]>
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 | ||
23dece19 | 428 | static void raw_parse_flags(int bdrv_flags, int *open_flags, bool has_writers) |
6a8dc042 | 429 | { |
23dece19 | 430 | bool read_write = false; |
6a8dc042 JC |
431 | assert(open_flags != NULL); |
432 | ||
433 | *open_flags |= O_BINARY; | |
434 | *open_flags &= ~O_ACCMODE; | |
23dece19 KW |
435 | |
436 | if (bdrv_flags & BDRV_O_AUTO_RDONLY) { | |
437 | read_write = has_writers; | |
438 | } else if (bdrv_flags & BDRV_O_RDWR) { | |
439 | read_write = true; | |
440 | } | |
441 | ||
442 | if (read_write) { | |
6a8dc042 JC |
443 | *open_flags |= O_RDWR; |
444 | } else { | |
445 | *open_flags |= O_RDONLY; | |
446 | } | |
447 | ||
448 | /* Use O_DSYNC for write-through caching, no flags for write-back caching, | |
449 | * and O_DIRECT for no caching. */ | |
450 | if ((bdrv_flags & BDRV_O_NOCACHE)) { | |
451 | *open_flags |= O_DIRECT; | |
452 | } | |
6a8dc042 JC |
453 | } |
454 | ||
078896a9 HR |
455 | static void raw_parse_filename(const char *filename, QDict *options, |
456 | Error **errp) | |
457 | { | |
03c320d8 | 458 | bdrv_parse_filename_strip_prefix(filename, "file:", options); |
078896a9 HR |
459 | } |
460 | ||
c66a6157 KW |
461 | static QemuOptsList raw_runtime_opts = { |
462 | .name = "raw", | |
463 | .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head), | |
464 | .desc = { | |
465 | { | |
466 | .name = "filename", | |
467 | .type = QEMU_OPT_STRING, | |
468 | .help = "File name of the image", | |
469 | }, | |
0a4279d9 KW |
470 | { |
471 | .name = "aio", | |
472 | .type = QEMU_OPT_STRING, | |
c6447510 | 473 | .help = "host AIO implementation (threads, native, io_uring)", |
0a4279d9 | 474 | }, |
16b48d5d FZ |
475 | { |
476 | .name = "locking", | |
477 | .type = QEMU_OPT_STRING, | |
478 | .help = "file locking mode (on/off/auto, default: auto)", | |
479 | }, | |
7c9e5276 PB |
480 | { |
481 | .name = "pr-manager", | |
482 | .type = QEMU_OPT_STRING, | |
483 | .help = "id of persistent reservation manager object (default: none)", | |
484 | }, | |
f357fcd8 SH |
485 | #if defined(__linux__) |
486 | { | |
487 | .name = "drop-cache", | |
488 | .type = QEMU_OPT_BOOL, | |
489 | .help = "invalidate page cache during live migration (default: on)", | |
490 | }, | |
491 | #endif | |
31be8a2a SH |
492 | { |
493 | .name = "x-check-cache-dropped", | |
494 | .type = QEMU_OPT_BOOL, | |
495 | .help = "check that page cache was dropped on live migration (default: off)" | |
496 | }, | |
c66a6157 KW |
497 | { /* end of list */ } |
498 | }, | |
499 | }; | |
500 | ||
8a2ce0bc AG |
501 | static const char *const mutable_opts[] = { "x-check-cache-dropped", NULL }; |
502 | ||
c66a6157 | 503 | static int raw_open_common(BlockDriverState *bs, QDict *options, |
230ff739 JS |
504 | int bdrv_flags, int open_flags, |
505 | bool device, Error **errp) | |
83f64091 FB |
506 | { |
507 | BDRVRawState *s = bs->opaque; | |
c66a6157 KW |
508 | QemuOpts *opts; |
509 | Error *local_err = NULL; | |
8bfea15d | 510 | const char *filename = NULL; |
7c9e5276 | 511 | const char *str; |
0a4279d9 | 512 | BlockdevAioOptions aio, aio_default; |
0e1d8f4c | 513 | int fd, ret; |
260a82e5 | 514 | struct stat st; |
244a5668 | 515 | OnOffAuto locking; |
83f64091 | 516 | |
87ea75d5 | 517 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); |
af175e85 | 518 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { |
c66a6157 KW |
519 | ret = -EINVAL; |
520 | goto fail; | |
521 | } | |
522 | ||
523 | filename = qemu_opt_get(opts, "filename"); | |
524 | ||
db0754df | 525 | ret = raw_normalize_devicepath(&filename, errp); |
1de1ae0a | 526 | if (ret != 0) { |
c66a6157 | 527 | goto fail; |
1de1ae0a CE |
528 | } |
529 | ||
c6447510 AM |
530 | if (bdrv_flags & BDRV_O_NATIVE_AIO) { |
531 | aio_default = BLOCKDEV_AIO_OPTIONS_NATIVE; | |
532 | #ifdef CONFIG_LINUX_IO_URING | |
533 | } else if (bdrv_flags & BDRV_O_IO_URING) { | |
534 | aio_default = BLOCKDEV_AIO_OPTIONS_IO_URING; | |
535 | #endif | |
536 | } else { | |
537 | aio_default = BLOCKDEV_AIO_OPTIONS_THREADS; | |
538 | } | |
539 | ||
f7abe0ec MAL |
540 | aio = qapi_enum_parse(&BlockdevAioOptions_lookup, |
541 | qemu_opt_get(opts, "aio"), | |
06c60b6c | 542 | aio_default, &local_err); |
0a4279d9 KW |
543 | if (local_err) { |
544 | error_propagate(errp, local_err); | |
545 | ret = -EINVAL; | |
546 | goto fail; | |
547 | } | |
c6447510 | 548 | |
0a4279d9 | 549 | s->use_linux_aio = (aio == BLOCKDEV_AIO_OPTIONS_NATIVE); |
c6447510 AM |
550 | #ifdef CONFIG_LINUX_IO_URING |
551 | s->use_linux_io_uring = (aio == BLOCKDEV_AIO_OPTIONS_IO_URING); | |
552 | #endif | |
0a4279d9 | 553 | |
f7abe0ec MAL |
554 | locking = qapi_enum_parse(&OnOffAuto_lookup, |
555 | qemu_opt_get(opts, "locking"), | |
06c60b6c | 556 | ON_OFF_AUTO_AUTO, &local_err); |
244a5668 FZ |
557 | if (local_err) { |
558 | error_propagate(errp, local_err); | |
559 | ret = -EINVAL; | |
560 | goto fail; | |
561 | } | |
562 | switch (locking) { | |
563 | case ON_OFF_AUTO_ON: | |
564 | s->use_lock = true; | |
2b218f5d | 565 | if (!qemu_has_ofd_lock()) { |
db0754df FZ |
566 | warn_report("File lock requested but OFD locking syscall is " |
567 | "unavailable, falling back to POSIX file locks"); | |
568 | error_printf("Due to the implementation, locks can be lost " | |
569 | "unexpectedly.\n"); | |
2b218f5d | 570 | } |
244a5668 FZ |
571 | break; |
572 | case ON_OFF_AUTO_OFF: | |
573 | s->use_lock = false; | |
574 | break; | |
575 | case ON_OFF_AUTO_AUTO: | |
2b218f5d | 576 | s->use_lock = qemu_has_ofd_lock(); |
244a5668 FZ |
577 | break; |
578 | default: | |
579 | abort(); | |
580 | } | |
581 | ||
7c9e5276 PB |
582 | str = qemu_opt_get(opts, "pr-manager"); |
583 | if (str) { | |
584 | s->pr_mgr = pr_manager_lookup(str, &local_err); | |
585 | if (local_err) { | |
586 | error_propagate(errp, local_err); | |
587 | ret = -EINVAL; | |
588 | goto fail; | |
589 | } | |
590 | } | |
591 | ||
f357fcd8 | 592 | s->drop_cache = qemu_opt_get_bool(opts, "drop-cache", true); |
31be8a2a SH |
593 | s->check_cache_dropped = qemu_opt_get_bool(opts, "x-check-cache-dropped", |
594 | false); | |
595 | ||
6a8dc042 | 596 | s->open_flags = open_flags; |
23dece19 | 597 | raw_parse_flags(bdrv_flags, &s->open_flags, false); |
83f64091 | 598 | |
90babde0 | 599 | s->fd = -1; |
40ff6d7e | 600 | fd = qemu_open(filename, s->open_flags, 0644); |
64107dc0 KW |
601 | ret = fd < 0 ? -errno : 0; |
602 | ||
64107dc0 | 603 | if (ret < 0) { |
f6fc1e30 | 604 | error_setg_file_open(errp, -ret, filename); |
c66a6157 | 605 | if (ret == -EROFS) { |
19cb3738 | 606 | ret = -EACCES; |
c66a6157 KW |
607 | } |
608 | goto fail; | |
19cb3738 | 609 | } |
83f64091 | 610 | s->fd = fd; |
9ef91a67 | 611 | |
244a5668 FZ |
612 | s->perm = 0; |
613 | s->shared_perm = BLK_PERM_ALL; | |
614 | ||
5c6c3a6c | 615 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 616 | /* Currently Linux does AIO only for files opened with O_DIRECT */ |
ed6e2161 NA |
617 | if (s->use_linux_aio) { |
618 | if (!(s->open_flags & O_DIRECT)) { | |
619 | error_setg(errp, "aio=native was specified, but it requires " | |
620 | "cache.direct=on, which was not specified."); | |
621 | ret = -EINVAL; | |
622 | goto fail; | |
623 | } | |
624 | if (!aio_setup_linux_aio(bdrv_get_aio_context(bs), errp)) { | |
625 | error_prepend(errp, "Unable to use native AIO: "); | |
626 | goto fail; | |
627 | } | |
96518254 | 628 | } |
1501ecc1 | 629 | #else |
0a4279d9 | 630 | if (s->use_linux_aio) { |
d657c0c2 KW |
631 | error_setg(errp, "aio=native was specified, but is not supported " |
632 | "in this build."); | |
633 | ret = -EINVAL; | |
634 | goto fail; | |
1501ecc1 SH |
635 | } |
636 | #endif /* !defined(CONFIG_LINUX_AIO) */ | |
9ef91a67 | 637 | |
c6447510 AM |
638 | #ifdef CONFIG_LINUX_IO_URING |
639 | if (s->use_linux_io_uring) { | |
640 | if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs), errp)) { | |
641 | error_prepend(errp, "Unable to use io_uring: "); | |
642 | goto fail; | |
643 | } | |
644 | } | |
645 | #else | |
646 | if (s->use_linux_io_uring) { | |
647 | error_setg(errp, "aio=io_uring was specified, but is not supported " | |
648 | "in this build."); | |
649 | ret = -EINVAL; | |
650 | goto fail; | |
651 | } | |
652 | #endif /* !defined(CONFIG_LINUX_IO_URING) */ | |
653 | ||
7ce21016 | 654 | s->has_discard = true; |
97a2ae34 | 655 | s->has_write_zeroes = true; |
5edc8557 | 656 | if ((bs->open_flags & BDRV_O_NOCACHE) != 0 && !dio_byte_aligned(s->fd)) { |
3cad8307 RPM |
657 | s->needs_alignment = true; |
658 | } | |
260a82e5 PB |
659 | |
660 | if (fstat(s->fd, &st) < 0) { | |
01212d4e | 661 | ret = -errno; |
260a82e5 PB |
662 | error_setg_errno(errp, errno, "Could not stat file"); |
663 | goto fail; | |
664 | } | |
230ff739 JS |
665 | |
666 | if (!device) { | |
667 | if (S_ISBLK(st.st_mode)) { | |
668 | warn_report("Opening a block device as a file using the '%s' " | |
669 | "driver is deprecated", bs->drv->format_name); | |
670 | } else if (S_ISCHR(st.st_mode)) { | |
671 | warn_report("Opening a character device as a file using the '%s' " | |
672 | "driver is deprecated", bs->drv->format_name); | |
673 | } else if (!S_ISREG(st.st_mode)) { | |
674 | error_setg(errp, "A regular file was expected by the '%s' driver, " | |
675 | "but something else was given", bs->drv->format_name); | |
676 | ret = -EINVAL; | |
677 | goto fail; | |
678 | } else { | |
679 | s->discard_zeroes = true; | |
680 | s->has_fallocate = true; | |
681 | } | |
682 | } else { | |
683 | if (!(S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
684 | error_setg(errp, "'%s' driver expects either " | |
685 | "a character or block device", bs->drv->format_name); | |
686 | ret = -EINVAL; | |
687 | goto fail; | |
688 | } | |
260a82e5 | 689 | } |
230ff739 | 690 | |
d0b4503e PB |
691 | if (S_ISBLK(st.st_mode)) { |
692 | #ifdef BLKDISCARDZEROES | |
693 | unsigned int arg; | |
694 | if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) { | |
695 | s->discard_zeroes = true; | |
696 | } | |
697 | #endif | |
698 | #ifdef __linux__ | |
699 | /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do | |
700 | * not rely on the contents of discarded blocks unless using O_DIRECT. | |
97a2ae34 | 701 | * Same for BLKZEROOUT. |
d0b4503e PB |
702 | */ |
703 | if (!(bs->open_flags & BDRV_O_NOCACHE)) { | |
704 | s->discard_zeroes = false; | |
97a2ae34 | 705 | s->has_write_zeroes = false; |
d0b4503e PB |
706 | } |
707 | #endif | |
708 | } | |
3cad8307 RPM |
709 | #ifdef __FreeBSD__ |
710 | if (S_ISCHR(st.st_mode)) { | |
711 | /* | |
712 | * The file is a char device (disk), which on FreeBSD isn't behind | |
713 | * a pager, so force all requests to be aligned. This is needed | |
714 | * so QEMU makes sure all IO operations on the device are aligned | |
715 | * to sector size, or else FreeBSD will reject them with EINVAL. | |
716 | */ | |
717 | s->needs_alignment = true; | |
718 | } | |
719 | #endif | |
260a82e5 | 720 | |
dce512de CH |
721 | #ifdef CONFIG_XFS |
722 | if (platform_test_xfs_fd(s->fd)) { | |
7ce21016 | 723 | s->is_xfs = true; |
dce512de CH |
724 | } |
725 | #endif | |
726 | ||
738301e1 | 727 | bs->supported_zero_flags = BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK; |
2f0c6e7a KW |
728 | if (S_ISREG(st.st_mode)) { |
729 | /* When extending regular files, we get zeros from the OS */ | |
730 | bs->supported_truncate_flags = BDRV_REQ_ZERO_WRITE; | |
731 | } | |
c66a6157 KW |
732 | ret = 0; |
733 | fail: | |
8bfea15d KW |
734 | if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) { |
735 | unlink(filename); | |
736 | } | |
c66a6157 KW |
737 | qemu_opts_del(opts); |
738 | return ret; | |
83f64091 FB |
739 | } |
740 | ||
015a1036 HR |
741 | static int raw_open(BlockDriverState *bs, QDict *options, int flags, |
742 | Error **errp) | |
90babde0 CH |
743 | { |
744 | BDRVRawState *s = bs->opaque; | |
745 | ||
746 | s->type = FTYPE_FILE; | |
230ff739 | 747 | return raw_open_common(bs, options, flags, 0, false, errp); |
90babde0 CH |
748 | } |
749 | ||
244a5668 FZ |
750 | typedef enum { |
751 | RAW_PL_PREPARE, | |
752 | RAW_PL_COMMIT, | |
753 | RAW_PL_ABORT, | |
754 | } RawPermLockOp; | |
755 | ||
756 | #define PERM_FOREACH(i) \ | |
757 | for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++) | |
758 | ||
759 | /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the | |
760 | * file; if @unlock == true, also unlock the unneeded bytes. | |
761 | * @shared_perm_lock_bits is the mask of all permissions that are NOT shared. | |
762 | */ | |
2996ffad | 763 | static int raw_apply_lock_bytes(BDRVRawState *s, int fd, |
244a5668 FZ |
764 | uint64_t perm_lock_bits, |
765 | uint64_t shared_perm_lock_bits, | |
766 | bool unlock, Error **errp) | |
767 | { | |
768 | int ret; | |
769 | int i; | |
2996ffad FZ |
770 | uint64_t locked_perm, locked_shared_perm; |
771 | ||
772 | if (s) { | |
773 | locked_perm = s->locked_perm; | |
774 | locked_shared_perm = s->locked_shared_perm; | |
775 | } else { | |
776 | /* | |
777 | * We don't have the previous bits, just lock/unlock for each of the | |
778 | * requested bits. | |
779 | */ | |
780 | if (unlock) { | |
781 | locked_perm = BLK_PERM_ALL; | |
782 | locked_shared_perm = BLK_PERM_ALL; | |
783 | } else { | |
784 | locked_perm = 0; | |
785 | locked_shared_perm = 0; | |
786 | } | |
787 | } | |
244a5668 FZ |
788 | |
789 | PERM_FOREACH(i) { | |
790 | int off = RAW_LOCK_PERM_BASE + i; | |
2996ffad FZ |
791 | uint64_t bit = (1ULL << i); |
792 | if ((perm_lock_bits & bit) && !(locked_perm & bit)) { | |
d0a96155 | 793 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
794 | if (ret) { |
795 | error_setg(errp, "Failed to lock byte %d", off); | |
796 | return ret; | |
2996ffad FZ |
797 | } else if (s) { |
798 | s->locked_perm |= bit; | |
244a5668 | 799 | } |
2996ffad | 800 | } else if (unlock && (locked_perm & bit) && !(perm_lock_bits & bit)) { |
d0a96155 | 801 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
802 | if (ret) { |
803 | error_setg(errp, "Failed to unlock byte %d", off); | |
804 | return ret; | |
2996ffad FZ |
805 | } else if (s) { |
806 | s->locked_perm &= ~bit; | |
244a5668 FZ |
807 | } |
808 | } | |
809 | } | |
810 | PERM_FOREACH(i) { | |
811 | int off = RAW_LOCK_SHARED_BASE + i; | |
2996ffad FZ |
812 | uint64_t bit = (1ULL << i); |
813 | if ((shared_perm_lock_bits & bit) && !(locked_shared_perm & bit)) { | |
d0a96155 | 814 | ret = qemu_lock_fd(fd, off, 1, false); |
244a5668 FZ |
815 | if (ret) { |
816 | error_setg(errp, "Failed to lock byte %d", off); | |
817 | return ret; | |
2996ffad FZ |
818 | } else if (s) { |
819 | s->locked_shared_perm |= bit; | |
244a5668 | 820 | } |
2996ffad FZ |
821 | } else if (unlock && (locked_shared_perm & bit) && |
822 | !(shared_perm_lock_bits & bit)) { | |
d0a96155 | 823 | ret = qemu_unlock_fd(fd, off, 1); |
244a5668 FZ |
824 | if (ret) { |
825 | error_setg(errp, "Failed to unlock byte %d", off); | |
826 | return ret; | |
2996ffad FZ |
827 | } else if (s) { |
828 | s->locked_shared_perm &= ~bit; | |
244a5668 FZ |
829 | } |
830 | } | |
831 | } | |
832 | return 0; | |
833 | } | |
834 | ||
835 | /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */ | |
d0a96155 | 836 | static int raw_check_lock_bytes(int fd, uint64_t perm, uint64_t shared_perm, |
244a5668 FZ |
837 | Error **errp) |
838 | { | |
839 | int ret; | |
840 | int i; | |
841 | ||
842 | PERM_FOREACH(i) { | |
843 | int off = RAW_LOCK_SHARED_BASE + i; | |
844 | uint64_t p = 1ULL << i; | |
845 | if (perm & p) { | |
d0a96155 | 846 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
847 | if (ret) { |
848 | char *perm_name = bdrv_perm_names(p); | |
849 | error_setg(errp, | |
850 | "Failed to get \"%s\" lock", | |
851 | perm_name); | |
852 | g_free(perm_name); | |
244a5668 FZ |
853 | return ret; |
854 | } | |
855 | } | |
856 | } | |
857 | PERM_FOREACH(i) { | |
858 | int off = RAW_LOCK_PERM_BASE + i; | |
859 | uint64_t p = 1ULL << i; | |
860 | if (!(shared_perm & p)) { | |
d0a96155 | 861 | ret = qemu_lock_fd_test(fd, off, 1, true); |
244a5668 FZ |
862 | if (ret) { |
863 | char *perm_name = bdrv_perm_names(p); | |
