]>
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" |
ffa244c8 | 33 | #include "qemu/units.h" |
de81a169 | 34 | #include "trace.h" |
737e150e | 35 | #include "block/thread-pool.h" |
1de7afc9 | 36 | #include "qemu/iov.h" |
0187f5c9 | 37 | #include "block/raw-aio.h" |
452fcdbc | 38 | #include "qapi/qmp/qdict.h" |
d49b6836 | 39 | #include "qapi/qmp/qstring.h" |
83f64091 | 40 | |
7c9e5276 PB |
41 | #include "scsi/pr-manager.h" |
42 | #include "scsi/constants.h" | |
43 | ||
83affaa6 | 44 | #if defined(__APPLE__) && (__MACH__) |
14176c8d JD |
45 | #include <sys/ioctl.h> |
46 | #if defined(HAVE_HOST_BLOCK_DEVICE) | |
83f64091 FB |
47 | #include <paths.h> |
48 | #include <sys/param.h> | |
0dfc7af2 | 49 | #include <sys/mount.h> |
83f64091 FB |
50 | #include <IOKit/IOKitLib.h> |
51 | #include <IOKit/IOBSD.h> | |
52 | #include <IOKit/storage/IOMediaBSDClient.h> | |
53 | #include <IOKit/storage/IOMedia.h> | |
54 | #include <IOKit/storage/IOCDMedia.h> | |
55 | //#include <IOKit/storage/IOCDTypes.h> | |
d0855f12 | 56 | #include <IOKit/storage/IODVDMedia.h> |
83f64091 | 57 | #include <CoreFoundation/CoreFoundation.h> |
14176c8d | 58 | #endif /* defined(HAVE_HOST_BLOCK_DEVICE) */ |
83f64091 FB |
59 | #endif |
60 | ||
61 | #ifdef __sun__ | |
2e9671da | 62 | #define _POSIX_PTHREAD_SEMANTICS 1 |
83f64091 FB |
63 | #include <sys/dkio.h> |
64 | #endif | |
19cb3738 FB |
65 | #ifdef __linux__ |
66 | #include <sys/ioctl.h> | |
05acda4d | 67 | #include <sys/param.h> |
1efad060 | 68 | #include <sys/syscall.h> |
5edc8557 | 69 | #include <sys/vfs.h> |
19cb3738 FB |
70 | #include <linux/cdrom.h> |
71 | #include <linux/fd.h> | |
5500316d | 72 | #include <linux/fs.h> |
1a9335e4 | 73 | #include <linux/hdreg.h> |
5edc8557 | 74 | #include <linux/magic.h> |
3307ed7b | 75 | #include <scsi/sg.h> |
1a9335e4 ET |
76 | #ifdef __s390__ |
77 | #include <asm/dasd.h> | |
78 | #endif | |
4ab15590 CL |
79 | #ifndef FS_NOCOW_FL |
80 | #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ | |
81 | #endif | |
5500316d | 82 | #endif |
b953f075 | 83 | #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) |
3d4fa43e KK |
84 | #include <linux/falloc.h> |
85 | #endif | |
a167ba50 | 86 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
1cb6c3fd | 87 | #include <sys/disk.h> |
9f23011a | 88 | #include <sys/cdio.h> |
1cb6c3fd | 89 | #endif |
83f64091 | 90 | |
128ab2ff BS |
91 | #ifdef __OpenBSD__ |
92 | #include <sys/ioctl.h> | |
93 | #include <sys/disklabel.h> | |
94 | #include <sys/dkio.h> | |
95 | #endif | |
96 | ||
d1f6fd8d CE |
97 | #ifdef __NetBSD__ |
98 | #include <sys/ioctl.h> | |
99 | #include <sys/disklabel.h> | |
100 | #include <sys/dkio.h> | |
101 | #include <sys/disk.h> | |
102 | #endif | |
103 | ||
c5e97233 BS |
104 | #ifdef __DragonFly__ |
105 | #include <sys/ioctl.h> | |
106 | #include <sys/diskslice.h> | |
107 | #endif | |
108 | ||
dce512de CH |
109 | #ifdef CONFIG_XFS |
110 | #include <xfs/xfs.h> | |
111 | #endif | |
112 | ||
f6465578 AL |
113 | /* OS X does not have O_DSYNC */ |
114 | #ifndef O_DSYNC | |
1c27a8b3 | 115 | #ifdef O_SYNC |
7ab064d2 | 116 | #define O_DSYNC O_SYNC |
1c27a8b3 JA |
117 | #elif defined(O_FSYNC) |
118 | #define O_DSYNC O_FSYNC | |
119 | #endif | |
f6465578 AL |
120 | #endif |
121 | ||
9f7965c7 AL |
122 | /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ |
123 | #ifndef O_DIRECT | |
124 | #define O_DIRECT O_DSYNC | |
125 | #endif | |
126 | ||
19cb3738 FB |
127 | #define FTYPE_FILE 0 |
128 | #define FTYPE_CD 1 | |
83f64091 | 129 | |
581b9e29 CH |
130 | #define MAX_BLOCKSIZE 4096 |
131 | ||
244a5668 FZ |
132 | /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, |
133 | * leaving a few more bytes for its future use. */ | |
134 | #define RAW_LOCK_PERM_BASE 100 | |
135 | #define RAW_LOCK_SHARED_BASE 200 | |
136 | ||
19cb3738 FB |
137 | typedef struct BDRVRawState { |
138 | int fd; | |
244a5668 | 139 | bool use_lock; |
19cb3738 | 140 | int type; |
0e1d8f4c | 141 | int open_flags; |
c25f53b0 PB |
142 | size_t buf_align; |
143 | ||
244a5668 FZ |
144 | /* The current permissions. */ |
145 | uint64_t perm; | |
146 | uint64_t shared_perm; | |
147 | ||
2996ffad | 148 | /* The perms bits whose corresponding bytes are already locked in |
f2e3af29 | 149 | * s->fd. */ |
2996ffad FZ |
150 | uint64_t locked_perm; |
151 | uint64_t locked_shared_perm; | |
152 | ||
684960d4 SG |
153 | uint64_t aio_max_batch; |
154 | ||
6ceabe6f | 155 | int perm_change_fd; |
094e3639 | 156 | int perm_change_flags; |
e0c9cf3a KW |
157 | BDRVReopenState *reopen_state; |
158 | ||
dce512de | 159 | #ifdef CONFIG_XFS |
260a82e5 | 160 | bool is_xfs:1; |
dce512de | 161 | #endif |
260a82e5 | 162 | bool has_discard:1; |
97a2ae34 | 163 | bool has_write_zeroes:1; |
260a82e5 | 164 | bool discard_zeroes:1; |
0a4279d9 | 165 | bool use_linux_aio:1; |
c6447510 | 166 | bool use_linux_io_uring:1; |
c7ddc882 | 167 | int page_cache_inconsistent; /* errno from fdatasync failure */ |
d50d8222 | 168 | bool has_fallocate; |
3cad8307 | 169 | bool needs_alignment; |
5dbd0ce1 | 170 | bool force_alignment; |
f357fcd8 | 171 | bool drop_cache; |
31be8a2a | 172 | bool check_cache_dropped; |
1c450366 AN |
173 | struct { |
174 | uint64_t discard_nb_ok; | |
175 | uint64_t discard_nb_failed; | |
176 | uint64_t discard_bytes_ok; | |
177 | } stats; | |
7c9e5276 PB |
178 | |
179 | PRManager *pr_mgr; | |
19cb3738 FB |
180 | } BDRVRawState; |
181 | ||
eeb6b45d | 182 | typedef struct BDRVRawReopenState { |
eeb6b45d | 183 | int open_flags; |
f357fcd8 | 184 | bool drop_cache; |
31be8a2a | 185 | bool check_cache_dropped; |
eeb6b45d JC |
186 | } BDRVRawReopenState; |
187 | ||
14176c8d JD |
188 | static int fd_open(BlockDriverState *bs) |
189 | { | |
190 | BDRVRawState *s = bs->opaque; | |
191 | ||
192 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
193 | if (s->fd >= 0) { | |
194 | return 0; | |
195 | } | |
196 | return -EIO; | |
197 | } | |
198 | ||
22afa7b5 | 199 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 200 | |
de81a169 PB |
201 | typedef struct RawPosixAIOData { |
202 | BlockDriverState *bs; | |
d57c44d0 | 203 | int aio_type; |
de81a169 | 204 | int aio_fildes; |
d57c44d0 | 205 | |
de81a169 | 206 | off_t aio_offset; |
d57c44d0 KW |
207 | uint64_t aio_nbytes; |
208 | ||
93f4e2ff | 209 | union { |
d57c44d0 KW |
210 | struct { |
211 | struct iovec *iov; | |
212 | int niov; | |
213 | } io; | |
214 | struct { | |
215 | uint64_t cmd; | |
216 | void *buf; | |
217 | } ioctl; | |
93f4e2ff KW |
218 | struct { |
219 | int aio_fd2; | |
220 | off_t aio_offset2; | |
d57c44d0 | 221 | } copy_range; |
93f4e2ff KW |
222 | struct { |
223 | PreallocMode prealloc; | |
224 | Error **errp; | |
d57c44d0 | 225 | } truncate; |
93f4e2ff | 226 | }; |
de81a169 PB |
227 | } RawPosixAIOData; |
228 | ||
a167ba50 | 229 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 230 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
231 | #endif |
232 | ||
797e3e38 DE |
233 | /* |
234 | * Elide EAGAIN and EACCES details when failing to lock, as this | |
235 | * indicates that the specified file region is already locked by | |
236 | * another process, which is considered a common scenario. | |
237 | */ | |
238 | #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ | |
239 | do { \ | |
240 | if ((err) == EAGAIN || (err) == EACCES) { \ | |
241 | error_setg((errp), (fmt), ## __VA_ARGS__); \ | |
242 | } else { \ | |
243 | error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ | |
244 | } \ | |
245 | } while (0) | |
246 | ||
1de1ae0a | 247 | #if defined(__NetBSD__) |
db0754df | 248 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
249 | { |
250 | static char namebuf[PATH_MAX]; | |
251 | const char *dp, *fname; | |
252 | struct stat sb; | |
253 | ||
254 | fname = *filename; | |
255 | dp = strrchr(fname, '/'); | |
256 | if (lstat(fname, &sb) < 0) { | |
f6fc1e30 | 257 | error_setg_file_open(errp, errno, fname); |
1de1ae0a CE |
258 | return -errno; |
259 | } | |
260 | ||
261 | if (!S_ISBLK(sb.st_mode)) { | |
262 | return 0; | |
263 | } | |
264 | ||
265 | if (dp == NULL) { | |
266 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
267 | } else { | |
268 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
269 | (int)(dp - fname), fname, dp + 1); | |
270 | } | |
1de1ae0a | 271 | *filename = namebuf; |
db0754df | 272 | warn_report("%s is a block device, using %s", fname, *filename); |
1de1ae0a CE |
273 | |
274 | return 0; | |
275 | } | |
276 | #else | |
db0754df | 277 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
278 | { |
279 | return 0; | |
280 | } | |
281 | #endif | |
282 | ||
8a4ed0d1 ET |
283 | /* |
284 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
285 | */ | |
286 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 287 | { |
c25f53b0 | 288 | unsigned int sector_size; |
8a4ed0d1 | 289 | bool success = false; |
700f9ce0 | 290 | int i; |
c25f53b0 | 291 | |
8a4ed0d1 | 292 | errno = ENOTSUP; |
700f9ce0 | 293 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 294 | #ifdef BLKSSZGET |
700f9ce0 | 295 | BLKSSZGET, |
c25f53b0 PB |
296 | #endif |
297 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 298 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
299 | #endif |
300 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 301 | DIOCGSECTORSIZE, |
c25f53b0 | 302 | #endif |
700f9ce0 PM |
303 | }; |
304 | ||
305 | /* Try a few ioctls to get the right size */ | |
306 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
307 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
308 | *sector_size_p = sector_size; | |
309 | success = true; | |
310 | } | |
311 | } | |
8a4ed0d1 ET |
312 | |
313 | return success ? 0 : -errno; | |
314 | } | |
315 | ||
1a9335e4 ET |
316 | /** |
317 | * Get physical block size of @fd. | |
318 | * On success, store it in @blk_size and return 0. | |
319 | * On failure, return -errno. | |
320 | */ | |
321 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
322 | { | |
323 | #ifdef BLKPBSZGET | |
324 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
325 | return -errno; | |
326 | } | |
327 | return 0; | |
328 | #else | |
329 | return -ENOTSUP; | |
330 | #endif | |
331 | } | |
332 | ||
5edc8557 KW |
333 | /* |
334 | * Returns true if no alignment restrictions are necessary even for files | |
335 | * opened with O_DIRECT. | |
336 | * | |
337 | * raw_probe_alignment() probes the required alignment and assume that 1 means | |
338 | * the probing failed, so it falls back to a safe default of 4k. This can be | |
339 | * avoided if we know that byte alignment is okay for the file. | |
340 | */ | |
341 | static bool dio_byte_aligned(int fd) | |
342 | { | |
343 | #ifdef __linux__ | |
344 | struct statfs buf; | |
345 | int ret; | |
346 | ||
347 | ret = fstatfs(fd, &buf); | |
348 | if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { | |
349 | return true; | |
350 | } | |
351 | #endif | |
352 | return false; | |
353 | } | |
354 | ||
5dbd0ce1 KW |
355 | static bool raw_needs_alignment(BlockDriverState *bs) |
356 | { | |
357 | BDRVRawState *s = bs->opaque; | |
358 | ||
359 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { | |
360 | return true; | |
361 | } | |
362 | ||
363 | return s->force_alignment; | |
364 | } | |
365 | ||
22d182e8 SH |
366 | /* Check if read is allowed with given memory buffer and length. |
367 | * | |
368 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
369 | */ | |
370 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
371 | { | |
372 | ssize_t ret = pread(fd, buf, len, 0); | |
373 | ||
374 | if (ret >= 0) { | |
375 | return true; | |
376 | } | |
377 | ||
378 | #ifdef __linux__ | |
379 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
380 | * other errors (e.g. real I/O error), which could happen on a failed | |
381 | * drive, since we only care about probing alignment. | |
382 | */ | |
383 | if (errno != EINVAL) { | |
384 | return true; | |
385 | } | |
386 | #endif | |
387 | ||
388 | return false; | |
389 | } | |
390 | ||
8a4ed0d1 ET |
391 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
392 | { | |
393 | BDRVRawState *s = bs->opaque; | |
394 | char *buf; | |
038adc2f | 395 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); |
a6b257a0 | 396 | size_t alignments[] = {1, 512, 1024, 2048, 4096}; |
8a4ed0d1 | 397 | |
b192af8a | 398 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 399 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 400 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 401 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
402 | s->buf_align = 1; |
403 | return; | |
404 | } | |
405 | ||
a5b8dd2c | 406 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
407 | s->buf_align = 0; |
408 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
409 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
410 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 411 | } |
c25f53b0 PB |
412 | #ifdef CONFIG_XFS |
413 | if (s->is_xfs) { | |
414 | struct dioattr da; | |
df26a350 | 415 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
a5b8dd2c | 416 | bs->bl.request_alignment = da.d_miniosz; |
c25f53b0 PB |
417 | /* The kernel returns wrong information for d_mem */ |
418 | /* s->buf_align = da.d_mem; */ | |
419 | } | |
420 | } | |
421 | #endif | |
422 | ||
a6b257a0 NS |
423 | /* |
424 | * If we could not get the sizes so far, we can only guess them. First try | |
425 | * to detect request alignment, since it is more likely to succeed. Then | |
426 | * try to detect buf_align, which cannot be detected in some cases (e.g. | |
427 | * Gluster). If buf_align cannot be detected, we fallback to the value of | |
428 | * request_alignment. | |
429 | */ | |
430 | ||
431 | if (!bs->bl.request_alignment) { | |
432 | int i; | |
c25f53b0 | 433 | size_t align; |
a6b257a0 NS |
434 | buf = qemu_memalign(max_align, max_align); |
435 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
436 | align = alignments[i]; | |
437 | if (raw_is_io_aligned(fd, buf, align)) { | |
438 | /* Fallback to safe value. */ | |
439 | bs->bl.request_alignment = (align != 1) ? align : max_align; | |
c25f53b0 PB |
440 | break; |
441 | } | |
442 | } | |
443 | qemu_vfree(buf); | |
444 | } | |
445 | ||
a6b257a0 NS |
446 | if (!s->buf_align) { |
447 | int i; | |
c25f53b0 | 448 | size_t align; |
a6b257a0 NS |
449 | buf = qemu_memalign(max_align, 2 * max_align); |
450 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
451 | align = alignments[i]; | |
452 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
236094c7 | 453 | /* Fallback to request_alignment. */ |
a6b257a0 | 454 | s->buf_align = (align != 1) ? align : bs->bl.request_alignment; |
c25f53b0 PB |
455 | break; |
456 | } | |
457 | } | |
458 | qemu_vfree(buf); | |
459 | } | |
df26a350 | 460 | |
a5b8dd2c | 461 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
462 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
463 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 464 | } |
c25f53b0 PB |
465 | } |
466 | ||
bca5283b | 467 | static int check_hdev_writable(int fd) |
20eaf1bf KW |
468 | { |
469 | #if defined(BLKROGET) | |
470 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
471 | * This is independent of device node permissions and therefore open(2) | |
472 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
473 | * | |
474 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
475 | * check for read-only block devices so that Linux block devices behave | |
476 | * properly. | |
477 | */ | |
478 | struct stat st; | |
479 | int readonly = 0; | |
480 | ||
bca5283b | 481 | if (fstat(fd, &st)) { |
20eaf1bf KW |
482 | return -errno; |
483 | } | |
484 | ||
485 | if (!S_ISBLK(st.st_mode)) { | |
486 | return 0; | |
487 | } | |
488 | ||
bca5283b | 489 | if (ioctl(fd, BLKROGET, &readonly) < 0) { |
20eaf1bf KW |
490 | return -errno; |
491 | } | |
492 | ||
493 | if (readonly) { | |
494 | return -EACCES; | |
495 | } | |
496 | #endif /* defined(BLKROGET) */ | |
497 | return 0; | |
498 | } | |
499 | ||
23dece19 | 500 | static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) |
6a8dc042 | 501 | { |
23dece19 | 502 | bool read_write = false; |
6a8dc042 JC |
503 | assert(open_flags != NULL); |
504 | ||
505 | *open_flags |= O_BINARY; | |
506 | *open_flags &= ~O_ACCMODE; | |
23dece19 KW |
507 | |
508 | if (bdrv_flags & BDRV_O_AUTO_RDONLY) { | |
509 | read_write = has_writers; | |
510 | } else if (bdrv_flags & BDRV_O_RDWR) { | |
511 | read_write = true; | |
512 | } | |
513 | ||
514 | if (read_write) { | |
6a8dc042 JC |
515 | *open_flags |= O_RDWR; |
516 | } else { | |
517 | *open_flags |= O_RDONLY; | |
518 | } | |
519 | ||
520 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
521 | * and O_DIRECT for no caching. */ | |
522 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
523 | *open_flags |= O_DIRECT; | |
524 | } | |
6a8dc042 JC |
525 | } |
526 | ||
078896a9 HR |
527 | static void raw_parse_filename(const char *filename, QDict *options, |
528 | Error **errp) | |
529 | { | |
03c320d8 | 530 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
531 | } |
532 | ||
c66a6157 KW |
533 | static QemuOptsList raw_runtime_opts = { |
534 | .name = "raw", | |
535 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
536 | .desc = { | |
537 | { | |
538 | .name = "filename", | |
539 | .type = QEMU_OPT_STRING, | |
540 | .help = "File name of the image", | |
541 | }, | |
0a4279d9 KW |
542 | { |
543 | .name = "aio", | |
544 | .type = QEMU_OPT_STRING, | |
c6447510 | 545 | .help = "host AIO implementation (threads, native, io_uring)", |
0a4279d9 | 546 | }, |
684960d4 SG |
547 | { |
548 | .name = "aio-max-batch", | |
549 | .type = QEMU_OPT_NUMBER, | |
550 | .help = "AIO max batch size (0 = auto handled by AIO backend, default: 0)", | |
551 | }, | |
