]>
Commit | Line | Data |
---|---|---|
83f64091 | 1 | /* |
223d4670 | 2 | * Block driver for RAW files (posix) |
5fafdf24 | 3 | * |
83f64091 | 4 | * Copyright (c) 2006 Fabrice Bellard |
5fafdf24 | 5 | * |
83f64091 FB |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
922a01a0 | 24 | |
80c71a24 | 25 | #include "qemu/osdep.h" |
da34e65c | 26 | #include "qapi/error.h" |
f348b6d1 | 27 | #include "qemu/cutils.h" |
d49b6836 | 28 | #include "qemu/error-report.h" |
737e150e | 29 | #include "block/block_int.h" |
1de7afc9 | 30 | #include "qemu/module.h" |
922a01a0 | 31 | #include "qemu/option.h" |
ffa244c8 | 32 | #include "qemu/units.h" |
5df022cf | 33 | #include "qemu/memalign.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 | ||
f6465578 AL |
109 | /* OS X does not have O_DSYNC */ |
110 | #ifndef O_DSYNC | |
1c27a8b3 | 111 | #ifdef O_SYNC |
7ab064d2 | 112 | #define O_DSYNC O_SYNC |
1c27a8b3 JA |
113 | #elif defined(O_FSYNC) |
114 | #define O_DSYNC O_FSYNC | |
115 | #endif | |
f6465578 AL |
116 | #endif |
117 | ||
9f7965c7 AL |
118 | /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ |
119 | #ifndef O_DIRECT | |
120 | #define O_DIRECT O_DSYNC | |
121 | #endif | |
122 | ||
19cb3738 FB |
123 | #define FTYPE_FILE 0 |
124 | #define FTYPE_CD 1 | |
83f64091 | 125 | |
581b9e29 CH |
126 | #define MAX_BLOCKSIZE 4096 |
127 | ||
244a5668 FZ |
128 | /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, |
129 | * leaving a few more bytes for its future use. */ | |
130 | #define RAW_LOCK_PERM_BASE 100 | |
131 | #define RAW_LOCK_SHARED_BASE 200 | |
132 | ||
19cb3738 FB |
133 | typedef struct BDRVRawState { |
134 | int fd; | |
244a5668 | 135 | bool use_lock; |
19cb3738 | 136 | int type; |
0e1d8f4c | 137 | int open_flags; |
c25f53b0 PB |
138 | size_t buf_align; |
139 | ||
244a5668 FZ |
140 | /* The current permissions. */ |
141 | uint64_t perm; | |
142 | uint64_t shared_perm; | |
143 | ||
2996ffad | 144 | /* The perms bits whose corresponding bytes are already locked in |
f2e3af29 | 145 | * s->fd. */ |
2996ffad FZ |
146 | uint64_t locked_perm; |
147 | uint64_t locked_shared_perm; | |
148 | ||
684960d4 SG |
149 | uint64_t aio_max_batch; |
150 | ||
6ceabe6f | 151 | int perm_change_fd; |
094e3639 | 152 | int perm_change_flags; |
e0c9cf3a KW |
153 | BDRVReopenState *reopen_state; |
154 | ||
260a82e5 | 155 | bool has_discard:1; |
97a2ae34 | 156 | bool has_write_zeroes:1; |
0a4279d9 | 157 | bool use_linux_aio:1; |
c6447510 | 158 | bool use_linux_io_uring:1; |
c7ddc882 | 159 | int page_cache_inconsistent; /* errno from fdatasync failure */ |
d50d8222 | 160 | bool has_fallocate; |
3cad8307 | 161 | bool needs_alignment; |
5dbd0ce1 | 162 | bool force_alignment; |
f357fcd8 | 163 | bool drop_cache; |
31be8a2a | 164 | bool check_cache_dropped; |
1c450366 AN |
165 | struct { |
166 | uint64_t discard_nb_ok; | |
167 | uint64_t discard_nb_failed; | |
168 | uint64_t discard_bytes_ok; | |
169 | } stats; | |
7c9e5276 PB |
170 | |
171 | PRManager *pr_mgr; | |
19cb3738 FB |
172 | } BDRVRawState; |
173 | ||
eeb6b45d | 174 | typedef struct BDRVRawReopenState { |
eeb6b45d | 175 | int open_flags; |
f357fcd8 | 176 | bool drop_cache; |
31be8a2a | 177 | bool check_cache_dropped; |
eeb6b45d JC |
178 | } BDRVRawReopenState; |
179 | ||
14176c8d JD |
180 | static int fd_open(BlockDriverState *bs) |
181 | { | |
182 | BDRVRawState *s = bs->opaque; | |
183 | ||
184 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
185 | if (s->fd >= 0) { | |
186 | return 0; | |
187 | } | |
188 | return -EIO; | |
189 | } | |
190 | ||
22afa7b5 | 191 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 192 | |
de81a169 PB |
193 | typedef struct RawPosixAIOData { |
194 | BlockDriverState *bs; | |
d57c44d0 | 195 | int aio_type; |
de81a169 | 196 | int aio_fildes; |
d57c44d0 | 197 | |
de81a169 | 198 | off_t aio_offset; |
d57c44d0 KW |
199 | uint64_t aio_nbytes; |
200 | ||
93f4e2ff | 201 | union { |
d57c44d0 KW |
202 | struct { |
203 | struct iovec *iov; | |
204 | int niov; | |
205 | } io; | |
206 | struct { | |
207 | uint64_t cmd; | |
208 | void *buf; | |
209 | } ioctl; | |
93f4e2ff KW |
210 | struct { |
211 | int aio_fd2; | |
212 | off_t aio_offset2; | |
d57c44d0 | 213 | } copy_range; |
93f4e2ff KW |
214 | struct { |
215 | PreallocMode prealloc; | |
216 | Error **errp; | |
d57c44d0 | 217 | } truncate; |
93f4e2ff | 218 | }; |
de81a169 PB |
219 | } RawPosixAIOData; |
220 | ||
a167ba50 | 221 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 222 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
223 | #endif |
224 | ||
797e3e38 DE |
225 | /* |
226 | * Elide EAGAIN and EACCES details when failing to lock, as this | |
227 | * indicates that the specified file region is already locked by | |
228 | * another process, which is considered a common scenario. | |
229 | */ | |
230 | #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ | |
231 | do { \ | |
232 | if ((err) == EAGAIN || (err) == EACCES) { \ | |
233 | error_setg((errp), (fmt), ## __VA_ARGS__); \ | |
234 | } else { \ | |
235 | error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ | |
236 | } \ | |
237 | } while (0) | |
238 | ||
1de1ae0a | 239 | #if defined(__NetBSD__) |
db0754df | 240 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
241 | { |
242 | static char namebuf[PATH_MAX]; | |
243 | const char *dp, *fname; | |
244 | struct stat sb; | |
245 | ||
246 | fname = *filename; | |
247 | dp = strrchr(fname, '/'); | |
248 | if (lstat(fname, &sb) < 0) { | |
f6fc1e30 | 249 | error_setg_file_open(errp, errno, fname); |
1de1ae0a CE |
250 | return -errno; |
251 | } | |
252 | ||
253 | if (!S_ISBLK(sb.st_mode)) { | |
254 | return 0; | |
255 | } | |
256 | ||
257 | if (dp == NULL) { | |
258 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
259 | } else { | |
260 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
261 | (int)(dp - fname), fname, dp + 1); | |
262 | } | |
1de1ae0a | 263 | *filename = namebuf; |
db0754df | 264 | warn_report("%s is a block device, using %s", fname, *filename); |
1de1ae0a CE |
265 | |
266 | return 0; | |
267 | } | |
268 | #else | |
db0754df | 269 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
270 | { |
271 | return 0; | |
272 | } | |
273 | #endif | |
274 | ||
8a4ed0d1 ET |
275 | /* |
276 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
277 | */ | |
278 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 279 | { |
c25f53b0 | 280 | unsigned int sector_size; |
8a4ed0d1 | 281 | bool success = false; |
700f9ce0 | 282 | int i; |
c25f53b0 | 283 | |
8a4ed0d1 | 284 | errno = ENOTSUP; |
700f9ce0 | 285 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 286 | #ifdef BLKSSZGET |
700f9ce0 | 287 | BLKSSZGET, |
c25f53b0 PB |
288 | #endif |
289 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 290 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
291 | #endif |
292 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 293 | DIOCGSECTORSIZE, |
c25f53b0 | 294 | #endif |
700f9ce0 PM |
295 | }; |
296 | ||
297 | /* Try a few ioctls to get the right size */ | |
298 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
299 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
300 | *sector_size_p = sector_size; | |
301 | success = true; | |
302 | } | |
303 | } | |
8a4ed0d1 ET |
304 | |
305 | return success ? 0 : -errno; | |
306 | } | |
307 | ||
1a9335e4 ET |
308 | /** |
309 | * Get physical block size of @fd. | |
310 | * On success, store it in @blk_size and return 0. | |
311 | * On failure, return -errno. | |
312 | */ | |
313 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
314 | { | |
315 | #ifdef BLKPBSZGET | |
316 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
317 | return -errno; | |
318 | } | |
319 | return 0; | |
320 | #else | |
321 | return -ENOTSUP; | |
322 | #endif | |
323 | } | |
324 | ||
5edc8557 KW |
325 | /* |
326 | * Returns true if no alignment restrictions are necessary even for files | |
327 | * opened with O_DIRECT. | |
328 | * | |
329 | * raw_probe_alignment() probes the required alignment and assume that 1 means | |
330 | * the probing failed, so it falls back to a safe default of 4k. This can be | |
331 | * avoided if we know that byte alignment is okay for the file. | |
332 | */ | |
333 | static bool dio_byte_aligned(int fd) | |
334 | { | |
335 | #ifdef __linux__ | |
336 | struct statfs buf; | |
337 | int ret; | |
338 | ||
339 | ret = fstatfs(fd, &buf); | |
340 | if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { | |
341 | return true; | |
342 | } | |
343 | #endif | |
344 | return false; | |
345 | } | |
346 | ||
5dbd0ce1 KW |
347 | static bool raw_needs_alignment(BlockDriverState *bs) |
348 | { | |
349 | BDRVRawState *s = bs->opaque; | |
350 | ||
351 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { | |
352 | return true; | |
353 | } | |
354 | ||
355 | return s->force_alignment; | |
356 | } | |
357 | ||
22d182e8 SH |
358 | /* Check if read is allowed with given memory buffer and length. |
359 | * | |
360 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
361 | */ | |
362 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
363 | { | |
364 | ssize_t ret = pread(fd, buf, len, 0); | |
365 | ||
366 | if (ret >= 0) { | |
367 | return true; | |
368 | } | |
369 | ||
370 | #ifdef __linux__ | |
371 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
372 | * other errors (e.g. real I/O error), which could happen on a failed | |
373 | * drive, since we only care about probing alignment. | |
374 | */ | |
375 | if (errno != EINVAL) { | |
376 | return true; | |
377 | } | |
378 | #endif | |
379 | ||
380 | return false; | |
381 | } | |
382 | ||
8a4ed0d1 ET |
383 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
384 | { | |
385 | BDRVRawState *s = bs->opaque; | |
386 | char *buf; | |
8e3b0cbb | 387 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size()); |
a6b257a0 | 388 | size_t alignments[] = {1, 512, 1024, 2048, 4096}; |
8a4ed0d1 | 389 | |
b192af8a | 390 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 391 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 392 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 393 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
394 | s->buf_align = 1; |
395 | return; | |
396 | } | |
397 | ||
a5b8dd2c | 398 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
399 | s->buf_align = 0; |
400 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
401 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
402 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 403 | } |
a5730b8b TH |
404 | |
405 | #ifdef __linux__ | |
406 | /* | |
407 | * The XFS ioctl definitions are shipped in extra packages that might | |
408 | * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl | |
409 | * here, we simply use our own definition instead: | |
410 | */ | |
411 | struct xfs_dioattr { | |
412 | uint32_t d_mem; | |
413 | uint32_t d_miniosz; | |
414 | uint32_t d_maxiosz; | |
415 | } da; | |
416 | if (ioctl(fd, _IOR('X', 30, struct xfs_dioattr), &da) >= 0) { | |
417 | bs->bl.request_alignment = da.d_miniosz; | |
418 | /* The kernel returns wrong information for d_mem */ | |
419 | /* s->buf_align = da.d_mem; */ | |
c25f53b0 PB |
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 { | |
230ff739 JS |
757 | s->has_fallocate = true; |
758 | } | |
759 | } else { | |
760 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
8d17adf3 DB |
761 | error_setg(errp, "'%s' driver requires '%s' to be either " |
762 | "a character or block device", | |
763 | bs->drv->format_name, bs->filename); | |
230ff739 JS |
764 | ret = -EINVAL; |
765 | goto fail; | |
766 | } | |
260a82e5 | 767 | } |
230ff739 | 768 | |
d0b4503e | 769 | if (S_ISBLK(st.st_mode)) { |
d0b4503e PB |
770 | #ifdef __linux__ |
771 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
772 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 773 | * Same for BLKZEROOUT. |
d0b4503e PB |
774 | */ |
775 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
97a2ae34 | 776 | s->has_write_zeroes = false; |
d0b4503e PB |
777 | } |
778 | #endif | |
779 | } | |
3cad8307 RPM |
780 | #ifdef __FreeBSD__ |
781 | if (S_ISCHR(st.st_mode)) { | |
782 | /* | |
783 | * The file is a char device (disk), which on FreeBSD isn't behind | |
784 | * a pager, so force all requests to be aligned. This is needed | |
785 | * so QEMU makes sure all IO operations on the device are aligned | |
786 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
787 | */ | |
5dbd0ce1 | 788 | s->force_alignment = true; |
3cad8307 RPM |
789 | } |
790 | #endif | |
5dbd0ce1 | 791 | s->needs_alignment = raw_needs_alignment(bs); |
260a82e5 | 792 | |
738301e1 | 793 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
2f0c6e7a KW |
794 | if (S_ISREG(st.st_mode)) { |
795 | /* When extending regular files, we get zeros from the OS */ | |
796 | bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; | |
797 | } | |
c66a6157 KW |
798 | ret = 0; |
799 | fail: | |
a8c5cf27 KW |
800 | if (ret < 0 && s->fd != -1) { |
801 | qemu_close(s->fd); | |
802 | } | |
8bfea15d KW |
803 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
804 | unlink(filename); | |
805 | } | |
c66a6157 KW |
806 | qemu_opts_del(opts); |
807 | return ret; | |
83f64091 FB |
808 | } |
809 | ||
015a1036 HR |
810 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
811 | Error **errp) | |
90babde0 CH |
812 | { |
813 | BDRVRawState *s = bs->opaque; | |
814 | ||
815 | s->type = FTYPE_FILE; | |
230ff739 | 816 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
817 | } |
818 | ||
244a5668 FZ |
819 | typedef enum { |
820 | RAW_PL_PREPARE, | |
821 | RAW_PL_COMMIT, | |
822 | RAW_PL_ABORT, | |
823 | } RawPermLockOp; | |
824 | ||
825 | #define PERM_FOREACH(i) \ | |
826 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
827 | ||
828 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
829 | * file; if @unlock == true, also unlock the unneeded bytes. | |
830 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
831 | */ | |
2996ffad | 832 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
833 | uint64_t perm_lock_bits, |
834 | uint64_t shared_perm_lock_bits, | |
835 | bool unlock, Error **errp) | |
836 | { | |
837 | int ret; | |
838 | int i; | |
2996ffad FZ |
839 | uint64_t locked_perm, locked_shared_perm; |
840 | ||
841 | if (s) { | |
842 | locked_perm = s->locked_perm; | |
843 | locked_shared_perm = s->locked_shared_perm; | |
844 | } else { | |
845 | /* | |
846 | * We don't have the previous bits, just lock/unlock for each of the | |
847 | * requested bits. | |
848 | */ | |
849 | if (unlock) { | |
850 | locked_perm = BLK_PERM_ALL; | |
851 | locked_shared_perm = BLK_PERM_ALL; | |
852 | } else { | |
853 | locked_perm = 0; | |
854 | locked_shared_perm = 0; | |
855 | } | |
856 | } | |
244a5668 FZ |
857 | |
858 | PERM_FOREACH(i) { | |
859 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
860 | uint64_t bit = (1ULL << i); |
861 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 862 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 | 863 | if (ret) { |
797e3e38 DE |
864 | raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", |
865 | off); | |
244a5668 | 866 | return ret; |
2996ffad FZ |
867 | } else if (s) { |
868 | s->locked_perm |= bit; | |
244a5668 | 869 | } |
2996ffad | 870 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 871 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 | 872 | if (ret) { |
797e3e38 | 873 | error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); |
244a5668 | 874 | return ret; |
2996ffad FZ |
875 | } else if (s) { |
876 | s->locked_perm &= ~bit; | |
244a5668 FZ |
877 | } |
878 | } | |
879 | } | |
880 | PERM_FOREACH(i) { | |
881 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
882 | uint64_t bit = (1ULL << i); |
883 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 884 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 | 885 | if (ret) { |
797e3e38 DE |
886 | raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", |
887 | off); | |
244a5668 | 888 | return ret; |
2996ffad FZ |
889 | } else if (s) { |
890 | s->locked_shared_perm |= bit; | |
