]>
Commit | Line | Data |
---|---|---|
83f64091 | 1 | /* |
223d4670 | 2 | * Block driver for RAW files (posix) |
5fafdf24 | 3 | * |
83f64091 | 4 | * Copyright (c) 2006 Fabrice Bellard |
5fafdf24 | 5 | * |
83f64091 FB |
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
7 | * of this software and associated documentation files (the "Software"), to deal | |
8 | * in the Software without restriction, including without limitation the rights | |
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
10 | * copies of the Software, and to permit persons to whom the Software is | |
11 | * furnished to do so, subject to the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice shall be included in | |
14 | * all copies or substantial portions of the Software. | |
15 | * | |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
22 | * THE SOFTWARE. | |
23 | */ | |
922a01a0 | 24 | |
80c71a24 | 25 | #include "qemu/osdep.h" |
a8d25326 | 26 | #include "qemu-common.h" |
da34e65c | 27 | #include "qapi/error.h" |
f348b6d1 | 28 | #include "qemu/cutils.h" |
d49b6836 | 29 | #include "qemu/error-report.h" |
737e150e | 30 | #include "block/block_int.h" |
1de7afc9 | 31 | #include "qemu/module.h" |
922a01a0 | 32 | #include "qemu/option.h" |
ffa244c8 | 33 | #include "qemu/units.h" |
de81a169 | 34 | #include "trace.h" |
737e150e | 35 | #include "block/thread-pool.h" |
1de7afc9 | 36 | #include "qemu/iov.h" |
0187f5c9 | 37 | #include "block/raw-aio.h" |
452fcdbc | 38 | #include "qapi/qmp/qdict.h" |
d49b6836 | 39 | #include "qapi/qmp/qstring.h" |
83f64091 | 40 | |
7c9e5276 PB |
41 | #include "scsi/pr-manager.h" |
42 | #include "scsi/constants.h" | |
43 | ||
83affaa6 | 44 | #if defined(__APPLE__) && (__MACH__) |
83f64091 FB |
45 | #include <paths.h> |
46 | #include <sys/param.h> | |
47 | #include <IOKit/IOKitLib.h> | |
48 | #include <IOKit/IOBSD.h> | |
49 | #include <IOKit/storage/IOMediaBSDClient.h> | |
50 | #include <IOKit/storage/IOMedia.h> | |
51 | #include <IOKit/storage/IOCDMedia.h> | |
52 | //#include <IOKit/storage/IOCDTypes.h> | |
d0855f12 | 53 | #include <IOKit/storage/IODVDMedia.h> |
83f64091 FB |
54 | #include <CoreFoundation/CoreFoundation.h> |
55 | #endif | |
56 | ||
57 | #ifdef __sun__ | |
2e9671da | 58 | #define _POSIX_PTHREAD_SEMANTICS 1 |
83f64091 FB |
59 | #include <sys/dkio.h> |
60 | #endif | |
19cb3738 FB |
61 | #ifdef __linux__ |
62 | #include <sys/ioctl.h> | |
05acda4d | 63 | #include <sys/param.h> |
1efad060 | 64 | #include <sys/syscall.h> |
5edc8557 | 65 | #include <sys/vfs.h> |
19cb3738 FB |
66 | #include <linux/cdrom.h> |
67 | #include <linux/fd.h> | |
5500316d | 68 | #include <linux/fs.h> |
1a9335e4 | 69 | #include <linux/hdreg.h> |
5edc8557 | 70 | #include <linux/magic.h> |
3307ed7b | 71 | #include <scsi/sg.h> |
1a9335e4 ET |
72 | #ifdef __s390__ |
73 | #include <asm/dasd.h> | |
74 | #endif | |
4ab15590 CL |
75 | #ifndef FS_NOCOW_FL |
76 | #define FS_NOCOW_FL 0x00800000 /* Do not cow file */ | |
77 | #endif | |
5500316d | 78 | #endif |
b953f075 | 79 | #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE) |
3d4fa43e KK |
80 | #include <linux/falloc.h> |
81 | #endif | |
a167ba50 | 82 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
1cb6c3fd | 83 | #include <sys/disk.h> |
9f23011a | 84 | #include <sys/cdio.h> |
1cb6c3fd | 85 | #endif |
83f64091 | 86 | |
128ab2ff BS |
87 | #ifdef __OpenBSD__ |
88 | #include <sys/ioctl.h> | |
89 | #include <sys/disklabel.h> | |
90 | #include <sys/dkio.h> | |
91 | #endif | |
92 | ||
d1f6fd8d CE |
93 | #ifdef __NetBSD__ |
94 | #include <sys/ioctl.h> | |
95 | #include <sys/disklabel.h> | |
96 | #include <sys/dkio.h> | |
97 | #include <sys/disk.h> | |
98 | #endif | |
99 | ||
c5e97233 BS |
100 | #ifdef __DragonFly__ |
101 | #include <sys/ioctl.h> | |
102 | #include <sys/diskslice.h> | |
103 | #endif | |
104 | ||
dce512de CH |
105 | #ifdef CONFIG_XFS |
106 | #include <xfs/xfs.h> | |
107 | #endif | |
108 | ||
4f7d28d7 | 109 | #include "trace.h" |
8c05dbf9 | 110 | |
f6465578 AL |
111 | /* OS X does not have O_DSYNC */ |
112 | #ifndef O_DSYNC | |
1c27a8b3 | 113 | #ifdef O_SYNC |
7ab064d2 | 114 | #define O_DSYNC O_SYNC |
1c27a8b3 JA |
115 | #elif defined(O_FSYNC) |
116 | #define O_DSYNC O_FSYNC | |
117 | #endif | |
f6465578 AL |
118 | #endif |
119 | ||
9f7965c7 AL |
120 | /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */ |
121 | #ifndef O_DIRECT | |
122 | #define O_DIRECT O_DSYNC | |
123 | #endif | |
124 | ||
19cb3738 FB |
125 | #define FTYPE_FILE 0 |
126 | #define FTYPE_CD 1 | |
83f64091 | 127 | |
581b9e29 CH |
128 | #define MAX_BLOCKSIZE 4096 |
129 | ||
244a5668 FZ |
130 | /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes, |
131 | * leaving a few more bytes for its future use. */ | |
132 | #define RAW_LOCK_PERM_BASE 100 | |
133 | #define RAW_LOCK_SHARED_BASE 200 | |
134 | ||
19cb3738 FB |
135 | typedef struct BDRVRawState { |
136 | int fd; | |
244a5668 | 137 | bool use_lock; |
19cb3738 | 138 | int type; |
0e1d8f4c | 139 | int open_flags; |
c25f53b0 PB |
140 | size_t buf_align; |
141 | ||
244a5668 FZ |
142 | /* The current permissions. */ |
143 | uint64_t perm; | |
144 | uint64_t shared_perm; | |
145 | ||
2996ffad | 146 | /* The perms bits whose corresponding bytes are already locked in |
f2e3af29 | 147 | * s->fd. */ |
2996ffad FZ |
148 | uint64_t locked_perm; |
149 | uint64_t locked_shared_perm; | |
150 | ||
6ceabe6f | 151 | int perm_change_fd; |
094e3639 | 152 | int perm_change_flags; |
e0c9cf3a KW |
153 | BDRVReopenState *reopen_state; |
154 | ||
dce512de | 155 | #ifdef CONFIG_XFS |
260a82e5 | 156 | bool is_xfs:1; |
dce512de | 157 | #endif |
260a82e5 | 158 | bool has_discard:1; |
97a2ae34 | 159 | bool has_write_zeroes:1; |
260a82e5 | 160 | bool discard_zeroes:1; |
0a4279d9 | 161 | bool use_linux_aio:1; |
c6447510 | 162 | bool use_linux_io_uring:1; |
e5bcf967 | 163 | bool page_cache_inconsistent:1; |
d50d8222 | 164 | bool has_fallocate; |
3cad8307 | 165 | bool needs_alignment; |
f357fcd8 | 166 | bool drop_cache; |
31be8a2a | 167 | bool check_cache_dropped; |
1c450366 AN |
168 | struct { |
169 | uint64_t discard_nb_ok; | |
170 | uint64_t discard_nb_failed; | |
171 | uint64_t discard_bytes_ok; | |
172 | } stats; | |
7c9e5276 PB |
173 | |
174 | PRManager *pr_mgr; | |
19cb3738 FB |
175 | } BDRVRawState; |
176 | ||
eeb6b45d JC |
177 | typedef struct BDRVRawReopenState { |
178 | int fd; | |
179 | int open_flags; | |
f357fcd8 | 180 | bool drop_cache; |
31be8a2a | 181 | bool check_cache_dropped; |
eeb6b45d JC |
182 | } BDRVRawReopenState; |
183 | ||
19cb3738 | 184 | static int fd_open(BlockDriverState *bs); |
22afa7b5 | 185 | static int64_t raw_getlength(BlockDriverState *bs); |
83f64091 | 186 | |
de81a169 PB |
187 | typedef struct RawPosixAIOData { |
188 | BlockDriverState *bs; | |
d57c44d0 | 189 | int aio_type; |
de81a169 | 190 | int aio_fildes; |
d57c44d0 | 191 | |
de81a169 | 192 | off_t aio_offset; |
d57c44d0 KW |
193 | uint64_t aio_nbytes; |
194 | ||
93f4e2ff | 195 | union { |
d57c44d0 KW |
196 | struct { |
197 | struct iovec *iov; | |
198 | int niov; | |
199 | } io; | |
200 | struct { | |
201 | uint64_t cmd; | |
202 | void *buf; | |
203 | } ioctl; | |
93f4e2ff KW |
204 | struct { |
205 | int aio_fd2; | |
206 | off_t aio_offset2; | |
d57c44d0 | 207 | } copy_range; |
93f4e2ff KW |
208 | struct { |
209 | PreallocMode prealloc; | |
210 | Error **errp; | |
d57c44d0 | 211 | } truncate; |
93f4e2ff | 212 | }; |
de81a169 PB |
213 | } RawPosixAIOData; |
214 | ||
a167ba50 | 215 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 | 216 | static int cdrom_reopen(BlockDriverState *bs); |
9f23011a BS |
217 | #endif |
218 | ||
1de1ae0a | 219 | #if defined(__NetBSD__) |
db0754df | 220 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
221 | { |
222 | static char namebuf[PATH_MAX]; | |
223 | const char *dp, *fname; | |
224 | struct stat sb; | |
225 | ||
226 | fname = *filename; | |
227 | dp = strrchr(fname, '/'); | |
228 | if (lstat(fname, &sb) < 0) { | |
f6fc1e30 | 229 | error_setg_file_open(errp, errno, fname); |
1de1ae0a CE |
230 | return -errno; |
231 | } | |
232 | ||
233 | if (!S_ISBLK(sb.st_mode)) { | |
234 | return 0; | |
235 | } | |
236 | ||
237 | if (dp == NULL) { | |
238 | snprintf(namebuf, PATH_MAX, "r%s", fname); | |
239 | } else { | |
240 | snprintf(namebuf, PATH_MAX, "%.*s/r%s", | |
241 | (int)(dp - fname), fname, dp + 1); | |
242 | } | |
1de1ae0a | 243 | *filename = namebuf; |
db0754df | 244 | warn_report("%s is a block device, using %s", fname, *filename); |
1de1ae0a CE |
245 | |
246 | return 0; | |
247 | } | |
248 | #else | |
db0754df | 249 | static int raw_normalize_devicepath(const char **filename, Error **errp) |
1de1ae0a CE |
250 | { |
251 | return 0; | |
252 | } | |
253 | #endif | |
254 | ||
8a4ed0d1 ET |
255 | /* |
256 | * Get logical block size via ioctl. On success store it in @sector_size_p. | |
257 | */ | |
258 | static int probe_logical_blocksize(int fd, unsigned int *sector_size_p) | |
c25f53b0 | 259 | { |
c25f53b0 | 260 | unsigned int sector_size; |
8a4ed0d1 | 261 | bool success = false; |
700f9ce0 | 262 | int i; |
c25f53b0 | 263 | |
8a4ed0d1 | 264 | errno = ENOTSUP; |
700f9ce0 | 265 | static const unsigned long ioctl_list[] = { |
c25f53b0 | 266 | #ifdef BLKSSZGET |
700f9ce0 | 267 | BLKSSZGET, |
c25f53b0 PB |
268 | #endif |
269 | #ifdef DKIOCGETBLOCKSIZE | |
700f9ce0 | 270 | DKIOCGETBLOCKSIZE, |
c25f53b0 PB |
271 | #endif |
272 | #ifdef DIOCGSECTORSIZE | |
700f9ce0 | 273 | DIOCGSECTORSIZE, |
c25f53b0 | 274 | #endif |
700f9ce0 PM |
275 | }; |
276 | ||
277 | /* Try a few ioctls to get the right size */ | |
278 | for (i = 0; i < (int)ARRAY_SIZE(ioctl_list); i++) { | |
279 | if (ioctl(fd, ioctl_list[i], §or_size) >= 0) { | |
280 | *sector_size_p = sector_size; | |
281 | success = true; | |
282 | } | |
283 | } | |
8a4ed0d1 ET |
284 | |
285 | return success ? 0 : -errno; | |
286 | } | |
287 | ||
1a9335e4 ET |
288 | /** |
289 | * Get physical block size of @fd. | |
290 | * On success, store it in @blk_size and return 0. | |
291 | * On failure, return -errno. | |
292 | */ | |
293 | static int probe_physical_blocksize(int fd, unsigned int *blk_size) | |
294 | { | |
295 | #ifdef BLKPBSZGET | |
296 | if (ioctl(fd, BLKPBSZGET, blk_size) < 0) { | |
297 | return -errno; | |
298 | } | |
299 | return 0; | |
300 | #else | |
301 | return -ENOTSUP; | |
302 | #endif | |
303 | } | |
304 | ||
5edc8557 KW |
305 | /* |
306 | * Returns true if no alignment restrictions are necessary even for files | |
307 | * opened with O_DIRECT. | |
308 | * | |
309 | * raw_probe_alignment() probes the required alignment and assume that 1 means | |
310 | * the probing failed, so it falls back to a safe default of 4k. This can be | |
311 | * avoided if we know that byte alignment is okay for the file. | |
312 | */ | |
313 | static bool dio_byte_aligned(int fd) | |
314 | { | |
315 | #ifdef __linux__ | |
316 | struct statfs buf; | |
317 | int ret; | |
318 | ||
319 | ret = fstatfs(fd, &buf); | |
320 | if (ret == 0 && buf.f_type == NFS_SUPER_MAGIC) { | |
321 | return true; | |
322 | } | |
323 | #endif | |
324 | return false; | |
325 | } | |
326 | ||
22d182e8 SH |
327 | /* Check if read is allowed with given memory buffer and length. |
328 | * | |
329 | * This function is used to check O_DIRECT memory buffer and request alignment. | |
330 | */ | |
331 | static bool raw_is_io_aligned(int fd, void *buf, size_t len) | |
332 | { | |
333 | ssize_t ret = pread(fd, buf, len, 0); | |
334 | ||
335 | if (ret >= 0) { | |
336 | return true; | |
337 | } | |
338 | ||
339 | #ifdef __linux__ | |
340 | /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore | |
341 | * other errors (e.g. real I/O error), which could happen on a failed | |
342 | * drive, since we only care about probing alignment. | |
343 | */ | |
344 | if (errno != EINVAL) { | |
345 | return true; | |
346 | } | |
347 | #endif | |
348 | ||
349 | return false; | |
350 | } | |
351 | ||
8a4ed0d1 ET |
352 | static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp) |
353 | { | |
354 | BDRVRawState *s = bs->opaque; | |
355 | char *buf; | |
038adc2f | 356 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); |
a6b257a0 | 357 | size_t alignments[] = {1, 512, 1024, 2048, 4096}; |
8a4ed0d1 | 358 | |
b192af8a | 359 | /* For SCSI generic devices the alignment is not really used. |
8a4ed0d1 | 360 | With buffered I/O, we don't have any restrictions. */ |
b192af8a | 361 | if (bdrv_is_sg(bs) || !s->needs_alignment) { |
a5b8dd2c | 362 | bs->bl.request_alignment = 1; |
8a4ed0d1 ET |
363 | s->buf_align = 1; |
364 | return; | |
365 | } | |
366 | ||
a5b8dd2c | 367 | bs->bl.request_alignment = 0; |
8a4ed0d1 ET |
368 | s->buf_align = 0; |
369 | /* Let's try to use the logical blocksize for the alignment. */ | |
a5b8dd2c EB |
370 | if (probe_logical_blocksize(fd, &bs->bl.request_alignment) < 0) { |
371 | bs->bl.request_alignment = 0; | |
8a4ed0d1 | 372 | } |
c25f53b0 PB |
373 | #ifdef CONFIG_XFS |
374 | if (s->is_xfs) { | |
375 | struct dioattr da; | |
df26a350 | 376 | if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) { |
a5b8dd2c | 377 | bs->bl.request_alignment = da.d_miniosz; |
c25f53b0 PB |
378 | /* The kernel returns wrong information for d_mem */ |
379 | /* s->buf_align = da.d_mem; */ | |
380 | } | |
381 | } | |
382 | #endif | |
383 | ||
a6b257a0 NS |
384 | /* |
385 | * If we could not get the sizes so far, we can only guess them. First try | |
386 | * to detect request alignment, since it is more likely to succeed. Then | |
387 | * try to detect buf_align, which cannot be detected in some cases (e.g. | |
388 | * Gluster). If buf_align cannot be detected, we fallback to the value of | |
389 | * request_alignment. | |
390 | */ | |
391 | ||
392 | if (!bs->bl.request_alignment) { | |
393 | int i; | |
c25f53b0 | 394 | size_t align; |
a6b257a0 NS |
395 | buf = qemu_memalign(max_align, max_align); |
396 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
397 | align = alignments[i]; | |
398 | if (raw_is_io_aligned(fd, buf, align)) { | |
399 | /* Fallback to safe value. */ | |
400 | bs->bl.request_alignment = (align != 1) ? align : max_align; | |
c25f53b0 PB |
401 | break; |
402 | } | |
403 | } | |
404 | qemu_vfree(buf); | |
405 | } | |
406 | ||
a6b257a0 NS |
407 | if (!s->buf_align) { |
408 | int i; | |
c25f53b0 | 409 | size_t align; |
a6b257a0 NS |
410 | buf = qemu_memalign(max_align, 2 * max_align); |
411 | for (i = 0; i < ARRAY_SIZE(alignments); i++) { | |
412 | align = alignments[i]; | |
413 | if (raw_is_io_aligned(fd, buf + align, max_align)) { | |
236094c7 | 414 | /* Fallback to request_alignment. */ |
a6b257a0 | 415 | s->buf_align = (align != 1) ? align : bs->bl.request_alignment; |
c25f53b0 PB |
416 | break; |
417 | } | |
418 | } | |
419 | qemu_vfree(buf); | |
420 | } | |
df26a350 | 421 | |
a5b8dd2c | 422 | if (!s->buf_align || !bs->bl.request_alignment) { |
8cc9c6e9 EB |
423 | error_setg(errp, "Could not find working O_DIRECT alignment"); |
424 | error_append_hint(errp, "Try cache.direct=off\n"); | |
df26a350 | 425 | } |
c25f53b0 PB |
426 | } |
427 | ||
bca5283b | 428 | static int check_hdev_writable(int fd) |
20eaf1bf KW |
429 | { |
430 | #if defined(BLKROGET) | |
431 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
432 | * This is independent of device node permissions and therefore open(2) | |
433 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
434 | * | |
435 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
436 | * check for read-only block devices so that Linux block devices behave | |
437 | * properly. | |
438 | */ | |
439 | struct stat st; | |
440 | int readonly = 0; | |
441 | ||
bca5283b | 442 | if (fstat(fd, &st)) { |
20eaf1bf KW |
443 | return -errno; |
444 | } | |
445 | ||
446 | if (!S_ISBLK(st.st_mode)) { | |
447 | return 0; | |
448 | } | |
449 | ||
bca5283b | 450 | if (ioctl(fd, BLKROGET, &readonly) < 0) { |
20eaf1bf KW |
451 | return -errno; |
452 | } | |
453 | ||
454 | if (readonly) { | |
455 | return -EACCES; | |
456 | } | |
457 | #endif /* defined(BLKROGET) */ | |
458 | return 0; | |
459 | } | |
460 | ||
23dece19 | 461 | static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) |
6a8dc042 | 462 | { |
23dece19 | 463 | bool read_write = false; |
6a8dc042 JC |
464 | assert(open_flags != NULL); |
465 | ||
466 | *open_flags |= O_BINARY; | |
467 | *open_flags &= ~O_ACCMODE; | |
23dece19 KW |
468 | |
469 | if (bdrv_flags & BDRV_O_AUTO_RDONLY) { | |
470 | read_write = has_writers; | |
471 | } else if (bdrv_flags & BDRV_O_RDWR) { | |
472 | read_write = true; | |
473 | } | |
474 | ||
475 | if (read_write) { | |
6a8dc042 JC |
476 | *open_flags |= O_RDWR; |
477 | } else { | |
478 | *open_flags |= O_RDONLY; | |
479 | } | |
480 | ||
481 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
482 | * and O_DIRECT for no caching. */ | |
483 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
484 | *open_flags |= O_DIRECT; | |
485 | } | |
6a8dc042 JC |
486 | } |
487 | ||
078896a9 HR |
488 | static void raw_parse_filename(const char *filename, QDict *options, |
489 | Error **errp) | |
490 | { | |
03c320d8 | 491 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
492 | } |
493 | ||
c66a6157 KW |
494 | static QemuOptsList raw_runtime_opts = { |
495 | .name = "raw", | |
496 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
497 | .desc = { | |
498 | { | |
499 | .name = "filename", | |
500 | .type = QEMU_OPT_STRING, | |
501 | .help = "File name of the image", | |
502 | }, | |
