]>
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; |
260a82e5 | 157 | bool discard_zeroes:1; |
0a4279d9 | 158 | bool use_linux_aio:1; |
c6447510 | 159 | bool use_linux_io_uring:1; |
c7ddc882 | 160 | int page_cache_inconsistent; /* errno from fdatasync failure */ |
d50d8222 | 161 | bool has_fallocate; |
3cad8307 | 162 | bool needs_alignment; |
5dbd0ce1 | 163 | bool force_alignment; |
f357fcd8 | 164 | bool drop_cache; |
31be8a2a | 165 | bool check_cache_dropped; |
1c450366 AN |
166 | struct { |
167 | uint64_t discard_nb_ok; | |
168 | uint64_t discard_nb_failed; | |
169 | uint64_t discard_bytes_ok; | |
170 | } stats; | |
7c9e5276 PB |
171 | |
172 | PRManager *pr_mgr; | |
19cb3738 FB |
173 | } BDRVRawState; |
174 | ||
eeb6b45d | 175 | typedef struct BDRVRawReopenState { |
eeb6b45d | 176 | int open_flags; |
f357fcd8 | 177 | bool drop_cache; |
31be8a2a | 178 | bool check_cache_dropped; |
eeb6b45d JC |
179 | } BDRVRawReopenState; |
180 | ||
14176c8d JD |
181 | static int fd_open(BlockDriverState *bs) |
182 | { | |
183 | BDRVRawState *s = bs->opaque; | |
184 | ||
185 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
186 | if (s->fd >= 0) { | |
187 | return 0; | |
188 | } | |
189 | return -EIO; | |
190 | } | |
191 | ||
22afa7b5 | 192 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 193 | |
de81a169 PB |
194 | typedef struct RawPosixAIOData { |
195 | BlockDriverState *bs; | |
d57c44d0 | 196 | int aio_type; |
de81a169 | 197 | int aio_fildes; |
d57c44d0 | 198 | |
de81a169 | 199 | off_t aio_offset; |
d57c44d0 KW |
200 | uint64_t aio_nbytes; |
201 | ||
93f4e2ff | 202 | union { |
d57c44d0 KW |
203 | struct { |
204 | struct iovec *iov; | |
205 | int niov; | |
206 | } io; | |
207 | struct { | |
208 | uint64_t cmd; | |
209 | void *buf; | |
210 | } ioctl; | |
93f4e2ff KW |
211 | struct { |
212 | int aio_fd2; | |
213 | off_t aio_offset2; | |
d57c44d0 | 214 | } copy_range; |
93f4e2ff KW |
215 | struct { |
216 | PreallocMode prealloc; | |
217 | Error **errp; | |
d57c44d0 | 218 | } truncate; |
93f4e2ff | 219 | }; |
de81a169 PB |
220 | } RawPosixAIOData; |
221 | ||
a167ba50 | 222 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 223 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
224 | #endif |
225 | ||
797e3e38 DE |
226 | /* |
227 | * Elide EAGAIN and EACCES details when failing to lock, as this | |
228 | * indicates that the specified file region is already locked by | |
229 | * another process, which is considered a common scenario. | |
230 | */ | |
231 | #define raw_lock_error_setg_errno(errp, err, fmt, ...) \ | |
232 | do { \ | |
233 | if ((err) == EAGAIN || (err) == EACCES) { \ | |
234 | error_setg((errp), (fmt), ## __VA_ARGS__); \ | |
235 | } else { \ | |
236 | error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \ | |
237 | } \ | |
238 | } while (0) | |
239 | ||
1de1ae0a | 240 | #if defined(__NetBSD__) |
db0754df | 241 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
242 | { |
243 | static char namebuf[PATH_MAX]; | |
244 | const char *dp, *fname; | |
245 | struct stat sb; | |
246 | ||
247 | fname = *filename; | |
248 | dp = strrchr(fname, '/'); | |
249 | if (lstat(fname, &sb) < 0) { | |
f6fc1e30 | 250 | error_setg_file_open(errp, errno, fname); |
1de1ae0a CE |
251 | return -errno; |
252 | } | |
253 | ||
254 | if (!S_ISBLK(sb.st_mode)) { | |
255 | return 0; | |
256 | } | |
257 | ||
258 | if (dp == NULL) { | |
259 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
260 | } else { | |
261 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
262 | (int)(dp - fname), fname, dp + 1); | |
263 | } | |
1de1ae0a | 264 | *filename = namebuf; |
db0754df | 265 | warn_report("%s is a block device, using %s", fname, *filename); |
1de1ae0a CE |
266 | |
267 | return 0; | |
268 | } | |
269 | #else | |
db0754df | 270 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
271 | { |
272 | return 0; | |
273 | } | |
274 | #endif | |
275 | ||
8a4ed0d1 ET |
276 | /* |
277 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
278 | */ | |
279 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 280 | { |
c25f53b0 | 281 | unsigned int sector_size; |
8a4ed0d1 | 282 | bool success = false; |
700f9ce0 | 283 | int i; |
c25f53b0 | 284 | |
8a4ed0d1 | 285 | errno = ENOTSUP; |
700f9ce0 | 286 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 287 | #ifdef BLKSSZGET |
700f9ce0 | 288 | BLKSSZGET, |
c25f53b0 PB |
289 | #endif |
290 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 291 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
292 | #endif |
293 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 294 | DIOCGSECTORSIZE, |
c25f53b0 | 295 | #endif |
700f9ce0 PM |
296 | }; |
297 | ||
298 | /* Try a few ioctls to get the right size */ | |
299 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
300 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
301 | *sector_size_p = sector_size; | |
302 | success = true; | |
303 | } | |
304 | } | |
8a4ed0d1 ET |
305 | |
306 | return success ? 0 : -errno; | |
307 | } | |
308 | ||
1a9335e4 ET |
309 | /** |
310 | * Get physical block size of @fd. | |
311 | * On success, store it in @blk_size and return 0. | |
312 | * On failure, return -errno. | |
313 | */ | |
314 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
315 | { | |
316 | #ifdef BLKPBSZGET | |
317 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
318 | return -errno; | |
319 | } | |
320 | return 0; | |
321 | #else | |
322 | return -ENOTSUP; | |
323 | #endif | |
324 | } | |
325 | ||
5edc8557 KW |
326 | /* |
327 | * Returns true if no alignment restrictions are necessary even for files | |
328 | * opened with O_DIRECT. | |
329 | * | |
330 | * raw_probe_alignment() probes the required alignment and assume that 1 means | |
331 | * the probing failed, so it falls back to a safe default of 4k. This can be | |
332 | * avoided if we know that byte alignment is okay for the file. | |
333 | */ | |
334 | static bool dio_byte_aligned(int fd) | |
335 | { | |
336 | #ifdef __linux__ | |
337 | struct statfs buf; | |
338 | int ret; | |
339 | ||
340 | ret = fstatfs(fd, &buf); | |
341 | if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { | |
342 | return true; | |
343 | } | |
344 | #endif | |
345 | return false; | |
346 | } | |
347 | ||
5dbd0ce1 KW |
348 | static bool raw_needs_alignment(BlockDriverState *bs) |
349 | { | |
350 | BDRVRawState *s = bs->opaque; | |
351 | ||
352 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { | |
353 | return true; | |
354 | } | |
355 | ||
356 | return s->force_alignment; | |
357 | } | |
358 | ||
22d182e8 SH |
359 | /* Check if read is allowed with given memory buffer and length. |
360 | * | |
361 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
362 | */ | |
363 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
364 | { | |
365 | ssize_t ret = pread(fd, buf, len, 0); | |
366 | ||
367 | if (ret >= 0) { | |
368 | return true; | |
369 | } | |
370 | ||
371 | #ifdef __linux__ | |
372 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
373 | * other errors (e.g. real I/O error), which could happen on a failed | |
374 | * drive, since we only care about probing alignment. | |
375 | */ | |
376 | if (errno != EINVAL) { | |
377 | return true; | |
378 | } | |
379 | #endif | |
380 | ||
381 | return false; | |
382 | } | |
383 | ||
8a4ed0d1 ET |
384 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
385 | { | |
386 | BDRVRawState *s = bs->opaque; | |
387 | char *buf; | |
8e3b0cbb | 388 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size()); |
a6b257a0 | 389 | size_t alignments[] = {1, 512, 1024, 2048, 4096}; |
8a4ed0d1 | 390 | |
b192af8a | 391 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 392 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 393 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 394 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
395 | s->buf_align = 1; |
396 | return; | |
397 | } | |
398 | ||
a5b8dd2c | 399 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
400 | s->buf_align = 0; |
401 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
402 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
403 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 404 | } |
a5730b8b TH |
405 | |
406 | #ifdef __linux__ | |
407 | /* | |
408 | * The XFS ioctl definitions are shipped in extra packages that might | |
409 | * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl | |
410 | * here, we simply use our own definition instead: | |
411 | */ | |
412 | struct xfs_dioattr { | |
413 | uint32_t d_mem; | |
414 | uint32_t d_miniosz; | |
415 | uint32_t d_maxiosz; | |
416 | } da; | |
417 | if (ioctl(fd, _IOR('X', 30, struct xfs_dioattr), &da) >= 0) { | |
418 | bs->bl.request_alignment = da.d_miniosz; | |
419 | /* The kernel returns wrong information for d_mem */ | |
420 | /* s->buf_align = da.d_mem; */ | |
c25f53b0 PB |
421 | } |
422 | #endif | |
423 | ||
a6b257a0 NS |
424 | /* |
425 | * If we could not get the sizes so far, we can only guess them. First try | |
426 | * to detect request alignment, since it is more likely to succeed. Then | |
427 | * try to detect buf_align, which cannot be detected in some cases (e.g. | |
428 | * Gluster). If buf_align cannot be detected, we fallback to the value of | |
429 | * request_alignment. | |
430 | */ | |
431 | ||
432 | if (!bs->bl.request_alignment) { | |
433 | int i; | |
c25f53b0 | 434 | size_t align; |
a6b257a0 NS |
435 | buf = qemu_memalign(max_align, max_align); |
436 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
437 | align = alignments[i]; | |
438 | if (raw_is_io_aligned(fd, buf, align)) { | |
439 | /* Fallback to safe value. */ | |
440 | bs->bl.request_alignment = (align != 1) ? align : max_align; | |
c25f53b0 PB |
441 | break; |
442 | } | |
443 | } | |
444 | qemu_vfree(buf); | |
445 | } | |
446 | ||
a6b257a0 NS |
447 | if (!s->buf_align) { |
448 | int i; | |
c25f53b0 | 449 | size_t align; |
a6b257a0 NS |
450 | buf = qemu_memalign(max_align, 2 * max_align); |
451 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
452 | align = alignments[i]; | |
453 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
236094c7 | 454 | /* Fallback to request_alignment. */ |
a6b257a0 | 455 | s->buf_align = (align != 1) ? align : bs->bl.request_alignment; |
c25f53b0 PB |
456 | break; |
457 | } | |
458 | } | |
459 | qemu_vfree(buf); | |
460 | } | |
df26a350 | 461 | |
a5b8dd2c | 462 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
463 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
464 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 465 | } |
c25f53b0 PB |
466 | } |
467 | ||
bca5283b | 468 | static int check_hdev_writable(int fd) |
20eaf1bf KW |
469 | { |
470 | #if defined(BLKROGET) | |
471 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
472 | * This is independent of device node permissions and therefore open(2) | |
473 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
474 | * | |
475 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
476 | * check for read-only block devices so that Linux block devices behave | |
477 | * properly. | |
478 | */ | |
479 | struct stat st; | |
480 | int readonly = 0; | |
481 | ||
bca5283b | 482 | if (fstat(fd, &st)) { |
20eaf1bf KW |
483 | return -errno; |
484 | } | |
485 | ||
486 | if (!S_ISBLK(st.st_mode)) { | |
487 | return 0; | |
488 | } | |
489 | ||
bca5283b | 490 | if (ioctl(fd, BLKROGET, &readonly) < 0) { |
20eaf1bf KW |
491 | return -errno; |
492 | } | |
493 | ||
494 | if (readonly) { | |
495 | return -EACCES; | |
496 | } | |
497 | #endif /* defined(BLKROGET) */ | |
498 | return 0; | |
499 | } | |
500 | ||
23dece19 | 501 | static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) |
6a8dc042 | 502 | { |
23dece19 | 503 | bool read_write = false; |
6a8dc042 JC |
504 | assert(open_flags != NULL); |
505 | ||
506 | *open_flags |= O_BINARY; | |
507 | *open_flags &= ~O_ACCMODE; | |
23dece19 KW |
508 | |
509 | if (bdrv_flags & BDRV_O_AUTO_RDONLY) { | |
510 | read_write = has_writers; | |
511 | } else if (bdrv_flags & BDRV_O_RDWR) { | |
512 | read_write = true; | |
513 | } | |
514 | ||
515 | if (read_write) { | |
6a8dc042 JC |
516 | *open_flags |= O_RDWR; |
517 | } else { | |
518 | *open_flags |= O_RDONLY; | |
519 | } | |
520 | ||
521 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
522 | * and O_DIRECT for no caching. */ | |
523 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
524 | *open_flags |= O_DIRECT; | |
525 | } | |
6a8dc042 JC |
526 | } |
527 | ||
078896a9 HR |
528 | static void raw_parse_filename(const char *filename, QDict *options, |
529 | Error **errp) | |
530 | { | |
03c320d8 | 531 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
532 | } |
533 | ||
c66a6157 KW |
534 | static QemuOptsList raw_runtime_opts = { |
535 | .name = "raw", | |
536 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
537 | .desc = { | |
538 | { | |
539 | .name = "filename", | |
540 | .type = QEMU_OPT_STRING, | |
541 | .help = "File name of the image", | |
542 | }, | |
0a4279d9 KW |
543 | { |
544 | .name = "aio", | |
545 | .type = QEMU_OPT_STRING, | |
c6447510 | 546 | .help = "host AIO implementation (threads, native, io_uring)", |
0a4279d9 | 547 | }, |
684960d4 SG |
548 | { |
549 | .name = "aio-max-batch", | |
550 | .type = QEMU_OPT_NUMBER, | |
551 | .help = "AIO max batch size (0 = auto handled by AIO backend, default: 0)", | |
552 | }, | |
16b48d5d FZ |
553 | { |
554 | .name = "locking", | |
555 | .type = QEMU_OPT_STRING, | |
556 | .help = "file locking mode (on/off/auto, default: auto)", | |
557 | }, | |
7c9e5276 PB |
558 | { |
559 | .name = "pr-manager", | |
560 | .type = QEMU_OPT_STRING, | |
561 | .help = "id of persistent reservation manager object (default: none)", | |
562 | }, | |
f357fcd8 SH |
563 | #if defined(__linux__) |
564 | { | |
565 | .name = "drop-cache", | |
566 | .type = QEMU_OPT_BOOL, | |
567 | .help = "invalidate page cache during live migration (default: on)", | |
568 | }, | |
569 | #endif | |
31be8a2a SH |
570 | { |
571 | .name = "x-check-cache-dropped", | |
572 | .type = QEMU_OPT_BOOL, | |
573 | .help = "check that page cache was dropped on live migration (default: off)" | |
574 | }, | |
c66a6157 KW |
575 | { /* end of list */ } |
576 | }, | |
577 | }; | |
578 | ||
8a2ce0bc AG |
579 | static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; |
580 | ||
c66a6157 | 581 | static int raw_open_common(BlockDriverState *bs, QDict *options, |
230ff739 JS |
582 | int bdrv_flags, int open_flags, |
583 | bool device, Error **errp) | |
83f64091 FB |
584 | { |
585 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
586 | QemuOpts *opts; |
587 | Error *local_err = NULL; | |
8bfea15d | 588 | const char *filename = NULL; |
7c9e5276 | 589 | const char *str; |
0a4279d9 | 590 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 591 | int fd, ret; |
260a82e5 | 592 | struct stat st; |
244a5668 | 593 | OnOffAuto locking; |
83f64091 | 594 | |
87ea75d5 | 595 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 596 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
c66a6157 KW |
597 | ret = -EINVAL; |
598 | goto fail; | |
599 | } | |
600 | ||
601 | filename = qemu_opt_get(opts, "filename"); | |
602 | ||
db0754df | 603 | ret = raw_normalize_devicepath(&filename, errp); |
1de1ae0a | 604 | if (ret != 0) { |
c66a6157 | 605 | goto fail; |
1de1ae0a CE |
606 | } |
607 | ||
c6447510 AM |
608 | if (bdrv_flags & BDRV_O_NATIVE_AIO) { |
609 | aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; | |
610 | #ifdef CONFIG_LINUX_IO_URING | |
611 | } else if (bdrv_flags & BDRV_O_IO_URING) { | |
612 | aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; | |
613 | #endif | |
614 | } else { | |
615 | aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; | |
616 | } | |
617 | ||
f7abe0ec MAL |
618 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
619 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 620 | aio_default, &local_err); |
0a4279d9 KW |
621 | if (local_err) { |
622 | error_propagate(errp, local_err); | |
623 | ret = -EINVAL; | |
624 | goto fail; | |
625 | } | |
c6447510 | 626 | |
0a4279d9 | 627 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); |
c6447510 AM |
628 | #ifdef CONFIG_LINUX_IO_URING |
629 | s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); | |
630 | #endif | |
0a4279d9 | 631 | |
684960d4 SG |
632 | s->aio_max_batch = qemu_opt_get_number(opts, "aio-max-batch", 0); |
633 | ||
f7abe0ec MAL |
634 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
635 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 636 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
637 | if (local_err) { |
638 | error_propagate(errp, local_err); | |
639 | ret = -EINVAL; | |
640 | goto fail; | |
641 | } | |
642 | switch (locking) { | |
643 | case ON_OFF_AUTO_ON: | |
644 | s->use_lock = true; | |
2b218f5d | 645 | if (!qemu_has_ofd_lock()) { |
db0754df FZ |
646 | warn_report("File lock requested but OFD locking syscall is " |
647 | "unavailable, falling back to POSIX file locks"); | |
648 | error_printf("Due to the implementation, locks can be lost " | |
649 | "unexpectedly.\n"); | |
2b218f5d | 650 | } |
244a5668 FZ |
651 | break; |
652 | case ON_OFF_AUTO_OFF: | |
653 | s->use_lock = false; | |
654 | break; | |
