | |
1787 | * system driver to do less allocations and possibly avoid | |
1788 | * fragmentation of the file. | |
1789 | */ | |
1790 | if (ftruncate(fd, offset) != 0) { | |
1791 | result = -errno; | |
1792 | error_setg_errno(errp, -result, "Could not resize file"); | |
1793 | goto out; | |
1794 | } | |
1795 | ||
1796 | buf = g_malloc0(65536); | |
1797 | ||
1798 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1799 | if (seek_result < 0) { | |
1800 | result = -errno; | |
1801 | error_setg_errno(errp, -result, | |
1802 | "Failed to seek to the old end of file"); | |
1803 | goto out; | |
1804 | } | |
1805 | ||
1806 | while (left > 0) { | |
1807 | num = MIN(left, 65536); | |
1808 | result = write(fd, buf, num); | |
1809 | if (result < 0) { | |
a1c81f4f FZ |
1810 | if (errno == EINTR) { |
1811 | continue; | |
1812 | } | |
93f4e2ff KW |
1813 | result = -errno; |
1814 | error_setg_errno(errp, -result, | |
1815 | "Could not write zeros for preallocation"); | |
1816 | goto out; | |
1817 | } | |
1818 | left -= result; | |
1819 | } | |
1820 | if (result >= 0) { | |
1821 | result = fsync(fd); | |
1822 | if (result < 0) { | |
1823 | result = -errno; | |
1824 | error_setg_errno(errp, -result, | |
1825 | "Could not flush file to disk"); | |
1826 | goto out; | |
1827 | } | |
1828 | } | |
1829 | goto out; | |
1830 | } | |
1831 | case PREALLOC_MODE_OFF: | |
1832 | if (ftruncate(fd, offset) != 0) { | |
1833 | result = -errno; | |
1834 | error_setg_errno(errp, -result, "Could not resize file"); | |
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, | |
2755 | .help = "Preallocation mode (allowed values: off, falloc, full)" | |
2756 | }, | |
6f482f74 CL |
2757 | { /* end of list */ } |
2758 | } | |
0e7e1989 KW |
2759 | }; |
2760 | ||
244a5668 FZ |
2761 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
2762 | Error **errp) | |
2763 | { | |
6ceabe6f KW |
2764 | BDRVRawState *s = bs->opaque; |
2765 | BDRVRawReopenState *rs = NULL; | |
2766 | int open_flags; | |
2767 | int ret; | |
2768 | ||
2769 | if (s->perm_change_fd) { | |
2770 | /* | |
2771 | * In the context of reopen, this function may be called several times | |
2772 | * (directly and recursively while change permissions of the parent). | |
2773 | * This is even true for children that don't inherit from the original | |
2774 | * reopen node, so s->reopen_state is not set. | |
2775 | * | |
2776 | * Ignore all but the first call. | |
2777 | */ | |
2778 | return 0; | |
2779 | } | |
2780 | ||
2781 | if (s->reopen_state) { | |
2782 | /* We already have a new file descriptor to set permissions for */ | |
2783 | assert(s->reopen_state->perm == perm); | |
2784 | assert(s->reopen_state->shared_perm == shared); | |
2785 | rs = s->reopen_state->opaque; | |
2786 | s->perm_change_fd = rs->fd; | |
2787 | } else { | |
2788 | /* We may need a new fd if auto-read-only switches the mode */ | |
23dece19 | 2789 | ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, |
6ceabe6f KW |
2790 | false, errp); |
2791 | if (ret < 0) { | |
2792 | return ret; | |
2793 | } else if (ret != s->fd) { | |
2794 | s->perm_change_fd = ret; | |
2795 | } | |
2796 | } | |
2797 | ||
2798 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
2799 | * but keep everything locked that new will need. */ | |
2800 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
2801 | if (ret < 0) { | |
2802 | goto fail; | |
2803 | } | |
2804 | ||
2805 | /* Copy locks to the new fd */ | |
2806 | if (s->perm_change_fd) { | |
2807 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, | |
2808 | false, errp); | |
2809 | if (ret < 0) { | |
2810 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
2811 | goto fail; | |
2812 | } | |
2813 | } | |
2814 | return 0; | |
2815 | ||
2816 | fail: | |
2817 | if (s->perm_change_fd && !s->reopen_state) { | |
2818 | qemu_close(s->perm_change_fd); | |
2819 | } | |
2820 | s->perm_change_fd = 0; | |
2821 | return ret; | |
244a5668 FZ |
2822 | } |
2823 | ||
2824 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
2825 | { | |
2826 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
2827 | |
2828 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
2829 | * called after .bdrv_reopen_commit) */ | |
2830 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
2831 | qemu_close(s->fd); | |
2832 | s->fd = s->perm_change_fd; | |
2833 | } | |
2834 | s->perm_change_fd = 0; | |
2835 | ||
244a5668 FZ |
2836 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
2837 | s->perm = perm; | |
2838 | s->shared_perm = shared; | |
2839 | } | |
2840 | ||