0a4279d9 KW |
503 | { |
504 | .name = "aio", | |
505 | .type = QEMU_OPT_STRING, | |
c6447510 | 506 | .help = "host AIO implementation (threads, native, io_uring)", |
0a4279d9 | 507 | }, |
16b48d5d FZ |
508 | { |
509 | .name = "locking", | |
510 | .type = QEMU_OPT_STRING, | |
511 | .help = "file locking mode (on/off/auto, default: auto)", | |
512 | }, | |
7c9e5276 PB |
513 | { |
514 | .name = "pr-manager", | |
515 | .type = QEMU_OPT_STRING, | |
516 | .help = "id of persistent reservation manager object (default: none)", | |
517 | }, | |
f357fcd8 SH |
518 | #if defined(__linux__) |
519 | { | |
520 | .name = "drop-cache", | |
521 | .type = QEMU_OPT_BOOL, | |
522 | .help = "invalidate page cache during live migration (default: on)", | |
523 | }, | |
524 | #endif | |
31be8a2a SH |
525 | { |
526 | .name = "x-check-cache-dropped", | |
527 | .type = QEMU_OPT_BOOL, | |
528 | .help = "check that page cache was dropped on live migration (default: off)" | |
529 | }, | |
c66a6157 KW |
530 | { /* end of list */ } |
531 | }, | |
532 | }; | |
533 | ||
8a2ce0bc AG |
534 | static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; |
535 | ||
c66a6157 | 536 | static int raw_open_common(BlockDriverState *bs, QDict *options, |
230ff739 JS |
537 | int bdrv_flags, int open_flags, |
538 | bool device, Error **errp) | |
83f64091 FB |
539 | { |
540 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
541 | QemuOpts *opts; |
542 | Error *local_err = NULL; | |
8bfea15d | 543 | const char *filename = NULL; |
7c9e5276 | 544 | const char *str; |
0a4279d9 | 545 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 546 | int fd, ret; |
260a82e5 | 547 | struct stat st; |
244a5668 | 548 | OnOffAuto locking; |
83f64091 | 549 | |
87ea75d5 | 550 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 551 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
c66a6157 KW |
552 | ret = -EINVAL; |
553 | goto fail; | |
554 | } | |
555 | ||
556 | filename = qemu_opt_get(opts, "filename"); | |
557 | ||
db0754df | 558 | ret = raw_normalize_devicepath(&filename, errp); |
1de1ae0a | 559 | if (ret != 0) { |
c66a6157 | 560 | goto fail; |
1de1ae0a CE |
561 | } |
562 | ||
c6447510 AM |
563 | if (bdrv_flags & BDRV_O_NATIVE_AIO) { |
564 | aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; | |
565 | #ifdef CONFIG_LINUX_IO_URING | |
566 | } else if (bdrv_flags & BDRV_O_IO_URING) { | |
567 | aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; | |
568 | #endif | |
569 | } else { | |
570 | aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; | |
571 | } | |
572 | ||
f7abe0ec MAL |
573 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
574 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 575 | aio_default, &local_err); |
0a4279d9 KW |
576 | if (local_err) { |
577 | error_propagate(errp, local_err); | |
578 | ret = -EINVAL; | |
579 | goto fail; | |
580 | } | |
c6447510 | 581 | |
0a4279d9 | 582 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); |
c6447510 AM |
583 | #ifdef CONFIG_LINUX_IO_URING |
584 | s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); | |
585 | #endif | |
0a4279d9 | 586 | |
f7abe0ec MAL |
587 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
588 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 589 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
590 | if (local_err) { |
591 | error_propagate(errp, local_err); | |
592 | ret = -EINVAL; | |
593 | goto fail; | |
594 | } | |
595 | switch (locking) { | |
596 | case ON_OFF_AUTO_ON: | |
597 | s->use_lock = true; | |
2b218f5d | 598 | if (!qemu_has_ofd_lock()) { |
db0754df FZ |
599 | warn_report("File lock requested but OFD locking syscall is " |
600 | "unavailable, falling back to POSIX file locks"); | |
601 | error_printf("Due to the implementation, locks can be lost " | |
602 | "unexpectedly.\n"); | |
2b218f5d | 603 | } |
244a5668 FZ |
604 | break; |
605 | case ON_OFF_AUTO_OFF: | |
606 | s->use_lock = false; | |
607 | break; | |
608 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 609 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
610 | break; |
611 | default: | |
612 | abort(); | |
613 | } | |
614 | ||
7c9e5276 PB |
615 | str = qemu_opt_get(opts, "pr-manager"); |
616 | if (str) { | |
617 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
618 | if (local_err) { | |
619 | error_propagate(errp, local_err); | |
620 | ret = -EINVAL; | |
621 | goto fail; | |
622 | } | |
623 | } | |
624 | ||
f357fcd8 | 625 | s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); |
31be8a2a SH |
626 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
627 | false); | |
628 | ||
6a8dc042 | 629 | s->open_flags = open_flags; |
23dece19 | 630 | raw_parse_flags(bdrv_flags, &s->open_flags, false); |
83f64091 | 631 | |
90babde0 | 632 | s->fd = -1; |
b18a24a9 | 633 | fd = qemu_open(filename, s->open_flags, errp); |
64107dc0 KW |
634 | ret = fd < 0 ? -errno : 0; |
635 | ||
64107dc0 | 636 | if (ret < 0) { |
c66a6157 | 637 | if (ret == -EROFS) { |
19cb3738 | 638 | ret = -EACCES; |
c66a6157 KW |
639 | } |
640 | goto fail; | |
19cb3738 | 641 | } |
83f64091 | 642 | s->fd = fd; |
9ef91a67 | 643 | |
bca5283b KW |
644 | /* Check s->open_flags rather than bdrv_flags due to auto-read-only */ |
645 | if (s->open_flags & O_RDWR) { | |
646 | ret = check_hdev_writable(s->fd); | |
647 | if (ret < 0) { | |
648 | error_setg_errno(errp, -ret, "The device is not writable"); | |
649 | goto fail; | |
650 | } | |
651 | } | |
652 | ||
244a5668 FZ |
653 | s->perm = 0; |
654 | s->shared_perm = BLK_PERM_ALL; | |
655 | ||
5c6c3a6c | 656 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 657 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
658 | if (s->use_linux_aio) { |
659 | if (!(s->open_flags & O_DIRECT)) { | |
660 | error_setg(errp, "aio=native was specified, but it requires " | |
661 | "cache.direct=on, which was not specified."); | |
662 | ret = -EINVAL; | |
663 | goto fail; | |
664 | } | |
665 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
666 | error_prepend(errp, "Unable to use native AIO: "); | |
667 | goto fail; | |
668 | } | |
96518254 | 669 | } |
1501ecc1 | 670 | #else |
0a4279d9 | 671 | if (s->use_linux_aio) { |
d657c0c2 KW |
672 | error_setg(errp, "aio=native was specified, but is not supported " |
673 | "in this build."); | |
674 | ret = -EINVAL; | |
675 | goto fail; | |
1501ecc1 SH |
676 | } |
677 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 678 | |
c6447510 AM |
679 | #ifdef CONFIG_LINUX_IO_URING |
680 | if (s->use_linux_io_uring) { | |
681 | if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { | |
682 | error_prepend(errp, "Unable to use io_uring: "); | |
683 | goto fail; | |
684 | } | |
685 | } | |
686 | #else | |
687 | if (s->use_linux_io_uring) { | |
688 | error_setg(errp, "aio=io_uring was specified, but is not supported " | |
689 | "in this build."); | |
690 | ret = -EINVAL; | |
691 | goto fail; | |
692 | } | |
693 | #endif /* !defined(CONFIG_LINUX_IO_URING) */ | |
694 | ||
7ce21016 | 695 | s->has_discard = true; |
97a2ae34 | 696 | s->has_write_zeroes = true; |
5edc8557 | 697 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { |
3cad8307 RPM |
698 | s->needs_alignment = true; |
699 | } | |
260a82e5 PB |
700 | |
701 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 702 | ret = -errno; |
260a82e5 PB |
703 | error_setg_errno(errp, errno, "Could not stat file"); |
704 | goto fail; | |
705 | } | |
230ff739 JS |
706 | |
707 | if (!device) { | |
708 | if (S_ISBLK(st.st_mode)) { | |
709 | warn_report("Opening a block device as a file using the '%s' " | |
710 | "driver is deprecated", bs->drv->format_name); | |
711 | } else if (S_ISCHR(st.st_mode)) { | |
712 | warn_report("Opening a character device as a file using the '%s' " | |
713 | "driver is deprecated", bs->drv->format_name); | |
714 | } else if (!S_ISREG(st.st_mode)) { | |
715 | error_setg(errp, "A regular file was expected by the '%s' driver, " | |
716 | "but something else was given", bs->drv->format_name); | |
717 | ret = -EINVAL; | |
718 | goto fail; | |
719 | } else { | |
720 | s->discard_zeroes = true; | |
721 | s->has_fallocate = true; | |
722 | } | |
723 | } else { | |
724 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
725 | error_setg(errp, "'%s' driver expects either " | |
726 | "a character or block device", bs->drv->format_name); | |
727 | ret = -EINVAL; | |
728 | goto fail; | |
729 | } | |
260a82e5 | 730 | } |
230ff739 | 731 | |
d0b4503e PB |
732 | if (S_ISBLK(st.st_mode)) { |
733 | #ifdef BLKDISCARDZEROES | |
734 | unsigned int arg; | |
735 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
736 | s->discard_zeroes = true; | |
737 | } | |
738 | #endif | |
739 | #ifdef __linux__ | |
740 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
741 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 742 | * Same for BLKZEROOUT. |
d0b4503e PB |
743 | */ |
744 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
745 | s->discard_zeroes = false; | |
97a2ae34 | 746 | s->has_write_zeroes = false; |
d0b4503e PB |
747 | } |
748 | #endif | |
749 | } | |
3cad8307 RPM |
750 | #ifdef __FreeBSD__ |
751 | if (S_ISCHR(st.st_mode)) { | |
752 | /* | |
753 | * The file is a char device (disk), which on FreeBSD isn't behind | |
754 | * a pager, so force all requests to be aligned. This is needed | |
755 | * so QEMU makes sure all IO operations on the device are aligned | |
756 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
757 | */ | |
758 | s->needs_alignment = true; | |
759 | } | |
760 | #endif | |
260a82e5 | 761 | |
dce512de CH |
762 | #ifdef CONFIG_XFS |
763 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 764 | s->is_xfs = true; |
dce512de CH |
765 | } |
766 | #endif | |
767 | ||
738301e1 | 768 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
2f0c6e7a KW |
769 | if (S_ISREG(st.st_mode)) { |
770 | /* When extending regular files, we get zeros from the OS */ | |
771 | bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; | |
772 | } | |
c66a6157 KW |
773 | ret = 0; |
774 | fail: | |
a8c5cf27 KW |
775 | if (ret < 0 && s->fd != -1) { |
776 | qemu_close(s->fd); | |
777 | } | |
8bfea15d KW |
778 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
779 | unlink(filename); | |
780 | } | |
c66a6157 KW |
781 | qemu_opts_del(opts); |
782 | return ret; | |
83f64091 FB |
783 | } |
784 | ||
015a1036 HR |
785 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
786 | Error **errp) | |
90babde0 CH |
787 | { |
788 | BDRVRawState *s = bs->opaque; | |
789 | ||
790 | s->type = FTYPE_FILE; | |
230ff739 | 791 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
792 | } |
793 | ||
244a5668 FZ |
794 | typedef enum { |
795 | RAW_PL_PREPARE, | |
796 | RAW_PL_COMMIT, | |
797 | RAW_PL_ABORT, | |
798 | } RawPermLockOp; | |
799 | ||
800 | #define PERM_FOREACH(i) \ | |
801 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
802 | ||
803 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
804 | * file; if @unlock == true, also unlock the unneeded bytes. | |
805 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
806 | */ | |
2996ffad | 807 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
808 | uint64_t perm_lock_bits, |
809 | uint64_t shared_perm_lock_bits, | |
810 | bool unlock, Error **errp) | |
811 | { | |
812 | int ret; | |
813 | int i; | |
2996ffad FZ |
814 | uint64_t locked_perm, locked_shared_perm; |
815 | ||
816 | if (s) { | |
817 | locked_perm = s->locked_perm; | |
818 | locked_shared_perm = s->locked_shared_perm; | |
819 | } else { | |
820 | /* | |
821 | * We don't have the previous bits, just lock/unlock for each of the | |
822 | * requested bits. | |
823 | */ | |
824 | if (unlock) { | |
825 | locked_perm = BLK_PERM_ALL; | |
826 | locked_shared_perm = BLK_PERM_ALL; | |
827 | } else { | |
828 | locked_perm = 0; | |
829 | locked_shared_perm = 0; | |
830 | } | |
831 | } | |
244a5668 FZ |
832 | |
833 | PERM_FOREACH(i) { | |
834 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
835 | uint64_t bit = (1ULL << i); |
836 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 837 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
838 | if (ret) { |
839 | error_setg(errp, "Failed to lock byte %d", off); | |
840 | return ret; | |
2996ffad FZ |
841 | } else if (s) { |
842 | s->locked_perm |= bit; | |
244a5668 | 843 | } |
2996ffad | 844 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 845 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
846 | if (ret) { |
847 | error_setg(errp, "Failed to unlock byte %d", off); | |
848 | return ret; | |
2996ffad FZ |
849 | } else if (s) { |
850 | s->locked_perm &= ~bit; | |
244a5668 FZ |
851 | } |
852 | } | |
853 | } | |
854 | PERM_FOREACH(i) { | |
855 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
856 | uint64_t bit = (1ULL << i); |
857 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 858 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
859 | if (ret) { |
860 | error_setg(errp, "Failed to lock byte %d", off); | |
861 | return ret; | |
2996ffad FZ |
862 | } else if (s) { |
863 | s->locked_shared_perm |= bit; | |
244a5668 | 864 | } |
2996ffad FZ |
865 | } else if (unlock && (locked_shared_perm & bit) && |
866 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 867 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
868 | if (ret) { |
869 | error_setg(errp, "Failed to unlock byte %d", off); | |
870 | return ret; | |
2996ffad FZ |
871 | } else if (s) { |
872 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
873 | } |
874 | } | |
875 | } | |
876 | return 0; | |
877 | } | |
878 | ||
879 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 880 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
881 | Error **errp) |
882 | { | |
883 | int ret; | |
884 | int i; | |
885 | ||
886 | PERM_FOREACH(i) { | |
887 | int off = RAW_LOCK_SHARED_BASE + i; | |
888 | uint64_t p = 1ULL << i; | |
889 | if (perm & p) { | |
d0a96155 | 890 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
891 | if (ret) { |
892 | char *perm_name = bdrv_perm_names(p); | |
893 | error_setg(errp, | |
894 | "Failed to get \"%s\" lock", | |
895 | perm_name); | |
896 | g_free(perm_name); | |
244a5668 FZ |
897 | return ret; |
898 | } | |
899 | } | |
900 | } | |
901 | PERM_FOREACH(i) { | |
902 | int off = RAW_LOCK_PERM_BASE + i; | |
903 | uint64_t p = 1ULL << i; | |
904 | if (!(shared_perm & p)) { | |
d0a96155 | 905 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
906 | if (ret) { |
907 | char *perm_name = bdrv_perm_names(p); | |
908 | error_setg(errp, | |
909 | "Failed to get shared \"%s\" lock", | |
910 | perm_name); | |
911 | g_free(perm_name); | |
244a5668 FZ |
912 | return ret; |
913 | } | |
914 | } | |
915 | } | |
916 | return 0; | |
917 | } | |
918 | ||
919 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
920 | RawPermLockOp op, | |
921 | uint64_t new_perm, uint64_t new_shared, | |
922 | Error **errp) | |
923 | { | |
924 | BDRVRawState *s = bs->opaque; | |
925 | int ret = 0; | |
926 | Error *local_err = NULL; | |
927 | ||
928 | if (!s->use_lock) { | |
929 | return 0; | |
930 | } | |
931 | ||
932 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
933 | return 0; | |
934 | } | |
935 | ||
244a5668 FZ |
936 | switch (op) { |
937 | case RAW_PL_PREPARE: | |
696aaaed VSO |
938 | if ((s->perm | new_perm) == s->perm && |
939 | (s->shared_perm & new_shared) == s->shared_perm) | |
940 | { | |
941 | /* | |
942 | * We are going to unlock bytes, it should not fail. If it fail due | |
943 | * to some fs-dependent permission-unrelated reasons (which occurs | |
944 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
945 | * can't prevent such errors by any check here. And we ignore them | |
946 | * anyway in ABORT and COMMIT. | |
947 | */ | |
948 | return 0; | |
949 | } | |
f2e3af29 | 950 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
951 | ~s->shared_perm | ~new_shared, |
952 | false, errp); | |
953 | if (!ret) { | |
f2e3af29 | 954 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
955 | if (!ret) { |
956 | return 0; | |
957 | } | |
b857431d FZ |
958 | error_append_hint(errp, |
959 | "Is another process using the image [%s]?\n", | |
960 | bs->filename); | |
244a5668 | 961 | } |
244a5668 FZ |
962 | /* fall through to unlock bytes. */ |
963 | case RAW_PL_ABORT: | |
f2e3af29 | 964 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 965 | true, &local_err); |
244a5668 FZ |
966 | if (local_err) { |
967 | /* Theoretically the above call only unlocks bytes and it cannot | |
968 | * fail. Something weird happened, report it. | |
969 | */ | |
db0754df | 970 | warn_report_err(local_err); |
244a5668 FZ |
971 | } |
972 | break; | |
973 | case RAW_PL_COMMIT: | |
f2e3af29 | 974 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 975 | true, &local_err); |
244a5668 FZ |
976 | if (local_err) { |
977 | /* Theoretically the above call only unlocks bytes and it cannot | |
978 | * fail. Something weird happened, report it. | |
979 | */ | |
db0754df | 980 | warn_report_err(local_err); |
244a5668 FZ |
981 | } |
982 | break; | |
983 | } | |
984 | return ret; | |
985 | } | |
986 | ||
5cec2870 | 987 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 988 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 989 | Error **errp) |
5cec2870 KW |
990 | { |
991 | BDRVRawState *s = bs->opaque; | |
992 | int fd = -1; | |
993 | int ret; | |
23dece19 KW |
994 | bool has_writers = perm & |
995 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