864 | error_setg(errp, | |
865 | "Failed to get shared \"%s\" lock", | |
866 | perm_name); | |
867 | g_free(perm_name); | |
244a5668 FZ |
868 | return ret; |
869 | } | |
870 | } | |
871 | } | |
872 | return 0; | |
873 | } | |
874 | ||
875 | static int raw_handle_perm_lock(BlockDriverState *bs, | |
876 | RawPermLockOp op, | |
877 | uint64_t new_perm, uint64_t new_shared, | |
878 | Error **errp) | |
879 | { | |
880 | BDRVRawState *s = bs->opaque; | |
881 | int ret = 0; | |
882 | Error *local_err = NULL; | |
883 | ||
884 | if (!s->use_lock) { | |
885 | return 0; | |
886 | } | |
887 | ||
888 | if (bdrv_get_flags(bs) & BDRV_O_INACTIVE) { | |
889 | return 0; | |
890 | } | |
891 | ||
244a5668 FZ |
892 | switch (op) { |
893 | case RAW_PL_PREPARE: | |
696aaaed VSO |
894 | if ((s->perm | new_perm) == s->perm && |
895 | (s->shared_perm & new_shared) == s->shared_perm) | |
896 | { | |
897 | /* | |
898 | * We are going to unlock bytes, it should not fail. If it fail due | |
899 | * to some fs-dependent permission-unrelated reasons (which occurs | |
900 | * sometimes on NFS and leads to abort in bdrv_replace_child) we | |
901 | * can't prevent such errors by any check here. And we ignore them | |
902 | * anyway in ABORT and COMMIT. | |
903 | */ | |
904 | return 0; | |
905 | } | |
f2e3af29 | 906 | ret = raw_apply_lock_bytes(s, s->fd, s->perm | new_perm, |
244a5668 FZ |
907 | ~s->shared_perm | ~new_shared, |
908 | false, errp); | |
909 | if (!ret) { | |
f2e3af29 | 910 | ret = raw_check_lock_bytes(s->fd, new_perm, new_shared, errp); |
244a5668 FZ |
911 | if (!ret) { |
912 | return 0; | |
913 | } | |
b857431d FZ |
914 | error_append_hint(errp, |
915 | "Is another process using the image [%s]?\n", | |
916 | bs->filename); | |
244a5668 | 917 | } |
244a5668 FZ |
918 | /* fall through to unlock bytes. */ |
919 | case RAW_PL_ABORT: | |
f2e3af29 | 920 | raw_apply_lock_bytes(s, s->fd, s->perm, ~s->shared_perm, |
d0a96155 | 921 | true, &local_err); |
244a5668 FZ |
922 | if (local_err) { |
923 | /* Theoretically the above call only unlocks bytes and it cannot | |
924 | * fail. Something weird happened, report it. | |
925 | */ | |
db0754df | 926 | warn_report_err(local_err); |
244a5668 FZ |
927 | } |
928 | break; | |
929 | case RAW_PL_COMMIT: | |
f2e3af29 | 930 | raw_apply_lock_bytes(s, s->fd, new_perm, ~new_shared, |
d0a96155 | 931 | true, &local_err); |
244a5668 FZ |
932 | if (local_err) { |
933 | /* Theoretically the above call only unlocks bytes and it cannot | |
934 | * fail. Something weird happened, report it. | |
935 | */ | |
db0754df | 936 | warn_report_err(local_err); |
244a5668 FZ |
937 | } |
938 | break; | |
939 | } | |
940 | return ret; | |
941 | } | |
942 | ||
5cec2870 | 943 | static int raw_reconfigure_getfd(BlockDriverState *bs, int flags, |
23dece19 | 944 | int *open_flags, uint64_t perm, bool force_dup, |
6ceabe6f | 945 | Error **errp) |
5cec2870 KW |
946 | { |
947 | BDRVRawState *s = bs->opaque; | |
948 | int fd = -1; | |
949 | int ret; | |
23dece19 KW |
950 | bool has_writers = perm & |
951 | (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED | BLK_PERM_RESIZE); | |
5cec2870 KW |
952 | int fcntl_flags = O_APPEND | O_NONBLOCK; |
953 | #ifdef O_NOATIME | |
954 | fcntl_flags |= O_NOATIME; | |
955 | #endif | |
956 | ||
957 | *open_flags = 0; | |
958 | if (s->type == FTYPE_CD) { | |
959 | *open_flags |= O_NONBLOCK; | |
960 | } | |
961 | ||
23dece19 | 962 | raw_parse_flags(flags, open_flags, has_writers); |
5cec2870 KW |
963 | |
964 | #ifdef O_ASYNC | |
965 | /* Not all operating systems have O_ASYNC, and those that don't | |
966 | * will not let us track the state into rs->open_flags (typically | |
967 | * you achieve the same effect with an ioctl, for example I_SETSIG | |
968 | * on Solaris). But we do not use O_ASYNC, so that's fine. | |
969 | */ | |
970 | assert((s->open_flags & O_ASYNC) == 0); | |
971 | #endif | |
972 | ||
6ceabe6f KW |
973 | if (!force_dup && *open_flags == s->open_flags) { |
974 | /* We're lucky, the existing fd is fine */ | |
975 | return s->fd; | |
976 | } | |
977 | ||
5cec2870 KW |
978 | if ((*open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) { |
979 | /* dup the original fd */ | |
980 | fd = qemu_dup(s->fd); | |
981 | if (fd >= 0) { | |
982 | ret = fcntl_setfl(fd, *open_flags); | |
983 | if (ret) { | |
984 | qemu_close(fd); | |
985 | fd = -1; | |
986 | } | |
987 | } | |
988 | } | |
989 | ||
990 | /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */ | |
991 | if (fd == -1) { | |
992 | const char *normalized_filename = bs->filename; | |
993 | ret = raw_normalize_devicepath(&normalized_filename, errp); | |
994 | if (ret >= 0) { | |
995 | assert(!(*open_flags & O_CREAT)); | |
996 | fd = qemu_open(normalized_filename, *open_flags); | |
997 | if (fd == -1) { | |
998 | error_setg_errno(errp, errno, "Could not reopen file"); | |
999 | return -1; | |
1000 | } | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | return fd; | |
1005 | } | |
1006 | ||
eeb6b45d JC |
1007 | static int raw_reopen_prepare(BDRVReopenState *state, |
1008 | BlockReopenQueue *queue, Error **errp) | |
1009 | { | |
1010 | BDRVRawState *s; | |
4e6d13c9 | 1011 | BDRVRawReopenState *rs; |
31be8a2a | 1012 | QemuOpts *opts; |
a6aeca0c | 1013 | int ret; |
df26a350 | 1014 | Error *local_err = NULL; |
eeb6b45d JC |
1015 | |
1016 | assert(state != NULL); | |
1017 | assert(state->bs != NULL); | |
1018 | ||
1019 | s = state->bs->opaque; | |
1020 | ||
5839e53b | 1021 | state->opaque = g_new0(BDRVRawReopenState, 1); |
4e6d13c9 | 1022 | rs = state->opaque; |
31be8a2a SH |
1023 | |
1024 | /* Handle options changes */ | |
1025 | opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort); | |
af175e85 | 1026 | if (!qemu_opts_absorb_qdict(opts, state->options, errp)) { |
31be8a2a SH |
1027 | ret = -EINVAL; |
1028 | goto out; | |
1029 | } | |
1030 | ||
f357fcd8 | 1031 | rs->drop_cache = qemu_opt_get_bool_del(opts, "drop-cache", true); |
8d324575 AG |
1032 | rs->check_cache_dropped = |
1033 | qemu_opt_get_bool_del(opts, "x-check-cache-dropped", false); | |
1034 | ||
1035 | /* This driver's reopen function doesn't currently allow changing | |
1036 | * other options, so let's put them back in the original QDict and | |
1037 | * bdrv_reopen_prepare() will detect changes and complain. */ | |
1038 | qemu_opts_to_qdict(opts, state->options); | |
eeb6b45d | 1039 | |
5cec2870 | 1040 | rs->fd = raw_reconfigure_getfd(state->bs, state->flags, &rs->open_flags, |
23dece19 | 1041 | state->perm, true, &local_err); |
5cec2870 KW |
1042 | if (local_err) { |
1043 | error_propagate(errp, local_err); | |
1044 | ret = -1; | |
1045 | goto out; | |
eeb6b45d | 1046 | } |
df26a350 KW |
1047 | |
1048 | /* Fail already reopen_prepare() if we can't get a working O_DIRECT | |
1049 | * alignment with the new fd. */ | |
4e6d13c9 FZ |
1050 | if (rs->fd != -1) { |
1051 | raw_probe_alignment(state->bs, rs->fd, &local_err); | |
df26a350 | 1052 | if (local_err) { |
df26a350 KW |
1053 | error_propagate(errp, local_err); |
1054 | ret = -EINVAL; | |
a6aeca0c KW |
1055 | goto out_fd; |
1056 | } | |
df26a350 KW |
1057 | } |
1058 | ||
e0c9cf3a | 1059 | s->reopen_state = state; |
a6aeca0c KW |
1060 | ret = 0; |
1061 | out_fd: | |
1062 | if (ret < 0) { | |
1063 | qemu_close(rs->fd); | |
1064 | rs->fd = -1; | |
1065 | } | |
31be8a2a SH |
1066 | out: |
1067 | qemu_opts_del(opts); | |
eeb6b45d JC |
1068 | return ret; |
1069 | } | |
1070 | ||
eeb6b45d JC |
1071 | static void raw_reopen_commit(BDRVReopenState *state) |
1072 | { | |
4e6d13c9 | 1073 | BDRVRawReopenState *rs = state->opaque; |
eeb6b45d JC |
1074 | BDRVRawState *s = state->bs->opaque; |
1075 | ||
f357fcd8 | 1076 | s->drop_cache = rs->drop_cache; |
31be8a2a | 1077 | s->check_cache_dropped = rs->check_cache_dropped; |
4e6d13c9 | 1078 | s->open_flags = rs->open_flags; |
eeb6b45d JC |
1079 | |
1080 | qemu_close(s->fd); | |
4e6d13c9 | 1081 | s->fd = rs->fd; |
eeb6b45d JC |
1082 | |
1083 | g_free(state->opaque); | |
1084 | state->opaque = NULL; | |
e0c9cf3a KW |
1085 | |
1086 | assert(s->reopen_state == state); | |
1087 | s->reopen_state = NULL; | |
eeb6b45d JC |
1088 | } |
1089 | ||
1090 | ||
1091 | static void raw_reopen_abort(BDRVReopenState *state) | |
1092 | { | |
4e6d13c9 | 1093 | BDRVRawReopenState *rs = state->opaque; |
e0c9cf3a | 1094 | BDRVRawState *s = state->bs->opaque; |
eeb6b45d JC |
1095 | |
1096 | /* nothing to do if NULL, we didn't get far enough */ | |
4e6d13c9 | 1097 | if (rs == NULL) { |
eeb6b45d JC |
1098 | return; |
1099 | } | |
1100 | ||
4e6d13c9 FZ |
1101 | if (rs->fd >= 0) { |
1102 | qemu_close(rs->fd); | |
1103 | rs->fd = -1; | |
eeb6b45d JC |
1104 | } |
1105 | g_free(state->opaque); | |
1106 | state->opaque = NULL; | |
e0c9cf3a KW |
1107 | |
1108 | assert(s->reopen_state == state); | |
1109 | s->reopen_state = NULL; | |
eeb6b45d JC |
1110 | } |
1111 | ||
867eccfe | 1112 | static int sg_get_max_transfer_length(int fd) |
6f607174 FZ |
1113 | { |
1114 | #ifdef BLKSECTGET | |
48265250 | 1115 | int max_bytes = 0; |
867eccfe ML |
1116 | |
1117 | if (ioctl(fd, BLKSECTGET, &max_bytes) == 0) { | |
48265250 | 1118 | return max_bytes; |
6f607174 FZ |
1119 | } else { |
1120 | return -errno; | |
1121 | } | |
1122 | #else | |
1123 | return -ENOSYS; | |
1124 | #endif | |
1125 | } | |
1126 | ||
867eccfe | 1127 | static int sg_get_max_segments(int fd) |
9103f1ce FZ |
1128 | { |
1129 | #ifdef CONFIG_LINUX | |
1130 | char buf[32]; | |
1131 | const char *end; | |
867eccfe | 1132 | char *sysfspath = NULL; |
9103f1ce | 1133 | int ret; |
867eccfe | 1134 | int sysfd = -1; |
9103f1ce | 1135 | long max_segments; |
867eccfe ML |
1136 | struct stat st; |
1137 | ||
1138 | if (fstat(fd, &st)) { | |
1139 | ret = -errno; | |
1140 | goto out; | |
1141 | } | |
9103f1ce FZ |
1142 | |
1143 | sysfspath = g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments", | |
867eccfe ML |
1144 | major(st.st_rdev), minor(st.st_rdev)); |
1145 | sysfd = open(sysfspath, O_RDONLY); | |
1146 | if (sysfd == -1) { | |
9103f1ce FZ |
1147 | ret = -errno; |
1148 | goto out; | |
1149 | } | |
1150 | do { | |
867eccfe | 1151 | ret = read(sysfd, buf, sizeof(buf) - 1); |
9103f1ce FZ |
1152 | } while (ret == -1 && errno == EINTR); |
1153 | if (ret < 0) { | |
1154 | ret = -errno; | |
1155 | goto out; | |
1156 | } else if (ret == 0) { | |
1157 | ret = -EIO; | |
1158 | goto out; | |
1159 | } | |
1160 | buf[ret] = 0; | |
1161 | /* The file is ended with '\n', pass 'end' to accept that. */ | |
1162 | ret = qemu_strtol(buf, &end, 10, &max_segments); | |
1163 | if (ret == 0 && end && *end == '\n') { | |
1164 | ret = max_segments; | |
1165 | } | |
1166 | ||
1167 | out: | |
867eccfe ML |
1168 | if (sysfd != -1) { |
1169 | close(sysfd); | |
fed414df | 1170 | } |
9103f1ce FZ |
1171 | g_free(sysfspath); |
1172 | return ret; | |
1173 | #else | |
1174 | return -ENOTSUP; | |
1175 | #endif | |
1176 | } | |
1177 | ||
3baca891 | 1178 | static void raw_refresh_limits(BlockDriverState *bs, Error **errp) |
c25f53b0 PB |
1179 | { |
1180 | BDRVRawState *s = bs->opaque; | |
6f607174 | 1181 | |
867eccfe ML |
1182 | if (bs->sg) { |
1183 | int ret = sg_get_max_transfer_length(s->fd); | |
1184 | ||
1185 | if (ret > 0 && ret <= BDRV_REQUEST_MAX_BYTES) { | |
1186 | bs->bl.max_transfer = pow2floor(ret); | |
1187 | } | |
1188 | ||
1189 | ret = sg_get_max_segments(s->fd); | |
1190 | if (ret > 0) { | |
038adc2f WY |
1191 | bs->bl.max_transfer = MIN(bs->bl.max_transfer, |
1192 | ret * qemu_real_host_page_size); | |
6f607174 FZ |
1193 | } |
1194 | } | |
eeb6b45d | 1195 | |
df26a350 | 1196 | raw_probe_alignment(bs, s->fd, errp); |
4196d2f0 | 1197 | bs->bl.min_mem_alignment = s->buf_align; |
038adc2f | 1198 | bs->bl.opt_mem_alignment = MAX(s->buf_align, qemu_real_host_page_size); |
c25f53b0 | 1199 | } |
83f64091 | 1200 | |
1a9335e4 ET |
1201 | static int check_for_dasd(int fd) |
1202 | { | |
1203 | #ifdef BIODASDINFO2 | |
1204 | struct dasd_information2_t info = {0}; | |
1205 | ||
1206 | return ioctl(fd, BIODASDINFO2, &info); | |
1207 | #else | |
1208 | return -1; | |
1209 | #endif | |
1210 | } | |
1211 | ||
1212 | /** | |
1213 | * Try to get @bs's logical and physical block size. | |
1214 | * On success, store them in @bsz and return zero. | |
1215 | * On failure, return negative errno. | |
1216 | */ | |
1217 | static int hdev_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz) | |
1218 | { | |
1219 | BDRVRawState *s = bs->opaque; | |
1220 | int ret; | |
1221 | ||
1222 | /* If DASD, get blocksizes */ | |
1223 | if (check_for_dasd(s->fd) < 0) { | |
1224 | return -ENOTSUP; | |
1225 | } | |
1226 | ret = probe_logical_blocksize(s->fd, &bsz->log); | |
1227 | if (ret < 0) { | |
1228 | return ret; | |
1229 | } | |
1230 | return probe_physical_blocksize(s->fd, &bsz->phys); | |
1231 | } | |
1232 | ||
1233 | /** | |
1234 | * Try to get @bs's geometry: cyls, heads, sectors. | |
1235 | * On success, store them in @geo and return 0. | |
1236 | * On failure return -errno. | |
1237 | * (Allows block driver to assign default geometry values that guest sees) | |
1238 | */ | |
1239 | #ifdef __linux__ | |
1240 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1241 | { | |
1242 | BDRVRawState *s = bs->opaque; | |
1243 | struct hd_geometry ioctl_geo = {0}; | |
1a9335e4 ET |
1244 | |
1245 | /* If DASD, get its geometry */ | |
1246 | if (check_for_dasd(s->fd) < 0) { | |
1247 | return -ENOTSUP; | |
1248 | } | |
1249 | if (ioctl(s->fd, HDIO_GETGEO, &ioctl_geo) < 0) { | |
1250 | return -errno; | |
1251 | } | |
1252 | /* HDIO_GETGEO may return success even though geo contains zeros | |
1253 | (e.g. certain multipath setups) */ | |
1254 | if (!ioctl_geo.heads || !ioctl_geo.sectors || !ioctl_geo.cylinders) { | |
1255 | return -ENOTSUP; | |
1256 | } | |
1257 | /* Do not return a geometry for partition */ | |
1258 | if (ioctl_geo.start != 0) { | |
1259 | return -ENOTSUP; | |
1260 | } | |
1261 | geo->heads = ioctl_geo.heads; | |
1262 | geo->sectors = ioctl_geo.sectors; | |
1a9335e4 ET |
1263 | geo->cylinders = ioctl_geo.cylinders; |
1264 | ||
1265 | return 0; | |
1266 | } | |
1267 | #else /* __linux__ */ | |
1268 | static int hdev_probe_geometry(BlockDriverState *bs, HDGeometry *geo) | |
1269 | { | |
1270 | return -ENOTSUP; | |
1271 | } | |
1272 | #endif | |
1273 | ||
03425671 KW |
1274 | #if defined(__linux__) |
1275 | static int handle_aiocb_ioctl(void *opaque) | |
de81a169 | 1276 | { |
03425671 | 1277 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1278 | int ret; |
1279 | ||
d57c44d0 | 1280 | ret = ioctl(aiocb->aio_fildes, aiocb->ioctl.cmd, aiocb->ioctl.buf); |
de81a169 PB |
1281 | if (ret == -1) { |
1282 | return -errno; | |
1283 | } | |
1284 | ||
b608c8dc | 1285 | return 0; |
de81a169 | 1286 | } |
03425671 | 1287 | #endif /* linux */ |
de81a169 | 1288 | |
06dc9bd5 | 1289 | static int handle_aiocb_flush(void *opaque) |
de81a169 | 1290 | { |
06dc9bd5 | 1291 | RawPosixAIOData *aiocb = opaque; |
e5bcf967 | 1292 | BDRVRawState *s = aiocb->bs->opaque; |
de81a169 PB |
1293 | int ret; |
1294 | ||
e5bcf967 KW |
1295 | if (s->page_cache_inconsistent) { |
1296 | return -EIO; | |
1297 | } | |
1298 | ||
de81a169 PB |
1299 | ret = qemu_fdatasync(aiocb->aio_fildes); |
1300 | if (ret == -1) { | |
e5bcf967 KW |
1301 | /* There is no clear definition of the semantics of a failing fsync(), |
1302 | * so we may have to assume the worst. The sad truth is that this | |
1303 | * assumption is correct for Linux. Some pages are now probably marked | |
1304 | * clean in the page cache even though they are inconsistent with the | |
1305 | * on-disk contents. The next fdatasync() call would succeed, but no | |
1306 | * further writeback attempt will be made. We can't get back to a state | |
1307 | * in which we know what is on disk (we would have to rewrite | |
1308 | * everything that was touched since the last fdatasync() at least), so | |
1309 | * make bdrv_flush() fail permanently. Given that the behaviour isn't | |
1310 | * really defined, I have little hope that other OSes are doing better. | |
1311 | * | |