655 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 656 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
657 | break; |
658 | default: | |
659 | abort(); | |
660 | } | |
661 | ||
7c9e5276 PB |
662 | str = qemu_opt_get(opts, "pr-manager"); |
663 | if (str) { | |
664 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
665 | if (local_err) { | |
666 | error_propagate(errp, local_err); | |
667 | ret = -EINVAL; | |
668 | goto fail; | |
669 | } | |
670 | } | |
671 | ||
f357fcd8 | 672 | s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); |
31be8a2a SH |
673 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
674 | false); | |
675 | ||
6a8dc042 | 676 | s->open_flags = open_flags; |
23dece19 | 677 | raw_parse_flags(bdrv_flags, &s->open_flags, false); |
83f64091 | 678 | |
90babde0 | 679 | s->fd = -1; |
b18a24a9 | 680 | fd = qemu_open(filename, s->open_flags, errp); |
64107dc0 KW |
681 | ret = fd < 0 ? -errno : 0; |
682 | ||
64107dc0 | 683 | if (ret < 0) { |
c66a6157 | 684 | if (ret == -EROFS) { |
19cb3738 | 685 | ret = -EACCES; |
c66a6157 KW |
686 | } |
687 | goto fail; | |
19cb3738 | 688 | } |
83f64091 | 689 | s->fd = fd; |
9ef91a67 | 690 | |
bca5283b KW |
691 | /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ |
692 | if (s->open_flags & O_RDWR) { | |
693 | ret = check_hdev_writable(s->fd); | |
694 | if (ret < 0) { | |
695 | error_setg_errno(errp, -ret, "The device is not writable"); | |
696 | goto fail; | |
697 | } | |
698 | } | |
699 | ||
244a5668 FZ |
700 | s->perm = 0; |
701 | s->shared_perm = BLK_PERM_ALL; | |
702 | ||
5c6c3a6c | 703 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 704 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
705 | if (s->use_linux_aio) { |
706 | if (!(s->open_flags & O_DIRECT)) { | |
707 | error_setg(errp, "aio=native was specified, but it requires " | |
708 | "cache.direct=on, which was not specified."); | |
709 | ret = -EINVAL; | |
710 | goto fail; | |
711 | } | |
712 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
713 | error_prepend(errp, "Unable to use native AIO: "); | |
714 | goto fail; | |
715 | } | |
96518254 | 716 | } |
1501ecc1 | 717 | #else |
0a4279d9 | 718 | if (s->use_linux_aio) { |
d657c0c2 KW |
719 | error_setg(errp, "aio=native was specified, but is not supported " |
720 | "in this build."); | |
721 | ret = -EINVAL; | |
722 | goto fail; | |
1501ecc1 SH |
723 | } |
724 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 725 | |
c6447510 AM |
726 | #ifdef CONFIG_LINUX_IO_URING |
727 | if (s->use_linux_io_uring) { | |
728 | if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { | |
729 | error_prepend(errp, "Unable to use io_uring: "); | |
730 | goto fail; | |
731 | } | |
732 | } | |
733 | #else | |
734 | if (s->use_linux_io_uring) { | |
735 | error_setg(errp, "aio=io_uring was specified, but is not supported " | |
736 | "in this build."); | |
737 | ret = -EINVAL; | |
738 | goto fail; | |
739 | } | |
740 | #endif /* !defined(CONFIG_LINUX_IO_URING) */ | |
741 | ||
7ce21016 | 742 | s->has_discard = true; |
97a2ae34 | 743 | s->has_write_zeroes = true; |
260a82e5 PB |
744 | |
745 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 746 | ret = -errno; |
260a82e5 PB |
747 | error_setg_errno(errp, errno, "Could not stat file"); |
748 | goto fail; | |
749 | } | |
230ff739 JS |
750 | |
751 | if (!device) { | |
8d17adf3 DB |
752 | if (!S_ISREG(st.st_mode)) { |
753 | error_setg(errp, "'%s' driver requires '%s' to be a regular file", | |
754 | bs->drv->format_name, bs->filename); | |
230ff739 JS |
755 | ret = -EINVAL; |
756 | goto fail; | |
757 | } else { | |
758 | s->discard_zeroes = true; | |
759 | s->has_fallocate = true; | |
760 | } | |
761 | } else { | |
762 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
8d17adf3 DB |
763 | error_setg(errp, "'%s' driver requires '%s' to be either " |
764 | "a character or block device", | |
765 | bs->drv->format_name, bs->filename); | |
230ff739 JS |
766 | ret = -EINVAL; |
767 | goto fail; | |
768 | } | |
260a82e5 | 769 | } |
230ff739 | 770 | |
d0b4503e PB |
771 | if (S_ISBLK(st.st_mode)) { |
772 | #ifdef BLKDISCARDZEROES | |
773 | unsigned int arg; | |
774 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
775 | s->discard_zeroes = true; | |
776 | } | |
777 | #endif | |
778 | #ifdef __linux__ | |
779 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
780 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 781 | * Same for BLKZEROOUT. |
d0b4503e PB |
782 | */ |
783 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
784 | s->discard_zeroes = false; | |
97a2ae34 | 785 | s->has_write_zeroes = false; |
d0b4503e PB |
786 | } |
787 | #endif | |
788 | } | |
3cad8307 RPM |
789 | #ifdef __FreeBSD__ |
790 | if (S_ISCHR(st.st_mode)) { | |
791 | /* | |
792 | * The file is a char device (disk), which on FreeBSD isn't behind | |
793 | * a pager, so force all requests to be aligned. This is needed | |
794 | * so QEMU makes sure all IO operations on the device are aligned | |
795 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
796 | */ | |
5dbd0ce1 | 797 | s->force_alignment = true; |
3cad8307 RPM |
798 | } |
799 | #endif | |
5dbd0ce1 | 800 | s->needs_alignment = raw_needs_alignment(bs); |
260a82e5 | 801 | |
738301e1 | 802 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
2f0c6e7a KW |
803 | if (S_ISREG(st.st_mode)) { |
804 | /* When extending regular files, we get zeros from the OS */ | |
805 | bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; | |
806 | } | |
c66a6157 KW |
807 | ret = 0; |
808 | fail: | |
a8c5cf27 KW |
809 | if (ret < 0 && s->fd != -1) { |
810 | qemu_close(s->fd); | |
811 | } | |
8bfea15d KW |
812 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
813 | unlink(filename); | |
814 | } | |
c66a6157 KW |
815 | qemu_opts_del(opts); |
816 | return ret; | |
83f64091 FB |
817 | } |
818 | ||
015a1036 HR |
819 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
820 | Error **errp) | |
90babde0 CH |
821 | { |
822 | BDRVRawState *s = bs->opaque; | |
823 | ||
824 | s->type = FTYPE_FILE; | |
230ff739 | 825 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
826 | } |
827 | ||
244a5668 FZ |
828 | typedef enum { |
829 | RAW_PL_PREPARE, | |
830 | RAW_PL_COMMIT, | |
831 | RAW_PL_ABORT, | |
832 | } RawPermLockOp; | |
833 | ||
834 | #define PERM_FOREACH(i) \ | |
835 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
836 | ||
837 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
838 | * file; if @unlock == true, also unlock the unneeded bytes. | |
839 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
840 | */ | |
2996ffad | 841 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
842 | uint64_t perm_lock_bits, |
843 | uint64_t shared_perm_lock_bits, | |
844 | bool unlock, Error **errp) | |
845 | { | |
846 | int ret; | |
847 | int i; | |
2996ffad FZ |
848 | uint64_t locked_perm, locked_shared_perm; |
849 | ||
850 | if (s) { | |
851 | locked_perm = s->locked_perm; | |
852 | locked_shared_perm = s->locked_shared_perm; | |
853 | } else { | |
854 | /* | |
855 | * We don't have the previous bits, just lock/unlock for each of the | |
856 | * requested bits. | |
857 | */ | |
858 | if (unlock) { | |
859 | locked_perm = BLK_PERM_ALL; | |
860 | locked_shared_perm = BLK_PERM_ALL; | |
861 | } else { | |
862 | locked_perm = 0; | |
863 | locked_shared_perm = 0; | |
864 | } | |
865 | } | |
244a5668 FZ |
866 | |
867 | PERM_FOREACH(i) { | |
868 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
869 | uint64_t bit = (1ULL << i); |
870 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 871 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 | 872 | if (ret) { |
797e3e38 DE |
873 | raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", |
874 | off); | |
244a5668 | 875 | return ret; |
2996ffad FZ |
876 | } else if (s) { |
877 | s->locked_perm |= bit; | |
244a5668 | 878 | } |
2996ffad | 879 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 880 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 | 881 | if (ret) { |
797e3e38 | 882 | error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); |
244a5668 | 883 | return ret; |
2996ffad FZ |
884 | } else if (s) { |
885 | s->locked_perm &= ~bit; | |
244a5668 FZ |
886 | } |
887 | } | |
888 | } | |
889 | PERM_FOREACH(i) { | |
890 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
891 | uint64_t bit = (1ULL << i); |
892 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 893 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 | 894 | if (ret) { |
797e3e38 DE |
895 | raw_lock_error_setg_errno(errp, -ret, "Failed to lock byte %d", |
896 | off); | |
244a5668 | 897 | return ret; |
2996ffad FZ |
898 | } else if (s) { |
899 | s->locked_shared_perm |= bit; | |
244a5668 | 900 | } |
2996ffad FZ |
901 | } else if (unlock && (locked_shared_perm & bit) && |
902 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 903 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 | 904 | if (ret) { |
797e3e38 | 905 | error_setg_errno(errp, -ret, "Failed to unlock byte %d", off); |
244a5668 | 906 | return ret; |
2996ffad FZ |
907 | } else if (s) { |
908 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
909 | } |
910 | } | |
911 | } | |
912 | return 0; | |
913 | } | |
914 | ||
915 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 916 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
917 | Error **errp) |
918 | { | |
919 | int ret; | |
920 | int i; | |
921 | ||
922 | PERM_FOREACH(i) { | |
923 | int off = RAW_LOCK_SHARED_BASE + i; | |
924 | uint64_t p = 1ULL << i; | |
925 | if (perm & p) { | |
d0a96155 | 926 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
927 | if (ret) { |
928 | char *perm_name = bdrv_perm_names(p); | |
797e3e38 DE |
929 | |
930 | raw_lock_error_setg_errno(errp, -ret, | |
931 | "Failed to get \"%s\" lock", | |
932 | perm_name); | |
244a5668 | 933 | g_free(perm_name); |
244a5668 FZ |
934 | return ret; |
935 | } | |
936 | } | |
937 | } | |
938 | PERM_FOREACH(i) { | |
939 | int off = RAW_LOCK_PERM_BASE + i; | |
940 | uint64_t p = 1ULL << i; | |
941 | if (!(shared_perm & p)) { | |
d0a96155 | 942 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
943 | if (ret) { |
944 | char *perm_name = bdrv_perm_names(p); | |
797e3e38 DE |
945 | |
946 | raw_lock_error_setg_errno(errp, -ret, | |
947 | "Failed to get shared \"%s\" lock", | |
948 | perm_name); | |
244a5668 | 949 | g_free(perm_name); |
244a5668 FZ |
950 | return ret; |
951 | } | |
952 | } | |
953 | } | |
954 | return 0; | |
955 | } | |
956 | ||
957 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
958 | RawPermLockOp op, | |
959 | uint64_t new_perm, uint64_t new_shared, | |
960 | Error **errp) | |
961 | { | |
962 | BDRVRawState *s = bs->opaque; | |
963 | int ret = 0; | |
964 | Error *local_err = NULL; | |
965 | ||
966 | if (!s->use_lock) { | |
967 | return 0; | |
968 | } | |
969 | ||
970 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
971 | return 0; | |
972 | } | |
973 | ||
244a5668 FZ |
974 | switch (op) { |
975 | case RAW_PL_PREPARE: | |
696aaaed VSO |
976 | if ((s->perm | new_perm) == s->perm && |
977 | (s->shared_perm & new_shared) == s->shared_perm) | |
978 | { | |
979 | /* | |
980 | * We are going to unlock bytes, it should not fail. If it fail due | |
981 | * to some fs-dependent permission-unrelated reasons (which occurs | |
982 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
983 | * can't prevent such errors by any check here. And we ignore them | |
984 | * anyway in ABORT and COMMIT. | |
985 | */ | |
986 | return 0; | |
987 | } | |
f2e3af29 | 988 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
989 | ~s->shared_perm | ~new_shared, |
990 | false, errp); | |
991 | if (!ret) { | |
f2e3af29 | 992 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
993 | if (!ret) { |
994 | return 0; | |
995 | } | |
b857431d FZ |
996 | error_append_hint(errp, |
997 | "Is another process using the image [%s]?\n", | |
998 | bs->filename); | |
244a5668 | 999 | } |
244a5668 FZ |
1000 | /* fall through to unlock bytes. */ |
1001 | case RAW_PL_ABORT: | |
f2e3af29 | 1002 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 1003 | true, &local_err); |
244a5668 FZ |
1004 | if (local_err) { |
1005 | /* Theoretically the above call only unlocks bytes and it cannot | |
1006 | * fail. Something weird happened, report it. | |
1007 | */ | |
db0754df | 1008 | warn_report_err(local_err); |
244a5668 FZ |
1009 | } |
1010 | break; | |
1011 | case RAW_PL_COMMIT: | |
f2e3af29 | 1012 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 1013 | true, &local_err); |
244a5668 FZ |
1014 | if (local_err) { |
1015 | /* Theoretically the above call only unlocks bytes and it cannot | |
1016 | * fail. Something weird happened, report it. | |
1017 | */ | |
db0754df | 1018 | warn_report_err(local_err); |
244a5668 FZ |
1019 | } |
1020 | break; | |
1021 | } | |
1022 | return ret; | |
1023 | } | |
1024 | ||
ad24b679 MAL |
1025 | /* Sets a specific flag */ |
1026 | static int fcntl_setfl(int fd, int flag) | |
1027 | { | |
1028 | int flags; | |
1029 | ||
1030 | flags = fcntl(fd, F_GETFL); | |
1031 | if (flags == -1) { | |
1032 | return -errno; | |
1033 | } | |
1034 | if (fcntl(fd, F_SETFL, flags | flag) == -1) { | |
1035 | return -errno; | |
1036 | } | |
1037 | return 0; | |
1038 | } | |
1039 | ||
5cec2870 | 1040 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 1041 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 1042 | Error **errp) |
5cec2870 KW |
1043 | { |
1044 | BDRVRawState *s = bs->opaque; | |
1045 | int fd = -1; | |
1046 | int ret; | |
23dece19 KW |
1047 | bool has_writers = perm & |
1048 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
1049 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
1050 | #ifdef O_NOATIME | |
1051 | fcntl_flags |= O_NOATIME; | |
1052 | #endif | |
1053 | ||
1054 | *open_flags = 0; | |
1055 | if (s->type == FTYPE_CD) { | |
1056 | *open_flags |= O_NONBLOCK; | |
1057 | } | |
1058 | ||
23dece19 | 1059 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
1060 | |
1061 | #ifdef O_ASYNC | |
1062 | /* Not all operating systems have O_ASYNC, and those that don't | |
1063 | * will not let us track the state into rs->open_flags (typically | |
1064 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
1065 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
1066 | */ | |
1067 | assert((s->open_flags & O_ASYNC) == 0); | |
1068 | #endif | |
1069 | ||
6ceabe6f KW |
1070 | if (!force_dup && *open_flags == s->open_flags) { |
1071 | /* We're lucky, the existing fd is fine */ | |
1072 | return s->fd; | |
1073 | } | |
1074 | ||
5cec2870 KW |
1075 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
1076 | /* dup the original fd */ | |
1077 | fd = qemu_dup(s->fd); | |
1078 | if (fd >= 0) { | |
1079 | ret = fcntl_setfl(fd, *open_flags); | |
1080 | if (ret) { | |
1081 | qemu_close(fd); | |
1082 | fd = -1; | |
1083 | } | |
1084 | } | |
1085 | } | |
1086 | ||
b18a24a9 | 1087 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ |
5cec2870 KW |
1088 | if (fd == -1) { |
1089 | const char *normalized_filename = bs->filename; | |
1090 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
1091 | if (ret >= 0) { | |
b18a24a9 | 1092 | fd = qemu_open(normalized_filename, *open_flags, errp); |
5cec2870 | 1093 | if (fd == -1) { |
5cec2870 KW |
1094 | return -1; |
1095 | } | |
1096 | } | |
1097 | } | |
1098 | ||
bca5283b KW |
1099 | if (fd != -1 && (*open_flags & O_RDWR)) { |
1100 | ret = check_hdev_writable(fd); | |
1101 | if (ret < 0) { | |
1102 | qemu_close(fd); | |
1103 | error_setg_errno(errp, -ret, "The device is not writable"); | |
1104 | return -1; | |
1105 | } | |
1106 | } | |
1107 | ||
5cec2870 KW |
1108 | return fd; |
1109 | } | |
1110 | ||
eeb6b45d JC |
1111 | static int raw_reopen_prepare(BDRVReopenState *state, |
1112 | BlockReopenQueue *queue, Error **errp) | |
1113 | { | |
1114 | BDRVRawState *s; | |
4e6d13c9 | 1115 | BDRVRawReopenState *rs; |
31be8a2a | 1116 | QemuOpts *opts; |
a6aeca0c | 1117 | int ret; |
eeb6b45d JC |
1118 | |
1119 | assert(state != NULL); | |
1120 | assert(state->bs != NULL); | |
1121 | ||
1122 | s = state->bs->opaque; | |
1123 | ||
5839e53b | 1124 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 1125 | rs = state->opaque; |
31be8a2a SH |
1126 | |
1127 | /* Handle options changes */ | |
1128 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 1129 | if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { |
31be8a2a SH |
1130 | ret = -EINVAL; |
1131 | goto out; | |
1132 | } | |
1133 | ||
f357fcd8 | 1134 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
1135 | rs->check_cache_dropped = |
1136 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
1137 | ||
1138 | /* This driver's reopen function doesn't currently allow changing | |
1139 | * other options, so let's put them back in the original QDict and | |
1140 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
1141 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 1142 | |
72373e40 VSO |
1143 | /* |
1144 | * As part of reopen prepare we also want to create new fd by | |
1145 | * raw_reconfigure_getfd(). But it wants updated "perm", when in | |
1146 | * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to | |