2841 | static void raw_abort_perm_update(BlockDriverState *bs) | |
2842 | { | |
6ceabe6f KW |
2843 | BDRVRawState *s = bs->opaque; |
2844 | ||
2845 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
2846 | * the file descriptor. */ | |
2847 | if (s->perm_change_fd && !s->reopen_state) { | |
2848 | qemu_close(s->perm_change_fd); | |
2849 | } | |
2850 | s->perm_change_fd = 0; | |
2851 | ||
244a5668 FZ |
2852 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
2853 | } | |
2854 | ||
67b51fb9 VSO |
2855 | static int coroutine_fn raw_co_copy_range_from( |
2856 | BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, | |
2857 | BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, | |
2858 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) | |
1efad060 | 2859 | { |
67b51fb9 VSO |
2860 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
2861 | read_flags, write_flags); | |
1efad060 FZ |
2862 | } |
2863 | ||
2864 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 VSO |
2865 | BdrvChild *src, |
2866 | uint64_t src_offset, | |
2867 | BdrvChild *dst, | |
2868 | uint64_t dst_offset, | |
2869 | uint64_t bytes, | |
2870 | BdrvRequestFlags read_flags, | |
2871 | BdrvRequestFlags write_flags) | |
1efad060 | 2872 | { |
58a209c4 | 2873 | RawPosixAIOData acb; |
1efad060 FZ |
2874 | BDRVRawState *s = bs->opaque; |
2875 | BDRVRawState *src_s; | |
2876 | ||
2877 | assert(dst->bs == bs); | |
2878 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
2879 | return -ENOTSUP; | |
2880 | } | |
2881 | ||
2882 | src_s = src->bs->opaque; | |
9f850f67 | 2883 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
2884 | return -EIO; |
2885 | } | |
58a209c4 KW |
2886 | |
2887 | acb = (RawPosixAIOData) { | |
2888 | .bs = bs, | |
2889 | .aio_type = QEMU_AIO_COPY_RANGE, | |
2890 | .aio_fildes = src_s->fd, | |
2891 | .aio_offset = src_offset, | |
2892 | .aio_nbytes = bytes, | |
2893 | .copy_range = { | |
2894 | .aio_fd2 = s->fd, | |
2895 | .aio_offset2 = dst_offset, | |
2896 | }, | |
2897 | }; | |
2898 | ||
2899 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
2900 | } |
2901 | ||
5f535a94 | 2902 | BlockDriver bdrv_file = { |
84a12e66 CH |
2903 | .format_name = "file", |
2904 | .protocol_name = "file", | |
856ae5c3 | 2905 | .instance_size = sizeof(BDRVRawState), |
030be321 | 2906 | .bdrv_needs_filename = true, |
856ae5c3 | 2907 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 2908 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 2909 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
2910 | .bdrv_reopen_prepare = raw_reopen_prepare, |
2911 | .bdrv_reopen_commit = raw_reopen_commit, | |
2912 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 2913 | .bdrv_close = raw_close, |
927f11e1 | 2914 | .bdrv_co_create = raw_co_create, |
efc75e2a | 2915 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 2916 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 2917 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 2918 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 2919 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
3b46e624 | 2920 | |
9d52aa3c KW |
2921 | .bdrv_co_preadv = raw_co_preadv, |
2922 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
2923 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
2924 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
2925 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
2926 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 2927 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
2928 | .bdrv_io_plug = raw_aio_plug, |
2929 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 2930 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 2931 | |
061ca8a3 | 2932 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 2933 | .bdrv_getlength = raw_getlength, |
260a82e5 | 2934 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
2935 | .bdrv_get_allocated_file_size |
2936 | = raw_get_allocated_file_size, | |
244a5668 FZ |
2937 | .bdrv_check_perm = raw_check_perm, |
2938 | .bdrv_set_perm = raw_set_perm, | |
2939 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 2940 | .create_opts = &raw_create_opts, |
8a2ce0bc | 2941 | .mutable_opts = mutable_opts, |
83f64091 FB |
2942 | }; |
2943 | ||
19cb3738 FB |
2944 | /***********************************************/ |
2945 | /* host device */ | |
2946 | ||
83affaa6 | 2947 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
2948 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2949 | CFIndex maxPathSize, int flags); | |
d0855f12 | 2950 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 2951 | { |