244a5668 | 891 | } |
2996ffad FZ |
892 | } else if (unlock && (locked_shared_perm & bit) && |
893 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 894 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 | 895 | if (ret) { |
797e3e38 | 896 | error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); |
244a5668 | 897 | return ret; |
2996ffad FZ |
898 | } else if (s) { |
899 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
900 | } |
901 | } | |
902 | } | |
903 | return 0; | |
904 | } | |
905 | ||
906 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 907 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
908 | Error **errp) |
909 | { | |
910 | int ret; | |
911 | int i; | |
912 | ||
913 | PERM_FOREACH(i) { | |
914 | int off = RAW_LOCK_SHARED_BASE + i; | |
915 | uint64_t p = 1ULL << i; | |
916 | if (perm & p) { | |
d0a96155 | 917 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
918 | if (ret) { |
919 | char *perm_name = bdrv_perm_names(p); | |
797e3e38 DE |
920 | |
921 | raw_lock_error_setg_errno(errp, -ret, | |
922 | "Failed to get \"%s\" lock", | |
923 | perm_name); | |
244a5668 | 924 | g_free(perm_name); |
244a5668 FZ |
925 | return ret; |
926 | } | |
927 | } | |
928 | } | |
929 | PERM_FOREACH(i) { | |
930 | int off = RAW_LOCK_PERM_BASE + i; | |
931 | uint64_t p = 1ULL << i; | |
932 | if (!(shared_perm & p)) { | |
d0a96155 | 933 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
934 | if (ret) { |
935 | char *perm_name = bdrv_perm_names(p); | |
797e3e38 DE |
936 | |
937 | raw_lock_error_setg_errno(errp, -ret, | |
938 | "Failed to get shared \"%s\" lock", | |
939 | perm_name); | |
244a5668 | 940 | g_free(perm_name); |
244a5668 FZ |
941 | return ret; |
942 | } | |
943 | } | |
944 | } | |
945 | return 0; | |
946 | } | |
947 | ||
948 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
949 | RawPermLockOp op, | |
950 | uint64_t new_perm, uint64_t new_shared, | |
951 | Error **errp) | |
952 | { | |
953 | BDRVRawState *s = bs->opaque; | |
954 | int ret = 0; | |
955 | Error *local_err = NULL; | |
956 | ||
957 | if (!s->use_lock) { | |
958 | return 0; | |
959 | } | |
960 | ||
961 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
962 | return 0; | |
963 | } | |
964 | ||
244a5668 FZ |
965 | switch (op) { |
966 | case RAW_PL_PREPARE: | |
696aaaed VSO |
967 | if ((s->perm | new_perm) == s->perm && |
968 | (s->shared_perm & new_shared) == s->shared_perm) | |
969 | { | |
970 | /* | |
971 | * We are going to unlock bytes, it should not fail. If it fail due | |
972 | * to some fs-dependent permission-unrelated reasons (which occurs | |
973 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
974 | * can't prevent such errors by any check here. And we ignore them | |
975 | * anyway in ABORT and COMMIT. | |
976 | */ | |
977 | return 0; | |
978 | } | |
f2e3af29 | 979 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
980 | ~s->shared_perm | ~new_shared, |
981 | false, errp); | |
982 | if (!ret) { | |
f2e3af29 | 983 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
984 | if (!ret) { |
985 | return 0; | |
986 | } | |
b857431d FZ |
987 | error_append_hint(errp, |
988 | "Is another process using the image [%s]?\n", | |
989 | bs->filename); | |
244a5668 | 990 | } |
244a5668 FZ |
991 | /* fall through to unlock bytes. */ |
992 | case RAW_PL_ABORT: | |
f2e3af29 | 993 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 994 | true, &local_err); |
244a5668 FZ |
995 | if (local_err) { |
996 | /* Theoretically the above call only unlocks bytes and it cannot | |
997 | * fail. Something weird happened, report it. | |
998 | */ | |
db0754df | 999 | warn_report_err(local_err); |
244a5668 FZ |
1000 | } |
1001 | break; | |
1002 | case RAW_PL_COMMIT: | |
f2e3af29 | 1003 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 1004 | true, &local_err); |
244a5668 FZ |
1005 | if (local_err) { |
1006 | /* Theoretically the above call only unlocks bytes and it cannot | |
1007 | * fail. Something weird happened, report it. | |
1008 | */ | |
db0754df | 1009 | warn_report_err(local_err); |
244a5668 FZ |
1010 | } |
1011 | break; | |
1012 | } | |
1013 | return ret; | |
1014 | } | |
1015 | ||
ad24b679 MAL |
1016 | /* Sets a specific flag */ |
1017 | static int fcntl_setfl(int fd, int flag) | |
1018 | { | |
1019 | int flags; | |
1020 | ||
1021 | flags = fcntl(fd, F_GETFL); | |
1022 | if (flags == -1) { | |
1023 | return -errno; | |
1024 | } | |
1025 | if (fcntl(fd, F_SETFL, flags | flag) == -1) { | |
1026 | return -errno; | |
1027 | } | |
1028 | return 0; | |
1029 | } | |
1030 | ||
5cec2870 | 1031 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 1032 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 1033 | Error **errp) |
5cec2870 KW |
1034 | { |
1035 | BDRVRawState *s = bs->opaque; | |
1036 | int fd = -1; | |
1037 | int ret; | |
23dece19 KW |
1038 | bool has_writers = perm & |
1039 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
1040 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
1041 | #ifdef O_NOATIME | |
1042 | fcntl_flags |= O_NOATIME; | |
1043 | #endif | |
1044 | ||
1045 | *open_flags = 0; | |
1046 | if (s->type == FTYPE_CD) { | |
1047 | *open_flags |= O_NONBLOCK; | |
1048 | } | |
1049 | ||
23dece19 | 1050 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
1051 | |
1052 | #ifdef O_ASYNC | |
1053 | /* Not all operating systems have O_ASYNC, and those that don't | |
1054 | * will not let us track the state into rs->open_flags (typically | |
1055 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
1056 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
1057 | */ | |
1058 | assert((s->open_flags & O_ASYNC) == 0); | |
1059 | #endif | |
1060 | ||
6ceabe6f KW |
1061 | if (!force_dup && *open_flags == s->open_flags) { |
1062 | /* We're lucky, the existing fd is fine */ | |
1063 | return s->fd; | |
1064 | } | |
1065 | ||
5cec2870 KW |
1066 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
1067 | /* dup the original fd */ | |
1068 | fd = qemu_dup(s->fd); | |
1069 | if (fd >= 0) { | |
1070 | ret = fcntl_setfl(fd, *open_flags); | |
1071 | if (ret) { | |
1072 | qemu_close(fd); | |
1073 | fd = -1; | |
1074 | } | |
1075 | } | |
1076 | } | |
1077 | ||
b18a24a9 | 1078 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ |
5cec2870 KW |
1079 | if (fd == -1) { |
1080 | const char *normalized_filename = bs->filename; | |
1081 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
1082 | if (ret >= 0) { | |
b18a24a9 | 1083 | fd = qemu_open(normalized_filename, *open_flags, errp); |
5cec2870 | 1084 | if (fd == -1) { |
5cec2870 KW |
1085 | return -1; |
1086 | } | |
1087 | } | |
1088 | } | |
1089 | ||
bca5283b KW |
1090 | if (fd != -1 && (*open_flags & O_RDWR)) { |
1091 | ret = check_hdev_writable(fd); | |
1092 | if (ret < 0) { | |
1093 | qemu_close(fd); | |
1094 | error_setg_errno(errp, -ret, "The device is not writable"); | |
1095 | return -1; | |
1096 | } | |
1097 | } | |
1098 | ||
5cec2870 KW |
1099 | return fd; |
1100 | } | |
1101 | ||
eeb6b45d JC |
1102 | static int raw_reopen_prepare(BDRVReopenState *state, |
1103 | BlockReopenQueue *queue, Error **errp) | |
1104 | { | |
1105 | BDRVRawState *s; | |
4e6d13c9 | 1106 | BDRVRawReopenState *rs; |
31be8a2a | 1107 | QemuOpts *opts; |
a6aeca0c | 1108 | int ret; |
eeb6b45d JC |
1109 | |
1110 | assert(state != NULL); | |
1111 | assert(state->bs != NULL); | |
1112 | ||
1113 | s = state->bs->opaque; | |
1114 | ||
5839e53b | 1115 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 1116 | rs = state->opaque; |
31be8a2a SH |
1117 | |
1118 | /* Handle options changes */ | |
1119 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 1120 | if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { |
31be8a2a SH |
1121 | ret = -EINVAL; |
1122 | goto out; | |
1123 | } | |
1124 | ||
f357fcd8 | 1125 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
1126 | rs->check_cache_dropped = |
1127 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
1128 | ||
1129 | /* This driver's reopen function doesn't currently allow changing | |
1130 | * other options, so let's put them back in the original QDict and | |
1131 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
1132 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 1133 | |
72373e40 VSO |
1134 | /* |
1135 | * As part of reopen prepare we also want to create new fd by | |
1136 | * raw_reconfigure_getfd(). But it wants updated "perm", when in | |
1137 | * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to | |
1138 | * permission update. Happily, permission update is always a part (a seprate | |
1139 | * stage) of bdrv_reopen_multiple() so we can rely on this fact and | |
1140 | * reconfigure fd in raw_check_perm(). | |
1141 | */ | |
df26a350 | 1142 | |
e0c9cf3a | 1143 | s->reopen_state = state; |
a6aeca0c | 1144 | ret = 0; |
72373e40 | 1145 | |
31be8a2a SH |
1146 | out: |
1147 | qemu_opts_del(opts); | |
eeb6b45d JC |
1148 | return ret; |
1149 | } | |
1150 | ||
eeb6b45d JC |
1151 | static void raw_reopen_commit(BDRVReopenState *state) |
1152 | { | |
4e6d13c9 | 1153 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1154 | BDRVRawState *s = state->bs->opaque; |
1155 | ||
f357fcd8 | 1156 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1157 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1158 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1159 | g_free(state->opaque); |
1160 | state->opaque = NULL; | |
e0c9cf3a KW |
1161 | |
1162 | assert(s->reopen_state == state); | |
1163 | s->reopen_state = NULL; | |
eeb6b45d JC |
1164 | } |
1165 | ||
1166 | ||
1167 | static void raw_reopen_abort(BDRVReopenState *state) | |
1168 | { | |
4e6d13c9 | 1169 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1170 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1171 | |
1172 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1173 | if (rs == NULL) { |
eeb6b45d JC |
1174 | return; |
1175 | } | |
1176 | ||
eeb6b45d JC |
1177 | g_free(state->opaque); |
1178 | state->opaque = NULL; | |
e0c9cf3a KW |
1179 | |
1180 | assert(s->reopen_state == state); | |
1181 | s->reopen_state = NULL; | |
eeb6b45d JC |
1182 | } |
1183 | ||
18473467 | 1184 | static int hdev_get_max_hw_transfer(int fd, struct stat *st) |
6f607174 FZ |
1185 | { |
1186 | #ifdef BLKSECTGET | |
18473467 PB |
1187 | if (S_ISBLK(st->st_mode)) { |
1188 | unsigned short max_sectors = 0; | |
1189 | if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) { | |
1190 | return max_sectors * 512; | |
1191 | } | |
6f607174 | 1192 | } else { |
18473467 PB |
1193 | int max_bytes = 0; |
1194 | if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
1195 | return max_bytes; | |
1196 | } | |
6f607174 | 1197 | } |
18473467 | 1198 | return -errno; |
6f607174 FZ |
1199 | #else |
1200 | return -ENOSYS; | |
1201 | #endif | |
1202 | } | |
1203 | ||
18473467 | 1204 | static int hdev_get_max_segments(int fd, struct stat *st) |
9103f1ce FZ |
1205 | { |
1206 | #ifdef CONFIG_LINUX | |
1207 | char buf[32]; | |
1208 | const char *end; | |
867eccfe | 1209 | char *sysfspath = NULL; |
9103f1ce | 1210 | int ret; |
867eccfe | 1211 | int sysfd = -1; |
9103f1ce | 1212 | long max_segments; |
867eccfe | 1213 | |
18473467 | 1214 | if (S_ISCHR(st->st_mode)) { |
8ad5ab61 PB |
1215 | if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) { |
1216 | return ret; | |
1217 | } | |
1218 | return -ENOTSUP; | |
1219 | } | |
1220 | ||
18473467 | 1221 | if (!S_ISBLK(st->st_mode)) { |
8ad5ab61 PB |
1222 | return -ENOTSUP; |
1223 | } | |
1224 | ||
9103f1ce | 1225 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", |
18473467 | 1226 | major(st->st_rdev), minor(st->st_rdev)); |
867eccfe ML |
1227 | sysfd = open(sysfspath, O_RDONLY); |
1228 | if (sysfd == -1) { | |
9103f1ce FZ |
1229 | ret = -errno; |
1230 | goto out; | |
1231 | } | |
1232 | do { | |
867eccfe | 1233 | ret = read(sysfd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1234 | } while (ret == -1 && errno == EINTR); |
1235 | if (ret < 0) { | |
1236 | ret = -errno; | |
1237 | goto out; | |
1238 | } else if (ret == 0) { | |
1239 | ret = -EIO; | |
1240 | goto out; | |
1241 | } | |
1242 | buf[ret] = 0; | |
1243 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1244 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1245 | if (ret == 0 && end && *end == '\n') { | |
1246 | ret = max_segments; | |
1247 | } | |
1248 | ||
1249 | out: | |
867eccfe ML |
1250 | if (sysfd != -1) { |
1251 | close(sysfd); | |
fed414df | 1252 | } |
9103f1ce FZ |
1253 | g_free(sysfspath); |
1254 | return ret; | |
1255 | #else | |
1256 | return -ENOTSUP; | |
1257 | #endif | |
1258 | } | |
1259 | ||
3baca891 | 1260 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1261 | { |
1262 | BDRVRawState *s = bs->opaque; | |
18473467 | 1263 | struct stat st; |
6f607174 | 1264 | |
5dbd0ce1 | 1265 | s->needs_alignment = raw_needs_alignment(bs); |
18473467 | 1266 | raw_probe_alignment(bs, s->fd, errp); |
5dbd0ce1 | 1267 | |
18473467 | 1268 | bs->bl.min_mem_alignment = s->buf_align; |
8e3b0cbb | 1269 | bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size()); |
18473467 PB |
1270 | |
1271 | /* | |
1272 | * Maximum transfers are best effort, so it is okay to ignore any | |
1273 | * errors. That said, based on the man page errors in fstat would be | |
1274 | * very much unexpected; the only possible case seems to be ENOMEM. | |
1275 | */ | |
1276 | if (fstat(s->fd, &st)) { | |
1277 | return; | |
1278 | } | |
1279 | ||
0dfc7af2 AO |
1280 | #if defined(__APPLE__) && (__MACH__) |
1281 | struct statfs buf; | |
1282 | ||
1283 | if (!fstatfs(s->fd, &buf)) { | |
1284 | bs->bl.opt_transfer = buf.f_iosize; | |
1285 | bs->bl.pdiscard_alignment = buf.f_bsize; | |
1286 | } | |
1287 | #endif | |
1288 | ||
006e1962 | 1289 | if (bdrv_is_sg(bs) || S_ISBLK(st.st_mode)) { |
18473467 | 1290 | int ret = hdev_get_max_hw_transfer(s->fd, &st); |
867eccfe ML |
1291 | |
1292 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
18473467 | 1293 | bs->bl.max_hw_transfer = ret; |
867eccfe ML |
1294 | } |
1295 | ||
18473467 | 1296 | ret = hdev_get_max_segments(s->fd, &st); |
867eccfe | 1297 | if (ret > 0) { |
cc071629 | 1298 | bs->bl.max_hw_iov = ret; |
6f607174 FZ |
1299 | } |
1300 | } | |
c25f53b0 | 1301 | } |
83f64091 | 1302 | |
1a9335e4 ET |
1303 | static int check_for_dasd(int fd) |
1304 | { | |
1305 | #ifdef BIODASDINFO2 | |
1306 | struct dasd_information2_t info = {0}; | |
1307 | ||
1308 | return ioctl(fd, BIODASDINFO2, &info); | |
1309 | #else | |
1310 | return -1; | |
1311 | #endif | |
1312 | } | |
1313 | ||
1314 | /** | |
1315 | * Try to get @bs's logical and physical block size. | |
1316 | * On success, store them in @bsz and return zero. | |
1317 | * On failure, return negative errno. | |
1318 | */ | |
1319 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1320 | { | |
1321 | BDRVRawState *s = bs->opaque; | |
1322 | int ret; | |
1323 | ||
1324 | /* If DASD, get blocksizes */ | |
1325 | if (check_for_dasd(s->fd) < 0) { | |
1326 | return -ENOTSUP; | |
1327 | } | |
1328 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1329 | if (ret < 0) { | |
1330 | return ret; | |
1331 | } | |
1332 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1333 | } | |
1334 | ||
1335 | /** | |
1336 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1337 | * On success, store them in @geo and return 0. | |
1338 | * On failure return -errno. | |
1339 | * (Allows block driver to assign default geometry values that guest sees) | |
1340 | */ | |
1341 | #ifdef __linux__ | |
1342 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1343 | { | |
1344 | BDRVRawState *s = bs->opaque; | |
1345 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1346 | |
1347 | /* If DASD, get its geometry */ | |
1348 | if (check_for_dasd(s->fd) < 0) { | |
1349 | return -ENOTSUP; | |
1350 | } | |
1351 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1352 | return -errno; | |
1353 | } | |
1354 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1355 | (e.g. certain multipath setups) */ | |
1356 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1357 | return -ENOTSUP; | |
1358 | } | |