1147 | * permission update. Happily, permission update is always a part (a seprate | |
1148 | * stage) of bdrv_reopen_multiple() so we can rely on this fact and | |
1149 | * reconfigure fd in raw_check_perm(). | |
1150 | */ | |
df26a350 | 1151 | |
e0c9cf3a | 1152 | s->reopen_state = state; |
a6aeca0c | 1153 | ret = 0; |
72373e40 | 1154 | |
31be8a2a SH |
1155 | out: |
1156 | qemu_opts_del(opts); | |
eeb6b45d JC |
1157 | return ret; |
1158 | } | |
1159 | ||
eeb6b45d JC |
1160 | static void raw_reopen_commit(BDRVReopenState *state) |
1161 | { | |
4e6d13c9 | 1162 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1163 | BDRVRawState *s = state->bs->opaque; |
1164 | ||
f357fcd8 | 1165 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1166 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1167 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1168 | g_free(state->opaque); |
1169 | state->opaque = NULL; | |
e0c9cf3a KW |
1170 | |
1171 | assert(s->reopen_state == state); | |
1172 | s->reopen_state = NULL; | |
eeb6b45d JC |
1173 | } |
1174 | ||
1175 | ||
1176 | static void raw_reopen_abort(BDRVReopenState *state) | |
1177 | { | |
4e6d13c9 | 1178 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1179 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1180 | |
1181 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1182 | if (rs == NULL) { |
eeb6b45d JC |
1183 | return; |
1184 | } | |
1185 | ||
eeb6b45d JC |
1186 | g_free(state->opaque); |
1187 | state->opaque = NULL; | |
e0c9cf3a KW |
1188 | |
1189 | assert(s->reopen_state == state); | |
1190 | s->reopen_state = NULL; | |
eeb6b45d JC |
1191 | } |
1192 | ||
18473467 | 1193 | static int hdev_get_max_hw_transfer(int fd, struct stat *st) |
6f607174 FZ |
1194 | { |
1195 | #ifdef BLKSECTGET | |
18473467 PB |
1196 | if (S_ISBLK(st->st_mode)) { |
1197 | unsigned short max_sectors = 0; | |
1198 | if (ioctl(fd, BLKSECTGET, &max_sectors) == 0) { | |
1199 | return max_sectors * 512; | |
1200 | } | |
6f607174 | 1201 | } else { |
18473467 PB |
1202 | int max_bytes = 0; |
1203 | if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
1204 | return max_bytes; | |
1205 | } | |
6f607174 | 1206 | } |
18473467 | 1207 | return -errno; |
6f607174 FZ |
1208 | #else |
1209 | return -ENOSYS; | |
1210 | #endif | |
1211 | } | |
1212 | ||
18473467 | 1213 | static int hdev_get_max_segments(int fd, struct stat *st) |
9103f1ce FZ |
1214 | { |
1215 | #ifdef CONFIG_LINUX | |
1216 | char buf[32]; | |
1217 | const char *end; | |
867eccfe | 1218 | char *sysfspath = NULL; |
9103f1ce | 1219 | int ret; |
867eccfe | 1220 | int sysfd = -1; |
9103f1ce | 1221 | long max_segments; |
867eccfe | 1222 | |
18473467 | 1223 | if (S_ISCHR(st->st_mode)) { |
8ad5ab61 PB |
1224 | if (ioctl(fd, SG_GET_SG_TABLESIZE, &ret) == 0) { |
1225 | return ret; | |
1226 | } | |
1227 | return -ENOTSUP; | |
1228 | } | |
1229 | ||
18473467 | 1230 | if (!S_ISBLK(st->st_mode)) { |
8ad5ab61 PB |
1231 | return -ENOTSUP; |
1232 | } | |
1233 | ||
9103f1ce | 1234 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", |
18473467 | 1235 | major(st->st_rdev), minor(st->st_rdev)); |
867eccfe ML |
1236 | sysfd = open(sysfspath, O_RDONLY); |
1237 | if (sysfd == -1) { | |
9103f1ce FZ |
1238 | ret = -errno; |
1239 | goto out; | |
1240 | } | |
1241 | do { | |
867eccfe | 1242 | ret = read(sysfd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1243 | } while (ret == -1 && errno == EINTR); |
1244 | if (ret < 0) { | |
1245 | ret = -errno; | |
1246 | goto out; | |
1247 | } else if (ret == 0) { | |
1248 | ret = -EIO; | |
1249 | goto out; | |
1250 | } | |
1251 | buf[ret] = 0; | |
1252 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1253 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1254 | if (ret == 0 && end && *end == '\n') { | |
1255 | ret = max_segments; | |
1256 | } | |
1257 | ||
1258 | out: | |
867eccfe ML |
1259 | if (sysfd != -1) { |
1260 | close(sysfd); | |
fed414df | 1261 | } |
9103f1ce FZ |
1262 | g_free(sysfspath); |
1263 | return ret; | |
1264 | #else | |
1265 | return -ENOTSUP; | |
1266 | #endif | |
1267 | } | |
1268 | ||
3baca891 | 1269 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1270 | { |
1271 | BDRVRawState *s = bs->opaque; | |
18473467 | 1272 | struct stat st; |
6f607174 | 1273 | |
5dbd0ce1 | 1274 | s->needs_alignment = raw_needs_alignment(bs); |
18473467 | 1275 | raw_probe_alignment(bs, s->fd, errp); |
5dbd0ce1 | 1276 | |
18473467 | 1277 | bs->bl.min_mem_alignment = s->buf_align; |
8e3b0cbb | 1278 | bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size()); |
18473467 PB |
1279 | |
1280 | /* | |
1281 | * Maximum transfers are best effort, so it is okay to ignore any | |
1282 | * errors. That said, based on the man page errors in fstat would be | |
1283 | * very much unexpected; the only possible case seems to be ENOMEM. | |
1284 | */ | |
1285 | if (fstat(s->fd, &st)) { | |
1286 | return; | |
1287 | } | |
1288 | ||
0dfc7af2 AO |
1289 | #if defined(__APPLE__) && (__MACH__) |
1290 | struct statfs buf; | |
1291 | ||
1292 | if (!fstatfs(s->fd, &buf)) { | |
1293 | bs->bl.opt_transfer = buf.f_iosize; | |
1294 | bs->bl.pdiscard_alignment = buf.f_bsize; | |
1295 | } | |
1296 | #endif | |
1297 | ||
18473467 PB |
1298 | if (bs->sg || S_ISBLK(st.st_mode)) { |
1299 | int ret = hdev_get_max_hw_transfer(s->fd, &st); | |
867eccfe ML |
1300 | |
1301 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
18473467 | 1302 | bs->bl.max_hw_transfer = ret; |
867eccfe ML |
1303 | } |
1304 | ||
18473467 | 1305 | ret = hdev_get_max_segments(s->fd, &st); |
867eccfe | 1306 | if (ret > 0) { |
cc071629 | 1307 | bs->bl.max_hw_iov = ret; |
6f607174 FZ |
1308 | } |
1309 | } | |
c25f53b0 | 1310 | } |
83f64091 | 1311 | |
1a9335e4 ET |
1312 | static int check_for_dasd(int fd) |
1313 | { | |
1314 | #ifdef BIODASDINFO2 | |
1315 | struct dasd_information2_t info = {0}; | |
1316 | ||
1317 | return ioctl(fd, BIODASDINFO2, &info); | |
1318 | #else | |
1319 | return -1; | |
1320 | #endif | |
1321 | } | |
1322 | ||
1323 | /** | |
1324 | * Try to get @bs's logical and physical block size. | |
1325 | * On success, store them in @bsz and return zero. | |
1326 | * On failure, return negative errno. | |
1327 | */ | |
1328 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1329 | { | |
1330 | BDRVRawState *s = bs->opaque; | |
1331 | int ret; | |
1332 | ||
1333 | /* If DASD, get blocksizes */ | |
1334 | if (check_for_dasd(s->fd) < 0) { | |
1335 | return -ENOTSUP; | |
1336 | } | |
1337 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1338 | if (ret < 0) { | |
1339 | return ret; | |
1340 | } | |
1341 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1342 | } | |
1343 | ||
1344 | /** | |
1345 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1346 | * On success, store them in @geo and return 0. | |
1347 | * On failure return -errno. | |
1348 | * (Allows block driver to assign default geometry values that guest sees) | |
1349 | */ | |
1350 | #ifdef __linux__ | |
1351 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1352 | { | |
1353 | BDRVRawState *s = bs->opaque; | |
1354 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1355 | |
1356 | /* If DASD, get its geometry */ | |
1357 | if (check_for_dasd(s->fd) < 0) { | |
1358 | return -ENOTSUP; | |
1359 | } | |
1360 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1361 | return -errno; | |
1362 | } | |
1363 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1364 | (e.g. certain multipath setups) */ | |
1365 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1366 | return -ENOTSUP; | |
1367 | } | |
1368 | /* Do not return a geometry for partition */ | |
1369 | if (ioctl_geo.start != 0) { | |
1370 | return -ENOTSUP; | |
1371 | } | |
1372 | geo->heads = ioctl_geo.heads; | |
1373 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1374 | geo->cylinders = ioctl_geo.cylinders; |
1375 | ||
1376 | return 0; | |
1377 | } | |
1378 | #else /* __linux__ */ | |
1379 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1380 | { | |
1381 | return -ENOTSUP; | |
1382 | } | |
1383 | #endif | |
1384 | ||
03425671 KW |
1385 | #if defined(__linux__) |
1386 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1387 | { |
03425671 | 1388 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1389 | int ret; |
1390 | ||
bd80936a PB |
1391 | do { |
1392 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); | |
1393 | } while (ret == -1 && errno == EINTR); | |
de81a169 PB |
1394 | if (ret == -1) { |
1395 | return -errno; | |
1396 | } | |
1397 | ||
b608c8dc | 1398 | return 0; |
de81a169 | 1399 | } |
03425671 | 1400 | #endif /* linux */ |
de81a169 | 1401 | |
06dc9bd5 | 1402 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1403 | { |
06dc9bd5 | 1404 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1405 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1406 | int ret; |
1407 | ||
e5bcf967 | 1408 | if (s->page_cache_inconsistent) { |
c7ddc882 | 1409 | return -s->page_cache_inconsistent; |
e5bcf967 KW |
1410 | } |
1411 | ||
de81a169 PB |
1412 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1413 | if (ret == -1) { | |
60ff2ae2 DB |
1414 | trace_file_flush_fdatasync_failed(errno); |
1415 | ||
e5bcf967 KW |
1416 | /* There is no clear definition of the semantics of a failing fsync(), |
1417 | * so we may have to assume the worst. The sad truth is that this | |
1418 | * assumption is correct for Linux. Some pages are now probably marked | |
1419 | * clean in the page cache even though they are inconsistent with the | |
1420 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1421 | * further writeback attempt will be made. We can't get back to a state | |
1422 | * in which we know what is on disk (we would have to rewrite | |
1423 | * everything that was touched since the last fdatasync() at least), so | |
1424 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1425 | * really defined, I have little hope that other OSes are doing better. | |
1426 | * | |
1427 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1428 | * cache. */ | |
1429 | if ((s->open_flags & O_DIRECT) == 0) { | |
c7ddc882 | 1430 | s->page_cache_inconsistent = errno; |
e5bcf967 | 1431 | } |
de81a169 PB |
1432 | return -errno; |
1433 | } | |
1434 | return 0; | |
1435 | } | |
1436 | ||
1437 | #ifdef CONFIG_PREADV | |
1438 | ||
1439 | static bool preadv_present = true; | |
1440 | ||
1441 | static ssize_t | |
1442 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1443 | { | |
1444 | return preadv(fd, iov, nr_iov, offset); | |
1445 | } | |
1446 | ||
1447 | static ssize_t | |
1448 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1449 | { | |
1450 | return pwritev(fd, iov, nr_iov, offset); | |
1451 | } | |
1452 | ||
1453 | #else | |
1454 | ||
1455 | static bool preadv_present = false; | |
1456 | ||
1457 | static ssize_t | |
1458 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1459 | { | |
1460 | return -ENOSYS; | |
1461 | } | |
1462 | ||
1463 | static ssize_t | |
1464 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1465 | { | |
1466 | return -ENOSYS; | |
1467 | } | |
1468 | ||
1469 | #endif | |
1470 | ||
1471 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1472 | { | |
1473 | ssize_t len; | |
1474 | ||
1475 | do { | |
1476 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1477 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1478 | aiocb->io.iov, |
1479 | aiocb->io.niov, | |
de81a169 PB |
1480 | aiocb->aio_offset); |
1481 | else | |
1482 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1483 | aiocb->io.iov, |
1484 | aiocb->io.niov, | |
de81a169 PB |
1485 | aiocb->aio_offset); |
1486 | } while (len == -1 && errno == EINTR); | |
1487 | ||
1488 | if (len == -1) { | |
1489 | return -errno; | |
1490 | } | |
1491 | return len; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Read/writes the data to/from a given linear buffer. | |
1496 | * | |
1497 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1498 | * reads are only returned if the end of the file is reached. | |
1499 | */ | |
1500 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1501 | { | |
1502 | ssize_t offset = 0; | |
1503 | ssize_t len; | |
1504 | ||
1505 | while (offset < aiocb->aio_nbytes) { | |
1506 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1507 | len = pwrite(aiocb->aio_fildes, | |
1508 | (const char *)buf + offset, | |
1509 | aiocb->aio_nbytes - offset, | |
1510 | aiocb->aio_offset + offset); | |
1511 | } else { | |
1512 | len = pread(aiocb->aio_fildes, | |
1513 | buf + offset, | |
1514 | aiocb->aio_nbytes - offset, | |
1515 | aiocb->aio_offset + offset); | |
1516 | } | |
1517 | if (len == -1 && errno == EINTR) { | |
1518 | continue; | |
61ed73cf SH |
1519 | } else if (len == -1 && errno == EINVAL && |
1520 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1521 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1522 | offset > 0) { | |
1523 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1524 | * after a short read. Assume that O_DIRECT short reads only occur | |
1525 | * at EOF. Therefore this is a short read, not an I/O error. | |
1526 | */ | |
1527 | break; | |
de81a169 PB |
1528 | } else if (len == -1) { |
1529 | offset = -errno; | |
1530 | break; | |
1531 | } else if (len == 0) { | |
1532 | break; | |
1533 | } | |
1534 | offset += len; | |
1535 | } | |
1536 | ||
1537 | return offset; | |
1538 | } | |
1539 | ||
999e6b69 | 1540 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1541 | { |
999e6b69 | 1542 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1543 | ssize_t nbytes; |
1544 | char *buf; | |
1545 | ||
1546 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1547 | /* | |
1548 | * If there is just a single buffer, and it is properly aligned | |
1549 | * we can just use plain pread/pwrite without any problems. | |
1550 | */ | |
d57c44d0 | 1551 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1552 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1553 | goto out; | |
de81a169 PB |
1554 | } |
1555 | /* | |
1556 | * We have more than one iovec, and all are properly aligned. | |
1557 | * | |
1558 | * Try preadv/pwritev first and fall back to linearizing the | |
1559 | * buffer if it's not supported. | |
1560 | */ | |
1561 | if (preadv_present) { | |
1562 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1563 | if (nbytes == aiocb->aio_nbytes || | |
1564 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1565 | goto out; |
de81a169 PB |
1566 | } |
1567 | preadv_present = false; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1572 | * using these interfaces. For now retry using plain | |
1573 | * pread/pwrite? | |
1574 | */ | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * Ok, we have to do it the hard way, copy all segments into | |
1579 | * a single aligned buffer. | |
1580 | */ | |
50d4a858 KW |
1581 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1582 | if (buf == NULL) { | |
54c7ca1b KW |
1583 | nbytes = -ENOMEM; |
1584 | goto out; | |
50d4a858 KW |
1585 | } |
1586 | ||
de81a169 PB |
1587 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1588 | char *p = buf; | |
1589 | int i; | |
1590 | ||
d57c44d0 KW |
1591 | for (i = 0; i < aiocb->io.niov; ++i) { |
1592 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1593 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1594 | } |
8eb029c2 | 1595 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1596 | } |
1597 | ||
1598 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1599 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1600 | char *p = buf; | |
1601 | size_t count = aiocb->aio_nbytes, copy; | |
1602 | int i; | |
1603 | ||
d57c44d0 | 1604 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1605 | copy = count; |
d57c44d0 KW |
1606 | if (copy > aiocb->io.iov[i].iov_len) { |
1607 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1608 | } |
d57c44d0 | 1609 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1610 | assert(count >= copy); |
de81a169 PB |
1611 | p += copy; |
1612 | count -= copy; | |
1613 | } | |
8eb029c2 | 1614 | assert(count == 0); |
de81a169 PB |
1615 | } |
1616 | qemu_vfree(buf); | |
1617 | ||
54c7ca1b KW |
1618 | out: |
1619 | if (nbytes == aiocb->aio_nbytes) { | |
1620 | return 0; | |
1621 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1622 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1623 | return -EINVAL; | |
1624 | } else { | |
1625 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1626 | 0, aiocb->aio_nbytes - nbytes); | |
1627 | return 0; | |
1628 | } | |
1629 | } else { | |
1630 | assert(nbytes < 0); | |
1631 | return nbytes; | |
1632 | } | |
de81a169 PB |
1633 | } |
1634 | ||
0dfc7af2 | 1635 | #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD) |
1486df0e DL |
1636 | static int translate_err(int err) |
1637 | { | |
1638 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1639 | err == -ENOTTY) { | |
1640 | err = -ENOTSUP; | |
1641 | } | |
1642 | return err; | |
1643 | } | |
0dfc7af2 | 1644 | #endif |
1486df0e | 1645 | |
d50d8222 | 1646 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1647 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1648 | { | |
1649 | do { | |
1650 | if (fallocate(fd, mode, offset, len) == 0) { | |
1651 | return 0; | |
1652 | } | |
1653 | } while (errno == EINTR); | |