d0855f12 | 2952 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
2953 | mach_port_t masterPort; |
2954 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
2955 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
2956 | char *mediaType = NULL; | |
19cb3738 FB |
2957 | |
2958 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
2959 | if ( KERN_SUCCESS != kernResult ) { | |
2960 | printf( "IOMasterPort returned %d\n", kernResult ); | |
2961 | } | |
3b46e624 | 2962 | |
d0855f12 JA |
2963 | int index; |
2964 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
2965 | classesToMatch = IOServiceMatching(matching_array[index]); | |
2966 | if (classesToMatch == NULL) { | |
2967 | error_report("IOServiceMatching returned NULL for %s", | |
2968 | matching_array[index]); | |
2969 | continue; | |
2970 | } | |
2971 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
2972 | kCFBooleanTrue); | |
2973 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
2974 | mediaIterator); | |
2975 | if (kernResult != KERN_SUCCESS) { | |
2976 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
2977 | kernResult); | |
2978 | continue; | |
2979 | } | |
3b46e624 | 2980 | |
d0855f12 JA |
2981 | /* If a match was found, leave the loop */ |
2982 | if (*mediaIterator != 0) { | |
4f7d28d7 | 2983 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
2984 | mediaType = g_strdup(matching_array[index]); |
2985 | break; | |
2986 | } | |
2987 | } | |
2988 | return mediaType; | |
19cb3738 FB |
2989 | } |
2990 | ||
98caa5bc JA |
2991 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
2992 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
2993 | { |
2994 | io_object_t nextMedia; | |
2995 | kern_return_t kernResult = KERN_FAILURE; | |
2996 | *bsdPath = '\0'; | |
2997 | nextMedia = IOIteratorNext( mediaIterator ); | |
2998 | if ( nextMedia ) | |
2999 | { | |
3000 | CFTypeRef bsdPathAsCFString; | |
3001 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3002 | if ( bsdPathAsCFString ) { | |
3003 | size_t devPathLength; | |
3004 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3005 | if (flags & BDRV_O_NOCACHE) { |
3006 | strcat(bsdPath, "r"); | |
3007 | } | |
19cb3738 FB |
3008 | devPathLength = strlen( bsdPath ); |
3009 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3010 | kernResult = KERN_SUCCESS; | |
3011 | } | |
3012 | CFRelease( bsdPathAsCFString ); | |
3013 | } | |
3014 | IOObjectRelease( nextMedia ); | |
3015 | } | |
3b46e624 | 3016 | |
19cb3738 FB |
3017 | return kernResult; |
3018 | } | |
3019 | ||
d0855f12 JA |
3020 | /* Sets up a real cdrom for use in QEMU */ |
3021 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3022 | { | |
3023 | int index, num_of_test_partitions = 2, fd; | |
3024 | char test_partition[MAXPATHLEN]; | |
3025 | bool partition_found = false; | |
3026 | ||
3027 | /* look for a working partition */ | |
3028 | for (index = 0; index < num_of_test_partitions; index++) { | |
3029 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3030 | index); | |
3031 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); | |
3032 | if (fd >= 0) { | |
3033 | partition_found = true; | |
3034 | qemu_close(fd); | |
3035 | break; | |
3036 | } | |
3037 | } | |
3038 | ||
3039 | /* if a working partition on the device was not found */ | |
3040 | if (partition_found == false) { | |
3041 | error_setg(errp, "Failed to find a working partition on disc"); | |
3042 | } else { | |
4f7d28d7 | 3043 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3044 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3045 | } | |
3046 | return partition_found; | |
3047 | } | |
3048 | ||
3049 | /* Prints directions on mounting and unmounting a device */ | |
3050 | static void print_unmounting_directions(const char *file_name) | |
3051 | { | |
3052 | error_report("If device %s is mounted on the desktop, unmount" | |
3053 | " it first before using it in QEMU", file_name); | |
3054 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3055 | file_name); | |
3056 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3057 | } | |
3058 | ||
3059 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3060 | |
508c7cb3 CH |
3061 | static int hdev_probe_device(const char *filename) |
3062 | { | |
3063 | struct stat st; | |
3064 | ||
3065 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3066 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3067 | return 50; | |
3068 | ||
3069 | if (stat(filename, &st) >= 0 && | |
3070 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3071 | return 100; | |
3072 | } | |