996 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
997 | #ifdef O_NOATIME | |
998 | fcntl_flags |= O_NOATIME; | |
999 | #endif | |
1000 | ||
1001 | *open_flags = 0; | |
1002 | if (s->type == FTYPE_CD) { | |
1003 | *open_flags |= O_NONBLOCK; | |
1004 | } | |
1005 | ||
23dece19 | 1006 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
1007 | |
1008 | #ifdef O_ASYNC | |
1009 | /* Not all operating systems have O_ASYNC, and those that don't | |
1010 | * will not let us track the state into rs->open_flags (typically | |
1011 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
1012 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
1013 | */ | |
1014 | assert((s->open_flags & O_ASYNC) == 0); | |
1015 | #endif | |
1016 | ||
6ceabe6f KW |
1017 | if (!force_dup && *open_flags == s->open_flags) { |
1018 | /* We're lucky, the existing fd is fine */ | |
1019 | return s->fd; | |
1020 | } | |
1021 | ||
5cec2870 KW |
1022 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
1023 | /* dup the original fd */ | |
1024 | fd = qemu_dup(s->fd); | |
1025 | if (fd >= 0) { | |
1026 | ret = fcntl_setfl(fd, *open_flags); | |
1027 | if (ret) { | |
1028 | qemu_close(fd); | |
1029 | fd = -1; | |
1030 | } | |
1031 | } | |
1032 | } | |
1033 | ||
b18a24a9 | 1034 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ |
5cec2870 KW |
1035 | if (fd == -1) { |
1036 | const char *normalized_filename = bs->filename; | |
1037 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
1038 | if (ret >= 0) { | |
b18a24a9 | 1039 | fd = qemu_open(normalized_filename, *open_flags, errp); |
5cec2870 | 1040 | if (fd == -1) { |
5cec2870 KW |
1041 | return -1; |
1042 | } | |
1043 | } | |
1044 | } | |
1045 | ||
bca5283b KW |
1046 | if (fd != -1 && (*open_flags & O_RDWR)) { |
1047 | ret = check_hdev_writable(fd); | |
1048 | if (ret < 0) { | |
1049 | qemu_close(fd); | |
1050 | error_setg_errno(errp, -ret, "The device is not writable"); | |
1051 | return -1; | |
1052 | } | |
1053 | } | |
1054 | ||
5cec2870 KW |
1055 | return fd; |
1056 | } | |
1057 | ||
eeb6b45d JC |
1058 | static int raw_reopen_prepare(BDRVReopenState *state, |
1059 | BlockReopenQueue *queue, Error **errp) | |
1060 | { | |
1061 | BDRVRawState *s; | |
4e6d13c9 | 1062 | BDRVRawReopenState *rs; |
31be8a2a | 1063 | QemuOpts *opts; |
a6aeca0c | 1064 | int ret; |
df26a350 | 1065 | Error *local_err = NULL; |
eeb6b45d JC |
1066 | |
1067 | assert(state != NULL); | |
1068 | assert(state->bs != NULL); | |
1069 | ||
1070 | s = state->bs->opaque; | |
1071 | ||
5839e53b | 1072 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 1073 | rs = state->opaque; |
31be8a2a SH |
1074 | |
1075 | /* Handle options changes */ | |
1076 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 1077 | if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { |
31be8a2a SH |
1078 | ret = -EINVAL; |
1079 | goto out; | |
1080 | } | |
1081 | ||
f357fcd8 | 1082 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
1083 | rs->check_cache_dropped = |
1084 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
1085 | ||
1086 | /* This driver's reopen function doesn't currently allow changing | |
1087 | * other options, so let's put them back in the original QDict and | |
1088 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
1089 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 1090 | |
5cec2870 | 1091 | rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, |
23dece19 | 1092 | state->perm, true, &local_err); |
5cec2870 KW |
1093 | if (local_err) { |
1094 | error_propagate(errp, local_err); | |
1095 | ret = -1; | |
1096 | goto out; | |
eeb6b45d | 1097 | } |
df26a350 KW |
1098 | |
1099 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
1100 | * alignment with the new fd. */ | |
4e6d13c9 FZ |
1101 | if (rs->fd != -1) { |
1102 | raw_probe_alignment(state->bs, rs->fd, &local_err); | |
df26a350 | 1103 | if (local_err) { |
df26a350 KW |
1104 | error_propagate(errp, local_err); |
1105 | ret = -EINVAL; | |
a6aeca0c KW |
1106 | goto out_fd; |
1107 | } | |
df26a350 KW |
1108 | } |
1109 | ||
e0c9cf3a | 1110 | s->reopen_state = state; |
a6aeca0c KW |
1111 | ret = 0; |
1112 | out_fd: | |
1113 | if (ret < 0) { | |
1114 | qemu_close(rs->fd); | |
1115 | rs->fd = -1; | |
1116 | } | |
31be8a2a SH |
1117 | out: |
1118 | qemu_opts_del(opts); | |
eeb6b45d JC |
1119 | return ret; |
1120 | } | |
1121 | ||
eeb6b45d JC |
1122 | static void raw_reopen_commit(BDRVReopenState *state) |
1123 | { | |
4e6d13c9 | 1124 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1125 | BDRVRawState *s = state->bs->opaque; |
1126 | ||
f357fcd8 | 1127 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1128 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1129 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1130 | |
1131 | qemu_close(s->fd); | |
4e6d13c9 | 1132 | s->fd = rs->fd; |
eeb6b45d JC |
1133 | |
1134 | g_free(state->opaque); | |
1135 | state->opaque = NULL; | |
e0c9cf3a KW |
1136 | |
1137 | assert(s->reopen_state == state); | |
1138 | s->reopen_state = NULL; | |
eeb6b45d JC |
1139 | } |
1140 | ||
1141 | ||
1142 | static void raw_reopen_abort(BDRVReopenState *state) | |
1143 | { | |
4e6d13c9 | 1144 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1145 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1146 | |
1147 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1148 | if (rs == NULL) { |
eeb6b45d JC |
1149 | return; |
1150 | } | |
1151 | ||
4e6d13c9 FZ |
1152 | if (rs->fd >= 0) { |
1153 | qemu_close(rs->fd); | |
1154 | rs->fd = -1; | |
eeb6b45d JC |
1155 | } |
1156 | g_free(state->opaque); | |
1157 | state->opaque = NULL; | |
e0c9cf3a KW |
1158 | |
1159 | assert(s->reopen_state == state); | |
1160 | s->reopen_state = NULL; | |
eeb6b45d JC |
1161 | } |
1162 | ||
867eccfe | 1163 | static int sg_get_max_transfer_length(int fd) |
6f607174 FZ |
1164 | { |
1165 | #ifdef BLKSECTGET | |
48265250 | 1166 | int max_bytes = 0; |
867eccfe ML |
1167 | |
1168 | if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
48265250 | 1169 | return max_bytes; |
6f607174 FZ |
1170 | } else { |
1171 | return -errno; | |
1172 | } | |
1173 | #else | |
1174 | return -ENOSYS; | |
1175 | #endif | |
1176 | } | |
1177 | ||
867eccfe | 1178 | static int sg_get_max_segments(int fd) |
9103f1ce FZ |
1179 | { |
1180 | #ifdef CONFIG_LINUX | |
1181 | char buf[32]; | |
1182 | const char *end; | |
867eccfe | 1183 | char *sysfspath = NULL; |
9103f1ce | 1184 | int ret; |
867eccfe | 1185 | int sysfd = -1; |
9103f1ce | 1186 | long max_segments; |
867eccfe ML |
1187 | struct stat st; |
1188 | ||
1189 | if (fstat(fd, &st)) { | |
1190 | ret = -errno; | |
1191 | goto out; | |
1192 | } | |
9103f1ce FZ |
1193 | |
1194 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", | |
867eccfe ML |
1195 | major(st.st_rdev), minor(st.st_rdev)); |
1196 | sysfd = open(sysfspath, O_RDONLY); | |
1197 | if (sysfd == -1) { | |
9103f1ce FZ |
1198 | ret = -errno; |
1199 | goto out; | |
1200 | } | |
1201 | do { | |
867eccfe | 1202 | ret = read(sysfd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1203 | } while (ret == -1 && errno == EINTR); |
1204 | if (ret < 0) { | |
1205 | ret = -errno; | |
1206 | goto out; | |
1207 | } else if (ret == 0) { | |
1208 | ret = -EIO; | |
1209 | goto out; | |
1210 | } | |
1211 | buf[ret] = 0; | |
1212 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1213 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1214 | if (ret == 0 && end && *end == '\n') { | |
1215 | ret = max_segments; | |
1216 | } | |
1217 | ||
1218 | out: | |
867eccfe ML |
1219 | if (sysfd != -1) { |
1220 | close(sysfd); | |
fed414df | 1221 | } |
9103f1ce FZ |
1222 | g_free(sysfspath); |
1223 | return ret; | |
1224 | #else | |
1225 | return -ENOTSUP; | |
1226 | #endif | |
1227 | } | |
1228 | ||
3baca891 | 1229 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1230 | { |
1231 | BDRVRawState *s = bs->opaque; | |
6f607174 | 1232 | |
867eccfe ML |
1233 | if (bs->sg) { |
1234 | int ret = sg_get_max_transfer_length(s->fd); | |
1235 | ||
1236 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
1237 | bs->bl.max_transfer = pow2floor(ret); | |
1238 | } | |
1239 | ||
1240 | ret = sg_get_max_segments(s->fd); | |
1241 | if (ret > 0) { | |
038adc2f WY |
1242 | bs->bl.max_transfer = MIN(bs->bl.max_transfer, |
1243 | ret * qemu_real_host_page_size); | |
6f607174 FZ |
1244 | } |
1245 | } | |
eeb6b45d | 1246 | |
df26a350 | 1247 | raw_probe_alignment(bs, s->fd, errp); |
4196d2f0 | 1248 | bs->bl.min_mem_alignment = s->buf_align; |
038adc2f | 1249 | bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); |
c25f53b0 | 1250 | } |
83f64091 | 1251 | |
1a9335e4 ET |
1252 | static int check_for_dasd(int fd) |
1253 | { | |
1254 | #ifdef BIODASDINFO2 | |
1255 | struct dasd_information2_t info = {0}; | |
1256 | ||
1257 | return ioctl(fd, BIODASDINFO2, &info); | |
1258 | #else | |
1259 | return -1; | |
1260 | #endif | |
1261 | } | |
1262 | ||
1263 | /** | |
1264 | * Try to get @bs's logical and physical block size. | |
1265 | * On success, store them in @bsz and return zero. | |
1266 | * On failure, return negative errno. | |
1267 | */ | |
1268 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1269 | { | |
1270 | BDRVRawState *s = bs->opaque; | |
1271 | int ret; | |
1272 | ||
1273 | /* If DASD, get blocksizes */ | |
1274 | if (check_for_dasd(s->fd) < 0) { | |
1275 | return -ENOTSUP; | |
1276 | } | |
1277 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1278 | if (ret < 0) { | |
1279 | return ret; | |
1280 | } | |
1281 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1282 | } | |
1283 | ||
1284 | /** | |
1285 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1286 | * On success, store them in @geo and return 0. | |
1287 | * On failure return -errno. | |
1288 | * (Allows block driver to assign default geometry values that guest sees) | |
1289 | */ | |
1290 | #ifdef __linux__ | |
1291 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1292 | { | |
1293 | BDRVRawState *s = bs->opaque; | |
1294 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1295 | |
1296 | /* If DASD, get its geometry */ | |
1297 | if (check_for_dasd(s->fd) < 0) { | |
1298 | return -ENOTSUP; | |
1299 | } | |
1300 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1301 | return -errno; | |
1302 | } | |
1303 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1304 | (e.g. certain multipath setups) */ | |
1305 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1306 | return -ENOTSUP; | |
1307 | } | |
1308 | /* Do not return a geometry for partition */ | |
1309 | if (ioctl_geo.start != 0) { | |
1310 | return -ENOTSUP; | |
1311 | } | |
1312 | geo->heads = ioctl_geo.heads; | |
1313 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1314 | geo->cylinders = ioctl_geo.cylinders; |
1315 | ||
1316 | return 0; | |
1317 | } | |
1318 | #else /* __linux__ */ | |
1319 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1320 | { | |
1321 | return -ENOTSUP; | |
1322 | } | |
1323 | #endif | |
1324 | ||
03425671 KW |
1325 | #if defined(__linux__) |
1326 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1327 | { |
03425671 | 1328 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1329 | int ret; |
1330 | ||
d57c44d0 | 1331 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); |
de81a169 PB |
1332 | if (ret == -1) { |
1333 | return -errno; | |
1334 | } | |
1335 | ||
b608c8dc | 1336 | return 0; |
de81a169 | 1337 | } |
03425671 | 1338 | #endif /* linux */ |
de81a169 | 1339 | |
06dc9bd5 | 1340 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1341 | { |
06dc9bd5 | 1342 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1343 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1344 | int ret; |
1345 | ||
e5bcf967 KW |
1346 | if (s->page_cache_inconsistent) { |
1347 | return -EIO; | |
1348 | } | |
1349 | ||
de81a169 PB |
1350 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1351 | if (ret == -1) { | |
e5bcf967 KW |
1352 | /* There is no clear definition of the semantics of a failing fsync(), |
1353 | * so we may have to assume the worst. The sad truth is that this | |
1354 | * assumption is correct for Linux. Some pages are now probably marked | |
1355 | * clean in the page cache even though they are inconsistent with the | |
1356 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1357 | * further writeback attempt will be made. We can't get back to a state | |
1358 | * in which we know what is on disk (we would have to rewrite | |
1359 | * everything that was touched since the last fdatasync() at least), so | |
1360 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1361 | * really defined, I have little hope that other OSes are doing better. | |
1362 | * | |
1363 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1364 | * cache. */ | |
1365 | if ((s->open_flags & O_DIRECT) == 0) { | |
1366 | s->page_cache_inconsistent = true; | |
1367 | } | |
de81a169 PB |
1368 | return -errno; |
1369 | } | |
1370 | return 0; | |
1371 | } | |
1372 | ||
1373 | #ifdef CONFIG_PREADV | |
1374 | ||
1375 | static bool preadv_present = true; | |
1376 | ||
1377 | static ssize_t | |
1378 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1379 | { | |
1380 | return preadv(fd, iov, nr_iov, offset); | |
1381 | } | |
1382 | ||
1383 | static ssize_t | |
1384 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1385 | { | |
1386 | return pwritev(fd, iov, nr_iov, offset); | |
1387 | } | |
1388 | ||
1389 | #else | |
1390 | ||
1391 | static bool preadv_present = false; | |
1392 | ||
1393 | static ssize_t | |
1394 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1395 | { | |
1396 | return -ENOSYS; | |
1397 | } | |
1398 | ||
1399 | static ssize_t | |
1400 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1401 | { | |
1402 | return -ENOSYS; | |
1403 | } | |
1404 | ||
1405 | #endif | |
1406 | ||
1407 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1408 | { | |
1409 | ssize_t len; | |
1410 | ||
1411 | do { | |
1412 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1413 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1414 | aiocb->io.iov, |
1415 | aiocb->io.niov, | |
de81a169 PB |
1416 | aiocb->aio_offset); |
1417 | else | |
1418 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1419 | aiocb->io.iov, |
1420 | aiocb->io.niov, | |
de81a169 PB |
1421 | aiocb->aio_offset); |
1422 | } while (len == -1 && errno == EINTR); | |
1423 | ||
1424 | if (len == -1) { | |
1425 | return -errno; | |
1426 | } | |
1427 | return len; | |
1428 | } | |
1429 | ||
1430 | /* | |
1431 | * Read/writes the data to/from a given linear buffer. | |
1432 | * | |
1433 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1434 | * reads are only returned if the end of the file is reached. | |
1435 | */ | |
1436 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1437 | { | |
1438 | ssize_t offset = 0; | |
1439 | ssize_t len; | |
1440 | ||
1441 | while (offset < aiocb->aio_nbytes) { | |
1442 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1443 | len = pwrite(aiocb->aio_fildes, | |
1444 | (const char *)buf + offset, | |
1445 | aiocb->aio_nbytes - offset, | |
1446 | aiocb->aio_offset + offset); | |
1447 | } else { | |
1448 | len = pread(aiocb->aio_fildes, | |
1449 | buf + offset, | |
1450 | aiocb->aio_nbytes - offset, | |
1451 | aiocb->aio_offset + offset); | |
1452 | } | |
1453 | if (len == -1 && errno == EINTR) { | |
1454 | continue; | |
61ed73cf SH |
1455 | } else if (len == -1 && errno == EINVAL && |
1456 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1457 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1458 | offset > 0) { | |
1459 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1460 | * after a short read. Assume that O_DIRECT short reads only occur | |
1461 | * at EOF. Therefore this is a short read, not an I/O error. | |
1462 | */ | |
1463 | break; | |
de81a169 PB |
1464 | } else if (len == -1) { |
1465 | offset = -errno; | |
1466 | break; | |
1467 | } else if (len == 0) { | |
1468 | break; | |
1469 | } | |
1470 | offset += len; | |
1471 | } | |
1472 | ||
1473 | return offset; | |
1474 | } | |
1475 | ||
999e6b69 | 1476 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1477 | { |
999e6b69 | 1478 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1479 | ssize_t nbytes; |
1480 | char *buf; | |
1481 | ||
1482 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1483 | /* | |
1484 | * If there is just a single buffer, and it is properly aligned | |
1485 | * we can just use plain pread/pwrite without any problems. | |
1486 | */ | |
d57c44d0 | 1487 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1488 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1489 | goto out; | |
de81a169 PB |
1490 | } |
1491 | /* | |
1492 | * We have more than one iovec, and all are properly aligned. | |
1493 | * | |
1494 | * Try preadv/pwritev first and fall back to linearizing the | |
1495 | * buffer if it's not supported. | |
1496 | */ | |
1497 | if (preadv_present) { | |
1498 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1499 | if (nbytes == aiocb->aio_nbytes || | |
1500 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1501 | goto out; |
de81a169 PB |
1502 | } |
1503 | preadv_present = false; | |
1504 | } | |
1505 | ||
1506 | /* | |
1507 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1508 | * using these interfaces. For now retry using plain | |
1509 | * pread/pwrite? | |
1510 | */ | |