1654 | return translate_err(-errno); | |
1655 | } | |
1656 | #endif | |
1657 | ||
37cc9f7f | 1658 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1659 | { |
37cc9f7f | 1660 | int ret = -ENOTSUP; |
97a2ae34 PB |
1661 | BDRVRawState *s = aiocb->bs->opaque; |
1662 | ||
37cc9f7f | 1663 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1664 | return -ENOTSUP; |
1665 | } | |
1666 | ||
97a2ae34 | 1667 | #ifdef BLKZEROOUT |
738301e1 KW |
1668 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1669 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1670 | * fallbacks. */ | |
1671 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1672 | do { | |
1673 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1674 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1675 | return 0; | |
1676 | } | |
1677 | } while (errno == EINTR); | |
37cc9f7f | 1678 | |
738301e1 | 1679 | ret = translate_err(-errno); |
effecce6 KW |
1680 | if (ret == -ENOTSUP) { |
1681 | s->has_write_zeroes = false; | |
1682 | } | |
738301e1 | 1683 | } |
97a2ae34 | 1684 | #endif |
97a2ae34 | 1685 | |
97a2ae34 PB |
1686 | return ret; |
1687 | } | |
1688 | ||
7154d8ae | 1689 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1690 | { |
7154d8ae | 1691 | RawPosixAIOData *aiocb = opaque; |
70d9110b | 1692 | #ifdef CONFIG_FALLOCATE |
b2c6f23f | 1693 | BDRVRawState *s = aiocb->bs->opaque; |
70d9110b DL |
1694 | int64_t len; |
1695 | #endif | |
37cc9f7f DL |
1696 | |
1697 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1698 | return handle_aiocb_write_zeroes_block(aiocb); | |
1699 | } | |
1700 | ||
b953f075 DL |
1701 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1702 | if (s->has_write_zeroes) { | |
1703 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1704 | aiocb->aio_offset, aiocb->aio_nbytes); | |
fa95e9fb TH |
1705 | if (ret == -ENOTSUP) { |
1706 | s->has_write_zeroes = false; | |
1707 | } else if (ret == 0 || ret != -EINVAL) { | |
b953f075 DL |
1708 | return ret; |
1709 | } | |
fa95e9fb TH |
1710 | /* |
1711 | * Note: Some file systems do not like unaligned byte ranges, and | |
1712 | * return EINVAL in such a case, though they should not do it according | |
1713 | * to the man-page of fallocate(). Thus we simply ignore this return | |
1714 | * value and try the other fallbacks instead. | |
1715 | */ | |
b953f075 DL |
1716 | } |
1717 | #endif | |
1718 | ||
1cdc3239 DL |
1719 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1720 | if (s->has_discard && s->has_fallocate) { | |
1721 | int ret = do_fallocate(s->fd, | |
1722 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1723 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1724 | if (ret == 0) { | |
1725 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1726 | if (ret == 0 || ret != -ENOTSUP) { | |
1727 | return ret; | |
1728 | } | |
1729 | s->has_fallocate = false; | |
73ebf297 TH |
1730 | } else if (ret == -EINVAL) { |
1731 | /* | |
1732 | * Some file systems like older versions of GPFS do not like un- | |
1733 | * aligned byte ranges, and return EINVAL in such a case, though | |
1734 | * they should not do it according to the man-page of fallocate(). | |
1735 | * Warn about the bad filesystem and try the final fallback instead. | |
1736 | */ | |
1737 | warn_report_once("Your file system is misbehaving: " | |
1738 | "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. " | |
68857f13 | 1739 | "Please report this bug to your file system " |
73ebf297 | 1740 | "vendor."); |
1cdc3239 DL |
1741 | } else if (ret != -ENOTSUP) { |
1742 | return ret; | |
1743 | } else { | |
1744 | s->has_discard = false; | |
1745 | } | |
1746 | } | |
1747 | #endif | |
1748 | ||
d50d8222 | 1749 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1750 | /* Last resort: we are trying to extend the file with zeroed data. This |
1751 | * can be done via fallocate(fd, 0) */ | |
1752 | len = bdrv_getlength(aiocb->bs); | |
1753 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1754 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1755 | if (ret == 0 || ret != -ENOTSUP) { | |
1756 | return ret; | |
1757 | } | |
1758 | s->has_fallocate = false; | |
1759 | } | |
1760 | #endif | |
1761 | ||
37cc9f7f DL |
1762 | return -ENOTSUP; |
1763 | } | |
1764 | ||
7154d8ae | 1765 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1766 | { |
7154d8ae | 1767 | RawPosixAIOData *aiocb = opaque; |
34fa110e | 1768 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
34fa110e KW |
1769 | |
1770 | /* First try to write zeros and unmap at the same time */ | |
1771 | ||
1772 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
b3ac2b94 SS |
1773 | int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1774 | aiocb->aio_offset, aiocb->aio_nbytes); | |
bae127d4 AD |
1775 | switch (ret) { |
1776 | case -ENOTSUP: | |
1777 | case -EINVAL: | |
ece4fa91 | 1778 | case -EBUSY: |
bae127d4 AD |
1779 | break; |
1780 | default: | |
34fa110e KW |
1781 | return ret; |
1782 | } | |
1783 | #endif | |
1784 | ||
34fa110e KW |
1785 | /* If we couldn't manage to unmap while guaranteed that the area reads as |
1786 | * all-zero afterwards, just write zeroes without unmapping */ | |
b3ac2b94 | 1787 | return handle_aiocb_write_zeroes(aiocb); |
34fa110e KW |
1788 | } |
1789 | ||
1efad060 FZ |
1790 | #ifndef HAVE_COPY_FILE_RANGE |
1791 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1792 | off_t *out_off, size_t len, unsigned int flags) | |
1793 | { | |
1794 | #ifdef __NR_copy_file_range | |
1795 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1796 | out_off, len, flags); | |
1797 | #else | |
1798 | errno = ENOSYS; | |
1799 | return -1; | |
1800 | #endif | |
1801 | } | |
1802 | #endif | |
1803 | ||
58a209c4 | 1804 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1805 | { |
58a209c4 | 1806 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1807 | uint64_t bytes = aiocb->aio_nbytes; |
1808 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1809 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1810 | |
1811 | while (bytes) { | |
1812 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1813 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1814 | bytes, 0); |
ecc983a5 | 1815 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1816 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1817 | 0, ret); | |
c436e3d0 FZ |
1818 | if (ret == 0) { |
1819 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1820 | * buffer I/O. */ | |
1821 | return -ENOSPC; | |
1efad060 FZ |
1822 | } |
1823 | if (ret < 0) { | |
c436e3d0 FZ |
1824 | switch (errno) { |
1825 | case ENOSYS: | |
1efad060 | 1826 | return -ENOTSUP; |
c436e3d0 FZ |
1827 | case EINTR: |
1828 | continue; | |
1829 | default: | |
1efad060 FZ |
1830 | return -errno; |
1831 | } | |
1832 | } | |
1efad060 FZ |
1833 | bytes -= ret; |
1834 | } | |
1835 | return 0; | |
1836 | } | |
1837 | ||
46ee0f46 | 1838 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1839 | { |
46ee0f46 | 1840 | RawPosixAIOData *aiocb = opaque; |
13a02833 | 1841 | int ret = -ENOTSUP; |
8238010b PB |
1842 | BDRVRawState *s = aiocb->bs->opaque; |
1843 | ||
7ce21016 PB |
1844 | if (!s->has_discard) { |
1845 | return -ENOTSUP; | |
8238010b PB |
1846 | } |
1847 | ||
1848 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1849 | #ifdef BLKDISCARD | |
1850 | do { | |
1851 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1852 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1853 | return 0; | |
1854 | } | |
1855 | } while (errno == EINTR); | |
1856 | ||
0dfc7af2 | 1857 | ret = translate_err(-errno); |
8238010b PB |
1858 | #endif |
1859 | } else { | |
8238010b | 1860 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
0b991712 DL |
1861 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1862 | aiocb->aio_offset, aiocb->aio_nbytes); | |
13a02833 | 1863 | ret = translate_err(ret); |
0dfc7af2 AO |
1864 | #elif defined(__APPLE__) && (__MACH__) |
1865 | fpunchhole_t fpunchhole; | |
1866 | fpunchhole.fp_flags = 0; | |
1867 | fpunchhole.reserved = 0; | |
1868 | fpunchhole.fp_offset = aiocb->aio_offset; | |
1869 | fpunchhole.fp_length = aiocb->aio_nbytes; | |
1870 | if (fcntl(s->fd, F_PUNCHHOLE, &fpunchhole) == -1) { | |
1871 | ret = errno == ENODEV ? -ENOTSUP : -errno; | |
1872 | } else { | |
1873 | ret = 0; | |
1874 | } | |
8238010b PB |
1875 | #endif |
1876 | } | |
1877 | ||
1486df0e | 1878 | if (ret == -ENOTSUP) { |
7ce21016 | 1879 | s->has_discard = false; |
8238010b PB |
1880 | } |
1881 | return ret; | |
1882 | } | |
1883 | ||
3a20013f NS |
1884 | /* |
1885 | * Help alignment probing by allocating the first block. | |
1886 | * | |
1887 | * When reading with direct I/O from unallocated area on Gluster backed by XFS, | |
1888 | * reading succeeds regardless of request length. In this case we fallback to | |
1889 | * safe alignment which is not optimal. Allocating the first block avoids this | |
1890 | * fallback. | |
1891 | * | |
1892 | * fd may be opened with O_DIRECT, but we don't know the buffer alignment or | |
1893 | * request alignment, so we use safe values. | |
1894 | * | |
1895 | * Returns: 0 on success, -errno on failure. Since this is an optimization, | |
1896 | * caller may ignore failures. | |
1897 | */ | |
1898 | static int allocate_first_block(int fd, size_t max_size) | |
1899 | { | |
1900 | size_t write_size = (max_size < MAX_BLOCKSIZE) | |
1901 | ? BDRV_SECTOR_SIZE | |
1902 | : MAX_BLOCKSIZE; | |
8e3b0cbb | 1903 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size()); |
3a20013f NS |
1904 | void *buf; |
1905 | ssize_t n; | |
1906 | int ret; | |
1907 | ||
1908 | buf = qemu_memalign(max_align, write_size); | |
1909 | memset(buf, 0, write_size); | |
1910 | ||
1911 | do { | |
1912 | n = pwrite(fd, buf, write_size, 0); | |
1913 | } while (n == -1 && errno == EINTR); | |
1914 | ||
1915 | ret = (n == -1) ? -errno : 0; | |
1916 | ||
1917 | qemu_vfree(buf); | |
1918 | return ret; | |
1919 | } | |
1920 | ||
29cb4c01 | 1921 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1922 | { |
29cb4c01 | 1923 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1924 | int result = 0; |
1925 | int64_t current_length = 0; | |
1926 | char *buf = NULL; | |
1927 | struct stat st; | |
1928 | int fd = aiocb->aio_fildes; | |
1929 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1930 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1931 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1932 | |
1933 | if (fstat(fd, &st) < 0) { | |
1934 | result = -errno; | |
1935 | error_setg_errno(errp, -result, "Could not stat file"); | |
1936 | return result; | |
1937 | } | |
1938 | ||
1939 | current_length = st.st_size; | |
d57c44d0 | 1940 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1941 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1942 | return -ENOTSUP; | |
1943 | } | |
1944 | ||
d57c44d0 | 1945 | switch (prealloc) { |
93f4e2ff KW |
1946 | #ifdef CONFIG_POSIX_FALLOCATE |
1947 | case PREALLOC_MODE_FALLOC: | |
1948 | /* | |
1949 | * Truncating before posix_fallocate() makes it about twice slower on | |
1950 | * file systems that do not support fallocate(), trying to check if a | |
1951 | * block is allocated before allocating it, so don't do that here. | |
1952 | */ | |
1953 | if (offset != current_length) { | |
1954 | result = -posix_fallocate(fd, current_length, | |
1955 | offset - current_length); | |
1956 | if (result != 0) { | |
1957 | /* posix_fallocate() doesn't set errno. */ | |
1958 | error_setg_errno(errp, -result, | |
1959 | "Could not preallocate new data"); | |
3a20013f NS |
1960 | } else if (current_length == 0) { |
1961 | /* | |
1962 | * posix_fallocate() uses fallocate() if the filesystem | |
1963 | * supports it, or fallback to manually writing zeroes. If | |
1964 | * fallocate() was used, unaligned reads from the fallocated | |
1965 | * area in raw_probe_alignment() will succeed, hence we need to | |
1966 | * allocate the first block. | |
1967 | * | |
1968 | * Optimize future alignment probing; ignore failures. | |
1969 | */ | |
1970 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1971 | } |
1972 | } else { | |
1973 | result = 0; | |
1974 | } | |
1975 | goto out; | |
1976 | #endif | |
1977 | case PREALLOC_MODE_FULL: | |
1978 | { | |
1979 | int64_t num = 0, left = offset - current_length; | |
1980 | off_t seek_result; | |
1981 | ||
1982 | /* | |
1983 | * Knowing the final size from the beginning could allow the file | |
1984 | * system driver to do less allocations and possibly avoid | |
1985 | * fragmentation of the file. | |
1986 | */ | |
1987 | if (ftruncate(fd, offset) != 0) { | |
1988 | result = -errno; | |
1989 | error_setg_errno(errp, -result, "Could not resize file"); | |
1990 | goto out; | |
1991 | } | |
1992 | ||
1993 | buf = g_malloc0(65536); | |
1994 | ||
1995 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1996 | if (seek_result < 0) { | |
1997 | result = -errno; | |
1998 | error_setg_errno(errp, -result, | |
1999 | "Failed to seek to the old end of file"); | |
2000 | goto out; | |
2001 | } | |
2002 | ||
2003 | while (left > 0) { | |
2004 | num = MIN(left, 65536); | |
2005 | result = write(fd, buf, num); | |
2006 | if (result < 0) { | |
a1c81f4f FZ |
2007 | if (errno == EINTR) { |
2008 | continue; | |
2009 | } | |
93f4e2ff KW |
2010 | result = -errno; |
2011 | error_setg_errno(errp, -result, | |
2012 | "Could not write zeros for preallocation"); | |
2013 | goto out; | |
2014 | } | |
2015 | left -= result; | |
2016 | } | |
2017 | if (result >= 0) { | |
2018 | result = fsync(fd); | |
2019 | if (result < 0) { | |
2020 | result = -errno; | |
2021 | error_setg_errno(errp, -result, | |
2022 | "Could not flush file to disk"); | |
2023 | goto out; | |
2024 | } | |
2025 | } | |
2026 | goto out; | |
2027 | } | |
2028 | case PREALLOC_MODE_OFF: | |
2029 | if (ftruncate(fd, offset) != 0) { | |
2030 | result = -errno; | |
2031 | error_setg_errno(errp, -result, "Could not resize file"); | |
3a20013f NS |
2032 | } else if (current_length == 0 && offset > current_length) { |
2033 | /* Optimize future alignment probing; ignore failures. */ | |
2034 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
2035 | } |
2036 | return result; | |
2037 | default: | |
2038 | result = -ENOTSUP; | |
2039 | error_setg(errp, "Unsupported preallocation mode: %s", | |
d57c44d0 | 2040 | PreallocMode_str(prealloc)); |
93f4e2ff KW |
2041 | return result; |
2042 | } | |
2043 | ||
2044 | out: | |
2045 | if (result < 0) { | |
2046 | if (ftruncate(fd, current_length) < 0) { | |
2047 | error_report("Failed to restore old file length: %s", | |
2048 | strerror(errno)); | |
2049 | } | |
2050 | } | |
2051 | ||
2052 | g_free(buf); | |
2053 | return result; | |
2054 | } | |
2055 | ||
5d5de250 KW |
2056 | static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, |
2057 | ThreadPoolFunc func, void *arg) | |
2058 | { | |
2059 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ | |
2060 | ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
2061 | return thread_pool_submit_co(pool, func, arg); | |
2062 | } | |
2063 | ||
9d52aa3c KW |
2064 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
2065 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 2066 | { |
ce1a14dc | 2067 | BDRVRawState *s = bs->opaque; |
999e6b69 | 2068 | RawPosixAIOData acb; |
ce1a14dc | 2069 | |
19cb3738 | 2070 | if (fd_open(bs) < 0) |
2174f12b | 2071 | return -EIO; |
19cb3738 | 2072 | |
f141eafe | 2073 | /* |
c6447510 AM |
2074 | * When using O_DIRECT, the request must be aligned to be able to use |
2075 | * either libaio or io_uring interface. If not fail back to regular thread | |
2076 | * pool read/write code which emulates this for us if we | |
2077 | * set QEMU_AIO_MISALIGNED. | |
f141eafe | 2078 | */ |
c6447510 AM |
2079 | if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { |
2080 | type |= QEMU_AIO_MISALIGNED; | |
2081 | #ifdef CONFIG_LINUX_IO_URING | |
2082 | } else if (s->use_linux_io_uring) { | |
2083 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2084 | assert(qiov->size == bytes); | |
2085 | return luring_co_submit(bs, aio, s->fd, offset, qiov, type); | |
2086 | #endif | |
e44bd6fc | 2087 | #ifdef CONFIG_LINUX_AIO |
c6447510 AM |
2088 | } else if (s->use_linux_aio) { |
2089 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); | |
2090 | assert(qiov->size == bytes); | |
512da211 SG |
2091 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type, |
2092 | s->aio_max_batch); | |
e44bd6fc | 2093 | #endif |
9ef91a67 | 2094 | } |
f141eafe | 2095 | |
999e6b69 KW |
2096 | acb = (RawPosixAIOData) { |
2097 | .bs = bs, | |
2098 | .aio_fildes = s->fd, | |
2099 | .aio_type = type, | |
2100 | .aio_offset = offset, | |
2101 | .aio_nbytes = bytes, | |
2102 | .io = { | |
2103 | .iov = qiov->iov, | |
2104 | .niov = qiov->niov, | |
2105 | }, | |
2106 | }; | |
2107 | ||
2108 | assert(qiov->size == bytes); | |
2109 | return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); | |
2174f12b KW |
2110 | } |
2111 | ||
f7ef38dd VSO |
2112 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, int64_t offset, |
2113 | int64_t bytes, QEMUIOVector *qiov, | |
2114 | BdrvRequestFlags flags) | |
2174f12b | 2115 | { |
9d52aa3c | 2116 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