1359 | /* Do not return a geometry for partition */ | |
1360 | if (ioctl_geo.start != 0) { | |
1361 | return -ENOTSUP; | |
1362 | } | |
1363 | geo->heads = ioctl_geo.heads; | |
1364 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1365 | geo->cylinders = ioctl_geo.cylinders; |
1366 | ||
1367 | return 0; | |
1368 | } | |
1369 | #else /* __linux__ */ | |
1370 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1371 | { | |
1372 | return -ENOTSUP; | |
1373 | } | |
1374 | #endif | |
1375 | ||
03425671 KW |
1376 | #if defined(__linux__) |
1377 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1378 | { |
03425671 | 1379 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1380 | int ret; |
1381 | ||
bd80936a PB |
1382 | do { |
1383 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); | |
1384 | } while (ret == -1 && errno == EINTR); | |
de81a169 PB |
1385 | if (ret == -1) { |
1386 | return -errno; | |
1387 | } | |
1388 | ||
b608c8dc | 1389 | return 0; |
de81a169 | 1390 | } |
03425671 | 1391 | #endif /* linux */ |
de81a169 | 1392 | |
06dc9bd5 | 1393 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1394 | { |
06dc9bd5 | 1395 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1396 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1397 | int ret; |
1398 | ||
e5bcf967 | 1399 | if (s->page_cache_inconsistent) { |
c7ddc882 | 1400 | return -s->page_cache_inconsistent; |
e5bcf967 KW |
1401 | } |
1402 | ||
de81a169 PB |
1403 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1404 | if (ret == -1) { | |
60ff2ae2 DB |
1405 | trace_file_flush_fdatasync_failed(errno); |
1406 | ||
e5bcf967 KW |
1407 | /* There is no clear definition of the semantics of a failing fsync(), |
1408 | * so we may have to assume the worst. The sad truth is that this | |
1409 | * assumption is correct for Linux. Some pages are now probably marked | |
1410 | * clean in the page cache even though they are inconsistent with the | |
1411 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1412 | * further writeback attempt will be made. We can't get back to a state | |
1413 | * in which we know what is on disk (we would have to rewrite | |
1414 | * everything that was touched since the last fdatasync() at least), so | |
1415 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1416 | * really defined, I have little hope that other OSes are doing better. | |
1417 | * | |
1418 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1419 | * cache. */ | |
1420 | if ((s->open_flags & O_DIRECT) == 0) { | |
c7ddc882 | 1421 | s->page_cache_inconsistent = errno; |
e5bcf967 | 1422 | } |
de81a169 PB |
1423 | return -errno; |
1424 | } | |
1425 | return 0; | |
1426 | } | |
1427 | ||
1428 | #ifdef CONFIG_PREADV | |
1429 | ||
1430 | static bool preadv_present = true; | |
1431 | ||
1432 | static ssize_t | |
1433 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1434 | { | |
1435 | return preadv(fd, iov, nr_iov, offset); | |
1436 | } | |
1437 | ||
1438 | static ssize_t | |
1439 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1440 | { | |
1441 | return pwritev(fd, iov, nr_iov, offset); | |
1442 | } | |
1443 | ||
1444 | #else | |
1445 | ||
1446 | static bool preadv_present = false; | |
1447 | ||
1448 | static ssize_t | |
1449 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1450 | { | |
1451 | return -ENOSYS; | |
1452 | } | |
1453 | ||
1454 | static ssize_t | |
1455 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1456 | { | |
1457 | return -ENOSYS; | |
1458 | } | |
1459 | ||
1460 | #endif | |
1461 | ||
1462 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1463 | { | |
1464 | ssize_t len; | |
1465 | ||
1466 | do { | |
1467 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1468 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1469 | aiocb->io.iov, |
1470 | aiocb->io.niov, | |
de81a169 PB |
1471 | aiocb->aio_offset); |
1472 | else | |
1473 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1474 | aiocb->io.iov, |
1475 | aiocb->io.niov, | |
de81a169 PB |
1476 | aiocb->aio_offset); |
1477 | } while (len == -1 && errno == EINTR); | |
1478 | ||
1479 | if (len == -1) { | |
1480 | return -errno; | |
1481 | } | |
1482 | return len; | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * Read/writes the data to/from a given linear buffer. | |
1487 | * | |
1488 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1489 | * reads are only returned if the end of the file is reached. | |
1490 | */ | |
1491 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1492 | { | |
1493 | ssize_t offset = 0; | |
1494 | ssize_t len; | |
1495 | ||
1496 | while (offset < aiocb->aio_nbytes) { | |
1497 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1498 | len = pwrite(aiocb->aio_fildes, | |
1499 | (const char *)buf + offset, | |
1500 | aiocb->aio_nbytes - offset, | |
1501 | aiocb->aio_offset + offset); | |
1502 | } else { | |
1503 | len = pread(aiocb->aio_fildes, | |
1504 | buf + offset, | |
1505 | aiocb->aio_nbytes - offset, | |
1506 | aiocb->aio_offset + offset); | |
1507 | } | |
1508 | if (len == -1 && errno == EINTR) { | |
1509 | continue; | |
61ed73cf SH |
1510 | } else if (len == -1 && errno == EINVAL && |
1511 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1512 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1513 | offset > 0) { | |
1514 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1515 | * after a short read. Assume that O_DIRECT short reads only occur | |
1516 | * at EOF. Therefore this is a short read, not an I/O error. | |
1517 | */ | |
1518 | break; | |
de81a169 PB |
1519 | } else if (len == -1) { |
1520 | offset = -errno; | |
1521 | break; | |
1522 | } else if (len == 0) { | |
1523 | break; | |
1524 | } | |
1525 | offset += len; | |
1526 | } | |
1527 | ||
1528 | return offset; | |
1529 | } | |
1530 | ||
999e6b69 | 1531 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1532 | { |
999e6b69 | 1533 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1534 | ssize_t nbytes; |
1535 | char *buf; | |
1536 | ||
1537 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1538 | /* | |
1539 | * If there is just a single buffer, and it is properly aligned | |
1540 | * we can just use plain pread/pwrite without any problems. | |
1541 | */ | |
d57c44d0 | 1542 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1543 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1544 | goto out; | |
de81a169 PB |
1545 | } |
1546 | /* | |
1547 | * We have more than one iovec, and all are properly aligned. | |
1548 | * | |
1549 | * Try preadv/pwritev first and fall back to linearizing the | |
1550 | * buffer if it's not supported. | |
1551 | */ | |
1552 | if (preadv_present) { | |
1553 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1554 | if (nbytes == aiocb->aio_nbytes || | |
1555 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1556 | goto out; |
de81a169 PB |
1557 | } |
1558 | preadv_present = false; | |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1563 | * using these interfaces. For now retry using plain | |
1564 | * pread/pwrite? | |
1565 | */ | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * Ok, we have to do it the hard way, copy all segments into | |
1570 | * a single aligned buffer. | |
1571 | */ | |
50d4a858 KW |
1572 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1573 | if (buf == NULL) { | |
54c7ca1b KW |
1574 | nbytes = -ENOMEM; |
1575 | goto out; | |
50d4a858 KW |
1576 | } |
1577 | ||
de81a169 PB |
1578 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1579 | char *p = buf; | |
1580 | int i; | |
1581 | ||
d57c44d0 KW |
1582 | for (i = 0; i < aiocb->io.niov; ++i) { |
1583 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1584 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1585 | } |
8eb029c2 | 1586 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1587 | } |
1588 | ||
1589 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1590 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1591 | char *p = buf; | |
1592 | size_t count = aiocb->aio_nbytes, copy; | |
1593 | int i; | |
1594 | ||
d57c44d0 | 1595 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1596 | copy = count; |
d57c44d0 KW |
1597 | if (copy > aiocb->io.iov[i].iov_len) { |
1598 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1599 | } |
d57c44d0 | 1600 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1601 | assert(count >= copy); |
de81a169 PB |
1602 | p += copy; |
1603 | count -= copy; | |
1604 | } | |
8eb029c2 | 1605 | assert(count == 0); |
de81a169 PB |
1606 | } |
1607 | qemu_vfree(buf); | |
1608 | ||
54c7ca1b KW |
1609 | out: |
1610 | if (nbytes == aiocb->aio_nbytes) { | |
1611 | return 0; | |
1612 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1613 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1614 | return -EINVAL; | |
1615 | } else { | |
1616 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1617 | 0, aiocb->aio_nbytes - nbytes); | |
1618 | return 0; | |
1619 | } | |
1620 | } else { | |
1621 | assert(nbytes < 0); | |
1622 | return nbytes; | |
1623 | } | |
de81a169 PB |
1624 | } |
1625 | ||
0dfc7af2 | 1626 | #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD) |
1486df0e DL |
1627 | static int translate_err(int err) |
1628 | { | |
1629 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1630 | err == -ENOTTY) { | |
1631 | err = -ENOTSUP; | |
1632 | } | |
1633 | return err; | |
1634 | } | |
0dfc7af2 | 1635 | #endif |
1486df0e | 1636 | |
d50d8222 | 1637 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1638 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1639 | { | |
1640 | do { | |
1641 | if (fallocate(fd, mode, offset, len) == 0) { | |
1642 | return 0; | |
1643 | } | |
1644 | } while (errno == EINTR); | |
1645 | return translate_err(-errno); | |
1646 | } | |
1647 | #endif | |
1648 | ||
37cc9f7f | 1649 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1650 | { |
37cc9f7f | 1651 | int ret = -ENOTSUP; |
97a2ae34 PB |
1652 | BDRVRawState *s = aiocb->bs->opaque; |
1653 | ||
37cc9f7f | 1654 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1655 | return -ENOTSUP; |
1656 | } | |
1657 | ||
97a2ae34 | 1658 | #ifdef BLKZEROOUT |
738301e1 KW |
1659 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1660 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1661 | * fallbacks. */ | |
1662 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1663 | do { | |
1664 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1665 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1666 | return 0; | |
1667 | } | |
1668 | } while (errno == EINTR); | |
37cc9f7f | 1669 | |
738301e1 | 1670 | ret = translate_err(-errno); |
effecce6 KW |
1671 | if (ret == -ENOTSUP) { |
1672 | s->has_write_zeroes = false; | |
1673 | } | |
738301e1 | 1674 | } |
97a2ae34 | 1675 | #endif |
97a2ae34 | 1676 | |
97a2ae34 PB |
1677 | return ret; |
1678 | } | |
1679 | ||
7154d8ae | 1680 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1681 | { |
7154d8ae | 1682 | RawPosixAIOData *aiocb = opaque; |
70d9110b | 1683 | #ifdef CONFIG_FALLOCATE |
b2c6f23f | 1684 | BDRVRawState *s = aiocb->bs->opaque; |
70d9110b DL |
1685 | int64_t len; |
1686 | #endif | |
37cc9f7f DL |
1687 | |
1688 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1689 | return handle_aiocb_write_zeroes_block(aiocb); | |
1690 | } | |
1691 | ||
b953f075 DL |
1692 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1693 | if (s->has_write_zeroes) { | |
1694 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1695 | aiocb->aio_offset, aiocb->aio_nbytes); | |
fa95e9fb TH |
1696 | if (ret == -ENOTSUP) { |
1697 | s->has_write_zeroes = false; | |
1698 | } else if (ret == 0 || ret != -EINVAL) { | |
b953f075 DL |
1699 | return ret; |
1700 | } | |
fa95e9fb TH |
1701 | /* |
1702 | * Note: Some file systems do not like unaligned byte ranges, and | |
1703 | * return EINVAL in such a case, though they should not do it according | |
1704 | * to the man-page of fallocate(). Thus we simply ignore this return | |
1705 | * value and try the other fallbacks instead. | |
1706 | */ | |
b953f075 DL |
1707 | } |
1708 | #endif | |
1709 | ||
1cdc3239 DL |
1710 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1711 | if (s->has_discard && s->has_fallocate) { | |
1712 | int ret = do_fallocate(s->fd, | |
1713 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1714 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1715 | if (ret == 0) { | |
1716 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1717 | if (ret == 0 || ret != -ENOTSUP) { | |
1718 | return ret; | |
1719 | } | |
1720 | s->has_fallocate = false; | |
73ebf297 TH |
1721 | } else if (ret == -EINVAL) { |
1722 | /* | |
1723 | * Some file systems like older versions of GPFS do not like un- | |
1724 | * aligned byte ranges, and return EINVAL in such a case, though | |
1725 | * they should not do it according to the man-page of fallocate(). | |
1726 | * Warn about the bad filesystem and try the final fallback instead. | |
1727 | */ | |
1728 | warn_report_once("Your file system is misbehaving: " | |
1729 | "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. " | |
68857f13 | 1730 | "Please report this bug to your file system " |
73ebf297 | 1731 | "vendor."); |
1cdc3239 DL |
1732 | } else if (ret != -ENOTSUP) { |
1733 | return ret; | |
1734 | } else { | |
1735 | s->has_discard = false; | |
1736 | } | |
1737 | } | |
1738 | #endif | |
1739 | ||
d50d8222 | 1740 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1741 | /* Last resort: we are trying to extend the file with zeroed data. This |
1742 | * can be done via fallocate(fd, 0) */ | |
1743 | len = bdrv_getlength(aiocb->bs); | |
1744 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1745 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1746 | if (ret == 0 || ret != -ENOTSUP) { | |
1747 | return ret; | |
1748 | } | |
1749 | s->has_fallocate = false; | |
1750 | } | |
1751 | #endif | |
1752 | ||
37cc9f7f DL |
1753 | return -ENOTSUP; |
1754 | } | |
1755 | ||
7154d8ae | 1756 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1757 | { |
7154d8ae | 1758 | RawPosixAIOData *aiocb = opaque; |
34fa110e | 1759 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
34fa110e KW |
1760 | |
1761 | /* First try to write zeros and unmap at the same time */ | |
1762 | ||
1763 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
b3ac2b94 SS |
1764 | int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1765 | aiocb->aio_offset, aiocb->aio_nbytes); | |
bae127d4 AD |
1766 | switch (ret) { |
1767 | case -ENOTSUP: | |
1768 | case -EINVAL: | |
ece4fa91 | 1769 | case -EBUSY: |
bae127d4 AD |
1770 | break; |
1771 | default: | |
34fa110e KW |
1772 | return ret; |
1773 | } | |
1774 | #endif | |
1775 | ||
34fa110e KW |
1776 | /* If we couldn't manage to unmap while guaranteed that the area reads as |
1777 | * all-zero afterwards, just write zeroes without unmapping */ | |
b3ac2b94 | 1778 | return handle_aiocb_write_zeroes(aiocb); |
34fa110e KW |
1779 | } |
1780 | ||
1efad060 FZ |
1781 | #ifndef HAVE_COPY_FILE_RANGE |
1782 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1783 | off_t *out_off, size_t len, unsigned int flags) | |
1784 | { | |
1785 | #ifdef __NR_copy_file_range | |
1786 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1787 | out_off, len, flags); | |
1788 | #else | |
1789 | errno = ENOSYS; | |
1790 | return -1; | |
1791 | #endif | |
1792 | } | |
1793 | #endif | |
1794 | ||
58a209c4 | 1795 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1796 | { |
58a209c4 | 1797 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1798 | uint64_t bytes = aiocb->aio_nbytes; |
1799 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1800 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1801 | |
1802 | while (bytes) { | |
1803 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1804 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1805 | bytes, 0); |
ecc983a5 | 1806 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1807 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1808 | 0, ret); | |
c436e3d0 FZ |
1809 | if (ret == 0) { |