3073 | ||
3074 | return 0; | |
3075 | } | |
3076 | ||
da888d37 SH |
3077 | static int check_hdev_writable(BDRVRawState *s) |
3078 | { | |
3079 | #if defined(BLKROGET) | |
3080 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
3081 | * This is independent of device node permissions and therefore open(2) | |
3082 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
3083 | * | |
3084 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
3085 | * check for read-only block devices so that Linux block devices behave | |
3086 | * properly. | |
3087 | */ | |
3088 | struct stat st; | |
3089 | int readonly = 0; | |
3090 | ||
3091 | if (fstat(s->fd, &st)) { | |
3092 | return -errno; | |
3093 | } | |
3094 | ||
3095 | if (!S_ISBLK(st.st_mode)) { | |
3096 | return 0; | |
3097 | } | |
3098 | ||
3099 | if (ioctl(s->fd, BLKROGET, &readonly) < 0) { | |
3100 | return -errno; | |
3101 | } | |
3102 | ||
3103 | if (readonly) { | |
3104 | return -EACCES; | |
3105 | } | |
3106 | #endif /* defined(BLKROGET) */ | |
3107 | return 0; | |
3108 | } | |
3109 | ||
7af803d4 HR |
3110 | static void hdev_parse_filename(const char *filename, QDict *options, |
3111 | Error **errp) | |
3112 | { | |
03c320d8 | 3113 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3114 | } |
3115 | ||
3307ed7b DA |
3116 | static bool hdev_is_sg(BlockDriverState *bs) |
3117 | { | |
3118 | ||
3119 | #if defined(__linux__) | |
3120 | ||
0d4377b3 | 3121 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3122 | struct stat st; |
3123 | struct sg_scsi_id scsiid; | |
3124 | int sg_version; | |
0d4377b3 KW |
3125 | int ret; |
3126 | ||
3127 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3128 | return false; | |
3129 | } | |
3307ed7b | 3130 | |
0d4377b3 KW |
3131 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3132 | if (ret < 0) { | |
3133 | return false; | |
3134 | } | |
3135 | ||
3136 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3137 | if (ret >= 0) { | |
4f7d28d7 | 3138 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3139 | return true; |
3140 | } | |
3141 | ||
3142 | #endif | |
3143 | ||
3144 | return false; | |
3145 | } | |
3146 | ||
015a1036 HR |
3147 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3148 | Error **errp) | |
19cb3738 FB |
3149 | { |
3150 | BDRVRawState *s = bs->opaque; | |
e428e439 | 3151 | Error *local_err = NULL; |
da888d37 | 3152 | int ret; |
a76bab49 | 3153 | |
83affaa6 | 3154 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3155 | /* |
3156 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3157 | * would require more care. When @options come from -blockdev or | |
3158 | * blockdev_add, its members are typed according to the QAPI | |
3159 | * schema, but when they come from -drive, they're all QString. | |
3160 | */ | |
3307ed7b | 3161 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3162 | char bsd_path[MAXPATHLEN] = ""; |
3163 | bool error_occurred = false; | |
3164 | ||
3165 | /* If using a real cdrom */ | |
3166 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3167 | char *mediaType = NULL; | |
3168 | kern_return_t ret_val; | |
3169 | io_iterator_t mediaIterator = 0; | |
3170 | ||
3171 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3172 | if (mediaType == NULL) { | |
3173 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3174 | " drive"); | |
3175 | error_occurred = true; | |
3176 | goto hdev_open_Mac_error; | |
3177 | } | |
3307ed7b | 3178 | |
d0855f12 JA |
3179 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3180 | if (ret_val != KERN_SUCCESS) { | |
3181 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3182 | error_occurred = true; | |
3183 | goto hdev_open_Mac_error; | |
3184 | } | |
3185 | ||
3186 | /* If a real optical drive was not found */ | |
3187 | if (bsd_path[0] == '\0') { | |
3188 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3189 | error_occurred = true; | |
3190 | goto hdev_open_Mac_error; | |
3191 | } | |
3192 | ||
3193 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3194 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3195 | setup_cdrom(bsd_path, errp) == false) { | |
3196 | print_unmounting_directions(bsd_path); | |
3197 | error_occurred = true; | |
3198 | goto hdev_open_Mac_error; | |
19cb3738 | 3199 | } |
3b46e624 | 3200 | |
46f5ac20 | 3201 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3202 | |
3203 | hdev_open_Mac_error: | |
3204 | g_free(mediaType); | |
3205 | if (mediaIterator) { | |
3206 | IOObjectRelease(mediaIterator); | |
3207 | } | |
3208 | if (error_occurred) { | |
3209 | return -ENOENT; | |