2117 | } |
2118 | ||
e75abeda VSO |
2119 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, int64_t offset, |
2120 | int64_t bytes, QEMUIOVector *qiov, | |
2121 | BdrvRequestFlags flags) | |
2174f12b | 2122 | { |
9d52aa3c KW |
2123 | assert(flags == 0); |
2124 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
2125 | } |
2126 | ||
1b3abdcc ML |
2127 | static void raw_aio_plug(BlockDriverState *bs) |
2128 | { | |
c6447510 | 2129 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2130 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2131 | if (s->use_linux_aio) { |
0187f5c9 PB |
2132 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
2133 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
2134 | } |
2135 | #endif | |
c6447510 AM |
2136 | #ifdef CONFIG_LINUX_IO_URING |
2137 | if (s->use_linux_io_uring) { | |
2138 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2139 | luring_io_plug(bs, aio); | |
2140 | } | |
2141 | #endif | |
1b3abdcc ML |
2142 | } |
2143 | ||
2144 | static void raw_aio_unplug(BlockDriverState *bs) | |
2145 | { | |
c6447510 | 2146 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2147 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2148 | if (s->use_linux_aio) { |
0187f5c9 | 2149 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
68d79466 | 2150 | laio_io_unplug(bs, aio, s->aio_max_batch); |
1b3abdcc ML |
2151 | } |
2152 | #endif | |
c6447510 AM |
2153 | #ifdef CONFIG_LINUX_IO_URING |
2154 | if (s->use_linux_io_uring) { | |
2155 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2156 | luring_io_unplug(bs, aio); | |
2157 | } | |
2158 | #endif | |
1b3abdcc ML |
2159 | } |
2160 | ||
33d70fb6 | 2161 | static int raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
2162 | { |
2163 | BDRVRawState *s = bs->opaque; | |
06dc9bd5 | 2164 | RawPosixAIOData acb; |
33d70fb6 | 2165 | int ret; |
b2e12bc6 | 2166 | |
33d70fb6 KW |
2167 | ret = fd_open(bs); |
2168 | if (ret < 0) { | |
2169 | return ret; | |
2170 | } | |
b2e12bc6 | 2171 | |
06dc9bd5 KW |
2172 | acb = (RawPosixAIOData) { |
2173 | .bs = bs, | |
2174 | .aio_fildes = s->fd, | |
2175 | .aio_type = QEMU_AIO_FLUSH, | |
2176 | }; | |
2177 | ||
c6447510 AM |
2178 | #ifdef CONFIG_LINUX_IO_URING |
2179 | if (s->use_linux_io_uring) { | |
2180 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2181 | return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); | |
2182 | } | |
2183 | #endif | |
06dc9bd5 | 2184 | return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); |
b2e12bc6 CH |
2185 | } |
2186 | ||
ed6e2161 NA |
2187 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
2188 | AioContext *new_context) | |
2189 | { | |
c6447510 | 2190 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
ed6e2161 | 2191 | #ifdef CONFIG_LINUX_AIO |
ed6e2161 | 2192 | if (s->use_linux_aio) { |
cb09104e | 2193 | Error *local_err = NULL; |
ed6e2161 NA |
2194 | if (!aio_setup_linux_aio(new_context, &local_err)) { |
2195 | error_reportf_err(local_err, "Unable to use native AIO, " | |
2196 | "falling back to thread pool: "); | |
2197 | s->use_linux_aio = false; | |
2198 | } | |
2199 | } | |
2200 | #endif | |
c6447510 AM |
2201 | #ifdef CONFIG_LINUX_IO_URING |
2202 | if (s->use_linux_io_uring) { | |
cb8d0851 | 2203 | Error *local_err = NULL; |
c6447510 AM |
2204 | if (!aio_setup_linux_io_uring(new_context, &local_err)) { |
2205 | error_reportf_err(local_err, "Unable to use linux io_uring, " | |
2206 | "falling back to thread pool: "); | |
2207 | s->use_linux_io_uring = false; | |
2208 | } | |
2209 | } | |
2210 | #endif | |
ed6e2161 NA |
2211 | } |
2212 | ||
83f64091 FB |
2213 | static void raw_close(BlockDriverState *bs) |
2214 | { | |
2215 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 2216 | |
19cb3738 | 2217 | if (s->fd >= 0) { |
2e1e79da | 2218 | qemu_close(s->fd); |
19cb3738 FB |
2219 | s->fd = -1; |
2220 | } | |
83f64091 FB |
2221 | } |
2222 | ||
d0bc9e5d HR |
2223 | /** |
2224 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
2225 | * new space according to @prealloc. | |
2226 | * | |
2227 | * Returns: 0 on success, -errno on failure. | |
2228 | */ | |
93f4e2ff KW |
2229 | static int coroutine_fn |
2230 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
2231 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 2232 | { |
29cb4c01 | 2233 | RawPosixAIOData acb; |
d0bc9e5d | 2234 | |
29cb4c01 | 2235 | acb = (RawPosixAIOData) { |
93f4e2ff KW |
2236 | .bs = bs, |
2237 | .aio_fildes = fd, | |
2238 | .aio_type = QEMU_AIO_TRUNCATE, | |
2239 | .aio_offset = offset, | |
d57c44d0 KW |
2240 | .truncate = { |
2241 | .prealloc = prealloc, | |
2242 | .errp = errp, | |
2243 | }, | |
93f4e2ff | 2244 | }; |
d0bc9e5d | 2245 | |
29cb4c01 | 2246 | return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); |
9f63b07e HR |
2247 | } |
2248 | ||
061ca8a3 | 2249 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
c80d8b06 | 2250 | bool exact, PreallocMode prealloc, |
92b92799 | 2251 | BdrvRequestFlags flags, Error **errp) |
83f64091 FB |
2252 | { |
2253 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 2254 | struct stat st; |
f59adb32 | 2255 | int ret; |
55b949c8 CH |
2256 | |
2257 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
2258 | ret = -errno; |
2259 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
2260 | return ret; | |
55b949c8 CH |
2261 | } |
2262 | ||
2263 | if (S_ISREG(st.st_mode)) { | |
82325ae5 | 2264 | /* Always resizes to the exact @offset */ |
93f4e2ff | 2265 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
2266 | } |
2267 | ||
2268 | if (prealloc != PREALLOC_MODE_OFF) { | |
2269 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 2270 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
2271 | return -ENOTSUP; |
2272 | } | |
2273 | ||
2274 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
82325ae5 HR |
2275 | int64_t cur_length = raw_getlength(bs); |
2276 | ||
2277 | if (offset != cur_length && exact) { | |
2278 | error_setg(errp, "Cannot resize device files"); | |
2279 | return -ENOTSUP; | |
2280 | } else if (offset > cur_length) { | |
f59adb32 HR |
2281 | error_setg(errp, "Cannot grow device files"); |
2282 | return -EINVAL; | |
2283 | } | |
55b949c8 | 2284 | } else { |
f59adb32 | 2285 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
2286 | return -ENOTSUP; |
2287 | } | |
2288 | ||
83f64091 FB |
2289 | return 0; |
2290 | } | |
2291 | ||
128ab2ff BS |
2292 | #ifdef __OpenBSD__ |
2293 | static int64_t raw_getlength(BlockDriverState *bs) | |
2294 | { | |
2295 | BDRVRawState *s = bs->opaque; | |
2296 | int fd = s->fd; | |
2297 | struct stat st; | |
2298 | ||
2299 | if (fstat(fd, &st)) | |
aa729704 | 2300 | return -errno; |
128ab2ff BS |
2301 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2302 | struct disklabel dl; | |
2303 | ||
2304 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2305 | return -errno; |
128ab2ff BS |
2306 | return (uint64_t)dl.d_secsize * |
2307 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2308 | } else | |
2309 | return st.st_size; | |
2310 | } | |
d1f6fd8d CE |
2311 | #elif defined(__NetBSD__) |
2312 | static int64_t raw_getlength(BlockDriverState *bs) | |
2313 | { | |
2314 | BDRVRawState *s = bs->opaque; | |
2315 | int fd = s->fd; | |
2316 | struct stat st; | |
2317 | ||
2318 | if (fstat(fd, &st)) | |
aa729704 | 2319 | return -errno; |
d1f6fd8d CE |
2320 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2321 | struct dkwedge_info dkw; | |
2322 | ||
2323 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
2324 | return dkw.dkw_size * 512; | |
2325 | } else { | |
2326 | struct disklabel dl; | |
2327 | ||
2328 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2329 | return -errno; |
d1f6fd8d CE |
2330 | return (uint64_t)dl.d_secsize * |
2331 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2332 | } | |
2333 | } else | |
2334 | return st.st_size; | |
2335 | } | |
50779cc2 CH |
2336 | #elif defined(__sun__) |
2337 | static int64_t raw_getlength(BlockDriverState *bs) | |
2338 | { | |
2339 | BDRVRawState *s = bs->opaque; | |
2340 | struct dk_minfo minfo; | |
2341 | int ret; | |
aa729704 | 2342 | int64_t size; |
50779cc2 CH |
2343 | |
2344 | ret = fd_open(bs); | |
2345 | if (ret < 0) { | |
2346 | return ret; | |
2347 | } | |
2348 | ||
2349 | /* | |
2350 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
2351 | */ | |
2352 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
2353 | if (ret != -1) { | |
2354 | return minfo.dki_lbsize * minfo.dki_capacity; | |
2355 | } | |
2356 | ||
2357 | /* | |
2358 | * There are reports that lseek on some devices fails, but | |
2359 | * irc discussion said that contingency on contingency was overkill. | |
2360 | */ | |
aa729704 MA |
2361 | size = lseek(s->fd, 0, SEEK_END); |
2362 | if (size < 0) { | |
2363 | return -errno; | |
2364 | } | |
2365 | return size; | |
50779cc2 CH |
2366 | } |
2367 | #elif defined(CONFIG_BSD) | |
2368 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2369 | { |
2370 | BDRVRawState *s = bs->opaque; | |
2371 | int fd = s->fd; | |
2372 | int64_t size; | |
83f64091 | 2373 | struct stat sb; |
a167ba50 | 2374 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2375 | int reopened = 0; |
83f64091 | 2376 | #endif |
19cb3738 FB |
2377 | int ret; |
2378 | ||
2379 | ret = fd_open(bs); | |
2380 | if (ret < 0) | |
2381 | return ret; | |
83f64091 | 2382 | |
a167ba50 | 2383 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2384 | again: |
2385 | #endif | |
83f64091 | 2386 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
267cd53f | 2387 | size = 0; |
83f64091 | 2388 | #ifdef DIOCGMEDIASIZE |
267cd53f PB |
2389 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) { |
2390 | size = 0; | |
2391 | } | |
2392 | #endif | |
2393 | #ifdef DIOCGPART | |
2394 | if (size == 0) { | |
2395 | struct partinfo pi; | |
2396 | if (ioctl(fd, DIOCGPART, &pi) == 0) { | |
2397 | size = pi.media_size; | |
2398 | } | |
c5e97233 | 2399 | } |
83f64091 | 2400 | #endif |
09e20abd | 2401 | #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE) |
267cd53f | 2402 | if (size == 0) { |
728dacbd JA |
2403 | uint64_t sectors = 0; |
2404 | uint32_t sector_size = 0; | |
2405 | ||
2406 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2407 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2408 | size = sectors * sector_size; | |
728dacbd JA |
2409 | } |
2410 | } | |
267cd53f PB |
2411 | #endif |
2412 | if (size == 0) { | |
2413 | size = lseek(fd, 0LL, SEEK_END); | |
2414 | } | |
aa729704 MA |
2415 | if (size < 0) { |
2416 | return -errno; | |
2417 | } | |
a167ba50 | 2418 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2419 | switch(s->type) { |
2420 | case FTYPE_CD: | |
2421 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2422 | if (size == 2048LL * (unsigned)-1) | |
2423 | size = 0; | |
2424 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2425 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2426 | reopened = 1; |
2427 | goto again; | |
2428 | } | |
2429 | } | |
83f64091 | 2430 | #endif |
50779cc2 | 2431 | } else { |
83f64091 | 2432 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2433 | if (size < 0) { |
2434 | return -errno; | |
2435 | } | |
83f64091 | 2436 | } |
83f64091 FB |
2437 | return size; |
2438 | } | |
50779cc2 CH |
2439 | #else |
2440 | static int64_t raw_getlength(BlockDriverState *bs) | |
2441 | { | |
2442 | BDRVRawState *s = bs->opaque; | |
2443 | int ret; | |
aa729704 | 2444 | int64_t size; |
50779cc2 CH |
2445 | |
2446 | ret = fd_open(bs); | |
2447 | if (ret < 0) { | |
2448 | return ret; | |
2449 | } | |
2450 | ||
aa729704 MA |
2451 | size = lseek(s->fd, 0, SEEK_END); |
2452 | if (size < 0) { | |
2453 | return -errno; | |
2454 | } | |
2455 | return size; | |
50779cc2 | 2456 | } |
128ab2ff | 2457 | #endif |
83f64091 | 2458 | |
4a1d5e1f FZ |
2459 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2460 | { | |
2461 | struct stat st; | |
2462 | BDRVRawState *s = bs->opaque; | |
2463 | ||
2464 | if (fstat(s->fd, &st) < 0) { | |
2465 | return -errno; | |
2466 | } | |
2467 | return (int64_t)st.st_blocks * 512; | |
2468 | } | |
2469 | ||
93f4e2ff KW |
2470 | static int coroutine_fn |
2471 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2472 | { |
927f11e1 | 2473 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2474 | Error *local_err = NULL; |
83f64091 | 2475 | int fd; |
d815efca | 2476 | uint64_t perm, shared; |
1e37d059 | 2477 | int result = 0; |
83f64091 | 2478 | |
927f11e1 KW |
2479 | /* Validate options and set default values */ |
2480 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2481 | file_opts = &options->u.file; | |
464d9f64 | 2482 | |
927f11e1 KW |
2483 | if (!file_opts->has_nocow) { |
2484 | file_opts->nocow = false; | |
2485 | } | |
2486 | if (!file_opts->has_preallocation) { | |
2487 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2488 | } |
ffa244c8 KW |
2489 | if (!file_opts->has_extent_size_hint) { |
2490 | file_opts->extent_size_hint = 1 * MiB; | |
2491 | } | |
2492 | if (file_opts->extent_size_hint > UINT32_MAX) { | |
2493 | result = -EINVAL; | |
2494 | error_setg(errp, "Extent size hint is too large"); | |
2495 | goto out; | |
2496 | } | |
83f64091 | 2497 | |
927f11e1 | 2498 | /* Create file */ |
b18a24a9 | 2499 | fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); |
1e37d059 SW |
2500 | if (fd < 0) { |
2501 | result = -errno; | |
06247428 HT |
2502 | goto out; |
2503 | } | |
2504 | ||
b8cf1913 HR |
2505 | /* Take permissions: We want to discard everything, so we need |
2506 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2507 | * BLK_PERM_RESIZE. | |
2508 | * On the other hand, we cannot share the RESIZE permission | |
2509 | * because we promise that after this function, the file has the | |
2510 | * size given in the options. If someone else were to resize it | |
2511 | * concurrently, we could not guarantee that. | |
2512 | * Note that after this function, we can no longer guarantee that | |
2513 | * the file is not touched by a third party, so it may be resized | |
2514 | * then. */ | |
2515 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2516 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2517 | ||
2518 | /* Step one: Take locks */ | |
2996ffad | 2519 | result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2520 | if (result < 0) { |
2521 | goto out_close; | |
2522 | } | |
2523 | ||
2524 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2525 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2526 | if (result < 0) { | |
b857431d FZ |
2527 | error_append_hint(errp, |
2528 | "Is another process using the image [%s]?\n", | |
2529 | file_opts->filename); | |
7c20c808 | 2530 | goto out_unlock; |
b8cf1913 HR |
2531 | } |
2532 | ||
2533 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2534 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2535 | if (result < 0) { |
7c20c808 | 2536 | goto out_unlock; |
b8cf1913 HR |
2537 | } |
2538 | ||
927f11e1 | 2539 | if (file_opts->nocow) { |
4ab15590 | 2540 | #ifdef __linux__ |
06247428 HT |
2541 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2542 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2543 | * will be ignored since any failure of this operation should not | |
2544 | * block the left work. | |
2545 | */ | |
2546 | int attr; | |
2547 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2548 | attr |= FS_NOCOW_FL; | |
2549 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2550 | } |
06247428 HT |
2551 | #endif |
2552 | } | |
ffa244c8 KW |
2553 | #ifdef FS_IOC_FSSETXATTR |
2554 | /* | |
2555 | * Try to set the extent size hint. Failure is not fatal, and a warning is | |
2556 | * only printed if the option was explicitly specified. | |
2557 | */ | |
2558 | { | |
2559 | struct fsxattr attr; | |
2560 | result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); | |
2561 | if (result == 0) { | |
2562 | attr.fsx_xflags |= FS_XFLAG_EXTSIZE; | |
2563 | attr.fsx_extsize = file_opts->extent_size_hint; | |
2564 | result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); | |
2565 | } | |
2566 | if (result < 0 && file_opts->has_extent_size_hint && | |
2567 | file_opts->extent_size_hint) | |
2568 | { | |
2569 | warn_report("Failed to set extent size hint: %s", | |
2570 | strerror(errno)); | |
2571 | } | |
2572 | } | |
2573 | #endif | |
06247428 | 2574 | |
b8cf1913 HR |
2575 | /* Resize and potentially preallocate the file to the desired |
2576 | * final size */ | |
93f4e2ff KW |
2577 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2578 | file_opts->preallocation, errp); | |
9f63b07e | 2579 | if (result < 0) { |
7c20c808 HR |
2580 | goto out_unlock; |
2581 | } | |
2582 | ||
2583 | out_unlock: | |
2996ffad | 2584 | raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); |
7c20c808 HR |
2585 | if (local_err) { |
2586 | /* The above call should not fail, and if it does, that does | |
2587 | * not mean the whole creation operation has failed. So | |
2588 | * report it the user for their convenience, but do not report | |
2589 | * it to the caller. */ | |
db0754df | 2590 | warn_report_err(local_err); |
1e37d059 | 2591 | } |
06247428 | 2592 | |
5a1dad9d | 2593 | out_close: |
06247428 HT |
2594 | if (qemu_close(fd) != 0 && result == 0) { |