1810 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1811 | * buffer I/O. */ | |
1812 | return -ENOSPC; | |
1efad060 FZ |
1813 | } |
1814 | if (ret < 0) { | |
c436e3d0 FZ |
1815 | switch (errno) { |
1816 | case ENOSYS: | |
1efad060 | 1817 | return -ENOTSUP; |
c436e3d0 FZ |
1818 | case EINTR: |
1819 | continue; | |
1820 | default: | |
1efad060 FZ |
1821 | return -errno; |
1822 | } | |
1823 | } | |
1efad060 FZ |
1824 | bytes -= ret; |
1825 | } | |
1826 | return 0; | |
1827 | } | |
1828 | ||
46ee0f46 | 1829 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1830 | { |
46ee0f46 | 1831 | RawPosixAIOData *aiocb = opaque; |
13a02833 | 1832 | int ret = -ENOTSUP; |
8238010b PB |
1833 | BDRVRawState *s = aiocb->bs->opaque; |
1834 | ||
7ce21016 PB |
1835 | if (!s->has_discard) { |
1836 | return -ENOTSUP; | |
8238010b PB |
1837 | } |
1838 | ||
1839 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1840 | #ifdef BLKDISCARD | |
1841 | do { | |
1842 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1843 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1844 | return 0; | |
1845 | } | |
1846 | } while (errno == EINTR); | |
1847 | ||
0dfc7af2 | 1848 | ret = translate_err(-errno); |
8238010b PB |
1849 | #endif |
1850 | } else { | |
8238010b | 1851 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
0b991712 DL |
1852 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1853 | aiocb->aio_offset, aiocb->aio_nbytes); | |
13a02833 | 1854 | ret = translate_err(ret); |
0dfc7af2 AO |
1855 | #elif defined(__APPLE__) && (__MACH__) |
1856 | fpunchhole_t fpunchhole; | |
1857 | fpunchhole.fp_flags = 0; | |
1858 | fpunchhole.reserved = 0; | |
1859 | fpunchhole.fp_offset = aiocb->aio_offset; | |
1860 | fpunchhole.fp_length = aiocb->aio_nbytes; | |
1861 | if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) { | |
1862 | ret = errno == ENODEV ? -ENOTSUP : -errno; | |
1863 | } else { | |
1864 | ret = 0; | |
1865 | } | |
8238010b PB |
1866 | #endif |
1867 | } | |
1868 | ||
1486df0e | 1869 | if (ret == -ENOTSUP) { |
7ce21016 | 1870 | s->has_discard = false; |
8238010b PB |
1871 | } |
1872 | return ret; | |
1873 | } | |
1874 | ||
3a20013f NS |
1875 | /* |
1876 | * Help alignment probing by allocating the first block. | |
1877 | * | |
1878 | * When reading with direct I/O from unallocated area on Gluster backed by XFS, | |
1879 | * reading succeeds regardless of request length. In this case we fallback to | |
1880 | * safe alignment which is not optimal. Allocating the first block avoids this | |
1881 | * fallback. | |
1882 | * | |
1883 | * fd may be opened with O_DIRECT, but we don't know the buffer alignment or | |
1884 | * request alignment, so we use safe values. | |
1885 | * | |
1886 | * Returns: 0 on success, -errno on failure. Since this is an optimization, | |
1887 | * caller may ignore failures. | |
1888 | */ | |
1889 | static int allocate_first_block(int fd, size_t max_size) | |
1890 | { | |
1891 | size_t write_size = (max_size < MAX_BLOCKSIZE) | |
1892 | ? BDRV_SECTOR_SIZE | |
1893 | : MAX_BLOCKSIZE; | |
8e3b0cbb | 1894 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size()); |
3a20013f NS |
1895 | void *buf; |
1896 | ssize_t n; | |
1897 | int ret; | |
1898 | ||
1899 | buf = qemu_memalign(max_align, write_size); | |
1900 | memset(buf, 0, write_size); | |
1901 | ||
1902 | do { | |
1903 | n = pwrite(fd, buf, write_size, 0); | |
1904 | } while (n == -1 && errno == EINTR); | |
1905 | ||
1906 | ret = (n == -1) ? -errno : 0; | |
1907 | ||
1908 | qemu_vfree(buf); | |
1909 | return ret; | |
1910 | } | |
1911 | ||
29cb4c01 | 1912 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1913 | { |
29cb4c01 | 1914 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1915 | int result = 0; |
1916 | int64_t current_length = 0; | |
1917 | char *buf = NULL; | |
1918 | struct stat st; | |
1919 | int fd = aiocb->aio_fildes; | |
1920 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1921 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1922 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1923 | |
1924 | if (fstat(fd, &st) < 0) { | |
1925 | result = -errno; | |
1926 | error_setg_errno(errp, -result, "Could not stat file"); | |
1927 | return result; | |
1928 | } | |
1929 | ||
1930 | current_length = st.st_size; | |
d57c44d0 | 1931 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1932 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1933 | return -ENOTSUP; | |
1934 | } | |
1935 | ||
d57c44d0 | 1936 | switch (prealloc) { |
93f4e2ff KW |
1937 | #ifdef CONFIG_POSIX_FALLOCATE |
1938 | case PREALLOC_MODE_FALLOC: | |
1939 | /* | |
1940 | * Truncating before posix_fallocate() makes it about twice slower on | |
1941 | * file systems that do not support fallocate(), trying to check if a | |
1942 | * block is allocated before allocating it, so don't do that here. | |
1943 | */ | |
1944 | if (offset != current_length) { | |
1945 | result = -posix_fallocate(fd, current_length, | |
1946 | offset - current_length); | |
1947 | if (result != 0) { | |
1948 | /* posix_fallocate() doesn't set errno. */ | |
1949 | error_setg_errno(errp, -result, | |
1950 | "Could not preallocate new data"); | |
3a20013f NS |
1951 | } else if (current_length == 0) { |
1952 | /* | |
1953 | * posix_fallocate() uses fallocate() if the filesystem | |
1954 | * supports it, or fallback to manually writing zeroes. If | |
1955 | * fallocate() was used, unaligned reads from the fallocated | |
1956 | * area in raw_probe_alignment() will succeed, hence we need to | |
1957 | * allocate the first block. | |
1958 | * | |
1959 | * Optimize future alignment probing; ignore failures. | |
1960 | */ | |
1961 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1962 | } |
1963 | } else { | |
1964 | result = 0; | |
1965 | } | |
1966 | goto out; | |
1967 | #endif | |
1968 | case PREALLOC_MODE_FULL: | |
1969 | { | |
1970 | int64_t num = 0, left = offset - current_length; | |
1971 | off_t seek_result; | |
1972 | ||
1973 | /* | |
1974 | * Knowing the final size from the beginning could allow the file | |
1975 | * system driver to do less allocations and possibly avoid | |
1976 | * fragmentation of the file. | |
1977 | */ | |
1978 | if (ftruncate(fd, offset) != 0) { | |
1979 | result = -errno; | |
1980 | error_setg_errno(errp, -result, "Could not resize file"); | |
1981 | goto out; | |
1982 | } | |
1983 | ||
1984 | buf = g_malloc0(65536); | |
1985 | ||
1986 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1987 | if (seek_result < 0) { | |
1988 | result = -errno; | |
1989 | error_setg_errno(errp, -result, | |
1990 | "Failed to seek to the old end of file"); | |
1991 | goto out; | |
1992 | } | |
1993 | ||
1994 | while (left > 0) { | |
1995 | num = MIN(left, 65536); | |
1996 | result = write(fd, buf, num); | |
1997 | if (result < 0) { | |
a1c81f4f FZ |
1998 | if (errno == EINTR) { |
1999 | continue; | |
2000 | } | |
93f4e2ff KW |
2001 | result = -errno; |
2002 | error_setg_errno(errp, -result, | |
2003 | "Could not write zeros for preallocation"); | |
2004 | goto out; | |
2005 | } | |
2006 | left -= result; | |
2007 | } | |
2008 | if (result >= 0) { | |
2009 | result = fsync(fd); | |
2010 | if (result < 0) { | |
2011 | result = -errno; | |
2012 | error_setg_errno(errp, -result, | |
2013 | "Could not flush file to disk"); | |
2014 | goto out; | |
2015 | } | |
2016 | } | |
2017 | goto out; | |
2018 | } | |
2019 | case PREALLOC_MODE_OFF: | |
2020 | if (ftruncate(fd, offset) != 0) { | |
2021 | result = -errno; | |
2022 | error_setg_errno(errp, -result, "Could not resize file"); | |
3a20013f NS |
2023 | } else if (current_length == 0 && offset > current_length) { |
2024 | /* Optimize future alignment probing; ignore failures. */ | |
2025 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
2026 | } |
2027 | return result; | |
2028 | default: | |
2029 | result = -ENOTSUP; | |
2030 | error_setg(errp, "Unsupported preallocation mode: %s", | |
d57c44d0 | 2031 | PreallocMode_str(prealloc)); |
93f4e2ff KW |
2032 | return result; |
2033 | } | |
2034 | ||
2035 | out: | |
2036 | if (result < 0) { | |
2037 | if (ftruncate(fd, current_length) < 0) { | |
2038 | error_report("Failed to restore old file length: %s", | |
2039 | strerror(errno)); | |
2040 | } | |
2041 | } | |
2042 | ||
2043 | g_free(buf); | |
2044 | return result; | |
2045 | } | |
2046 | ||
5d5de250 KW |
2047 | static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, |
2048 | ThreadPoolFunc func, void *arg) | |
2049 | { | |
2050 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ | |
2051 | ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
2052 | return thread_pool_submit_co(pool, func, arg); | |
2053 | } | |
2054 | ||
a7c5f67a KB |
2055 | /* |
2056 | * Check if all memory in this vector is sector aligned. | |
2057 | */ | |
2058 | static bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
2059 | { | |
2060 | int i; | |
2061 | size_t alignment = bdrv_min_mem_align(bs); | |
25474d90 | 2062 | size_t len = bs->bl.request_alignment; |
a7c5f67a KB |
2063 | IO_CODE(); |
2064 | ||
2065 | for (i = 0; i < qiov->niov; i++) { | |
2066 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
2067 | return false; | |
2068 | } | |
25474d90 | 2069 | if (qiov->iov[i].iov_len % len) { |
a7c5f67a KB |
2070 | return false; |
2071 | } | |
2072 | } | |
2073 | ||
2074 | return true; | |
2075 | } | |
2076 | ||
9d52aa3c KW |
2077 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
2078 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 2079 | { |
ce1a14dc | 2080 | BDRVRawState *s = bs->opaque; |
999e6b69 | 2081 | RawPosixAIOData acb; |
ce1a14dc | 2082 | |
19cb3738 | 2083 | if (fd_open(bs) < 0) |
2174f12b | 2084 | return -EIO; |
19cb3738 | 2085 | |
f141eafe | 2086 | /* |
c6447510 AM |
2087 | * When using O_DIRECT, the request must be aligned to be able to use |
2088 | * either libaio or io_uring interface. If not fail back to regular thread | |
2089 | * pool read/write code which emulates this for us if we | |
2090 | * set QEMU_AIO_MISALIGNED. | |
f141eafe | 2091 | */ |
c6447510 AM |
2092 | if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { |
2093 | type |= QEMU_AIO_MISALIGNED; | |
2094 | #ifdef CONFIG_LINUX_IO_URING | |
2095 | } else if (s->use_linux_io_uring) { | |
2096 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2097 | assert(qiov->size == bytes); | |
2098 | return luring_co_submit(bs, aio, s->fd, offset, qiov, type); | |
2099 | #endif | |
e44bd6fc | 2100 | #ifdef CONFIG_LINUX_AIO |
c6447510 AM |
2101 | } else if (s->use_linux_aio) { |
2102 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); | |
2103 | assert(qiov->size == bytes); | |
512da211 SG |
2104 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type, |
2105 | s->aio_max_batch); | |
e44bd6fc | 2106 | #endif |
9ef91a67 | 2107 | } |
f141eafe | 2108 | |
999e6b69 KW |
2109 | acb = (RawPosixAIOData) { |
2110 | .bs = bs, | |
2111 | .aio_fildes = s->fd, | |
2112 | .aio_type = type, | |
2113 | .aio_offset = offset, | |
2114 | .aio_nbytes = bytes, | |
2115 | .io = { | |
2116 | .iov = qiov->iov, | |
2117 | .niov = qiov->niov, | |
2118 | }, | |
2119 | }; | |
2120 | ||
2121 | assert(qiov->size == bytes); | |
2122 | return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); | |
2174f12b KW |
2123 | } |
2124 | ||
f7ef38dd VSO |
2125 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset, |
2126 | int64_t bytes, QEMUIOVector *qiov, | |
2127 | BdrvRequestFlags flags) | |
2174f12b | 2128 | { |
9d52aa3c | 2129 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
2130 | } |
2131 | ||
e75abeda VSO |
2132 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset, |
2133 | int64_t bytes, QEMUIOVector *qiov, | |
2134 | BdrvRequestFlags flags) | |
2174f12b | 2135 | { |
9d52aa3c KW |
2136 | assert(flags == 0); |
2137 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
2138 | } |
2139 | ||
1b3abdcc ML |
2140 | static void raw_aio_plug(BlockDriverState *bs) |
2141 | { | |
c6447510 | 2142 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2143 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2144 | if (s->use_linux_aio) { |
0187f5c9 PB |
2145 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
2146 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
2147 | } |
2148 | #endif | |
c6447510 AM |
2149 | #ifdef CONFIG_LINUX_IO_URING |
2150 | if (s->use_linux_io_uring) { | |
2151 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2152 | luring_io_plug(bs, aio); | |
2153 | } | |
2154 | #endif | |
1b3abdcc ML |
2155 | } |
2156 | ||
2157 | static void raw_aio_unplug(BlockDriverState *bs) | |
2158 | { | |
c6447510 | 2159 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2160 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2161 | if (s->use_linux_aio) { |
0187f5c9 | 2162 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
68d79466 | 2163 | laio_io_unplug(bs, aio, s->aio_max_batch); |
1b3abdcc ML |
2164 | } |
2165 | #endif | |
c6447510 AM |
2166 | #ifdef CONFIG_LINUX_IO_URING |
2167 | if (s->use_linux_io_uring) { | |
2168 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2169 | luring_io_unplug(bs, aio); | |
2170 | } | |
2171 | #endif | |
1b3abdcc ML |
2172 | } |
2173 | ||
dda56b75 | 2174 | static int coroutine_fn raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
2175 | { |
2176 | BDRVRawState *s = bs->opaque; | |
06dc9bd5 | 2177 | RawPosixAIOData acb; |
33d70fb6 | 2178 | int ret; |
b2e12bc6 | 2179 | |
33d70fb6 KW |
2180 | ret = fd_open(bs); |
2181 | if (ret < 0) { | |
2182 | return ret; | |
2183 | } | |
b2e12bc6 | 2184 | |
06dc9bd5 KW |
2185 | acb = (RawPosixAIOData) { |
2186 | .bs = bs, | |
2187 | .aio_fildes = s->fd, | |
2188 | .aio_type = QEMU_AIO_FLUSH, | |
2189 | }; | |
2190 | ||
c6447510 AM |
2191 | #ifdef CONFIG_LINUX_IO_URING |
2192 | if (s->use_linux_io_uring) { | |
2193 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2194 | return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); | |
2195 | } | |
2196 | #endif | |
06dc9bd5 | 2197 | return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); |
b2e12bc6 CH |
2198 | } |
2199 | ||
ed6e2161 NA |
2200 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
2201 | AioContext *new_context) | |
2202 | { | |
c6447510 | 2203 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
ed6e2161 | 2204 | #ifdef CONFIG_LINUX_AIO |
ed6e2161 | 2205 | if (s->use_linux_aio) { |
cb09104e | 2206 | Error *local_err = NULL; |
ed6e2161 NA |
2207 | if (!aio_setup_linux_aio(new_context, &local_err)) { |
2208 | error_reportf_err(local_err, "Unable to use native AIO, " | |
2209 | "falling back to thread pool: "); | |
2210 | s->use_linux_aio = false; | |
2211 | } | |
2212 | } | |
2213 | #endif | |
c6447510 AM |
2214 | #ifdef CONFIG_LINUX_IO_URING |
2215 | if (s->use_linux_io_uring) { | |
cb8d0851 | 2216 | Error *local_err = NULL; |
c6447510 AM |
2217 | if (!aio_setup_linux_io_uring(new_context, &local_err)) { |
2218 | error_reportf_err(local_err, "Unable to use linux io_uring, " | |
2219 | "falling back to thread pool: "); | |
2220 | s->use_linux_io_uring = false; | |
2221 | } | |
2222 | } | |
2223 | #endif | |
ed6e2161 NA |
2224 | } |
2225 | ||
83f64091 FB |
2226 | static void raw_close(BlockDriverState *bs) |
2227 | { | |
2228 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 2229 | |
19cb3738 | 2230 | if (s->fd >= 0) { |
2e1e79da | 2231 | qemu_close(s->fd); |
19cb3738 FB |
2232 | s->fd = -1; |
2233 | } | |
83f64091 FB |
2234 | } |
2235 | ||
d0bc9e5d HR |
2236 | /** |
2237 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
2238 | * new space according to @prealloc. | |
2239 | * | |
2240 | * Returns: 0 on success, -errno on failure. | |
2241 | */ | |
93f4e2ff KW |
2242 | static int coroutine_fn |
2243 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
2244 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 2245 | { |
29cb4c01 | 2246 | RawPosixAIOData acb; |
d0bc9e5d | 2247 | |
29cb4c01 | 2248 | acb = (RawPosixAIOData) { |
93f4e2ff KW |
2249 | .bs = bs, |
2250 | .aio_fildes = fd, | |
2251 | .aio_type = QEMU_AIO_TRUNCATE, | |
2252 | .aio_offset = offset, | |