16b48d5d FZ |
552 | { |
553 | .name = "locking", | |
554 | .type = QEMU_OPT_STRING, | |
555 | .help = "file locking mode (on/off/auto, default: auto)", | |
556 | }, | |
7c9e5276 PB |
557 | { |
558 | .name = "pr-manager", | |
559 | .type = QEMU_OPT_STRING, | |
560 | .help = "id of persistent reservation manager object (default: none)", | |
561 | }, | |
f357fcd8 SH |
562 | #if defined(__linux__) |
563 | { | |
564 | .name = "drop-cache", | |
565 | .type = QEMU_OPT_BOOL, | |
566 | .help = "invalidate page cache during live migration (default: on)", | |
567 | }, | |
568 | #endif | |
31be8a2a SH |
569 | { |
570 | .name = "x-check-cache-dropped", | |
571 | .type = QEMU_OPT_BOOL, | |
572 | .help = "check that page cache was dropped on live migration (default: off)" | |
573 | }, | |
c66a6157 KW |
574 | { /* end of list */ } |
575 | }, | |
576 | }; | |
577 | ||
8a2ce0bc AG |
578 | static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; |
579 | ||
c66a6157 | 580 | static int raw_open_common(BlockDriverState *bs, QDict *options, |
230ff739 JS |
581 | int bdrv_flags, int open_flags, |
582 | bool device, Error **errp) | |
83f64091 FB |
583 | { |
584 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
585 | QemuOpts *opts; |
586 | Error *local_err = NULL; | |
8bfea15d | 587 | const char *filename = NULL; |
7c9e5276 | 588 | const char *str; |
0a4279d9 | 589 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 590 | int fd, ret; |
260a82e5 | 591 | struct stat st; |
244a5668 | 592 | OnOffAuto locking; |
83f64091 | 593 | |
87ea75d5 | 594 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 595 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
c66a6157 KW |
596 | ret = -EINVAL; |
597 | goto fail; | |
598 | } | |
599 | ||
600 | filename = qemu_opt_get(opts, "filename"); | |
601 | ||
db0754df | 602 | ret = raw_normalize_devicepath(&filename, errp); |
1de1ae0a | 603 | if (ret != 0) { |
c66a6157 | 604 | goto fail; |
1de1ae0a CE |
605 | } |
606 | ||
c6447510 AM |
607 | if (bdrv_flags & BDRV_O_NATIVE_AIO) { |
608 | aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; | |
609 | #ifdef CONFIG_LINUX_IO_URING | |
610 | } else if (bdrv_flags & BDRV_O_IO_URING) { | |
611 | aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; | |
612 | #endif | |
613 | } else { | |
614 | aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; | |
615 | } | |
616 | ||
f7abe0ec MAL |
617 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
618 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 619 | aio_default, &local_err); |
0a4279d9 KW |
620 | if (local_err) { |
621 | error_propagate(errp, local_err); | |
622 | ret = -EINVAL; | |
623 | goto fail; | |
624 | } | |
c6447510 | 625 | |
0a4279d9 | 626 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); |
c6447510 AM |
627 | #ifdef CONFIG_LINUX_IO_URING |
628 | s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); | |
629 | #endif | |
0a4279d9 | 630 | |
684960d4 SG |
631 | s->aio_max_batch = qemu_opt_get_number(opts, "aio-max-batch", 0); |
632 | ||
f7abe0ec MAL |
633 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
634 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 635 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
636 | if (local_err) { |
637 | error_propagate(errp, local_err); | |
638 | ret = -EINVAL; | |
639 | goto fail; | |
640 | } | |
641 | switch (locking) { | |
642 | case ON_OFF_AUTO_ON: | |
643 | s->use_lock = true; | |
2b218f5d | 644 | if (!qemu_has_ofd_lock()) { |
db0754df FZ |
645 | warn_report("File lock requested but OFD locking syscall is " |
646 | "unavailable, falling back to POSIX file locks"); | |
647 | error_printf("Due to the implementation, locks can be lost " | |
648 | "unexpectedly.\n"); | |
2b218f5d | 649 | } |
244a5668 FZ |
650 | break; |
651 | case ON_OFF_AUTO_OFF: | |
652 | s->use_lock = false; | |
653 | break; | |
654 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 655 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
656 | break; |
657 | default: | |
658 | abort(); | |
659 | } | |
660 | ||
7c9e5276 PB |
661 | str = qemu_opt_get(opts, "pr-manager"); |
662 | if (str) { | |
663 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
664 | if (local_err) { | |
665 | error_propagate(errp, local_err); | |
666 | ret = -EINVAL; | |
667 | goto fail; | |
668 | } | |
669 | } | |
670 | ||
f357fcd8 | 671 | s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); |
31be8a2a SH |
672 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
673 | false); | |
674 | ||
6a8dc042 | 675 | s->open_flags = open_flags; |
23dece19 | 676 | raw_parse_flags(bdrv_flags, &s->open_flags, false); |
83f64091 | 677 | |
90babde0 | 678 | s->fd = -1; |
b18a24a9 | 679 | fd = qemu_open(filename, s->open_flags, errp); |
64107dc0 KW |
680 | ret = fd < 0 ? -errno : 0; |
681 | ||
64107dc0 | 682 | if (ret < 0) { |
c66a6157 | 683 | if (ret == -EROFS) { |
19cb3738 | 684 | ret = -EACCES; |
c66a6157 KW |
685 | } |
686 | goto fail; | |
19cb3738 | 687 | } |
83f64091 | 688 | s->fd = fd; |
9ef91a67 | 689 | |
bca5283b KW |
690 | /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ |
691 | if (s->open_flags & O_RDWR) { | |
692 | ret = check_hdev_writable(s->fd); | |
693 | if (ret < 0) { | |
694 | error_setg_errno(errp, -ret, "The device is not writable"); | |
695 | goto fail; | |
696 | } | |
697 | } | |
698 | ||
244a5668 FZ |
699 | s->perm = 0; |
700 | s->shared_perm = BLK_PERM_ALL; | |
701 | ||
5c6c3a6c | 702 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 703 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
704 | if (s->use_linux_aio) { |
705 | if (!(s->open_flags & O_DIRECT)) { | |
706 | error_setg(errp, "aio=native was specified, but it requires " | |
707 | "cache.direct=on, which was not specified."); | |
708 | ret = -EINVAL; | |
709 | goto fail; | |
710 | } | |
711 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
712 | error_prepend(errp, "Unable to use native AIO: "); | |
713 | goto fail; | |
714 | } | |
96518254 | 715 | } |
1501ecc1 | 716 | #else |
0a4279d9 | 717 | if (s->use_linux_aio) { |
d657c0c2 KW |
718 | error_setg(errp, "aio=native was specified, but is not supported " |
719 | "in this build."); | |
720 | ret = -EINVAL; | |
721 | goto fail; | |
1501ecc1 SH |
722 | } |
723 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 724 | |
c6447510 AM |
725 | #ifdef CONFIG_LINUX_IO_URING |
726 | if (s->use_linux_io_uring) { | |
727 | if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { | |
728 | error_prepend(errp, "Unable to use io_uring: "); | |
729 | goto fail; | |
730 | } | |
731 | } | |
732 | #else | |
733 | if (s->use_linux_io_uring) { | |
734 | error_setg(errp, "aio=io_uring was specified, but is not supported " | |
735 | "in this build."); | |
736 | ret = -EINVAL; | |
737 | goto fail; | |
738 | } | |
739 | #endif /* !defined(CONFIG_LINUX_IO_URING) */ | |
740 | ||
7ce21016 | 741 | s->has_discard = true; |
97a2ae34 | 742 | s->has_write_zeroes = true; |
260a82e5 PB |
743 | |
744 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 745 | ret = -errno; |
260a82e5 PB |
746 | error_setg_errno(errp, errno, "Could not stat file"); |
747 | goto fail; | |
748 | } | |
230ff739 JS |
749 | |
750 | if (!device) { | |
8d17adf3 DB |
751 | if (!S_ISREG(st.st_mode)) { |
752 | error_setg(errp, "'%s' driver requires '%s' to be a regular file", | |
753 | bs->drv->format_name, bs->filename); | |
230ff739 JS |
754 | ret = -EINVAL; |
755 | goto fail; | |
756 | } else { | |
757 | s->discard_zeroes = true; | |
758 | s->has_fallocate = true; | |
759 | } | |
760 | } else { | |
761 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
8d17adf3 DB |
762 | error_setg(errp, "'%s' driver requires '%s' to be either " |
763 | "a character or block device", | |
764 | bs->drv->format_name, bs->filename); | |
230ff739 JS |
765 | ret = -EINVAL; |
766 | goto fail; | |
767 | } | |
260a82e5 | 768 | } |
230ff739 | 769 | |
d0b4503e PB |
770 | if (S_ISBLK(st.st_mode)) { |
771 | #ifdef BLKDISCARDZEROES | |
772 | unsigned int arg; | |
773 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
774 | s->discard_zeroes = true; | |
775 | } | |
776 | #endif | |
777 | #ifdef __linux__ | |
778 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
779 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 780 | * Same for BLKZEROOUT. |
d0b4503e PB |
781 | */ |
782 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
783 | s->discard_zeroes = false; | |
97a2ae34 | 784 | s->has_write_zeroes = false; |
d0b4503e PB |
785 | } |
786 | #endif | |
787 | } | |
3cad8307 RPM |
788 | #ifdef __FreeBSD__ |
789 | if (S_ISCHR(st.st_mode)) { | |
790 | /* | |
791 | * The file is a char device (disk), which on FreeBSD isn't behind | |
792 | * a pager, so force all requests to be aligned. This is needed | |
793 | * so QEMU makes sure all IO operations on the device are aligned | |
794 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
795 | */ | |
5dbd0ce1 | 796 | s->force_alignment = true; |
3cad8307 RPM |
797 | } |
798 | #endif | |
5dbd0ce1 | 799 | s->needs_alignment = raw_needs_alignment(bs); |
260a82e5 | 800 | |
dce512de CH |
801 | #ifdef CONFIG_XFS |
802 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 803 | s->is_xfs = true; |
dce512de CH |
804 | } |
805 | #endif | |
806 | ||
738301e1 | 807 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
2f0c6e7a KW |
808 | if (S_ISREG(st.st_mode)) { |
809 | /* When extending regular files, we get zeros from the OS */ | |
810 | bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; | |
811 | } | |
c66a6157 KW |
812 | ret = 0; |
813 | fail: | |
a8c5cf27 KW |
814 | if (ret < 0 && s->fd != -1) { |
815 | qemu_close(s->fd); | |
816 | } | |
8bfea15d KW |
817 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
818 | unlink(filename); | |
819 | } | |
c66a6157 KW |
820 | qemu_opts_del(opts); |
821 | return ret; | |
83f64091 FB |
822 | } |
823 | ||
015a1036 HR |
824 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
825 | Error **errp) | |
90babde0 CH |
826 | { |
827 | BDRVRawState *s = bs->opaque; | |
828 | ||
829 | s->type = FTYPE_FILE; | |
230ff739 | 830 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
831 | } |
832 | ||
244a5668 FZ |
833 | typedef enum { |
834 | RAW_PL_PREPARE, | |
835 | RAW_PL_COMMIT, | |
836 | RAW_PL_ABORT, | |
837 | } RawPermLockOp; | |
838 | ||
839 | #define PERM_FOREACH(i) \ | |
840 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
841 | ||
842 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
843 | * file; if @unlock == true, also unlock the unneeded bytes. | |
844 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
845 | */ | |
2996ffad | 846 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
847 | uint64_t perm_lock_bits, |
848 | uint64_t shared_perm_lock_bits, | |
849 | bool unlock, Error **errp) | |
850 | { | |
851 | int ret; | |
852 | int i; | |
2996ffad FZ |
853 | uint64_t locked_perm, locked_shared_perm; |
854 | ||
855 | if (s) { | |
856 | locked_perm = s->locked_perm; | |
857 | locked_shared_perm = s->locked_shared_perm; | |
858 | } else { | |
859 | /* | |
860 | * We don't have the previous bits, just lock/unlock for each of the | |
861 | * requested bits. | |
862 | */ | |
863 | if (unlock) { | |
864 | locked_perm = BLK_PERM_ALL; | |
865 | locked_shared_perm = BLK_PERM_ALL; | |
866 | } else { | |
867 | locked_perm = 0; | |
868 | locked_shared_perm = 0; | |
869 | } | |
870 | } | |
244a5668 FZ |
871 | |
872 | PERM_FOREACH(i) { | |
873 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
874 | uint64_t bit = (1ULL << i); |
875 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 876 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 | 877 | if (ret) { |
797e3e38 DE |
878 | raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", |
879 | off); | |
244a5668 | 880 | return ret; |
2996ffad FZ |
881 | } else if (s) { |
882 | s->locked_perm |= bit; | |
244a5668 | 883 | } |
2996ffad | 884 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 885 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 | 886 | if (ret) { |
797e3e38 | 887 | error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); |
244a5668 | 888 | return ret; |
2996ffad FZ |
889 | } else if (s) { |
890 | s->locked_perm &= ~bit; | |
244a5668 FZ |
891 | } |
892 | } | |
893 | } | |
894 | PERM_FOREACH(i) { | |
895 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
896 | uint64_t bit = (1ULL << i); |
897 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 898 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 | 899 | if (ret) { |
797e3e38 DE |
900 | raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", |
901 | off); | |
244a5668 | 902 | return ret; |
2996ffad FZ |
903 | } else if (s) { |
904 | s->locked_shared_perm |= bit; | |
244a5668 | 905 | } |
2996ffad FZ |
906 | } else if (unlock && (locked_shared_perm & bit) && |
907 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 908 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 | 909 | if (ret) { |
797e3e38 | 910 | error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); |
244a5668 | 911 | return ret; |
2996ffad FZ |
912 | } else if (s) { |
913 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
914 | } |
915 | } | |
916 | } | |
917 | return 0; | |
918 | } | |
919 | ||
920 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 921 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
922 | Error **errp) |
923 | { | |
924 | int ret; | |
925 | int i; | |
926 | ||
927 | PERM_FOREACH(i) { | |
928 | int off = RAW_LOCK_SHARED_BASE + i; | |
929 | uint64_t p = 1ULL << i; | |
930 | if (perm & p) { | |
d0a96155 | 931 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
932 | if (ret) { |
933 | char *perm_name = bdrv_perm_names(p); | |
797e3e38 DE |
934 | |
935 | raw_lock_error_setg_errno(errp, -ret, | |
936 | "Failed to get \"%s\" lock", | |
937 | perm_name); | |
244a5668 | 938 | g_free(perm_name); |
244a5668 FZ |
939 | return ret; |
940 | } | |
941 | } | |
942 | } | |
943 | PERM_FOREACH(i) { | |
944 | int off = RAW_LOCK_PERM_BASE + i; | |
945 | uint64_t p = 1ULL << i; | |
946 | if (!(shared_perm & p)) { | |
d0a96155 | 947 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
948 | if (ret) { |
949 | char *perm_name = bdrv_perm_names(p); | |
797e3e38 DE |
950 | |
951 | raw_lock_error_setg_errno(errp, -ret, | |
952 | "Failed to get shared \"%s\" lock", | |
953 | perm_name); | |
244a5668 | 954 | g_free(perm_name); |
244a5668 FZ |
955 | return ret; |
956 | } | |
957 | } | |
958 | } | |
959 | return 0; | |
960 | } | |
961 | ||
962 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
963 | RawPermLockOp op, | |
964 | uint64_t new_perm, uint64_t new_shared, | |
965 | Error **errp) | |
966 | { | |
967 | BDRVRawState *s = bs->opaque; | |
968 | int ret = 0; | |
969 | Error *local_err = NULL; | |
970 | ||
971 | if (!s->use_lock) { | |
972 | return 0; | |
973 | } | |
974 | ||
975 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
976 | return 0; | |
977 | } | |
978 | ||
244a5668 FZ |
979 | switch (op) { |
980 | case RAW_PL_PREPARE: | |
696aaaed VSO |
981 | if ((s->perm | new_perm) == s->perm && |
982 | (s->shared_perm & new_shared) == s->shared_perm) | |
983 | { | |
984 | /* | |
985 | * We are going to unlock bytes, it should not fail. If it fail due | |
986 | * to some fs-dependent permission-unrelated reasons (which occurs | |
987 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
988 | * can't prevent such errors by any check here. And we ignore them | |
989 | * anyway in ABORT and COMMIT. | |
990 | */ | |
991 | return 0; | |
992 | } | |
f2e3af29 | 993 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
994 | ~s->shared_perm | ~new_shared, |
995 | false, errp); | |
996 | if (!ret) { | |
f2e3af29 | 997 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
998 | if (!ret) { |
999 | return 0; | |
1000 | } | |
b857431d FZ |
1001 | error_append_hint(errp, |
1002 | "Is another process using the image [%s]?\n", | |
1003 | bs->filename); | |
244a5668 | 1004 | } |
244a5668 FZ |
1005 | /* fall through to unlock bytes. */ |
1006 | case RAW_PL_ABORT: | |
f2e3af29 | 1007 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 1008 | true, &local_err); |
244a5668 FZ |
1009 | if (local_err) { |
1010 | /* Theoretically the above call only unlocks bytes and it cannot | |
1011 | * fail. Something weird happened, report it. | |
1012 | */ | |
db0754df | 1013 | warn_report_err(local_err); |
244a5668 FZ |
1014 | } |
1015 | break; | |
1016 | case RAW_PL_COMMIT: | |
f2e3af29 | 1017 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 1018 | true, &local_err); |
244a5668 FZ |
1019 | if (local_err) { |
1020 | /* Theoretically the above call only unlocks bytes and it cannot | |
1021 | * fail. Something weird happened, report it. | |
1022 | */ | |
db0754df | 1023 | warn_report_err(local_err); |
244a5668 FZ |
1024 | } |
1025 | break; | |
1026 | } | |
1027 | return ret; | |
1028 | } | |
1029 | ||
5cec2870 | 1030 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 1031 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 1032 | Error **errp) |
5cec2870 KW |
1033 | { |
1034 | BDRVRawState *s = bs->opaque; | |
1035 | int fd = -1; | |
1036 | int ret; | |
23dece19 KW |
1037 | bool has_writers = perm & |
1038 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
1039 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
1040 | #ifdef O_NOATIME | |
1041 | fcntl_flags |= O_NOATIME; | |
1042 | #endif | |
1043 | ||
1044 | *open_flags = 0; | |
1045 | if (s->type == FTYPE_CD) { | |
1046 | *open_flags |= O_NONBLOCK; | |
1047 | } | |
1048 | ||
23dece19 | 1049 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
1050 | |
1051 | #ifdef O_ASYNC | |
1052 | /* Not all operating systems have O_ASYNC, and those that don't | |