2595 | result = -errno; | |
2596 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2597 | } | |
2598 | out: | |
1e37d059 | 2599 | return result; |
83f64091 FB |
2600 | } |
2601 | ||
b92902df ML |
2602 | static int coroutine_fn raw_co_create_opts(BlockDriver *drv, |
2603 | const char *filename, | |
2604 | QemuOpts *opts, | |
927f11e1 KW |
2605 | Error **errp) |
2606 | { | |
2607 | BlockdevCreateOptions options; | |
2608 | int64_t total_size = 0; | |
ffa244c8 KW |
2609 | int64_t extent_size_hint = 0; |
2610 | bool has_extent_size_hint = false; | |
927f11e1 KW |
2611 | bool nocow = false; |
2612 | PreallocMode prealloc; | |
2613 | char *buf = NULL; | |
2614 | Error *local_err = NULL; | |
2615 | ||
2616 | /* Skip file: protocol prefix */ | |
2617 | strstart(filename, "file:", &filename); | |
2618 | ||
2619 | /* Read out options */ | |
2620 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2621 | BDRV_SECTOR_SIZE); | |
ffa244c8 KW |
2622 | if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { |
2623 | has_extent_size_hint = true; | |
2624 | extent_size_hint = | |
2625 | qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); | |
2626 | } | |
927f11e1 KW |
2627 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); |
2628 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2629 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2630 | PREALLOC_MODE_OFF, &local_err); | |
2631 | g_free(buf); | |
2632 | if (local_err) { | |
2633 | error_propagate(errp, local_err); | |
2634 | return -EINVAL; | |
2635 | } | |
2636 | ||
2637 | options = (BlockdevCreateOptions) { | |
2638 | .driver = BLOCKDEV_DRIVER_FILE, | |
2639 | .u.file = { | |
2640 | .filename = (char *) filename, | |
2641 | .size = total_size, | |
2642 | .has_preallocation = true, | |
2643 | .preallocation = prealloc, | |
2644 | .has_nocow = true, | |
2645 | .nocow = nocow, | |
ffa244c8 KW |
2646 | .has_extent_size_hint = has_extent_size_hint, |
2647 | .extent_size_hint = extent_size_hint, | |
927f11e1 KW |
2648 | }, |
2649 | }; | |
2650 | return raw_co_create(&options, errp); | |
2651 | } | |
2652 | ||
9bffae14 DHB |
2653 | static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, |
2654 | Error **errp) | |
2655 | { | |
2656 | struct stat st; | |
2657 | int ret; | |
2658 | ||
2659 | if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { | |
2660 | error_setg_errno(errp, ENOENT, "%s is not a regular file", | |
2661 | bs->filename); | |
2662 | return -ENOENT; | |
2663 | } | |
2664 | ||
2665 | ret = unlink(bs->filename); | |
2666 | if (ret < 0) { | |
2667 | ret = -errno; | |
2668 | error_setg_errno(errp, -ret, "Error when deleting file %s", | |
2669 | bs->filename); | |
2670 | } | |
2671 | ||
2672 | return ret; | |
2673 | } | |
2674 | ||
d1f06fe6 MA |
2675 | /* |
2676 | * Find allocation range in @bs around offset @start. | |
2677 | * May change underlying file descriptor's file offset. | |
2678 | * If @start is not in a hole, store @start in @data, and the | |
2679 | * beginning of the next hole in @hole, and return 0. | |
2680 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2681 | * beginning of the next non-hole in @data, and return 0. | |
2682 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2683 | * If we can't find out, return a negative errno other than -ENXIO. | |
2684 | */ | |
2685 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2686 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2687 | { |
2688 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2689 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2690 | off_t offs; |
94282e71 | 2691 | |
d1f06fe6 MA |
2692 | /* |
2693 | * SEEK_DATA cases: | |
2694 | * D1. offs == start: start is in data | |
2695 | * D2. offs > start: start is in a hole, next data at offs | |
2696 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2697 | * or start is beyond EOF | |
2698 | * If the latter happens, the file has been truncated behind | |
2699 | * our back since we opened it. All bets are off then. | |
2700 | * Treating like a trailing hole is simplest. | |
2701 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2702 | */ | |
2703 | offs = lseek(s->fd, start, SEEK_DATA); | |
2704 | if (offs < 0) { | |
2705 | return -errno; /* D3 or D4 */ | |
2706 | } | |
a03083a0 JC |
2707 | |
2708 | if (offs < start) { | |
2709 | /* This is not a valid return by lseek(). We are safe to just return | |
2710 | * -EIO in this case, and we'll treat it like D4. */ | |
2711 | return -EIO; | |
2712 | } | |
d1f06fe6 MA |
2713 | |
2714 | if (offs > start) { | |
2715 | /* D2: in hole, next data at offs */ | |
2716 | *hole = start; | |
2717 | *data = offs; | |
2718 | return 0; | |
5500316d PB |
2719 | } |
2720 | ||
d1f06fe6 MA |
2721 | /* D1: in data, end not yet known */ |
2722 | ||
2723 | /* | |
2724 | * SEEK_HOLE cases: | |
2725 | * H1. offs == start: start is in a hole | |
2726 | * If this happens here, a hole has been dug behind our back | |
2727 | * since the previous lseek(). | |
2728 | * H2. offs > start: either start is in data, next hole at offs, | |
2729 | * or start is in trailing hole, EOF at offs | |
2730 | * Linux treats trailing holes like any other hole: offs == | |
2731 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2732 | * If that happens here, a hole has been dug behind our back | |
2733 | * since the previous lseek(). | |
2734 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2735 | * If this happens, the file has been truncated behind our | |
2736 | * back since we opened it. Treat it like a trailing hole. | |
2737 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2738 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2739 | */ | |
2740 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2741 | if (offs < 0) { | |
2742 | return -errno; /* D1 and (H3 or H4) */ | |
2743 | } | |
a03083a0 JC |
2744 | |
2745 | if (offs < start) { | |
2746 | /* This is not a valid return by lseek(). We are safe to just return | |
2747 | * -EIO in this case, and we'll treat it like H4. */ | |
2748 | return -EIO; | |
2749 | } | |
d1f06fe6 MA |
2750 | |
2751 | if (offs > start) { | |
2752 | /* | |
2753 | * D1 and H2: either in data, next hole at offs, or it was in | |
2754 | * data but is now in a trailing hole. In the latter case, | |
2755 | * all bets are off. Treating it as if it there was data all | |
2756 | * the way to EOF is safe, so simply do that. | |
2757 | */ | |
4f11aa8a | 2758 | *data = start; |
d1f06fe6 MA |
2759 | *hole = offs; |
2760 | return 0; | |
5500316d | 2761 | } |
4f11aa8a | 2762 | |
d1f06fe6 MA |
2763 | /* D1 and H1 */ |
2764 | return -EBUSY; | |
5500316d | 2765 | #else |
4f11aa8a | 2766 | return -ENOTSUP; |
5500316d | 2767 | #endif |
4f11aa8a HR |
2768 | } |
2769 | ||
2770 | /* | |
a290f085 | 2771 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2772 | * |
a290f085 | 2773 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2774 | * |
a290f085 EB |
2775 | * 'pnum' is set to the number of bytes (including and immediately following |
2776 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2777 | * allocated/unallocated state. |
2778 | * | |
869e7ee8 HR |
2779 | * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may |
2780 | * well exceed it. | |
4f11aa8a | 2781 | */ |
a290f085 EB |
2782 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2783 | bool want_zero, | |
2784 | int64_t offset, | |
2785 | int64_t bytes, int64_t *pnum, | |
2786 | int64_t *map, | |
2787 | BlockDriverState **file) | |
2788 | { | |
2789 | off_t data = 0, hole = 0; | |
d7f62751 | 2790 | int ret; |
4f11aa8a | 2791 | |
9c3db310 HR |
2792 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); |
2793 | ||
4f11aa8a HR |
2794 | ret = fd_open(bs); |
2795 | if (ret < 0) { | |
2796 | return ret; | |
2797 | } | |
2798 | ||
a290f085 EB |
2799 | if (!want_zero) { |
2800 | *pnum = bytes; | |
2801 | *map = offset; | |
2802 | *file = bs; | |
2803 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2804 | } |
4f11aa8a | 2805 | |
a290f085 | 2806 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2807 | if (ret == -ENXIO) { |
2808 | /* Trailing hole */ | |
a290f085 | 2809 | *pnum = bytes; |
d1f06fe6 MA |
2810 | ret = BDRV_BLOCK_ZERO; |
2811 | } else if (ret < 0) { | |
2812 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2813 | *pnum = bytes; |
d1f06fe6 | 2814 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2815 | } else if (data == offset) { |
2816 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2817 | * possibly including a partial sector at EOF. */ |
869e7ee8 | 2818 | *pnum = hole - offset; |
9c3db310 HR |
2819 | |
2820 | /* | |
2821 | * We are not allowed to return partial sectors, though, so | |
2822 | * round up if necessary. | |
2823 | */ | |
2824 | if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { | |
2825 | int64_t file_length = raw_getlength(bs); | |
2826 | if (file_length > 0) { | |
2827 | /* Ignore errors, this is just a safeguard */ | |
2828 | assert(hole == file_length); | |
2829 | } | |
2830 | *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); | |
2831 | } | |
2832 | ||
d1f06fe6 | 2833 | ret = BDRV_BLOCK_DATA; |
5500316d | 2834 | } else { |
a290f085 EB |
2835 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2836 | assert(hole == offset); | |
869e7ee8 | 2837 | *pnum = data - offset; |
d1f06fe6 | 2838 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2839 | } |
a290f085 | 2840 | *map = offset; |
02650acb | 2841 | *file = bs; |
a290f085 | 2842 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2843 | } |
2844 | ||
31be8a2a SH |
2845 | #if defined(__linux__) |
2846 | /* Verify that the file is not in the page cache */ | |
2847 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2848 | { | |
2849 | const size_t window_size = 128 * 1024 * 1024; | |
2850 | BDRVRawState *s = bs->opaque; | |
2851 | void *window = NULL; | |
2852 | size_t length = 0; | |
2853 | unsigned char *vec; | |
2854 | size_t page_size; | |
2855 | off_t offset; | |
2856 | off_t end; | |
2857 | ||
2858 | /* mincore(2) page status information requires 1 byte per page */ | |
2859 | page_size = sysconf(_SC_PAGESIZE); | |
2860 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2861 | ||
2862 | end = raw_getlength(bs); | |
2863 | ||
2864 | for (offset = 0; offset < end; offset += window_size) { | |
2865 | void *new_window; | |
2866 | size_t new_length; | |
2867 | size_t vec_end; | |
2868 | size_t i; | |
2869 | int ret; | |
2870 | ||
2871 | /* Unmap previous window if size has changed */ | |
2872 | new_length = MIN(end - offset, window_size); | |
2873 | if (new_length != length) { | |
2874 | munmap(window, length); | |
2875 | window = NULL; | |
2876 | length = 0; | |
2877 | } | |
2878 | ||
2879 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2880 | s->fd, offset); | |
2881 | if (new_window == MAP_FAILED) { | |
2882 | error_setg_errno(errp, errno, "mmap failed"); | |
2883 | break; | |
2884 | } | |
2885 | ||
2886 | window = new_window; | |
2887 | length = new_length; | |
2888 | ||
2889 | ret = mincore(window, length, vec); | |
2890 | if (ret < 0) { | |
2891 | error_setg_errno(errp, errno, "mincore failed"); | |
2892 | break; | |
2893 | } | |
2894 | ||
2895 | vec_end = DIV_ROUND_UP(length, page_size); | |
2896 | for (i = 0; i < vec_end; i++) { | |
2897 | if (vec[i] & 0x1) { | |
31be8a2a SH |
2898 | break; |
2899 | } | |
2900 | } | |
77ed971b MA |
2901 | if (i < vec_end) { |
2902 | error_setg(errp, "page cache still in use!"); | |
2903 | break; | |
2904 | } | |
31be8a2a SH |
2905 | } |
2906 | ||
2907 | if (window) { | |
2908 | munmap(window, length); | |
2909 | } | |
2910 | ||
2911 | g_free(vec); | |
2912 | } | |
2913 | #endif /* __linux__ */ | |
2914 | ||
dd577a26 SH |
2915 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2916 | Error **errp) | |
2917 | { | |
2918 | BDRVRawState *s = bs->opaque; | |
2919 | int ret; | |
2920 | ||
2921 | ret = fd_open(bs); | |
2922 | if (ret < 0) { | |
2923 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2924 | return; | |
2925 | } | |
2926 | ||
f357fcd8 SH |
2927 | if (!s->drop_cache) { |
2928 | return; | |
2929 | } | |
2930 | ||
dd577a26 SH |
2931 | if (s->open_flags & O_DIRECT) { |
2932 | return; /* No host kernel page cache */ | |
2933 | } | |
2934 | ||
2935 | #if defined(__linux__) | |
2936 | /* This sets the scene for the next syscall... */ | |
2937 | ret = bdrv_co_flush(bs); | |
2938 | if (ret < 0) { | |
2939 | error_setg_errno(errp, -ret, "flush failed"); | |
2940 | return; | |
2941 | } | |
2942 | ||
2943 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2944 | * process. These limitations are okay because we just fsynced the file, | |
2945 | * we don't use mmap, and the file should not be in use by other processes. | |
2946 | */ | |
2947 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2948 | if (ret != 0) { /* the return value is a positive errno */ | |
2949 | error_setg_errno(errp, ret, "fadvise failed"); | |
2950 | return; | |
2951 | } | |
31be8a2a SH |
2952 | |
2953 | if (s->check_cache_dropped) { | |
2954 | check_cache_dropped(bs, errp); | |
2955 | } | |
dd577a26 SH |
2956 | #else /* __linux__ */ |
2957 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2958 | * unsupported on other host operating systems. Cache consistency issues | |
2959 | * may occur but no error is reported here, partly because that's the | |
2960 | * historical behavior and partly because it's hard to differentiate valid | |
2961 | * configurations that should not cause errors. | |
2962 | */ | |
2963 | #endif /* !__linux__ */ | |
2964 | } | |
2965 | ||
1c450366 AN |
2966 | static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) |
2967 | { | |
2968 | if (ret) { | |
2969 | s->stats.discard_nb_failed++; | |
2970 | } else { | |
2971 | s->stats.discard_nb_ok++; | |
2972 | s->stats.discard_bytes_ok += nbytes; | |
2973 | } | |
2974 | } | |
2975 | ||
33d70fb6 | 2976 | static coroutine_fn int |
0c802287 VSO |
2977 | raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes, |
2978 | bool blkdev) | |
dce512de | 2979 | { |
dce512de | 2980 | BDRVRawState *s = bs->opaque; |
46ee0f46 | 2981 | RawPosixAIOData acb; |
1c450366 | 2982 | int ret; |
46ee0f46 KW |
2983 | |
2984 | acb = (RawPosixAIOData) { | |
2985 | .bs = bs, | |
2986 | .aio_fildes = s->fd, | |
2987 | .aio_type = QEMU_AIO_DISCARD, | |
2988 | .aio_offset = offset, | |
2989 | .aio_nbytes = bytes, | |
2990 | }; | |
dce512de | 2991 | |
46ee0f46 KW |
2992 | if (blkdev) { |
2993 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2994 | } | |
2995 | ||
1c450366 AN |
2996 | ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); |
2997 | raw_account_discard(s, bytes, ret); | |
2998 | return ret; | |
46ee0f46 KW |
2999 | } |
3000 | ||
3001 | static coroutine_fn int | |
0c802287 | 3002 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
46ee0f46 KW |
3003 | { |
3004 | return raw_do_pdiscard(bs, offset, bytes, false); | |
dce512de | 3005 | } |
0e7e1989 | 3006 | |
7154d8ae | 3007 | static int coroutine_fn |
f34b2bcf | 3008 | raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, |
7154d8ae | 3009 | BdrvRequestFlags flags, bool blkdev) |
260a82e5 PB |
3010 | { |
3011 | BDRVRawState *s = bs->opaque; | |
7154d8ae KW |
3012 | RawPosixAIOData acb; |
3013 | ThreadPoolFunc *handler; | |
3014 | ||
292d06b9 HR |
3015 | #ifdef CONFIG_FALLOCATE |
3016 | if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { | |
3017 | BdrvTrackedRequest *req; | |
292d06b9 HR |
3018 | |
3019 | /* | |
3020 | * This is a workaround for a bug in the Linux XFS driver, | |
3021 | * where writes submitted through the AIO interface will be | |
3022 | * discarded if they happen beyond a concurrently running | |
3023 | * fallocate() that increases the file length (i.e., both the | |
3024 | * write and the fallocate() happen beyond the EOF). | |
3025 | * | |
3026 | * To work around it, we extend the tracked request for this | |
3027 | * zero write until INT64_MAX (effectively infinity), and mark | |
3028 | * it as serializing. | |
3029 | * | |
3030 | * We have to enable this workaround for all filesystems and | |
3031 | * AIO modes (not just XFS with aio=native), because for | |
3032 | * remote filesystems we do not know the host configuration. | |
3033 | */ | |
3034 | ||
3035 | req = bdrv_co_get_self_request(bs); | |
3036 | assert(req); | |
3037 | assert(req->type == BDRV_TRACKED_WRITE); | |
3038 | assert(req->offset <= offset); | |
3039 | assert(req->offset + req->bytes >= offset + bytes); | |
3040 | ||
8b117001 VSO |
3041 | req->bytes = BDRV_MAX_LENGTH - req->offset; |
3042 | ||
69b55e03 | 3043 | bdrv_check_request(req->offset, req->bytes, &error_abort); |
292d06b9 | 3044 | |
8ac5aab2 | 3045 | bdrv_make_request_serialising(req, bs->bl.request_alignment); |
292d06b9 HR |
3046 | } |
3047 | #endif | |
3048 | ||
7154d8ae KW |
3049 | acb = (RawPosixAIOData) { |
3050 | .bs = bs, | |
3051 | .aio_fildes = s->fd, | |
3052 | .aio_type = QEMU_AIO_WRITE_ZEROES, | |
3053 | .aio_offset = offset, | |
3054 | .aio_nbytes = bytes, | |
3055 | }; | |
3056 | ||
3057 | if (blkdev) { | |
3058 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
3059 | } | |
738301e1 KW |
3060 | if (flags & BDRV_REQ_NO_FALLBACK) { |