d57c44d0 KW |
2253 | .truncate = { |
2254 | .prealloc = prealloc, | |
2255 | .errp = errp, | |
2256 | }, | |
93f4e2ff | 2257 | }; |
d0bc9e5d | 2258 | |
29cb4c01 | 2259 | return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); |
9f63b07e HR |
2260 | } |
2261 | ||
061ca8a3 | 2262 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
c80d8b06 | 2263 | bool exact, PreallocMode prealloc, |
92b92799 | 2264 | BdrvRequestFlags flags, Error **errp) |
83f64091 FB |
2265 | { |
2266 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 2267 | struct stat st; |
f59adb32 | 2268 | int ret; |
55b949c8 CH |
2269 | |
2270 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
2271 | ret = -errno; |
2272 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
2273 | return ret; | |
55b949c8 CH |
2274 | } |
2275 | ||
2276 | if (S_ISREG(st.st_mode)) { | |
82325ae5 | 2277 | /* Always resizes to the exact @offset */ |
93f4e2ff | 2278 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
2279 | } |
2280 | ||
2281 | if (prealloc != PREALLOC_MODE_OFF) { | |
2282 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 2283 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
2284 | return -ENOTSUP; |
2285 | } | |
2286 | ||
2287 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
82325ae5 HR |
2288 | int64_t cur_length = raw_getlength(bs); |
2289 | ||
2290 | if (offset != cur_length && exact) { | |
2291 | error_setg(errp, "Cannot resize device files"); | |
2292 | return -ENOTSUP; | |
2293 | } else if (offset > cur_length) { | |
f59adb32 HR |
2294 | error_setg(errp, "Cannot grow device files"); |
2295 | return -EINVAL; | |
2296 | } | |
55b949c8 | 2297 | } else { |
f59adb32 | 2298 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
2299 | return -ENOTSUP; |
2300 | } | |
2301 | ||
83f64091 FB |
2302 | return 0; |
2303 | } | |
2304 | ||
128ab2ff BS |
2305 | #ifdef __OpenBSD__ |
2306 | static int64_t raw_getlength(BlockDriverState *bs) | |
2307 | { | |
2308 | BDRVRawState *s = bs->opaque; | |
2309 | int fd = s->fd; | |
2310 | struct stat st; | |
2311 | ||
2312 | if (fstat(fd, &st)) | |
aa729704 | 2313 | return -errno; |
128ab2ff BS |
2314 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2315 | struct disklabel dl; | |
2316 | ||
2317 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2318 | return -errno; |
128ab2ff BS |
2319 | return (uint64_t)dl.d_secsize * |
2320 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2321 | } else | |
2322 | return st.st_size; | |
2323 | } | |
d1f6fd8d CE |
2324 | #elif defined(__NetBSD__) |
2325 | static int64_t raw_getlength(BlockDriverState *bs) | |
2326 | { | |
2327 | BDRVRawState *s = bs->opaque; | |
2328 | int fd = s->fd; | |
2329 | struct stat st; | |
2330 | ||
2331 | if (fstat(fd, &st)) | |
aa729704 | 2332 | return -errno; |
d1f6fd8d CE |
2333 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2334 | struct dkwedge_info dkw; | |
2335 | ||
2336 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
2337 | return dkw.dkw_size * 512; | |
2338 | } else { | |
2339 | struct disklabel dl; | |
2340 | ||
2341 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2342 | return -errno; |
d1f6fd8d CE |
2343 | return (uint64_t)dl.d_secsize * |
2344 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2345 | } | |
2346 | } else | |
2347 | return st.st_size; | |
2348 | } | |
50779cc2 CH |
2349 | #elif defined(__sun__) |
2350 | static int64_t raw_getlength(BlockDriverState *bs) | |
2351 | { | |
2352 | BDRVRawState *s = bs->opaque; | |
2353 | struct dk_minfo minfo; | |
2354 | int ret; | |
aa729704 | 2355 | int64_t size; |
50779cc2 CH |
2356 | |
2357 | ret = fd_open(bs); | |
2358 | if (ret < 0) { | |
2359 | return ret; | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
2364 | */ | |
2365 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
2366 | if (ret != -1) { | |
2367 | return minfo.dki_lbsize * minfo.dki_capacity; | |
2368 | } | |
2369 | ||
2370 | /* | |
2371 | * There are reports that lseek on some devices fails, but | |
2372 | * irc discussion said that contingency on contingency was overkill. | |
2373 | */ | |
aa729704 MA |
2374 | size = lseek(s->fd, 0, SEEK_END); |
2375 | if (size < 0) { | |
2376 | return -errno; | |
2377 | } | |
2378 | return size; | |
50779cc2 CH |
2379 | } |
2380 | #elif defined(CONFIG_BSD) | |
2381 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2382 | { |
2383 | BDRVRawState *s = bs->opaque; | |
2384 | int fd = s->fd; | |
2385 | int64_t size; | |
83f64091 | 2386 | struct stat sb; |
a167ba50 | 2387 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2388 | int reopened = 0; |
83f64091 | 2389 | #endif |
19cb3738 FB |
2390 | int ret; |
2391 | ||
2392 | ret = fd_open(bs); | |
2393 | if (ret < 0) | |
2394 | return ret; | |
83f64091 | 2395 | |
a167ba50 | 2396 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2397 | again: |
2398 | #endif | |
83f64091 | 2399 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
267cd53f | 2400 | size = 0; |
83f64091 | 2401 | #ifdef DIOCGMEDIASIZE |
267cd53f PB |
2402 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) { |
2403 | size = 0; | |
2404 | } | |
2405 | #endif | |
2406 | #ifdef DIOCGPART | |
2407 | if (size == 0) { | |
2408 | struct partinfo pi; | |
2409 | if (ioctl(fd, DIOCGPART, &pi) == 0) { | |
2410 | size = pi.media_size; | |
2411 | } | |
c5e97233 | 2412 | } |
83f64091 | 2413 | #endif |
09e20abd | 2414 | #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE) |
267cd53f | 2415 | if (size == 0) { |
728dacbd JA |
2416 | uint64_t sectors = 0; |
2417 | uint32_t sector_size = 0; | |
2418 | ||
2419 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2420 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2421 | size = sectors * sector_size; | |
728dacbd JA |
2422 | } |
2423 | } | |
267cd53f PB |
2424 | #endif |
2425 | if (size == 0) { | |
2426 | size = lseek(fd, 0LL, SEEK_END); | |
2427 | } | |
aa729704 MA |
2428 | if (size < 0) { |
2429 | return -errno; | |
2430 | } | |
a167ba50 | 2431 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2432 | switch(s->type) { |
2433 | case FTYPE_CD: | |
2434 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2435 | if (size == 2048LL * (unsigned)-1) | |
2436 | size = 0; | |
2437 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2438 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2439 | reopened = 1; |
2440 | goto again; | |
2441 | } | |
2442 | } | |
83f64091 | 2443 | #endif |
50779cc2 | 2444 | } else { |
83f64091 | 2445 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2446 | if (size < 0) { |
2447 | return -errno; | |
2448 | } | |
83f64091 | 2449 | } |
83f64091 FB |
2450 | return size; |
2451 | } | |
50779cc2 CH |
2452 | #else |
2453 | static int64_t raw_getlength(BlockDriverState *bs) | |
2454 | { | |
2455 | BDRVRawState *s = bs->opaque; | |
2456 | int ret; | |
aa729704 | 2457 | int64_t size; |
50779cc2 CH |
2458 | |
2459 | ret = fd_open(bs); | |
2460 | if (ret < 0) { | |
2461 | return ret; | |
2462 | } | |
2463 | ||
aa729704 MA |
2464 | size = lseek(s->fd, 0, SEEK_END); |
2465 | if (size < 0) { | |
2466 | return -errno; | |
2467 | } | |
2468 | return size; | |
50779cc2 | 2469 | } |
128ab2ff | 2470 | #endif |
83f64091 | 2471 | |
4a1d5e1f FZ |
2472 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2473 | { | |
2474 | struct stat st; | |
2475 | BDRVRawState *s = bs->opaque; | |
2476 | ||
2477 | if (fstat(s->fd, &st) < 0) { | |
2478 | return -errno; | |
2479 | } | |
2480 | return (int64_t)st.st_blocks * 512; | |
2481 | } | |
2482 | ||
93f4e2ff KW |
2483 | static int coroutine_fn |
2484 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2485 | { |
927f11e1 | 2486 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2487 | Error *local_err = NULL; |
83f64091 | 2488 | int fd; |
d815efca | 2489 | uint64_t perm, shared; |
1e37d059 | 2490 | int result = 0; |
83f64091 | 2491 | |
927f11e1 KW |
2492 | /* Validate options and set default values */ |
2493 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2494 | file_opts = &options->u.file; | |
464d9f64 | 2495 | |
927f11e1 KW |
2496 | if (!file_opts->has_nocow) { |
2497 | file_opts->nocow = false; | |
2498 | } | |
2499 | if (!file_opts->has_preallocation) { | |
2500 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2501 | } |
ffa244c8 KW |
2502 | if (!file_opts->has_extent_size_hint) { |
2503 | file_opts->extent_size_hint = 1 * MiB; | |
2504 | } | |
2505 | if (file_opts->extent_size_hint > UINT32_MAX) { | |
2506 | result = -EINVAL; | |
2507 | error_setg(errp, "Extent size hint is too large"); | |
2508 | goto out; | |
2509 | } | |
83f64091 | 2510 | |
927f11e1 | 2511 | /* Create file */ |
b18a24a9 | 2512 | fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); |
1e37d059 SW |
2513 | if (fd < 0) { |
2514 | result = -errno; | |
06247428 HT |
2515 | goto out; |
2516 | } | |
2517 | ||
b8cf1913 HR |
2518 | /* Take permissions: We want to discard everything, so we need |
2519 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2520 | * BLK_PERM_RESIZE. | |
2521 | * On the other hand, we cannot share the RESIZE permission | |
2522 | * because we promise that after this function, the file has the | |
2523 | * size given in the options. If someone else were to resize it | |
2524 | * concurrently, we could not guarantee that. | |
2525 | * Note that after this function, we can no longer guarantee that | |
2526 | * the file is not touched by a third party, so it may be resized | |
2527 | * then. */ | |
2528 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2529 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2530 | ||
2531 | /* Step one: Take locks */ | |
2996ffad | 2532 | result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2533 | if (result < 0) { |
2534 | goto out_close; | |
2535 | } | |
2536 | ||
2537 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2538 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2539 | if (result < 0) { | |
b857431d FZ |
2540 | error_append_hint(errp, |
2541 | "Is another process using the image [%s]?\n", | |
2542 | file_opts->filename); | |
7c20c808 | 2543 | goto out_unlock; |
b8cf1913 HR |
2544 | } |
2545 | ||
2546 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2547 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2548 | if (result < 0) { |
7c20c808 | 2549 | goto out_unlock; |
b8cf1913 HR |
2550 | } |
2551 | ||
927f11e1 | 2552 | if (file_opts->nocow) { |
4ab15590 | 2553 | #ifdef __linux__ |
06247428 HT |
2554 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2555 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2556 | * will be ignored since any failure of this operation should not | |
2557 | * block the left work. | |
2558 | */ | |
2559 | int attr; | |
2560 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2561 | attr |= FS_NOCOW_FL; | |
2562 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2563 | } |
06247428 HT |
2564 | #endif |
2565 | } | |
ffa244c8 KW |
2566 | #ifdef FS_IOC_FSSETXATTR |
2567 | /* | |
2568 | * Try to set the extent size hint. Failure is not fatal, and a warning is | |
2569 | * only printed if the option was explicitly specified. | |
2570 | */ | |
2571 | { | |
2572 | struct fsxattr attr; | |
2573 | result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); | |
2574 | if (result == 0) { | |
2575 | attr.fsx_xflags |= FS_XFLAG_EXTSIZE; | |
2576 | attr.fsx_extsize = file_opts->extent_size_hint; | |
2577 | result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); | |
2578 | } | |
2579 | if (result < 0 && file_opts->has_extent_size_hint && | |
2580 | file_opts->extent_size_hint) | |
2581 | { | |
2582 | warn_report("Failed to set extent size hint: %s", | |
2583 | strerror(errno)); | |
2584 | } | |
2585 | } | |
2586 | #endif | |
06247428 | 2587 | |
b8cf1913 HR |
2588 | /* Resize and potentially preallocate the file to the desired |
2589 | * final size */ | |
93f4e2ff KW |
2590 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2591 | file_opts->preallocation, errp); | |
9f63b07e | 2592 | if (result < 0) { |
7c20c808 HR |
2593 | goto out_unlock; |
2594 | } | |
2595 | ||
2596 | out_unlock: | |
2996ffad | 2597 | raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); |
7c20c808 HR |
2598 | if (local_err) { |
2599 | /* The above call should not fail, and if it does, that does | |
2600 | * not mean the whole creation operation has failed. So | |
2601 | * report it the user for their convenience, but do not report | |
2602 | * it to the caller. */ | |
db0754df | 2603 | warn_report_err(local_err); |
1e37d059 | 2604 | } |
06247428 | 2605 | |
5a1dad9d | 2606 | out_close: |
06247428 HT |
2607 | if (qemu_close(fd) != 0 && result == 0) { |
2608 | result = -errno; | |
2609 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2610 | } | |
2611 | out: | |
1e37d059 | 2612 | return result; |
83f64091 FB |
2613 | } |
2614 | ||
b92902df ML |
2615 | static int coroutine_fn raw_co_create_opts(BlockDriver *drv, |
2616 | const char *filename, | |
2617 | QemuOpts *opts, | |
927f11e1 KW |
2618 | Error **errp) |
2619 | { | |
2620 | BlockdevCreateOptions options; | |
2621 | int64_t total_size = 0; | |
ffa244c8 KW |
2622 | int64_t extent_size_hint = 0; |
2623 | bool has_extent_size_hint = false; | |
927f11e1 KW |
2624 | bool nocow = false; |
2625 | PreallocMode prealloc; | |
2626 | char *buf = NULL; | |
2627 | Error *local_err = NULL; | |
2628 | ||
2629 | /* Skip file: protocol prefix */ | |
2630 | strstart(filename, "file:", &filename); | |
2631 | ||
2632 | /* Read out options */ | |
2633 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2634 | BDRV_SECTOR_SIZE); | |
ffa244c8 KW |
2635 | if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { |
2636 | has_extent_size_hint = true; | |
2637 | extent_size_hint = | |
2638 | qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); | |
2639 | } | |
927f11e1 KW |
2640 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); |
2641 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2642 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2643 | PREALLOC_MODE_OFF, &local_err); | |
2644 | g_free(buf); | |
2645 | if (local_err) { | |
2646 | error_propagate(errp, local_err); | |
2647 | return -EINVAL; | |
2648 | } | |
2649 | ||
2650 | options = (BlockdevCreateOptions) { | |
2651 | .driver = BLOCKDEV_DRIVER_FILE, | |
2652 | .u.file = { | |
2653 | .filename = (char *) filename, | |
2654 | .size = total_size, | |
2655 | .has_preallocation = true, | |
2656 | .preallocation = prealloc, | |
2657 | .has_nocow = true, | |
2658 | .nocow = nocow, | |
ffa244c8 KW |
2659 | .has_extent_size_hint = has_extent_size_hint, |
2660 | .extent_size_hint = extent_size_hint, | |
927f11e1 KW |
2661 | }, |
2662 | }; | |
2663 | return raw_co_create(&options, errp); | |
2664 | } | |
2665 | ||
9bffae14 DHB |
2666 | static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, |
2667 | Error **errp) | |
2668 | { | |
2669 | struct stat st; | |
2670 | int ret; | |
2671 | ||
2672 | if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { | |
2673 | error_setg_errno(errp, ENOENT, "%s is not a regular file", | |
2674 | bs->filename); | |
2675 | return -ENOENT; | |
2676 | } | |
2677 | ||
2678 | ret = unlink(bs->filename); | |
2679 | if (ret < 0) { | |
2680 | ret = -errno; | |
2681 | error_setg_errno(errp, -ret, "Error when deleting file %s", | |
2682 | bs->filename); | |
2683 | } | |
2684 | ||
2685 | return ret; | |
2686 | } | |
2687 | ||
d1f06fe6 MA |
2688 | /* |
2689 | * Find allocation range in @bs around offset @start. | |
2690 | * May change underlying file descriptor's file offset. | |
2691 | * If @start is not in a hole, store @start in @data, and the | |
2692 | * beginning of the next hole in @hole, and return 0. | |
2693 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2694 | * beginning of the next non-hole in @data, and return 0. | |
2695 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2696 | * If we can't find out, return a negative errno other than -ENXIO. | |
2697 | */ | |
2698 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2699 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2700 | { |