3210 | } | |
19cb3738 | 3211 | } |
d0855f12 | 3212 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3213 | |
3214 | s->type = FTYPE_FILE; | |
90babde0 | 3215 | |
230ff739 | 3216 | ret = raw_open_common(bs, options, flags, 0, true, &local_err); |
da888d37 | 3217 | if (ret < 0) { |
621ff94d | 3218 | error_propagate(errp, local_err); |
d0855f12 JA |
3219 | #if defined(__APPLE__) && defined(__MACH__) |
3220 | if (*bsd_path) { | |
3221 | filename = bsd_path; | |
3222 | } | |
3223 | /* if a physical device experienced an error while being opened */ | |
3224 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3225 | print_unmounting_directions(filename); | |
3226 | } | |
3227 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3228 | return ret; |
3229 | } | |
3230 | ||
3307ed7b DA |
3231 | /* Since this does ioctl the device must be already opened */ |
3232 | bs->sg = hdev_is_sg(bs); | |
3233 | ||
da888d37 SH |
3234 | if (flags & BDRV_O_RDWR) { |
3235 | ret = check_hdev_writable(s); | |
3236 | if (ret < 0) { | |
3237 | raw_close(bs); | |
e428e439 | 3238 | error_setg_errno(errp, -ret, "The device is not writable"); |
da888d37 SH |
3239 | return ret; |
3240 | } | |
3241 | } | |
3242 | ||
3243 | return ret; | |
19cb3738 FB |
3244 | } |
3245 | ||
03ff3ca3 | 3246 | #if defined(__linux__) |
2f3a7ab3 KW |
3247 | static int coroutine_fn |
3248 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3249 | { |
f141eafe | 3250 | BDRVRawState *s = bs->opaque; |
03425671 | 3251 | RawPosixAIOData acb; |
2f3a7ab3 | 3252 | int ret; |
221f715d | 3253 | |
2f3a7ab3 KW |
3254 | ret = fd_open(bs); |
3255 | if (ret < 0) { | |
3256 | return ret; | |
3257 | } | |
c208e8c2 | 3258 | |
7c9e5276 PB |
3259 | if (req == SG_IO && s->pr_mgr) { |
3260 | struct sg_io_hdr *io_hdr = buf; | |
3261 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3262 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3263 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3264 | s->fd, io_hdr); |
7c9e5276 PB |
3265 | } |
3266 | } | |
3267 | ||
03425671 KW |
3268 | acb = (RawPosixAIOData) { |
3269 | .bs = bs, | |
3270 | .aio_type = QEMU_AIO_IOCTL, | |
3271 | .aio_fildes = s->fd, | |
3272 | .aio_offset = 0, | |
3273 | .ioctl = { | |
3274 | .buf = buf, | |
3275 | .cmd = req, | |
3276 | }, | |
3277 | }; | |
3278 | ||
3279 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3280 | } |
f709623b | 3281 | #endif /* linux */ |
221f715d | 3282 | |
9f23011a BS |
3283 | static int fd_open(BlockDriverState *bs) |
3284 | { | |
3285 | BDRVRawState *s = bs->opaque; | |
3286 | ||
3287 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
3288 | if (s->fd >= 0) | |
3289 | return 0; | |
3290 | return -EIO; | |
3291 | } | |
04eeb8b6 | 3292 | |
33d70fb6 KW |
3293 | static coroutine_fn int |
3294 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
c36dd8a0 | 3295 | { |
33d70fb6 | 3296 | int ret; |
c36dd8a0 | 3297 | |
33d70fb6 KW |
3298 | ret = fd_open(bs); |
3299 | if (ret < 0) { | |
3300 | return ret; | |
c36dd8a0 | 3301 | } |
46ee0f46 | 3302 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3303 | } |
3304 | ||
2ffa76c2 | 3305 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3306 | int64_t offset, int bytes, BdrvRequestFlags flags) |
d0b4503e | 3307 | { |
d0b4503e PB |
3308 | int rc; |
3309 | ||
3310 | rc = fd_open(bs); | |
3311 | if (rc < 0) { | |
3312 | return rc; | |
3313 | } | |
34fa110e | 3314 | |
7154d8ae | 3315 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3316 | } |
3317 | ||
efc75e2a SH |
3318 | static int coroutine_fn hdev_co_create_opts(const char *filename, QemuOpts *opts, |
3319 | Error **errp) | |
93c65b47 AL |
3320 | { |
3321 | int fd; | |
3322 | int ret = 0; | |
3323 | struct stat stat_buf; | |
0e7e1989 | 3324 | int64_t total_size = 0; |
cc28c6aa HR |
3325 | bool has_prefix; |
3326 | ||
f709623b HR |
3327 | /* This function is used by both protocol block drivers and therefore either |
3328 | * of these prefixes may be given. | |
cc28c6aa HR |
3329 | * The return value has to be stored somewhere, otherwise this is an error |
3330 | * due to -Werror=unused-value. */ | |
3331 | has_prefix = | |
3332 | strstart(filename, "host_device:", &filename) || | |
f709623b | 3333 | strstart(filename, "host_cdrom:" , &filename); |
cc28c6aa HR |
3334 | |
3335 | (void)has_prefix; | |
93c65b47 | 3336 | |
db0754df | 3337 | ret = raw_normalize_devicepath(&filename, errp); |
bdd03cdf | 3338 | if (ret < 0) { |
bdd03cdf HR |
3339 | return ret; |
3340 | } | |
3341 | ||
0e7e1989 | 3342 | /* Read out options */ |
180e9526 HT |