1312 | * Obviously, this doesn't affect O_DIRECT, which bypasses the page | |
1313 | * cache. */ | |
1314 | if ((s->open_flags & O_DIRECT) == 0) { | |
1315 | s->page_cache_inconsistent = true; | |
1316 | } | |
de81a169 PB |
1317 | return -errno; |
1318 | } | |
1319 | return 0; | |
1320 | } | |
1321 | ||
1322 | #ifdef CONFIG_PREADV | |
1323 | ||
1324 | static bool preadv_present = true; | |
1325 | ||
1326 | static ssize_t | |
1327 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1328 | { | |
1329 | return preadv(fd, iov, nr_iov, offset); | |
1330 | } | |
1331 | ||
1332 | static ssize_t | |
1333 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1334 | { | |
1335 | return pwritev(fd, iov, nr_iov, offset); | |
1336 | } | |
1337 | ||
1338 | #else | |
1339 | ||
1340 | static bool preadv_present = false; | |
1341 | ||
1342 | static ssize_t | |
1343 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1344 | { | |
1345 | return -ENOSYS; | |
1346 | } | |
1347 | ||
1348 | static ssize_t | |
1349 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
1350 | { | |
1351 | return -ENOSYS; | |
1352 | } | |
1353 | ||
1354 | #endif | |
1355 | ||
1356 | static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb) | |
1357 | { | |
1358 | ssize_t len; | |
1359 | ||
1360 | do { | |
1361 | if (aiocb->aio_type & QEMU_AIO_WRITE) | |
1362 | len = qemu_pwritev(aiocb->aio_fildes, | |
d57c44d0 KW |
1363 | aiocb->io.iov, |
1364 | aiocb->io.niov, | |
de81a169 PB |
1365 | aiocb->aio_offset); |
1366 | else | |
1367 | len = qemu_preadv(aiocb->aio_fildes, | |
d57c44d0 KW |
1368 | aiocb->io.iov, |
1369 | aiocb->io.niov, | |
de81a169 PB |
1370 | aiocb->aio_offset); |
1371 | } while (len == -1 && errno == EINTR); | |
1372 | ||
1373 | if (len == -1) { | |
1374 | return -errno; | |
1375 | } | |
1376 | return len; | |
1377 | } | |
1378 | ||
1379 | /* | |
1380 | * Read/writes the data to/from a given linear buffer. | |
1381 | * | |
1382 | * Returns the number of bytes handles or -errno in case of an error. Short | |
1383 | * reads are only returned if the end of the file is reached. | |
1384 | */ | |
1385 | static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf) | |
1386 | { | |
1387 | ssize_t offset = 0; | |
1388 | ssize_t len; | |
1389 | ||
1390 | while (offset < aiocb->aio_nbytes) { | |
1391 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1392 | len = pwrite(aiocb->aio_fildes, | |
1393 | (const char *)buf + offset, | |
1394 | aiocb->aio_nbytes - offset, | |
1395 | aiocb->aio_offset + offset); | |
1396 | } else { | |
1397 | len = pread(aiocb->aio_fildes, | |
1398 | buf + offset, | |
1399 | aiocb->aio_nbytes - offset, | |
1400 | aiocb->aio_offset + offset); | |
1401 | } | |
1402 | if (len == -1 && errno == EINTR) { | |
1403 | continue; | |
61ed73cf SH |
1404 | } else if (len == -1 && errno == EINVAL && |
1405 | (aiocb->bs->open_flags & BDRV_O_NOCACHE) && | |
1406 | !(aiocb->aio_type & QEMU_AIO_WRITE) && | |
1407 | offset > 0) { | |
1408 | /* O_DIRECT pread() may fail with EINVAL when offset is unaligned | |
1409 | * after a short read. Assume that O_DIRECT short reads only occur | |
1410 | * at EOF. Therefore this is a short read, not an I/O error. | |
1411 | */ | |
1412 | break; | |
de81a169 PB |
1413 | } else if (len == -1) { |
1414 | offset = -errno; | |
1415 | break; | |
1416 | } else if (len == 0) { | |
1417 | break; | |
1418 | } | |
1419 | offset += len; | |
1420 | } | |
1421 | ||
1422 | return offset; | |
1423 | } | |
1424 | ||
999e6b69 | 1425 | static int handle_aiocb_rw(void *opaque) |
de81a169 | 1426 | { |
999e6b69 | 1427 | RawPosixAIOData *aiocb = opaque; |
de81a169 PB |
1428 | ssize_t nbytes; |
1429 | char *buf; | |
1430 | ||
1431 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { | |
1432 | /* | |
1433 | * If there is just a single buffer, and it is properly aligned | |
1434 | * we can just use plain pread/pwrite without any problems. | |
1435 | */ | |
d57c44d0 | 1436 | if (aiocb->io.niov == 1) { |
54c7ca1b KW |
1437 | nbytes = handle_aiocb_rw_linear(aiocb, aiocb->io.iov->iov_base); |
1438 | goto out; | |
de81a169 PB |
1439 | } |
1440 | /* | |
1441 | * We have more than one iovec, and all are properly aligned. | |
1442 | * | |
1443 | * Try preadv/pwritev first and fall back to linearizing the | |
1444 | * buffer if it's not supported. | |
1445 | */ | |
1446 | if (preadv_present) { | |
1447 | nbytes = handle_aiocb_rw_vector(aiocb); | |
1448 | if (nbytes == aiocb->aio_nbytes || | |
1449 | (nbytes < 0 && nbytes != -ENOSYS)) { | |
54c7ca1b | 1450 | goto out; |
de81a169 PB |
1451 | } |
1452 | preadv_present = false; | |
1453 | } | |
1454 | ||
1455 | /* | |
1456 | * XXX(hch): short read/write. no easy way to handle the reminder | |
1457 | * using these interfaces. For now retry using plain | |
1458 | * pread/pwrite? | |
1459 | */ | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * Ok, we have to do it the hard way, copy all segments into | |
1464 | * a single aligned buffer. | |
1465 | */ | |
50d4a858 KW |
1466 | buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes); |
1467 | if (buf == NULL) { | |
54c7ca1b KW |
1468 | nbytes = -ENOMEM; |
1469 | goto out; | |
50d4a858 KW |
1470 | } |
1471 | ||
de81a169 PB |
1472 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
1473 | char *p = buf; | |
1474 | int i; | |
1475 | ||
d57c44d0 KW |
1476 | for (i = 0; i < aiocb->io.niov; ++i) { |
1477 | memcpy(p, aiocb->io.iov[i].iov_base, aiocb->io.iov[i].iov_len); | |
1478 | p += aiocb->io.iov[i].iov_len; | |
de81a169 | 1479 | } |
8eb029c2 | 1480 | assert(p - buf == aiocb->aio_nbytes); |
de81a169 PB |
1481 | } |
1482 | ||
1483 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
1484 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { | |
1485 | char *p = buf; | |
1486 | size_t count = aiocb->aio_nbytes, copy; | |
1487 | int i; | |
1488 | ||
d57c44d0 | 1489 | for (i = 0; i < aiocb->io.niov && count; ++i) { |
de81a169 | 1490 | copy = count; |
d57c44d0 KW |
1491 | if (copy > aiocb->io.iov[i].iov_len) { |
1492 | copy = aiocb->io.iov[i].iov_len; | |
de81a169 | 1493 | } |
d57c44d0 | 1494 | memcpy(aiocb->io.iov[i].iov_base, p, copy); |
8eb029c2 | 1495 | assert(count >= copy); |
de81a169 PB |
1496 | p += copy; |
1497 | count -= copy; | |
1498 | } | |
8eb029c2 | 1499 | assert(count == 0); |
de81a169 PB |
1500 | } |
1501 | qemu_vfree(buf); | |
1502 | ||
54c7ca1b KW |
1503 | out: |
1504 | if (nbytes == aiocb->aio_nbytes) { | |
1505 | return 0; | |
1506 | } else if (nbytes >= 0 && nbytes < aiocb->aio_nbytes) { | |
1507 | if (aiocb->aio_type & QEMU_AIO_WRITE) { | |
1508 | return -EINVAL; | |
1509 | } else { | |
1510 | iov_memset(aiocb->io.iov, aiocb->io.niov, nbytes, | |
1511 | 0, aiocb->aio_nbytes - nbytes); | |
1512 | return 0; | |
1513 | } | |
1514 | } else { | |
1515 | assert(nbytes < 0); | |
1516 | return nbytes; | |
1517 | } | |
de81a169 PB |
1518 | } |
1519 | ||
1486df0e DL |
1520 | static int translate_err(int err) |
1521 | { | |
1522 | if (err == -ENODEV || err == -ENOSYS || err == -EOPNOTSUPP || | |
1523 | err == -ENOTTY) { | |
1524 | err = -ENOTSUP; | |
1525 | } | |
1526 | return err; | |
1527 | } | |
1528 | ||
d50d8222 | 1529 | #ifdef CONFIG_FALLOCATE |
0b991712 DL |
1530 | static int do_fallocate(int fd, int mode, off_t offset, off_t len) |
1531 | { | |
1532 | do { | |
1533 | if (fallocate(fd, mode, offset, len) == 0) { | |
1534 | return 0; | |
1535 | } | |
1536 | } while (errno == EINTR); | |
1537 | return translate_err(-errno); | |
1538 | } | |
1539 | #endif | |
1540 | ||
37cc9f7f | 1541 | static ssize_t handle_aiocb_write_zeroes_block(RawPosixAIOData *aiocb) |
97a2ae34 | 1542 | { |
37cc9f7f | 1543 | int ret = -ENOTSUP; |
97a2ae34 PB |
1544 | BDRVRawState *s = aiocb->bs->opaque; |
1545 | ||
37cc9f7f | 1546 | if (!s->has_write_zeroes) { |
97a2ae34 PB |
1547 | return -ENOTSUP; |
1548 | } | |
1549 | ||
97a2ae34 | 1550 | #ifdef BLKZEROOUT |
738301e1 KW |
1551 | /* The BLKZEROOUT implementation in the kernel doesn't set |
1552 | * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow | |
1553 | * fallbacks. */ | |
1554 | if (!(aiocb->aio_type & QEMU_AIO_NO_FALLBACK)) { | |
1555 | do { | |
1556 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1557 | if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) { | |
1558 | return 0; | |
1559 | } | |
1560 | } while (errno == EINTR); | |
37cc9f7f | 1561 | |
738301e1 | 1562 | ret = translate_err(-errno); |
effecce6 KW |
1563 | if (ret == -ENOTSUP) { |
1564 | s->has_write_zeroes = false; | |
1565 | } | |
738301e1 | 1566 | } |
97a2ae34 | 1567 | #endif |
97a2ae34 | 1568 | |
97a2ae34 PB |
1569 | return ret; |
1570 | } | |
1571 | ||
7154d8ae | 1572 | static int handle_aiocb_write_zeroes(void *opaque) |
37cc9f7f | 1573 | { |
7154d8ae | 1574 | RawPosixAIOData *aiocb = opaque; |
70d9110b | 1575 | #ifdef CONFIG_FALLOCATE |
b2c6f23f | 1576 | BDRVRawState *s = aiocb->bs->opaque; |
70d9110b DL |
1577 | int64_t len; |
1578 | #endif | |
37cc9f7f DL |
1579 | |
1580 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1581 | return handle_aiocb_write_zeroes_block(aiocb); | |
1582 | } | |
1583 | ||
b953f075 DL |
1584 | #ifdef CONFIG_FALLOCATE_ZERO_RANGE |
1585 | if (s->has_write_zeroes) { | |
1586 | int ret = do_fallocate(s->fd, FALLOC_FL_ZERO_RANGE, | |
1587 | aiocb->aio_offset, aiocb->aio_nbytes); | |
294682cc AS |
1588 | if (ret == -EINVAL) { |
1589 | /* | |
1590 | * Allow falling back to pwrite for file systems that | |
1591 | * do not support fallocate() for an unaligned byte range. | |
1592 | */ | |
1593 | return -ENOTSUP; | |
1594 | } | |
b953f075 DL |
1595 | if (ret == 0 || ret != -ENOTSUP) { |
1596 | return ret; | |
1597 | } | |
1598 | s->has_write_zeroes = false; | |
1599 | } | |
1600 | #endif | |
1601 | ||
1cdc3239 DL |
1602 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
1603 | if (s->has_discard && s->has_fallocate) { | |
1604 | int ret = do_fallocate(s->fd, | |
1605 | FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
1606 | aiocb->aio_offset, aiocb->aio_nbytes); | |
1607 | if (ret == 0) { | |
1608 | ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); | |
1609 | if (ret == 0 || ret != -ENOTSUP) { | |
1610 | return ret; | |
1611 | } | |
1612 | s->has_fallocate = false; | |
1613 | } else if (ret != -ENOTSUP) { | |
1614 | return ret; | |
1615 | } else { | |
1616 | s->has_discard = false; | |
1617 | } | |
1618 | } | |
1619 | #endif | |
1620 | ||
d50d8222 | 1621 | #ifdef CONFIG_FALLOCATE |
70d9110b DL |
1622 | /* Last resort: we are trying to extend the file with zeroed data. This |
1623 | * can be done via fallocate(fd, 0) */ | |
1624 | len = bdrv_getlength(aiocb->bs); | |
1625 | if (s->has_fallocate && len >= 0 && aiocb->aio_offset >= len) { | |
d50d8222 DL |
1626 | int ret = do_fallocate(s->fd, 0, aiocb->aio_offset, aiocb->aio_nbytes); |
1627 | if (ret == 0 || ret != -ENOTSUP) { | |
1628 | return ret; | |
1629 | } | |
1630 | s->has_fallocate = false; | |
1631 | } | |
1632 | #endif | |
1633 | ||
37cc9f7f DL |
1634 | return -ENOTSUP; |
1635 | } | |
1636 | ||
7154d8ae | 1637 | static int handle_aiocb_write_zeroes_unmap(void *opaque) |
34fa110e | 1638 | { |
7154d8ae | 1639 | RawPosixAIOData *aiocb = opaque; |
34fa110e | 1640 | BDRVRawState *s G_GNUC_UNUSED = aiocb->bs->opaque; |
34fa110e KW |
1641 | |
1642 | /* First try to write zeros and unmap at the same time */ | |
1643 | ||
1644 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
b3ac2b94 SS |
1645 | int ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1646 | aiocb->aio_offset, aiocb->aio_nbytes); | |
34fa110e KW |
1647 | if (ret != -ENOTSUP) { |
1648 | return ret; | |
1649 | } | |
1650 | #endif | |
1651 | ||
34fa110e KW |
1652 | /* If we couldn't manage to unmap while guaranteed that the area reads as |
1653 | * all-zero afterwards, just write zeroes without unmapping */ | |
b3ac2b94 | 1654 | return handle_aiocb_write_zeroes(aiocb); |
34fa110e KW |
1655 | } |
1656 | ||
1efad060 FZ |
1657 | #ifndef HAVE_COPY_FILE_RANGE |
1658 | static off_t copy_file_range(int in_fd, off_t *in_off, int out_fd, | |
1659 | off_t *out_off, size_t len, unsigned int flags) | |
1660 | { | |
1661 | #ifdef __NR_copy_file_range | |
1662 | return syscall(__NR_copy_file_range, in_fd, in_off, out_fd, | |
1663 | out_off, len, flags); | |
1664 | #else | |
1665 | errno = ENOSYS; | |
1666 | return -1; | |
1667 | #endif | |
1668 | } | |
1669 | #endif | |
1670 | ||
58a209c4 | 1671 | static int handle_aiocb_copy_range(void *opaque) |
1efad060 | 1672 | { |
58a209c4 | 1673 | RawPosixAIOData *aiocb = opaque; |
1efad060 FZ |
1674 | uint64_t bytes = aiocb->aio_nbytes; |
1675 | off_t in_off = aiocb->aio_offset; | |
d57c44d0 | 1676 | off_t out_off = aiocb->copy_range.aio_offset2; |
1efad060 FZ |
1677 | |
1678 | while (bytes) { | |
1679 | ssize_t ret = copy_file_range(aiocb->aio_fildes, &in_off, | |
d57c44d0 | 1680 | aiocb->copy_range.aio_fd2, &out_off, |
1efad060 | 1681 | bytes, 0); |
ecc983a5 | 1682 | trace_file_copy_file_range(aiocb->bs, aiocb->aio_fildes, in_off, |
d57c44d0 KW |
1683 | aiocb->copy_range.aio_fd2, out_off, bytes, |
1684 | 0, ret); | |
c436e3d0 FZ |
1685 | if (ret == 0) { |
1686 | /* No progress (e.g. when beyond EOF), let the caller fall back to | |
1687 | * buffer I/O. */ | |
1688 | return -ENOSPC; | |
1efad060 FZ |
1689 | } |
1690 | if (ret < 0) { | |
c436e3d0 FZ |
1691 | switch (errno) { |
1692 | case ENOSYS: | |
1efad060 | 1693 | return -ENOTSUP; |
c436e3d0 FZ |
1694 | case EINTR: |
1695 | continue; | |
1696 | default: | |
1efad060 FZ |
1697 | return -errno; |
1698 | } | |
1699 | } | |
1efad060 FZ |
1700 | bytes -= ret; |
1701 | } | |
1702 | return 0; | |
1703 | } | |
1704 | ||
46ee0f46 | 1705 | static int handle_aiocb_discard(void *opaque) |
8238010b | 1706 | { |
46ee0f46 | 1707 | RawPosixAIOData *aiocb = opaque; |
8238010b PB |
1708 | int ret = -EOPNOTSUPP; |
1709 | BDRVRawState *s = aiocb->bs->opaque; | |
1710 | ||
7ce21016 PB |
1711 | if (!s->has_discard) { |
1712 | return -ENOTSUP; | |
8238010b PB |
1713 | } |
1714 | ||
1715 | if (aiocb->aio_type & QEMU_AIO_BLKDEV) { | |
1716 | #ifdef BLKDISCARD | |
1717 | do { | |
1718 | uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes }; | |
1719 | if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) { | |
1720 | return 0; | |
1721 | } | |
1722 | } while (errno == EINTR); | |
1723 | ||
1724 | ret = -errno; | |
1725 | #endif | |
1726 | } else { | |
8238010b | 1727 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
0b991712 DL |
1728 | ret = do_fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, |
1729 | aiocb->aio_offset, aiocb->aio_nbytes); | |
8238010b PB |
1730 | #endif |
1731 | } | |
1732 | ||
1486df0e DL |
1733 | ret = translate_err(ret); |
1734 | if (ret == -ENOTSUP) { | |
7ce21016 | 1735 | s->has_discard = false; |
8238010b PB |
1736 | } |
1737 | return ret; | |
1738 | } | |
1739 | ||
3a20013f NS |
1740 | /* |
1741 | * Help alignment probing by allocating the first block. | |
1742 | * | |
1743 | * When reading with direct I/O from unallocated area on Gluster backed by XFS, | |
1744 | * reading succeeds regardless of request length. In this case we fallback to | |
1745 | * safe alignment which is not optimal. Allocating the first block avoids this | |
1746 | * fallback. | |
1747 | * | |
1748 | * fd may be opened with O_DIRECT, but we don't know the buffer alignment or | |
1749 | * request alignment, so we use safe values. | |
1750 | * | |
1751 | * Returns: 0 on success, -errno on failure. Since this is an optimization, | |
1752 | * caller may ignore failures. | |
1753 | */ | |
1754 | static int allocate_first_block(int fd, size_t max_size) | |
1755 | { | |
1756 | size_t write_size = (max_size < MAX_BLOCKSIZE) | |
1757 | ? BDRV_SECTOR_SIZE | |
1758 | : MAX_BLOCKSIZE; | |
038adc2f | 1759 | size_t max_align = MAX(MAX_BLOCKSIZE, qemu_real_host_page_size); |
3a20013f NS |
1760 | void *buf; |
1761 | ssize_t n; | |
1762 | int ret; | |
1763 | ||
1764 | buf = qemu_memalign(max_align, write_size); | |
1765 | memset(buf, 0, write_size); | |
1766 | ||
1767 | do { | |
1768 | n = pwrite(fd, buf, write_size, 0); | |