1053 | * will not let us track the state into rs->open_flags (typically | |
1054 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
1055 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
1056 | */ | |
1057 | assert((s->open_flags & O_ASYNC) == 0); | |
1058 | #endif | |
1059 | ||
6ceabe6f KW |
1060 | if (!force_dup && *open_flags == s->open_flags) { |
1061 | /* We're lucky, the existing fd is fine */ | |
1062 | return s->fd; | |
1063 | } | |
1064 | ||
5cec2870 KW |
1065 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
1066 | /* dup the original fd */ | |
1067 | fd = qemu_dup(s->fd); | |
1068 | if (fd >= 0) { | |
1069 | ret = fcntl_setfl(fd, *open_flags); | |
1070 | if (ret) { | |
1071 | qemu_close(fd); | |
1072 | fd = -1; | |
1073 | } | |
1074 | } | |
1075 | } | |
1076 | ||
b18a24a9 | 1077 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ |
5cec2870 KW |
1078 | if (fd == -1) { |
1079 | const char *normalized_filename = bs->filename; | |
1080 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
1081 | if (ret >= 0) { | |
b18a24a9 | 1082 | fd = qemu_open(normalized_filename, *open_flags, errp); |
5cec2870 | 1083 | if (fd == -1) { |
5cec2870 KW |
1084 | return -1; |
1085 | } | |
1086 | } | |
1087 | } | |
1088 | ||
bca5283b KW |
1089 | if (fd != -1 && (*open_flags & O_RDWR)) { |
1090 | ret = check_hdev_writable(fd); | |
1091 | if (ret < 0) { | |
1092 | qemu_close(fd); | |
1093 | error_setg_errno(errp, -ret, "The device is not writable"); | |
1094 | return -1; | |
1095 | } | |
1096 | } | |
1097 | ||
5cec2870 KW |
1098 | return fd; |
1099 | } | |
1100 | ||
eeb6b45d JC |
1101 | static int raw_reopen_prepare(BDRVReopenState *state, |
1102 | BlockReopenQueue *queue, Error **errp) | |
1103 | { | |
1104 | BDRVRawState *s; | |
4e6d13c9 | 1105 | BDRVRawReopenState *rs; |
31be8a2a | 1106 | QemuOpts *opts; |
a6aeca0c | 1107 | int ret; |
eeb6b45d JC |
1108 | |
1109 | assert(state != NULL); | |
1110 | assert(state->bs != NULL); | |
1111 | ||
1112 | s = state->bs->opaque; | |
1113 | ||
5839e53b | 1114 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 1115 | rs = state->opaque; |
31be8a2a SH |
1116 | |
1117 | /* Handle options changes */ | |
1118 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 1119 | if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { |
31be8a2a SH |
1120 | ret = -EINVAL; |
1121 | goto out; | |
1122 | } | |
1123 | ||
f357fcd8 | 1124 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
1125 | rs->check_cache_dropped = |
1126 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
1127 | ||
1128 | /* This driver's reopen function doesn't currently allow changing | |
1129 | * other options, so let's put them back in the original QDict and | |
1130 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
1131 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 1132 | |
72373e40 VSO |
1133 | /* |
1134 | * As part of reopen prepare we also want to create new fd by | |
1135 | * raw_reconfigure_getfd(). But it wants updated "perm", when in | |
1136 | * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to | |
1137 | * permission update. Happily, permission update is always a part (a seprate | |
1138 | * stage) of bdrv_reopen_multiple() so we can rely on this fact and | |
1139 | * reconfigure fd in raw_check_perm(). | |
1140 | */ | |
df26a350 | 1141 | |
e0c9cf3a | 1142 | s->reopen_state = state; |
a6aeca0c | 1143 | ret = 0; |
72373e40 | 1144 | |
31be8a2a SH |
1145 | out: |
1146 | qemu_opts_del(opts); | |
eeb6b45d JC |
1147 | return ret; |
1148 | } | |
1149 | ||
eeb6b45d JC |
1150 | static void raw_reopen_commit(BDRVReopenState *state) |
1151 | { | |
4e6d13c9 | 1152 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1153 | BDRVRawState *s = state->bs->opaque; |
1154 | ||
f357fcd8 | 1155 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1156 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1157 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1158 | g_free(state->opaque); |
1159 | state->opaque = NULL; | |
e0c9cf3a KW |
1160 | |
1161 | assert(s->reopen_state == state); | |
1162 | s->reopen_state = NULL; | |
eeb6b45d JC |
1163 | } |
1164 | ||
1165 | ||
1166 | static void raw_reopen_abort(BDRVReopenState *state) | |
1167 | { | |
4e6d13c9 | 1168 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1169 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1170 | |
1171 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1172 | if (rs == NULL) { |
eeb6b45d JC |
1173 | return; |
1174 | } | |
1175 | ||
eeb6b45d JC |
1176 | g_free(state->opaque); |
1177 | state->opaque = NULL; | |
e0c9cf3a KW |
1178 | |
1179 | assert(s->reopen_state == state); | |
1180 | s->reopen_state = NULL; | |
eeb6b45d JC |
1181 | } |
1182 | ||
18473467 | 1183 | static int hdev_get_max_hw_transfer(int fd, struct stat *st) |
6f607174 FZ |
1184 | { |
1185 | #ifdef BLKSECTGET | |
18473467 PB |
1186 | if (S_ISBLK(st->st_mode)) { |
1187 | unsigned short max_sectors = 0; | |
1188 | if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) { | |
1189 | return max_sectors * 512; | |
1190 | } | |
6f607174 | 1191 | } else { |
18473467 PB |
1192 | int max_bytes = 0; |
1193 | if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
1194 | return max_bytes; | |
1195 | } | |
6f607174 | 1196 | } |
18473467 | 1197 | return -errno; |
6f607174 FZ |
1198 | #else |
1199 | return -ENOSYS; | |
1200 | #endif | |
1201 | } | |
1202 | ||
18473467 | 1203 | static int hdev_get_max_segments(int fd, struct stat *st) |
9103f1ce FZ |
1204 | { |
1205 | #ifdef CONFIG_LINUX | |
1206 | char buf[32]; | |
1207 | const char *end; | |
867eccfe | 1208 | char *sysfspath = NULL; |
9103f1ce | 1209 | int ret; |
867eccfe | 1210 | int sysfd = -1; |
9103f1ce | 1211 | long max_segments; |
867eccfe | 1212 | |
18473467 | 1213 | if (S_ISCHR(st->st_mode)) { |
8ad5ab61 PB |
1214 | if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) { |
1215 | return ret; | |
1216 | } | |
1217 | return -ENOTSUP; | |
1218 | } | |
1219 | ||
18473467 | 1220 | if (!S_ISBLK(st->st_mode)) { |
8ad5ab61 PB |
1221 | return -ENOTSUP; |
1222 | } | |
1223 | ||
9103f1ce | 1224 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", |
18473467 | 1225 | major(st->st_rdev), minor(st->st_rdev)); |
867eccfe ML |
1226 | sysfd = open(sysfspath, O_RDONLY); |
1227 | if (sysfd == -1) { | |
9103f1ce FZ |
1228 | ret = -errno; |
1229 | goto out; | |
1230 | } | |
1231 | do { | |
867eccfe | 1232 | ret = read(sysfd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1233 | } while (ret == -1 && errno == EINTR); |
1234 | if (ret < 0) { | |
1235 | ret = -errno; | |
1236 | goto out; | |
1237 | } else if (ret == 0) { | |
1238 | ret = -EIO; | |
1239 | goto out; | |
1240 | } | |
1241 | buf[ret] = 0; | |
1242 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1243 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1244 | if (ret == 0 && end && *end == '\n') { | |
1245 | ret = max_segments; | |
1246 | } | |
1247 | ||
1248 | out: | |
867eccfe ML |
1249 | if (sysfd != -1) { |
1250 | close(sysfd); | |
fed414df | 1251 | } |
9103f1ce FZ |
1252 | g_free(sysfspath); |
1253 | return ret; | |
1254 | #else | |
1255 | return -ENOTSUP; | |
1256 | #endif | |
1257 | } | |
1258 | ||
3baca891 | 1259 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1260 | { |
1261 | BDRVRawState *s = bs->opaque; | |
18473467 | 1262 | struct stat st; |
6f607174 | 1263 | |
5dbd0ce1 | 1264 | s->needs_alignment = raw_needs_alignment(bs); |
18473467 | 1265 | raw_probe_alignment(bs, s->fd, errp); |
5dbd0ce1 | 1266 | |
18473467 PB |
1267 | bs->bl.min_mem_alignment = s->buf_align; |
1268 | bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); | |
1269 | ||
1270 | /* | |
1271 | * Maximum transfers are best effort, so it is okay to ignore any | |
1272 | * errors. That said, based on the man page errors in fstat would be | |
1273 | * very much unexpected; the only possible case seems to be ENOMEM. | |
1274 | */ | |
1275 | if (fstat(s->fd, &st)) { | |
1276 | return; | |
1277 | } | |
1278 | ||
0dfc7af2 AO |
1279 | #if defined(__APPLE__) && (__MACH__) |
1280 | struct statfs buf; | |
1281 | ||
1282 | if (!fstatfs(s->fd, &buf)) { | |
1283 | bs->bl.opt_transfer = buf.f_iosize; | |
1284 | bs->bl.pdiscard_alignment = buf.f_bsize; | |
1285 | } | |
1286 | #endif | |
1287 | ||
18473467 PB |
1288 | if (bs->sg || S_ISBLK(st.st_mode)) { |
1289 | int ret = hdev_get_max_hw_transfer(s->fd, &st); | |
867eccfe ML |
1290 | |
1291 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
18473467 | 1292 | bs->bl.max_hw_transfer = ret; |
867eccfe ML |
1293 | } |
1294 | ||
18473467 | 1295 | ret = hdev_get_max_segments(s->fd, &st); |
867eccfe | 1296 | if (ret > 0) { |
cc071629 | 1297 | bs->bl.max_hw_iov = ret; |
6f607174 FZ |
1298 | } |
1299 | } | |
c25f53b0 | 1300 | } |
83f64091 | 1301 | |
1a9335e4 ET |
1302 | static int check_for_dasd(int fd) |
1303 | { | |
1304 | #ifdef BIODASDINFO2 | |
1305 | struct dasd_information2_t info = {0}; | |
1306 | ||
1307 | return ioctl(fd, BIODASDINFO2, &info); | |
1308 | #else | |
1309 | return -1; | |
1310 | #endif | |
1311 | } | |
1312 | ||
1313 | /** | |
1314 | * Try to get @bs's logical and physical block size. | |
1315 | * On success, store them in @bsz and return zero. | |
1316 | * On failure, return negative errno. | |
1317 | */ | |
1318 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1319 | { | |
1320 | BDRVRawState *s = bs->opaque; | |
1321 | int ret; | |
1322 | ||
1323 | /* If DASD, get blocksizes */ | |
1324 | if (check_for_dasd(s->fd) < 0) { | |
1325 | return -ENOTSUP; | |
1326 | } | |
1327 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1328 | if (ret < 0) { | |
1329 | return ret; | |
1330 | } | |
1331 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1332 | } | |
1333 | ||
1334 | /** | |
1335 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1336 | * On success, store them in @geo and return 0. | |
1337 | * On failure return -errno. | |
1338 | * (Allows block driver to assign default geometry values that guest sees) | |
1339 | */ | |
1340 | #ifdef __linux__ | |
1341 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1342 | { | |
1343 | BDRVRawState *s = bs->opaque; | |
1344 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1345 | |
1346 | /* If DASD, get its geometry */ | |
1347 | if (check_for_dasd(s->fd) < 0) { | |
1348 | return -ENOTSUP; | |
1349 | } | |
1350 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1351 | return -errno; | |
1352 | } | |
1353 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1354 | (e.g. certain multipath setups) */ | |
1355 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1356 | return -ENOTSUP; | |
1357 | } | |
1358 | /* Do not return a geometry for partition */ | |
1359 | if (ioctl_geo.start != 0) { | |
1360 | return -ENOTSUP; | |
1361 | } | |
1362 | geo->heads = ioctl_geo.heads; | |
1363 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1364 | geo->cylinders = ioctl_geo.cylinders; |
1365 | ||
1366 | return 0; | |
1367 | } | |
1368 | #else /* __linux__ */ | |
1369 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1370 | { | |
1371 | return -ENOTSUP; | |
1372 | } | |
1373 | #endif | |
1374 | ||
03425671 KW |
1375 | #if defined(__linux__) |
1376 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1377 | { |
03425671 | 1378 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1379 | int ret; |
1380 | ||
bd80936a PB |
1381 | do { |
1382 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); | |
1383 | } while (ret == -1 && errno == EINTR); | |
de81a169 PB |
1384 | if (ret == -1) { |
1385 | return -errno; | |
1386 | } | |
1387 | ||
b608c8dc | 1388 | return 0; |
de81a169 | 1389 | } |
03425671 | 1390 | #endif /* linux */ |
de81a169 | 1391 | |
06dc9bd5 | 1392 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1393 | { |
06dc9bd5 | 1394 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1395 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1396 | int ret; |
1397 | ||
e5bcf967 | 1398 | if (s->page_cache_inconsistent) { |
c7ddc882 | 1399 | return -s->page_cache_inconsistent; |
e5bcf967 KW |
1400 | } |
1401 | ||
de81a169 PB |
1402 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1403 | if (ret == -1) { | |
60ff2ae2 DB |
1404 | trace_file_flush_fdatasync_failed(errno); |
1405 | ||
e5bcf967 KW |
1406 | /* There is no clear definition of the semantics of a failing fsync(), |
1407 | * so we may have to assume the worst. The sad truth is that this | |
1408 | * assumption is correct for Linux. Some pages are now probably marked | |
1409 | * clean in the page cache even though they are inconsistent with the | |
1410 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1411 | * further writeback attempt will be made. We can't get back to a state | |
1412 | * in which we know what is on disk (we would have to rewrite | |
1413 | * everything that was touched since the last fdatasync() at least), so | |
1414 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1415 | * really defined, I have little hope that other OSes are doing better. | |
1416 | * | |
1417 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1418 | * cache. */ | |
1419 | if ((s->open_flags & O_DIRECT) == 0) { | |
c7ddc882 | 1420 | s->page_cache_inconsistent = errno; |
e5bcf967 | 1421 | } |
de81a169 PB |
1422 | return -errno; |
1423 | } | |
1424 | return 0; | |
1425 | } | |
1426 | ||
1427 | #ifdef CONFIG_PREADV | |
1428 | ||
1429 | static bool preadv_present = true; | |
1430 | ||
1431 | static ssize_t | |
1432 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1433 | { | |
1434 | return preadv(fd, iov, nr_iov, offset); | |
1435 | } | |
1436 | ||
1437 | static ssize_t | |
1438 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1439 | { | |
1440 | return pwritev(fd, iov, nr_iov, offset); | |
1441 | } | |
1442 | ||
1443 | #else | |
1444 | ||
1445 | static bool preadv_present = false; | |
1446 | ||
1447 | static ssize_t | |
1448 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1449 | { | |
1450 | return -ENOSYS; | |
1451 | } | |
1452 | ||
1453 | static ssize_t | |
1454 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1455 | { | |
1456 | return -ENOSYS; | |
1457 | } | |
1458 | ||
1459 | #endif | |
1460 | ||
1461 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1462 | { | |
1463 | ssize_t len; | |
1464 | ||
1465 | do { | |
1466 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1467 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1468 | aiocb->io.iov, |
1469 | aiocb->io.niov, | |
de81a169 PB |
1470 | aiocb->aio_offset); |
1471 | else | |
1472 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1473 | aiocb->io.iov, |
1474 | aiocb->io.niov, | |
de81a169 PB |
1475 | aiocb->aio_offset); |
1476 | } while (len == -1 && errno == EINTR); | |
1477 | ||
1478 | if (len == -1) { | |
1479 | return -errno; | |
1480 | } | |
1481 | return len; | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Read/writes the data to/from a given linear buffer. | |
1486 | * | |
1487 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1488 | * reads are only returned if the end of the file is reached. | |
1489 | */ | |
1490 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1491 | { | |
1492 | ssize_t offset = 0; | |
1493 | ssize_t len; | |
1494 | ||
1495 | while (offset < aiocb->aio_nbytes) { | |
1496 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1497 | len = pwrite(aiocb->aio_fildes, | |
1498 | (const char *)buf + offset, | |
1499 | aiocb->aio_nbytes - offset, | |
1500 | aiocb->aio_offset + offset); | |
1501 | } else { | |
1502 | len = pread(aiocb->aio_fildes, | |
1503 | buf + offset, | |
1504 | aiocb->aio_nbytes - offset, | |
1505 | aiocb->aio_offset + offset); | |
1506 | } | |
1507 | if (len == -1 && errno == EINTR) { | |
1508 | continue; | |
61ed73cf SH |
1509 | } else if (len == -1 && errno == EINVAL && |
1510 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1511 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1512 | offset > 0) { | |
1513 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1514 | * after a short read. Assume that O_DIRECT short reads only occur | |
1515 | * at EOF. Therefore this is a short read, not an I/O error. | |
1516 | */ | |
1517 | break; | |
de81a169 PB |
1518 | } else if (len == -1) { |
1519 | offset = -errno; | |
1520 | break; | |
1521 | } else if (len == 0) { | |
1522 | break; | |
1523 | } | |
1524 | offset += len; | |
1525 | } | |
1526 | ||
1527 | return offset; | |
1528 | } | |
1529 | ||
999e6b69 | 1530 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1531 | { |
999e6b69 | 1532 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1533 | ssize_t nbytes; |
1534 | char *buf; | |
1535 | ||
1536 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1537 | /* | |
1538 | * If there is just a single buffer, and it is properly aligned | |
1539 | * we can just use plain pread/pwrite without any problems. | |
1540 | */ | |
d57c44d0 | 1541 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1542 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1543 | goto out; | |
de81a169 PB |
1544 | } |
1545 | /* | |
1546 | * We have more than one iovec, and all are properly aligned. | |
1547 | * | |
1548 | * Try preadv/pwritev first and fall back to linearizing the | |
1549 | * buffer if it's not supported. | |
1550 | */ | |
1551 | if (preadv_present) { | |
1552 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1553 | if (nbytes == aiocb->aio_nbytes || | |
1554 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1555 | goto out; |
de81a169 PB |
1556 | } |
1557 | preadv_present = false; | |
1558 | } | |
1559 | ||
1560 | /* | |
1561 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1562 | * using these interfaces. For now retry using plain | |