3343 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
3344 | BDRV_SECTOR_SIZE); | |
93c65b47 | 3345 | |
6165f4d8 | 3346 | fd = qemu_open(filename, O_WRONLY | O_BINARY); |
e428e439 HR |
3347 | if (fd < 0) { |
3348 | ret = -errno; | |
3349 | error_setg_errno(errp, -ret, "Could not open device"); | |
3350 | return ret; | |
3351 | } | |
93c65b47 | 3352 | |
e428e439 | 3353 | if (fstat(fd, &stat_buf) < 0) { |
57e69b7d | 3354 | ret = -errno; |
e428e439 HR |
3355 | error_setg_errno(errp, -ret, "Could not stat device"); |
3356 | } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) { | |
3357 | error_setg(errp, | |
3358 | "The given file is neither a block nor a character device"); | |
57e69b7d | 3359 | ret = -ENODEV; |
180e9526 | 3360 | } else if (lseek(fd, 0, SEEK_END) < total_size) { |
e428e439 | 3361 | error_setg(errp, "Device is too small"); |
93c65b47 | 3362 | ret = -ENOSPC; |
e428e439 | 3363 | } |
93c65b47 | 3364 | |
97ec9117 FZ |
3365 | if (!ret && total_size) { |
3366 | uint8_t buf[BDRV_SECTOR_SIZE] = { 0 }; | |
3367 | int64_t zero_size = MIN(BDRV_SECTOR_SIZE, total_size); | |
3368 | if (lseek(fd, 0, SEEK_SET) == -1) { | |
3369 | ret = -errno; | |
3370 | } else { | |
3371 | ret = qemu_write_full(fd, buf, zero_size); | |
3372 | ret = ret == zero_size ? 0 : -errno; | |
3373 | } | |
3374 | } | |
2e1e79da | 3375 | qemu_close(fd); |
93c65b47 AL |
3376 | return ret; |
3377 | } | |
3378 | ||
5efa9d5a | 3379 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3380 | .format_name = "host_device", |
84a12e66 | 3381 | .protocol_name = "host_device", |
0b4ce02e | 3382 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3383 | .bdrv_needs_filename = true, |
0b4ce02e | 3384 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3385 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3386 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3387 | .bdrv_close = raw_close, |
1bc6b705 JC |
3388 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3389 | .bdrv_reopen_commit = raw_reopen_commit, | |
3390 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3391 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3392 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3393 | .mutable_opts = mutable_opts, |
dd577a26 | 3394 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3395 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3396 | |
9d52aa3c KW |
3397 | .bdrv_co_preadv = raw_co_preadv, |
3398 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3399 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3400 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3401 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3402 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3403 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3404 | .bdrv_io_plug = raw_aio_plug, |
3405 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3406 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3407 | |
061ca8a3 | 3408 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3409 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3410 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3411 | .bdrv_get_allocated_file_size |
3412 | = raw_get_allocated_file_size, | |
244a5668 FZ |
3413 | .bdrv_check_perm = raw_check_perm, |
3414 | .bdrv_set_perm = raw_set_perm, | |
3415 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3416 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3417 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3418 | |
f3a5d3f8 | 3419 | /* generic scsi device */ |
63ec93db | 3420 | #ifdef __linux__ |
2f3a7ab3 | 3421 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3422 | #endif |
f3a5d3f8 CH |
3423 | }; |
3424 | ||
18fa1c42 HR |
3425 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3426 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3427 | Error **errp) | |
3428 | { | |
03c320d8 | 3429 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3430 | } |
3431 | #endif | |
3432 | ||
3433 | #ifdef __linux__ | |
015a1036 HR |
3434 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3435 | Error **errp) | |
f3a5d3f8 CH |
3436 | { |
3437 | BDRVRawState *s = bs->opaque; | |
3438 | ||
f3a5d3f8 CH |
3439 | s->type = FTYPE_CD; |
3440 | ||
19a3da7f | 3441 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3442 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3443 | } |
3444 | ||
508c7cb3 CH |
3445 | static int cdrom_probe_device(const char *filename) |
3446 | { | |
3baf720e CR |
3447 | int fd, ret; |
3448 | int prio = 0; | |
343f8568 | 3449 | struct stat st; |