1769 | } while (n == -1 && errno == EINTR); | |
1770 | ||
1771 | ret = (n == -1) ? -errno : 0; | |
1772 | ||
1773 | qemu_vfree(buf); | |
1774 | return ret; | |
1775 | } | |
1776 | ||
29cb4c01 | 1777 | static int handle_aiocb_truncate(void *opaque) |
93f4e2ff | 1778 | { |
29cb4c01 | 1779 | RawPosixAIOData *aiocb = opaque; |
93f4e2ff KW |
1780 | int result = 0; |
1781 | int64_t current_length = 0; | |
1782 | char *buf = NULL; | |
1783 | struct stat st; | |
1784 | int fd = aiocb->aio_fildes; | |
1785 | int64_t offset = aiocb->aio_offset; | |
d57c44d0 KW |
1786 | PreallocMode prealloc = aiocb->truncate.prealloc; |
1787 | Error **errp = aiocb->truncate.errp; | |
93f4e2ff KW |
1788 | |
1789 | if (fstat(fd, &st) < 0) { | |
1790 | result = -errno; | |
1791 | error_setg_errno(errp, -result, "Could not stat file"); | |
1792 | return result; | |
1793 | } | |
1794 | ||
1795 | current_length = st.st_size; | |
d57c44d0 | 1796 | if (current_length > offset && prealloc != PREALLOC_MODE_OFF) { |
93f4e2ff KW |
1797 | error_setg(errp, "Cannot use preallocation for shrinking files"); |
1798 | return -ENOTSUP; | |
1799 | } | |
1800 | ||
d57c44d0 | 1801 | switch (prealloc) { |
93f4e2ff KW |
1802 | #ifdef CONFIG_POSIX_FALLOCATE |
1803 | case PREALLOC_MODE_FALLOC: | |
1804 | /* | |
1805 | * Truncating before posix_fallocate() makes it about twice slower on | |
1806 | * file systems that do not support fallocate(), trying to check if a | |
1807 | * block is allocated before allocating it, so don't do that here. | |
1808 | */ | |
1809 | if (offset != current_length) { | |
1810 | result = -posix_fallocate(fd, current_length, | |
1811 | offset - current_length); | |
1812 | if (result != 0) { | |
1813 | /* posix_fallocate() doesn't set errno. */ | |
1814 | error_setg_errno(errp, -result, | |
1815 | "Could not preallocate new data"); | |
3a20013f NS |
1816 | } else if (current_length == 0) { |
1817 | /* | |
1818 | * posix_fallocate() uses fallocate() if the filesystem | |
1819 | * supports it, or fallback to manually writing zeroes. If | |
1820 | * fallocate() was used, unaligned reads from the fallocated | |
1821 | * area in raw_probe_alignment() will succeed, hence we need to | |
1822 | * allocate the first block. | |
1823 | * | |
1824 | * Optimize future alignment probing; ignore failures. | |
1825 | */ | |
1826 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1827 | } |
1828 | } else { | |
1829 | result = 0; | |
1830 | } | |
1831 | goto out; | |
1832 | #endif | |
1833 | case PREALLOC_MODE_FULL: | |
1834 | { | |
1835 | int64_t num = 0, left = offset - current_length; | |
1836 | off_t seek_result; | |
1837 | ||
1838 | /* | |
1839 | * Knowing the final size from the beginning could allow the file | |
1840 | * system driver to do less allocations and possibly avoid | |
1841 | * fragmentation of the file. | |
1842 | */ | |
1843 | if (ftruncate(fd, offset) != 0) { | |
1844 | result = -errno; | |
1845 | error_setg_errno(errp, -result, "Could not resize file"); | |
1846 | goto out; | |
1847 | } | |
1848 | ||
1849 | buf = g_malloc0(65536); | |
1850 | ||
1851 | seek_result = lseek(fd, current_length, SEEK_SET); | |
1852 | if (seek_result < 0) { | |
1853 | result = -errno; | |
1854 | error_setg_errno(errp, -result, | |
1855 | "Failed to seek to the old end of file"); | |
1856 | goto out; | |
1857 | } | |
1858 | ||
1859 | while (left > 0) { | |
1860 | num = MIN(left, 65536); | |
1861 | result = write(fd, buf, num); | |
1862 | if (result < 0) { | |
a1c81f4f FZ |
1863 | if (errno == EINTR) { |
1864 | continue; | |
1865 | } | |
93f4e2ff KW |
1866 | result = -errno; |
1867 | error_setg_errno(errp, -result, | |
1868 | "Could not write zeros for preallocation"); | |
1869 | goto out; | |
1870 | } | |
1871 | left -= result; | |
1872 | } | |
1873 | if (result >= 0) { | |
1874 | result = fsync(fd); | |
1875 | if (result < 0) { | |
1876 | result = -errno; | |
1877 | error_setg_errno(errp, -result, | |
1878 | "Could not flush file to disk"); | |
1879 | goto out; | |
1880 | } | |
1881 | } | |
1882 | goto out; | |
1883 | } | |
1884 | case PREALLOC_MODE_OFF: | |
1885 | if (ftruncate(fd, offset) != 0) { | |
1886 | result = -errno; | |
1887 | error_setg_errno(errp, -result, "Could not resize file"); | |
3a20013f NS |
1888 | } else if (current_length == 0 && offset > current_length) { |
1889 | /* Optimize future alignment probing; ignore failures. */ | |
1890 | allocate_first_block(fd, offset); | |
93f4e2ff KW |
1891 | } |
1892 | return result; | |
1893 | default: | |
1894 | result = -ENOTSUP; | |
1895 | error_setg(errp, "Unsupported preallocation mode: %s", | |
d57c44d0 | 1896 | PreallocMode_str(prealloc)); |
93f4e2ff KW |
1897 | return result; |
1898 | } | |
1899 | ||
1900 | out: | |
1901 | if (result < 0) { | |
1902 | if (ftruncate(fd, current_length) < 0) { | |
1903 | error_report("Failed to restore old file length: %s", | |
1904 | strerror(errno)); | |
1905 | } | |
1906 | } | |
1907 | ||
1908 | g_free(buf); | |
1909 | return result; | |
1910 | } | |
1911 | ||
5d5de250 KW |
1912 | static int coroutine_fn raw_thread_pool_submit(BlockDriverState *bs, |
1913 | ThreadPoolFunc func, void *arg) | |
1914 | { | |
1915 | /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */ | |
1916 | ThreadPool *pool = aio_get_thread_pool(bdrv_get_aio_context(bs)); | |
1917 | return thread_pool_submit_co(pool, func, arg); | |
1918 | } | |
1919 | ||
9d52aa3c KW |
1920 | static int coroutine_fn raw_co_prw(BlockDriverState *bs, uint64_t offset, |
1921 | uint64_t bytes, QEMUIOVector *qiov, int type) | |
83f64091 | 1922 | { |
ce1a14dc | 1923 | BDRVRawState *s = bs->opaque; |
999e6b69 | 1924 | RawPosixAIOData acb; |
ce1a14dc | 1925 | |
19cb3738 | 1926 | if (fd_open(bs) < 0) |
2174f12b | 1927 | return -EIO; |
19cb3738 | 1928 | |
f141eafe | 1929 | /* |
c6447510 AM |
1930 | * When using O_DIRECT, the request must be aligned to be able to use |
1931 | * either libaio or io_uring interface. If not fail back to regular thread | |
1932 | * pool read/write code which emulates this for us if we | |
1933 | * set QEMU_AIO_MISALIGNED. | |
f141eafe | 1934 | */ |
c6447510 AM |
1935 | if (s->needs_alignment && !bdrv_qiov_is_aligned(bs, qiov)) { |
1936 | type |= QEMU_AIO_MISALIGNED; | |
1937 | #ifdef CONFIG_LINUX_IO_URING | |
1938 | } else if (s->use_linux_io_uring) { | |
1939 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
1940 | assert(qiov->size == bytes); | |
1941 | return luring_co_submit(bs, aio, s->fd, offset, qiov, type); | |
1942 | #endif | |
e44bd6fc | 1943 | #ifdef CONFIG_LINUX_AIO |
c6447510 AM |
1944 | } else if (s->use_linux_aio) { |
1945 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); | |
1946 | assert(qiov->size == bytes); | |
1947 | return laio_co_submit(bs, aio, s->fd, offset, qiov, type); | |
e44bd6fc | 1948 | #endif |
9ef91a67 | 1949 | } |
f141eafe | 1950 | |
999e6b69 KW |
1951 | acb = (RawPosixAIOData) { |
1952 | .bs = bs, | |
1953 | .aio_fildes = s->fd, | |
1954 | .aio_type = type, | |
1955 | .aio_offset = offset, | |
1956 | .aio_nbytes = bytes, | |
1957 | .io = { | |
1958 | .iov = qiov->iov, | |
1959 | .niov = qiov->niov, | |
1960 | }, | |
1961 | }; | |
1962 | ||
1963 | assert(qiov->size == bytes); | |
1964 | return raw_thread_pool_submit(bs, handle_aiocb_rw, &acb); | |
2174f12b KW |
1965 | } |
1966 | ||
9d52aa3c KW |
1967 | static int coroutine_fn raw_co_preadv(BlockDriverState *bs, uint64_t offset, |
1968 | uint64_t bytes, QEMUIOVector *qiov, | |
1969 | int flags) | |
2174f12b | 1970 | { |
9d52aa3c | 1971 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_READ); |
2174f12b KW |
1972 | } |
1973 | ||
9d52aa3c KW |
1974 | static int coroutine_fn raw_co_pwritev(BlockDriverState *bs, uint64_t offset, |
1975 | uint64_t bytes, QEMUIOVector *qiov, | |
1976 | int flags) | |
2174f12b | 1977 | { |
9d52aa3c KW |
1978 | assert(flags == 0); |
1979 | return raw_co_prw(bs, offset, bytes, qiov, QEMU_AIO_WRITE); | |
83f64091 FB |
1980 | } |
1981 | ||
1b3abdcc ML |
1982 | static void raw_aio_plug(BlockDriverState *bs) |
1983 | { | |
c6447510 | 1984 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 1985 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 1986 | if (s->use_linux_aio) { |
0187f5c9 PB |
1987 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
1988 | laio_io_plug(bs, aio); | |
1b3abdcc ML |
1989 | } |
1990 | #endif | |
c6447510 AM |
1991 | #ifdef CONFIG_LINUX_IO_URING |
1992 | if (s->use_linux_io_uring) { | |
1993 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
1994 | luring_io_plug(bs, aio); | |
1995 | } | |
1996 | #endif | |
1b3abdcc ML |
1997 | } |
1998 | ||
1999 | static void raw_aio_unplug(BlockDriverState *bs) | |
2000 | { | |
c6447510 | 2001 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
1b3abdcc | 2002 | #ifdef CONFIG_LINUX_AIO |
0a4279d9 | 2003 | if (s->use_linux_aio) { |
0187f5c9 PB |
2004 | LinuxAioState *aio = aio_get_linux_aio(bdrv_get_aio_context(bs)); |
2005 | laio_io_unplug(bs, aio); | |
1b3abdcc ML |
2006 | } |
2007 | #endif | |
c6447510 AM |
2008 | #ifdef CONFIG_LINUX_IO_URING |
2009 | if (s->use_linux_io_uring) { | |
2010 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2011 | luring_io_unplug(bs, aio); | |
2012 | } | |
2013 | #endif | |
1b3abdcc ML |
2014 | } |
2015 | ||
33d70fb6 | 2016 | static int raw_co_flush_to_disk(BlockDriverState *bs) |
b2e12bc6 CH |
2017 | { |
2018 | BDRVRawState *s = bs->opaque; | |
06dc9bd5 | 2019 | RawPosixAIOData acb; |
33d70fb6 | 2020 | int ret; |
b2e12bc6 | 2021 | |
33d70fb6 KW |
2022 | ret = fd_open(bs); |
2023 | if (ret < 0) { | |
2024 | return ret; | |
2025 | } | |
b2e12bc6 | 2026 | |
06dc9bd5 KW |
2027 | acb = (RawPosixAIOData) { |
2028 | .bs = bs, | |
2029 | .aio_fildes = s->fd, | |
2030 | .aio_type = QEMU_AIO_FLUSH, | |
2031 | }; | |
2032 | ||
c6447510 AM |
2033 | #ifdef CONFIG_LINUX_IO_URING |
2034 | if (s->use_linux_io_uring) { | |
2035 | LuringState *aio = aio_get_linux_io_uring(bdrv_get_aio_context(bs)); | |
2036 | return luring_co_submit(bs, aio, s->fd, 0, NULL, QEMU_AIO_FLUSH); | |
2037 | } | |
2038 | #endif | |
06dc9bd5 | 2039 | return raw_thread_pool_submit(bs, handle_aiocb_flush, &acb); |
b2e12bc6 CH |
2040 | } |
2041 | ||
ed6e2161 NA |
2042 | static void raw_aio_attach_aio_context(BlockDriverState *bs, |
2043 | AioContext *new_context) | |
2044 | { | |
c6447510 | 2045 | BDRVRawState __attribute__((unused)) *s = bs->opaque; |
ed6e2161 | 2046 | #ifdef CONFIG_LINUX_AIO |
ed6e2161 | 2047 | if (s->use_linux_aio) { |
cb09104e | 2048 | Error *local_err = NULL; |
ed6e2161 NA |
2049 | if (!aio_setup_linux_aio(new_context, &local_err)) { |
2050 | error_reportf_err(local_err, "Unable to use native AIO, " | |
2051 | "falling back to thread pool: "); | |
2052 | s->use_linux_aio = false; | |
2053 | } | |
2054 | } | |
2055 | #endif | |
c6447510 AM |
2056 | #ifdef CONFIG_LINUX_IO_URING |
2057 | if (s->use_linux_io_uring) { | |
2058 | Error *local_err; | |
2059 | if (!aio_setup_linux_io_uring(new_context, &local_err)) { | |
2060 | error_reportf_err(local_err, "Unable to use linux io_uring, " | |
2061 | "falling back to thread pool: "); | |
2062 | s->use_linux_io_uring = false; | |
2063 | } | |
2064 | } | |
2065 | #endif | |
ed6e2161 NA |
2066 | } |
2067 | ||
83f64091 FB |
2068 | static void raw_close(BlockDriverState *bs) |
2069 | { | |
2070 | BDRVRawState *s = bs->opaque; | |
c2f3426c | 2071 | |
19cb3738 | 2072 | if (s->fd >= 0) { |
2e1e79da | 2073 | qemu_close(s->fd); |
19cb3738 FB |
2074 | s->fd = -1; |
2075 | } | |
83f64091 FB |
2076 | } |
2077 | ||
d0bc9e5d HR |
2078 | /** |
2079 | * Truncates the given regular file @fd to @offset and, when growing, fills the | |
2080 | * new space according to @prealloc. | |
2081 | * | |
2082 | * Returns: 0 on success, -errno on failure. | |
2083 | */ | |
93f4e2ff KW |
2084 | static int coroutine_fn |
2085 | raw_regular_truncate(BlockDriverState *bs, int fd, int64_t offset, | |
2086 | PreallocMode prealloc, Error **errp) | |
9f63b07e | 2087 | { |
29cb4c01 | 2088 | RawPosixAIOData acb; |
d0bc9e5d | 2089 | |
29cb4c01 | 2090 | acb = (RawPosixAIOData) { |
93f4e2ff KW |
2091 | .bs = bs, |
2092 | .aio_fildes = fd, | |
2093 | .aio_type = QEMU_AIO_TRUNCATE, | |
2094 | .aio_offset = offset, | |
d57c44d0 KW |
2095 | .truncate = { |
2096 | .prealloc = prealloc, | |
2097 | .errp = errp, | |
2098 | }, | |
93f4e2ff | 2099 | }; |
d0bc9e5d | 2100 | |
29cb4c01 | 2101 | return raw_thread_pool_submit(bs, handle_aiocb_truncate, &acb); |
9f63b07e HR |
2102 | } |
2103 | ||
061ca8a3 | 2104 | static int coroutine_fn raw_co_truncate(BlockDriverState *bs, int64_t offset, |
c80d8b06 | 2105 | bool exact, PreallocMode prealloc, |
92b92799 | 2106 | BdrvRequestFlags flags, Error **errp) |
83f64091 FB |
2107 | { |
2108 | BDRVRawState *s = bs->opaque; | |
55b949c8 | 2109 | struct stat st; |
f59adb32 | 2110 | int ret; |
55b949c8 CH |
2111 | |
2112 | if (fstat(s->fd, &st)) { | |
f59adb32 HR |
2113 | ret = -errno; |
2114 | error_setg_errno(errp, -ret, "Failed to fstat() the file"); | |
2115 | return ret; | |
55b949c8 CH |
2116 | } |
2117 | ||
2118 | if (S_ISREG(st.st_mode)) { | |
82325ae5 | 2119 | /* Always resizes to the exact @offset */ |
93f4e2ff | 2120 | return raw_regular_truncate(bs, s->fd, offset, prealloc, errp); |
35d72602 HR |
2121 | } |
2122 | ||
2123 | if (prealloc != PREALLOC_MODE_OFF) { | |
2124 | error_setg(errp, "Preallocation mode '%s' unsupported for this " | |
977c736f | 2125 | "non-regular file", PreallocMode_str(prealloc)); |
35d72602 HR |
2126 | return -ENOTSUP; |
2127 | } | |
2128 | ||
2129 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { | |
82325ae5 HR |
2130 | int64_t cur_length = raw_getlength(bs); |
2131 | ||
2132 | if (offset != cur_length && exact) { | |
2133 | error_setg(errp, "Cannot resize device files"); | |
2134 | return -ENOTSUP; | |
2135 | } else if (offset > cur_length) { | |
f59adb32 HR |
2136 | error_setg(errp, "Cannot grow device files"); |
2137 | return -EINVAL; | |
2138 | } | |
55b949c8 | 2139 | } else { |
f59adb32 | 2140 | error_setg(errp, "Resizing this file is not supported"); |
55b949c8 CH |
2141 | return -ENOTSUP; |
2142 | } | |
2143 | ||
83f64091 FB |
2144 | return 0; |
2145 | } | |
2146 | ||
128ab2ff BS |
2147 | #ifdef __OpenBSD__ |
2148 | static int64_t raw_getlength(BlockDriverState *bs) | |
2149 | { | |
2150 | BDRVRawState *s = bs->opaque; | |
2151 | int fd = s->fd; | |
2152 | struct stat st; | |
2153 | ||
2154 | if (fstat(fd, &st)) | |
aa729704 | 2155 | return -errno; |
128ab2ff BS |
2156 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2157 | struct disklabel dl; | |
2158 | ||
2159 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2160 | return -errno; |
128ab2ff BS |
2161 | return (uint64_t)dl.d_secsize * |
2162 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2163 | } else | |
2164 | return st.st_size; | |
2165 | } | |
d1f6fd8d CE |
2166 | #elif defined(__NetBSD__) |
2167 | static int64_t raw_getlength(BlockDriverState *bs) | |
2168 | { | |
2169 | BDRVRawState *s = bs->opaque; | |
2170 | int fd = s->fd; | |
2171 | struct stat st; | |
2172 | ||
2173 | if (fstat(fd, &st)) | |
aa729704 | 2174 | return -errno; |
d1f6fd8d CE |
2175 | if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) { |
2176 | struct dkwedge_info dkw; | |
2177 | ||
2178 | if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) { | |
2179 | return dkw.dkw_size * 512; | |
2180 | } else { | |
2181 | struct disklabel dl; | |
2182 | ||
2183 | if (ioctl(fd, DIOCGDINFO, &dl)) | |
aa729704 | 2184 | return -errno; |
d1f6fd8d CE |
2185 | return (uint64_t)dl.d_secsize * |
2186 | dl.d_partitions[DISKPART(st.st_rdev)].p_size; | |
2187 | } | |
2188 | } else | |
2189 | return st.st_size; | |
2190 | } | |
50779cc2 CH |
2191 | #elif defined(__sun__) |
2192 | static int64_t raw_getlength(BlockDriverState *bs) | |
2193 | { | |
2194 | BDRVRawState *s = bs->opaque; | |
2195 | struct dk_minfo minfo; | |
2196 | int ret; | |
aa729704 | 2197 | int64_t size; |
50779cc2 CH |
2198 | |
2199 | ret = fd_open(bs); | |
2200 | if (ret < 0) { | |
2201 | return ret; | |
2202 | } | |