1511 | } | |
1512 | ||
1513 | /* | |
1514 | * Ok, we have to do it the hard way, copy all segments into | |
1515 | * a single aligned buffer. | |
1516 | */ | |
50d4a858 KW |
1517 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1518 | if (buf == NULL) { | |
54c7ca1b KW |
1519 | nbytes = -ENOMEM; |
1520 | goto out; | |
50d4a858 KW |
1521 | } |
1522 | ||
de81a169 PB |
1523 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1524 | char *p = buf; | |
1525 | int i; | |
1526 | ||
d57c44d0 KW |
1527 | for (i = 0; i < aiocb->io.niov; ++i) { |
1528 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1529 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1530 | } |
8eb029c2 | 1531 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1532 | } |
1533 | ||
1534 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1535 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1536 | char *p = buf; | |
1537 | size_t count = aiocb->aio_nbytes, copy; | |
1538 | int i; | |
1539 | ||
d57c44d0 | 1540 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1541 | copy = count; |
d57c44d0 KW |
1542 | if (copy > aiocb->io.iov[i].iov_len) { |
1543 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1544 | } |
d57c44d0 | 1545 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1546 | assert(count >= copy); |
de81a169 PB |
1547 | p += copy; |
1548 | count -= copy; | |
1549 | } | |
8eb029c2 | 1550 | assert(count == 0); |
de81a169 PB |
1551 | } |
1552 | qemu_vfree(buf); | |
1553 | ||
54c7ca1b KW |
1554 | out: |
1555 | if (nbytes == aiocb->aio_nbytes) { | |
1556 | return 0; | |
1557 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1558 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1559 | return -EINVAL; | |
1560 | } else { | |
1561 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1562 | 0, aiocb->aio_nbytes - nbytes); | |
1563 | return 0; | |
1564 | } | |
1565 | } else { | |
1566 | assert(nbytes < 0); | |
1567 | return nbytes; | |
1568 | } | |
de81a169 PB |
1569 | } |
1570 | ||
1486df0e DL |
1571 | static int translate_err(int err) |
1572 | { | |
1573 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1574 | err == -ENOTTY) { | |
1575 | err = -ENOTSUP; | |
1576 | } | |
1577 | return err; | |
1578 | } | |
1579 | ||
d50d8222 | 1580 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1581 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1582 | { | |
1583 | do { | |
1584 | if (fallocate(fd, mode, offset, len) == 0) { | |
1585 | return 0; | |
1586 | } | |
1587 | } while (errno == EINTR); | |
1588 | return translate_err(-errno); | |
1589 | } | |
1590 | #endif | |
1591 | ||
37cc9f7f | 1592 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1593 | { |
37cc9f7f | 1594 | int ret = -ENOTSUP; |
97a2ae34 PB |
1595 | BDRVRawState *s = aiocb->bs->opaque; |
1596 | ||
37cc9f7f | 1597 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1598 | return -ENOTSUP; |
1599 | } | |
1600 | ||
97a2ae34 | 1601 | #ifdef BLKZEROOUT |
738301e1 KW |
1602 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1603 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1604 | * fallbacks. */ | |
1605 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1606 | do { | |
1607 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1608 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1609 | return 0; | |
1610 | } | |
1611 | } while (errno == EINTR); | |
37cc9f7f | 1612 | |
738301e1 | 1613 | ret = translate_err(-errno); |
effecce6 KW |
1614 | if (ret == -ENOTSUP) { |
1615 | s->has_write_zeroes = false; | |
1616 | } | |
738301e1 | 1617 | } |
97a2ae34 | 1618 | #endif |
97a2ae34 | 1619 | |
97a2ae34 PB |
1620 | return ret; |
1621 | } | |
1622 | ||
7154d8ae | 1623 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1624 | { |
7154d8ae | 1625 | RawPosixAIOData *aiocb = opaque; |
70d9110b | 1626 | #ifdef CONFIG_FALLOCATE |
b2c6f23f | 1627 | BDRVRawState *s = aiocb->bs->opaque; |
70d9110b DL |
1628 | int64_t len; |
1629 | #endif | |
37cc9f7f DL |
1630 | |
1631 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1632 | return handle_aiocb_write_zeroes_block(aiocb); | |
1633 | } | |
1634 | ||
b953f075 DL |
1635 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1636 | if (s->has_write_zeroes) { | |
1637 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1638 | aiocb->aio_offset, aiocb->aio_nbytes); | |
294682cc AS |
1639 | if (ret == -EINVAL) { |
1640 | /* | |
1641 | * Allow falling back to pwrite for file systems that | |
1642 | * do not support fallocate() for an unaligned byte range. | |
1643 | */ | |
1644 | return -ENOTSUP; | |
1645 | } | |
b953f075 DL |
1646 | if (ret == 0 || ret != -ENOTSUP) { |
1647 | return ret; | |
1648 | } | |
1649 | s->has_write_zeroes = false; | |
1650 | } | |
1651 | #endif | |
1652 | ||
1cdc3239 DL |
1653 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1654 | if (s->has_discard && s->has_fallocate) { | |
1655 | int ret = do_fallocate(s->fd, | |
1656 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1657 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1658 | if (ret == 0) { | |
1659 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1660 | if (ret == 0 || ret != -ENOTSUP) { | |
1661 | return ret; | |
1662 | } | |
1663 | s->has_fallocate = false; | |
1664 | } else if (ret != -ENOTSUP) { | |
1665 | return ret; | |
1666 | } else { | |
1667 | s->has_discard = false; | |
1668 | } | |
1669 | } | |
1670 | #endif | |
1671 | ||
d50d8222 | 1672 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1673 | /* Last resort: we are trying to extend the file with zeroed data. This |
1674 | * can be done via fallocate(fd, 0) */ | |
1675 | len = bdrv_getlength(aiocb->bs); | |
1676 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1677 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1678 | if (ret == 0 || ret != -ENOTSUP) { | |
1679 | return ret; | |
1680 | } | |
1681 | s->has_fallocate = false; | |
1682 | } | |
1683 | #endif | |
1684 | ||
37cc9f7f DL |
1685 | return -ENOTSUP; |
1686 | } | |
1687 | ||
7154d8ae | 1688 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1689 | { |
7154d8ae | 1690 | RawPosixAIOData *aiocb = opaque; |
34fa110e | 1691 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
34fa110e KW |
1692 | |
1693 | /* First try to write zeros and unmap at the same time */ | |
1694 | ||
1695 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
b3ac2b94 SS |
1696 | int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1697 | aiocb->aio_offset, aiocb->aio_nbytes); | |
bae127d4 AD |
1698 | switch (ret) { |
1699 | case -ENOTSUP: | |
1700 | case -EINVAL: | |
ece4fa91 | 1701 | case -EBUSY: |
bae127d4 AD |
1702 | break; |
1703 | default: | |
34fa110e KW |
1704 | return ret; |
1705 | } | |
1706 | #endif | |
1707 | ||
34fa110e KW |
1708 | /* If we couldn't manage to unmap while guaranteed that the area reads as |
1709 | * all-zero afterwards, just write zeroes without unmapping */ | |
b3ac2b94 | 1710 | return handle_aiocb_write_zeroes(aiocb); |
34fa110e KW |
1711 | } |
1712 | ||
1efad060 FZ |
1713 | #ifndef HAVE_COPY_FILE_RANGE |
1714 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1715 | off_t *out_off, size_t len, unsigned int flags) | |
1716 | { | |
1717 | #ifdef __NR_copy_file_range | |
1718 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1719 | out_off, len, flags); | |
1720 | #else | |
1721 | errno = ENOSYS; | |
1722 | return -1; | |
1723 | #endif | |
1724 | } | |
1725 | #endif | |
1726 | ||
58a209c4 | 1727 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1728 | { |
58a209c4 | 1729 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1730 | uint64_t bytes = aiocb->aio_nbytes; |
1731 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1732 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1733 | |
1734 | while (bytes) { | |
1735 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1736 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1737 | bytes, 0); |
ecc983a5 | 1738 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1739 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1740 | 0, ret); | |
c436e3d0 FZ |
1741 | if (ret == 0) { |
1742 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1743 | * buffer I/O. */ | |
1744 | return -ENOSPC; | |
1efad060 FZ |
1745 | } |
1746 | if (ret < 0) { | |
c436e3d0 FZ |
1747 | switch (errno) { |
1748 | case ENOSYS: | |
1efad060 | 1749 | return -ENOTSUP; |
c436e3d0 FZ |
1750 | case EINTR: |
1751 | continue; | |
1752 | default: | |
1efad060 FZ |
1753 | return -errno; |
1754 | } | |
1755 | } | |
1efad060 FZ |
1756 | bytes -= ret; |
1757 | } | |
1758 | return 0; | |
1759 | } | |
1760 | ||
46ee0f46 | 1761 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1762 | { |
46ee0f46 | 1763 | RawPosixAIOData *aiocb = opaque; |
8238010b PB |
1764 | int ret = -EOPNOTSUPP; |
1765 | BDRVRawState *s = aiocb->bs->opaque; | |
1766 | ||
7ce21016 PB |
1767 | if (!s->has_discard) { |
1768 | return -ENOTSUP; | |
8238010b PB |
1769 | } |
1770 | ||
1771 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1772 | #ifdef BLKDISCARD | |
1773 | do { | |
1774 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1775 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1776 | return 0; | |
1777 | } | |
1778 | } while (errno == EINTR); | |
1779 | ||
1780 | ret = -errno; | |
1781 | #endif | |
1782 | } else { | |
8238010b | 1783 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
0b991712 DL |
1784 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1785 | aiocb->aio_offset, aiocb->aio_nbytes); | |
8238010b PB |
1786 | #endif |
1787 | } | |
1788 | ||
1486df0e DL |
1789 | ret = translate_err(ret); |
1790 | if (ret == -ENOTSUP) { | |
7ce21016 | 1791 | s->has_discard = false; |
8238010b PB |
1792 | } |
1793 | return ret; | |
1794 | } | |
1795 | ||
3a20013f NS |
1796 | /* |
1797 | * Help alignment probing by allocating the first block. | |
1798 | * | |
1799 | * When reading with direct I/O from unallocated area on Gluster backed by XFS, | |
1800 | * reading succeeds regardless of request length. In this case we fallback to | |
1801 | * safe alignment which is not optimal. Allocating the first block avoids this | |
1802 | * fallback. | |
1803 | * | |
1804 | * fd may be opened with O_DIRECT, but we don't know the buffer alignment or | |
1805 | * request alignment, so we use safe values. | |
1806 | * | |
1807 | * Returns: 0 on success, -errno on failure. Since this is an optimization, | |
1808 | * caller may ignore failures. | |
1809 | */ | |
1810 | static int allocate_first_block(int fd, size_t max_size) | |
1811 | { | |
1812 | size_t write_size = (max_size < MAX_BLOCKSIZE) | |
1813 | ? BDRV_SECTOR_SIZE | |
1814 | : MAX_BLOCKSIZE; | |
038adc2f | 1815 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); |
3a20013f NS |
1816 | void *buf; |
1817 | ssize_t n; | |
1818 | int ret; | |
1819 | ||
1820 | buf = qemu_memalign(max_align, write_size); | |
1821 | memset(buf, 0, write_size); | |
1822 | ||
1823 | do { | |
1824 | n = pwrite(fd, buf, write_size, 0); | |
1825 | } while (n == -1 && errno == EINTR); | |
1826 | ||
1827 | ret = (n == -1) ? -errno : 0; | |
1828 | ||
1829 | qemu_vfree(buf); | |
1830 | return ret; | |
1831 | } | |
1832 | ||
29cb4c01 | 1833 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1834 | { |
29cb4c01 | 1835 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1836 | int result = 0; |
1837 | int64_t current_length = 0; | |
1838 | char *buf = NULL; | |
1839 | struct stat st; | |
1840 | int fd = aiocb->aio_fildes; | |
1841 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1842 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1843 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1844 | |
1845 | if (fstat(fd, &st) < 0) { | |
1846 | result = -errno; | |
1847 | error_setg_errno(errp, -result, "Could not stat file"); | |
1848 | return result; | |
1849 | } | |
1850 | ||
1851 | current_length = st.st_size; | |
d57c44d0 | 1852 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1853 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1854 | return -ENOTSUP; | |
1855 | } | |
1856 | ||
d57c44d0 | 1857 | switch (prealloc) { |
93f4e2ff KW |
1858 | #ifdef CONFIG_POSIX_FALLOCATE |
1859 | case PREALLOC_MODE_FALLOC: | |
1860 | /* | |
1861 | * Truncating before posix_fallocate() makes it about twice slower on | |
1862 | * file systems that do not support fallocate(), trying to check if a | |
1863 | * block is allocated before allocating it, so don't do that here. | |
1864 | */ | |
1865 | if (offset != current_length) { | |
1866 | result = -posix_fallocate(fd, current_length, | |
1867 | offset - current_length); | |
1868 | if (result != 0) { | |
1869 | /* posix_fallocate() doesn't set errno. */ | |
1870 | error_setg_errno(errp, -result, | |
1871 | "Could not preallocate new data"); | |
3a20013f NS |
1872 | } else if (current_length == 0) { |
1873 | /* | |
1874 | * posix_fallocate() uses fallocate() if the filesystem | |
1875 | * supports it, or fallback to manually writing zeroes. If | |
1876 | * fallocate() was used, unaligned reads from the fallocated | |
1877 | * area in raw_probe_alignment() will succeed, hence we need to | |
1878 | * allocate the first block. | |
1879 | * | |
1880 | * Optimize future alignment probing; ignore failures. | |
1881 | */ | |
1882 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1883 | } |
1884 | } else { | |
1885 | result = 0; | |
1886 | } | |
1887 | goto out; | |
1888 | #endif | |
1889 | case PREALLOC_MODE_FULL: | |
1890 | { | |
1891 | int64_t num = 0, left = offset - current_length; | |
1892 | off_t seek_result; | |
1893 | ||
1894 | /* | |
1895 | * Knowing the final size from the beginning could allow the file | |
1896 | * system driver to do less allocations and possibly avoid | |
1897 | * fragmentation of the file. | |
1898 | */ | |
1899 | if (ftruncate(fd, offset) != 0) { | |
1900 | result = -errno; | |
1901 | error_setg_errno(errp, -result, "Could not resize file"); | |
1902 | goto out; | |
1903 | } | |
1904 | ||
1905 | buf = g_malloc0(65536); | |
1906 | ||
1907 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1908 | if (seek_result < 0) { | |
1909 | result = -errno; | |
1910 | error_setg_errno(errp, -result, | |
1911 | "Failed to seek to the old end of file"); | |
1912 | goto out; | |
1913 | } | |
1914 | ||
1915 | while (left > 0) { | |
1916 | num = MIN(left, 65536); | |
1917 | result = write(fd, buf, num); | |
1918 | if (result < 0) { | |
a1c81f4f FZ |
1919 | if (errno == EINTR) { |
1920 | continue; | |
1921 | } | |
93f4e2ff KW |
1922 | result = -errno; |
1923 | error_setg_errno(errp, -result, | |
1924 | "Could not write zeros for preallocation"); | |
1925 | goto out; | |
1926 | } | |
1927 | left -= result; | |
1928 | } | |
1929 | if (result >= 0) { | |
1930 | result = fsync(fd); | |
1931 | if (result < 0) { | |
1932 | result = -errno; | |
1933 | error_setg_errno(errp, -result, | |
1934 | "Could not flush file to disk"); | |
1935 | goto out; | |
1936 | } | |
1937 | } | |
1938 | goto out; | |
1939 | } | |
1940 | case PREALLOC_MODE_OFF: | |
1941 | if (ftruncate(fd, offset) != 0) { | |
1942 | result = -errno; | |
1943 | error_setg_errno(errp, -result, "Could not resize file"); | |
3a20013f NS |
1944 | } else if (current_length == 0 && offset > current_length) { |
1945 | /* Optimize future alignment probing; ignore failures. */ | |
1946 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1947 | } |
1948 | return result; | |
1949 | default: | |
1950 | result = -ENOTSUP; | |
1951 | error_setg(errp, "Unsupported preallocation mode: %s", | |
d57c44d0 | 1952 | PreallocMode_str(prealloc)); |
93f4e2ff KW |
1953 | return result; |
1954 | } | |
1955 | ||
1956 | out: | |
1957 | if (result < 0) { | |
1958 | if (ftruncate(fd, current_length) < 0) { | |
1959 | error_report("Failed to restore old file length: %s", | |
1960 | strerror(errno)); | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | g_free(buf); | |
1965 | return result; | |
1966 | } | |
1967 | ||
5d5de250 KW |
1968 | static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, |
1969 | ThreadPoolFunc func, void *arg) | |
1970 | { | |
1971 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ | |
1972 | ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
1973 | return thread_pool_submit_co(pool, func, arg); | |
1974 | } | |
1975 | ||
9d52aa3c KW |
1976 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
1977 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 1978 | { |
ce1a14dc | 1979 | BDRVRawState *s = bs->opaque; |
999e6b69 | 1980 | RawPosixAIOData acb; |
ce1a14dc | 1981 | |
19cb3738 | 1982 | if (fd_open(bs) < 0) |
2174f12b | 1983 | return -EIO; |
19cb3738 | 1984 | |
f141eafe | 1985 | /* |
c6447510 AM |
1986 | * When using O_DIRECT, the request must be aligned to be able to use |
1987 | * either libaio or io_uring interface. If not fail back to regular thread | |
1988 | * pool read/write code which emulates this for us if we | |
1989 | * set QEMU_AIO_MISALIGNED. | |
f141eafe | 1990 | */ |
c6447510 AM |
1991 | if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { |
1992 | type |= QEMU_AIO_MISALIGNED; | |
1993 | #ifdef CONFIG_LINUX_IO_URING | |
1994 | } else if (s->use_linux_io_uring) { | |
1995 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
1996 | assert(qiov->size == bytes); | |
1997 | return luring_co_submit(bs, aio, s->fd, offset, qiov, type); | |
1998 | #endif | |
e44bd6fc | 1999 | #ifdef CONFIG_LINUX_AIO |