3baf720e | 3450 | |
6165f4d8 | 3451 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
3baf720e CR |
3452 | if (fd < 0) { |
3453 | goto out; | |
3454 | } | |
343f8568 JS |
3455 | ret = fstat(fd, &st); |
3456 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3457 | goto outc; | |
3458 | } | |
3baf720e CR |
3459 | |
3460 | /* Attempt to detect via a CDROM specific ioctl */ | |
3461 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3462 | if (ret >= 0) | |
3463 | prio = 100; | |
3464 | ||
343f8568 | 3465 | outc: |
2e1e79da | 3466 | qemu_close(fd); |
3baf720e CR |
3467 | out: |
3468 | return prio; | |
508c7cb3 CH |
3469 | } |
3470 | ||
e031f750 | 3471 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3472 | { |
3473 | BDRVRawState *s = bs->opaque; | |
3474 | int ret; | |
3475 | ||
3476 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3477 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3478 | } |
3479 | ||
f36f3949 | 3480 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3481 | { |
3482 | BDRVRawState *s = bs->opaque; | |
3483 | ||
3484 | if (eject_flag) { | |
3485 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3486 | perror("CDROMEJECT"); | |
3487 | } else { | |
3488 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3489 | perror("CDROMEJECT"); | |
3490 | } | |
f3a5d3f8 CH |
3491 | } |
3492 | ||
025e849a | 3493 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3494 | { |
3495 | BDRVRawState *s = bs->opaque; | |
3496 | ||
3497 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3498 | /* | |
3499 | * Note: an error can happen if the distribution automatically | |
3500 | * mounts the CD-ROM | |
3501 | */ | |
3502 | /* perror("CDROM_LOCKDOOR"); */ | |
3503 | } | |
f3a5d3f8 CH |
3504 | } |
3505 | ||
3506 | static BlockDriver bdrv_host_cdrom = { | |
3507 | .format_name = "host_cdrom", | |
84a12e66 | 3508 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3509 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3510 | .bdrv_needs_filename = true, |
508c7cb3 | 3511 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3512 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3513 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3514 | .bdrv_close = raw_close, |
1bc6b705 JC |
3515 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3516 | .bdrv_reopen_commit = raw_reopen_commit, | |
3517 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3518 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3519 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3520 | .mutable_opts = mutable_opts, |
dd577a26 | 3521 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3522 | |
9d52aa3c KW |
3523 | |
3524 | .bdrv_co_preadv = raw_co_preadv, | |
3525 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3526 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3527 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3528 | .bdrv_io_plug = raw_aio_plug, |
3529 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3530 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3531 | |
061ca8a3 | 3532 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3533 | .bdrv_getlength = raw_getlength, |
3534 | .has_variable_length = true, | |
4a1d5e1f FZ |
3535 | .bdrv_get_allocated_file_size |
3536 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3537 | |
3538 | /* removable device support */ | |
3539 | .bdrv_is_inserted = cdrom_is_inserted, | |
3540 | .bdrv_eject = cdrom_eject, | |
025e849a | 3541 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3542 | |
3543 | /* generic scsi device */ | |
2f3a7ab3 | 3544 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3545 | }; |
3546 | #endif /* __linux__ */ | |
3547 | ||
a167ba50 | 3548 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3549 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3550 | Error **errp) | |
f3a5d3f8 CH |
3551 | { |
3552 | BDRVRawState *s = bs->opaque; | |
e428e439 | 3553 | Error *local_err = NULL; |
f3a5d3f8 CH |
3554 | int ret; |
3555 | ||
3556 | s->type = FTYPE_CD; | |
3557 | ||
230ff739 | 3558 | ret = raw_open_common(bs, options, flags, 0, true, &local_err); |
e428e439 | 3559 | if (ret) { |
621ff94d | 3560 | error_propagate(errp, local_err); |
f3a5d3f8 | 3561 | return ret; |
e428e439 | 3562 | } |
f3a5d3f8 | 3563 | |
9b2260cb | 3564 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3565 | ioctl(s->fd, CDIOCALLOW); |
3566 | return 0; | |
3567 | } | |
3568 | ||
508c7cb3 CH |
3569 | static int cdrom_probe_device(const char *filename) |
3570 | { | |
3571 | if (strstart(filename, "/dev/cd", NULL) || | |