2701 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2702 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2703 | off_t offs; |
94282e71 | 2704 | |
d1f06fe6 MA |
2705 | /* |
2706 | * SEEK_DATA cases: | |
2707 | * D1. offs == start: start is in data | |
2708 | * D2. offs > start: start is in a hole, next data at offs | |
2709 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2710 | * or start is beyond EOF | |
2711 | * If the latter happens, the file has been truncated behind | |
2712 | * our back since we opened it. All bets are off then. | |
2713 | * Treating like a trailing hole is simplest. | |
2714 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2715 | */ | |
2716 | offs = lseek(s->fd, start, SEEK_DATA); | |
2717 | if (offs < 0) { | |
2718 | return -errno; /* D3 or D4 */ | |
2719 | } | |
a03083a0 JC |
2720 | |
2721 | if (offs < start) { | |
2722 | /* This is not a valid return by lseek(). We are safe to just return | |
2723 | * -EIO in this case, and we'll treat it like D4. */ | |
2724 | return -EIO; | |
2725 | } | |
d1f06fe6 MA |
2726 | |
2727 | if (offs > start) { | |
2728 | /* D2: in hole, next data at offs */ | |
2729 | *hole = start; | |
2730 | *data = offs; | |
2731 | return 0; | |
5500316d PB |
2732 | } |
2733 | ||
d1f06fe6 MA |
2734 | /* D1: in data, end not yet known */ |
2735 | ||
2736 | /* | |
2737 | * SEEK_HOLE cases: | |
2738 | * H1. offs == start: start is in a hole | |
2739 | * If this happens here, a hole has been dug behind our back | |
2740 | * since the previous lseek(). | |
2741 | * H2. offs > start: either start is in data, next hole at offs, | |
2742 | * or start is in trailing hole, EOF at offs | |
2743 | * Linux treats trailing holes like any other hole: offs == | |
2744 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2745 | * If that happens here, a hole has been dug behind our back | |
2746 | * since the previous lseek(). | |
2747 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2748 | * If this happens, the file has been truncated behind our | |
2749 | * back since we opened it. Treat it like a trailing hole. | |
2750 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2751 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2752 | */ | |
2753 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2754 | if (offs < 0) { | |
2755 | return -errno; /* D1 and (H3 or H4) */ | |
2756 | } | |
a03083a0 JC |
2757 | |
2758 | if (offs < start) { | |
2759 | /* This is not a valid return by lseek(). We are safe to just return | |
2760 | * -EIO in this case, and we'll treat it like H4. */ | |
2761 | return -EIO; | |
2762 | } | |
d1f06fe6 MA |
2763 | |
2764 | if (offs > start) { | |
2765 | /* | |
2766 | * D1 and H2: either in data, next hole at offs, or it was in | |
2767 | * data but is now in a trailing hole. In the latter case, | |
2768 | * all bets are off. Treating it as if it there was data all | |
2769 | * the way to EOF is safe, so simply do that. | |
2770 | */ | |
4f11aa8a | 2771 | *data = start; |
d1f06fe6 MA |
2772 | *hole = offs; |
2773 | return 0; | |
5500316d | 2774 | } |
4f11aa8a | 2775 | |
d1f06fe6 MA |
2776 | /* D1 and H1 */ |
2777 | return -EBUSY; | |
5500316d | 2778 | #else |
4f11aa8a | 2779 | return -ENOTSUP; |
5500316d | 2780 | #endif |
4f11aa8a HR |
2781 | } |
2782 | ||
2783 | /* | |
a290f085 | 2784 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2785 | * |
a290f085 | 2786 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2787 | * |
a290f085 EB |
2788 | * 'pnum' is set to the number of bytes (including and immediately following |
2789 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2790 | * allocated/unallocated state. |
2791 | * | |
869e7ee8 HR |
2792 | * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may |
2793 | * well exceed it. | |
4f11aa8a | 2794 | */ |
a290f085 EB |
2795 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2796 | bool want_zero, | |
2797 | int64_t offset, | |
2798 | int64_t bytes, int64_t *pnum, | |
2799 | int64_t *map, | |
2800 | BlockDriverState **file) | |
2801 | { | |
2802 | off_t data = 0, hole = 0; | |
d7f62751 | 2803 | int ret; |
4f11aa8a | 2804 | |
9c3db310 HR |
2805 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); |
2806 | ||
4f11aa8a HR |
2807 | ret = fd_open(bs); |
2808 | if (ret < 0) { | |
2809 | return ret; | |
2810 | } | |
2811 | ||
a290f085 EB |
2812 | if (!want_zero) { |
2813 | *pnum = bytes; | |
2814 | *map = offset; | |
2815 | *file = bs; | |
2816 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2817 | } |
4f11aa8a | 2818 | |
a290f085 | 2819 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2820 | if (ret == -ENXIO) { |
2821 | /* Trailing hole */ | |
a290f085 | 2822 | *pnum = bytes; |
d1f06fe6 MA |
2823 | ret = BDRV_BLOCK_ZERO; |
2824 | } else if (ret < 0) { | |
2825 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2826 | *pnum = bytes; |
d1f06fe6 | 2827 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2828 | } else if (data == offset) { |
2829 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2830 | * possibly including a partial sector at EOF. */ |
869e7ee8 | 2831 | *pnum = hole - offset; |
9c3db310 HR |
2832 | |
2833 | /* | |
2834 | * We are not allowed to return partial sectors, though, so | |
2835 | * round up if necessary. | |
2836 | */ | |
2837 | if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { | |
2838 | int64_t file_length = raw_getlength(bs); | |
2839 | if (file_length > 0) { | |
2840 | /* Ignore errors, this is just a safeguard */ | |
2841 | assert(hole == file_length); | |
2842 | } | |
2843 | *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); | |
2844 | } | |
2845 | ||
d1f06fe6 | 2846 | ret = BDRV_BLOCK_DATA; |
5500316d | 2847 | } else { |
a290f085 EB |
2848 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2849 | assert(hole == offset); | |
869e7ee8 | 2850 | *pnum = data - offset; |
d1f06fe6 | 2851 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2852 | } |
a290f085 | 2853 | *map = offset; |
02650acb | 2854 | *file = bs; |
a290f085 | 2855 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2856 | } |
2857 | ||
31be8a2a SH |
2858 | #if defined(__linux__) |
2859 | /* Verify that the file is not in the page cache */ | |
2860 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2861 | { | |
2862 | const size_t window_size = 128 * 1024 * 1024; | |
2863 | BDRVRawState *s = bs->opaque; | |
2864 | void *window = NULL; | |
2865 | size_t length = 0; | |
2866 | unsigned char *vec; | |
2867 | size_t page_size; | |
2868 | off_t offset; | |
2869 | off_t end; | |
2870 | ||
2871 | /* mincore(2) page status information requires 1 byte per page */ | |
2872 | page_size = sysconf(_SC_PAGESIZE); | |
2873 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2874 | ||
2875 | end = raw_getlength(bs); | |
2876 | ||
2877 | for (offset = 0; offset < end; offset += window_size) { | |
2878 | void *new_window; | |
2879 | size_t new_length; | |
2880 | size_t vec_end; | |
2881 | size_t i; | |
2882 | int ret; | |
2883 | ||
2884 | /* Unmap previous window if size has changed */ | |
2885 | new_length = MIN(end - offset, window_size); | |
2886 | if (new_length != length) { | |
2887 | munmap(window, length); | |
2888 | window = NULL; | |
2889 | length = 0; | |
2890 | } | |
2891 | ||
2892 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2893 | s->fd, offset); | |
2894 | if (new_window == MAP_FAILED) { | |
2895 | error_setg_errno(errp, errno, "mmap failed"); | |
2896 | break; | |
2897 | } | |
2898 | ||
2899 | window = new_window; | |
2900 | length = new_length; | |
2901 | ||
2902 | ret = mincore(window, length, vec); | |
2903 | if (ret < 0) { | |
2904 | error_setg_errno(errp, errno, "mincore failed"); | |
2905 | break; | |
2906 | } | |
2907 | ||
2908 | vec_end = DIV_ROUND_UP(length, page_size); | |
2909 | for (i = 0; i < vec_end; i++) { | |
2910 | if (vec[i] & 0x1) { | |
31be8a2a SH |
2911 | break; |
2912 | } | |
2913 | } | |
77ed971b MA |
2914 | if (i < vec_end) { |
2915 | error_setg(errp, "page cache still in use!"); | |
2916 | break; | |
2917 | } | |
31be8a2a SH |
2918 | } |
2919 | ||
2920 | if (window) { | |
2921 | munmap(window, length); | |
2922 | } | |
2923 | ||
2924 | g_free(vec); | |
2925 | } | |
2926 | #endif /* __linux__ */ | |
2927 | ||
dd577a26 SH |
2928 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2929 | Error **errp) | |
2930 | { | |
2931 | BDRVRawState *s = bs->opaque; | |
2932 | int ret; | |
2933 | ||
2934 | ret = fd_open(bs); | |
2935 | if (ret < 0) { | |
2936 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2937 | return; | |
2938 | } | |
2939 | ||
f357fcd8 SH |
2940 | if (!s->drop_cache) { |
2941 | return; | |
2942 | } | |
2943 | ||
dd577a26 SH |
2944 | if (s->open_flags & O_DIRECT) { |
2945 | return; /* No host kernel page cache */ | |
2946 | } | |
2947 | ||
2948 | #if defined(__linux__) | |
2949 | /* This sets the scene for the next syscall... */ | |
2950 | ret = bdrv_co_flush(bs); | |
2951 | if (ret < 0) { | |
2952 | error_setg_errno(errp, -ret, "flush failed"); | |
2953 | return; | |
2954 | } | |
2955 | ||
2956 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2957 | * process. These limitations are okay because we just fsynced the file, | |
2958 | * we don't use mmap, and the file should not be in use by other processes. | |
2959 | */ | |
2960 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2961 | if (ret != 0) { /* the return value is a positive errno */ | |
2962 | error_setg_errno(errp, ret, "fadvise failed"); | |
2963 | return; | |
2964 | } | |
31be8a2a SH |
2965 | |
2966 | if (s->check_cache_dropped) { | |
2967 | check_cache_dropped(bs, errp); | |
2968 | } | |
dd577a26 SH |
2969 | #else /* __linux__ */ |
2970 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2971 | * unsupported on other host operating systems. Cache consistency issues | |
2972 | * may occur but no error is reported here, partly because that's the | |
2973 | * historical behavior and partly because it's hard to differentiate valid | |
2974 | * configurations that should not cause errors. | |
2975 | */ | |
2976 | #endif /* !__linux__ */ | |
2977 | } | |
2978 | ||
1c450366 AN |
2979 | static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) |
2980 | { | |
2981 | if (ret) { | |
2982 | s->stats.discard_nb_failed++; | |
2983 | } else { | |
2984 | s->stats.discard_nb_ok++; | |
2985 | s->stats.discard_bytes_ok += nbytes; | |
2986 | } | |
2987 | } | |
2988 | ||
33d70fb6 | 2989 | static coroutine_fn int |
0c802287 VSO |
2990 | raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2991 | bool blkdev) | |
dce512de | 2992 | { |
dce512de | 2993 | BDRVRawState *s = bs->opaque; |
46ee0f46 | 2994 | RawPosixAIOData acb; |
1c450366 | 2995 | int ret; |
46ee0f46 KW |
2996 | |
2997 | acb = (RawPosixAIOData) { | |
2998 | .bs = bs, | |
2999 | .aio_fildes = s->fd, | |
3000 | .aio_type = QEMU_AIO_DISCARD, | |
3001 | .aio_offset = offset, | |
3002 | .aio_nbytes = bytes, | |
3003 | }; | |
dce512de | 3004 | |
46ee0f46 KW |
3005 | if (blkdev) { |
3006 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
3007 | } | |
3008 | ||
1c450366 AN |
3009 | ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); |
3010 | raw_account_discard(s, bytes, ret); | |
3011 | return ret; | |
46ee0f46 KW |
3012 | } |
3013 | ||
3014 | static coroutine_fn int | |
0c802287 | 3015 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
46ee0f46 KW |
3016 | { |
3017 | return raw_do_pdiscard(bs, offset, bytes, false); | |
dce512de | 3018 | } |
0e7e1989 | 3019 | |
7154d8ae | 3020 | static int coroutine_fn |
f34b2bcf | 3021 | raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, |
7154d8ae | 3022 | BdrvRequestFlags flags, bool blkdev) |
260a82e5 PB |
3023 | { |
3024 | BDRVRawState *s = bs->opaque; | |
7154d8ae KW |
3025 | RawPosixAIOData acb; |
3026 | ThreadPoolFunc *handler; | |
3027 | ||
292d06b9 HR |
3028 | #ifdef CONFIG_FALLOCATE |
3029 | if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { | |
3030 | BdrvTrackedRequest *req; | |
292d06b9 HR |
3031 | |
3032 | /* | |
3033 | * This is a workaround for a bug in the Linux XFS driver, | |
3034 | * where writes submitted through the AIO interface will be | |
3035 | * discarded if they happen beyond a concurrently running | |
3036 | * fallocate() that increases the file length (i.e., both the | |
3037 | * write and the fallocate() happen beyond the EOF). | |
3038 | * | |
3039 | * To work around it, we extend the tracked request for this | |
3040 | * zero write until INT64_MAX (effectively infinity), and mark | |
3041 | * it as serializing. | |
3042 | * | |
3043 | * We have to enable this workaround for all filesystems and | |
3044 | * AIO modes (not just XFS with aio=native), because for | |
3045 | * remote filesystems we do not know the host configuration. | |
3046 | */ | |
3047 | ||
3048 | req = bdrv_co_get_self_request(bs); | |
3049 | assert(req); | |
3050 | assert(req->type == BDRV_TRACKED_WRITE); | |
3051 | assert(req->offset <= offset); | |
3052 | assert(req->offset + req->bytes >= offset + bytes); | |
3053 | ||
8b117001 VSO |
3054 | req->bytes = BDRV_MAX_LENGTH - req->offset; |
3055 | ||
69b55e03 | 3056 | bdrv_check_request(req->offset, req->bytes, &error_abort); |
292d06b9 | 3057 | |
8ac5aab2 | 3058 | bdrv_make_request_serialising(req, bs->bl.request_alignment); |
292d06b9 HR |
3059 | } |
3060 | #endif | |
3061 | ||
7154d8ae KW |
3062 | acb = (RawPosixAIOData) { |
3063 | .bs = bs, | |
3064 | .aio_fildes = s->fd, | |
3065 | .aio_type = QEMU_AIO_WRITE_ZEROES, | |
3066 | .aio_offset = offset, | |
3067 | .aio_nbytes = bytes, | |
3068 | }; | |
3069 | ||
3070 | if (blkdev) { | |
3071 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
3072 | } | |
738301e1 KW |
3073 | if (flags & BDRV_REQ_NO_FALLBACK) { |
3074 | acb.aio_type |= QEMU_AIO_NO_FALLBACK; | |
3075 | } | |
260a82e5 | 3076 | |
34fa110e | 3077 | if (flags & BDRV_REQ_MAY_UNMAP) { |
7154d8ae KW |
3078 | acb.aio_type |= QEMU_AIO_DISCARD; |
3079 | handler = handle_aiocb_write_zeroes_unmap; | |
3080 | } else { | |
3081 | handler = handle_aiocb_write_zeroes; | |
260a82e5 | 3082 | } |
34fa110e | 3083 | |
7154d8ae KW |
3084 | return raw_thread_pool_submit(bs, handler, &acb); |
3085 | } | |
3086 | ||
3087 | static int coroutine_fn raw_co_pwrite_zeroes( | |
3088 | BlockDriverState *bs, int64_t offset, | |
f34b2bcf | 3089 | int64_t bytes, BdrvRequestFlags flags) |
7154d8ae KW |
3090 | { |
3091 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); | |
260a82e5 PB |
3092 | } |
3093 | ||
3094 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
3095 | { | |
260a82e5 PB |
3096 | return 0; |
3097 | } | |
3098 | ||
d9245599 AN |
3099 | static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) |
3100 | { | |
3101 | BDRVRawState *s = bs->opaque; | |
3102 | return (BlockStatsSpecificFile) { | |
3103 | .discard_nb_ok = s->stats.discard_nb_ok, | |
3104 | .discard_nb_failed = s->stats.discard_nb_failed, | |
3105 | .discard_bytes_ok = s->stats.discard_bytes_ok, | |
3106 | }; | |
3107 | } | |
3108 | ||
3109 | static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) | |
3110 | { | |
3111 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3112 | ||
3113 | stats->driver = BLOCKDEV_DRIVER_FILE; | |
3114 | stats->u.file = get_blockstats_specific_file(bs); | |
3115 | ||
3116 | return stats; | |
3117 | } | |
3118 | ||
14176c8d | 3119 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
d9245599 AN |
3120 | static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) |
3121 | { | |
3122 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3123 | ||
3124 | stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; | |
3125 | stats->u.host_device = get_blockstats_specific_file(bs); | |
3126 | ||
3127 | return stats; | |
3128 | } | |
14176c8d | 3129 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
d9245599 | 3130 | |
6f482f74 CL |
3131 | static QemuOptsList raw_create_opts = { |
3132 | .name = "raw-create-opts", | |
3133 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
3134 | .desc = { | |
3135 | { | |
3136 | .name = BLOCK_OPT_SIZE, | |