2203 | ||
2204 | /* | |
2205 | * Use the DKIOCGMEDIAINFO ioctl to read the size. | |
2206 | */ | |
2207 | ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo); | |
2208 | if (ret != -1) { | |
2209 | return minfo.dki_lbsize * minfo.dki_capacity; | |
2210 | } | |
2211 | ||
2212 | /* | |
2213 | * There are reports that lseek on some devices fails, but | |
2214 | * irc discussion said that contingency on contingency was overkill. | |
2215 | */ | |
aa729704 MA |
2216 | size = lseek(s->fd, 0, SEEK_END); |
2217 | if (size < 0) { | |
2218 | return -errno; | |
2219 | } | |
2220 | return size; | |
50779cc2 CH |
2221 | } |
2222 | #elif defined(CONFIG_BSD) | |
2223 | static int64_t raw_getlength(BlockDriverState *bs) | |
83f64091 FB |
2224 | { |
2225 | BDRVRawState *s = bs->opaque; | |
2226 | int fd = s->fd; | |
2227 | int64_t size; | |
83f64091 | 2228 | struct stat sb; |
a167ba50 | 2229 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a | 2230 | int reopened = 0; |
83f64091 | 2231 | #endif |
19cb3738 FB |
2232 | int ret; |
2233 | ||
2234 | ret = fd_open(bs); | |
2235 | if (ret < 0) | |
2236 | return ret; | |
83f64091 | 2237 | |
a167ba50 | 2238 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2239 | again: |
2240 | #endif | |
83f64091 FB |
2241 | if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) { |
2242 | #ifdef DIOCGMEDIASIZE | |
7d37435b | 2243 | if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size)) |
c5e97233 BS |
2244 | #elif defined(DIOCGPART) |
2245 | { | |
2246 | struct partinfo pi; | |
2247 | if (ioctl(fd, DIOCGPART, &pi) == 0) | |
2248 | size = pi.media_size; | |
2249 | else | |
2250 | size = 0; | |
2251 | } | |
2252 | if (size == 0) | |
83f64091 | 2253 | #endif |
83affaa6 | 2254 | #if defined(__APPLE__) && defined(__MACH__) |
728dacbd JA |
2255 | { |
2256 | uint64_t sectors = 0; | |
2257 | uint32_t sector_size = 0; | |
2258 | ||
2259 | if (ioctl(fd, DKIOCGETBLOCKCOUNT, §ors) == 0 | |
2260 | && ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) == 0) { | |
2261 | size = sectors * sector_size; | |
2262 | } else { | |
2263 | size = lseek(fd, 0LL, SEEK_END); | |
2264 | if (size < 0) { | |
2265 | return -errno; | |
2266 | } | |
2267 | } | |
2268 | } | |
83f64091 FB |
2269 | #else |
2270 | size = lseek(fd, 0LL, SEEK_END); | |
aa729704 MA |
2271 | if (size < 0) { |
2272 | return -errno; | |
2273 | } | |
9f23011a | 2274 | #endif |
a167ba50 | 2275 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
9f23011a BS |
2276 | switch(s->type) { |
2277 | case FTYPE_CD: | |
2278 | /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */ | |
2279 | if (size == 2048LL * (unsigned)-1) | |
2280 | size = 0; | |
2281 | /* XXX no disc? maybe we need to reopen... */ | |
f3a5d3f8 | 2282 | if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) { |
9f23011a BS |
2283 | reopened = 1; |
2284 | goto again; | |
2285 | } | |
2286 | } | |
83f64091 | 2287 | #endif |
50779cc2 | 2288 | } else { |
83f64091 | 2289 | size = lseek(fd, 0, SEEK_END); |
aa729704 MA |
2290 | if (size < 0) { |
2291 | return -errno; | |
2292 | } | |
83f64091 | 2293 | } |
83f64091 FB |
2294 | return size; |
2295 | } | |
50779cc2 CH |
2296 | #else |
2297 | static int64_t raw_getlength(BlockDriverState *bs) | |
2298 | { | |
2299 | BDRVRawState *s = bs->opaque; | |
2300 | int ret; | |
aa729704 | 2301 | int64_t size; |
50779cc2 CH |
2302 | |
2303 | ret = fd_open(bs); | |
2304 | if (ret < 0) { | |
2305 | return ret; | |
2306 | } | |
2307 | ||
aa729704 MA |
2308 | size = lseek(s->fd, 0, SEEK_END); |
2309 | if (size < 0) { | |
2310 | return -errno; | |
2311 | } | |
2312 | return size; | |
50779cc2 | 2313 | } |
128ab2ff | 2314 | #endif |
83f64091 | 2315 | |
4a1d5e1f FZ |
2316 | static int64_t raw_get_allocated_file_size(BlockDriverState *bs) |
2317 | { | |
2318 | struct stat st; | |
2319 | BDRVRawState *s = bs->opaque; | |
2320 | ||
2321 | if (fstat(s->fd, &st) < 0) { | |
2322 | return -errno; | |
2323 | } | |
2324 | return (int64_t)st.st_blocks * 512; | |
2325 | } | |
2326 | ||
93f4e2ff KW |
2327 | static int coroutine_fn |
2328 | raw_co_create(BlockdevCreateOptions *options, Error **errp) | |
83f64091 | 2329 | { |
927f11e1 | 2330 | BlockdevCreateOptionsFile *file_opts; |
7c20c808 | 2331 | Error *local_err = NULL; |
83f64091 | 2332 | int fd; |
d815efca | 2333 | uint64_t perm, shared; |
1e37d059 | 2334 | int result = 0; |
83f64091 | 2335 | |
927f11e1 KW |
2336 | /* Validate options and set default values */ |
2337 | assert(options->driver == BLOCKDEV_DRIVER_FILE); | |
2338 | file_opts = &options->u.file; | |
464d9f64 | 2339 | |
927f11e1 KW |
2340 | if (!file_opts->has_nocow) { |
2341 | file_opts->nocow = false; | |
2342 | } | |
2343 | if (!file_opts->has_preallocation) { | |
2344 | file_opts->preallocation = PREALLOC_MODE_OFF; | |
06247428 | 2345 | } |
ffa244c8 KW |
2346 | if (!file_opts->has_extent_size_hint) { |
2347 | file_opts->extent_size_hint = 1 * MiB; | |
2348 | } | |
2349 | if (file_opts->extent_size_hint > UINT32_MAX) { | |
2350 | result = -EINVAL; | |
2351 | error_setg(errp, "Extent size hint is too large"); | |
2352 | goto out; | |
2353 | } | |
83f64091 | 2354 | |
927f11e1 | 2355 | /* Create file */ |
b8cf1913 | 2356 | fd = qemu_open(file_opts->filename, O_RDWR | O_CREAT | O_BINARY, 0644); |
1e37d059 SW |
2357 | if (fd < 0) { |
2358 | result = -errno; | |
e428e439 | 2359 | error_setg_errno(errp, -result, "Could not create file"); |
06247428 HT |
2360 | goto out; |
2361 | } | |
2362 | ||
b8cf1913 HR |
2363 | /* Take permissions: We want to discard everything, so we need |
2364 | * BLK_PERM_WRITE; and truncation to the desired size requires | |
2365 | * BLK_PERM_RESIZE. | |
2366 | * On the other hand, we cannot share the RESIZE permission | |
2367 | * because we promise that after this function, the file has the | |
2368 | * size given in the options. If someone else were to resize it | |
2369 | * concurrently, we could not guarantee that. | |
2370 | * Note that after this function, we can no longer guarantee that | |
2371 | * the file is not touched by a third party, so it may be resized | |
2372 | * then. */ | |
2373 | perm = BLK_PERM_WRITE | BLK_PERM_RESIZE; | |
2374 | shared = BLK_PERM_ALL & ~BLK_PERM_RESIZE; | |
2375 | ||
2376 | /* Step one: Take locks */ | |
2996ffad | 2377 | result = raw_apply_lock_bytes(NULL, fd, perm, ~shared, false, errp); |
b8cf1913 HR |
2378 | if (result < 0) { |
2379 | goto out_close; | |
2380 | } | |
2381 | ||
2382 | /* Step two: Check that nobody else has taken conflicting locks */ | |
2383 | result = raw_check_lock_bytes(fd, perm, shared, errp); | |
2384 | if (result < 0) { | |
b857431d FZ |
2385 | error_append_hint(errp, |
2386 | "Is another process using the image [%s]?\n", | |
2387 | file_opts->filename); | |
7c20c808 | 2388 | goto out_unlock; |
b8cf1913 HR |
2389 | } |
2390 | ||
2391 | /* Clear the file by truncating it to 0 */ | |
93f4e2ff | 2392 | result = raw_regular_truncate(NULL, fd, 0, PREALLOC_MODE_OFF, errp); |
b8cf1913 | 2393 | if (result < 0) { |
7c20c808 | 2394 | goto out_unlock; |
b8cf1913 HR |
2395 | } |
2396 | ||
927f11e1 | 2397 | if (file_opts->nocow) { |
4ab15590 | 2398 | #ifdef __linux__ |
06247428 HT |
2399 | /* Set NOCOW flag to solve performance issue on fs like btrfs. |
2400 | * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value | |
2401 | * will be ignored since any failure of this operation should not | |
2402 | * block the left work. | |
2403 | */ | |
2404 | int attr; | |
2405 | if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) { | |
2406 | attr |= FS_NOCOW_FL; | |
2407 | ioctl(fd, FS_IOC_SETFLAGS, &attr); | |
4ab15590 | 2408 | } |
06247428 HT |
2409 | #endif |
2410 | } | |
ffa244c8 KW |
2411 | #ifdef FS_IOC_FSSETXATTR |
2412 | /* | |
2413 | * Try to set the extent size hint. Failure is not fatal, and a warning is | |
2414 | * only printed if the option was explicitly specified. | |
2415 | */ | |
2416 | { | |
2417 | struct fsxattr attr; | |
2418 | result = ioctl(fd, FS_IOC_FSGETXATTR, &attr); | |
2419 | if (result == 0) { | |
2420 | attr.fsx_xflags |= FS_XFLAG_EXTSIZE; | |
2421 | attr.fsx_extsize = file_opts->extent_size_hint; | |
2422 | result = ioctl(fd, FS_IOC_FSSETXATTR, &attr); | |
2423 | } | |
2424 | if (result < 0 && file_opts->has_extent_size_hint && | |
2425 | file_opts->extent_size_hint) | |
2426 | { | |
2427 | warn_report("Failed to set extent size hint: %s", | |
2428 | strerror(errno)); | |
2429 | } | |
2430 | } | |
2431 | #endif | |
06247428 | 2432 | |
b8cf1913 HR |
2433 | /* Resize and potentially preallocate the file to the desired |
2434 | * final size */ | |
93f4e2ff KW |
2435 | result = raw_regular_truncate(NULL, fd, file_opts->size, |
2436 | file_opts->preallocation, errp); | |
9f63b07e | 2437 | if (result < 0) { |
7c20c808 HR |
2438 | goto out_unlock; |
2439 | } | |
2440 | ||
2441 | out_unlock: | |
2996ffad | 2442 | raw_apply_lock_bytes(NULL, fd, 0, 0, true, &local_err); |
7c20c808 HR |
2443 | if (local_err) { |
2444 | /* The above call should not fail, and if it does, that does | |
2445 | * not mean the whole creation operation has failed. So | |
2446 | * report it the user for their convenience, but do not report | |
2447 | * it to the caller. */ | |
db0754df | 2448 | warn_report_err(local_err); |
1e37d059 | 2449 | } |
06247428 | 2450 | |
5a1dad9d | 2451 | out_close: |
06247428 HT |
2452 | if (qemu_close(fd) != 0 && result == 0) { |
2453 | result = -errno; | |
2454 | error_setg_errno(errp, -result, "Could not close the new file"); | |
2455 | } | |
2456 | out: | |
1e37d059 | 2457 | return result; |
83f64091 FB |
2458 | } |
2459 | ||
b92902df ML |
2460 | static int coroutine_fn raw_co_create_opts(BlockDriver *drv, |
2461 | const char *filename, | |
2462 | QemuOpts *opts, | |
927f11e1 KW |
2463 | Error **errp) |
2464 | { | |
2465 | BlockdevCreateOptions options; | |
2466 | int64_t total_size = 0; | |
ffa244c8 KW |
2467 | int64_t extent_size_hint = 0; |
2468 | bool has_extent_size_hint = false; | |
927f11e1 KW |
2469 | bool nocow = false; |
2470 | PreallocMode prealloc; | |
2471 | char *buf = NULL; | |
2472 | Error *local_err = NULL; | |
2473 | ||
2474 | /* Skip file: protocol prefix */ | |
2475 | strstart(filename, "file:", &filename); | |
2476 | ||
2477 | /* Read out options */ | |
2478 | total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), | |
2479 | BDRV_SECTOR_SIZE); | |
ffa244c8 KW |
2480 | if (qemu_opt_get(opts, BLOCK_OPT_EXTENT_SIZE_HINT)) { |
2481 | has_extent_size_hint = true; | |
2482 | extent_size_hint = | |
2483 | qemu_opt_get_size_del(opts, BLOCK_OPT_EXTENT_SIZE_HINT, -1); | |
2484 | } | |
927f11e1 KW |
2485 | nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false); |
2486 | buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC); | |
2487 | prealloc = qapi_enum_parse(&PreallocMode_lookup, buf, | |
2488 | PREALLOC_MODE_OFF, &local_err); | |
2489 | g_free(buf); | |
2490 | if (local_err) { | |
2491 | error_propagate(errp, local_err); | |
2492 | return -EINVAL; | |
2493 | } | |
2494 | ||
2495 | options = (BlockdevCreateOptions) { | |
2496 | .driver = BLOCKDEV_DRIVER_FILE, | |
2497 | .u.file = { | |
2498 | .filename = (char *) filename, | |
2499 | .size = total_size, | |
2500 | .has_preallocation = true, | |
2501 | .preallocation = prealloc, | |
2502 | .has_nocow = true, | |
2503 | .nocow = nocow, | |
ffa244c8 KW |
2504 | .has_extent_size_hint = has_extent_size_hint, |
2505 | .extent_size_hint = extent_size_hint, | |
927f11e1 KW |
2506 | }, |
2507 | }; | |
2508 | return raw_co_create(&options, errp); | |
2509 | } | |
2510 | ||
9bffae14 DHB |
2511 | static int coroutine_fn raw_co_delete_file(BlockDriverState *bs, |
2512 | Error **errp) | |
2513 | { | |
2514 | struct stat st; | |
2515 | int ret; | |
2516 | ||
2517 | if (!(stat(bs->filename, &st) == 0) || !S_ISREG(st.st_mode)) { | |
2518 | error_setg_errno(errp, ENOENT, "%s is not a regular file", | |
2519 | bs->filename); | |
2520 | return -ENOENT; | |
2521 | } | |
2522 | ||
2523 | ret = unlink(bs->filename); | |
2524 | if (ret < 0) { | |
2525 | ret = -errno; | |
2526 | error_setg_errno(errp, -ret, "Error when deleting file %s", | |
2527 | bs->filename); | |
2528 | } | |
2529 | ||
2530 | return ret; | |
2531 | } | |
2532 | ||
d1f06fe6 MA |
2533 | /* |
2534 | * Find allocation range in @bs around offset @start. | |
2535 | * May change underlying file descriptor's file offset. | |
2536 | * If @start is not in a hole, store @start in @data, and the | |
2537 | * beginning of the next hole in @hole, and return 0. | |
2538 | * If @start is in a non-trailing hole, store @start in @hole and the | |
2539 | * beginning of the next non-hole in @data, and return 0. | |
2540 | * If @start is in a trailing hole or beyond EOF, return -ENXIO. | |
2541 | * If we can't find out, return a negative errno other than -ENXIO. | |
2542 | */ | |
2543 | static int find_allocation(BlockDriverState *bs, off_t start, | |
2544 | off_t *data, off_t *hole) | |
4f11aa8a HR |
2545 | { |
2546 | #if defined SEEK_HOLE && defined SEEK_DATA | |
94282e71 | 2547 | BDRVRawState *s = bs->opaque; |
d1f06fe6 | 2548 | off_t offs; |
94282e71 | 2549 | |
d1f06fe6 MA |
2550 | /* |
2551 | * SEEK_DATA cases: | |
2552 | * D1. offs == start: start is in data | |
2553 | * D2. offs > start: start is in a hole, next data at offs | |
2554 | * D3. offs < 0, errno = ENXIO: either start is in a trailing hole | |
2555 | * or start is beyond EOF | |
2556 | * If the latter happens, the file has been truncated behind | |
2557 | * our back since we opened it. All bets are off then. | |
2558 | * Treating like a trailing hole is simplest. | |
2559 | * D4. offs < 0, errno != ENXIO: we learned nothing | |
2560 | */ | |
2561 | offs = lseek(s->fd, start, SEEK_DATA); | |
2562 | if (offs < 0) { | |
2563 | return -errno; /* D3 or D4 */ | |
2564 | } | |
a03083a0 JC |
2565 | |
2566 | if (offs < start) { | |
2567 | /* This is not a valid return by lseek(). We are safe to just return | |
2568 | * -EIO in this case, and we'll treat it like D4. */ | |
2569 | return -EIO; | |
2570 | } | |
d1f06fe6 MA |
2571 | |
2572 | if (offs > start) { | |
2573 | /* D2: in hole, next data at offs */ | |
2574 | *hole = start; | |
2575 | *data = offs; | |
2576 | return 0; | |
5500316d PB |
2577 | } |
2578 | ||
d1f06fe6 MA |
2579 | /* D1: in data, end not yet known */ |
2580 | ||
2581 | /* | |
2582 | * SEEK_HOLE cases: | |
2583 | * H1. offs == start: start is in a hole | |
2584 | * If this happens here, a hole has been dug behind our back | |
2585 | * since the previous lseek(). | |
2586 | * H2. offs > start: either start is in data, next hole at offs, | |
2587 | * or start is in trailing hole, EOF at offs | |
2588 | * Linux treats trailing holes like any other hole: offs == | |
2589 | * start. Solaris seeks to EOF instead: offs > start (blech). | |
2590 | * If that happens here, a hole has been dug behind our back | |
2591 | * since the previous lseek(). | |
2592 | * H3. offs < 0, errno = ENXIO: start is beyond EOF | |
2593 | * If this happens, the file has been truncated behind our | |
2594 | * back since we opened it. Treat it like a trailing hole. | |
2595 | * H4. offs < 0, errno != ENXIO: we learned nothing | |
2596 | * Pretend we know nothing at all, i.e. "forget" about D1. | |
2597 | */ | |
2598 | offs = lseek(s->fd, start, SEEK_HOLE); | |
2599 | if (offs < 0) { | |
2600 | return -errno; /* D1 and (H3 or H4) */ | |
2601 | } | |
a03083a0 JC |
2602 | |
2603 | if (offs < start) { | |
2604 | /* This is not a valid return by lseek(). We are safe to just return | |
2605 | * -EIO in this case, and we'll treat it like H4. */ | |
2606 | return -EIO; | |
2607 | } | |
d1f06fe6 MA |
2608 | |
2609 | if (offs > start) { | |
2610 | /* | |
2611 | * D1 and H2: either in data, next hole at offs, or it was in | |
2612 | * data but is now in a trailing hole. In the latter case, | |
2613 | * all bets are off. Treating it as if it there was data all | |
2614 | * the way to EOF is safe, so simply do that. | |
2615 | */ | |
4f11aa8a | 2616 | *data = start; |
d1f06fe6 MA |
2617 | *hole = offs; |