3572 | strstart(filename, "/dev/acd", NULL)) | |
3573 | return 100; | |
3574 | return 0; | |
3575 | } | |
3576 | ||
f3a5d3f8 CH |
3577 | static int cdrom_reopen(BlockDriverState *bs) |
3578 | { | |
3579 | BDRVRawState *s = bs->opaque; | |
3580 | int fd; | |
3581 | ||
3582 | /* | |
3583 | * Force reread of possibly changed/newly loaded disc, | |
3584 | * FreeBSD seems to not notice sometimes... | |
3585 | */ | |
3586 | if (s->fd >= 0) | |
2e1e79da | 3587 | qemu_close(s->fd); |
6165f4d8 | 3588 | fd = qemu_open(bs->filename, s->open_flags, 0644); |
f3a5d3f8 CH |
3589 | if (fd < 0) { |
3590 | s->fd = -1; | |
3591 | return -EIO; | |
3592 | } | |
3593 | s->fd = fd; | |
3594 | ||
9b2260cb | 3595 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3596 | ioctl(s->fd, CDIOCALLOW); |
3597 | return 0; | |
3598 | } | |
3599 | ||
e031f750 | 3600 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3601 | { |
3602 | return raw_getlength(bs) > 0; | |
3603 | } | |
3604 | ||
f36f3949 | 3605 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3606 | { |
3607 | BDRVRawState *s = bs->opaque; | |
3608 | ||
3609 | if (s->fd < 0) | |
822e1cd1 | 3610 | return; |
f3a5d3f8 CH |
3611 | |
3612 | (void) ioctl(s->fd, CDIOCALLOW); | |
3613 | ||
3614 | if (eject_flag) { | |
3615 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3616 | perror("CDIOCEJECT"); | |
3617 | } else { | |
3618 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3619 | perror("CDIOCCLOSE"); | |
3620 | } | |
3621 | ||
822e1cd1 | 3622 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3623 | } |
3624 | ||
025e849a | 3625 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3626 | { |
3627 | BDRVRawState *s = bs->opaque; | |
3628 | ||
3629 | if (s->fd < 0) | |
7bf37fed | 3630 | return; |
f3a5d3f8 CH |
3631 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3632 | /* | |
3633 | * Note: an error can happen if the distribution automatically | |
3634 | * mounts the CD-ROM | |
3635 | */ | |
3636 | /* perror("CDROM_LOCKDOOR"); */ | |
3637 | } | |
f3a5d3f8 CH |
3638 | } |
3639 | ||
3640 | static BlockDriver bdrv_host_cdrom = { | |
3641 | .format_name = "host_cdrom", | |
84a12e66 | 3642 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3643 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3644 | .bdrv_needs_filename = true, |
508c7cb3 | 3645 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3646 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3647 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3648 | .bdrv_close = raw_close, |
1bc6b705 JC |
3649 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3650 | .bdrv_reopen_commit = raw_reopen_commit, | |
3651 | .bdrv_reopen_abort = raw_reopen_abort, | |
efc75e2a | 3652 | .bdrv_co_create_opts = hdev_co_create_opts, |
6f482f74 | 3653 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3654 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3655 | |
9d52aa3c KW |
3656 | .bdrv_co_preadv = raw_co_preadv, |
3657 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3658 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3659 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3660 | .bdrv_io_plug = raw_aio_plug, |
3661 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3662 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3663 | |
061ca8a3 | 3664 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3665 | .bdrv_getlength = raw_getlength, |
3666 | .has_variable_length = true, | |
4a1d5e1f FZ |
3667 | .bdrv_get_allocated_file_size |
3668 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3669 | |
19cb3738 | 3670 | /* removable device support */ |
f3a5d3f8 CH |
3671 | .bdrv_is_inserted = cdrom_is_inserted, |
3672 | .bdrv_eject = cdrom_eject, | |
025e849a | 3673 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3674 | }; |
f3a5d3f8 | 3675 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3676 | |
84a12e66 | 3677 | static void bdrv_file_init(void) |
5efa9d5a | 3678 | { |
508c7cb3 CH |
3679 | /* |
3680 | * Register all the drivers. Note that order is important, the driver | |
3681 | * registered last will get probed first. | |
3682 | */ | |
84a12e66 | 3683 | bdrv_register(&bdrv_file); |
5efa9d5a | 3684 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3685 | #ifdef __linux__ |
f3a5d3f8 CH |
3686 | bdrv_register(&bdrv_host_cdrom); |
3687 | #endif | |
a167ba50 | 3688 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3689 | bdrv_register(&bdrv_host_cdrom); |
3690 | #endif | |
5efa9d5a AL |
3691 | } |
3692 | ||
84a12e66 | 3693 | block_init(bdrv_file_init); |