1563 | * pread/pwrite? | |
1564 | */ | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * Ok, we have to do it the hard way, copy all segments into | |
1569 | * a single aligned buffer. | |
1570 | */ | |
50d4a858 KW |
1571 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1572 | if (buf == NULL) { | |
54c7ca1b KW |
1573 | nbytes = -ENOMEM; |
1574 | goto out; | |
50d4a858 KW |
1575 | } |
1576 | ||
de81a169 PB |
1577 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1578 | char *p = buf; | |
1579 | int i; | |
1580 | ||
d57c44d0 KW |
1581 | for (i = 0; i < aiocb->io.niov; ++i) { |
1582 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1583 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1584 | } |
8eb029c2 | 1585 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1586 | } |
1587 | ||
1588 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1589 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1590 | char *p = buf; | |
1591 | size_t count = aiocb->aio_nbytes, copy; | |
1592 | int i; | |
1593 | ||
d57c44d0 | 1594 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1595 | copy = count; |
d57c44d0 KW |
1596 | if (copy > aiocb->io.iov[i].iov_len) { |
1597 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1598 | } |
d57c44d0 | 1599 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1600 | assert(count >= copy); |
de81a169 PB |
1601 | p += copy; |
1602 | count -= copy; | |
1603 | } | |
8eb029c2 | 1604 | assert(count == 0); |
de81a169 PB |
1605 | } |
1606 | qemu_vfree(buf); | |
1607 | ||
54c7ca1b KW |
1608 | out: |
1609 | if (nbytes == aiocb->aio_nbytes) { | |
1610 | return 0; | |
1611 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1612 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1613 | return -EINVAL; | |
1614 | } else { | |
1615 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1616 | 0, aiocb->aio_nbytes - nbytes); | |
1617 | return 0; | |
1618 | } | |
1619 | } else { | |
1620 | assert(nbytes < 0); | |
1621 | return nbytes; | |
1622 | } | |
de81a169 PB |
1623 | } |
1624 | ||
0dfc7af2 | 1625 | #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD) |
1486df0e DL |
1626 | static int translate_err(int err) |
1627 | { | |
1628 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1629 | err == -ENOTTY) { | |
1630 | err = -ENOTSUP; | |
1631 | } | |
1632 | return err; | |
1633 | } | |
0dfc7af2 | 1634 | #endif |
1486df0e | 1635 | |
d50d8222 | 1636 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1637 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1638 | { | |
1639 | do { | |
1640 | if (fallocate(fd, mode, offset, len) == 0) { | |
1641 | return 0; | |
1642 | } | |
1643 | } while (errno == EINTR); | |
1644 | return translate_err(-errno); | |
1645 | } | |
1646 | #endif | |
1647 | ||
37cc9f7f | 1648 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1649 | { |
37cc9f7f | 1650 | int ret = -ENOTSUP; |
97a2ae34 PB |
1651 | BDRVRawState *s = aiocb->bs->opaque; |
1652 | ||
37cc9f7f | 1653 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1654 | return -ENOTSUP; |
1655 | } | |
1656 | ||
97a2ae34 | 1657 | #ifdef BLKZEROOUT |
738301e1 KW |
1658 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1659 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1660 | * fallbacks. */ | |
1661 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1662 | do { | |
1663 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1664 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1665 | return 0; | |
1666 | } | |
1667 | } while (errno == EINTR); | |
37cc9f7f | 1668 | |
738301e1 | 1669 | ret = translate_err(-errno); |
effecce6 KW |
1670 | if (ret == -ENOTSUP) { |
1671 | s->has_write_zeroes = false; | |
1672 | } | |
738301e1 | 1673 | } |
97a2ae34 | 1674 | #endif |
97a2ae34 | 1675 | |
97a2ae34 PB |
1676 | return ret; |
1677 | } | |
1678 | ||
7154d8ae | 1679 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1680 | { |
7154d8ae | 1681 | RawPosixAIOData *aiocb = opaque; |
70d9110b | 1682 | #ifdef CONFIG_FALLOCATE |
b2c6f23f | 1683 | BDRVRawState *s = aiocb->bs->opaque; |
70d9110b DL |
1684 | int64_t len; |
1685 | #endif | |
37cc9f7f DL |
1686 | |
1687 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1688 | return handle_aiocb_write_zeroes_block(aiocb); | |
1689 | } | |
1690 | ||
b953f075 DL |
1691 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1692 | if (s->has_write_zeroes) { | |
1693 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1694 | aiocb->aio_offset, aiocb->aio_nbytes); | |
fa95e9fb TH |
1695 | if (ret == -ENOTSUP) { |
1696 | s->has_write_zeroes = false; | |
1697 | } else if (ret == 0 || ret != -EINVAL) { | |
b953f075 DL |
1698 | return ret; |
1699 | } | |
fa95e9fb TH |
1700 | /* |
1701 | * Note: Some file systems do not like unaligned byte ranges, and | |
1702 | * return EINVAL in such a case, though they should not do it according | |
1703 | * to the man-page of fallocate(). Thus we simply ignore this return | |
1704 | * value and try the other fallbacks instead. | |
1705 | */ | |
b953f075 DL |
1706 | } |
1707 | #endif | |
1708 | ||
1cdc3239 DL |
1709 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1710 | if (s->has_discard && s->has_fallocate) { | |
1711 | int ret = do_fallocate(s->fd, | |
1712 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1713 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1714 | if (ret == 0) { | |
1715 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1716 | if (ret == 0 || ret != -ENOTSUP) { | |
1717 | return ret; | |
1718 | } | |
1719 | s->has_fallocate = false; | |
73ebf297 TH |
1720 | } else if (ret == -EINVAL) { |
1721 | /* | |
1722 | * Some file systems like older versions of GPFS do not like un- | |
1723 | * aligned byte ranges, and return EINVAL in such a case, though | |
1724 | * they should not do it according to the man-page of fallocate(). | |
1725 | * Warn about the bad filesystem and try the final fallback instead. | |
1726 | */ | |
1727 | warn_report_once("Your file system is misbehaving: " | |
1728 | "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. " | |
68857f13 | 1729 | "Please report this bug to your file system " |
73ebf297 | 1730 | "vendor."); |
1cdc3239 DL |
1731 | } else if (ret != -ENOTSUP) { |
1732 | return ret; | |
1733 | } else { | |
1734 | s->has_discard = false; | |
1735 | } | |
1736 | } | |
1737 | #endif | |
1738 | ||
d50d8222 | 1739 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1740 | /* Last resort: we are trying to extend the file with zeroed data. This |
1741 | * can be done via fallocate(fd, 0) */ | |
1742 | len = bdrv_getlength(aiocb->bs); | |
1743 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1744 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1745 | if (ret == 0 || ret != -ENOTSUP) { | |
1746 | return ret; | |
1747 | } | |
1748 | s->has_fallocate = false; | |
1749 | } | |
1750 | #endif | |
1751 | ||
37cc9f7f DL |
1752 | return -ENOTSUP; |
1753 | } | |
1754 | ||
7154d8ae | 1755 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1756 | { |
7154d8ae | 1757 | RawPosixAIOData *aiocb = opaque; |
34fa110e | 1758 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
34fa110e KW |
1759 | |
1760 | /* First try to write zeros and unmap at the same time */ | |
1761 | ||
1762 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
b3ac2b94 SS |
1763 | int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1764 | aiocb->aio_offset, aiocb->aio_nbytes); | |
bae127d4 AD |
1765 | switch (ret) { |
1766 | case -ENOTSUP: | |
1767 | case -EINVAL: | |
ece4fa91 | 1768 | case -EBUSY: |
bae127d4 AD |
1769 | break; |
1770 | default: | |
34fa110e KW |
1771 | return ret; |
1772 | } | |
1773 | #endif | |
1774 | ||
34fa110e KW |
1775 | /* If we couldn't manage to unmap while guaranteed that the area reads as |
1776 | * all-zero afterwards, just write zeroes without unmapping */ | |
b3ac2b94 | 1777 | return handle_aiocb_write_zeroes(aiocb); |
34fa110e KW |
1778 | } |
1779 | ||
1efad060 FZ |
1780 | #ifndef HAVE_COPY_FILE_RANGE |
1781 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1782 | off_t *out_off, size_t len, unsigned int flags) | |
1783 | { | |
1784 | #ifdef __NR_copy_file_range | |
1785 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1786 | out_off, len, flags); | |
1787 | #else | |
1788 | errno = ENOSYS; | |
1789 | return -1; | |
1790 | #endif | |
1791 | } | |
1792 | #endif | |
1793 | ||
58a209c4 | 1794 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1795 | { |
58a209c4 | 1796 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1797 | uint64_t bytes = aiocb->aio_nbytes; |
1798 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1799 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1800 | |
1801 | while (bytes) { | |
1802 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1803 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1804 | bytes, 0); |
ecc983a5 | 1805 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1806 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1807 | 0, ret); | |
c436e3d0 FZ |
1808 | if (ret == 0) { |
1809 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1810 | * buffer I/O. */ | |
1811 | return -ENOSPC; | |
1efad060 FZ |
1812 | } |
1813 | if (ret < 0) { | |
c436e3d0 FZ |
1814 | switch (errno) { |
1815 | case ENOSYS: | |
1efad060 | 1816 | return -ENOTSUP; |
c436e3d0 FZ |
1817 | case EINTR: |
1818 | continue; | |
1819 | default: | |
1efad060 FZ |
1820 | return -errno; |
1821 | } | |
1822 | } | |
1efad060 FZ |
1823 | bytes -= ret; |
1824 | } | |
1825 | return 0; | |
1826 | } | |
1827 | ||
46ee0f46 | 1828 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1829 | { |
46ee0f46 | 1830 | RawPosixAIOData *aiocb = opaque; |
13a02833 | 1831 | int ret = -ENOTSUP; |
8238010b PB |
1832 | BDRVRawState *s = aiocb->bs->opaque; |
1833 | ||
7ce21016 PB |
1834 | if (!s->has_discard) { |
1835 | return -ENOTSUP; | |
8238010b PB |
1836 | } |
1837 | ||
1838 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1839 | #ifdef BLKDISCARD | |
1840 | do { | |
1841 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1842 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1843 | return 0; | |
1844 | } | |
1845 | } while (errno == EINTR); | |
1846 | ||
0dfc7af2 | 1847 | ret = translate_err(-errno); |
8238010b PB |
1848 | #endif |
1849 | } else { | |
8238010b | 1850 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
0b991712 DL |
1851 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1852 | aiocb->aio_offset, aiocb->aio_nbytes); | |
13a02833 | 1853 | ret = translate_err(ret); |
0dfc7af2 AO |
1854 | #elif defined(__APPLE__) && (__MACH__) |
1855 | fpunchhole_t fpunchhole; | |
1856 | fpunchhole.fp_flags = 0; | |
1857 | fpunchhole.reserved = 0; | |
1858 | fpunchhole.fp_offset = aiocb->aio_offset; | |
1859 | fpunchhole.fp_length = aiocb->aio_nbytes; | |
1860 | if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) { | |
1861 | ret = errno == ENODEV ? -ENOTSUP : -errno; | |
1862 | } else { | |
1863 | ret = 0; | |
1864 | } | |
8238010b PB |
1865 | #endif |
1866 | } | |
1867 | ||
1486df0e | 1868 | if (ret == -ENOTSUP) { |
7ce21016 | 1869 | s->has_discard = false; |
8238010b PB |
1870 | } |
1871 | return ret; | |
1872 | } | |
1873 | ||
3a20013f NS |
1874 | /* |
1875 | * Help alignment probing by allocating the first block. | |
1876 | * | |
1877 | * When reading with direct I/O from unallocated area on Gluster backed by XFS, | |
1878 | * reading succeeds regardless of request length. In this case we fallback to | |
1879 | * safe alignment which is not optimal. Allocating the first block avoids this | |
1880 | * fallback. | |
1881 | * | |
1882 | * fd may be opened with O_DIRECT, but we don't know the buffer alignment or | |
1883 | * request alignment, so we use safe values. | |
1884 | * | |
1885 | * Returns: 0 on success, -errno on failure. Since this is an optimization, | |
1886 | * caller may ignore failures. | |
1887 | */ | |
1888 | static int allocate_first_block(int fd, size_t max_size) | |
1889 | { | |
1890 | size_t write_size = (max_size < MAX_BLOCKSIZE) | |
1891 | ? BDRV_SECTOR_SIZE | |
1892 | : MAX_BLOCKSIZE; | |
038adc2f | 1893 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); |
3a20013f NS |
1894 | void *buf; |
1895 | ssize_t n; | |
1896 | int ret; | |
1897 | ||
1898 | buf = qemu_memalign(max_align, write_size); | |
1899 | memset(buf, 0, write_size); | |
1900 | ||
1901 | do { | |
1902 | n = pwrite(fd, buf, write_size, 0); | |
1903 | } while (n == -1 && errno == EINTR); | |
1904 | ||
1905 | ret = (n == -1) ? -errno : 0; | |
1906 | ||
1907 | qemu_vfree(buf); | |
1908 | return ret; | |
1909 | } | |
1910 | ||
29cb4c01 | 1911 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1912 | { |
29cb4c01 | 1913 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1914 | int result = 0; |
1915 | int64_t current_length = 0; | |
1916 | char *buf = NULL; | |
1917 | struct stat st; | |
1918 | int fd = aiocb->aio_fildes; | |
1919 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1920 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1921 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1922 | |
1923 | if (fstat(fd, &st) < 0) { | |
1924 | result = -errno; | |
1925 | error_setg_errno(errp, -result, "Could not stat file"); | |
1926 | return result; | |
1927 | } | |
1928 | ||
1929 | current_length = st.st_size; | |
d57c44d0 | 1930 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1931 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1932 | return -ENOTSUP; | |
1933 | } | |
1934 | ||
d57c44d0 | 1935 | switch (prealloc) { |
93f4e2ff KW |
1936 | #ifdef CONFIG_POSIX_FALLOCATE |
1937 | case PREALLOC_MODE_FALLOC: | |
1938 | /* | |
1939 | * Truncating before posix_fallocate() makes it about twice slower on | |
1940 | * file systems that do not support fallocate(), trying to check if a | |
1941 | * block is allocated before allocating it, so don't do that here. | |
1942 | */ | |
1943 | if (offset != current_length) { | |
1944 | result = -posix_fallocate(fd, current_length, | |
1945 | offset - current_length); | |
1946 | if (result != 0) { | |
1947 | /* posix_fallocate() doesn't set errno. */ | |
1948 | error_setg_errno(errp, -result, | |
1949 | "Could not preallocate new data"); | |
3a20013f NS |
1950 | } else if (current_length == 0) { |
1951 | /* | |
1952 | * posix_fallocate() uses fallocate() if the filesystem | |
1953 | * supports it, or fallback to manually writing zeroes. If | |
1954 | * fallocate() was used, unaligned reads from the fallocated | |
1955 | * area in raw_probe_alignment() will succeed, hence we need to | |
1956 | * allocate the first block. | |
1957 | * | |
1958 | * Optimize future alignment probing; ignore failures. | |
1959 | */ | |
1960 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1961 | } |
1962 | } else { | |
1963 | result = 0; | |
1964 | } | |
1965 | goto out; | |
1966 | #endif | |
1967 | case PREALLOC_MODE_FULL: | |
1968 | { | |
1969 | int64_t num = 0, left = offset - current_length; | |
1970 | off_t seek_result; | |
1971 | ||
1972 | /* | |
1973 | * Knowing the final size from the beginning could allow the file | |
1974 | * system driver to do less allocations and possibly avoid | |
1975 | * fragmentation of the file. | |
1976 | */ | |
1977 | if (ftruncate(fd, offset) != 0) { | |
1978 | result = -errno; | |
1979 | error_setg_errno(errp, -result, "Could not resize file"); | |
1980 | goto out; | |
1981 | } | |
1982 | ||
1983 | buf = g_malloc0(65536); | |
1984 | ||
1985 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1986 | if (seek_result < 0) { | |
1987 | result = -errno; | |
1988 | error_setg_errno(errp, -result, | |
1989 | "Failed to seek to the old end of file"); | |
1990 | goto out; | |
1991 | } | |
1992 | ||
1993 | while (left > 0) { | |
1994 | num = MIN(left, 65536); | |
1995 | result = write(fd, buf, num); | |
1996 | if (result < 0) { | |
a1c81f4f FZ |
1997 | if (errno == EINTR) { |
1998 | continue; | |
1999 | } | |
93f4e2ff KW |
2000 | result = -errno; |
2001 | error_setg_errno(errp, -result, | |
2002 | "Could not write zeros for preallocation"); | |
2003 | goto out; | |
2004 | } | |
2005 | left -= result; | |
2006 | } | |
2007 | if (result >= 0) { | |
2008 | result = fsync(fd); | |
2009 | if (result < 0) { | |
2010 | result = -errno; | |
2011 | error_setg_errno(errp, -result, | |
2012 | "Could not flush file to disk"); | |
2013 | goto out; | |
2014 | } | |
2015 | } | |
2016 | goto out; | |
2017 | } | |
2018 | case PREALLOC_MODE_OFF: | |
2019 | if (ftruncate(fd, offset) != 0) { | |
2020 | result = -errno; | |
2021 | error_setg_errno(errp, -result, "Could not resize file"); | |
3a20013f NS |
2022 | } else if (current_length == 0 && offset > current_length) { |
2023 | /* Optimize future alignment probing; ignore failures. */ | |
2024 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
2025 | } |
2026 | return result; | |
2027 | default: | |
2028 | result = -ENOTSUP; | |
2029 | error_setg(errp, "Unsupported preallocation mode: %s", | |
d57c44d0 | 2030 | PreallocMode_str(prealloc)); |
93f4e2ff KW |
2031 | return result; |
2032 | } | |
2033 | ||
2034 | out: | |
2035 | if (result < 0) { | |
2036 | if (ftruncate(fd, current_length) < 0) { | |
2037 | error_report("Failed to restore old file length: %s", | |
2038 | strerror(errno)); | |
2039 | } | |
2040 | } | |
2041 | ||
2042 | g_free(buf); | |
2043 | return result; | |
2044 | } | |
2045 | ||
5d5de250 KW |
2046 | static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, |
2047 | ThreadPoolFunc func, void *arg) | |
2048 | { | |
2049 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ | |
2050 | ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