3061 | acb.aio_type |= QEMU_AIO_NO_FALLBACK; | |
3062 | } | |
260a82e5 | 3063 | |
34fa110e | 3064 | if (flags & BDRV_REQ_MAY_UNMAP) { |
7154d8ae KW |
3065 | acb.aio_type |= QEMU_AIO_DISCARD; |
3066 | handler = handle_aiocb_write_zeroes_unmap; | |
3067 | } else { | |
3068 | handler = handle_aiocb_write_zeroes; | |
260a82e5 | 3069 | } |
34fa110e | 3070 | |
7154d8ae KW |
3071 | return raw_thread_pool_submit(bs, handler, &acb); |
3072 | } | |
3073 | ||
3074 | static int coroutine_fn raw_co_pwrite_zeroes( | |
3075 | BlockDriverState *bs, int64_t offset, | |
f34b2bcf | 3076 | int64_t bytes, BdrvRequestFlags flags) |
7154d8ae KW |
3077 | { |
3078 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); | |
260a82e5 PB |
3079 | } |
3080 | ||
3081 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
3082 | { | |
260a82e5 PB |
3083 | return 0; |
3084 | } | |
3085 | ||
d9245599 AN |
3086 | static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) |
3087 | { | |
3088 | BDRVRawState *s = bs->opaque; | |
3089 | return (BlockStatsSpecificFile) { | |
3090 | .discard_nb_ok = s->stats.discard_nb_ok, | |
3091 | .discard_nb_failed = s->stats.discard_nb_failed, | |
3092 | .discard_bytes_ok = s->stats.discard_bytes_ok, | |
3093 | }; | |
3094 | } | |
3095 | ||
3096 | static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) | |
3097 | { | |
3098 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3099 | ||
3100 | stats->driver = BLOCKDEV_DRIVER_FILE; | |
3101 | stats->u.file = get_blockstats_specific_file(bs); | |
3102 | ||
3103 | return stats; | |
3104 | } | |
3105 | ||
14176c8d | 3106 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
d9245599 AN |
3107 | static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) |
3108 | { | |
3109 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3110 | ||
3111 | stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; | |
3112 | stats->u.host_device = get_blockstats_specific_file(bs); | |
3113 | ||
3114 | return stats; | |
3115 | } | |
14176c8d | 3116 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
d9245599 | 3117 | |
6f482f74 CL |
3118 | static QemuOptsList raw_create_opts = { |
3119 | .name = "raw-create-opts", | |
3120 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
3121 | .desc = { | |
3122 | { | |
3123 | .name = BLOCK_OPT_SIZE, | |
3124 | .type = QEMU_OPT_SIZE, | |
3125 | .help = "Virtual disk size" | |
3126 | }, | |
4ab15590 CL |
3127 | { |
3128 | .name = BLOCK_OPT_NOCOW, | |
3129 | .type = QEMU_OPT_BOOL, | |
3130 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
3131 | }, | |
06247428 HT |
3132 | { |
3133 | .name = BLOCK_OPT_PREALLOC, | |
3134 | .type = QEMU_OPT_STRING, | |
abea0053 SG |
3135 | .help = "Preallocation mode (allowed values: off" |
3136 | #ifdef CONFIG_POSIX_FALLOCATE | |
3137 | ", falloc" | |
3138 | #endif | |
3139 | ", full)" | |
06247428 | 3140 | }, |
ffa244c8 KW |
3141 | { |
3142 | .name = BLOCK_OPT_EXTENT_SIZE_HINT, | |
3143 | .type = QEMU_OPT_SIZE, | |
3144 | .help = "Extent size hint for the image file, 0 to disable" | |
3145 | }, | |
6f482f74 CL |
3146 | { /* end of list */ } |
3147 | } | |
0e7e1989 KW |
3148 | }; |
3149 | ||
244a5668 FZ |
3150 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
3151 | Error **errp) | |
3152 | { | |
6ceabe6f | 3153 | BDRVRawState *s = bs->opaque; |
72373e40 | 3154 | int input_flags = s->reopen_state ? s->reopen_state->flags : bs->open_flags; |
6ceabe6f KW |
3155 | int open_flags; |
3156 | int ret; | |
3157 | ||
72373e40 VSO |
3158 | /* We may need a new fd if auto-read-only switches the mode */ |
3159 | ret = raw_reconfigure_getfd(bs, input_flags, &open_flags, perm, | |
3160 | false, errp); | |
3161 | if (ret < 0) { | |
3162 | return ret; | |
3163 | } else if (ret != s->fd) { | |
3164 | Error *local_err = NULL; | |
3165 | ||
6ceabe6f | 3166 | /* |
72373e40 VSO |
3167 | * Fail already check_perm() if we can't get a working O_DIRECT |
3168 | * alignment with the new fd. | |
6ceabe6f | 3169 | */ |
72373e40 VSO |
3170 | raw_probe_alignment(bs, ret, &local_err); |
3171 | if (local_err) { | |
3172 | error_propagate(errp, local_err); | |
3173 | return -EINVAL; | |
6ceabe6f | 3174 | } |
72373e40 VSO |
3175 | |
3176 | s->perm_change_fd = ret; | |
3177 | s->perm_change_flags = open_flags; | |
6ceabe6f KW |
3178 | } |
3179 | ||
3180 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
3181 | * but keep everything locked that new will need. */ | |
3182 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
3183 | if (ret < 0) { | |
3184 | goto fail; | |
3185 | } | |
3186 | ||
3187 | /* Copy locks to the new fd */ | |
eb43ea16 | 3188 | if (s->perm_change_fd && s->use_lock) { |
6ceabe6f KW |
3189 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, |
3190 | false, errp); | |
3191 | if (ret < 0) { | |
3192 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
3193 | goto fail; | |
3194 | } | |
3195 | } | |
3196 | return 0; | |
3197 | ||
3198 | fail: | |
72373e40 | 3199 | if (s->perm_change_fd) { |
6ceabe6f KW |
3200 | qemu_close(s->perm_change_fd); |
3201 | } | |
3202 | s->perm_change_fd = 0; | |
3203 | return ret; | |
244a5668 FZ |
3204 | } |
3205 | ||
3206 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
3207 | { | |
3208 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
3209 | |
3210 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
3211 | * called after .bdrv_reopen_commit) */ | |
3212 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
3213 | qemu_close(s->fd); | |
3214 | s->fd = s->perm_change_fd; | |
094e3639 | 3215 | s->open_flags = s->perm_change_flags; |
6ceabe6f KW |
3216 | } |
3217 | s->perm_change_fd = 0; | |
3218 | ||
244a5668 FZ |
3219 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
3220 | s->perm = perm; | |
3221 | s->shared_perm = shared; | |
3222 | } | |
3223 | ||
3224 | static void raw_abort_perm_update(BlockDriverState *bs) | |
3225 | { | |
6ceabe6f KW |
3226 | BDRVRawState *s = bs->opaque; |
3227 | ||
3228 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
3229 | * the file descriptor. */ | |
72373e40 | 3230 | if (s->perm_change_fd) { |
6ceabe6f KW |
3231 | qemu_close(s->perm_change_fd); |
3232 | } | |
3233 | s->perm_change_fd = 0; | |
3234 | ||
244a5668 FZ |
3235 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
3236 | } | |
3237 | ||
67b51fb9 | 3238 | static int coroutine_fn raw_co_copy_range_from( |
48535049 VSO |
3239 | BlockDriverState *bs, BdrvChild *src, int64_t src_offset, |
3240 | BdrvChild *dst, int64_t dst_offset, int64_t bytes, | |
67b51fb9 | 3241 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) |
1efad060 | 3242 | { |
67b51fb9 VSO |
3243 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3244 | read_flags, write_flags); | |
1efad060 FZ |
3245 | } |
3246 | ||
3247 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 | 3248 | BdrvChild *src, |
48535049 | 3249 | int64_t src_offset, |
67b51fb9 | 3250 | BdrvChild *dst, |
48535049 VSO |
3251 | int64_t dst_offset, |
3252 | int64_t bytes, | |
67b51fb9 VSO |
3253 | BdrvRequestFlags read_flags, |
3254 | BdrvRequestFlags write_flags) | |
1efad060 | 3255 | { |
58a209c4 | 3256 | RawPosixAIOData acb; |
1efad060 FZ |
3257 | BDRVRawState *s = bs->opaque; |
3258 | BDRVRawState *src_s; | |
3259 | ||
3260 | assert(dst->bs == bs); | |
3261 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
3262 | return -ENOTSUP; | |
3263 | } | |
3264 | ||
3265 | src_s = src->bs->opaque; | |
9f850f67 | 3266 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
3267 | return -EIO; |
3268 | } | |
58a209c4 KW |
3269 | |
3270 | acb = (RawPosixAIOData) { | |
3271 | .bs = bs, | |
3272 | .aio_type = QEMU_AIO_COPY_RANGE, | |
3273 | .aio_fildes = src_s->fd, | |
3274 | .aio_offset = src_offset, | |
3275 | .aio_nbytes = bytes, | |
3276 | .copy_range = { | |
3277 | .aio_fd2 = s->fd, | |
3278 | .aio_offset2 = dst_offset, | |
3279 | }, | |
3280 | }; | |
3281 | ||
3282 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
3283 | } |
3284 | ||
5f535a94 | 3285 | BlockDriver bdrv_file = { |
84a12e66 CH |
3286 | .format_name = "file", |
3287 | .protocol_name = "file", | |
856ae5c3 | 3288 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3289 | .bdrv_needs_filename = true, |
856ae5c3 | 3290 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 3291 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 3292 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
3293 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3294 | .bdrv_reopen_commit = raw_reopen_commit, | |
3295 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 3296 | .bdrv_close = raw_close, |
927f11e1 | 3297 | .bdrv_co_create = raw_co_create, |
efc75e2a | 3298 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 3299 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 3300 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 3301 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3302 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
9bffae14 | 3303 | .bdrv_co_delete_file = raw_co_delete_file, |
3b46e624 | 3304 | |
9d52aa3c KW |
3305 | .bdrv_co_preadv = raw_co_preadv, |
3306 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3307 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3308 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
3309 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3310 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3311 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3312 | .bdrv_io_plug = raw_aio_plug, |
3313 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 3314 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3315 | |
061ca8a3 | 3316 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 3317 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3318 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3319 | .bdrv_get_allocated_file_size |
3320 | = raw_get_allocated_file_size, | |
d9245599 | 3321 | .bdrv_get_specific_stats = raw_get_specific_stats, |
244a5668 FZ |
3322 | .bdrv_check_perm = raw_check_perm, |
3323 | .bdrv_set_perm = raw_set_perm, | |
3324 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 3325 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3326 | .mutable_opts = mutable_opts, |
83f64091 FB |
3327 | }; |
3328 | ||
19cb3738 FB |
3329 | /***********************************************/ |
3330 | /* host device */ | |
3331 | ||
14176c8d JD |
3332 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
3333 | ||
83affaa6 | 3334 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
3335 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3336 | CFIndex maxPathSize, int flags); | |
aa44d3f6 PMD |
3337 | |
3338 | #if !defined(MAC_OS_VERSION_12_0) \ | |
3339 | || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0) | |
3340 | #define IOMainPort IOMasterPort | |
3341 | #endif | |
3342 | ||
d0855f12 | 3343 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 3344 | { |
d0855f12 | 3345 | kern_return_t kernResult = KERN_FAILURE; |
aa44d3f6 | 3346 | mach_port_t mainPort; |
19cb3738 | 3347 | CFMutableDictionaryRef classesToMatch; |
d0855f12 JA |
3348 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
3349 | char *mediaType = NULL; | |
19cb3738 | 3350 | |
aa44d3f6 | 3351 | kernResult = IOMainPort(MACH_PORT_NULL, &mainPort); |
19cb3738 | 3352 | if ( KERN_SUCCESS != kernResult ) { |
aa44d3f6 | 3353 | printf("IOMainPort returned %d\n", kernResult); |
19cb3738 | 3354 | } |
3b46e624 | 3355 | |
d0855f12 JA |
3356 | int index; |
3357 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
3358 | classesToMatch = IOServiceMatching(matching_array[index]); | |
3359 | if (classesToMatch == NULL) { | |
3360 | error_report("IOServiceMatching returned NULL for %s", | |
3361 | matching_array[index]); | |
3362 | continue; | |
3363 | } | |
3364 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
3365 | kCFBooleanTrue); | |
aa44d3f6 | 3366 | kernResult = IOServiceGetMatchingServices(mainPort, classesToMatch, |
d0855f12 JA |
3367 | mediaIterator); |
3368 | if (kernResult != KERN_SUCCESS) { | |
3369 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
3370 | kernResult); | |
3371 | continue; | |
3372 | } | |
3b46e624 | 3373 | |
d0855f12 JA |
3374 | /* If a match was found, leave the loop */ |
3375 | if (*mediaIterator != 0) { | |
4f7d28d7 | 3376 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
3377 | mediaType = g_strdup(matching_array[index]); |
3378 | break; | |
3379 | } | |
3380 | } | |
3381 | return mediaType; | |
19cb3738 FB |
3382 | } |
3383 | ||
98caa5bc JA |
3384 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3385 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
3386 | { |
3387 | io_object_t nextMedia; | |
3388 | kern_return_t kernResult = KERN_FAILURE; | |
3389 | *bsdPath = '\0'; | |
3390 | nextMedia = IOIteratorNext( mediaIterator ); | |
3391 | if ( nextMedia ) | |
3392 | { | |
3393 | CFTypeRef bsdPathAsCFString; | |
3394 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3395 | if ( bsdPathAsCFString ) { | |
3396 | size_t devPathLength; | |
3397 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3398 | if (flags & BDRV_O_NOCACHE) { |
3399 | strcat(bsdPath, "r"); | |
3400 | } | |
19cb3738 FB |
3401 | devPathLength = strlen( bsdPath ); |
3402 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3403 | kernResult = KERN_SUCCESS; | |
3404 | } | |
3405 | CFRelease( bsdPathAsCFString ); | |
3406 | } | |
3407 | IOObjectRelease( nextMedia ); | |
3408 | } | |
3b46e624 | 3409 | |
19cb3738 FB |
3410 | return kernResult; |
3411 | } | |
3412 | ||
d0855f12 JA |
3413 | /* Sets up a real cdrom for use in QEMU */ |
3414 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3415 | { | |
3416 | int index, num_of_test_partitions = 2, fd; | |
3417 | char test_partition[MAXPATHLEN]; | |
3418 | bool partition_found = false; | |
3419 | ||
3420 | /* look for a working partition */ | |
3421 | for (index = 0; index < num_of_test_partitions; index++) { | |
3422 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3423 | index); | |
b18a24a9 | 3424 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); |
d0855f12 JA |
3425 | if (fd >= 0) { |
3426 | partition_found = true; | |
3427 | qemu_close(fd); | |
3428 | break; | |
3429 | } | |
3430 | } | |
3431 | ||
3432 | /* if a working partition on the device was not found */ | |
3433 | if (partition_found == false) { | |
3434 | error_setg(errp, "Failed to find a working partition on disc"); | |
3435 | } else { | |
4f7d28d7 | 3436 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3437 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3438 | } | |
3439 | return partition_found; | |
3440 | } | |
3441 | ||
3442 | /* Prints directions on mounting and unmounting a device */ | |
3443 | static void print_unmounting_directions(const char *file_name) | |
3444 | { | |
3445 | error_report("If device %s is mounted on the desktop, unmount" | |
3446 | " it first before using it in QEMU", file_name); | |
3447 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3448 | file_name); | |
3449 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3450 | } | |
3451 | ||
3452 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3453 | |
508c7cb3 CH |
3454 | static int hdev_probe_device(const char *filename) |
3455 | { | |
3456 | struct stat st; | |
3457 | ||
3458 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3459 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3460 | return 50; | |
3461 | ||
3462 | if (stat(filename, &st) >= 0 && | |
3463 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3464 | return 100; | |
3465 | } | |
3466 | ||
3467 | return 0; | |
3468 | } | |
3469 | ||
7af803d4 HR |
3470 | static void hdev_parse_filename(const char *filename, QDict *options, |
3471 | Error **errp) | |
3472 | { | |
03c320d8 | 3473 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3474 | } |
3475 | ||
3307ed7b DA |
3476 | static bool hdev_is_sg(BlockDriverState *bs) |
3477 | { | |
3478 | ||
3479 | #if defined(__linux__) | |
3480 | ||
0d4377b3 | 3481 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3482 | struct stat st; |
3483 | struct sg_scsi_id scsiid; | |
3484 | int sg_version; | |
0d4377b3 KW |
3485 | int ret; |
3486 | ||
3487 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3488 | return false; | |
3489 | } | |
3307ed7b | 3490 | |
0d4377b3 KW |
3491 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3492 | if (ret < 0) { | |
3493 | return false; | |
3494 | } | |
3495 | ||
3496 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3497 | if (ret >= 0) { | |
4f7d28d7 | 3498 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3499 | return true; |
3500 | } | |
3501 | ||
3502 | #endif | |
3503 | ||
3504 | return false; | |
3505 | } | |
3506 | ||
015a1036 HR |
3507 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3508 | Error **errp) | |
19cb3738 FB |
3509 | { |
3510 | BDRVRawState *s = bs->opaque; | |
da888d37 | 3511 | int ret; |
a76bab49 | 3512 | |
83affaa6 | 3513 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3514 | /* |
3515 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3516 | * would require more care. When @options come from -blockdev or | |
3517 | * blockdev_add, its members are typed according to the QAPI | |
3518 | * schema, but when they come from -drive, they're all QString. | |