2618 | return 0; | |
5500316d | 2619 | } |
4f11aa8a | 2620 | |
d1f06fe6 MA |
2621 | /* D1 and H1 */ |
2622 | return -EBUSY; | |
5500316d | 2623 | #else |
4f11aa8a | 2624 | return -ENOTSUP; |
5500316d | 2625 | #endif |
4f11aa8a HR |
2626 | } |
2627 | ||
2628 | /* | |
a290f085 | 2629 | * Returns the allocation status of the specified offset. |
4f11aa8a | 2630 | * |
a290f085 | 2631 | * The block layer guarantees 'offset' and 'bytes' are within bounds. |
4f11aa8a | 2632 | * |
a290f085 EB |
2633 | * 'pnum' is set to the number of bytes (including and immediately following |
2634 | * the specified offset) that are known to be in the same | |
4f11aa8a HR |
2635 | * allocated/unallocated state. |
2636 | * | |
a290f085 | 2637 | * 'bytes' is the max value 'pnum' should be set to. |
4f11aa8a | 2638 | */ |
a290f085 EB |
2639 | static int coroutine_fn raw_co_block_status(BlockDriverState *bs, |
2640 | bool want_zero, | |
2641 | int64_t offset, | |
2642 | int64_t bytes, int64_t *pnum, | |
2643 | int64_t *map, | |
2644 | BlockDriverState **file) | |
2645 | { | |
2646 | off_t data = 0, hole = 0; | |
d7f62751 | 2647 | int ret; |
4f11aa8a | 2648 | |
9c3db310 HR |
2649 | assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment)); |
2650 | ||
4f11aa8a HR |
2651 | ret = fd_open(bs); |
2652 | if (ret < 0) { | |
2653 | return ret; | |
2654 | } | |
2655 | ||
a290f085 EB |
2656 | if (!want_zero) { |
2657 | *pnum = bytes; | |
2658 | *map = offset; | |
2659 | *file = bs; | |
2660 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
e6d7ec32 | 2661 | } |
4f11aa8a | 2662 | |
a290f085 | 2663 | ret = find_allocation(bs, offset, &data, &hole); |
d1f06fe6 MA |
2664 | if (ret == -ENXIO) { |
2665 | /* Trailing hole */ | |
a290f085 | 2666 | *pnum = bytes; |
d1f06fe6 MA |
2667 | ret = BDRV_BLOCK_ZERO; |
2668 | } else if (ret < 0) { | |
2669 | /* No info available, so pretend there are no holes */ | |
a290f085 | 2670 | *pnum = bytes; |
d1f06fe6 | 2671 | ret = BDRV_BLOCK_DATA; |
a290f085 EB |
2672 | } else if (data == offset) { |
2673 | /* On a data extent, compute bytes to the end of the extent, | |
f4a769ab | 2674 | * possibly including a partial sector at EOF. */ |
a290f085 | 2675 | *pnum = MIN(bytes, hole - offset); |
9c3db310 HR |
2676 | |
2677 | /* | |
2678 | * We are not allowed to return partial sectors, though, so | |
2679 | * round up if necessary. | |
2680 | */ | |
2681 | if (!QEMU_IS_ALIGNED(*pnum, bs->bl.request_alignment)) { | |
2682 | int64_t file_length = raw_getlength(bs); | |
2683 | if (file_length > 0) { | |
2684 | /* Ignore errors, this is just a safeguard */ | |
2685 | assert(hole == file_length); | |
2686 | } | |
2687 | *pnum = ROUND_UP(*pnum, bs->bl.request_alignment); | |
2688 | } | |
2689 | ||
d1f06fe6 | 2690 | ret = BDRV_BLOCK_DATA; |
5500316d | 2691 | } else { |
a290f085 EB |
2692 | /* On a hole, compute bytes to the beginning of the next extent. */ |
2693 | assert(hole == offset); | |
2694 | *pnum = MIN(bytes, data - offset); | |
d1f06fe6 | 2695 | ret = BDRV_BLOCK_ZERO; |
5500316d | 2696 | } |
a290f085 | 2697 | *map = offset; |
02650acb | 2698 | *file = bs; |
a290f085 | 2699 | return ret | BDRV_BLOCK_OFFSET_VALID; |
5500316d PB |
2700 | } |
2701 | ||
31be8a2a SH |
2702 | #if defined(__linux__) |
2703 | /* Verify that the file is not in the page cache */ | |
2704 | static void check_cache_dropped(BlockDriverState *bs, Error **errp) | |
2705 | { | |
2706 | const size_t window_size = 128 * 1024 * 1024; | |
2707 | BDRVRawState *s = bs->opaque; | |
2708 | void *window = NULL; | |
2709 | size_t length = 0; | |
2710 | unsigned char *vec; | |
2711 | size_t page_size; | |
2712 | off_t offset; | |
2713 | off_t end; | |
2714 | ||
2715 | /* mincore(2) page status information requires 1 byte per page */ | |
2716 | page_size = sysconf(_SC_PAGESIZE); | |
2717 | vec = g_malloc(DIV_ROUND_UP(window_size, page_size)); | |
2718 | ||
2719 | end = raw_getlength(bs); | |
2720 | ||
2721 | for (offset = 0; offset < end; offset += window_size) { | |
2722 | void *new_window; | |
2723 | size_t new_length; | |
2724 | size_t vec_end; | |
2725 | size_t i; | |
2726 | int ret; | |
2727 | ||
2728 | /* Unmap previous window if size has changed */ | |
2729 | new_length = MIN(end - offset, window_size); | |
2730 | if (new_length != length) { | |
2731 | munmap(window, length); | |
2732 | window = NULL; | |
2733 | length = 0; | |
2734 | } | |
2735 | ||
2736 | new_window = mmap(window, new_length, PROT_NONE, MAP_PRIVATE, | |
2737 | s->fd, offset); | |
2738 | if (new_window == MAP_FAILED) { | |
2739 | error_setg_errno(errp, errno, "mmap failed"); | |
2740 | break; | |
2741 | } | |
2742 | ||
2743 | window = new_window; | |
2744 | length = new_length; | |
2745 | ||
2746 | ret = mincore(window, length, vec); | |
2747 | if (ret < 0) { | |
2748 | error_setg_errno(errp, errno, "mincore failed"); | |
2749 | break; | |
2750 | } | |
2751 | ||
2752 | vec_end = DIV_ROUND_UP(length, page_size); | |
2753 | for (i = 0; i < vec_end; i++) { | |
2754 | if (vec[i] & 0x1) { | |
31be8a2a SH |
2755 | break; |
2756 | } | |
2757 | } | |
77ed971b MA |
2758 | if (i < vec_end) { |
2759 | error_setg(errp, "page cache still in use!"); | |
2760 | break; | |
2761 | } | |
31be8a2a SH |
2762 | } |
2763 | ||
2764 | if (window) { | |
2765 | munmap(window, length); | |
2766 | } | |
2767 | ||
2768 | g_free(vec); | |
2769 | } | |
2770 | #endif /* __linux__ */ | |
2771 | ||
dd577a26 SH |
2772 | static void coroutine_fn raw_co_invalidate_cache(BlockDriverState *bs, |
2773 | Error **errp) | |
2774 | { | |
2775 | BDRVRawState *s = bs->opaque; | |
2776 | int ret; | |
2777 | ||
2778 | ret = fd_open(bs); | |
2779 | if (ret < 0) { | |
2780 | error_setg_errno(errp, -ret, "The file descriptor is not open"); | |
2781 | return; | |
2782 | } | |
2783 | ||
f357fcd8 SH |
2784 | if (!s->drop_cache) { |
2785 | return; | |
2786 | } | |
2787 | ||
dd577a26 SH |
2788 | if (s->open_flags & O_DIRECT) { |
2789 | return; /* No host kernel page cache */ | |
2790 | } | |
2791 | ||
2792 | #if defined(__linux__) | |
2793 | /* This sets the scene for the next syscall... */ | |
2794 | ret = bdrv_co_flush(bs); | |
2795 | if (ret < 0) { | |
2796 | error_setg_errno(errp, -ret, "flush failed"); | |
2797 | return; | |
2798 | } | |
2799 | ||
2800 | /* Linux does not invalidate pages that are dirty, locked, or mmapped by a | |
2801 | * process. These limitations are okay because we just fsynced the file, | |
2802 | * we don't use mmap, and the file should not be in use by other processes. | |
2803 | */ | |
2804 | ret = posix_fadvise(s->fd, 0, 0, POSIX_FADV_DONTNEED); | |
2805 | if (ret != 0) { /* the return value is a positive errno */ | |
2806 | error_setg_errno(errp, ret, "fadvise failed"); | |
2807 | return; | |
2808 | } | |
31be8a2a SH |
2809 | |
2810 | if (s->check_cache_dropped) { | |
2811 | check_cache_dropped(bs, errp); | |
2812 | } | |
dd577a26 SH |
2813 | #else /* __linux__ */ |
2814 | /* Do nothing. Live migration to a remote host with cache.direct=off is | |
2815 | * unsupported on other host operating systems. Cache consistency issues | |
2816 | * may occur but no error is reported here, partly because that's the | |
2817 | * historical behavior and partly because it's hard to differentiate valid | |
2818 | * configurations that should not cause errors. | |
2819 | */ | |
2820 | #endif /* !__linux__ */ | |
2821 | } | |
2822 | ||
1c450366 AN |
2823 | static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret) |
2824 | { | |
2825 | if (ret) { | |
2826 | s->stats.discard_nb_failed++; | |
2827 | } else { | |
2828 | s->stats.discard_nb_ok++; | |
2829 | s->stats.discard_bytes_ok += nbytes; | |
2830 | } | |
2831 | } | |
2832 | ||
33d70fb6 | 2833 | static coroutine_fn int |
46ee0f46 | 2834 | raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev) |
dce512de | 2835 | { |
dce512de | 2836 | BDRVRawState *s = bs->opaque; |
46ee0f46 | 2837 | RawPosixAIOData acb; |
1c450366 | 2838 | int ret; |
46ee0f46 KW |
2839 | |
2840 | acb = (RawPosixAIOData) { | |
2841 | .bs = bs, | |
2842 | .aio_fildes = s->fd, | |
2843 | .aio_type = QEMU_AIO_DISCARD, | |
2844 | .aio_offset = offset, | |
2845 | .aio_nbytes = bytes, | |
2846 | }; | |
dce512de | 2847 | |
46ee0f46 KW |
2848 | if (blkdev) { |
2849 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2850 | } | |
2851 | ||
1c450366 AN |
2852 | ret = raw_thread_pool_submit(bs, handle_aiocb_discard, &acb); |
2853 | raw_account_discard(s, bytes, ret); | |
2854 | return ret; | |
46ee0f46 KW |
2855 | } |
2856 | ||
2857 | static coroutine_fn int | |
2858 | raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
2859 | { | |
2860 | return raw_do_pdiscard(bs, offset, bytes, false); | |
dce512de | 2861 | } |
0e7e1989 | 2862 | |
7154d8ae KW |
2863 | static int coroutine_fn |
2864 | raw_do_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes, | |
2865 | BdrvRequestFlags flags, bool blkdev) | |
260a82e5 PB |
2866 | { |
2867 | BDRVRawState *s = bs->opaque; | |
7154d8ae KW |
2868 | RawPosixAIOData acb; |
2869 | ThreadPoolFunc *handler; | |
2870 | ||
292d06b9 HR |
2871 | #ifdef CONFIG_FALLOCATE |
2872 | if (offset + bytes > bs->total_sectors * BDRV_SECTOR_SIZE) { | |
2873 | BdrvTrackedRequest *req; | |
2874 | uint64_t end; | |
2875 | ||
2876 | /* | |
2877 | * This is a workaround for a bug in the Linux XFS driver, | |
2878 | * where writes submitted through the AIO interface will be | |
2879 | * discarded if they happen beyond a concurrently running | |
2880 | * fallocate() that increases the file length (i.e., both the | |
2881 | * write and the fallocate() happen beyond the EOF). | |
2882 | * | |
2883 | * To work around it, we extend the tracked request for this | |
2884 | * zero write until INT64_MAX (effectively infinity), and mark | |
2885 | * it as serializing. | |
2886 | * | |
2887 | * We have to enable this workaround for all filesystems and | |
2888 | * AIO modes (not just XFS with aio=native), because for | |
2889 | * remote filesystems we do not know the host configuration. | |
2890 | */ | |
2891 | ||
2892 | req = bdrv_co_get_self_request(bs); | |
2893 | assert(req); | |
2894 | assert(req->type == BDRV_TRACKED_WRITE); | |
2895 | assert(req->offset <= offset); | |
2896 | assert(req->offset + req->bytes >= offset + bytes); | |
2897 | ||
2898 | end = INT64_MAX & -(uint64_t)bs->bl.request_alignment; | |
2899 | req->bytes = end - req->offset; | |
2900 | req->overlap_bytes = req->bytes; | |
2901 | ||
2902 | bdrv_mark_request_serialising(req, bs->bl.request_alignment); | |
292d06b9 HR |
2903 | } |
2904 | #endif | |
2905 | ||
7154d8ae KW |
2906 | acb = (RawPosixAIOData) { |
2907 | .bs = bs, | |
2908 | .aio_fildes = s->fd, | |
2909 | .aio_type = QEMU_AIO_WRITE_ZEROES, | |
2910 | .aio_offset = offset, | |
2911 | .aio_nbytes = bytes, | |
2912 | }; | |
2913 | ||
2914 | if (blkdev) { | |
2915 | acb.aio_type |= QEMU_AIO_BLKDEV; | |
2916 | } | |
738301e1 KW |
2917 | if (flags & BDRV_REQ_NO_FALLBACK) { |
2918 | acb.aio_type |= QEMU_AIO_NO_FALLBACK; | |
2919 | } | |
260a82e5 | 2920 | |
34fa110e | 2921 | if (flags & BDRV_REQ_MAY_UNMAP) { |
7154d8ae KW |
2922 | acb.aio_type |= QEMU_AIO_DISCARD; |
2923 | handler = handle_aiocb_write_zeroes_unmap; | |
2924 | } else { | |
2925 | handler = handle_aiocb_write_zeroes; | |
260a82e5 | 2926 | } |
34fa110e | 2927 | |
7154d8ae KW |
2928 | return raw_thread_pool_submit(bs, handler, &acb); |
2929 | } | |
2930 | ||
2931 | static int coroutine_fn raw_co_pwrite_zeroes( | |
2932 | BlockDriverState *bs, int64_t offset, | |
2933 | int bytes, BdrvRequestFlags flags) | |
2934 | { | |
2935 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, false); | |
260a82e5 PB |
2936 | } |
2937 | ||
2938 | static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
2939 | { | |
260a82e5 PB |
2940 | return 0; |
2941 | } | |
2942 | ||
d9245599 AN |
2943 | static BlockStatsSpecificFile get_blockstats_specific_file(BlockDriverState *bs) |
2944 | { | |
2945 | BDRVRawState *s = bs->opaque; | |
2946 | return (BlockStatsSpecificFile) { | |
2947 | .discard_nb_ok = s->stats.discard_nb_ok, | |
2948 | .discard_nb_failed = s->stats.discard_nb_failed, | |
2949 | .discard_bytes_ok = s->stats.discard_bytes_ok, | |
2950 | }; | |
2951 | } | |
2952 | ||
2953 | static BlockStatsSpecific *raw_get_specific_stats(BlockDriverState *bs) | |
2954 | { | |
2955 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
2956 | ||
2957 | stats->driver = BLOCKDEV_DRIVER_FILE; | |
2958 | stats->u.file = get_blockstats_specific_file(bs); | |
2959 | ||
2960 | return stats; | |
2961 | } | |
2962 | ||
2963 | static BlockStatsSpecific *hdev_get_specific_stats(BlockDriverState *bs) | |
2964 | { | |
2965 | BlockStatsSpecific *stats = g_new(BlockStatsSpecific, 1); | |
2966 | ||
2967 | stats->driver = BLOCKDEV_DRIVER_HOST_DEVICE; | |
2968 | stats->u.host_device = get_blockstats_specific_file(bs); | |
2969 | ||
2970 | return stats; | |
2971 | } | |
2972 | ||
6f482f74 CL |
2973 | static QemuOptsList raw_create_opts = { |
2974 | .name = "raw-create-opts", | |
2975 | .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head), | |
2976 | .desc = { | |
2977 | { | |
2978 | .name = BLOCK_OPT_SIZE, | |
2979 | .type = QEMU_OPT_SIZE, | |
2980 | .help = "Virtual disk size" | |
2981 | }, | |
4ab15590 CL |
2982 | { |
2983 | .name = BLOCK_OPT_NOCOW, | |
2984 | .type = QEMU_OPT_BOOL, | |
2985 | .help = "Turn off copy-on-write (valid only on btrfs)" | |
2986 | }, | |
06247428 HT |
2987 | { |
2988 | .name = BLOCK_OPT_PREALLOC, | |
2989 | .type = QEMU_OPT_STRING, | |
abea0053 SG |
2990 | .help = "Preallocation mode (allowed values: off" |
2991 | #ifdef CONFIG_POSIX_FALLOCATE | |
2992 | ", falloc" | |
2993 | #endif | |
2994 | ", full)" | |
06247428 | 2995 | }, |
ffa244c8 KW |
2996 | { |
2997 | .name = BLOCK_OPT_EXTENT_SIZE_HINT, | |
2998 | .type = QEMU_OPT_SIZE, | |
2999 | .help = "Extent size hint for the image file, 0 to disable" | |
3000 | }, | |
6f482f74 CL |
3001 | { /* end of list */ } |
3002 | } | |
0e7e1989 KW |
3003 | }; |
3004 | ||
244a5668 FZ |
3005 | static int raw_check_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared, |
3006 | Error **errp) | |
3007 | { | |
6ceabe6f KW |
3008 | BDRVRawState *s = bs->opaque; |
3009 | BDRVRawReopenState *rs = NULL; | |
3010 | int open_flags; | |
3011 | int ret; | |
3012 | ||
3013 | if (s->perm_change_fd) { | |
3014 | /* | |
3015 | * In the context of reopen, this function may be called several times | |
3016 | * (directly and recursively while change permissions of the parent). | |
3017 | * This is even true for children that don't inherit from the original | |
3018 | * reopen node, so s->reopen_state is not set. | |
3019 | * | |
3020 | * Ignore all but the first call. | |
3021 | */ | |
3022 | return 0; | |
3023 | } | |
3024 | ||
3025 | if (s->reopen_state) { | |
3026 | /* We already have a new file descriptor to set permissions for */ | |
3027 | assert(s->reopen_state->perm == perm); | |
3028 | assert(s->reopen_state->shared_perm == shared); | |
3029 | rs = s->reopen_state->opaque; | |
3030 | s->perm_change_fd = rs->fd; | |
094e3639 | 3031 | s->perm_change_flags = rs->open_flags; |
6ceabe6f KW |
3032 | } else { |
3033 | /* We may need a new fd if auto-read-only switches the mode */ | |
23dece19 | 3034 | ret = raw_reconfigure_getfd(bs, bs->open_flags, &open_flags, perm, |
6ceabe6f KW |
3035 | false, errp); |
3036 | if (ret < 0) { | |
3037 | return ret; | |
3038 | } else if (ret != s->fd) { | |
3039 | s->perm_change_fd = ret; | |
094e3639 | 3040 | s->perm_change_flags = open_flags; |
6ceabe6f KW |
3041 | } |
3042 | } | |
3043 | ||
3044 | /* Prepare permissions on old fd to avoid conflicts between old and new, | |
3045 | * but keep everything locked that new will need. */ | |
3046 | ret = raw_handle_perm_lock(bs, RAW_PL_PREPARE, perm, shared, errp); | |
3047 | if (ret < 0) { | |
3048 | goto fail; | |
3049 | } | |
3050 | ||
3051 | /* Copy locks to the new fd */ | |
3052 | if (s->perm_change_fd) { | |
3053 | ret = raw_apply_lock_bytes(NULL, s->perm_change_fd, perm, ~shared, | |
3054 | false, errp); | |
3055 | if (ret < 0) { | |