2051 | return thread_pool_submit_co(pool, func, arg); | |
2052 | } | |
2053 | ||
9d52aa3c KW |
2054 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
2055 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 2056 | { |
ce1a14dc | 2057 | BDRVRawState *s = bs->opaque; |
999e6b69 | 2058 | RawPosixAIOData acb; |
ce1a14dc | 2059 | |
19cb3738 | 2060 | if (fd_open(bs) < 0) |
2174f12b | 2061 | return -EIO; |
19cb3738 | 2062 | |
f141eafe | 2063 | /* |
c6447510 AM |
2064 | * When using O_DIRECT, the request must be aligned to be able to use |
2065 | * either libaio or io_uring interface. If not fail back to regular thread | |
2066 | * pool read/write code which emulates this for us if we | |
2067 | * set QEMU_AIO_MISALIGNED. | |
f141eafe | 2068 | */ |
c6447510 AM |
2069 | if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { |
2070 | type |= QEMU_AIO_MISALIGNED; | |
2071 | #ifdef CONFIG_LINUX_IO_URING | |
2072 | } else if (s->use_linux_io_uring) { | |
2073 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2074 | assert(qiov->size == bytes); | |
2075 | return luring_co_submit(bs, aio, s->fd, offset, qiov, type); | |
2076 | #endif | |
e44bd6fc | 2077 | #ifdef CONFIG_LINUX_AIO |
c6447510 AM |
2078 | } else if (s->use_linux_aio) { |
2079 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); | |
2080 | assert(qiov->size == bytes); | |
512da211 SG |
2081 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type, |
2082 | s->aio_max_batch); | |
e44bd6fc | 2083 | #endif |
9ef91a67 | 2084 | } |
f141eafe | 2085 | |
999e6b69 KW |
2086 | acb = (RawPosixAIOData) { |
2087 | .bs = bs, | |
2088 | .aio_fildes = s->fd, | |
2089 | .aio_type = type, | |
2090 | .aio_offset = offset, | |
2091 | .aio_nbytes = bytes, | |
2092 | .io = { | |
2093 | .iov = qiov->iov, | |
2094 | .niov = qiov->niov, | |
2095 | }, | |
2096 | }; | |
2097 | ||
2098 | assert(qiov->size == bytes); | |
2099 | return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); | |
2174f12b KW |
2100 | } |
2101 | ||
f7ef38dd VSO |
2102 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset, |
2103 | int64_t bytes, QEMUIOVector *qiov, | |
2104 | BdrvRequestFlags flags) | |
2174f12b | 2105 | { |
9d52aa3c | 2106 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
2107 | } |
2108 | ||
e75abeda VSO |
2109 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset, |
2110 | int64_t bytes, QEMUIOVector *qiov, | |
2111 | BdrvRequestFlags flags) | |
2174f12b | 2112 | { |
9d52aa3c KW |
2113 | assert(flags == 0); |
2114 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
2115 | } |
2116 | ||
1b3abdcc ML |
2117 | static void raw_aio_plug(BlockDriverState *bs) |
2118 | { | |
c6447510 | 2119 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2120 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2121 | if (s->use_linux_aio) { |
0187f5c9 PB |
2122 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
2123 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
2124 | } |
2125 | #endif | |
c6447510 AM |
2126 | #ifdef CONFIG_LINUX_IO_URING |
2127 | if (s->use_linux_io_uring) { | |
2128 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2129 | luring_io_plug(bs, aio); | |
2130 | } | |
2131 | #endif | |
1b3abdcc ML |
2132 | } |
2133 | ||
2134 | static void raw_aio_unplug(BlockDriverState *bs) | |
2135 | { | |
c6447510 | 2136 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2137 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2138 | if (s->use_linux_aio) { |
0187f5c9 | 2139 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
68d79466 | 2140 | laio_io_unplug(bs, aio, s->aio_max_batch); |
1b3abdcc ML |
2141 | } |
2142 | #endif | |
c6447510 AM |
2143 | #ifdef CONFIG_LINUX_IO_URING |
2144 | if (s->use_linux_io_uring) { | |
2145 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2146 | luring_io_unplug(bs, aio); | |
2147 | } | |
2148 | #endif | |
1b3abdcc ML |
2149 | } |
2150 | ||
33d70fb6 | 2151 | static int raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
2152 | { |
2153 | BDRVRawState *s = bs->opaque; | |
06dc9bd5 | 2154 | RawPosixAIOData acb; |
33d70fb6 | 2155 | int ret; |
b2e12bc6 | 2156 | |
33d70fb6 KW |
2157 | ret = fd_open(bs); |
2158 | if (ret < 0) { | |
2159 | return ret; | |
2160 | } | |
b2e12bc6 | 2161 | |
06dc9bd5 KW |
2162 | acb = (RawPosixAIOData) { |
2163 | .bs = bs, | |
2164 | .aio_fildes = s->fd, | |
2165 | .aio_type = QEMU_AIO_FLUSH, | |
2166 | }; | |
2167 | ||
c6447510 AM |
2168 | #ifdef CONFIG_LINUX_IO_URING |
2169 | if (s->use_linux_io_uring) { | |
2170 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2171 | return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); | |
2172 | } | |
2173 | #endif | |
06dc9bd5 | 2174 | return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); |
b2e12bc6 CH |
2175 | } |
2176 | ||
ed6e2161 NA |
2177 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
2178 | AioContext *new_context) | |
2179 | { | |
c6447510 | 2180 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
ed6e2161 | 2181 | #ifdef CONFIG_LINUX_AIO |
ed6e2161 | 2182 | if (s->use_linux_aio) { |
cb09104e | 2183 | Error *local_err = NULL; |
ed6e2161 NA |
2184 | if (!aio_setup_linux_aio(new_context, &local_err)) { |
2185 | error_reportf_err(local_err, "Unable to use native AIO, " | |
2186 | "falling back to thread pool: "); | |
2187 | s->use_linux_aio = false; | |
2188 | } | |
2189 | } | |
2190 | #endif | |
c6447510 AM |
2191 | #ifdef CONFIG_LINUX_IO_URING |
2192 | if (s->use_linux_io_uring) { | |
cb8d0851 | 2193 | Error *local_err = NULL; |
c6447510 AM |
2194 | if (!aio_setup_linux_io_uring(new_context, &local_err)) { |
2195 | error_reportf_err(local_err, "Unable to use linux io_uring, " | |
2196 | "falling back to thread pool: "); | |
2197 | s->use_linux_io_uring = false; | |
2198 | } | |
2199 | } | |
2200 | #endif | |
ed6e2161 NA |
2201 | } |
2202 | ||
83f64091 FB |
2203 | static void raw_close(BlockDriverState *bs) |
2204 | { | |
2205 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 2206 | |
19cb3738 | 2207 | if (s->fd >= 0) { |
2e1e79da | 2208 | qemu_close(s->fd); |
19cb3738 FB |
2209 | s->fd = -1; |
2210 | } | |
83f64091 FB |
2211 | } |
2212 | ||
d0bc9e5d HR |
2213 | /** |
2214 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
2215 | * new space according to @prealloc. | |
2216 | * | |
2217 | * Returns: 0 on success, -errno on failure. | |
2218 | */ | |
93f4e2ff KW |
2219 | static int coroutine_fn |
2220 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
2221 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 2222 | { |
29cb4c01 | 2223 | RawPosixAIOData acb; |
d0bc9e5d | 2224 | |
29cb4c01 | 2225 | acb = (RawPosixAIOData) { |
93f4e2ff KW |
2226 | .bs = bs, |
2227 | .aio_fildes = fd, | |
2228 | .aio_type = QEMU_AIO_TRUNCATE, | |
2229 | .aio_offset = offset, | |
d57c44d0 KW |
2230 | .truncate = { |
2231 | .prealloc = prealloc, | |
2232 | .errp = errp, | |
2233 | }, | |
93f4e2ff | 2234 | }; |
d0bc9e5d | 2235 | |
29cb4c01 | 2236 | return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); |
9f63b07e HR |
2237 | } |
2238 | ||
061ca8a3 | 2239 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
c80d8b06 | 2240 | bool exact, PreallocMode prealloc, |
92b92799 | 2241 | BdrvRequestFlags flags, Error **errp) |
83f64091 FB |
2242 | { |
2243 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 2244 | struct stat st; |
f59adb32 | 2245 | int ret; |
55b949c8 CH |
2246 | |
2247 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
2248 | ret = -errno; |
2249 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
2250 | return ret; | |
55b949c8 CH |
2251 | } |
2252 | ||
2253 | if (S_ISREG(st.st_mode)) { | |
82325ae5 | 2254 | /* Always resizes to the exact @offset */ |
93f4e2ff | 2255 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
2256 | } |
2257 | ||
2258 | if (prealloc != PREALLOC_MODE_OFF) { | |
2259 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 2260 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
2261 | return -ENOTSUP; |
2262 | } | |
2263 | ||
2264 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
82325ae5 HR |
2265 | int64_t cur_length = raw_getlength(bs); |
2266 | ||
2267 | if (offset != cur_length && exact) { | |
2268 | error_setg(errp, "Cannot resize device files"); | |
2269 | return -ENOTSUP; | |
2270 | } else if (offset > cur_length) { | |
f59adb32 HR |
2271 | error_setg(errp, "Cannot grow device files"); |
2272 | return -EINVAL; | |
2273 | } | |
55b949c8 | 2274 | } else { |
f59adb32 | 2275 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
2276 | return -ENOTSUP; |
2277 | } | |
2278 | ||
83f64091 FB |
2279 | return 0; |
2280 | } | |
2281 | ||
128ab2ff BS |
2282 | #ifdef __OpenBSD__ |
2283 | static int64_t raw_getlength(BlockDriverState *bs) | |
2284 | { | |
2285 | BDRVRawState *s = bs->opaque; | |
2286 | int fd = s->fd; | |
2287 | struct stat st; | |
2288 | ||
2289 | if (fstat(fd, &st)) | |
aa729704 | 2290 | return -errno; |
128ab2ff BS |
2291 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2292 | struct disklabel dl; | |
2293 | ||
2294 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2295 | return -errno; |
128ab2ff BS |
2296 | return (uint64_t)dl.d_secsize * |
2297 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2298 | } else | |
2299 | return st.st_size; | |
2300 | } | |
d1f6fd8d CE |
2301 | #elif defined(__NetBSD__) |
2302 | static int64_t raw_getlength(BlockDriverState *bs) | |
2303 | { | |
2304 | BDRVRawState *s = bs->opaque; | |
2305 | int fd = s->fd; | |
2306 | struct stat st; | |
2307 | ||
2308 | if (fstat(fd, &st)) | |
aa729704 | 2309 | return -errno; |
d1f6fd8d CE |
2310 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2311 | struct dkwedge_info dkw; | |
2312 | ||
2313 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
2314 | return dkw.dkw_size * 512; | |
2315 | } else { | |
2316 | struct disklabel dl; | |
2317 | ||
2318 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2319 | return -errno; |
d1f6fd8d CE |
2320 | return (uint64_t)dl.d_secsize * |
2321 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2322 | } | |
2323 | } else | |
2324 | return st.st_size; | |
2325 | } | |
50779cc2 CH |
2326 | #elif defined(__sun__) |
2327 | static int64_t raw_getlength(BlockDriverState *bs) | |
2328 | { | |
2329 | BDRVRawState *s = bs->opaque; | |
2330 | struct dk_minfo minfo; | |
2331 | int ret; | |
aa729704 | 2332 | int64_t size; |
50779cc2 CH |
2333 | |
2334 | ret = fd_open(bs); | |
2335 | if (ret < 0) { | |
2336 | return ret; | |
2337 | } | |
2338 | ||
2339 | /* | |
2340 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
2341 | */ | |
2342 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
2343 | if (ret != -1) { | |
2344 | return minfo.dki_lbsize * minfo.dki_capacity; | |
2345 | } | |
2346 | ||
2347 | /* | |
2348 | * There are reports that lseek on some devices fails, but | |
2349 | * irc discussion said that contingency on contingency was overkill. | |
2350 | */ | |
aa729704 MA |
2351 | size = lseek(s->fd, 0, SEEK_END); |
2352 | if (size < 0) { | |
2353 | return -errno; | |
2354 | } | |
2355 | return size; | |
50779cc2 CH |
2356 | } |
2357 | #elif defined(CONFIG_BSD) | |
2358 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2359 | { |
2360 | BDRVRawState *s = bs->opaque; | |
2361 | int fd = s->fd; | |
2362 | int64_t size; | |
83f64091 | 2363 | struct stat sb; |
a167ba50 | 2364 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2365 | int reopened = 0; |
83f64091 | 2366 | #endif |
19cb3738 FB |
2367 | int ret; |
2368 | ||
2369 | ret = fd_open(bs); | |
2370 | if (ret < 0) | |
2371 | return ret; | |
83f64091 | 2372 | |
a167ba50 | 2373 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2374 | again: |
2375 | #endif | |
83f64091 | 2376 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
267cd53f | 2377 | size = 0; |
83f64091 | 2378 | #ifdef DIOCGMEDIASIZE |
267cd53f PB |
2379 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) { |
2380 | size = 0; | |
2381 | } | |
2382 | #endif | |
2383 | #ifdef DIOCGPART | |
2384 | if (size == 0) { | |
2385 | struct partinfo pi; | |
2386 | if (ioctl(fd, DIOCGPART, &pi) == 0) { | |
2387 | size = pi.media_size; | |
2388 | } | |
c5e97233 | 2389 | } |
83f64091 | 2390 | #endif |
09e20abd | 2391 | #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE) |
267cd53f | 2392 | if (size == 0) { |
728dacbd JA |
2393 | uint64_t sectors = 0; |
2394 | uint32_t sector_size = 0; | |
2395 | ||
2396 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2397 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2398 | size = sectors * sector_size; | |
728dacbd JA |
2399 | } |
2400 | } | |
267cd53f PB |
2401 | #endif |
2402 | if (size == 0) { | |
2403 | size = lseek(fd, 0LL, SEEK_END); | |
2404 | } | |
aa729704 MA |
2405 | if (size < 0) { |
2406 | return -errno; | |
2407 | } | |
a167ba50 | 2408 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2409 | switch(s->type) { |
2410 | case FTYPE_CD: | |
2411 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2412 | if (size == 2048LL * (unsigned)-1) | |
2413 | size = 0; | |
2414 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2415 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2416 | reopened = 1; |
2417 | goto again; | |
2418 | } | |
2419 | } | |
83f64091 | 2420 | #endif |
50779cc2 | 2421 | } else { |
83f64091 | 2422 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2423 | if (size < 0) { |
2424 | return -errno; | |
2425 | } | |
83f64091 | 2426 | } |
83f64091 FB |
2427 | return size; |
2428 | } | |
50779cc2 CH |
2429 | #else |
2430 | static int64_t raw_getlength(BlockDriverState *bs) | |
2431 | { | |
2432 | BDRVRawState *s = bs->opaque; | |
2433 | int ret; | |
aa729704 | 2434 | int64_t size; |
50779cc2 CH |
2435 | |
2436 | ret = fd_open(bs); | |
2437 | if (ret < 0) { | |
2438 | return ret; | |
2439 | } | |
2440 | ||
aa729704 MA |
2441 | size = lseek(s->fd, 0, SEEK_END); |
2442 | if (size < 0) { | |
2443 | return -errno; | |
2444 | } | |
2445 | return size; | |
50779cc2 | 2446 | } |
128ab2ff | 2447 | #endif |
83f64091 | 2448 | |
4a1d5e1f FZ |
2449 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2450 | { | |
2451 | struct stat st; | |
2452 | BDRVRawState *s = bs->opaque; | |
2453 | ||
2454 | if (fstat(s->fd, &st) < 0) { | |
2455 | return -errno; | |
2456 | } | |
2457 | return (int64_t)st.st_blocks * 512; | |
2458 | } | |
2459 | ||
93f4e2ff KW |
2460 | static int coroutine_fn |
2461 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2462 | { |
927f11e1 | 2463 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2464 | Error *local_err = NULL; |
83f64091 | 2465 | int fd; |
d815efca | 2466 | uint64_t perm, shared; |
1e37d059 | 2467 | int result = 0; |
83f64091 | 2468 | |
927f11e1 KW |
2469 | /* Validate options and set default values */ |
2470 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2471 | file_opts = &options->u.file; | |
464d9f64 | 2472 | |
927f11e1 KW |
2473 | if (!file_opts->has_nocow) { |
2474 | file_opts->nocow = false; | |
2475 | } | |
2476 | if (!file_opts->has_preallocation) { | |
2477 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2478 | } |
ffa244c8 KW |
2479 | if (!file_opts->has_extent_size_hint) { |
2480 | file_opts->extent_size_hint = 1 * MiB; | |
2481 | } | |
2482 | if (file_opts->extent_size_hint > UINT32_MAX) { | |
2483 | result = -EINVAL; | |
2484 | error_setg(errp, "Extent size hint is too large"); | |
2485 | goto out; | |
2486 | } | |
83f64091 | 2487 | |
927f11e1 | 2488 | /* Create file */ |
b18a24a9 | 2489 | fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); |
1e37d059 SW |
2490 | if (fd < 0) { |
2491 | result = -errno; | |
06247428 HT |
2492 | goto out; |
2493 | } | |
2494 | ||
b8cf1913 HR |
2495 | /* Take permissions: We want to discard everything, so we need |
2496 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2497 | * BLK_PERM_RESIZE. | |
2498 | * On the other hand, we cannot share the RESIZE permission | |
2499 | * because we promise that after this function, the file has the | |
2500 | * size given in the options. If someone else were to resize it | |
2501 | * concurrently, we could not guarantee that. | |
2502 | * Note that after this function, we can no longer guarantee that | |
2503 | * the file is not touched by a third party, so it may be resized | |
2504 | * then. */ | |
2505 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2506 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2507 | ||
2508 | /* Step one: Take locks */ | |
2996ffad | 2509 | result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2510 | if (result < 0) { |
2511 | goto out_close; | |
2512 | } | |
2513 | ||
2514 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2515 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2516 | if (result < 0) { | |
b857431d FZ |
2517 | error_append_hint(errp, |
2518 | "Is another process using the image [%s]?\n", | |
2519 | file_opts->filename); | |
7c20c808 | 2520 | goto out_unlock; |
b8cf1913 HR |
2521 | } |
2522 | ||
2523 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2524 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2525 | if (result < 0) { |
7c20c808 | 2526 | goto out_unlock; |
b8cf1913 HR |
2527 | } |
2528 | ||
927f11e1 | 2529 | if (file_opts->nocow) { |
4ab15590 | 2530 | #ifdef __linux__ |
06247428 HT |
2531 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2532 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2533 | * will be ignored since any failure of this operation should not | |
2534 | * block the left work. | |
2535 | */ | |
2536 | int attr; | |
2537 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2538 | attr |= FS_NOCOW_FL; | |
2539 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2540 | } |
06247428 HT |
2541 | #endif |