3137 | .type = QEMU_OPT_SIZE, | |
3138 | .help = "Virtual disk size" | |
3139 | }, | |
4ab15590 CL |
3140 | { |
3141 | .name = BLOCK_OPT_NOCOW, | |
3142 | .type = QEMU_OPT_BOOL, | |
3143 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
3144 | }, | |
06247428 HT |
3145 | { |
3146 | .name = BLOCK_OPT_PREALLOC, | |
3147 | .type = QEMU_OPT_STRING, | |
abea0053 SG |
3148 | .help = "Preallocation mode (allowed values: off" |
3149 | #ifdef CONFIG_POSIX_FALLOCATE | |
3150 | ", falloc" | |
3151 | #endif | |
3152 | ", full)" | |
06247428 | 3153 | }, |
ffa244c8 KW |
3154 | { |
3155 | .name = BLOCK_OPT_EXTENT_SIZE_HINT, | |
3156 | .type = QEMU_OPT_SIZE, | |
3157 | .help = "Extent size hint for the image file, 0 to disable" | |
3158 | }, | |
6f482f74 CL |
3159 | { /* end of list */ } |
3160 | } | |
0e7e1989 KW |
3161 | }; |
3162 | ||
244a5668 FZ |
3163 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
3164 | Error **errp) | |
3165 | { | |
6ceabe6f | 3166 | BDRVRawState *s = bs->opaque; |
72373e40 | 3167 | int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags; |
6ceabe6f KW |
3168 | int open_flags; |
3169 | int ret; | |
3170 | ||
72373e40 VSO |
3171 | /* We may need a new fd if auto-read-only switches the mode */ |
3172 | ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, | |
3173 | false, errp); | |
3174 | if (ret < 0) { | |
3175 | return ret; | |
3176 | } else if (ret != s->fd) { | |
3177 | Error *local_err = NULL; | |
3178 | ||
6ceabe6f | 3179 | /* |
72373e40 VSO |
3180 | * Fail already check_perm() if we can't get a working O_DIRECT |
3181 | * alignment with the new fd. | |
6ceabe6f | 3182 | */ |
72373e40 VSO |
3183 | raw_probe_alignment(bs, ret, &local_err); |
3184 | if (local_err) { | |
3185 | error_propagate(errp, local_err); | |
3186 | return -EINVAL; | |
6ceabe6f | 3187 | } |
72373e40 VSO |
3188 | |
3189 | s->perm_change_fd = ret; | |
3190 | s->perm_change_flags = open_flags; | |
6ceabe6f KW |
3191 | } |
3192 | ||
3193 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
3194 | * but keep everything locked that new will need. */ | |
3195 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
3196 | if (ret < 0) { | |
3197 | goto fail; | |
3198 | } | |
3199 | ||
3200 | /* Copy locks to the new fd */ | |
eb43ea16 | 3201 | if (s->perm_change_fd && s->use_lock) { |
6ceabe6f KW |
3202 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, |
3203 | false, errp); | |
3204 | if (ret < 0) { | |
3205 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
3206 | goto fail; | |
3207 | } | |
3208 | } | |
3209 | return 0; | |
3210 | ||
3211 | fail: | |
72373e40 | 3212 | if (s->perm_change_fd) { |
6ceabe6f KW |
3213 | qemu_close(s->perm_change_fd); |
3214 | } | |
3215 | s->perm_change_fd = 0; | |
3216 | return ret; | |
244a5668 FZ |
3217 | } |
3218 | ||
3219 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
3220 | { | |
3221 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
3222 | |
3223 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
3224 | * called after .bdrv_reopen_commit) */ | |
3225 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
3226 | qemu_close(s->fd); | |
3227 | s->fd = s->perm_change_fd; | |
094e3639 | 3228 | s->open_flags = s->perm_change_flags; |
6ceabe6f KW |
3229 | } |
3230 | s->perm_change_fd = 0; | |
3231 | ||
244a5668 FZ |
3232 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
3233 | s->perm = perm; | |
3234 | s->shared_perm = shared; | |
3235 | } | |
3236 | ||
3237 | static void raw_abort_perm_update(BlockDriverState *bs) | |
3238 | { | |
6ceabe6f KW |
3239 | BDRVRawState *s = bs->opaque; |
3240 | ||
3241 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
3242 | * the file descriptor. */ | |
72373e40 | 3243 | if (s->perm_change_fd) { |
6ceabe6f KW |
3244 | qemu_close(s->perm_change_fd); |
3245 | } | |
3246 | s->perm_change_fd = 0; | |
3247 | ||
244a5668 FZ |
3248 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
3249 | } | |
3250 | ||
67b51fb9 | 3251 | static int coroutine_fn raw_co_copy_range_from( |
48535049 VSO |
3252 | BlockDriverState *bs, BdrvChild *src, int64_t src_offset, |
3253 | BdrvChild *dst, int64_t dst_offset, int64_t bytes, | |
67b51fb9 | 3254 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) |
1efad060 | 3255 | { |
67b51fb9 VSO |
3256 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3257 | read_flags, write_flags); | |
1efad060 FZ |
3258 | } |
3259 | ||
3260 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 | 3261 | BdrvChild *src, |
48535049 | 3262 | int64_t src_offset, |
67b51fb9 | 3263 | BdrvChild *dst, |
48535049 VSO |
3264 | int64_t dst_offset, |
3265 | int64_t bytes, | |
67b51fb9 VSO |
3266 | BdrvRequestFlags read_flags, |
3267 | BdrvRequestFlags write_flags) | |
1efad060 | 3268 | { |
58a209c4 | 3269 | RawPosixAIOData acb; |
1efad060 FZ |
3270 | BDRVRawState *s = bs->opaque; |
3271 | BDRVRawState *src_s; | |
3272 | ||
3273 | assert(dst->bs == bs); | |
3274 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
3275 | return -ENOTSUP; | |
3276 | } | |
3277 | ||
3278 | src_s = src->bs->opaque; | |
9f850f67 | 3279 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
3280 | return -EIO; |
3281 | } | |
58a209c4 KW |
3282 | |
3283 | acb = (RawPosixAIOData) { | |
3284 | .bs = bs, | |
3285 | .aio_type = QEMU_AIO_COPY_RANGE, | |
3286 | .aio_fildes = src_s->fd, | |
3287 | .aio_offset = src_offset, | |
3288 | .aio_nbytes = bytes, | |
3289 | .copy_range = { | |
3290 | .aio_fd2 = s->fd, | |
3291 | .aio_offset2 = dst_offset, | |
3292 | }, | |
3293 | }; | |
3294 | ||
3295 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
3296 | } |
3297 | ||
5f535a94 | 3298 | BlockDriver bdrv_file = { |
84a12e66 CH |
3299 | .format_name = "file", |
3300 | .protocol_name = "file", | |
856ae5c3 | 3301 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3302 | .bdrv_needs_filename = true, |
856ae5c3 | 3303 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 3304 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 3305 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
3306 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3307 | .bdrv_reopen_commit = raw_reopen_commit, | |
3308 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 3309 | .bdrv_close = raw_close, |
927f11e1 | 3310 | .bdrv_co_create = raw_co_create, |
efc75e2a | 3311 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 3312 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 3313 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 3314 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3315 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
9bffae14 | 3316 | .bdrv_co_delete_file = raw_co_delete_file, |
3b46e624 | 3317 | |
9d52aa3c KW |
3318 | .bdrv_co_preadv = raw_co_preadv, |
3319 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3320 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3321 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
3322 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3323 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3324 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3325 | .bdrv_io_plug = raw_aio_plug, |
3326 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 3327 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3328 | |
061ca8a3 | 3329 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 3330 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3331 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3332 | .bdrv_get_allocated_file_size |
3333 | = raw_get_allocated_file_size, | |
d9245599 | 3334 | .bdrv_get_specific_stats = raw_get_specific_stats, |
244a5668 FZ |
3335 | .bdrv_check_perm = raw_check_perm, |
3336 | .bdrv_set_perm = raw_set_perm, | |
3337 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 3338 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3339 | .mutable_opts = mutable_opts, |
83f64091 FB |
3340 | }; |
3341 | ||
19cb3738 FB |
3342 | /***********************************************/ |
3343 | /* host device */ | |
3344 | ||
14176c8d JD |
3345 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
3346 | ||
83affaa6 | 3347 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
3348 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3349 | CFIndex maxPathSize, int flags); | |
aa44d3f6 PMD |
3350 | |
3351 | #if !defined(MAC_OS_VERSION_12_0) \ | |
3352 | || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0) | |
3353 | #define IOMainPort IOMasterPort | |
3354 | #endif | |
3355 | ||
d0855f12 | 3356 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 3357 | { |
d0855f12 | 3358 | kern_return_t kernResult = KERN_FAILURE; |
aa44d3f6 | 3359 | mach_port_t mainPort; |
19cb3738 | 3360 | CFMutableDictionaryRef classesToMatch; |
d0855f12 JA |
3361 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
3362 | char *mediaType = NULL; | |
19cb3738 | 3363 | |
aa44d3f6 | 3364 | kernResult = IOMainPort(MACH_PORT_NULL, &mainPort); |
19cb3738 | 3365 | if ( KERN_SUCCESS != kernResult ) { |
aa44d3f6 | 3366 | printf("IOMainPort returned %d\n", kernResult); |
19cb3738 | 3367 | } |
3b46e624 | 3368 | |
d0855f12 JA |
3369 | int index; |
3370 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
3371 | classesToMatch = IOServiceMatching(matching_array[index]); | |
3372 | if (classesToMatch == NULL) { | |
3373 | error_report("IOServiceMatching returned NULL for %s", | |
3374 | matching_array[index]); | |
3375 | continue; | |
3376 | } | |
3377 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
3378 | kCFBooleanTrue); | |
aa44d3f6 | 3379 | kernResult = IOServiceGetMatchingServices(mainPort, classesToMatch, |
d0855f12 JA |
3380 | mediaIterator); |
3381 | if (kernResult != KERN_SUCCESS) { | |
3382 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
3383 | kernResult); | |
3384 | continue; | |
3385 | } | |
3b46e624 | 3386 | |
d0855f12 JA |
3387 | /* If a match was found, leave the loop */ |
3388 | if (*mediaIterator != 0) { | |
4f7d28d7 | 3389 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
3390 | mediaType = g_strdup(matching_array[index]); |
3391 | break; | |
3392 | } | |
3393 | } | |
3394 | return mediaType; | |
19cb3738 FB |
3395 | } |
3396 | ||
98caa5bc JA |
3397 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3398 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
3399 | { |
3400 | io_object_t nextMedia; | |
3401 | kern_return_t kernResult = KERN_FAILURE; | |
3402 | *bsdPath = '\0'; | |
3403 | nextMedia = IOIteratorNext( mediaIterator ); | |
3404 | if ( nextMedia ) | |
3405 | { | |
3406 | CFTypeRef bsdPathAsCFString; | |
3407 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3408 | if ( bsdPathAsCFString ) { | |
3409 | size_t devPathLength; | |
3410 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3411 | if (flags & BDRV_O_NOCACHE) { |
3412 | strcat(bsdPath, "r"); | |
3413 | } | |
19cb3738 FB |
3414 | devPathLength = strlen( bsdPath ); |
3415 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3416 | kernResult = KERN_SUCCESS; | |
3417 | } | |
3418 | CFRelease( bsdPathAsCFString ); | |
3419 | } | |
3420 | IOObjectRelease( nextMedia ); | |
3421 | } | |
3b46e624 | 3422 | |
19cb3738 FB |
3423 | return kernResult; |
3424 | } | |
3425 | ||
d0855f12 JA |
3426 | /* Sets up a real cdrom for use in QEMU */ |
3427 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3428 | { | |
3429 | int index, num_of_test_partitions = 2, fd; | |
3430 | char test_partition[MAXPATHLEN]; | |
3431 | bool partition_found = false; | |
3432 | ||
3433 | /* look for a working partition */ | |
3434 | for (index = 0; index < num_of_test_partitions; index++) { | |
3435 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3436 | index); | |
b18a24a9 | 3437 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); |
d0855f12 JA |
3438 | if (fd >= 0) { |
3439 | partition_found = true; | |
3440 | qemu_close(fd); | |
3441 | break; | |
3442 | } | |
3443 | } | |
3444 | ||
3445 | /* if a working partition on the device was not found */ | |
3446 | if (partition_found == false) { | |
3447 | error_setg(errp, "Failed to find a working partition on disc"); | |
3448 | } else { | |
4f7d28d7 | 3449 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3450 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3451 | } | |
3452 | return partition_found; | |
3453 | } | |
3454 | ||
3455 | /* Prints directions on mounting and unmounting a device */ | |
3456 | static void print_unmounting_directions(const char *file_name) | |
3457 | { | |
3458 | error_report("If device %s is mounted on the desktop, unmount" | |
3459 | " it first before using it in QEMU", file_name); | |
3460 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3461 | file_name); | |
3462 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3463 | } | |
3464 | ||
3465 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3466 | |
508c7cb3 CH |
3467 | static int hdev_probe_device(const char *filename) |
3468 | { | |
3469 | struct stat st; | |
3470 | ||
3471 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3472 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3473 | return 50; | |
3474 | ||
3475 | if (stat(filename, &st) >= 0 && | |
3476 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3477 | return 100; | |
3478 | } | |
3479 | ||
3480 | return 0; | |
3481 | } | |
3482 | ||
7af803d4 HR |
3483 | static void hdev_parse_filename(const char *filename, QDict *options, |
3484 | Error **errp) | |
3485 | { | |
03c320d8 | 3486 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3487 | } |
3488 | ||
3307ed7b DA |
3489 | static bool hdev_is_sg(BlockDriverState *bs) |
3490 | { | |
3491 | ||
3492 | #if defined(__linux__) | |
3493 | ||
0d4377b3 | 3494 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3495 | struct stat st; |
3496 | struct sg_scsi_id scsiid; | |
3497 | int sg_version; | |
0d4377b3 KW |
3498 | int ret; |
3499 | ||
3500 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3501 | return false; | |
3502 | } | |
3307ed7b | 3503 | |
0d4377b3 KW |
3504 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3505 | if (ret < 0) { | |
3506 | return false; | |
3507 | } | |
3508 | ||
3509 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3510 | if (ret >= 0) { | |
4f7d28d7 | 3511 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3512 | return true; |
3513 | } | |
3514 | ||
3515 | #endif | |
3516 | ||
3517 | return false; | |
3518 | } | |
3519 | ||
015a1036 HR |
3520 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3521 | Error **errp) | |
19cb3738 FB |
3522 | { |
3523 | BDRVRawState *s = bs->opaque; | |
da888d37 | 3524 | int ret; |
a76bab49 | 3525 | |
83affaa6 | 3526 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3527 | /* |
3528 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3529 | * would require more care. When @options come from -blockdev or | |
3530 | * blockdev_add, its members are typed according to the QAPI | |
3531 | * schema, but when they come from -drive, they're all QString. | |
3532 | */ | |
3307ed7b | 3533 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3534 | char bsd_path[MAXPATHLEN] = ""; |
3535 | bool error_occurred = false; | |
3536 | ||
3537 | /* If using a real cdrom */ | |
3538 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3539 | char *mediaType = NULL; | |
3540 | kern_return_t ret_val; | |
3541 | io_iterator_t mediaIterator = 0; | |
3542 | ||
3543 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3544 | if (mediaType == NULL) { | |
3545 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3546 | " drive"); | |
3547 | error_occurred = true; | |
3548 | goto hdev_open_Mac_error; | |
3549 | } | |
3307ed7b | 3550 | |
d0855f12 JA |
3551 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3552 | if (ret_val != KERN_SUCCESS) { | |
3553 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3554 | error_occurred = true; | |
3555 | goto hdev_open_Mac_error; | |
3556 | } | |
3557 | ||
3558 | /* If a real optical drive was not found */ | |
3559 | if (bsd_path[0] == '\0') { | |
3560 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3561 | error_occurred = true; | |
3562 | goto hdev_open_Mac_error; | |