c6447510 AM |
2000 | } else if (s->use_linux_aio) { |
2001 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); | |
2002 | assert(qiov->size == bytes); | |
2003 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type); | |
e44bd6fc | 2004 | #endif |
9ef91a67 | 2005 | } |
f141eafe | 2006 | |
999e6b69 KW |
2007 | acb = (RawPosixAIOData) { |
2008 | .bs = bs, | |
2009 | .aio_fildes = s->fd, | |
2010 | .aio_type = type, | |
2011 | .aio_offset = offset, | |
2012 | .aio_nbytes = bytes, | |
2013 | .io = { | |
2014 | .iov = qiov->iov, | |
2015 | .niov = qiov->niov, | |
2016 | }, | |
2017 | }; | |
2018 | ||
2019 | assert(qiov->size == bytes); | |
2020 | return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); | |
2174f12b KW |
2021 | } |
2022 | ||
9d52aa3c KW |
2023 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, |
2024 | uint64_t bytes, QEMUIOVector *qiov, | |
2025 | int flags) | |
2174f12b | 2026 | { |
9d52aa3c | 2027 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
2028 | } |
2029 | ||
9d52aa3c KW |
2030 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, |
2031 | uint64_t bytes, QEMUIOVector *qiov, | |
2032 | int flags) | |
2174f12b | 2033 | { |
9d52aa3c KW |
2034 | assert(flags == 0); |
2035 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
2036 | } |
2037 | ||
1b3abdcc ML |
2038 | static void raw_aio_plug(BlockDriverState *bs) |
2039 | { | |
c6447510 | 2040 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2041 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2042 | if (s->use_linux_aio) { |
0187f5c9 PB |
2043 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
2044 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
2045 | } |
2046 | #endif | |
c6447510 AM |
2047 | #ifdef CONFIG_LINUX_IO_URING |
2048 | if (s->use_linux_io_uring) { | |
2049 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2050 | luring_io_plug(bs, aio); | |
2051 | } | |
2052 | #endif | |
1b3abdcc ML |
2053 | } |
2054 | ||
2055 | static void raw_aio_unplug(BlockDriverState *bs) | |
2056 | { | |
c6447510 | 2057 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2058 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2059 | if (s->use_linux_aio) { |
0187f5c9 PB |
2060 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
2061 | laio_io_unplug(bs, aio); | |
1b3abdcc ML |
2062 | } |
2063 | #endif | |
c6447510 AM |
2064 | #ifdef CONFIG_LINUX_IO_URING |
2065 | if (s->use_linux_io_uring) { | |
2066 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2067 | luring_io_unplug(bs, aio); | |
2068 | } | |
2069 | #endif | |
1b3abdcc ML |
2070 | } |
2071 | ||
33d70fb6 | 2072 | static int raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
2073 | { |
2074 | BDRVRawState *s = bs->opaque; | |
06dc9bd5 | 2075 | RawPosixAIOData acb; |
33d70fb6 | 2076 | int ret; |
b2e12bc6 | 2077 | |
33d70fb6 KW |
2078 | ret = fd_open(bs); |
2079 | if (ret < 0) { | |
2080 | return ret; | |
2081 | } | |
b2e12bc6 | 2082 | |
06dc9bd5 KW |
2083 | acb = (RawPosixAIOData) { |
2084 | .bs = bs, | |
2085 | .aio_fildes = s->fd, | |
2086 | .aio_type = QEMU_AIO_FLUSH, | |
2087 | }; | |
2088 | ||
c6447510 AM |
2089 | #ifdef CONFIG_LINUX_IO_URING |
2090 | if (s->use_linux_io_uring) { | |
2091 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2092 | return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); | |
2093 | } | |
2094 | #endif | |
06dc9bd5 | 2095 | return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); |
b2e12bc6 CH |
2096 | } |
2097 | ||
ed6e2161 NA |
2098 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
2099 | AioContext *new_context) | |
2100 | { | |
c6447510 | 2101 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
ed6e2161 | 2102 | #ifdef CONFIG_LINUX_AIO |
ed6e2161 | 2103 | if (s->use_linux_aio) { |
cb09104e | 2104 | Error *local_err = NULL; |
ed6e2161 NA |
2105 | if (!aio_setup_linux_aio(new_context, &local_err)) { |
2106 | error_reportf_err(local_err, "Unable to use native AIO, " | |
2107 | "falling back to thread pool: "); | |
2108 | s->use_linux_aio = false; | |
2109 | } | |
2110 | } | |
2111 | #endif | |
c6447510 AM |
2112 | #ifdef CONFIG_LINUX_IO_URING |
2113 | if (s->use_linux_io_uring) { | |
cb8d0851 | 2114 | Error *local_err = NULL; |
c6447510 AM |
2115 | if (!aio_setup_linux_io_uring(new_context, &local_err)) { |
2116 | error_reportf_err(local_err, "Unable to use linux io_uring, " | |
2117 | "falling back to thread pool: "); | |
2118 | s->use_linux_io_uring = false; | |
2119 | } | |
2120 | } | |
2121 | #endif | |
ed6e2161 NA |
2122 | } |
2123 | ||
83f64091 FB |
2124 | static void raw_close(BlockDriverState *bs) |
2125 | { | |
2126 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 2127 | |
19cb3738 | 2128 | if (s->fd >= 0) { |
2e1e79da | 2129 | qemu_close(s->fd); |
19cb3738 FB |
2130 | s->fd = -1; |
2131 | } | |
83f64091 FB |
2132 | } |
2133 | ||
d0bc9e5d HR |
2134 | /** |
2135 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
2136 | * new space according to @prealloc. | |
2137 | * | |
2138 | * Returns: 0 on success, -errno on failure. | |
2139 | */ | |
93f4e2ff KW |
2140 | static int coroutine_fn |
2141 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
2142 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 2143 | { |
29cb4c01 | 2144 | RawPosixAIOData acb; |
d0bc9e5d | 2145 | |
29cb4c01 | 2146 | acb = (RawPosixAIOData) { |
93f4e2ff KW |
2147 | .bs = bs, |
2148 | .aio_fildes = fd, | |
2149 | .aio_type = QEMU_AIO_TRUNCATE, | |
2150 | .aio_offset = offset, | |
d57c44d0 KW |
2151 | .truncate = { |
2152 | .prealloc = prealloc, | |
2153 | .errp = errp, | |
2154 | }, | |
93f4e2ff | 2155 | }; |
d0bc9e5d | 2156 | |
29cb4c01 | 2157 | return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); |
9f63b07e HR |
2158 | } |
2159 | ||
061ca8a3 | 2160 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
c80d8b06 | 2161 | bool exact, PreallocMode prealloc, |
92b92799 | 2162 | BdrvRequestFlags flags, Error **errp) |
83f64091 FB |
2163 | { |
2164 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 2165 | struct stat st; |
f59adb32 | 2166 | int ret; |
55b949c8 CH |
2167 | |
2168 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
2169 | ret = -errno; |
2170 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
2171 | return ret; | |
55b949c8 CH |
2172 | } |
2173 | ||
2174 | if (S_ISREG(st.st_mode)) { | |
82325ae5 | 2175 | /* Always resizes to the exact @offset */ |
93f4e2ff | 2176 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
2177 | } |
2178 | ||
2179 | if (prealloc != PREALLOC_MODE_OFF) { | |
2180 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 2181 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
2182 | return -ENOTSUP; |
2183 | } | |
2184 | ||
2185 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
82325ae5 HR |
2186 | int64_t cur_length = raw_getlength(bs); |
2187 | ||
2188 | if (offset != cur_length && exact) { | |
2189 | error_setg(errp, "Cannot resize device files"); | |
2190 | return -ENOTSUP; | |
2191 | } else if (offset > cur_length) { | |
f59adb32 HR |
2192 | error_setg(errp, "Cannot grow device files"); |
2193 | return -EINVAL; | |
2194 | } | |
55b949c8 | 2195 | } else { |
f59adb32 | 2196 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
2197 | return -ENOTSUP; |
2198 | } | |
2199 | ||
83f64091 FB |
2200 | return 0; |
2201 | } | |
2202 | ||
128ab2ff BS |
2203 | #ifdef __OpenBSD__ |
2204 | static int64_t raw_getlength(BlockDriverState *bs) | |
2205 | { | |
2206 | BDRVRawState *s = bs->opaque; | |
2207 | int fd = s->fd; | |
2208 | struct stat st; | |
2209 | ||
2210 | if (fstat(fd, &st)) | |
aa729704 | 2211 | return -errno; |
128ab2ff BS |
2212 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2213 | struct disklabel dl; | |
2214 | ||
2215 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2216 | return -errno; |
128ab2ff BS |
2217 | return (uint64_t)dl.d_secsize * |
2218 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2219 | } else | |
2220 | return st.st_size; | |
2221 | } | |
d1f6fd8d CE |
2222 | #elif defined(__NetBSD__) |
2223 | static int64_t raw_getlength(BlockDriverState *bs) | |
2224 | { | |
2225 | BDRVRawState *s = bs->opaque; | |
2226 | int fd = s->fd; | |
2227 | struct stat st; | |
2228 | ||
2229 | if (fstat(fd, &st)) | |
aa729704 | 2230 | return -errno; |
d1f6fd8d CE |
2231 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2232 | struct dkwedge_info dkw; | |
2233 | ||
2234 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
2235 | return dkw.dkw_size * 512; | |
2236 | } else { | |
2237 | struct disklabel dl; | |
2238 | ||
2239 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2240 | return -errno; |
d1f6fd8d CE |
2241 | return (uint64_t)dl.d_secsize * |
2242 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2243 | } | |
2244 | } else | |
2245 | return st.st_size; | |
2246 | } | |
50779cc2 CH |
2247 | #elif defined(__sun__) |
2248 | static int64_t raw_getlength(BlockDriverState *bs) | |
2249 | { | |
2250 | BDRVRawState *s = bs->opaque; | |
2251 | struct dk_minfo minfo; | |
2252 | int ret; | |
aa729704 | 2253 | int64_t size; |
50779cc2 CH |
2254 | |
2255 | ret = fd_open(bs); | |
2256 | if (ret < 0) { | |
2257 | return ret; | |
2258 | } | |
2259 | ||
2260 | /* | |
2261 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
2262 | */ | |
2263 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
2264 | if (ret != -1) { | |
2265 | return minfo.dki_lbsize * minfo.dki_capacity; | |
2266 | } | |
2267 | ||
2268 | /* | |
2269 | * There are reports that lseek on some devices fails, but | |
2270 | * irc discussion said that contingency on contingency was overkill. | |
2271 | */ | |
aa729704 MA |
2272 | size = lseek(s->fd, 0, SEEK_END); |
2273 | if (size < 0) { | |
2274 | return -errno; | |
2275 | } | |
2276 | return size; | |
50779cc2 CH |
2277 | } |
2278 | #elif defined(CONFIG_BSD) | |
2279 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2280 | { |
2281 | BDRVRawState *s = bs->opaque; | |
2282 | int fd = s->fd; | |
2283 | int64_t size; | |
83f64091 | 2284 | struct stat sb; |
a167ba50 | 2285 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2286 | int reopened = 0; |
83f64091 | 2287 | #endif |
19cb3738 FB |
2288 | int ret; |
2289 | ||
2290 | ret = fd_open(bs); | |
2291 | if (ret < 0) | |
2292 | return ret; | |
83f64091 | 2293 | |
a167ba50 | 2294 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2295 | again: |
2296 | #endif | |
83f64091 FB |
2297 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
2298 | #ifdef DIOCGMEDIASIZE | |
7d37435b | 2299 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) |
c5e97233 BS |
2300 | #elif defined(DIOCGPART) |
2301 | { | |
2302 | struct partinfo pi; | |
2303 | if (ioctl(fd, DIOCGPART, &pi) == 0) | |
2304 | size = pi.media_size; | |
2305 | else | |
2306 | size = 0; | |
2307 | } | |
2308 | if (size == 0) | |
83f64091 | 2309 | #endif |
83affaa6 | 2310 | #if defined(__APPLE__) && defined(__MACH__) |
728dacbd JA |
2311 | { |
2312 | uint64_t sectors = 0; | |
2313 | uint32_t sector_size = 0; | |
2314 | ||
2315 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2316 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2317 | size = sectors * sector_size; | |
2318 | } else { | |
2319 | size = lseek(fd, 0LL, SEEK_END); | |
2320 | if (size < 0) { | |
2321 | return -errno; | |
2322 | } | |
2323 | } | |
2324 | } | |
83f64091 FB |
2325 | #else |
2326 | size = lseek(fd, 0LL, SEEK_END); | |
aa729704 MA |
2327 | if (size < 0) { |
2328 | return -errno; | |
2329 | } | |
9f23011a | 2330 | #endif |
a167ba50 | 2331 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2332 | switch(s->type) { |
2333 | case FTYPE_CD: | |
2334 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2335 | if (size == 2048LL * (unsigned)-1) | |
2336 | size = 0; | |
2337 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2338 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2339 | reopened = 1; |
2340 | goto again; | |
2341 | } | |
2342 | } | |
83f64091 | 2343 | #endif |
50779cc2 | 2344 | } else { |
83f64091 | 2345 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2346 | if (size < 0) { |
2347 | return -errno; | |
2348 | } | |
83f64091 | 2349 | } |
83f64091 FB |
2350 | return size; |
2351 | } | |
50779cc2 CH |
2352 | #else |
2353 | static int64_t raw_getlength(BlockDriverState *bs) | |
2354 | { | |
2355 | BDRVRawState *s = bs->opaque; | |
2356 | int ret; | |
aa729704 | 2357 | int64_t size; |
50779cc2 CH |
2358 | |
2359 | ret = fd_open(bs); | |
2360 | if (ret < 0) { | |
2361 | return ret; | |
2362 | } | |
2363 | ||
aa729704 MA |
2364 | size = lseek(s->fd, 0, SEEK_END); |
2365 | if (size < 0) { | |
2366 | return -errno; | |
2367 | } | |
2368 | return size; | |
50779cc2 | 2369 | } |
128ab2ff | 2370 | #endif |
83f64091 | 2371 | |
4a1d5e1f FZ |
2372 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2373 | { | |
2374 | struct stat st; | |
2375 | BDRVRawState *s = bs->opaque; | |
2376 | ||
2377 | if (fstat(s->fd, &st) < 0) { | |
2378 | return -errno; | |
2379 | } | |
2380 | return (int64_t)st.st_blocks * 512; | |
2381 | } | |
2382 | ||
93f4e2ff KW |
2383 | static int coroutine_fn |
2384 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2385 | { |
927f11e1 | 2386 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2387 | Error *local_err = NULL; |
83f64091 | 2388 | int fd; |
d815efca | 2389 | uint64_t perm, shared; |
1e37d059 | 2390 | int result = 0; |
83f64091 | 2391 | |
927f11e1 KW |
2392 | /* Validate options and set default values */ |
2393 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2394 | file_opts = &options->u.file; | |
464d9f64 | 2395 | |
927f11e1 KW |
2396 | if (!file_opts->has_nocow) { |
2397 | file_opts->nocow = false; | |
2398 | } | |
2399 | if (!file_opts->has_preallocation) { | |
2400 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2401 | } |
ffa244c8 KW |
2402 | if (!file_opts->has_extent_size_hint) { |
2403 | file_opts->extent_size_hint = 1 * MiB; | |
2404 | } | |
2405 | if (file_opts->extent_size_hint > UINT32_MAX) { | |
2406 | result = -EINVAL; | |
2407 | error_setg(errp, "Extent size hint is too large"); | |
2408 | goto out; | |
2409 | } | |
83f64091 | 2410 | |
927f11e1 | 2411 | /* Create file */ |
b18a24a9 | 2412 | fd = qemu_create(file_opts->filename, O_RDWR | O_BINARY, 0644, errp); |
1e37d059 SW |
2413 | if (fd < 0) { |
2414 | result = -errno; | |
06247428 HT |
2415 | goto out; |
2416 | } | |
2417 | ||
b8cf1913 HR |
2418 | /* Take permissions: We want to discard everything, so we need |
2419 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2420 | * BLK_PERM_RESIZE. | |
2421 | * On the other hand, we cannot share the RESIZE permission | |
2422 | * because we promise that after this function, the file has the | |
2423 | * size given in the options. If someone else were to resize it | |
2424 | * concurrently, we could not guarantee that. | |
2425 | * Note that after this function, we can no longer guarantee that | |
2426 | * the file is not touched by a third party, so it may be resized | |
2427 | * then. */ | |
2428 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2429 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2430 | ||
2431 | /* Step one: Take locks */ | |
2996ffad | 2432 | result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2433 | if (result < 0) { |
2434 | goto out_close; | |
2435 | } | |
2436 | ||
2437 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2438 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2439 | if (result < 0) { | |
b857431d FZ |
2440 | error_append_hint(errp, |
2441 | "Is another process using the image [%s]?\n", | |
2442 | file_opts->filename); | |
7c20c808 | 2443 | goto out_unlock; |
b8cf1913 HR |
2444 | } |
2445 | ||
2446 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2447 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2448 | if (result < 0) { |
7c20c808 | 2449 | goto out_unlock; |
b8cf1913 HR |
2450 | } |
2451 | ||
927f11e1 | 2452 | if (file_opts->nocow) { |
4ab15590 | 2453 | #ifdef __linux__ |
06247428 HT |
2454 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2455 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2456 | * will be ignored since any failure of this operation should not | |
2457 | * block the left work. | |
2458 | */ | |
2459 | int attr; | |
2460 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2461 | attr |= FS_NOCOW_FL; | |
2462 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2463 | } |
06247428 HT |
2464 | #endif |
2465 | } | |
ffa244c8 KW |
2466 | #ifdef FS_IOC_FSSETXATTR |
2467 | /* | |
2468 | * Try to set the extent size hint. Failure is not fatal, and a warning is | |
2469 | * only printed if the option was explicitly specified. | |
2470 | */ | |
2471 | { | |
2472 | struct fsxattr attr; | |
2473 | result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); | |
2474 | if (result == 0) { | |
2475 | attr.fsx_xflags |= FS_XFLAG_EXTSIZE; | |
2476 | attr.fsx_extsize = file_opts->extent_size_hint; | |
2477 | result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); | |
2478 | } | |
2479 | if (result < 0 && file_opts->has_extent_size_hint && | |
2480 | file_opts->extent_size_hint) | |
2481 | { | |
2482 | warn_report("Failed to set extent size hint: %s", | |
2483 | strerror(errno)); | |
2484 | } | |
2485 | } | |
2486 | #endif | |
06247428 | 2487 | |
b8cf1913 HR |
2488 | /* Resize and potentially preallocate the file to the desired |
2489 | * final size */ | |