2542 | } | |
ffa244c8 KW |
2543 | #ifdef FS_IOC_FSSETXATTR |
2544 | /* | |
2545 | * Try to set the extent size hint. Failure is not fatal, and a warning is | |
2546 | * only printed if the option was explicitly specified. | |
2547 | */ | |
2548 | { | |
2549 | struct fsxattr attr; | |
2550 | result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); | |
2551 | if (result == 0) { | |
2552 | attr.fsx_xflags |= FS_XFLAG_EXTSIZE; | |
2553 | attr.fsx_extsize = file_opts->extent_size_hint; | |
2554 | result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); | |
2555 | } | |
2556 | if (result < 0 && file_opts->has_extent_size_hint && | |
2557 | file_opts->extent_size_hint) | |
2558 | { | |
2559 | warn_report("Failed to set extent size hint: %s", | |
2560 | strerror(errno)); | |
2561 | } | |
2562 | } | |
2563 | #endif | |
06247428 | 2564 | |
b8cf1913 HR |
2565 | /* Resize and potentially preallocate the file to the desired |
2566 | * final size */ | |
93f4e2ff KW |
2567 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2568 | file_opts->preallocation, errp); | |
9f63b07e | 2569 | if (result < 0) { |
7c20c808 HR |
2570 | goto out_unlock; |
2571 | } | |
2572 | ||
2573 | out_unlock: | |
2996ffad | 2574 | raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); |
7c20c808 HR |
2575 | if (local_err) { |
2576 | /* The above call should not fail, and if it does, that does | |
2577 | * not mean the whole creation operation has failed. So | |
2578 | * report it the user for their convenience, but do not report | |
2579 | * it to the caller. */ | |
db0754df | 2580 | warn_report_err(local_err); |
1e37d059 | 2581 | } |
06247428 | 2582 | |
5a1dad9d | 2583 | out_close: |
06247428 HT |
2584 | if (qemu_close(fd) != 0 && result == 0) { |
2585 | result = -errno; | |
2586 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2587 | } | |
2588 | out: | |
1e37d059 | 2589 | return result; |
83f64091 FB |
2590 | } |
2591 | ||
b92902df ML |
2592 | static int coroutine_fn raw_co_create_opts(BlockDriver *drv, |
2593 | const char *filename, | |
2594 | QemuOpts *opts, | |
927f11e1 KW |
2595 | Error **errp) |
2596 | { | |
2597 | BlockdevCreateOptions options; | |
2598 | int64_t total_size = 0; | |
ffa244c8 KW |
2599 | int64_t extent_size_hint = 0; |
2600 | bool has_extent_size_hint = false; | |
927f11e1 KW |
2601 | bool nocow = false; |
2602 | PreallocMode prealloc; | |
2603 | char *buf = NULL; | |
2604 | Error *local_err = NULL; | |
2605 | ||
2606 | /* Skip file: protocol prefix */ | |
2607 | strstart(filename, "file:", &filename); | |
2608 | ||
2609 | /* Read out options */ | |
2610 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2611 | BDRV_SECTOR_SIZE); | |
ffa244c8 KW |
2612 | if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { |
2613 | has_extent_size_hint = true; | |
2614 | extent_size_hint = | |
2615 | qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); | |
2616 | } | |
927f11e1 KW |
2617 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); |
2618 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2619 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2620 | PREALLOC_MODE_OFF, &local_err); | |
2621 | g_free(buf); | |
2622 | if (local_err) { | |
2623 | error_propagate(errp, local_err); | |
2624 | return -EINVAL; | |
2625 | } | |
2626 | ||
2627 | options = (BlockdevCreateOptions) { | |
2628 | .driver = BLOCKDEV_DRIVER_FILE, | |
2629 | .u.file = { | |
2630 | .filename = (char *) filename, | |
2631 | .size = total_size, | |
2632 | .has_preallocation = true, | |
2633 | .preallocation = prealloc, | |
2634 | .has_nocow = true, | |
2635 | .nocow = nocow, | |
ffa244c8 KW |
2636 | .has_extent_size_hint = has_extent_size_hint, |
2637 | .extent_size_hint = extent_size_hint, | |
927f11e1 KW |
2638 | }, |
2639 | }; | |
2640 | return raw_co_create(&options, errp); | |
2641 | } | |
2642 | ||
9bffae14 DHB |
2643 | static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, |
2644 | Error **errp) | |
2645 | { | |
2646 | struct stat st; | |
2647 | int ret; | |
2648 | ||
2649 | if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { | |
2650 | error_setg_errno(errp, ENOENT, "%s is not a regular file", | |
2651 | bs->filename); | |
2652 | return -ENOENT; | |
2653 | } | |
2654 | ||
2655 | ret = unlink(bs->filename); | |
2656 | if (ret < 0) { | |
2657 | ret = -errno; | |
2658 | error_setg_errno(errp, -ret, "Error when deleting file %s", | |
2659 | bs->filename); | |
2660 | } | |
2661 | ||
2662 | return ret; | |
2663 | } | |
2664 | ||
d1f06fe6 MA |
2665 | /* |
2666 | * Find allocation range in @bs around offset @start. | |
2667 | * May change underlying file descriptor's file offset. | |
2668 | * If @start is not in a hole, store @start in @data, and the | |
2669 | * beginning of the next hole in @hole, and return 0. | |
2670 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2671 | * beginning of the next non-hole in @data, and return 0. | |
2672 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2673 | * If we can't find out, return a negative errno other than -ENXIO. | |
2674 | */ | |
2675 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2676 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2677 | { |
2678 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2679 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2680 | off_t offs; |
94282e71 | 2681 | |
d1f06fe6 MA |
2682 | /* |
2683 | * SEEK_DATA cases: | |
2684 | * D1. offs == start: start is in data | |
2685 | * D2. offs > start: start is in a hole, next data at offs | |
2686 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2687 | * or start is beyond EOF | |
2688 | * If the latter happens, the file has been truncated behind | |
2689 | * our back since we opened it. All bets are off then. | |
2690 | * Treating like a trailing hole is simplest. | |
2691 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2692 | */ | |
2693 | offs = lseek(s->fd, start, SEEK_DATA); | |
2694 | if (offs < 0) { | |
2695 | return -errno; /* D3 or D4 */ | |
2696 | } | |
a03083a0 JC |
2697 | |
2698 | if (offs < start) { | |
2699 | /* This is not a valid return by lseek(). We are safe to just return | |
2700 | * -EIO in this case, and we'll treat it like D4. */ | |
2701 | return -EIO; | |
2702 | } | |
d1f06fe6 MA |
2703 | |
2704 | if (offs > start) { | |
2705 | /* D2: in hole, next data at offs */ | |
2706 | *hole = start; | |
2707 | *data = offs; | |
2708 | return 0; | |
5500316d PB |
2709 | } |
2710 | ||
d1f06fe6 MA |
2711 | /* D1: in data, end not yet known */ |
2712 | ||
2713 | /* | |
2714 | * SEEK_HOLE cases: | |
2715 | * H1. offs == start: start is in a hole | |
2716 | * If this happens here, a hole has been dug behind our back | |
2717 | * since the previous lseek(). | |
2718 | * H2. offs > start: either start is in data, next hole at offs, | |
2719 | * or start is in trailing hole, EOF at offs | |
2720 | * Linux treats trailing holes like any other hole: offs == | |
2721 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2722 | * If that happens here, a hole has been dug behind our back | |
2723 | * since the previous lseek(). | |
2724 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2725 | * If this happens, the file has been truncated behind our | |
2726 | * back since we opened it. Treat it like a trailing hole. | |
2727 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2728 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2729 | */ | |
2730 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2731 | if (offs < 0) { | |
2732 | return -errno; /* D1 and (H3 or H4) */ | |
2733 | } | |
a03083a0 JC |
2734 | |
2735 | if (offs < start) { | |
2736 | /* This is not a valid return by lseek(). We are safe to just return | |
2737 | * -EIO in this case, and we'll treat it like H4. */ | |
2738 | return -EIO; | |
2739 | } | |
d1f06fe6 MA |
2740 | |
2741 | if (offs > start) { | |
2742 | /* | |
2743 | * D1 and H2: either in data, next hole at offs, or it was in | |
2744 | * data but is now in a trailing hole. In the latter case, | |
2745 | * all bets are off. Treating it as if it there was data all | |
2746 | * the way to EOF is safe, so simply do that. | |
2747 | */ | |
4f11aa8a | 2748 | *data = start; |
d1f06fe6 MA |
2749 | *hole = offs; |
2750 | return 0; | |
5500316d | 2751 | } |
4f11aa8a | 2752 | |
d1f06fe6 MA |
2753 | /* D1 and H1 */ |
2754 | return -EBUSY; | |
5500316d | 2755 | #else |
4f11aa8a | 2756 | return -ENOTSUP; |
5500316d | 2757 | #endif |
4f11aa8a HR |
2758 | } |
2759 | ||
2760 | /* | |
a290f085 | 2761 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2762 | * |
a290f085 | 2763 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2764 | * |
a290f085 EB |
2765 | * 'pnum' is set to the number of bytes (including and immediately following |
2766 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2767 | * allocated/unallocated state. |
2768 | * | |
869e7ee8 HR |
2769 | * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may |
2770 | * well exceed it. | |
4f11aa8a | 2771 | */ |
a290f085 EB |
2772 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2773 | bool want_zero, | |
2774 | int64_t offset, | |
2775 | int64_t bytes, int64_t *pnum, | |
2776 | int64_t *map, | |
2777 | BlockDriverState **file) | |
2778 | { | |
2779 | off_t data = 0, hole = 0; | |
d7f62751 | 2780 | int ret; |
4f11aa8a | 2781 | |
9c3db310 HR |
2782 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); |
2783 | ||
4f11aa8a HR |
2784 | ret = fd_open(bs); |
2785 | if (ret < 0) { | |
2786 | return ret; | |
2787 | } | |
2788 | ||
a290f085 EB |
2789 | if (!want_zero) { |
2790 | *pnum = bytes; | |
2791 | *map = offset; | |
2792 | *file = bs; | |
2793 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2794 | } |
4f11aa8a | 2795 | |
a290f085 | 2796 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2797 | if (ret == -ENXIO) { |
2798 | /* Trailing hole */ | |
a290f085 | 2799 | *pnum = bytes; |
d1f06fe6 MA |
2800 | ret = BDRV_BLOCK_ZERO; |
2801 | } else if (ret < 0) { | |
2802 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2803 | *pnum = bytes; |
d1f06fe6 | 2804 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2805 | } else if (data == offset) { |
2806 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2807 | * possibly including a partial sector at EOF. */ |
869e7ee8 | 2808 | *pnum = hole - offset; |
9c3db310 HR |
2809 | |
2810 | /* | |
2811 | * We are not allowed to return partial sectors, though, so | |
2812 | * round up if necessary. | |
2813 | */ | |
2814 | if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { | |
2815 | int64_t file_length = raw_getlength(bs); | |
2816 | if (file_length > 0) { | |
2817 | /* Ignore errors, this is just a safeguard */ | |
2818 | assert(hole == file_length); | |
2819 | } | |
2820 | *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); | |
2821 | } | |
2822 | ||
d1f06fe6 | 2823 | ret = BDRV_BLOCK_DATA; |
5500316d | 2824 | } else { |
a290f085 EB |
2825 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2826 | assert(hole == offset); | |
869e7ee8 | 2827 | *pnum = data - offset; |
d1f06fe6 | 2828 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2829 | } |
a290f085 | 2830 | *map = offset; |
02650acb | 2831 | *file = bs; |
a290f085 | 2832 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2833 | } |
2834 | ||
31be8a2a SH |
2835 | #if defined(__linux__) |
2836 | /* Verify that the file is not in the page cache */ | |
2837 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2838 | { | |
2839 | const size_t window_size = 128 * 1024 * 1024; | |
2840 | BDRVRawState *s = bs->opaque; | |
2841 | void *window = NULL; | |
2842 | size_t length = 0; | |
2843 | unsigned char *vec; | |
2844 | size_t page_size; | |
2845 | off_t offset; | |
2846 | off_t end; | |
2847 | ||
2848 | /* mincore(2) page status information requires 1 byte per page */ | |
2849 | page_size = sysconf(_SC_PAGESIZE); | |
2850 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2851 | ||
2852 | end = raw_getlength(bs); | |
2853 | ||
2854 | for (offset = 0; offset < end; offset += window_size) { | |
2855 | void *new_window; | |
2856 | size_t new_length; | |
2857 | size_t vec_end; | |
2858 | size_t i; | |
2859 | int ret; | |
2860 | ||
2861 | /* Unmap previous window if size has changed */ | |
2862 | new_length = MIN(end - offset, window_size); | |
2863 | if (new_length != length) { | |
2864 | munmap(window, length); | |
2865 | window = NULL; | |
2866 | length = 0; | |
2867 | } | |
2868 | ||
2869 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2870 | s->fd, offset); | |
2871 | if (new_window == MAP_FAILED) { | |
2872 | error_setg_errno(errp, errno, "mmap failed"); | |
2873 | break; | |
2874 | } | |
2875 | ||
2876 | window = new_window; | |
2877 | length = new_length; | |
2878 | ||
2879 | ret = mincore(window, length, vec); | |
2880 | if (ret < 0) { | |
2881 | error_setg_errno(errp, errno, "mincore failed"); | |
2882 | break; | |
2883 | } | |
2884 | ||
2885 | vec_end = DIV_ROUND_UP(length, page_size); | |
2886 | for (i = 0; i < vec_end; i++) { | |
2887 | if (vec[i] & 0x1) { | |
31be8a2a SH |
2888 | break; |
2889 | } | |
2890 | } | |
77ed971b MA |
2891 | if (i < vec_end) { |
2892 | error_setg(errp, "page cache still in use!"); | |
2893 | break; | |
2894 | } | |
31be8a2a SH |
2895 | } |
2896 | ||
2897 | if (window) { | |
2898 | munmap(window, length); | |
2899 | } | |
2900 | ||
2901 | g_free(vec); | |
2902 | } | |
2903 | #endif /* __linux__ */ | |
2904 | ||
dd577a26 SH |
2905 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2906 | Error **errp) | |
2907 | { | |
2908 | BDRVRawState *s = bs->opaque; | |
2909 | int ret; | |
2910 | ||
2911 | ret = fd_open(bs); | |
2912 | if (ret < 0) { | |
2913 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2914 | return; | |
2915 | } | |
2916 | ||
f357fcd8 SH |
2917 | if (!s->drop_cache) { |
2918 | return; | |
2919 | } | |
2920 | ||
dd577a26 SH |
2921 | if (s->open_flags & O_DIRECT) { |
2922 | return; /* No host kernel page cache */ | |
2923 | } | |
2924 | ||
2925 | #if defined(__linux__) | |
2926 | /* This sets the scene for the next syscall... */ | |
2927 | ret = bdrv_co_flush(bs); | |
2928 | if (ret < 0) { | |
2929 | error_setg_errno(errp, -ret, "flush failed"); | |
2930 | return; | |
2931 | } | |
2932 | ||
2933 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2934 | * process. These limitations are okay because we just fsynced the file, | |
2935 | * we don't use mmap, and the file should not be in use by other processes. | |
2936 | */ | |
2937 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2938 | if (ret != 0) { /* the return value is a positive errno */ | |
2939 | error_setg_errno(errp, ret, "fadvise failed"); | |
2940 | return; | |
2941 | } | |
31be8a2a SH |
2942 | |
2943 | if (s->check_cache_dropped) { | |
2944 | check_cache_dropped(bs, errp); | |
2945 | } | |
dd577a26 SH |
2946 | #else /* __linux__ */ |
2947 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2948 | * unsupported on other host operating systems. Cache consistency issues | |
2949 | * may occur but no error is reported here, partly because that's the | |
2950 | * historical behavior and partly because it's hard to differentiate valid | |
2951 | * configurations that should not cause errors. | |
2952 | */ | |
2953 | #endif /* !__linux__ */ | |
2954 | } | |
2955 | ||
1c450366 AN |
2956 | static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) |
2957 | { | |
2958 | if (ret) { | |
2959 | s->stats.discard_nb_failed++; | |
2960 | } else { | |
2961 | s->stats.discard_nb_ok++; | |
2962 | s->stats.discard_bytes_ok += nbytes; | |
2963 | } | |
2964 | } | |
2965 | ||
33d70fb6 | 2966 | static coroutine_fn int |
0c802287 VSO |
2967 | raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2968 | bool blkdev) | |
dce512de | 2969 | { |
dce512de | 2970 | BDRVRawState *s = bs->opaque; |
46ee0f46 | 2971 | RawPosixAIOData acb; |
1c450366 | 2972 | int ret; |
46ee0f46 KW |
2973 | |
2974 | acb = (RawPosixAIOData) { | |
2975 | .bs = bs, | |
2976 | .aio_fildes = s->fd, | |
2977 | .aio_type = QEMU_AIO_DISCARD, | |
2978 | .aio_offset = offset, | |
2979 | .aio_nbytes = bytes, | |
2980 | }; | |
dce512de | 2981 | |
46ee0f46 KW |
2982 | if (blkdev) { |
2983 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2984 | } | |
2985 | ||
1c450366 AN |
2986 | ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); |
2987 | raw_account_discard(s, bytes, ret); | |
2988 | return ret; | |
46ee0f46 KW |
2989 | } |
2990 | ||
2991 | static coroutine_fn int | |
0c802287 | 2992 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
46ee0f46 KW |
2993 | { |
2994 | return raw_do_pdiscard(bs, offset, bytes, false); | |
dce512de | 2995 | } |
0e7e1989 | 2996 | |
7154d8ae | 2997 | static int coroutine_fn |
f34b2bcf | 2998 | raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, |
7154d8ae | 2999 | BdrvRequestFlags flags, bool blkdev) |
260a82e5 PB |
3000 | { |
3001 | BDRVRawState *s = bs->opaque; | |
7154d8ae KW |
3002 | RawPosixAIOData acb; |
3003 | ThreadPoolFunc *handler; | |
3004 | ||
292d06b9 HR |
3005 | #ifdef CONFIG_FALLOCATE |
3006 | if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { | |
3007 | BdrvTrackedRequest *req; | |
292d06b9 HR |
3008 | |
3009 | /* | |
3010 | * This is a workaround for a bug in the Linux XFS driver, | |
3011 | * where writes submitted through the AIO interface will be | |
3012 | * discarded if they happen beyond a concurrently running | |
3013 | * fallocate() that increases the file length (i.e., both the | |
3014 | * write and the fallocate() happen beyond the EOF). | |
3015 | * | |
3016 | * To work around it, we extend the tracked request for this | |
3017 | * zero write until INT64_MAX (effectively infinity), and mark | |