3056 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); | |
3057 | goto fail; | |
3058 | } | |
3059 | } | |
3060 | return 0; | |
3061 | ||
3062 | fail: | |
3063 | if (s->perm_change_fd && !s->reopen_state) { | |
3064 | qemu_close(s->perm_change_fd); | |
3065 | } | |
3066 | s->perm_change_fd = 0; | |
3067 | return ret; | |
244a5668 FZ |
3068 | } |
3069 | ||
3070 | static void raw_set_perm(BlockDriverState *bs, uint64_t perm, uint64_t shared) | |
3071 | { | |
3072 | BDRVRawState *s = bs->opaque; | |
6ceabe6f KW |
3073 | |
3074 | /* For reopen, we have already switched to the new fd (.bdrv_set_perm is | |
3075 | * called after .bdrv_reopen_commit) */ | |
3076 | if (s->perm_change_fd && s->fd != s->perm_change_fd) { | |
3077 | qemu_close(s->fd); | |
3078 | s->fd = s->perm_change_fd; | |
094e3639 | 3079 | s->open_flags = s->perm_change_flags; |
6ceabe6f KW |
3080 | } |
3081 | s->perm_change_fd = 0; | |
3082 | ||
244a5668 FZ |
3083 | raw_handle_perm_lock(bs, RAW_PL_COMMIT, perm, shared, NULL); |
3084 | s->perm = perm; | |
3085 | s->shared_perm = shared; | |
3086 | } | |
3087 | ||
3088 | static void raw_abort_perm_update(BlockDriverState *bs) | |
3089 | { | |
6ceabe6f KW |
3090 | BDRVRawState *s = bs->opaque; |
3091 | ||
3092 | /* For reopen, .bdrv_reopen_abort is called afterwards and will close | |
3093 | * the file descriptor. */ | |
3094 | if (s->perm_change_fd && !s->reopen_state) { | |
3095 | qemu_close(s->perm_change_fd); | |
3096 | } | |
3097 | s->perm_change_fd = 0; | |
3098 | ||
244a5668 FZ |
3099 | raw_handle_perm_lock(bs, RAW_PL_ABORT, 0, 0, NULL); |
3100 | } | |
3101 | ||
67b51fb9 VSO |
3102 | static int coroutine_fn raw_co_copy_range_from( |
3103 | BlockDriverState *bs, BdrvChild *src, uint64_t src_offset, | |
3104 | BdrvChild *dst, uint64_t dst_offset, uint64_t bytes, | |
3105 | BdrvRequestFlags read_flags, BdrvRequestFlags write_flags) | |
1efad060 | 3106 | { |
67b51fb9 VSO |
3107 | return bdrv_co_copy_range_to(src, src_offset, dst, dst_offset, bytes, |
3108 | read_flags, write_flags); | |
1efad060 FZ |
3109 | } |
3110 | ||
3111 | static int coroutine_fn raw_co_copy_range_to(BlockDriverState *bs, | |
67b51fb9 VSO |
3112 | BdrvChild *src, |
3113 | uint64_t src_offset, | |
3114 | BdrvChild *dst, | |
3115 | uint64_t dst_offset, | |
3116 | uint64_t bytes, | |
3117 | BdrvRequestFlags read_flags, | |
3118 | BdrvRequestFlags write_flags) | |
1efad060 | 3119 | { |
58a209c4 | 3120 | RawPosixAIOData acb; |
1efad060 FZ |
3121 | BDRVRawState *s = bs->opaque; |
3122 | BDRVRawState *src_s; | |
3123 | ||
3124 | assert(dst->bs == bs); | |
3125 | if (src->bs->drv->bdrv_co_copy_range_to != raw_co_copy_range_to) { | |
3126 | return -ENOTSUP; | |
3127 | } | |
3128 | ||
3129 | src_s = src->bs->opaque; | |
9f850f67 | 3130 | if (fd_open(src->bs) < 0 || fd_open(dst->bs) < 0) { |
1efad060 FZ |
3131 | return -EIO; |
3132 | } | |
58a209c4 KW |
3133 | |
3134 | acb = (RawPosixAIOData) { | |
3135 | .bs = bs, | |
3136 | .aio_type = QEMU_AIO_COPY_RANGE, | |
3137 | .aio_fildes = src_s->fd, | |
3138 | .aio_offset = src_offset, | |
3139 | .aio_nbytes = bytes, | |
3140 | .copy_range = { | |
3141 | .aio_fd2 = s->fd, | |
3142 | .aio_offset2 = dst_offset, | |
3143 | }, | |
3144 | }; | |
3145 | ||
3146 | return raw_thread_pool_submit(bs, handle_aiocb_copy_range, &acb); | |
1efad060 FZ |
3147 | } |
3148 | ||
5f535a94 | 3149 | BlockDriver bdrv_file = { |
84a12e66 CH |
3150 | .format_name = "file", |
3151 | .protocol_name = "file", | |
856ae5c3 | 3152 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3153 | .bdrv_needs_filename = true, |
856ae5c3 | 3154 | .bdrv_probe = NULL, /* no probe for protocols */ |
078896a9 | 3155 | .bdrv_parse_filename = raw_parse_filename, |
66f82cee | 3156 | .bdrv_file_open = raw_open, |
eeb6b45d JC |
3157 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3158 | .bdrv_reopen_commit = raw_reopen_commit, | |
3159 | .bdrv_reopen_abort = raw_reopen_abort, | |
856ae5c3 | 3160 | .bdrv_close = raw_close, |
927f11e1 | 3161 | .bdrv_co_create = raw_co_create, |
efc75e2a | 3162 | .bdrv_co_create_opts = raw_co_create_opts, |
3ac21627 | 3163 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
a290f085 | 3164 | .bdrv_co_block_status = raw_co_block_status, |
dd577a26 | 3165 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3166 | .bdrv_co_pwrite_zeroes = raw_co_pwrite_zeroes, |
9bffae14 | 3167 | .bdrv_co_delete_file = raw_co_delete_file, |
3b46e624 | 3168 | |
9d52aa3c KW |
3169 | .bdrv_co_preadv = raw_co_preadv, |
3170 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3171 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3172 | .bdrv_co_pdiscard = raw_co_pdiscard, | |
1efad060 FZ |
3173 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3174 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3175 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3176 | .bdrv_io_plug = raw_aio_plug, |
3177 | .bdrv_io_unplug = raw_aio_unplug, | |
ed6e2161 | 3178 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3179 | |
061ca8a3 | 3180 | .bdrv_co_truncate = raw_co_truncate, |
83f64091 | 3181 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3182 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3183 | .bdrv_get_allocated_file_size |
3184 | = raw_get_allocated_file_size, | |
d9245599 | 3185 | .bdrv_get_specific_stats = raw_get_specific_stats, |
244a5668 FZ |
3186 | .bdrv_check_perm = raw_check_perm, |
3187 | .bdrv_set_perm = raw_set_perm, | |
3188 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
6f482f74 | 3189 | .create_opts = &raw_create_opts, |
8a2ce0bc | 3190 | .mutable_opts = mutable_opts, |
83f64091 FB |
3191 | }; |
3192 | ||
19cb3738 FB |
3193 | /***********************************************/ |
3194 | /* host device */ | |
3195 | ||
83affaa6 | 3196 | #if defined(__APPLE__) && defined(__MACH__) |
98caa5bc JA |
3197 | static kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3198 | CFIndex maxPathSize, int flags); | |
d0855f12 | 3199 | static char *FindEjectableOpticalMedia(io_iterator_t *mediaIterator) |
19cb3738 | 3200 | { |
d0855f12 | 3201 | kern_return_t kernResult = KERN_FAILURE; |
19cb3738 FB |
3202 | mach_port_t masterPort; |
3203 | CFMutableDictionaryRef classesToMatch; | |
d0855f12 JA |
3204 | const char *matching_array[] = {kIODVDMediaClass, kIOCDMediaClass}; |
3205 | char *mediaType = NULL; | |
19cb3738 FB |
3206 | |
3207 | kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort ); | |
3208 | if ( KERN_SUCCESS != kernResult ) { | |
3209 | printf( "IOMasterPort returned %d\n", kernResult ); | |
3210 | } | |
3b46e624 | 3211 | |
d0855f12 JA |
3212 | int index; |
3213 | for (index = 0; index < ARRAY_SIZE(matching_array); index++) { | |
3214 | classesToMatch = IOServiceMatching(matching_array[index]); | |
3215 | if (classesToMatch == NULL) { | |
3216 | error_report("IOServiceMatching returned NULL for %s", | |
3217 | matching_array[index]); | |
3218 | continue; | |
3219 | } | |
3220 | CFDictionarySetValue(classesToMatch, CFSTR(kIOMediaEjectableKey), | |
3221 | kCFBooleanTrue); | |
3222 | kernResult = IOServiceGetMatchingServices(masterPort, classesToMatch, | |
3223 | mediaIterator); | |
3224 | if (kernResult != KERN_SUCCESS) { | |
3225 | error_report("Note: IOServiceGetMatchingServices returned %d", | |
3226 | kernResult); | |
3227 | continue; | |
3228 | } | |
3b46e624 | 3229 | |
d0855f12 JA |
3230 | /* If a match was found, leave the loop */ |
3231 | if (*mediaIterator != 0) { | |
4f7d28d7 | 3232 | trace_file_FindEjectableOpticalMedia(matching_array[index]); |
d0855f12 JA |
3233 | mediaType = g_strdup(matching_array[index]); |
3234 | break; | |
3235 | } | |
3236 | } | |
3237 | return mediaType; | |
19cb3738 FB |
3238 | } |
3239 | ||
98caa5bc JA |
3240 | kern_return_t GetBSDPath(io_iterator_t mediaIterator, char *bsdPath, |
3241 | CFIndex maxPathSize, int flags) | |
19cb3738 FB |
3242 | { |
3243 | io_object_t nextMedia; | |
3244 | kern_return_t kernResult = KERN_FAILURE; | |
3245 | *bsdPath = '\0'; | |
3246 | nextMedia = IOIteratorNext( mediaIterator ); | |
3247 | if ( nextMedia ) | |
3248 | { | |
3249 | CFTypeRef bsdPathAsCFString; | |
3250 | bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 ); | |
3251 | if ( bsdPathAsCFString ) { | |
3252 | size_t devPathLength; | |
3253 | strcpy( bsdPath, _PATH_DEV ); | |
98caa5bc JA |
3254 | if (flags & BDRV_O_NOCACHE) { |
3255 | strcat(bsdPath, "r"); | |
3256 | } | |
19cb3738 FB |
3257 | devPathLength = strlen( bsdPath ); |
3258 | if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) { | |
3259 | kernResult = KERN_SUCCESS; | |
3260 | } | |
3261 | CFRelease( bsdPathAsCFString ); | |
3262 | } | |
3263 | IOObjectRelease( nextMedia ); | |
3264 | } | |
3b46e624 | 3265 | |
19cb3738 FB |
3266 | return kernResult; |
3267 | } | |
3268 | ||
d0855f12 JA |
3269 | /* Sets up a real cdrom for use in QEMU */ |
3270 | static bool setup_cdrom(char *bsd_path, Error **errp) | |
3271 | { | |
3272 | int index, num_of_test_partitions = 2, fd; | |
3273 | char test_partition[MAXPATHLEN]; | |
3274 | bool partition_found = false; | |
3275 | ||
3276 | /* look for a working partition */ | |
3277 | for (index = 0; index < num_of_test_partitions; index++) { | |
3278 | snprintf(test_partition, sizeof(test_partition), "%ss%d", bsd_path, | |
3279 | index); | |
3280 | fd = qemu_open(test_partition, O_RDONLY | O_BINARY | O_LARGEFILE); | |
3281 | if (fd >= 0) { | |
3282 | partition_found = true; | |
3283 | qemu_close(fd); | |
3284 | break; | |
3285 | } | |
3286 | } | |
3287 | ||
3288 | /* if a working partition on the device was not found */ | |
3289 | if (partition_found == false) { | |
3290 | error_setg(errp, "Failed to find a working partition on disc"); | |
3291 | } else { | |
4f7d28d7 | 3292 | trace_file_setup_cdrom(test_partition); |
d0855f12 JA |
3293 | pstrcpy(bsd_path, MAXPATHLEN, test_partition); |
3294 | } | |
3295 | return partition_found; | |
3296 | } | |
3297 | ||
3298 | /* Prints directions on mounting and unmounting a device */ | |
3299 | static void print_unmounting_directions(const char *file_name) | |
3300 | { | |
3301 | error_report("If device %s is mounted on the desktop, unmount" | |
3302 | " it first before using it in QEMU", file_name); | |
3303 | error_report("Command to unmount device: diskutil unmountDisk %s", | |
3304 | file_name); | |
3305 | error_report("Command to mount device: diskutil mountDisk %s", file_name); | |
3306 | } | |
3307 | ||
3308 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
19cb3738 | 3309 | |
508c7cb3 CH |
3310 | static int hdev_probe_device(const char *filename) |
3311 | { | |
3312 | struct stat st; | |
3313 | ||
3314 | /* allow a dedicated CD-ROM driver to match with a higher priority */ | |
3315 | if (strstart(filename, "/dev/cdrom", NULL)) | |
3316 | return 50; | |
3317 | ||
3318 | if (stat(filename, &st) >= 0 && | |
3319 | (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) { | |
3320 | return 100; | |
3321 | } | |
3322 | ||
3323 | return 0; | |
3324 | } | |
3325 | ||
da888d37 SH |
3326 | static int check_hdev_writable(BDRVRawState *s) |
3327 | { | |
3328 | #if defined(BLKROGET) | |
3329 | /* Linux block devices can be configured "read-only" using blockdev(8). | |
3330 | * This is independent of device node permissions and therefore open(2) | |
3331 | * with O_RDWR succeeds. Actual writes fail with EPERM. | |
3332 | * | |
3333 | * bdrv_open() is supposed to fail if the disk is read-only. Explicitly | |
3334 | * check for read-only block devices so that Linux block devices behave | |
3335 | * properly. | |
3336 | */ | |
3337 | struct stat st; | |
3338 | int readonly = 0; | |
3339 | ||
3340 | if (fstat(s->fd, &st)) { | |
3341 | return -errno; | |
3342 | } | |
3343 | ||
3344 | if (!S_ISBLK(st.st_mode)) { | |
3345 | return 0; | |
3346 | } | |
3347 | ||
3348 | if (ioctl(s->fd, BLKROGET, &readonly) < 0) { | |
3349 | return -errno; | |
3350 | } | |
3351 | ||
3352 | if (readonly) { | |
3353 | return -EACCES; | |
3354 | } | |
3355 | #endif /* defined(BLKROGET) */ | |
3356 | return 0; | |
3357 | } | |
3358 | ||
7af803d4 HR |
3359 | static void hdev_parse_filename(const char *filename, QDict *options, |
3360 | Error **errp) | |
3361 | { | |
03c320d8 | 3362 | bdrv_parse_filename_strip_prefix(filename, "host_device:", options); |
7af803d4 HR |
3363 | } |
3364 | ||
3307ed7b DA |
3365 | static bool hdev_is_sg(BlockDriverState *bs) |
3366 | { | |
3367 | ||
3368 | #if defined(__linux__) | |
3369 | ||
0d4377b3 | 3370 | BDRVRawState *s = bs->opaque; |
3307ed7b DA |
3371 | struct stat st; |
3372 | struct sg_scsi_id scsiid; | |
3373 | int sg_version; | |
0d4377b3 KW |
3374 | int ret; |
3375 | ||
3376 | if (stat(bs->filename, &st) < 0 || !S_ISCHR(st.st_mode)) { | |
3377 | return false; | |
3378 | } | |
3307ed7b | 3379 | |
0d4377b3 KW |
3380 | ret = ioctl(s->fd, SG_GET_VERSION_NUM, &sg_version); |
3381 | if (ret < 0) { | |
3382 | return false; | |
3383 | } | |
3384 | ||
3385 | ret = ioctl(s->fd, SG_GET_SCSI_ID, &scsiid); | |
3386 | if (ret >= 0) { | |
4f7d28d7 | 3387 | trace_file_hdev_is_sg(scsiid.scsi_type, sg_version); |
3307ed7b DA |
3388 | return true; |
3389 | } | |
3390 | ||
3391 | #endif | |
3392 | ||
3393 | return false; | |
3394 | } | |
3395 | ||
015a1036 HR |
3396 | static int hdev_open(BlockDriverState *bs, QDict *options, int flags, |
3397 | Error **errp) | |
19cb3738 FB |
3398 | { |
3399 | BDRVRawState *s = bs->opaque; | |
da888d37 | 3400 | int ret; |
a76bab49 | 3401 | |
83affaa6 | 3402 | #if defined(__APPLE__) && defined(__MACH__) |
129c7d1c MA |
3403 | /* |
3404 | * Caution: while qdict_get_str() is fine, getting non-string types | |
3405 | * would require more care. When @options come from -blockdev or | |
3406 | * blockdev_add, its members are typed according to the QAPI | |
3407 | * schema, but when they come from -drive, they're all QString. | |
3408 | */ | |
3307ed7b | 3409 | const char *filename = qdict_get_str(options, "filename"); |
d0855f12 JA |
3410 | char bsd_path[MAXPATHLEN] = ""; |
3411 | bool error_occurred = false; | |
3412 | ||
3413 | /* If using a real cdrom */ | |
3414 | if (strcmp(filename, "/dev/cdrom") == 0) { | |
3415 | char *mediaType = NULL; | |
3416 | kern_return_t ret_val; | |
3417 | io_iterator_t mediaIterator = 0; | |
3418 | ||
3419 | mediaType = FindEjectableOpticalMedia(&mediaIterator); | |
3420 | if (mediaType == NULL) { | |
3421 | error_setg(errp, "Please make sure your CD/DVD is in the optical" | |
3422 | " drive"); | |
3423 | error_occurred = true; | |
3424 | goto hdev_open_Mac_error; | |
3425 | } | |
3307ed7b | 3426 | |
d0855f12 JA |
3427 | ret_val = GetBSDPath(mediaIterator, bsd_path, sizeof(bsd_path), flags); |
3428 | if (ret_val != KERN_SUCCESS) { | |
3429 | error_setg(errp, "Could not get BSD path for optical drive"); | |
3430 | error_occurred = true; | |
3431 | goto hdev_open_Mac_error; | |
3432 | } | |
3433 | ||
3434 | /* If a real optical drive was not found */ | |
3435 | if (bsd_path[0] == '\0') { | |
3436 | error_setg(errp, "Failed to obtain bsd path for optical drive"); | |
3437 | error_occurred = true; | |
3438 | goto hdev_open_Mac_error; | |
3439 | } | |
3440 | ||
3441 | /* If using a cdrom disc and finding a partition on the disc failed */ | |
3442 | if (strncmp(mediaType, kIOCDMediaClass, 9) == 0 && | |
3443 | setup_cdrom(bsd_path, errp) == false) { | |
3444 | print_unmounting_directions(bsd_path); | |
3445 | error_occurred = true; | |
3446 | goto hdev_open_Mac_error; | |
19cb3738 | 3447 | } |
3b46e624 | 3448 | |
46f5ac20 | 3449 | qdict_put_str(options, "filename", bsd_path); |
d0855f12 JA |
3450 | |
3451 | hdev_open_Mac_error: | |
3452 | g_free(mediaType); | |
3453 | if (mediaIterator) { | |
3454 | IOObjectRelease(mediaIterator); | |
3455 | } | |
3456 | if (error_occurred) { | |
3457 | return -ENOENT; | |
3458 | } | |
19cb3738 | 3459 | } |
d0855f12 | 3460 | #endif /* defined(__APPLE__) && defined(__MACH__) */ |
19cb3738 FB |
3461 | |
3462 | s->type = FTYPE_FILE; | |
90babde0 | 3463 | |
668f62ec | 3464 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
da888d37 | 3465 | if (ret < 0) { |