93f4e2ff KW |
2490 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2491 | file_opts->preallocation, errp); | |
9f63b07e | 2492 | if (result < 0) { |
7c20c808 HR |
2493 | goto out_unlock; |
2494 | } | |
2495 | ||
2496 | out_unlock: | |
2996ffad | 2497 | raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); |
7c20c808 HR |
2498 | if (local_err) { |
2499 | /* The above call should not fail, and if it does, that does | |
2500 | * not mean the whole creation operation has failed. So | |
2501 | * report it the user for their convenience, but do not report | |
2502 | * it to the caller. */ | |
db0754df | 2503 | warn_report_err(local_err); |
1e37d059 | 2504 | } |
06247428 | 2505 | |
5a1dad9d | 2506 | out_close: |
06247428 HT |
2507 | if (qemu_close(fd) != 0 && result == 0) { |
2508 | result = -errno; | |
2509 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2510 | } | |
2511 | out: | |
1e37d059 | 2512 | return result; |
83f64091 FB |
2513 | } |
2514 | ||
b92902df ML |
2515 | static int coroutine_fn raw_co_create_opts(BlockDriver *drv, |
2516 | const char *filename, | |
2517 | QemuOpts *opts, | |
927f11e1 KW |
2518 | Error **errp) |
2519 | { | |
2520 | BlockdevCreateOptions options; | |
2521 | int64_t total_size = 0; | |
ffa244c8 KW |
2522 | int64_t extent_size_hint = 0; |
2523 | bool has_extent_size_hint = false; | |
927f11e1 KW |
2524 | bool nocow = false; |
2525 | PreallocMode prealloc; | |
2526 | char *buf = NULL; | |
2527 | Error *local_err = NULL; | |
2528 | ||
2529 | /* Skip file: protocol prefix */ | |
2530 | strstart(filename, "file:", &filename); | |
2531 | ||
2532 | /* Read out options */ | |
2533 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2534 | BDRV_SECTOR_SIZE); | |
ffa244c8 KW |
2535 | if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { |
2536 | has_extent_size_hint = true; | |
2537 | extent_size_hint = | |
2538 | qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); | |
2539 | } | |
927f11e1 KW |
2540 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); |
2541 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2542 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2543 | PREALLOC_MODE_OFF, &local_err); | |
2544 | g_free(buf); | |
2545 | if (local_err) { | |
2546 | error_propagate(errp, local_err); | |
2547 | return -EINVAL; | |
2548 | } | |
2549 | ||
2550 | options = (BlockdevCreateOptions) { | |
2551 | .driver = BLOCKDEV_DRIVER_FILE, | |
2552 | .u.file = { | |
2553 | .filename = (char *) filename, | |
2554 | .size = total_size, | |
2555 | .has_preallocation = true, | |
2556 | .preallocation = prealloc, | |
2557 | .has_nocow = true, | |
2558 | .nocow = nocow, | |
ffa244c8 KW |
2559 | .has_extent_size_hint = has_extent_size_hint, |
2560 | .extent_size_hint = extent_size_hint, | |
927f11e1 KW |
2561 | }, |
2562 | }; | |
2563 | return raw_co_create(&options, errp); | |
2564 | } | |
2565 | ||
9bffae14 DHB |
2566 | static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, |
2567 | Error **errp) | |
2568 | { | |
2569 | struct stat st; | |
2570 | int ret; | |
2571 | ||
2572 | if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { | |
2573 | error_setg_errno(errp, ENOENT, "%s is not a regular file", | |
2574 | bs->filename); | |
2575 | return -ENOENT; | |
2576 | } | |
2577 | ||
2578 | ret = unlink(bs->filename); | |
2579 | if (ret < 0) { | |
2580 | ret = -errno; | |
2581 | error_setg_errno(errp, -ret, "Error when deleting file %s", | |
2582 | bs->filename); | |
2583 | } | |
2584 | ||
2585 | return ret; | |
2586 | } | |
2587 | ||
d1f06fe6 MA |
2588 | /* |
2589 | * Find allocation range in @bs around offset @start. | |
2590 | * May change underlying file descriptor's file offset. | |
2591 | * If @start is not in a hole, store @start in @data, and the | |
2592 | * beginning of the next hole in @hole, and return 0. | |
2593 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2594 | * beginning of the next non-hole in @data, and return 0. | |
2595 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2596 | * If we can't find out, return a negative errno other than -ENXIO. | |
2597 | */ | |
2598 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2599 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2600 | { |
2601 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2602 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2603 | off_t offs; |
94282e71 | 2604 | |
d1f06fe6 MA |
2605 | /* |
2606 | * SEEK_DATA cases: | |
2607 | * D1. offs == start: start is in data | |
2608 | * D2. offs > start: start is in a hole, next data at offs | |
2609 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2610 | * or start is beyond EOF | |
2611 | * If the latter happens, the file has been truncated behind | |
2612 | * our back since we opened it. All bets are off then. | |
2613 | * Treating like a trailing hole is simplest. | |
2614 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2615 | */ | |
2616 | offs = lseek(s->fd, start, SEEK_DATA); | |
2617 | if (offs < 0) { | |
2618 | return -errno; /* D3 or D4 */ | |
2619 | } | |
a03083a0 JC |
2620 | |
2621 | if (offs < start) { | |
2622 | /* This is not a valid return by lseek(). We are safe to just return | |
2623 | * -EIO in this case, and we'll treat it like D4. */ | |
2624 | return -EIO; | |
2625 | } | |
d1f06fe6 MA |
2626 | |
2627 | if (offs > start) { | |
2628 | /* D2: in hole, next data at offs */ | |
2629 | *hole = start; | |
2630 | *data = offs; | |
2631 | return 0; | |
5500316d PB |
2632 | } |
2633 | ||
d1f06fe6 MA |
2634 | /* D1: in data, end not yet known */ |
2635 | ||
2636 | /* | |
2637 | * SEEK_HOLE cases: | |
2638 | * H1. offs == start: start is in a hole | |
2639 | * If this happens here, a hole has been dug behind our back | |
2640 | * since the previous lseek(). | |
2641 | * H2. offs > start: either start is in data, next hole at offs, | |
2642 | * or start is in trailing hole, EOF at offs | |
2643 | * Linux treats trailing holes like any other hole: offs == | |
2644 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2645 | * If that happens here, a hole has been dug behind our back | |
2646 | * since the previous lseek(). | |
2647 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2648 | * If this happens, the file has been truncated behind our | |
2649 | * back since we opened it. Treat it like a trailing hole. | |
2650 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2651 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2652 | */ | |
2653 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2654 | if (offs < 0) { | |
2655 | return -errno; /* D1 and (H3 or H4) */ | |
2656 | } | |
a03083a0 JC |
2657 | |
2658 | if (offs < start) { | |
2659 | /* This is not a valid return by lseek(). We are safe to just return | |
2660 | * -EIO in this case, and we'll treat it like H4. */ | |
2661 | return -EIO; | |
2662 | } | |
d1f06fe6 MA |
2663 | |
2664 | if (offs > start) { | |
2665 | /* | |
2666 | * D1 and H2: either in data, next hole at offs, or it was in | |
2667 | * data but is now in a trailing hole. In the latter case, | |
2668 | * all bets are off. Treating it as if it there was data all | |
2669 | * the way to EOF is safe, so simply do that. | |
2670 | */ | |
4f11aa8a | 2671 | *data = start; |
d1f06fe6 MA |
2672 | *hole = offs; |
2673 | return 0; | |
5500316d | 2674 | } |
4f11aa8a | 2675 | |
d1f06fe6 MA |
2676 | /* D1 and H1 */ |
2677 | return -EBUSY; | |
5500316d | 2678 | #else |
4f11aa8a | 2679 | return -ENOTSUP; |
5500316d | 2680 | #endif |
4f11aa8a HR |
2681 | } |
2682 | ||
2683 | /* | |
a290f085 | 2684 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2685 | * |
a290f085 | 2686 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2687 | * |
a290f085 EB |
2688 | * 'pnum' is set to the number of bytes (including and immediately following |
2689 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2690 | * allocated/unallocated state. |
2691 | * | |
a290f085 | 2692 | * 'bytes' is the max value 'pnum' should be set to. |
4f11aa8a | 2693 | */ |
a290f085 EB |
2694 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2695 | bool want_zero, | |
2696 | int64_t offset, | |
2697 | int64_t bytes, int64_t *pnum, | |
2698 | int64_t *map, | |
2699 | BlockDriverState **file) | |
2700 | { | |
2701 | off_t data = 0, hole = 0; | |
d7f62751 | 2702 | int ret; |
4f11aa8a | 2703 | |
9c3db310 HR |
2704 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); |
2705 | ||
4f11aa8a HR |
2706 | ret = fd_open(bs); |
2707 | if (ret < 0) { | |
2708 | return ret; | |
2709 | } | |
2710 | ||
a290f085 EB |
2711 | if (!want_zero) { |
2712 | *pnum = bytes; | |
2713 | *map = offset; | |
2714 | *file = bs; | |
2715 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2716 | } |
4f11aa8a | 2717 | |
a290f085 | 2718 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2719 | if (ret == -ENXIO) { |
2720 | /* Trailing hole */ | |
a290f085 | 2721 | *pnum = bytes; |
d1f06fe6 MA |
2722 | ret = BDRV_BLOCK_ZERO; |
2723 | } else if (ret < 0) { | |
2724 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2725 | *pnum = bytes; |
d1f06fe6 | 2726 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2727 | } else if (data == offset) { |
2728 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2729 | * possibly including a partial sector at EOF. */ |
a290f085 | 2730 | *pnum = MIN(bytes, hole - offset); |
9c3db310 HR |
2731 | |
2732 | /* | |
2733 | * We are not allowed to return partial sectors, though, so | |
2734 | * round up if necessary. | |
2735 | */ | |
2736 | if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { | |
2737 | int64_t file_length = raw_getlength(bs); | |
2738 | if (file_length > 0) { | |
2739 | /* Ignore errors, this is just a safeguard */ | |
2740 | assert(hole == file_length); | |
2741 | } | |
2742 | *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); | |
2743 | } | |
2744 | ||
d1f06fe6 | 2745 | ret = BDRV_BLOCK_DATA; |
5500316d | 2746 | } else { |
a290f085 EB |
2747 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2748 | assert(hole == offset); | |
2749 | *pnum = MIN(bytes, data - offset); | |
d1f06fe6 | 2750 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2751 | } |
a290f085 | 2752 | *map = offset; |
02650acb | 2753 | *file = bs; |
a290f085 | 2754 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2755 | } |
2756 | ||
31be8a2a SH |
2757 | #if defined(__linux__) |
2758 | /* Verify that the file is not in the page cache */ | |
2759 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2760 | { | |
2761 | const size_t window_size = 128 * 1024 * 1024; | |
2762 | BDRVRawState *s = bs->opaque; | |
2763 | void *window = NULL; | |
2764 | size_t length = 0; | |
2765 | unsigned char *vec; | |
2766 | size_t page_size; | |
2767 | off_t offset; | |
2768 | off_t end; | |
2769 | ||
2770 | /* mincore(2) page status information requires 1 byte per page */ | |
2771 | page_size = sysconf(_SC_PAGESIZE); | |
2772 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2773 | ||
2774 | end = raw_getlength(bs); | |
2775 | ||
2776 | for (offset = 0; offset < end; offset += window_size) { | |
2777 | void *new_window; | |
2778 | size_t new_length; | |
2779 | size_t vec_end; | |
2780 | size_t i; | |
2781 | int ret; | |
2782 | ||
2783 | /* Unmap previous window if size has changed */ | |
2784 | new_length = MIN(end - offset, window_size); | |
2785 | if (new_length != length) { | |
2786 | munmap(window, length); | |
2787 | window = NULL; | |
2788 | length = 0; | |
2789 | } | |
2790 | ||
2791 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2792 | s->fd, offset); | |
2793 | if (new_window == MAP_FAILED) { | |
2794 | error_setg_errno(errp, errno, "mmap failed"); | |
2795 | break; | |
2796 | } | |
2797 | ||
2798 | window = new_window; | |
2799 | length = new_length; | |
2800 | ||
2801 | ret = mincore(window, length, vec); | |
2802 | if (ret < 0) { | |
2803 | error_setg_errno(errp, errno, "mincore failed"); | |
2804 | break; | |
2805 | } | |
2806 | ||
2807 | vec_end = DIV_ROUND_UP(length, page_size); | |
2808 | for (i = 0; i < vec_end; i++) { | |
2809 | if (vec[i] & 0x1) { | |
31be8a2a SH |
2810 | break; |
2811 | } | |
2812 | } | |
77ed971b MA |
2813 | if (i < vec_end) { |
2814 | error_setg(errp, "page cache still in use!"); | |
2815 | break; | |
2816 | } | |
31be8a2a SH |
2817 | } |
2818 | ||
2819 | if (window) { | |
2820 | munmap(window, length); | |
2821 | } | |
2822 | ||
2823 | g_free(vec); | |
2824 | } | |
2825 | #endif /* __linux__ */ | |
2826 | ||
dd577a26 SH |
2827 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2828 | Error **errp) | |
2829 | { | |
2830 | BDRVRawState *s = bs->opaque; | |
2831 | int ret; | |
2832 | ||
2833 | ret = fd_open(bs); | |
2834 | if (ret < 0) { | |
2835 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2836 | return; | |
2837 | } | |
2838 | ||
f357fcd8 SH |
2839 | if (!s->drop_cache) { |
2840 | return; | |
2841 | } | |
2842 | ||
dd577a26 SH |
2843 | if (s->open_flags & O_DIRECT) { |
2844 | return; /* No host kernel page cache */ | |
2845 | } | |
2846 | ||
2847 | #if defined(__linux__) | |
2848 | /* This sets the scene for the next syscall... */ | |
2849 | ret = bdrv_co_flush(bs); | |
2850 | if (ret < 0) { | |
2851 | error_setg_errno(errp, -ret, "flush failed"); | |
2852 | return; | |
2853 | } | |
2854 | ||
2855 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2856 | * process. These limitations are okay because we just fsynced the file, | |
2857 | * we don't use mmap, and the file should not be in use by other processes. | |
2858 | */ | |
2859 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2860 | if (ret != 0) { /* the return value is a positive errno */ | |
2861 | error_setg_errno(errp, ret, "fadvise failed"); | |
2862 | return; | |
2863 | } | |
31be8a2a SH |
2864 | |
2865 | if (s->check_cache_dropped) { | |
2866 | check_cache_dropped(bs, errp); | |
2867 | } | |
dd577a26 SH |
2868 | #else /* __linux__ */ |
2869 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2870 | * unsupported on other host operating systems. Cache consistency issues | |
2871 | * may occur but no error is reported here, partly because that's the | |
2872 | * historical behavior and partly because it's hard to differentiate valid | |
2873 | * configurations that should not cause errors. | |
2874 | */ | |
2875 | #endif /* !__linux__ */ | |
2876 | } | |
2877 | ||
1c450366 AN |
2878 | static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) |
2879 | { | |
2880 | if (ret) { | |
2881 | s->stats.discard_nb_failed++; | |
2882 | } else { | |
2883 | s->stats.discard_nb_ok++; | |
2884 | s->stats.discard_bytes_ok += nbytes; | |
2885 | } | |
2886 | } | |
2887 | ||
33d70fb6 | 2888 | static coroutine_fn int |
46ee0f46 | 2889 | raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) |
dce512de | 2890 | { |
dce512de | 2891 | BDRVRawState *s = bs->opaque; |
46ee0f46 | 2892 | RawPosixAIOData acb; |
1c450366 | 2893 | int ret; |
46ee0f46 KW |
2894 | |
2895 | acb = (RawPosixAIOData) { | |
2896 | .bs = bs, | |
2897 | .aio_fildes = s->fd, | |
2898 | .aio_type = QEMU_AIO_DISCARD, | |
2899 | .aio_offset = offset, | |
2900 | .aio_nbytes = bytes, | |
2901 | }; | |
dce512de | 2902 | |
46ee0f46 KW |
2903 | if (blkdev) { |
2904 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2905 | } | |
2906 | ||
1c450366 AN |
2907 | ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); |
2908 | raw_account_discard(s, bytes, ret); | |
2909 | return ret; | |
46ee0f46 KW |
2910 | } |
2911 | ||
2912 | static coroutine_fn int | |
2913 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
2914 | { | |
2915 | return raw_do_pdiscard(bs, offset, bytes, false); | |
dce512de | 2916 | } |
0e7e1989 | 2917 | |
7154d8ae KW |
2918 | static int coroutine_fn |
2919 | raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, | |
2920 | BdrvRequestFlags flags, bool blkdev) | |
260a82e5 PB |
2921 | { |
2922 | BDRVRawState *s = bs->opaque; | |
7154d8ae KW |
2923 | RawPosixAIOData acb; |
2924 | ThreadPoolFunc *handler; | |
2925 | ||
292d06b9 HR |
2926 | #ifdef CONFIG_FALLOCATE |
2927 | if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { | |
2928 | BdrvTrackedRequest *req; | |
292d06b9 HR |
2929 | |
2930 | /* | |
2931 | * This is a workaround for a bug in the Linux XFS driver, | |
2932 | * where writes submitted through the AIO interface will be | |
2933 | * discarded if they happen beyond a concurrently running | |
2934 | * fallocate() that increases the file length (i.e., both the | |
2935 | * write and the fallocate() happen beyond the EOF). | |
2936 | * | |
2937 | * To work around it, we extend the tracked request for this | |
2938 | * zero write until INT64_MAX (effectively infinity), and mark | |
2939 | * it as serializing. | |
2940 | * | |
2941 | * We have to enable this workaround for all filesystems and | |
2942 | * AIO modes (not just XFS with aio=native), because for | |
2943 | * remote filesystems we do not know the host configuration. | |
2944 | */ | |
2945 | ||
2946 | req = bdrv_co_get_self_request(bs); | |
2947 | assert(req); | |
2948 | assert(req->type == BDRV_TRACKED_WRITE); | |
2949 | assert(req->offset <= offset); | |
2950 | assert(req->offset + req->bytes >= offset + bytes); | |
2951 | ||
8b117001 VSO |
2952 | req->bytes = BDRV_MAX_LENGTH - req->offset; |
2953 | ||
2954 | assert(bdrv_check_request(req->offset, req->bytes) == 0); | |
292d06b9 | 2955 | |