3519 | */ | |
3307ed7b | 3520 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3521 | char bsd_path[MAXPATHLEN] = ""; |
3522 | bool error_occurred = false; | |
3523 | ||
3524 | /* If using a real cdrom */ | |
3525 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3526 | char *mediaType = NULL; | |
3527 | kern_return_t ret_val; | |
3528 | io_iterator_t mediaIterator = 0; | |
3529 | ||
3530 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3531 | if (mediaType == NULL) { | |
3532 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3533 | " drive"); | |
3534 | error_occurred = true; | |
3535 | goto hdev_open_Mac_error; | |
3536 | } | |
3307ed7b | 3537 | |
d0855f12 JA |
3538 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3539 | if (ret_val != KERN_SUCCESS) { | |
3540 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3541 | error_occurred = true; | |
3542 | goto hdev_open_Mac_error; | |
3543 | } | |
3544 | ||
3545 | /* If a real optical drive was not found */ | |
3546 | if (bsd_path[0] == '\0') { | |
3547 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3548 | error_occurred = true; | |
3549 | goto hdev_open_Mac_error; | |
3550 | } | |
3551 | ||
3552 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3553 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3554 | setup_cdrom(bsd_path, errp) == false) { | |
3555 | print_unmounting_directions(bsd_path); | |
3556 | error_occurred = true; | |
3557 | goto hdev_open_Mac_error; | |
19cb3738 | 3558 | } |
3b46e624 | 3559 | |
46f5ac20 | 3560 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3561 | |
3562 | hdev_open_Mac_error: | |
3563 | g_free(mediaType); | |
3564 | if (mediaIterator) { | |
3565 | IOObjectRelease(mediaIterator); | |
3566 | } | |
3567 | if (error_occurred) { | |
3568 | return -ENOENT; | |
3569 | } | |
19cb3738 | 3570 | } |
d0855f12 | 3571 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3572 | |
3573 | s->type = FTYPE_FILE; | |
90babde0 | 3574 | |
668f62ec | 3575 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
da888d37 | 3576 | if (ret < 0) { |
d0855f12 JA |
3577 | #if defined(__APPLE__) && defined(__MACH__) |
3578 | if (*bsd_path) { | |
3579 | filename = bsd_path; | |
3580 | } | |
3581 | /* if a physical device experienced an error while being opened */ | |
3582 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3583 | print_unmounting_directions(filename); | |
3584 | } | |
3585 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3586 | return ret; |
3587 | } | |
3588 | ||
3307ed7b DA |
3589 | /* Since this does ioctl the device must be already opened */ |
3590 | bs->sg = hdev_is_sg(bs); | |
3591 | ||
da888d37 | 3592 | return ret; |
19cb3738 FB |
3593 | } |
3594 | ||
03ff3ca3 | 3595 | #if defined(__linux__) |
2f3a7ab3 KW |
3596 | static int coroutine_fn |
3597 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3598 | { |
f141eafe | 3599 | BDRVRawState *s = bs->opaque; |
03425671 | 3600 | RawPosixAIOData acb; |
2f3a7ab3 | 3601 | int ret; |
221f715d | 3602 | |
2f3a7ab3 KW |
3603 | ret = fd_open(bs); |
3604 | if (ret < 0) { | |
3605 | return ret; | |
3606 | } | |
c208e8c2 | 3607 | |
7c9e5276 PB |
3608 | if (req == SG_IO && s->pr_mgr) { |
3609 | struct sg_io_hdr *io_hdr = buf; | |
3610 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3611 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3612 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3613 | s->fd, io_hdr); |
7c9e5276 PB |
3614 | } |
3615 | } | |
3616 | ||
03425671 KW |
3617 | acb = (RawPosixAIOData) { |
3618 | .bs = bs, | |
3619 | .aio_type = QEMU_AIO_IOCTL, | |
3620 | .aio_fildes = s->fd, | |
3621 | .aio_offset = 0, | |
3622 | .ioctl = { | |
3623 | .buf = buf, | |
3624 | .cmd = req, | |
3625 | }, | |
3626 | }; | |
3627 | ||
3628 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3629 | } |
f709623b | 3630 | #endif /* linux */ |
221f715d | 3631 | |
33d70fb6 | 3632 | static coroutine_fn int |
0c802287 | 3633 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
c36dd8a0 | 3634 | { |
1c450366 | 3635 | BDRVRawState *s = bs->opaque; |
33d70fb6 | 3636 | int ret; |
c36dd8a0 | 3637 | |
33d70fb6 KW |
3638 | ret = fd_open(bs); |
3639 | if (ret < 0) { | |
1c450366 | 3640 | raw_account_discard(s, bytes, ret); |
33d70fb6 | 3641 | return ret; |
c36dd8a0 | 3642 | } |
46ee0f46 | 3643 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3644 | } |
3645 | ||
2ffa76c2 | 3646 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f34b2bcf | 3647 | int64_t offset, int64_t bytes, BdrvRequestFlags flags) |
d0b4503e | 3648 | { |
d0b4503e PB |
3649 | int rc; |
3650 | ||
3651 | rc = fd_open(bs); | |
3652 | if (rc < 0) { | |
3653 | return rc; | |
3654 | } | |
34fa110e | 3655 | |
7154d8ae | 3656 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3657 | } |
3658 | ||
5efa9d5a | 3659 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3660 | .format_name = "host_device", |
84a12e66 | 3661 | .protocol_name = "host_device", |
0b4ce02e | 3662 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3663 | .bdrv_needs_filename = true, |
0b4ce02e | 3664 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3665 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3666 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3667 | .bdrv_close = raw_close, |
1bc6b705 JC |
3668 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3669 | .bdrv_reopen_commit = raw_reopen_commit, | |
3670 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3671 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3672 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3673 | .mutable_opts = mutable_opts, |
dd577a26 | 3674 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3675 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3676 | |
9d52aa3c KW |
3677 | .bdrv_co_preadv = raw_co_preadv, |
3678 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3679 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3680 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3681 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3682 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3683 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3684 | .bdrv_io_plug = raw_aio_plug, |
3685 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3686 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3687 | |
061ca8a3 | 3688 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3689 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3690 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3691 | .bdrv_get_allocated_file_size |
3692 | = raw_get_allocated_file_size, | |
d9245599 | 3693 | .bdrv_get_specific_stats = hdev_get_specific_stats, |
244a5668 FZ |
3694 | .bdrv_check_perm = raw_check_perm, |
3695 | .bdrv_set_perm = raw_set_perm, | |
3696 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3697 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3698 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3699 | |
f3a5d3f8 | 3700 | /* generic scsi device */ |
63ec93db | 3701 | #ifdef __linux__ |
2f3a7ab3 | 3702 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3703 | #endif |
f3a5d3f8 CH |
3704 | }; |
3705 | ||
18fa1c42 HR |
3706 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3707 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3708 | Error **errp) | |
3709 | { | |
03c320d8 | 3710 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3711 | } |
3712 | #endif | |
3713 | ||
3714 | #ifdef __linux__ | |
015a1036 HR |
3715 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3716 | Error **errp) | |
f3a5d3f8 CH |
3717 | { |
3718 | BDRVRawState *s = bs->opaque; | |
3719 | ||
f3a5d3f8 CH |
3720 | s->type = FTYPE_CD; |
3721 | ||
19a3da7f | 3722 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3723 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3724 | } |
3725 | ||
508c7cb3 CH |
3726 | static int cdrom_probe_device(const char *filename) |
3727 | { | |
3baf720e CR |
3728 | int fd, ret; |
3729 | int prio = 0; | |
343f8568 | 3730 | struct stat st; |
3baf720e | 3731 | |
b18a24a9 | 3732 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); |
3baf720e CR |
3733 | if (fd < 0) { |
3734 | goto out; | |
3735 | } | |
343f8568 JS |
3736 | ret = fstat(fd, &st); |
3737 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3738 | goto outc; | |
3739 | } | |
3baf720e CR |
3740 | |
3741 | /* Attempt to detect via a CDROM specific ioctl */ | |
3742 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3743 | if (ret >= 0) | |
3744 | prio = 100; | |
3745 | ||
343f8568 | 3746 | outc: |
2e1e79da | 3747 | qemu_close(fd); |
3baf720e CR |
3748 | out: |
3749 | return prio; | |
508c7cb3 CH |
3750 | } |
3751 | ||
e031f750 | 3752 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3753 | { |
3754 | BDRVRawState *s = bs->opaque; | |
3755 | int ret; | |
3756 | ||
3757 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3758 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3759 | } |
3760 | ||
f36f3949 | 3761 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3762 | { |
3763 | BDRVRawState *s = bs->opaque; | |
3764 | ||
3765 | if (eject_flag) { | |
3766 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3767 | perror("CDROMEJECT"); | |
3768 | } else { | |
3769 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3770 | perror("CDROMEJECT"); | |
3771 | } | |
f3a5d3f8 CH |
3772 | } |
3773 | ||
025e849a | 3774 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3775 | { |
3776 | BDRVRawState *s = bs->opaque; | |
3777 | ||
3778 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3779 | /* | |
3780 | * Note: an error can happen if the distribution automatically | |
3781 | * mounts the CD-ROM | |
3782 | */ | |
3783 | /* perror("CDROM_LOCKDOOR"); */ | |
3784 | } | |
f3a5d3f8 CH |
3785 | } |
3786 | ||
3787 | static BlockDriver bdrv_host_cdrom = { | |
3788 | .format_name = "host_cdrom", | |
84a12e66 | 3789 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3790 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3791 | .bdrv_needs_filename = true, |
508c7cb3 | 3792 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3793 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3794 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3795 | .bdrv_close = raw_close, |
1bc6b705 JC |
3796 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3797 | .bdrv_reopen_commit = raw_reopen_commit, | |
3798 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3799 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3800 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3801 | .mutable_opts = mutable_opts, |
dd577a26 | 3802 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3803 | |
9d52aa3c KW |
3804 | .bdrv_co_preadv = raw_co_preadv, |
3805 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3806 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3807 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3808 | .bdrv_io_plug = raw_aio_plug, |
3809 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3810 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3811 | |
061ca8a3 | 3812 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3813 | .bdrv_getlength = raw_getlength, |
3814 | .has_variable_length = true, | |
4a1d5e1f FZ |
3815 | .bdrv_get_allocated_file_size |
3816 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3817 | |
3818 | /* removable device support */ | |
3819 | .bdrv_is_inserted = cdrom_is_inserted, | |
3820 | .bdrv_eject = cdrom_eject, | |
025e849a | 3821 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3822 | |
3823 | /* generic scsi device */ | |
2f3a7ab3 | 3824 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3825 | }; |
3826 | #endif /* __linux__ */ | |
3827 | ||
a167ba50 | 3828 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3829 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3830 | Error **errp) | |
f3a5d3f8 CH |
3831 | { |
3832 | BDRVRawState *s = bs->opaque; | |
3833 | int ret; | |
3834 | ||
3835 | s->type = FTYPE_CD; | |
3836 | ||
668f62ec | 3837 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
e428e439 | 3838 | if (ret) { |
f3a5d3f8 | 3839 | return ret; |
e428e439 | 3840 | } |
f3a5d3f8 | 3841 | |
9b2260cb | 3842 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3843 | ioctl(s->fd, CDIOCALLOW); |
3844 | return 0; | |
3845 | } | |
3846 | ||
508c7cb3 CH |
3847 | static int cdrom_probe_device(const char *filename) |
3848 | { | |
3849 | if (strstart(filename, "/dev/cd", NULL) || | |
3850 | strstart(filename, "/dev/acd", NULL)) | |
3851 | return 100; | |
3852 | return 0; | |
3853 | } | |
3854 | ||
f3a5d3f8 CH |
3855 | static int cdrom_reopen(BlockDriverState *bs) |
3856 | { | |
3857 | BDRVRawState *s = bs->opaque; | |
3858 | int fd; | |
3859 | ||
3860 | /* | |
3861 | * Force reread of possibly changed/newly loaded disc, | |
3862 | * FreeBSD seems to not notice sometimes... | |
3863 | */ | |
3864 | if (s->fd >= 0) | |
2e1e79da | 3865 | qemu_close(s->fd); |
b18a24a9 | 3866 | fd = qemu_open(bs->filename, s->open_flags, NULL); |
f3a5d3f8 CH |
3867 | if (fd < 0) { |
3868 | s->fd = -1; | |
3869 | return -EIO; | |
3870 | } | |
3871 | s->fd = fd; | |
3872 | ||
9b2260cb | 3873 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3874 | ioctl(s->fd, CDIOCALLOW); |
3875 | return 0; | |
3876 | } | |
3877 | ||
e031f750 | 3878 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3879 | { |
3880 | return raw_getlength(bs) > 0; | |
3881 | } | |
3882 | ||
f36f3949 | 3883 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3884 | { |
3885 | BDRVRawState *s = bs->opaque; | |
3886 | ||
3887 | if (s->fd < 0) | |
822e1cd1 | 3888 | return; |
f3a5d3f8 CH |
3889 | |
3890 | (void) ioctl(s->fd, CDIOCALLOW); | |
3891 | ||
3892 | if (eject_flag) { | |
3893 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3894 | perror("CDIOCEJECT"); | |
3895 | } else { | |
3896 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3897 | perror("CDIOCCLOSE"); | |
3898 | } | |
3899 | ||
822e1cd1 | 3900 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3901 | } |
3902 | ||
025e849a | 3903 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3904 | { |
3905 | BDRVRawState *s = bs->opaque; | |
3906 | ||
3907 | if (s->fd < 0) | |
7bf37fed | 3908 | return; |
f3a5d3f8 CH |
3909 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3910 | /* | |
3911 | * Note: an error can happen if the distribution automatically | |
3912 | * mounts the CD-ROM | |
3913 | */ | |
3914 | /* perror("CDROM_LOCKDOOR"); */ | |
3915 | } | |
f3a5d3f8 CH |
3916 | } |
3917 | ||
3918 | static BlockDriver bdrv_host_cdrom = { | |
3919 | .format_name = "host_cdrom", | |
84a12e66 | 3920 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3921 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3922 | .bdrv_needs_filename = true, |
508c7cb3 | 3923 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3924 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3925 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3926 | .bdrv_close = raw_close, |
1bc6b705 JC |
3927 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3928 | .bdrv_reopen_commit = raw_reopen_commit, | |
3929 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3930 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3931 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3932 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3933 | |
9d52aa3c KW |
3934 | .bdrv_co_preadv = raw_co_preadv, |
3935 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3936 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3937 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3938 | .bdrv_io_plug = raw_aio_plug, |
3939 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3940 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3941 | |
061ca8a3 | 3942 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3943 | .bdrv_getlength = raw_getlength, |
3944 | .has_variable_length = true, | |
4a1d5e1f FZ |
3945 | .bdrv_get_allocated_file_size |
3946 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3947 | |
19cb3738 | 3948 | /* removable device support */ |
f3a5d3f8 CH |
3949 | .bdrv_is_inserted = cdrom_is_inserted, |
3950 | .bdrv_eject = cdrom_eject, | |
025e849a | 3951 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3952 | }; |
f3a5d3f8 | 3953 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3954 | |
14176c8d JD |
3955 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
3956 | ||
84a12e66 | 3957 | static void bdrv_file_init(void) |
5efa9d5a | 3958 | { |
508c7cb3 CH |
3959 | /* |
3960 | * Register all the drivers. Note that order is important, the driver | |
3961 | * registered last will get probed first. | |
3962 | */ | |
84a12e66 | 3963 | bdrv_register(&bdrv_file); |
14176c8d | 3964 | #if defined(HAVE_HOST_BLOCK_DEVICE) |
5efa9d5a | 3965 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3966 | #ifdef __linux__ |
f3a5d3f8 CH |
3967 | bdrv_register(&bdrv_host_cdrom); |
3968 | #endif | |
a167ba50 | 3969 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3970 | bdrv_register(&bdrv_host_cdrom); |
3971 | #endif | |
14176c8d | 3972 | #endif /* HAVE_HOST_BLOCK_DEVICE */ |
5efa9d5a AL |
3973 | } |
3974 | ||
84a12e66 | 3975 | block_init(bdrv_file_init); |