3018 | * it as serializing. | |
3019 | * | |
3020 | * We have to enable this workaround for all filesystems and | |
3021 | * AIO modes (not just XFS with aio=native), because for | |
3022 | * remote filesystems we do not know the host configuration. | |
3023 | */ | |
3024 | ||
3025 | req = bdrv_co_get_self_request(bs); | |
3026 | assert(req); | |
3027 | assert(req->type == BDRV_TRACKED_WRITE); | |
3028 | assert(req->offset <= offset); | |
3029 | assert(req->offset + req->bytes >= offset + bytes); | |
3030 | ||
8b117001 VSO |
3031 | req->bytes = BDRV_MAX_LENGTH - req->offset; |
3032 | ||
69b55e03 | 3033 | bdrv_check_request(req->offset, req->bytes, &error_abort); |
292d06b9 | 3034 | |
8ac5aab2 | 3035 | bdrv_make_request_serialising(req, bs->bl.request_alignment); |
292d06b9 HR |
3036 | } |
3037 | #endif | |
3038 | ||
7154d8ae KW |
3039 | acb = (RawPosixAIOData) { |
3040 | .bs = bs, | |
3041 | .aio_fildes = s->fd, | |
3042 | .aio_type = QEMU_AIO_WRITE_ZEROES, | |
3043 | .aio_offset = offset, | |
3044 | .aio_nbytes = bytes, | |
3045 | }; | |
3046 | ||
3047 | if (blkdev) { | |
3048 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
3049 | } | |
738301e1 KW |
3050 | if (flags & BDRV_REQ_NO_FALLBACK) { |
3051 | acb.aio_type |= QEMU_AIO_NO_FALLBACK; | |
3052 | } | |
260a82e5 | 3053 | |
34fa110e | 3054 | if (flags & BDRV_REQ_MAY_UNMAP) { |
7154d8ae KW |
3055 | acb.aio_type |= QEMU_AIO_DISCARD; |
3056 | handler = handle_aiocb_write_zeroes_unmap; | |
3057 | } else { | |
3058 | handler = handle_aiocb_write_zeroes; | |
260a82e5 | 3059 | } |
34fa110e | 3060 | |
7154d8ae KW |
3061 | return raw_thread_pool_submit(bs, handler, &acb); |
3062 | } | |
3063 | ||
3064 | static int coroutine_fn raw_co_pwrite_zeroes( | |
3065 | BlockDriverState *bs, int64_t offset, | |
f34b2bcf | 3066 | int64_t bytes, BdrvRequestFlags flags) |
7154d8ae KW |
3067 | { |
3068 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); | |
260a82e5 PB |
3069 | } |
3070 | ||
3071 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
3072 | { | |
260a82e5 PB |
3073 | return 0; |
3074 | } | |
3075 | ||
d9245599 AN |
3076 | static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) |
3077 | { | |
3078 | BDRVRawState *s = bs->opaque; | |
3079 | return (BlockStatsSpecificFile) { | |
3080 | .discard_nb_ok = s->stats.discard_nb_ok, | |
3081 | .discard_nb_failed = s->stats.discard_nb_failed, | |
3082 | .discard_bytes_ok = s->stats.discard_bytes_ok, | |
3083 | }; | |
3084 | } | |
3085 | ||
3086 | static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) | |
3087 | { | |
3088 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3089 | ||
3090 | stats->driver = BLOCKDEV_DRIVER_FILE; | |
3091 | stats->u.file = get_blockstats_specific_file(bs); | |
3092 | ||
3093 | return stats; | |
3094 | } | |
3095 | ||
14176c8d | 3096 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
d9245599 AN |
3097 | static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) |
3098 | { | |
3099 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3100 | ||
3101 | stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; | |
3102 | stats->u.host_device = get_blockstats_specific_file(bs); | |
3103 | ||
3104 | return stats; | |
3105 | } | |
14176c8d | 3106 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
d9245599 | 3107 | |
6f482f74 CL |
3108 | static QemuOptsList raw_create_opts = { |
3109 | .name = "raw-create-opts", | |
3110 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
3111 | .desc = { | |
3112 | { | |
3113 | .name = BLOCK_OPT_SIZE, | |
3114 | .type = QEMU_OPT_SIZE, | |
3115 | .help = "Virtual disk size" | |
3116 | }, | |
4ab15590 CL |
3117 | { |
3118 | .name = BLOCK_OPT_NOCOW, | |
3119 | .type = QEMU_OPT_BOOL, | |
3120 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
3121 | }, | |
06247428 HT |
3122 | { |
3123 | .name = BLOCK_OPT_PREALLOC, | |
3124 | .type = QEMU_OPT_STRING, | |
abea0053 SG |
3125 | .help = "Preallocation mode (allowed values: off" |
3126 | #ifdef CONFIG_POSIX_FALLOCATE | |
3127 | ", falloc" | |
3128 | #endif | |
3129 | ", full)" | |
06247428 | 3130 | }, |
ffa244c8 KW |
3131 | { |
3132 | .name = BLOCK_OPT_EXTENT_SIZE_HINT, | |
3133 | .type = QEMU_OPT_SIZE, | |
3134 | .help = "Extent size hint for the image file, 0 to disable" | |
3135 | }, | |
6f482f74 CL |
3136 | { /* end of list */ } |
3137 | } | |
0e7e1989 KW |
3138 | }; |
3139 | ||
244a5668 FZ |
3140 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
3141 | Error **errp) | |
3142 | { | |
6ceabe6f | 3143 | BDRVRawState *s = bs->opaque; |
72373e40 | 3144 | int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags; |
6ceabe6f KW |
3145 | int open_flags; |
3146 | int ret; | |
3147 | ||
72373e40 VSO |
3148 | /* We may need a new fd if auto-read-only switches the mode */ |
3149 | ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, | |
3150 | false, errp); | |
3151 | if (ret < 0) { | |
3152 | return ret; | |
3153 | } else if (ret != s->fd) { | |
3154 | Error *local_err = NULL; | |
3155 | ||
6ceabe6f | 3156 | /* |
72373e40 VSO |
3157 | * Fail already check_perm() if we can't get a working O_DIRECT |
3158 | * alignment with the new fd. | |
6ceabe6f | 3159 | */ |
72373e40 VSO |
3160 | raw_probe_alignment(bs, ret, &local_err); |
3161 | if (local_err) { | |
3162 | error_propagate(errp, local_err); | |
3163 | return -EINVAL; | |
6ceabe6f | 3164 | } |
72373e40 VSO |
3165 | |
3166 | s->perm_change_fd = ret; | |
3167 | s->perm_change_flags = open_flags; | |
6ceabe6f KW |
3168 | } |
3169 | ||
3170 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
3171 | * but keep everything locked that new will need. */ | |
3172 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
3173 | if (ret < 0) { | |
3174 | goto fail; | |
3175 | } | |
3176 | ||
3177 | /* Copy locks to the new fd */ | |
eb43ea16 | 3178 | if (s->perm_change_fd && s->use_lock) { |
6ceabe6f KW |
3179 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, |
3180 | false, errp); | |
3181 | if (ret < 0) { | |
3182 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
3183 | goto fail; | |
3184 | } | |
3185 | } | |
3186 | return 0; | |
3187 | ||
3188 | fail: | |
72373e40 | 3189 | if (s->perm_change_fd) { |
6ceabe6f KW |
3190 | qemu_close(s->perm_change_fd); |
3191 | } | |
3192 | s->perm_change_fd = 0; | |
3193 | return ret; | |
244a5668 FZ |
3194 | } |
3195 | ||
3196 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
3197 | { | |
3198 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
3199 | |
3200 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
3201 | * called after .bdrv_reopen_commit) */ | |
3202 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
3203 | qemu_close(s->fd); | |
3204 | s->fd = s->perm_change_fd; | |
094e3639 | 3205 | s->open_flags = s->perm_change_flags; |
6ceabe6f KW |
3206 | } |
3207 | s->perm_change_fd = 0; | |
3208 | ||
244a5668 FZ |
3209 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
3210 | s->perm = perm; | |
3211 | s->shared_perm = shared; | |
3212 | } | |
3213 | ||
3214 | static void raw_abort_perm_update(BlockDriverState *bs) | |
3215 | { | |
6ceabe6f KW |
3216 | BDRVRawState *s = bs->opaque; |
3217 | ||
3218 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
3219 | * the file descriptor. */ | |
72373e40 | 3220 | if (s->perm_change_fd) { |
6ceabe6f KW |
3221 | qemu_close(s->perm_change_fd); |
3222 | } | |
3223 | s->perm_change_fd = 0; | |
3224 | ||
244a5668 FZ |
3225 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
3226 | } | |
3227 | ||
67b51fb9 | 3228 | static int coroutine_fn raw_co_copy_range_from( |
48535049 VSO |
3229 | BlockDriverState *bs, BdrvChild *src, int64_t src_offset, |
3230 | BdrvChild *dst, int64_t dst_offset, int64_t bytes, | |
67b51fb9 | 3231 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) |
1efad060 | 3232 | { |
67b51fb9 VSO |
3233 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3234 | read_flags, write_flags); | |
1efad060 FZ |
3235 | } |
3236 | ||
3237 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 | 3238 | BdrvChild *src, |
48535049 | 3239 | int64_t src_offset, |
67b51fb9 | 3240 | BdrvChild *dst, |
48535049 VSO |
3241 | int64_t dst_offset, |
3242 | int64_t bytes, | |
67b51fb9 VSO |
3243 | BdrvRequestFlags read_flags, |
3244 | BdrvRequestFlags write_flags) | |
1efad060 | 3245 | { |
58a209c4 | 3246 | RawPosixAIOData acb; |
1efad060 FZ |
3247 | BDRVRawState *s = bs->opaque; |
3248 | BDRVRawState *src_s; | |
3249 | ||
3250 | assert(dst->bs == bs); | |
3251 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
3252 | return -ENOTSUP; | |
3253 | } | |
3254 | ||
3255 | src_s = src->bs->opaque; | |
9f850f67 | 3256 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
3257 | return -EIO; |
3258 | } | |
58a209c4 KW |
3259 | |
3260 | acb = (RawPosixAIOData) { | |
3261 | .bs = bs, | |
3262 | .aio_type = QEMU_AIO_COPY_RANGE, | |
3263 | .aio_fildes = src_s->fd, | |
3264 | .aio_offset = src_offset, | |
3265 | .aio_nbytes = bytes, | |
3266 | .copy_range = { | |
3267 | .aio_fd2 = s->fd, | |
3268 | .aio_offset2 = dst_offset, | |
3269 | }, | |
3270 | }; | |
3271 | ||
3272 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
3273 | } |
3274 | ||
5f535a94 | 3275 | BlockDriver bdrv_file = { |
84a12e66 CH |
3276 | .format_name = "file", |
3277 | .protocol_name = "file", | |
856ae5c3 | 3278 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3279 | .bdrv_needs_filename = true, |
856ae5c3 | 3280 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 3281 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 3282 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
3283 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3284 | .bdrv_reopen_commit = raw_reopen_commit, | |
3285 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 3286 | .bdrv_close = raw_close, |
927f11e1 | 3287 | .bdrv_co_create = raw_co_create, |
efc75e2a | 3288 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 3289 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 3290 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 3291 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3292 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
9bffae14 | 3293 | .bdrv_co_delete_file = raw_co_delete_file, |
3b46e624 | 3294 | |
9d52aa3c KW |
3295 | .bdrv_co_preadv = raw_co_preadv, |
3296 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3297 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3298 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
3299 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3300 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3301 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3302 | .bdrv_io_plug = raw_aio_plug, |
3303 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 3304 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3305 | |
061ca8a3 | 3306 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 3307 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3308 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3309 | .bdrv_get_allocated_file_size |
3310 | = raw_get_allocated_file_size, | |
d9245599 | 3311 | .bdrv_get_specific_stats = raw_get_specific_stats, |
244a5668 FZ |
3312 | .bdrv_check_perm = raw_check_perm, |
3313 | .bdrv_set_perm = raw_set_perm, | |
3314 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 3315 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3316 | .mutable_opts = mutable_opts, |
83f64091 FB |
3317 | }; |
3318 | ||
19cb3738 FB |
3319 | /***********************************************/ |
3320 | /* host device */ | |
3321 | ||
14176c8d JD |
3322 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
3323 | ||
83affaa6 | 3324 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
3325 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3326 | CFIndex maxPathSize, int flags); | |
d0855f12 | 3327 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 3328 | { |
d0855f12 | 3329 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
3330 | mach_port_t masterPort; |
3331 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
3332 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
3333 | char *mediaType = NULL; | |
19cb3738 FB |
3334 | |
3335 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
3336 | if ( KERN_SUCCESS != kernResult ) { | |
3337 | printf( "IOMasterPort returned %d\n", kernResult ); | |
3338 | } | |
3b46e624 | 3339 | |
d0855f12 JA |
3340 | int index; |
3341 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
3342 | classesToMatch = IOServiceMatching(matching_array[index]); | |
3343 | if (classesToMatch == NULL) { | |
3344 | error_report("IOServiceMatching returned NULL for %s", | |
3345 | matching_array[index]); | |
3346 | continue; | |
3347 | } | |
3348 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
3349 | kCFBooleanTrue); | |
3350 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
3351 | mediaIterator); | |
3352 | if (kernResult != KERN_SUCCESS) { | |
3353 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
3354 | kernResult); | |
3355 | continue; | |
3356 | } | |
3b46e624 | 3357 | |
d0855f12 JA |
3358 | /* If a match was found, leave the loop */ |
3359 | if (*mediaIterator != 0) { | |
4f7d28d7 | 3360 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
3361 | mediaType = g_strdup(matching_array[index]); |
3362 | break; | |
3363 | } | |
3364 | } | |
3365 | return mediaType; | |
19cb3738 FB |
3366 | } |
3367 | ||
98caa5bc JA |
3368 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3369 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
3370 | { |
3371 | io_object_t nextMedia; | |
3372 | kern_return_t kernResult = KERN_FAILURE; | |
3373 | *bsdPath = '\0'; | |
3374 | nextMedia = IOIteratorNext( mediaIterator ); | |
3375 | if ( nextMedia ) | |
3376 | { | |
3377 | CFTypeRef bsdPathAsCFString; | |
3378 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3379 | if ( bsdPathAsCFString ) { | |
3380 | size_t devPathLength; | |
3381 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3382 | if (flags & BDRV_O_NOCACHE) { |
3383 | strcat(bsdPath, "r"); | |
3384 | } | |
19cb3738 FB |
3385 | devPathLength = strlen( bsdPath ); |
3386 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3387 | kernResult = KERN_SUCCESS; | |
3388 | } | |
3389 | CFRelease( bsdPathAsCFString ); | |
3390 | } | |
3391 | IOObjectRelease( nextMedia ); | |
3392 | } | |
3b46e624 | 3393 | |
19cb3738 FB |
3394 | return kernResult; |
3395 | } | |
3396 | ||
d0855f12 JA |
3397 | /* Sets up a real cdrom for use in QEMU */ |
3398 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3399 | { | |
3400 | int index, num_of_test_partitions = 2, fd; | |
3401 | char test_partition[MAXPATHLEN]; | |
3402 | bool partition_found = false; | |
3403 | ||
3404 | /* look for a working partition */ | |
3405 | for (index = 0; index < num_of_test_partitions; index++) { | |
3406 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3407 | index); | |
b18a24a9 | 3408 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); |
d0855f12 JA |
3409 | if (fd >= 0) { |
3410 | partition_found = true; | |
3411 | qemu_close(fd); | |
3412 | break; | |
3413 | } | |
3414 | } | |
3415 | ||
3416 | /* if a working partition on the device was not found */ | |
3417 | if (partition_found == false) { | |
3418 | error_setg(errp, "Failed to find a working partition on disc"); | |
3419 | } else { | |
4f7d28d7 | 3420 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3421 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3422 | } | |
3423 | return partition_found; | |
3424 | } | |
3425 | ||
3426 | /* Prints directions on mounting and unmounting a device */ | |
3427 | static void print_unmounting_directions(const char *file_name) | |
3428 | { | |
3429 | error_report("If device %s is mounted on the desktop, unmount" | |
3430 | " it first before using it in QEMU", file_name); | |
3431 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3432 | file_name); | |
3433 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3434 | } | |
3435 | ||
3436 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3437 | |
508c7cb3 CH |
3438 | static int hdev_probe_device(const char *filename) |
3439 | { | |
3440 | struct stat st; | |
3441 | ||
3442 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3443 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3444 | return 50; | |
3445 | ||
3446 | if (stat(filename, &st) >= 0 && | |
3447 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3448 | return 100; | |
3449 | } | |
3450 | ||
3451 | return 0; | |
3452 | } | |
3453 | ||
7af803d4 HR |
3454 | static void hdev_parse_filename(const char *filename, QDict *options, |
3455 | Error **errp) | |
3456 | { | |
03c320d8 | 3457 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3458 | } |
3459 | ||
3307ed7b DA |
3460 | static bool hdev_is_sg(BlockDriverState *bs) |
3461 | { | |
3462 | ||
3463 | #if defined(__linux__) | |
3464 | ||
0d4377b3 | 3465 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3466 | struct stat st; |
3467 | struct sg_scsi_id scsiid; | |
3468 | int sg_version; | |
0d4377b3 KW |
3469 | int ret; |
3470 | ||
3471 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3472 | return false; | |
3473 | } | |
3307ed7b | 3474 | |
0d4377b3 KW |
3475 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3476 | if (ret < 0) { | |
3477 | return false; | |
3478 | } | |
3479 | ||
3480 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3481 | if (ret >= 0) { | |
4f7d28d7 | 3482 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3483 | return true; |
3484 | } | |
3485 | ||
3486 | #endif | |
3487 | ||
3488 | return false; | |
3489 | } | |
3490 | ||
015a1036 HR |
3491 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3492 | Error **errp) | |