8ac5aab2 | 2956 | bdrv_make_request_serialising(req, bs->bl.request_alignment); |
292d06b9 HR |
2957 | } |
2958 | #endif | |
2959 | ||
7154d8ae KW |
2960 | acb = (RawPosixAIOData) { |
2961 | .bs = bs, | |
2962 | .aio_fildes = s->fd, | |
2963 | .aio_type = QEMU_AIO_WRITE_ZEROES, | |
2964 | .aio_offset = offset, | |
2965 | .aio_nbytes = bytes, | |
2966 | }; | |
2967 | ||
2968 | if (blkdev) { | |
2969 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2970 | } | |
738301e1 KW |
2971 | if (flags & BDRV_REQ_NO_FALLBACK) { |
2972 | acb.aio_type |= QEMU_AIO_NO_FALLBACK; | |
2973 | } | |
260a82e5 | 2974 | |
34fa110e | 2975 | if (flags & BDRV_REQ_MAY_UNMAP) { |
7154d8ae KW |
2976 | acb.aio_type |= QEMU_AIO_DISCARD; |
2977 | handler = handle_aiocb_write_zeroes_unmap; | |
2978 | } else { | |
2979 | handler = handle_aiocb_write_zeroes; | |
260a82e5 | 2980 | } |
34fa110e | 2981 | |
7154d8ae KW |
2982 | return raw_thread_pool_submit(bs, handler, &acb); |
2983 | } | |
2984 | ||
2985 | static int coroutine_fn raw_co_pwrite_zeroes( | |
2986 | BlockDriverState *bs, int64_t offset, | |
2987 | int bytes, BdrvRequestFlags flags) | |
2988 | { | |
2989 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); | |
260a82e5 PB |
2990 | } |
2991 | ||
2992 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
2993 | { | |
260a82e5 PB |
2994 | return 0; |
2995 | } | |
2996 | ||
d9245599 AN |
2997 | static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) |
2998 | { | |
2999 | BDRVRawState *s = bs->opaque; | |
3000 | return (BlockStatsSpecificFile) { | |
3001 | .discard_nb_ok = s->stats.discard_nb_ok, | |
3002 | .discard_nb_failed = s->stats.discard_nb_failed, | |
3003 | .discard_bytes_ok = s->stats.discard_bytes_ok, | |
3004 | }; | |
3005 | } | |
3006 | ||
3007 | static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) | |
3008 | { | |
3009 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3010 | ||
3011 | stats->driver = BLOCKDEV_DRIVER_FILE; | |
3012 | stats->u.file = get_blockstats_specific_file(bs); | |
3013 | ||
3014 | return stats; | |
3015 | } | |
3016 | ||
3017 | static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) | |
3018 | { | |
3019 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
3020 | ||
3021 | stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; | |
3022 | stats->u.host_device = get_blockstats_specific_file(bs); | |
3023 | ||
3024 | return stats; | |
3025 | } | |
3026 | ||
6f482f74 CL |
3027 | static QemuOptsList raw_create_opts = { |
3028 | .name = "raw-create-opts", | |
3029 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
3030 | .desc = { | |
3031 | { | |
3032 | .name = BLOCK_OPT_SIZE, | |
3033 | .type = QEMU_OPT_SIZE, | |
3034 | .help = "Virtual disk size" | |
3035 | }, | |
4ab15590 CL |
3036 | { |
3037 | .name = BLOCK_OPT_NOCOW, | |
3038 | .type = QEMU_OPT_BOOL, | |
3039 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
3040 | }, | |
06247428 HT |
3041 | { |
3042 | .name = BLOCK_OPT_PREALLOC, | |
3043 | .type = QEMU_OPT_STRING, | |
abea0053 SG |
3044 | .help = "Preallocation mode (allowed values: off" |
3045 | #ifdef CONFIG_POSIX_FALLOCATE | |
3046 | ", falloc" | |
3047 | #endif | |
3048 | ", full)" | |
06247428 | 3049 | }, |
ffa244c8 KW |
3050 | { |
3051 | .name = BLOCK_OPT_EXTENT_SIZE_HINT, | |
3052 | .type = QEMU_OPT_SIZE, | |
3053 | .help = "Extent size hint for the image file, 0 to disable" | |
3054 | }, | |
6f482f74 CL |
3055 | { /* end of list */ } |
3056 | } | |
0e7e1989 KW |
3057 | }; |
3058 | ||
244a5668 FZ |
3059 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
3060 | Error **errp) | |
3061 | { | |
6ceabe6f KW |
3062 | BDRVRawState *s = bs->opaque; |
3063 | BDRVRawReopenState *rs = NULL; | |
3064 | int open_flags; | |
3065 | int ret; | |
3066 | ||
3067 | if (s->perm_change_fd) { | |
3068 | /* | |
3069 | * In the context of reopen, this function may be called several times | |
3070 | * (directly and recursively while change permissions of the parent). | |
3071 | * This is even true for children that don't inherit from the original | |
3072 | * reopen node, so s->reopen_state is not set. | |
3073 | * | |
3074 | * Ignore all but the first call. | |
3075 | */ | |
3076 | return 0; | |
3077 | } | |
3078 | ||
3079 | if (s->reopen_state) { | |
3080 | /* We already have a new file descriptor to set permissions for */ | |
3081 | assert(s->reopen_state->perm == perm); | |
3082 | assert(s->reopen_state->shared_perm == shared); | |
3083 | rs = s->reopen_state->opaque; | |
3084 | s->perm_change_fd = rs->fd; | |
094e3639 | 3085 | s->perm_change_flags = rs->open_flags; |
6ceabe6f KW |
3086 | } else { |
3087 | /* We may need a new fd if auto-read-only switches the mode */ | |
23dece19 | 3088 | ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, |
6ceabe6f KW |
3089 | false, errp); |
3090 | if (ret < 0) { | |
3091 | return ret; | |
3092 | } else if (ret != s->fd) { | |
3093 | s->perm_change_fd = ret; | |
094e3639 | 3094 | s->perm_change_flags = open_flags; |
6ceabe6f KW |
3095 | } |
3096 | } | |
3097 | ||
3098 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
3099 | * but keep everything locked that new will need. */ | |
3100 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
3101 | if (ret < 0) { | |
3102 | goto fail; | |
3103 | } | |
3104 | ||
3105 | /* Copy locks to the new fd */ | |
eb43ea16 | 3106 | if (s->perm_change_fd && s->use_lock) { |
6ceabe6f KW |
3107 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, |
3108 | false, errp); | |
3109 | if (ret < 0) { | |
3110 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
3111 | goto fail; | |
3112 | } | |
3113 | } | |
3114 | return 0; | |
3115 | ||
3116 | fail: | |
3117 | if (s->perm_change_fd && !s->reopen_state) { | |
3118 | qemu_close(s->perm_change_fd); | |
3119 | } | |
3120 | s->perm_change_fd = 0; | |
3121 | return ret; | |
244a5668 FZ |
3122 | } |
3123 | ||
3124 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
3125 | { | |
3126 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
3127 | |
3128 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
3129 | * called after .bdrv_reopen_commit) */ | |
3130 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
3131 | qemu_close(s->fd); | |
3132 | s->fd = s->perm_change_fd; | |
094e3639 | 3133 | s->open_flags = s->perm_change_flags; |
6ceabe6f KW |
3134 | } |
3135 | s->perm_change_fd = 0; | |
3136 | ||
244a5668 FZ |
3137 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
3138 | s->perm = perm; | |
3139 | s->shared_perm = shared; | |
3140 | } | |
3141 | ||
3142 | static void raw_abort_perm_update(BlockDriverState *bs) | |
3143 | { | |
6ceabe6f KW |
3144 | BDRVRawState *s = bs->opaque; |
3145 | ||
3146 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
3147 | * the file descriptor. */ | |
3148 | if (s->perm_change_fd && !s->reopen_state) { | |
3149 | qemu_close(s->perm_change_fd); | |
3150 | } | |
3151 | s->perm_change_fd = 0; | |
3152 | ||
244a5668 FZ |
3153 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
3154 | } | |
3155 | ||
67b51fb9 VSO |
3156 | static int coroutine_fn raw_co_copy_range_from( |
3157 | BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, | |
3158 | BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, | |
3159 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) | |
1efad060 | 3160 | { |
67b51fb9 VSO |
3161 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3162 | read_flags, write_flags); | |
1efad060 FZ |
3163 | } |
3164 | ||
3165 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 VSO |
3166 | BdrvChild *src, |
3167 | uint64_t src_offset, | |
3168 | BdrvChild *dst, | |
3169 | uint64_t dst_offset, | |
3170 | uint64_t bytes, | |
3171 | BdrvRequestFlags read_flags, | |
3172 | BdrvRequestFlags write_flags) | |
1efad060 | 3173 | { |
58a209c4 | 3174 | RawPosixAIOData acb; |
1efad060 FZ |
3175 | BDRVRawState *s = bs->opaque; |
3176 | BDRVRawState *src_s; | |
3177 | ||
3178 | assert(dst->bs == bs); | |
3179 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
3180 | return -ENOTSUP; | |
3181 | } | |
3182 | ||
3183 | src_s = src->bs->opaque; | |
9f850f67 | 3184 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
3185 | return -EIO; |
3186 | } | |
58a209c4 KW |
3187 | |
3188 | acb = (RawPosixAIOData) { | |
3189 | .bs = bs, | |
3190 | .aio_type = QEMU_AIO_COPY_RANGE, | |
3191 | .aio_fildes = src_s->fd, | |
3192 | .aio_offset = src_offset, | |
3193 | .aio_nbytes = bytes, | |
3194 | .copy_range = { | |
3195 | .aio_fd2 = s->fd, | |
3196 | .aio_offset2 = dst_offset, | |
3197 | }, | |
3198 | }; | |
3199 | ||
3200 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
3201 | } |
3202 | ||
5f535a94 | 3203 | BlockDriver bdrv_file = { |
84a12e66 CH |
3204 | .format_name = "file", |
3205 | .protocol_name = "file", | |
856ae5c3 | 3206 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3207 | .bdrv_needs_filename = true, |
856ae5c3 | 3208 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 3209 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 3210 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
3211 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3212 | .bdrv_reopen_commit = raw_reopen_commit, | |
3213 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 3214 | .bdrv_close = raw_close, |
927f11e1 | 3215 | .bdrv_co_create = raw_co_create, |
efc75e2a | 3216 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 3217 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 3218 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 3219 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3220 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
9bffae14 | 3221 | .bdrv_co_delete_file = raw_co_delete_file, |
3b46e624 | 3222 | |
9d52aa3c KW |
3223 | .bdrv_co_preadv = raw_co_preadv, |
3224 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3225 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3226 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
3227 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3228 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3229 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3230 | .bdrv_io_plug = raw_aio_plug, |
3231 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 3232 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3233 | |
061ca8a3 | 3234 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 3235 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3236 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3237 | .bdrv_get_allocated_file_size |
3238 | = raw_get_allocated_file_size, | |
d9245599 | 3239 | .bdrv_get_specific_stats = raw_get_specific_stats, |
244a5668 FZ |
3240 | .bdrv_check_perm = raw_check_perm, |
3241 | .bdrv_set_perm = raw_set_perm, | |
3242 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 3243 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3244 | .mutable_opts = mutable_opts, |
83f64091 FB |
3245 | }; |
3246 | ||
19cb3738 FB |
3247 | /***********************************************/ |
3248 | /* host device */ | |
3249 | ||
83affaa6 | 3250 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
3251 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3252 | CFIndex maxPathSize, int flags); | |
d0855f12 | 3253 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 3254 | { |
d0855f12 | 3255 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
3256 | mach_port_t masterPort; |
3257 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
3258 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
3259 | char *mediaType = NULL; | |
19cb3738 FB |
3260 | |
3261 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
3262 | if ( KERN_SUCCESS != kernResult ) { | |
3263 | printf( "IOMasterPort returned %d\n", kernResult ); | |
3264 | } | |
3b46e624 | 3265 | |
d0855f12 JA |
3266 | int index; |
3267 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
3268 | classesToMatch = IOServiceMatching(matching_array[index]); | |
3269 | if (classesToMatch == NULL) { | |
3270 | error_report("IOServiceMatching returned NULL for %s", | |
3271 | matching_array[index]); | |
3272 | continue; | |
3273 | } | |
3274 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
3275 | kCFBooleanTrue); | |
3276 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
3277 | mediaIterator); | |
3278 | if (kernResult != KERN_SUCCESS) { | |
3279 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
3280 | kernResult); | |
3281 | continue; | |
3282 | } | |
3b46e624 | 3283 | |
d0855f12 JA |
3284 | /* If a match was found, leave the loop */ |
3285 | if (*mediaIterator != 0) { | |
4f7d28d7 | 3286 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
3287 | mediaType = g_strdup(matching_array[index]); |
3288 | break; | |
3289 | } | |
3290 | } | |
3291 | return mediaType; | |
19cb3738 FB |
3292 | } |
3293 | ||
98caa5bc JA |
3294 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3295 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
3296 | { |
3297 | io_object_t nextMedia; | |
3298 | kern_return_t kernResult = KERN_FAILURE; | |
3299 | *bsdPath = '\0'; | |
3300 | nextMedia = IOIteratorNext( mediaIterator ); | |
3301 | if ( nextMedia ) | |
3302 | { | |
3303 | CFTypeRef bsdPathAsCFString; | |
3304 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3305 | if ( bsdPathAsCFString ) { | |
3306 | size_t devPathLength; | |
3307 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3308 | if (flags & BDRV_O_NOCACHE) { |
3309 | strcat(bsdPath, "r"); | |
3310 | } | |
19cb3738 FB |
3311 | devPathLength = strlen( bsdPath ); |
3312 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3313 | kernResult = KERN_SUCCESS; | |
3314 | } | |
3315 | CFRelease( bsdPathAsCFString ); | |
3316 | } | |
3317 | IOObjectRelease( nextMedia ); | |
3318 | } | |
3b46e624 | 3319 | |
19cb3738 FB |
3320 | return kernResult; |
3321 | } | |
3322 | ||
d0855f12 JA |
3323 | /* Sets up a real cdrom for use in QEMU */ |
3324 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3325 | { | |
3326 | int index, num_of_test_partitions = 2, fd; | |
3327 | char test_partition[MAXPATHLEN]; | |
3328 | bool partition_found = false; | |
3329 | ||
3330 | /* look for a working partition */ | |
3331 | for (index = 0; index < num_of_test_partitions; index++) { | |
3332 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3333 | index); | |
b18a24a9 | 3334 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE, NULL); |
d0855f12 JA |
3335 | if (fd >= 0) { |
3336 | partition_found = true; | |
3337 | qemu_close(fd); | |
3338 | break; | |
3339 | } | |
3340 | } | |
3341 | ||
3342 | /* if a working partition on the device was not found */ | |
3343 | if (partition_found == false) { | |
3344 | error_setg(errp, "Failed to find a working partition on disc"); | |
3345 | } else { | |
4f7d28d7 | 3346 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3347 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3348 | } | |
3349 | return partition_found; | |
3350 | } | |
3351 | ||
3352 | /* Prints directions on mounting and unmounting a device */ | |
3353 | static void print_unmounting_directions(const char *file_name) | |
3354 | { | |
3355 | error_report("If device %s is mounted on the desktop, unmount" | |
3356 | " it first before using it in QEMU", file_name); | |
3357 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3358 | file_name); | |
3359 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3360 | } | |
3361 | ||
3362 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3363 | |
508c7cb3 CH |
3364 | static int hdev_probe_device(const char *filename) |
3365 | { | |
3366 | struct stat st; | |
3367 | ||
3368 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3369 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3370 | return 50; | |
3371 | ||
3372 | if (stat(filename, &st) >= 0 && | |
3373 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3374 | return 100; | |
3375 | } | |
3376 | ||
3377 | return 0; | |
3378 | } | |
3379 | ||
7af803d4 HR |
3380 | static void hdev_parse_filename(const char *filename, QDict *options, |
3381 | Error **errp) | |
3382 | { | |
03c320d8 | 3383 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3384 | } |
3385 | ||
3307ed7b DA |
3386 | static bool hdev_is_sg(BlockDriverState *bs) |
3387 | { | |
3388 | ||
3389 | #if defined(__linux__) | |
3390 | ||
0d4377b3 | 3391 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3392 | struct stat st; |
3393 | struct sg_scsi_id scsiid; | |
3394 | int sg_version; | |
0d4377b3 KW |
3395 | int ret; |
3396 | ||
3397 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3398 | return false; | |
3399 | } | |
3307ed7b | 3400 | |
0d4377b3 KW |
3401 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3402 | if (ret < 0) { | |
3403 | return false; | |
3404 | } | |
3405 | ||
3406 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3407 | if (ret >= 0) { | |
4f7d28d7 | 3408 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3409 | return true; |
3410 | } | |
3411 | ||
3412 | #endif | |
3413 | ||
3414 | return false; | |
3415 | } | |
3416 | ||
015a1036 HR |
3417 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3418 | Error **errp) | |
19cb3738 FB |
3419 | { |
3420 | BDRVRawState *s = bs->opaque; | |
da888d37 | 3421 | int ret; |
a76bab49 | 3422 | |
83affaa6 | 3423 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3424 | /* |
3425 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3426 | * would require more care. When @options come from -blockdev or | |