3563 | } | |
3564 | ||
3565 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3566 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3567 | setup_cdrom(bsd_path, errp) == false) { | |
3568 | print_unmounting_directions(bsd_path); | |
3569 | error_occurred = true; | |
3570 | goto hdev_open_Mac_error; | |
19cb3738 | 3571 | } |
3b46e624 | 3572 | |
46f5ac20 | 3573 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3574 | |
3575 | hdev_open_Mac_error: | |
3576 | g_free(mediaType); | |
3577 | if (mediaIterator) { | |
3578 | IOObjectRelease(mediaIterator); | |
3579 | } | |
3580 | if (error_occurred) { | |
3581 | return -ENOENT; | |
3582 | } | |
19cb3738 | 3583 | } |
d0855f12 | 3584 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3585 | |
3586 | s->type = FTYPE_FILE; | |
90babde0 | 3587 | |
668f62ec | 3588 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
da888d37 | 3589 | if (ret < 0) { |
d0855f12 JA |
3590 | #if defined(__APPLE__) && defined(__MACH__) |
3591 | if (*bsd_path) { | |
3592 | filename = bsd_path; | |
3593 | } | |
3594 | /* if a physical device experienced an error while being opened */ | |
3595 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3596 | print_unmounting_directions(filename); | |
3597 | } | |
3598 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3599 | return ret; |
3600 | } | |
3601 | ||
3307ed7b DA |
3602 | /* Since this does ioctl the device must be already opened */ |
3603 | bs->sg = hdev_is_sg(bs); | |
3604 | ||
da888d37 | 3605 | return ret; |
19cb3738 FB |
3606 | } |
3607 | ||
03ff3ca3 | 3608 | #if defined(__linux__) |
2f3a7ab3 KW |
3609 | static int coroutine_fn |
3610 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3611 | { |
f141eafe | 3612 | BDRVRawState *s = bs->opaque; |
03425671 | 3613 | RawPosixAIOData acb; |
2f3a7ab3 | 3614 | int ret; |
221f715d | 3615 | |
2f3a7ab3 KW |
3616 | ret = fd_open(bs); |
3617 | if (ret < 0) { | |
3618 | return ret; | |
3619 | } | |
c208e8c2 | 3620 | |
7c9e5276 PB |
3621 | if (req == SG_IO && s->pr_mgr) { |
3622 | struct sg_io_hdr *io_hdr = buf; | |
3623 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3624 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3625 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3626 | s->fd, io_hdr); |
7c9e5276 PB |
3627 | } |
3628 | } | |
3629 | ||
03425671 KW |
3630 | acb = (RawPosixAIOData) { |
3631 | .bs = bs, | |
3632 | .aio_type = QEMU_AIO_IOCTL, | |
3633 | .aio_fildes = s->fd, | |
3634 | .aio_offset = 0, | |
3635 | .ioctl = { | |
3636 | .buf = buf, | |
3637 | .cmd = req, | |
3638 | }, | |
3639 | }; | |
3640 | ||
3641 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3642 | } |
f709623b | 3643 | #endif /* linux */ |
221f715d | 3644 | |
33d70fb6 | 3645 | static coroutine_fn int |
0c802287 | 3646 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
c36dd8a0 | 3647 | { |
1c450366 | 3648 | BDRVRawState *s = bs->opaque; |
33d70fb6 | 3649 | int ret; |
c36dd8a0 | 3650 | |
33d70fb6 KW |
3651 | ret = fd_open(bs); |
3652 | if (ret < 0) { | |
1c450366 | 3653 | raw_account_discard(s, bytes, ret); |
33d70fb6 | 3654 | return ret; |
c36dd8a0 | 3655 | } |
46ee0f46 | 3656 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3657 | } |
3658 | ||
2ffa76c2 | 3659 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f34b2bcf | 3660 | int64_t offset, int64_t bytes, BdrvRequestFlags flags) |
d0b4503e | 3661 | { |
d0b4503e PB |
3662 | int rc; |
3663 | ||
3664 | rc = fd_open(bs); | |
3665 | if (rc < 0) { | |
3666 | return rc; | |
3667 | } | |
34fa110e | 3668 | |
7154d8ae | 3669 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3670 | } |
3671 | ||
5efa9d5a | 3672 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3673 | .format_name = "host_device", |
84a12e66 | 3674 | .protocol_name = "host_device", |
0b4ce02e | 3675 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3676 | .bdrv_needs_filename = true, |
0b4ce02e | 3677 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3678 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3679 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3680 | .bdrv_close = raw_close, |
1bc6b705 JC |
3681 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3682 | .bdrv_reopen_commit = raw_reopen_commit, | |
3683 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3684 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3685 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3686 | .mutable_opts = mutable_opts, |
dd577a26 | 3687 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3688 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3689 | |
9d52aa3c KW |
3690 | .bdrv_co_preadv = raw_co_preadv, |
3691 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3692 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3693 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3694 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3695 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3696 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3697 | .bdrv_io_plug = raw_aio_plug, |
3698 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3699 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3700 | |
061ca8a3 | 3701 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3702 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3703 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3704 | .bdrv_get_allocated_file_size |
3705 | = raw_get_allocated_file_size, | |
d9245599 | 3706 | .bdrv_get_specific_stats = hdev_get_specific_stats, |
244a5668 FZ |
3707 | .bdrv_check_perm = raw_check_perm, |
3708 | .bdrv_set_perm = raw_set_perm, | |
3709 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3710 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3711 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3712 | |
f3a5d3f8 | 3713 | /* generic scsi device */ |
63ec93db | 3714 | #ifdef __linux__ |
2f3a7ab3 | 3715 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3716 | #endif |
f3a5d3f8 CH |
3717 | }; |
3718 | ||
18fa1c42 HR |
3719 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3720 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3721 | Error **errp) | |
3722 | { | |
03c320d8 | 3723 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3724 | } |
3725 | #endif | |
3726 | ||
3727 | #ifdef __linux__ | |
015a1036 HR |
3728 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3729 | Error **errp) | |
f3a5d3f8 CH |
3730 | { |
3731 | BDRVRawState *s = bs->opaque; | |
3732 | ||
f3a5d3f8 CH |
3733 | s->type = FTYPE_CD; |
3734 | ||
19a3da7f | 3735 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3736 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3737 | } |
3738 | ||
508c7cb3 CH |
3739 | static int cdrom_probe_device(const char *filename) |
3740 | { | |
3baf720e CR |
3741 | int fd, ret; |
3742 | int prio = 0; | |
343f8568 | 3743 | struct stat st; |
3baf720e | 3744 | |
b18a24a9 | 3745 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); |
3baf720e CR |
3746 | if (fd < 0) { |
3747 | goto out; | |
3748 | } | |
343f8568 JS |
3749 | ret = fstat(fd, &st); |
3750 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3751 | goto outc; | |
3752 | } | |
3baf720e CR |
3753 | |
3754 | /* Attempt to detect via a CDROM specific ioctl */ | |
3755 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3756 | if (ret >= 0) | |
3757 | prio = 100; | |
3758 | ||
343f8568 | 3759 | outc: |
2e1e79da | 3760 | qemu_close(fd); |
3baf720e CR |
3761 | out: |
3762 | return prio; | |
508c7cb3 CH |
3763 | } |
3764 | ||
e031f750 | 3765 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3766 | { |
3767 | BDRVRawState *s = bs->opaque; | |
3768 | int ret; | |
3769 | ||
3770 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3771 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3772 | } |
3773 | ||
f36f3949 | 3774 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3775 | { |
3776 | BDRVRawState *s = bs->opaque; | |
3777 | ||
3778 | if (eject_flag) { | |
3779 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3780 | perror("CDROMEJECT"); | |
3781 | } else { | |
3782 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3783 | perror("CDROMEJECT"); | |
3784 | } | |
f3a5d3f8 CH |
3785 | } |
3786 | ||
025e849a | 3787 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3788 | { |
3789 | BDRVRawState *s = bs->opaque; | |
3790 | ||
3791 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3792 | /* | |
3793 | * Note: an error can happen if the distribution automatically | |
3794 | * mounts the CD-ROM | |
3795 | */ | |
3796 | /* perror("CDROM_LOCKDOOR"); */ | |
3797 | } | |
f3a5d3f8 CH |
3798 | } |
3799 | ||
3800 | static BlockDriver bdrv_host_cdrom = { | |
3801 | .format_name = "host_cdrom", | |
84a12e66 | 3802 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3803 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3804 | .bdrv_needs_filename = true, |
508c7cb3 | 3805 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3806 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3807 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3808 | .bdrv_close = raw_close, |
1bc6b705 JC |
3809 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3810 | .bdrv_reopen_commit = raw_reopen_commit, | |
3811 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3812 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3813 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3814 | .mutable_opts = mutable_opts, |
dd577a26 | 3815 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3816 | |
9d52aa3c KW |
3817 | .bdrv_co_preadv = raw_co_preadv, |
3818 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3819 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3820 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3821 | .bdrv_io_plug = raw_aio_plug, |
3822 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3823 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3824 | |
061ca8a3 | 3825 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3826 | .bdrv_getlength = raw_getlength, |
3827 | .has_variable_length = true, | |
4a1d5e1f FZ |
3828 | .bdrv_get_allocated_file_size |
3829 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3830 | |
3831 | /* removable device support */ | |
3832 | .bdrv_is_inserted = cdrom_is_inserted, | |
3833 | .bdrv_eject = cdrom_eject, | |
025e849a | 3834 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3835 | |
3836 | /* generic scsi device */ | |
2f3a7ab3 | 3837 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3838 | }; |
3839 | #endif /* __linux__ */ | |
3840 | ||
a167ba50 | 3841 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3842 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3843 | Error **errp) | |
f3a5d3f8 CH |
3844 | { |
3845 | BDRVRawState *s = bs->opaque; | |
3846 | int ret; | |
3847 | ||
3848 | s->type = FTYPE_CD; | |
3849 | ||
668f62ec | 3850 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
e428e439 | 3851 | if (ret) { |
f3a5d3f8 | 3852 | return ret; |
e428e439 | 3853 | } |
f3a5d3f8 | 3854 | |
9b2260cb | 3855 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3856 | ioctl(s->fd, CDIOCALLOW); |
3857 | return 0; | |
3858 | } | |
3859 | ||
508c7cb3 CH |
3860 | static int cdrom_probe_device(const char *filename) |
3861 | { | |
3862 | if (strstart(filename, "/dev/cd", NULL) || | |
3863 | strstart(filename, "/dev/acd", NULL)) | |
3864 | return 100; | |
3865 | return 0; | |
3866 | } | |
3867 | ||
f3a5d3f8 CH |
3868 | static int cdrom_reopen(BlockDriverState *bs) |
3869 | { | |
3870 | BDRVRawState *s = bs->opaque; | |
3871 | int fd; | |
3872 | ||
3873 | /* | |
3874 | * Force reread of possibly changed/newly loaded disc, | |
3875 | * FreeBSD seems to not notice sometimes... | |
3876 | */ | |
3877 | if (s->fd >= 0) | |
2e1e79da | 3878 | qemu_close(s->fd); |
b18a24a9 | 3879 | fd = qemu_open(bs->filename, s->open_flags, NULL); |
f3a5d3f8 CH |
3880 | if (fd < 0) { |
3881 | s->fd = -1; | |
3882 | return -EIO; | |
3883 | } | |
3884 | s->fd = fd; | |
3885 | ||
9b2260cb | 3886 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3887 | ioctl(s->fd, CDIOCALLOW); |
3888 | return 0; | |
3889 | } | |
3890 | ||
e031f750 | 3891 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3892 | { |
3893 | return raw_getlength(bs) > 0; | |
3894 | } | |
3895 | ||
f36f3949 | 3896 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3897 | { |
3898 | BDRVRawState *s = bs->opaque; | |
3899 | ||
3900 | if (s->fd < 0) | |
822e1cd1 | 3901 | return; |
f3a5d3f8 CH |
3902 | |
3903 | (void) ioctl(s->fd, CDIOCALLOW); | |
3904 | ||
3905 | if (eject_flag) { | |
3906 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3907 | perror("CDIOCEJECT"); | |
3908 | } else { | |
3909 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3910 | perror("CDIOCCLOSE"); | |
3911 | } | |
3912 | ||
822e1cd1 | 3913 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3914 | } |
3915 | ||
025e849a | 3916 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3917 | { |
3918 | BDRVRawState *s = bs->opaque; | |
3919 | ||
3920 | if (s->fd < 0) | |
7bf37fed | 3921 | return; |
f3a5d3f8 CH |
3922 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3923 | /* | |
3924 | * Note: an error can happen if the distribution automatically | |
3925 | * mounts the CD-ROM | |
3926 | */ | |
3927 | /* perror("CDROM_LOCKDOOR"); */ | |
3928 | } | |
f3a5d3f8 CH |
3929 | } |
3930 | ||
3931 | static BlockDriver bdrv_host_cdrom = { | |
3932 | .format_name = "host_cdrom", | |
84a12e66 | 3933 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3934 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3935 | .bdrv_needs_filename = true, |
508c7cb3 | 3936 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3937 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3938 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3939 | .bdrv_close = raw_close, |
1bc6b705 JC |
3940 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3941 | .bdrv_reopen_commit = raw_reopen_commit, | |
3942 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3943 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3944 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3945 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3946 | |
9d52aa3c KW |
3947 | .bdrv_co_preadv = raw_co_preadv, |
3948 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3949 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3950 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3951 | .bdrv_io_plug = raw_aio_plug, |
3952 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3953 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3954 | |
061ca8a3 | 3955 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3956 | .bdrv_getlength = raw_getlength, |
3957 | .has_variable_length = true, | |
4a1d5e1f FZ |
3958 | .bdrv_get_allocated_file_size |
3959 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3960 | |
19cb3738 | 3961 | /* removable device support */ |
f3a5d3f8 CH |
3962 | .bdrv_is_inserted = cdrom_is_inserted, |
3963 | .bdrv_eject = cdrom_eject, | |
025e849a | 3964 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3965 | }; |
f3a5d3f8 | 3966 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3967 | |
14176c8d JD |
3968 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
3969 | ||
84a12e66 | 3970 | static void bdrv_file_init(void) |
5efa9d5a | 3971 | { |
508c7cb3 CH |
3972 | /* |
3973 | * Register all the drivers. Note that order is important, the driver | |
3974 | * registered last will get probed first. | |
3975 | */ | |
84a12e66 | 3976 | bdrv_register(&bdrv_file); |
14176c8d | 3977 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
5efa9d5a | 3978 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3979 | #ifdef __linux__ |
f3a5d3f8 CH |
3980 | bdrv_register(&bdrv_host_cdrom); |
3981 | #endif | |
a167ba50 | 3982 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3983 | bdrv_register(&bdrv_host_cdrom); |
3984 | #endif | |
14176c8d | 3985 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
5efa9d5a AL |
3986 | } |
3987 | ||
84a12e66 | 3988 | block_init(bdrv_file_init); |