19cb3738 FB |
3493 | { |
3494 | BDRVRawState *s = bs->opaque; | |
da888d37 | 3495 | int ret; |
a76bab49 | 3496 | |
83affaa6 | 3497 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3498 | /* |
3499 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3500 | * would require more care. When @options come from -blockdev or | |
3501 | * blockdev_add, its members are typed according to the QAPI | |
3502 | * schema, but when they come from -drive, they're all QString. | |
3503 | */ | |
3307ed7b | 3504 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3505 | char bsd_path[MAXPATHLEN] = ""; |
3506 | bool error_occurred = false; | |
3507 | ||
3508 | /* If using a real cdrom */ | |
3509 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3510 | char *mediaType = NULL; | |
3511 | kern_return_t ret_val; | |
3512 | io_iterator_t mediaIterator = 0; | |
3513 | ||
3514 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3515 | if (mediaType == NULL) { | |
3516 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3517 | " drive"); | |
3518 | error_occurred = true; | |
3519 | goto hdev_open_Mac_error; | |
3520 | } | |
3307ed7b | 3521 | |
d0855f12 JA |
3522 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3523 | if (ret_val != KERN_SUCCESS) { | |
3524 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3525 | error_occurred = true; | |
3526 | goto hdev_open_Mac_error; | |
3527 | } | |
3528 | ||
3529 | /* If a real optical drive was not found */ | |
3530 | if (bsd_path[0] == '\0') { | |
3531 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3532 | error_occurred = true; | |
3533 | goto hdev_open_Mac_error; | |
3534 | } | |
3535 | ||
3536 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3537 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3538 | setup_cdrom(bsd_path, errp) == false) { | |
3539 | print_unmounting_directions(bsd_path); | |
3540 | error_occurred = true; | |
3541 | goto hdev_open_Mac_error; | |
19cb3738 | 3542 | } |
3b46e624 | 3543 | |
46f5ac20 | 3544 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3545 | |
3546 | hdev_open_Mac_error: | |
3547 | g_free(mediaType); | |
3548 | if (mediaIterator) { | |
3549 | IOObjectRelease(mediaIterator); | |
3550 | } | |
3551 | if (error_occurred) { | |
3552 | return -ENOENT; | |
3553 | } | |
19cb3738 | 3554 | } |
d0855f12 | 3555 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3556 | |
3557 | s->type = FTYPE_FILE; | |
90babde0 | 3558 | |
668f62ec | 3559 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
da888d37 | 3560 | if (ret < 0) { |
d0855f12 JA |
3561 | #if defined(__APPLE__) && defined(__MACH__) |
3562 | if (*bsd_path) { | |
3563 | filename = bsd_path; | |
3564 | } | |
3565 | /* if a physical device experienced an error while being opened */ | |
3566 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3567 | print_unmounting_directions(filename); | |
3568 | } | |
3569 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3570 | return ret; |
3571 | } | |
3572 | ||
3307ed7b DA |
3573 | /* Since this does ioctl the device must be already opened */ |
3574 | bs->sg = hdev_is_sg(bs); | |
3575 | ||
da888d37 | 3576 | return ret; |
19cb3738 FB |
3577 | } |
3578 | ||
03ff3ca3 | 3579 | #if defined(__linux__) |
2f3a7ab3 KW |
3580 | static int coroutine_fn |
3581 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3582 | { |
f141eafe | 3583 | BDRVRawState *s = bs->opaque; |
03425671 | 3584 | RawPosixAIOData acb; |
2f3a7ab3 | 3585 | int ret; |
221f715d | 3586 | |
2f3a7ab3 KW |
3587 | ret = fd_open(bs); |
3588 | if (ret < 0) { | |
3589 | return ret; | |
3590 | } | |
c208e8c2 | 3591 | |
7c9e5276 PB |
3592 | if (req == SG_IO && s->pr_mgr) { |
3593 | struct sg_io_hdr *io_hdr = buf; | |
3594 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3595 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3596 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3597 | s->fd, io_hdr); |
7c9e5276 PB |
3598 | } |
3599 | } | |
3600 | ||
03425671 KW |
3601 | acb = (RawPosixAIOData) { |
3602 | .bs = bs, | |
3603 | .aio_type = QEMU_AIO_IOCTL, | |
3604 | .aio_fildes = s->fd, | |
3605 | .aio_offset = 0, | |
3606 | .ioctl = { | |
3607 | .buf = buf, | |
3608 | .cmd = req, | |
3609 | }, | |
3610 | }; | |
3611 | ||
3612 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3613 | } |
f709623b | 3614 | #endif /* linux */ |
221f715d | 3615 | |
33d70fb6 | 3616 | static coroutine_fn int |
0c802287 | 3617 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
c36dd8a0 | 3618 | { |
1c450366 | 3619 | BDRVRawState *s = bs->opaque; |
33d70fb6 | 3620 | int ret; |
c36dd8a0 | 3621 | |
33d70fb6 KW |
3622 | ret = fd_open(bs); |
3623 | if (ret < 0) { | |
1c450366 | 3624 | raw_account_discard(s, bytes, ret); |
33d70fb6 | 3625 | return ret; |
c36dd8a0 | 3626 | } |
46ee0f46 | 3627 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3628 | } |
3629 | ||
2ffa76c2 | 3630 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f34b2bcf | 3631 | int64_t offset, int64_t bytes, BdrvRequestFlags flags) |
d0b4503e | 3632 | { |
d0b4503e PB |
3633 | int rc; |
3634 | ||
3635 | rc = fd_open(bs); | |
3636 | if (rc < 0) { | |
3637 | return rc; | |
3638 | } | |
34fa110e | 3639 | |
7154d8ae | 3640 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3641 | } |
3642 | ||
5efa9d5a | 3643 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3644 | .format_name = "host_device", |
84a12e66 | 3645 | .protocol_name = "host_device", |
0b4ce02e | 3646 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3647 | .bdrv_needs_filename = true, |
0b4ce02e | 3648 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3649 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3650 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3651 | .bdrv_close = raw_close, |
1bc6b705 JC |
3652 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3653 | .bdrv_reopen_commit = raw_reopen_commit, | |
3654 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3655 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3656 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3657 | .mutable_opts = mutable_opts, |
dd577a26 | 3658 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3659 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3660 | |
9d52aa3c KW |
3661 | .bdrv_co_preadv = raw_co_preadv, |
3662 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3663 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3664 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3665 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3666 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3667 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3668 | .bdrv_io_plug = raw_aio_plug, |
3669 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3670 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3671 | |
061ca8a3 | 3672 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3673 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3674 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3675 | .bdrv_get_allocated_file_size |
3676 | = raw_get_allocated_file_size, | |
d9245599 | 3677 | .bdrv_get_specific_stats = hdev_get_specific_stats, |
244a5668 FZ |
3678 | .bdrv_check_perm = raw_check_perm, |
3679 | .bdrv_set_perm = raw_set_perm, | |
3680 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3681 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3682 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3683 | |
f3a5d3f8 | 3684 | /* generic scsi device */ |
63ec93db | 3685 | #ifdef __linux__ |
2f3a7ab3 | 3686 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3687 | #endif |
f3a5d3f8 CH |
3688 | }; |
3689 | ||
18fa1c42 HR |
3690 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3691 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3692 | Error **errp) | |
3693 | { | |
03c320d8 | 3694 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3695 | } |
3696 | #endif | |
3697 | ||
3698 | #ifdef __linux__ | |
015a1036 HR |
3699 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3700 | Error **errp) | |
f3a5d3f8 CH |
3701 | { |
3702 | BDRVRawState *s = bs->opaque; | |
3703 | ||
f3a5d3f8 CH |
3704 | s->type = FTYPE_CD; |
3705 | ||
19a3da7f | 3706 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3707 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3708 | } |
3709 | ||
508c7cb3 CH |
3710 | static int cdrom_probe_device(const char *filename) |
3711 | { | |
3baf720e CR |
3712 | int fd, ret; |
3713 | int prio = 0; | |
343f8568 | 3714 | struct stat st; |
3baf720e | 3715 | |
b18a24a9 | 3716 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); |
3baf720e CR |
3717 | if (fd < 0) { |
3718 | goto out; | |
3719 | } | |
343f8568 JS |
3720 | ret = fstat(fd, &st); |
3721 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3722 | goto outc; | |
3723 | } | |
3baf720e CR |
3724 | |
3725 | /* Attempt to detect via a CDROM specific ioctl */ | |
3726 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3727 | if (ret >= 0) | |
3728 | prio = 100; | |
3729 | ||
343f8568 | 3730 | outc: |
2e1e79da | 3731 | qemu_close(fd); |
3baf720e CR |
3732 | out: |
3733 | return prio; | |
508c7cb3 CH |
3734 | } |
3735 | ||
e031f750 | 3736 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3737 | { |
3738 | BDRVRawState *s = bs->opaque; | |
3739 | int ret; | |
3740 | ||
3741 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3742 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3743 | } |
3744 | ||
f36f3949 | 3745 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3746 | { |
3747 | BDRVRawState *s = bs->opaque; | |
3748 | ||
3749 | if (eject_flag) { | |
3750 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3751 | perror("CDROMEJECT"); | |
3752 | } else { | |
3753 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3754 | perror("CDROMEJECT"); | |
3755 | } | |
f3a5d3f8 CH |
3756 | } |
3757 | ||
025e849a | 3758 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3759 | { |
3760 | BDRVRawState *s = bs->opaque; | |
3761 | ||
3762 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3763 | /* | |
3764 | * Note: an error can happen if the distribution automatically | |
3765 | * mounts the CD-ROM | |
3766 | */ | |
3767 | /* perror("CDROM_LOCKDOOR"); */ | |
3768 | } | |
f3a5d3f8 CH |
3769 | } |
3770 | ||
3771 | static BlockDriver bdrv_host_cdrom = { | |
3772 | .format_name = "host_cdrom", | |
84a12e66 | 3773 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3774 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3775 | .bdrv_needs_filename = true, |
508c7cb3 | 3776 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3777 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3778 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3779 | .bdrv_close = raw_close, |
1bc6b705 JC |
3780 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3781 | .bdrv_reopen_commit = raw_reopen_commit, | |
3782 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3783 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3784 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3785 | .mutable_opts = mutable_opts, |
dd577a26 | 3786 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3787 | |
9d52aa3c KW |
3788 | .bdrv_co_preadv = raw_co_preadv, |
3789 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3790 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3791 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3792 | .bdrv_io_plug = raw_aio_plug, |
3793 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3794 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3795 | |
061ca8a3 | 3796 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3797 | .bdrv_getlength = raw_getlength, |
3798 | .has_variable_length = true, | |
4a1d5e1f FZ |
3799 | .bdrv_get_allocated_file_size |
3800 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3801 | |
3802 | /* removable device support */ | |
3803 | .bdrv_is_inserted = cdrom_is_inserted, | |
3804 | .bdrv_eject = cdrom_eject, | |
025e849a | 3805 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3806 | |
3807 | /* generic scsi device */ | |
2f3a7ab3 | 3808 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3809 | }; |
3810 | #endif /* __linux__ */ | |
3811 | ||
a167ba50 | 3812 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3813 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3814 | Error **errp) | |
f3a5d3f8 CH |
3815 | { |
3816 | BDRVRawState *s = bs->opaque; | |
3817 | int ret; | |
3818 | ||
3819 | s->type = FTYPE_CD; | |
3820 | ||
668f62ec | 3821 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
e428e439 | 3822 | if (ret) { |
f3a5d3f8 | 3823 | return ret; |
e428e439 | 3824 | } |
f3a5d3f8 | 3825 | |
9b2260cb | 3826 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3827 | ioctl(s->fd, CDIOCALLOW); |
3828 | return 0; | |
3829 | } | |
3830 | ||
508c7cb3 CH |
3831 | static int cdrom_probe_device(const char *filename) |
3832 | { | |
3833 | if (strstart(filename, "/dev/cd", NULL) || | |
3834 | strstart(filename, "/dev/acd", NULL)) | |
3835 | return 100; | |
3836 | return 0; | |
3837 | } | |
3838 | ||
f3a5d3f8 CH |
3839 | static int cdrom_reopen(BlockDriverState *bs) |
3840 | { | |
3841 | BDRVRawState *s = bs->opaque; | |
3842 | int fd; | |
3843 | ||
3844 | /* | |
3845 | * Force reread of possibly changed/newly loaded disc, | |
3846 | * FreeBSD seems to not notice sometimes... | |
3847 | */ | |
3848 | if (s->fd >= 0) | |
2e1e79da | 3849 | qemu_close(s->fd); |
b18a24a9 | 3850 | fd = qemu_open(bs->filename, s->open_flags, NULL); |
f3a5d3f8 CH |
3851 | if (fd < 0) { |
3852 | s->fd = -1; | |
3853 | return -EIO; | |
3854 | } | |
3855 | s->fd = fd; | |
3856 | ||
9b2260cb | 3857 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3858 | ioctl(s->fd, CDIOCALLOW); |
3859 | return 0; | |
3860 | } | |
3861 | ||
e031f750 | 3862 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3863 | { |
3864 | return raw_getlength(bs) > 0; | |
3865 | } | |
3866 | ||
f36f3949 | 3867 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3868 | { |
3869 | BDRVRawState *s = bs->opaque; | |
3870 | ||
3871 | if (s->fd < 0) | |
822e1cd1 | 3872 | return; |
f3a5d3f8 CH |
3873 | |
3874 | (void) ioctl(s->fd, CDIOCALLOW); | |
3875 | ||
3876 | if (eject_flag) { | |
3877 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3878 | perror("CDIOCEJECT"); | |
3879 | } else { | |
3880 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3881 | perror("CDIOCCLOSE"); | |
3882 | } | |
3883 | ||
822e1cd1 | 3884 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3885 | } |
3886 | ||
025e849a | 3887 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3888 | { |
3889 | BDRVRawState *s = bs->opaque; | |
3890 | ||
3891 | if (s->fd < 0) | |
7bf37fed | 3892 | return; |
f3a5d3f8 CH |
3893 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3894 | /* | |
3895 | * Note: an error can happen if the distribution automatically | |
3896 | * mounts the CD-ROM | |
3897 | */ | |
3898 | /* perror("CDROM_LOCKDOOR"); */ | |
3899 | } | |
f3a5d3f8 CH |
3900 | } |
3901 | ||
3902 | static BlockDriver bdrv_host_cdrom = { | |
3903 | .format_name = "host_cdrom", | |
84a12e66 | 3904 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3905 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3906 | .bdrv_needs_filename = true, |
508c7cb3 | 3907 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3908 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3909 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3910 | .bdrv_close = raw_close, |
1bc6b705 JC |
3911 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3912 | .bdrv_reopen_commit = raw_reopen_commit, | |
3913 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3914 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3915 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3916 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3917 | |
9d52aa3c KW |
3918 | .bdrv_co_preadv = raw_co_preadv, |
3919 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3920 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3921 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3922 | .bdrv_io_plug = raw_aio_plug, |
3923 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3924 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3925 | |
061ca8a3 | 3926 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3927 | .bdrv_getlength = raw_getlength, |
3928 | .has_variable_length = true, | |
4a1d5e1f FZ |
3929 | .bdrv_get_allocated_file_size |
3930 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3931 | |
19cb3738 | 3932 | /* removable device support */ |
f3a5d3f8 CH |
3933 | .bdrv_is_inserted = cdrom_is_inserted, |
3934 | .bdrv_eject = cdrom_eject, | |
025e849a | 3935 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3936 | }; |
f3a5d3f8 | 3937 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3938 | |
14176c8d JD |
3939 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
3940 | ||
84a12e66 | 3941 | static void bdrv_file_init(void) |
5efa9d5a | 3942 | { |
508c7cb3 CH |
3943 | /* |
3944 | * Register all the drivers. Note that order is important, the driver | |
3945 | * registered last will get probed first. | |
3946 | */ | |
84a12e66 | 3947 | bdrv_register(&bdrv_file); |
14176c8d | 3948 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
5efa9d5a | 3949 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3950 | #ifdef __linux__ |
f3a5d3f8 CH |
3951 | bdrv_register(&bdrv_host_cdrom); |
3952 | #endif | |
a167ba50 | 3953 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3954 | bdrv_register(&bdrv_host_cdrom); |
3955 | #endif | |
14176c8d | 3956 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
5efa9d5a AL |
3957 | } |
3958 | ||
84a12e66 | 3959 | block_init(bdrv_file_init); |