3427 | * blockdev_add, its members are typed according to the QAPI | |
3428 | * schema, but when they come from -drive, they're all QString. | |
3429 | */ | |
3307ed7b | 3430 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3431 | char bsd_path[MAXPATHLEN] = ""; |
3432 | bool error_occurred = false; | |
3433 | ||
3434 | /* If using a real cdrom */ | |
3435 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3436 | char *mediaType = NULL; | |
3437 | kern_return_t ret_val; | |
3438 | io_iterator_t mediaIterator = 0; | |
3439 | ||
3440 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3441 | if (mediaType == NULL) { | |
3442 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3443 | " drive"); | |
3444 | error_occurred = true; | |
3445 | goto hdev_open_Mac_error; | |
3446 | } | |
3307ed7b | 3447 | |
d0855f12 JA |
3448 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3449 | if (ret_val != KERN_SUCCESS) { | |
3450 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3451 | error_occurred = true; | |
3452 | goto hdev_open_Mac_error; | |
3453 | } | |
3454 | ||
3455 | /* If a real optical drive was not found */ | |
3456 | if (bsd_path[0] == '\0') { | |
3457 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3458 | error_occurred = true; | |
3459 | goto hdev_open_Mac_error; | |
3460 | } | |
3461 | ||
3462 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3463 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3464 | setup_cdrom(bsd_path, errp) == false) { | |
3465 | print_unmounting_directions(bsd_path); | |
3466 | error_occurred = true; | |
3467 | goto hdev_open_Mac_error; | |
19cb3738 | 3468 | } |
3b46e624 | 3469 | |
46f5ac20 | 3470 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3471 | |
3472 | hdev_open_Mac_error: | |
3473 | g_free(mediaType); | |
3474 | if (mediaIterator) { | |
3475 | IOObjectRelease(mediaIterator); | |
3476 | } | |
3477 | if (error_occurred) { | |
3478 | return -ENOENT; | |
3479 | } | |
19cb3738 | 3480 | } |
d0855f12 | 3481 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3482 | |
3483 | s->type = FTYPE_FILE; | |
90babde0 | 3484 | |
668f62ec | 3485 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
da888d37 | 3486 | if (ret < 0) { |
d0855f12 JA |
3487 | #if defined(__APPLE__) && defined(__MACH__) |
3488 | if (*bsd_path) { | |
3489 | filename = bsd_path; | |
3490 | } | |
3491 | /* if a physical device experienced an error while being opened */ | |
3492 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3493 | print_unmounting_directions(filename); | |
3494 | } | |
3495 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3496 | return ret; |
3497 | } | |
3498 | ||
3307ed7b DA |
3499 | /* Since this does ioctl the device must be already opened */ |
3500 | bs->sg = hdev_is_sg(bs); | |
3501 | ||
da888d37 | 3502 | return ret; |
19cb3738 FB |
3503 | } |
3504 | ||
03ff3ca3 | 3505 | #if defined(__linux__) |
2f3a7ab3 KW |
3506 | static int coroutine_fn |
3507 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3508 | { |
f141eafe | 3509 | BDRVRawState *s = bs->opaque; |
03425671 | 3510 | RawPosixAIOData acb; |
2f3a7ab3 | 3511 | int ret; |
221f715d | 3512 | |
2f3a7ab3 KW |
3513 | ret = fd_open(bs); |
3514 | if (ret < 0) { | |
3515 | return ret; | |
3516 | } | |
c208e8c2 | 3517 | |
7c9e5276 PB |
3518 | if (req == SG_IO && s->pr_mgr) { |
3519 | struct sg_io_hdr *io_hdr = buf; | |
3520 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3521 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3522 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3523 | s->fd, io_hdr); |
7c9e5276 PB |
3524 | } |
3525 | } | |
3526 | ||
03425671 KW |
3527 | acb = (RawPosixAIOData) { |
3528 | .bs = bs, | |
3529 | .aio_type = QEMU_AIO_IOCTL, | |
3530 | .aio_fildes = s->fd, | |
3531 | .aio_offset = 0, | |
3532 | .ioctl = { | |
3533 | .buf = buf, | |
3534 | .cmd = req, | |
3535 | }, | |
3536 | }; | |
3537 | ||
3538 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3539 | } |
f709623b | 3540 | #endif /* linux */ |
221f715d | 3541 | |
9f23011a BS |
3542 | static int fd_open(BlockDriverState *bs) |
3543 | { | |
3544 | BDRVRawState *s = bs->opaque; | |
3545 | ||
3546 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
3547 | if (s->fd >= 0) | |
3548 | return 0; | |
3549 | return -EIO; | |
3550 | } | |
04eeb8b6 | 3551 | |
33d70fb6 KW |
3552 | static coroutine_fn int |
3553 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
c36dd8a0 | 3554 | { |
1c450366 | 3555 | BDRVRawState *s = bs->opaque; |
33d70fb6 | 3556 | int ret; |
c36dd8a0 | 3557 | |
33d70fb6 KW |
3558 | ret = fd_open(bs); |
3559 | if (ret < 0) { | |
1c450366 | 3560 | raw_account_discard(s, bytes, ret); |
33d70fb6 | 3561 | return ret; |
c36dd8a0 | 3562 | } |
46ee0f46 | 3563 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3564 | } |
3565 | ||
2ffa76c2 | 3566 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3567 | int64_t offset, int bytes, BdrvRequestFlags flags) |
d0b4503e | 3568 | { |
d0b4503e PB |
3569 | int rc; |
3570 | ||
3571 | rc = fd_open(bs); | |
3572 | if (rc < 0) { | |
3573 | return rc; | |
3574 | } | |
34fa110e | 3575 | |
7154d8ae | 3576 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3577 | } |
3578 | ||
5efa9d5a | 3579 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3580 | .format_name = "host_device", |
84a12e66 | 3581 | .protocol_name = "host_device", |
0b4ce02e | 3582 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3583 | .bdrv_needs_filename = true, |
0b4ce02e | 3584 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3585 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3586 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3587 | .bdrv_close = raw_close, |
1bc6b705 JC |
3588 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3589 | .bdrv_reopen_commit = raw_reopen_commit, | |
3590 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3591 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3592 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3593 | .mutable_opts = mutable_opts, |
dd577a26 | 3594 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3595 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3596 | |
9d52aa3c KW |
3597 | .bdrv_co_preadv = raw_co_preadv, |
3598 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3599 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3600 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3601 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3602 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3603 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3604 | .bdrv_io_plug = raw_aio_plug, |
3605 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3606 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3607 | |
061ca8a3 | 3608 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3609 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3610 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3611 | .bdrv_get_allocated_file_size |
3612 | = raw_get_allocated_file_size, | |
d9245599 | 3613 | .bdrv_get_specific_stats = hdev_get_specific_stats, |
244a5668 FZ |
3614 | .bdrv_check_perm = raw_check_perm, |
3615 | .bdrv_set_perm = raw_set_perm, | |
3616 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3617 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3618 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3619 | |
f3a5d3f8 | 3620 | /* generic scsi device */ |
63ec93db | 3621 | #ifdef __linux__ |
2f3a7ab3 | 3622 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3623 | #endif |
f3a5d3f8 CH |
3624 | }; |
3625 | ||
18fa1c42 HR |
3626 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3627 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3628 | Error **errp) | |
3629 | { | |
03c320d8 | 3630 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3631 | } |
3632 | #endif | |
3633 | ||
3634 | #ifdef __linux__ | |
015a1036 HR |
3635 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3636 | Error **errp) | |
f3a5d3f8 CH |
3637 | { |
3638 | BDRVRawState *s = bs->opaque; | |
3639 | ||
f3a5d3f8 CH |
3640 | s->type = FTYPE_CD; |
3641 | ||
19a3da7f | 3642 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3643 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3644 | } |
3645 | ||
508c7cb3 CH |
3646 | static int cdrom_probe_device(const char *filename) |
3647 | { | |
3baf720e CR |
3648 | int fd, ret; |
3649 | int prio = 0; | |
343f8568 | 3650 | struct stat st; |
3baf720e | 3651 | |
b18a24a9 | 3652 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK, NULL); |
3baf720e CR |
3653 | if (fd < 0) { |
3654 | goto out; | |
3655 | } | |
343f8568 JS |
3656 | ret = fstat(fd, &st); |
3657 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3658 | goto outc; | |
3659 | } | |
3baf720e CR |
3660 | |
3661 | /* Attempt to detect via a CDROM specific ioctl */ | |
3662 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3663 | if (ret >= 0) | |
3664 | prio = 100; | |
3665 | ||
343f8568 | 3666 | outc: |
2e1e79da | 3667 | qemu_close(fd); |
3baf720e CR |
3668 | out: |
3669 | return prio; | |
508c7cb3 CH |
3670 | } |
3671 | ||
e031f750 | 3672 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3673 | { |
3674 | BDRVRawState *s = bs->opaque; | |
3675 | int ret; | |
3676 | ||
3677 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3678 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3679 | } |
3680 | ||
f36f3949 | 3681 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3682 | { |
3683 | BDRVRawState *s = bs->opaque; | |
3684 | ||
3685 | if (eject_flag) { | |
3686 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3687 | perror("CDROMEJECT"); | |
3688 | } else { | |
3689 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3690 | perror("CDROMEJECT"); | |
3691 | } | |
f3a5d3f8 CH |
3692 | } |
3693 | ||
025e849a | 3694 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3695 | { |
3696 | BDRVRawState *s = bs->opaque; | |
3697 | ||
3698 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3699 | /* | |
3700 | * Note: an error can happen if the distribution automatically | |
3701 | * mounts the CD-ROM | |
3702 | */ | |
3703 | /* perror("CDROM_LOCKDOOR"); */ | |
3704 | } | |
f3a5d3f8 CH |
3705 | } |
3706 | ||
3707 | static BlockDriver bdrv_host_cdrom = { | |
3708 | .format_name = "host_cdrom", | |
84a12e66 | 3709 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3710 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3711 | .bdrv_needs_filename = true, |
508c7cb3 | 3712 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3713 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3714 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3715 | .bdrv_close = raw_close, |
1bc6b705 JC |
3716 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3717 | .bdrv_reopen_commit = raw_reopen_commit, | |
3718 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3719 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3720 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3721 | .mutable_opts = mutable_opts, |
dd577a26 | 3722 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3723 | |
9d52aa3c KW |
3724 | .bdrv_co_preadv = raw_co_preadv, |
3725 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3726 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3727 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3728 | .bdrv_io_plug = raw_aio_plug, |
3729 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3730 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3731 | |
061ca8a3 | 3732 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3733 | .bdrv_getlength = raw_getlength, |
3734 | .has_variable_length = true, | |
4a1d5e1f FZ |
3735 | .bdrv_get_allocated_file_size |
3736 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3737 | |
3738 | /* removable device support */ | |
3739 | .bdrv_is_inserted = cdrom_is_inserted, | |
3740 | .bdrv_eject = cdrom_eject, | |
025e849a | 3741 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3742 | |
3743 | /* generic scsi device */ | |
2f3a7ab3 | 3744 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3745 | }; |
3746 | #endif /* __linux__ */ | |
3747 | ||
a167ba50 | 3748 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3749 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3750 | Error **errp) | |
f3a5d3f8 CH |
3751 | { |
3752 | BDRVRawState *s = bs->opaque; | |
3753 | int ret; | |
3754 | ||
3755 | s->type = FTYPE_CD; | |
3756 | ||
668f62ec | 3757 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
e428e439 | 3758 | if (ret) { |
f3a5d3f8 | 3759 | return ret; |
e428e439 | 3760 | } |
f3a5d3f8 | 3761 | |
9b2260cb | 3762 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3763 | ioctl(s->fd, CDIOCALLOW); |
3764 | return 0; | |
3765 | } | |
3766 | ||
508c7cb3 CH |
3767 | static int cdrom_probe_device(const char *filename) |
3768 | { | |
3769 | if (strstart(filename, "/dev/cd", NULL) || | |
3770 | strstart(filename, "/dev/acd", NULL)) | |
3771 | return 100; | |
3772 | return 0; | |
3773 | } | |
3774 | ||
f3a5d3f8 CH |
3775 | static int cdrom_reopen(BlockDriverState *bs) |
3776 | { | |
3777 | BDRVRawState *s = bs->opaque; | |
3778 | int fd; | |
3779 | ||
3780 | /* | |
3781 | * Force reread of possibly changed/newly loaded disc, | |
3782 | * FreeBSD seems to not notice sometimes... | |
3783 | */ | |
3784 | if (s->fd >= 0) | |
2e1e79da | 3785 | qemu_close(s->fd); |
b18a24a9 | 3786 | fd = qemu_open(bs->filename, s->open_flags, NULL); |
f3a5d3f8 CH |
3787 | if (fd < 0) { |
3788 | s->fd = -1; | |
3789 | return -EIO; | |
3790 | } | |
3791 | s->fd = fd; | |
3792 | ||
9b2260cb | 3793 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3794 | ioctl(s->fd, CDIOCALLOW); |
3795 | return 0; | |
3796 | } | |
3797 | ||
e031f750 | 3798 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3799 | { |
3800 | return raw_getlength(bs) > 0; | |
3801 | } | |
3802 | ||
f36f3949 | 3803 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3804 | { |
3805 | BDRVRawState *s = bs->opaque; | |
3806 | ||
3807 | if (s->fd < 0) | |
822e1cd1 | 3808 | return; |
f3a5d3f8 CH |
3809 | |
3810 | (void) ioctl(s->fd, CDIOCALLOW); | |
3811 | ||
3812 | if (eject_flag) { | |
3813 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3814 | perror("CDIOCEJECT"); | |
3815 | } else { | |
3816 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3817 | perror("CDIOCCLOSE"); | |
3818 | } | |
3819 | ||
822e1cd1 | 3820 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3821 | } |
3822 | ||
025e849a | 3823 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3824 | { |
3825 | BDRVRawState *s = bs->opaque; | |
3826 | ||
3827 | if (s->fd < 0) | |
7bf37fed | 3828 | return; |
f3a5d3f8 CH |
3829 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3830 | /* | |
3831 | * Note: an error can happen if the distribution automatically | |
3832 | * mounts the CD-ROM | |
3833 | */ | |
3834 | /* perror("CDROM_LOCKDOOR"); */ | |
3835 | } | |
f3a5d3f8 CH |
3836 | } |
3837 | ||
3838 | static BlockDriver bdrv_host_cdrom = { | |
3839 | .format_name = "host_cdrom", | |
84a12e66 | 3840 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3841 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3842 | .bdrv_needs_filename = true, |
508c7cb3 | 3843 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3844 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3845 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3846 | .bdrv_close = raw_close, |
1bc6b705 JC |
3847 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3848 | .bdrv_reopen_commit = raw_reopen_commit, | |
3849 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3850 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3851 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3852 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3853 | |
9d52aa3c KW |
3854 | .bdrv_co_preadv = raw_co_preadv, |
3855 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3856 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3857 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3858 | .bdrv_io_plug = raw_aio_plug, |
3859 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3860 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3861 | |
061ca8a3 | 3862 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3863 | .bdrv_getlength = raw_getlength, |
3864 | .has_variable_length = true, | |
4a1d5e1f FZ |
3865 | .bdrv_get_allocated_file_size |
3866 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3867 | |
19cb3738 | 3868 | /* removable device support */ |
f3a5d3f8 CH |
3869 | .bdrv_is_inserted = cdrom_is_inserted, |
3870 | .bdrv_eject = cdrom_eject, | |
025e849a | 3871 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3872 | }; |
f3a5d3f8 | 3873 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3874 | |
84a12e66 | 3875 | static void bdrv_file_init(void) |
5efa9d5a | 3876 | { |
508c7cb3 CH |
3877 | /* |
3878 | * Register all the drivers. Note that order is important, the driver | |
3879 | * registered last will get probed first. | |
3880 | */ | |
84a12e66 | 3881 | bdrv_register(&bdrv_file); |
5efa9d5a | 3882 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3883 | #ifdef __linux__ |
f3a5d3f8 CH |
3884 | bdrv_register(&bdrv_host_cdrom); |
3885 | #endif | |
a167ba50 | 3886 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3887 | bdrv_register(&bdrv_host_cdrom); |
3888 | #endif | |
5efa9d5a AL |
3889 | } |
3890 | ||
84a12e66 | 3891 | block_init(bdrv_file_init); |