d0855f12 JA |
3466 | #if defined(__APPLE__) && defined(__MACH__) |
3467 | if (*bsd_path) { | |
3468 | filename = bsd_path; | |
3469 | } | |
3470 | /* if a physical device experienced an error while being opened */ | |
3471 | if (strncmp(filename, "/dev/", 5) == 0) { | |
3472 | print_unmounting_directions(filename); | |
3473 | } | |
3474 | #endif /* defined(__APPLE__) && defined(__MACH__) */ | |
da888d37 SH |
3475 | return ret; |
3476 | } | |
3477 | ||
3307ed7b DA |
3478 | /* Since this does ioctl the device must be already opened */ |
3479 | bs->sg = hdev_is_sg(bs); | |
3480 | ||
da888d37 SH |
3481 | if (flags & BDRV_O_RDWR) { |
3482 | ret = check_hdev_writable(s); | |
3483 | if (ret < 0) { | |
3484 | raw_close(bs); | |
e428e439 | 3485 | error_setg_errno(errp, -ret, "The device is not writable"); |
da888d37 SH |
3486 | return ret; |
3487 | } | |
3488 | } | |
3489 | ||
3490 | return ret; | |
19cb3738 FB |
3491 | } |
3492 | ||
03ff3ca3 | 3493 | #if defined(__linux__) |
2f3a7ab3 KW |
3494 | static int coroutine_fn |
3495 | hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
221f715d | 3496 | { |
f141eafe | 3497 | BDRVRawState *s = bs->opaque; |
03425671 | 3498 | RawPosixAIOData acb; |
2f3a7ab3 | 3499 | int ret; |
221f715d | 3500 | |
2f3a7ab3 KW |
3501 | ret = fd_open(bs); |
3502 | if (ret < 0) { | |
3503 | return ret; | |
3504 | } | |
c208e8c2 | 3505 | |
7c9e5276 PB |
3506 | if (req == SG_IO && s->pr_mgr) { |
3507 | struct sg_io_hdr *io_hdr = buf; | |
3508 | if (io_hdr->cmdp[0] == PERSISTENT_RESERVE_OUT || | |
3509 | io_hdr->cmdp[0] == PERSISTENT_RESERVE_IN) { | |
3510 | return pr_manager_execute(s->pr_mgr, bdrv_get_aio_context(bs), | |
2f3a7ab3 | 3511 | s->fd, io_hdr); |
7c9e5276 PB |
3512 | } |
3513 | } | |
3514 | ||
03425671 KW |
3515 | acb = (RawPosixAIOData) { |
3516 | .bs = bs, | |
3517 | .aio_type = QEMU_AIO_IOCTL, | |
3518 | .aio_fildes = s->fd, | |
3519 | .aio_offset = 0, | |
3520 | .ioctl = { | |
3521 | .buf = buf, | |
3522 | .cmd = req, | |
3523 | }, | |
3524 | }; | |
3525 | ||
3526 | return raw_thread_pool_submit(bs, handle_aiocb_ioctl, &acb); | |
221f715d | 3527 | } |
f709623b | 3528 | #endif /* linux */ |
221f715d | 3529 | |
9f23011a BS |
3530 | static int fd_open(BlockDriverState *bs) |
3531 | { | |
3532 | BDRVRawState *s = bs->opaque; | |
3533 | ||
3534 | /* this is just to ensure s->fd is sane (its called by io ops) */ | |
3535 | if (s->fd >= 0) | |
3536 | return 0; | |
3537 | return -EIO; | |
3538 | } | |
04eeb8b6 | 3539 | |
33d70fb6 KW |
3540 | static coroutine_fn int |
3541 | hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes) | |
c36dd8a0 | 3542 | { |
1c450366 | 3543 | BDRVRawState *s = bs->opaque; |
33d70fb6 | 3544 | int ret; |
c36dd8a0 | 3545 | |
33d70fb6 KW |
3546 | ret = fd_open(bs); |
3547 | if (ret < 0) { | |
1c450366 | 3548 | raw_account_discard(s, bytes, ret); |
33d70fb6 | 3549 | return ret; |
c36dd8a0 | 3550 | } |
46ee0f46 | 3551 | return raw_do_pdiscard(bs, offset, bytes, true); |
c36dd8a0 AF |
3552 | } |
3553 | ||
2ffa76c2 | 3554 | static coroutine_fn int hdev_co_pwrite_zeroes(BlockDriverState *bs, |
f5a5ca79 | 3555 | int64_t offset, int bytes, BdrvRequestFlags flags) |
d0b4503e | 3556 | { |
d0b4503e PB |
3557 | int rc; |
3558 | ||
3559 | rc = fd_open(bs); | |
3560 | if (rc < 0) { | |
3561 | return rc; | |
3562 | } | |
34fa110e | 3563 | |
7154d8ae | 3564 | return raw_do_pwrite_zeroes(bs, offset, bytes, flags, true); |
d0b4503e PB |
3565 | } |
3566 | ||
5efa9d5a | 3567 | static BlockDriver bdrv_host_device = { |
0b4ce02e | 3568 | .format_name = "host_device", |
84a12e66 | 3569 | .protocol_name = "host_device", |
0b4ce02e | 3570 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3571 | .bdrv_needs_filename = true, |
0b4ce02e | 3572 | .bdrv_probe_device = hdev_probe_device, |
7af803d4 | 3573 | .bdrv_parse_filename = hdev_parse_filename, |
66f82cee | 3574 | .bdrv_file_open = hdev_open, |
0b4ce02e | 3575 | .bdrv_close = raw_close, |
1bc6b705 JC |
3576 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3577 | .bdrv_reopen_commit = raw_reopen_commit, | |
3578 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3579 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3580 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3581 | .mutable_opts = mutable_opts, |
dd577a26 | 3582 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
2ffa76c2 | 3583 | .bdrv_co_pwrite_zeroes = hdev_co_pwrite_zeroes, |
3b46e624 | 3584 | |
9d52aa3c KW |
3585 | .bdrv_co_preadv = raw_co_preadv, |
3586 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 KW |
3587 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
3588 | .bdrv_co_pdiscard = hdev_co_pdiscard, | |
1efad060 FZ |
3589 | .bdrv_co_copy_range_from = raw_co_copy_range_from, |
3590 | .bdrv_co_copy_range_to = raw_co_copy_range_to, | |
c25f53b0 | 3591 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3592 | .bdrv_io_plug = raw_aio_plug, |
3593 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3594 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
3c529d93 | 3595 | |
061ca8a3 | 3596 | .bdrv_co_truncate = raw_co_truncate, |
e60f469c | 3597 | .bdrv_getlength = raw_getlength, |
260a82e5 | 3598 | .bdrv_get_info = raw_get_info, |
4a1d5e1f FZ |
3599 | .bdrv_get_allocated_file_size |
3600 | = raw_get_allocated_file_size, | |
d9245599 | 3601 | .bdrv_get_specific_stats = hdev_get_specific_stats, |
244a5668 FZ |
3602 | .bdrv_check_perm = raw_check_perm, |
3603 | .bdrv_set_perm = raw_set_perm, | |
3604 | .bdrv_abort_perm_update = raw_abort_perm_update, | |
1a9335e4 ET |
3605 | .bdrv_probe_blocksizes = hdev_probe_blocksizes, |
3606 | .bdrv_probe_geometry = hdev_probe_geometry, | |
19cb3738 | 3607 | |
f3a5d3f8 | 3608 | /* generic scsi device */ |
63ec93db | 3609 | #ifdef __linux__ |
2f3a7ab3 | 3610 | .bdrv_co_ioctl = hdev_co_ioctl, |
63ec93db | 3611 | #endif |
f3a5d3f8 CH |
3612 | }; |
3613 | ||
18fa1c42 HR |
3614 | #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
3615 | static void cdrom_parse_filename(const char *filename, QDict *options, | |
3616 | Error **errp) | |
3617 | { | |
03c320d8 | 3618 | bdrv_parse_filename_strip_prefix(filename, "host_cdrom:", options); |
18fa1c42 HR |
3619 | } |
3620 | #endif | |
3621 | ||
3622 | #ifdef __linux__ | |
015a1036 HR |
3623 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3624 | Error **errp) | |
f3a5d3f8 CH |
3625 | { |
3626 | BDRVRawState *s = bs->opaque; | |
3627 | ||
f3a5d3f8 CH |
3628 | s->type = FTYPE_CD; |
3629 | ||
19a3da7f | 3630 | /* open will not fail even if no CD is inserted, so add O_NONBLOCK */ |
230ff739 | 3631 | return raw_open_common(bs, options, flags, O_NONBLOCK, true, errp); |
f3a5d3f8 CH |
3632 | } |
3633 | ||
508c7cb3 CH |
3634 | static int cdrom_probe_device(const char *filename) |
3635 | { | |
3baf720e CR |
3636 | int fd, ret; |
3637 | int prio = 0; | |
343f8568 | 3638 | struct stat st; |
3baf720e | 3639 | |
6165f4d8 | 3640 | fd = qemu_open(filename, O_RDONLY | O_NONBLOCK); |
3baf720e CR |
3641 | if (fd < 0) { |
3642 | goto out; | |
3643 | } | |
343f8568 JS |
3644 | ret = fstat(fd, &st); |
3645 | if (ret == -1 || !S_ISBLK(st.st_mode)) { | |
3646 | goto outc; | |
3647 | } | |
3baf720e CR |
3648 | |
3649 | /* Attempt to detect via a CDROM specific ioctl */ | |
3650 | ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
3651 | if (ret >= 0) | |
3652 | prio = 100; | |
3653 | ||
343f8568 | 3654 | outc: |
2e1e79da | 3655 | qemu_close(fd); |
3baf720e CR |
3656 | out: |
3657 | return prio; | |
508c7cb3 CH |
3658 | } |
3659 | ||
e031f750 | 3660 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3661 | { |
3662 | BDRVRawState *s = bs->opaque; | |
3663 | int ret; | |
3664 | ||
3665 | ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT); | |
e031f750 | 3666 | return ret == CDS_DISC_OK; |
f3a5d3f8 CH |
3667 | } |
3668 | ||
f36f3949 | 3669 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3670 | { |
3671 | BDRVRawState *s = bs->opaque; | |
3672 | ||
3673 | if (eject_flag) { | |
3674 | if (ioctl(s->fd, CDROMEJECT, NULL) < 0) | |
3675 | perror("CDROMEJECT"); | |
3676 | } else { | |
3677 | if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0) | |
3678 | perror("CDROMEJECT"); | |
3679 | } | |
f3a5d3f8 CH |
3680 | } |
3681 | ||
025e849a | 3682 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3683 | { |
3684 | BDRVRawState *s = bs->opaque; | |
3685 | ||
3686 | if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) { | |
3687 | /* | |
3688 | * Note: an error can happen if the distribution automatically | |
3689 | * mounts the CD-ROM | |
3690 | */ | |
3691 | /* perror("CDROM_LOCKDOOR"); */ | |
3692 | } | |
f3a5d3f8 CH |
3693 | } |
3694 | ||
3695 | static BlockDriver bdrv_host_cdrom = { | |
3696 | .format_name = "host_cdrom", | |
84a12e66 | 3697 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3698 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3699 | .bdrv_needs_filename = true, |
508c7cb3 | 3700 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3701 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3702 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3703 | .bdrv_close = raw_close, |
1bc6b705 JC |
3704 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3705 | .bdrv_reopen_commit = raw_reopen_commit, | |
3706 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3707 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3708 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3709 | .mutable_opts = mutable_opts, |
dd577a26 | 3710 | .bdrv_co_invalidate_cache = raw_co_invalidate_cache, |
f3a5d3f8 | 3711 | |
9d52aa3c KW |
3712 | .bdrv_co_preadv = raw_co_preadv, |
3713 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3714 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3715 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3716 | .bdrv_io_plug = raw_aio_plug, |
3717 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3718 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3719 | |
061ca8a3 | 3720 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3721 | .bdrv_getlength = raw_getlength, |
3722 | .has_variable_length = true, | |
4a1d5e1f FZ |
3723 | .bdrv_get_allocated_file_size |
3724 | = raw_get_allocated_file_size, | |
f3a5d3f8 CH |
3725 | |
3726 | /* removable device support */ | |
3727 | .bdrv_is_inserted = cdrom_is_inserted, | |
3728 | .bdrv_eject = cdrom_eject, | |
025e849a | 3729 | .bdrv_lock_medium = cdrom_lock_medium, |
f3a5d3f8 CH |
3730 | |
3731 | /* generic scsi device */ | |
2f3a7ab3 | 3732 | .bdrv_co_ioctl = hdev_co_ioctl, |
f3a5d3f8 CH |
3733 | }; |
3734 | #endif /* __linux__ */ | |
3735 | ||
a167ba50 | 3736 | #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__) |
511018e4 AT |
3737 | static int cdrom_open(BlockDriverState *bs, QDict *options, int flags, |
3738 | Error **errp) | |
f3a5d3f8 CH |
3739 | { |
3740 | BDRVRawState *s = bs->opaque; | |
3741 | int ret; | |
3742 | ||
3743 | s->type = FTYPE_CD; | |
3744 | ||
668f62ec | 3745 | ret = raw_open_common(bs, options, flags, 0, true, errp); |
e428e439 | 3746 | if (ret) { |
f3a5d3f8 | 3747 | return ret; |
e428e439 | 3748 | } |
f3a5d3f8 | 3749 | |
9b2260cb | 3750 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3751 | ioctl(s->fd, CDIOCALLOW); |
3752 | return 0; | |
3753 | } | |
3754 | ||
508c7cb3 CH |
3755 | static int cdrom_probe_device(const char *filename) |
3756 | { | |
3757 | if (strstart(filename, "/dev/cd", NULL) || | |
3758 | strstart(filename, "/dev/acd", NULL)) | |
3759 | return 100; | |
3760 | return 0; | |
3761 | } | |
3762 | ||
f3a5d3f8 CH |
3763 | static int cdrom_reopen(BlockDriverState *bs) |
3764 | { | |
3765 | BDRVRawState *s = bs->opaque; | |
3766 | int fd; | |
3767 | ||
3768 | /* | |
3769 | * Force reread of possibly changed/newly loaded disc, | |
3770 | * FreeBSD seems to not notice sometimes... | |
3771 | */ | |
3772 | if (s->fd >= 0) | |
2e1e79da | 3773 | qemu_close(s->fd); |
6165f4d8 | 3774 | fd = qemu_open(bs->filename, s->open_flags, 0644); |
f3a5d3f8 CH |
3775 | if (fd < 0) { |
3776 | s->fd = -1; | |
3777 | return -EIO; | |
3778 | } | |
3779 | s->fd = fd; | |
3780 | ||
9b2260cb | 3781 | /* make sure the door isn't locked at this time */ |
f3a5d3f8 CH |
3782 | ioctl(s->fd, CDIOCALLOW); |
3783 | return 0; | |
3784 | } | |
3785 | ||
e031f750 | 3786 | static bool cdrom_is_inserted(BlockDriverState *bs) |
f3a5d3f8 CH |
3787 | { |
3788 | return raw_getlength(bs) > 0; | |
3789 | } | |
3790 | ||
f36f3949 | 3791 | static void cdrom_eject(BlockDriverState *bs, bool eject_flag) |
f3a5d3f8 CH |
3792 | { |
3793 | BDRVRawState *s = bs->opaque; | |
3794 | ||
3795 | if (s->fd < 0) | |
822e1cd1 | 3796 | return; |
f3a5d3f8 CH |
3797 | |
3798 | (void) ioctl(s->fd, CDIOCALLOW); | |
3799 | ||
3800 | if (eject_flag) { | |
3801 | if (ioctl(s->fd, CDIOCEJECT) < 0) | |
3802 | perror("CDIOCEJECT"); | |
3803 | } else { | |
3804 | if (ioctl(s->fd, CDIOCCLOSE) < 0) | |
3805 | perror("CDIOCCLOSE"); | |
3806 | } | |
3807 | ||
822e1cd1 | 3808 | cdrom_reopen(bs); |
f3a5d3f8 CH |
3809 | } |
3810 | ||
025e849a | 3811 | static void cdrom_lock_medium(BlockDriverState *bs, bool locked) |
f3a5d3f8 CH |
3812 | { |
3813 | BDRVRawState *s = bs->opaque; | |
3814 | ||
3815 | if (s->fd < 0) | |
7bf37fed | 3816 | return; |
f3a5d3f8 CH |
3817 | if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) { |
3818 | /* | |
3819 | * Note: an error can happen if the distribution automatically | |
3820 | * mounts the CD-ROM | |
3821 | */ | |
3822 | /* perror("CDROM_LOCKDOOR"); */ | |
3823 | } | |
f3a5d3f8 CH |
3824 | } |
3825 | ||
3826 | static BlockDriver bdrv_host_cdrom = { | |
3827 | .format_name = "host_cdrom", | |
84a12e66 | 3828 | .protocol_name = "host_cdrom", |
f3a5d3f8 | 3829 | .instance_size = sizeof(BDRVRawState), |
030be321 | 3830 | .bdrv_needs_filename = true, |
508c7cb3 | 3831 | .bdrv_probe_device = cdrom_probe_device, |
18fa1c42 | 3832 | .bdrv_parse_filename = cdrom_parse_filename, |
66f82cee | 3833 | .bdrv_file_open = cdrom_open, |
f3a5d3f8 | 3834 | .bdrv_close = raw_close, |
1bc6b705 JC |
3835 | .bdrv_reopen_prepare = raw_reopen_prepare, |
3836 | .bdrv_reopen_commit = raw_reopen_commit, | |
3837 | .bdrv_reopen_abort = raw_reopen_abort, | |
5a5e7f8c ML |
3838 | .bdrv_co_create_opts = bdrv_co_create_opts_simple, |
3839 | .create_opts = &bdrv_create_opts_simple, | |
8a2ce0bc | 3840 | .mutable_opts = mutable_opts, |
f3a5d3f8 | 3841 | |
9d52aa3c KW |
3842 | .bdrv_co_preadv = raw_co_preadv, |
3843 | .bdrv_co_pwritev = raw_co_pwritev, | |
33d70fb6 | 3844 | .bdrv_co_flush_to_disk = raw_co_flush_to_disk, |
c25f53b0 | 3845 | .bdrv_refresh_limits = raw_refresh_limits, |
1b3abdcc ML |
3846 | .bdrv_io_plug = raw_aio_plug, |
3847 | .bdrv_io_unplug = raw_aio_unplug, | |
042b757c | 3848 | .bdrv_attach_aio_context = raw_aio_attach_aio_context, |
f3a5d3f8 | 3849 | |
061ca8a3 | 3850 | .bdrv_co_truncate = raw_co_truncate, |
b94a2610 KW |
3851 | .bdrv_getlength = raw_getlength, |
3852 | .has_variable_length = true, | |
4a1d5e1f FZ |
3853 | .bdrv_get_allocated_file_size |
3854 | = raw_get_allocated_file_size, | |
f3a5d3f8 | 3855 | |
19cb3738 | 3856 | /* removable device support */ |
f3a5d3f8 CH |
3857 | .bdrv_is_inserted = cdrom_is_inserted, |
3858 | .bdrv_eject = cdrom_eject, | |
025e849a | 3859 | .bdrv_lock_medium = cdrom_lock_medium, |
19cb3738 | 3860 | }; |
f3a5d3f8 | 3861 | #endif /* __FreeBSD__ */ |
5efa9d5a | 3862 | |
84a12e66 | 3863 | static void bdrv_file_init(void) |
5efa9d5a | 3864 | { |
508c7cb3 CH |
3865 | /* |
3866 | * Register all the drivers. Note that order is important, the driver | |
3867 | * registered last will get probed first. | |
3868 | */ | |
84a12e66 | 3869 | bdrv_register(&bdrv_file); |
5efa9d5a | 3870 | bdrv_register(&bdrv_host_device); |
f3a5d3f8 | 3871 | #ifdef __linux__ |
f3a5d3f8 CH |
3872 | bdrv_register(&bdrv_host_cdrom); |
3873 | #endif | |
a167ba50 | 3874 | #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) |
f3a5d3f8 CH |
3875 | bdrv_register(&bdrv_host_cdrom); |
3876 | #endif | |
5efa9d5a AL |
3877 | } |
3878 | ||
84a12e66 | 3879 | block_init(bdrv_file_init); |